{"generatedAt":"2026-11-10T12:00:00Z","builtAt":"2026-11-11T01:01:00.058Z","growthWindowDays":7,"downloadsWindowDays":7,"targets":[{"slug":"photoshop","name":"Adobe Photoshop","short":"Photoshop","vendor":"Adobe","category":"image"},{"slug":"illustrator","name":"Adobe Illustrator","short":"Illustrator","vendor":"Adobe","category":"design_cad"},{"slug":"premiere-pro","name":"Adobe Premiere Pro","short":"Premiere Pro","vendor":"Adobe","category":"video"},{"slug":"after-effects","name":"Adobe After Effects","short":"After Effects","vendor":"Adobe","category":"video"},{"slug":"lightroom","name":"Adobe Lightroom","short":"Lightroom","vendor":"Adobe","category":"image"},{"slug":"indesign","name":"Adobe InDesign","short":"InDesign","vendor":"Adobe","category":"design_cad"},{"slug":"acrobat","name":"Adobe Acrobat","short":"Acrobat","vendor":"Adobe","category":"office"},{"slug":"word","name":"Microsoft Word","short":"Word","vendor":"Microsoft","category":"office"},{"slug":"excel","name":"Microsoft Excel","short":"Excel","vendor":"Microsoft","category":"office"},{"slug":"powerpoint","name":"Microsoft PowerPoint","short":"PowerPoint","vendor":"Microsoft","category":"office"},{"slug":"autocad","name":"Autodesk AutoCAD","short":"AutoCAD","vendor":"Autodesk","category":"design_cad"},{"slug":"figma","name":"Figma","short":"Figma","vendor":"Figma","category":"design_cad"},{"slug":"cursor","name":"Cursor","short":"Cursor","vendor":"Anysphere","category":"development","ai":true},{"slug":"github-copilot","name":"GitHub Copilot","short":"Copilot","vendor":"GitHub","category":"development","ai":true},{"slug":"snyk","name":"Snyk","short":"Snyk","vendor":"Snyk","category":"development"},{"slug":"sublime-text","name":"Sublime Text","short":"Sublime Text","vendor":"Sublime HQ","category":"development"},{"slug":"midjourney","name":"Midjourney","short":"Midjourney","vendor":"Midjourney","category":"image","ai":true},{"slug":"autodesk-maya","name":"Autodesk Maya","short":"Maya","vendor":"Autodesk","category":"design_cad"},{"slug":"autodesk-3ds-max","name":"Autodesk 3ds Max","short":"3ds Max","vendor":"Autodesk","category":"design_cad"},{"slug":"cinema-4d","name":"Cinema 4D","short":"Cinema 4D","vendor":"Maxon","category":"design_cad"},{"slug":"typingmind","name":"TypingMind","short":"TypingMind","vendor":"TypingMind","category":"local_ai","ai":true},{"slug":"runway","name":"Runway","short":"Runway","vendor":"Runway","category":"video","ai":true},{"slug":"datagrip","name":"DataGrip","short":"DataGrip","vendor":"JetBrains","category":"development"},{"slug":"navicat","name":"Navicat","short":"Navicat","vendor":"PremiumSoft CyberTech","category":"development"},{"slug":"miro","name":"Miro","short":"Miro","vendor":"Miro","category":"productivity"},{"slug":"lucidchart","name":"Lucidchart","short":"Lucidchart","vendor":"Lucid Software","category":"productivity"},{"slug":"solidworks","name":"SolidWorks","short":"SolidWorks","vendor":"Dassault Systèmes","category":"design_cad"},{"slug":"autodesk-fusion","name":"Autodesk Fusion","short":"Fusion","vendor":"Autodesk","category":"design_cad"},{"slug":"vysor","name":"Vysor","short":"Vysor","vendor":"ClockworkMod","category":"hardware"},{"slug":"squarespace","name":"Squarespace","short":"Squarespace","vendor":"Squarespace","category":"development"},{"slug":"wix","name":"Wix","short":"Wix","vendor":"Wix.com","category":"development"},{"slug":"pictory","name":"Pictory","short":"Pictory","vendor":"Pictory","category":"video","ai":true},{"slug":"invideo-ai","name":"InVideo AI","short":"AI","vendor":"InVideo","category":"video","ai":true},{"slug":"google-photos","name":"Google Photos","short":"Photos","vendor":"Google","category":"image"},{"slug":"coreldraw","name":"CorelDRAW","short":"CorelDRAW","vendor":"Corel","category":"design_cad"},{"slug":"jev","name":"Jev","short":"Jev","vendor":"TypeSafe","category":"development"},{"slug":"microsoft-visio","name":"Microsoft Visio","short":"Visio","vendor":"Microsoft","category":"office"},{"slug":"clip-studio-paint","name":"Clip Studio Paint","short":"Clip Studio Paint","vendor":"Celsys","category":"image"},{"slug":"corel-painter","name":"Corel Painter","short":"Painter","vendor":"Corel","category":"image"},{"slug":"feedly","name":"Feedly","short":"Feedly","vendor":"Feedly","category":"productivity"},{"slug":"postman","name":"Postman","short":"Postman","vendor":"Postman","category":"development"},{"slug":"browserbase","name":"Browserbase","short":"Browserbase","vendor":"Browserbase","category":"development"},{"slug":"roam-research","name":"Roam Research","short":"Roam Research","vendor":"Roam Research","category":"productivity"},{"slug":"notion","name":"Notion","short":"Notion","vendor":"Notion Labs","category":"productivity"},{"slug":"displayfusion","name":"DisplayFusion","short":"DisplayFusion","vendor":"Binary Fortress Software","category":"network_system"},{"slug":"jetbrains-ides","name":"JetBrains IDEs","short":"IDEs","vendor":"JetBrains","category":"development"},{"slug":"zapier","name":"Zapier","short":"Zapier","vendor":"Zapier","category":"automation"},{"slug":"make","name":"Make","short":"Make","vendor":"Make","category":"automation"},{"slug":"nuke","name":"Nuke","short":"Nuke","vendor":"Foundry","category":"video"},{"slug":"blackmagic-fusion-studio","name":"Blackmagic Fusion Studio","short":"Blackmagic Fusion Studio","vendor":"Blackmagic Design","category":"video"},{"slug":"chatgpt-plus","name":"ChatGPT Plus","short":"ChatGPT Plus","vendor":"OpenAI","category":"local_ai","ai":true},{"slug":"xsplit-broadcaster","name":"XSplit Broadcaster","short":"XSplit Broadcaster","vendor":"SplitmediaLabs","category":"video"},{"slug":"camtasia","name":"Camtasia","short":"Camtasia","vendor":"TechSmith","category":"video"},{"slug":"claude-code","name":"Claude Code","short":"Claude Code","vendor":"Anthropic","category":"development","ai":true},{"slug":"otter-ai","name":"Otter.ai","short":"Otter.ai","vendor":"Otter.ai","category":"audio","ai":true},{"slug":"adobe-camera-raw","name":"Adobe Camera Raw","short":"Camera Raw","vendor":"Adobe","category":"image"},{"slug":"capture-one","name":"Capture One","short":"Capture One","vendor":"Capture One","category":"image"},{"slug":"arq-backup","name":"Arq Backup","short":"Arq Backup","vendor":"Haystack Software","category":"network_system"},{"slug":"pixelmator-pro","name":"Pixelmator Pro","short":"Pixelmator Pro","vendor":"Apple","category":"image"},{"slug":"inoreader","name":"Inoreader","short":"Inoreader","vendor":"Innologica","category":"productivity"},{"slug":"teamviewer","name":"TeamViewer","short":"TeamViewer","vendor":"TeamViewer","category":"network_system"},{"slug":"anydesk","name":"AnyDesk","short":"AnyDesk","vendor":"AnyDesk Software","category":"network_system"},{"slug":"quarkxpress","name":"QuarkXPress","short":"QuarkXPress","vendor":"Quark Software","category":"design_cad"},{"slug":"microsoft-publisher","name":"Microsoft Publisher","short":"Publisher","vendor":"Microsoft","category":"office"},{"slug":"tailwind-plus","name":"Tailwind Plus","short":"Tailwind Plus","vendor":"Tailwind Labs","category":"development"},{"slug":"snagit","name":"Snagit","short":"Snagit","vendor":"TechSmith","category":"image"},{"slug":"carlinkit-wireless-carplay-adapter","name":"Carlinkit wireless CarPlay adapter","short":"wireless CarPlay adapter","vendor":"Carlinkit","category":"hardware"},{"slug":"adobe-creative-cloud","name":"Adobe Creative Cloud","short":"Creative Cloud","vendor":"Adobe","category":"design_cad"},{"slug":"dropbox","name":"Dropbox","short":"Dropbox","vendor":"Dropbox","category":"productivity"},{"slug":"resilio-sync","name":"Resilio Sync","short":"Sync","vendor":"Resilio","category":"network_system"},{"slug":"topaz-gigapixel-ai","name":"Topaz Gigapixel AI","short":"Topaz Gigapixel AI","vendor":"Topaz Labs","category":"image","ai":true},{"slug":"insomnia","name":"Insomnia","short":"Insomnia","vendor":"Kong","category":"development"},{"slug":"4k-video-downloader","name":"4K Video Downloader","short":"4K Video Downloader","vendor":"Open Media","category":"video"},{"slug":"heygen","name":"HeyGen","short":"HeyGen","vendor":"HeyGen","category":"video","ai":true},{"slug":"synthesia","name":"Synthesia","short":"Synthesia","vendor":"Synthesia","category":"video","ai":true},{"slug":"projectivy-launcher","name":"Projectivy Launcher","short":"Launcher","vendor":"Projectivy","category":"hardware"},{"slug":"searchlight-exposure-platform","name":"Searchlight Exposure Platform","short":"Searchlight Exposure Platform","vendor":"Searchlight Cyber","category":"network_system"},{"slug":"elevenlabs","name":"ElevenLabs","short":"ElevenLabs","vendor":"ElevenLabs","category":"audio","ai":true},{"slug":"descript","name":"Descript","short":"Descript","vendor":"Descript","category":"audio","ai":true},{"slug":"ancestry","name":"Ancestry","short":"Ancestry","vendor":"Ancestry","category":"productivity"},{"slug":"myheritage","name":"MyHeritage","short":"MyHeritage","vendor":"MyHeritage","category":"productivity"},{"slug":"quizlet","name":"Quizlet","short":"Quizlet","vendor":"Quizlet","category":"productivity"},{"slug":"maimemo","name":"Maimemo","short":"Maimemo","vendor":"Maimemo","category":"productivity"},{"slug":"microsoft-outlook","name":"Microsoft Outlook","short":"Outlook","vendor":"Microsoft","category":"office"},{"slug":"gitkraken","name":"GitKraken","short":"GitKraken","vendor":"Axosoft","category":"development"},{"slug":"notebooklm","name":"NotebookLM","short":"NotebookLM","vendor":"Google","category":"local_ai","ai":true},{"slug":"tradingview","name":"TradingView","short":"TradingView","vendor":"TradingView","category":"other"},{"slug":"benzinga-pro","name":"Benzinga Pro","short":"Pro","vendor":"Benzinga","category":"other"},{"slug":"v0-by-vercel","name":"v0 by Vercel","short":"v0 by Vercel","vendor":"Vercel","category":"development","ai":true},{"slug":"ideogram","name":"Ideogram","short":"Ideogram","vendor":"Ideogram","category":"image","ai":true},{"slug":"adobe-firefly","name":"Adobe Firefly","short":"Firefly","vendor":"Adobe","category":"image","ai":true},{"slug":"bright-data","name":"Bright Data","short":"Bright Data","vendor":"Bright Data","category":"development"},{"slug":"x-twitter-api","name":"X (Twitter) API","short":"X (Twitter) API","vendor":"X Corp","category":"social_marketing"},{"slug":"coderabbit","name":"CodeRabbit","short":"CodeRabbit","vendor":"CodeRabbit","category":"development","ai":true},{"slug":"polycam","name":"Polycam","short":"Polycam","vendor":"Polycam","category":"design_cad"},{"slug":"matterport","name":"Matterport","short":"Matterport","vendor":"Matterport","category":"design_cad"},{"slug":"winrar","name":"WinRAR","short":"WinRAR","vendor":"win.rar GmbH","category":"network_system"},{"slug":"winzip","name":"WinZip","short":"WinZip","vendor":"Corel","category":"network_system"},{"slug":"claude-motion","name":"Claude Motion","short":"Claude Motion","vendor":"Anthropic","category":"video","ai":true},{"slug":"steam-deck","name":"Steam Deck","short":"Steam Deck","vendor":"Valve","category":"games"},{"slug":"excalidraw","name":"Excalidraw+","short":"Excalidraw+","vendor":"Excalidraw","category":"design_cad"},{"slug":"buffer","name":"Buffer","short":"Buffer","vendor":"Buffer","category":"social_marketing"},{"slug":"hootsuite","name":"Hootsuite","short":"Hootsuite","vendor":"Hootsuite","category":"social_marketing"},{"slug":"later","name":"Later","short":"Later","vendor":"Later","category":"social_marketing"},{"slug":"iverify","name":"iVerify","short":"iVerify","vendor":"iVerify","category":"network_system"},{"slug":"unity","name":"Unity","short":"Unity","vendor":"Unity Technologies","category":"games"},{"slug":"playcanvas","name":"PlayCanvas","short":"PlayCanvas","vendor":"PlayCanvas","category":"games"},{"slug":"sketch","name":"Sketch","short":"Sketch","vendor":"Sketch B.V.","category":"design_cad"},{"slug":"cluely","name":"Cluely","short":"Cluely","vendor":"Cluely","category":"productivity","ai":true},{"slug":"hopper","name":"Hopper","short":"Hopper","vendor":"Cryptic Apps","category":"development"},{"slug":"parsec","name":"Parsec","short":"Parsec","vendor":"Parsec","category":"network_system"},{"slug":"remotion","name":"Remotion","short":"Remotion","vendor":"Remotion","category":"video"},{"slug":"gamma","name":"Gamma","short":"Gamma","vendor":"Gamma","category":"office","ai":true},{"slug":"microsoft-365-copilot","name":"Microsoft 365 Copilot","short":"365 Copilot","vendor":"Microsoft","category":"office","ai":true},{"slug":"langsmith","name":"LangSmith","short":"LangSmith","vendor":"LangChain","category":"development"},{"slug":"braintrust","name":"Braintrust","short":"Braintrust","vendor":"Braintrust","category":"development"},{"slug":"move-ai","name":"Move AI","short":"Move AI","vendor":"Move AI","category":"design_cad","ai":true},{"slug":"deepmotion","name":"DeepMotion","short":"DeepMotion","vendor":"DeepMotion","category":"design_cad","ai":true},{"slug":"wintousb","name":"WinToUSB","short":"WinToUSB","vendor":"EasyUEFI Development Team","category":"network_system"},{"slug":"openrouter","name":"OpenRouter","short":"OpenRouter","vendor":"OpenRouter","category":"development"},{"slug":"portkey","name":"Portkey","short":"Portkey","vendor":"Portkey","category":"development"},{"slug":"mem0","name":"Mem0","short":"Mem0","vendor":"Mem0","category":"development"},{"slug":"zep","name":"Zep","short":"Zep","vendor":"Zep","category":"development"},{"slug":"origin-wireless","name":"Origin Wireless","short":"Origin Wireless","vendor":"Origin Wireless","category":"hardware"},{"slug":"goodfire-ember","name":"Goodfire Ember","short":"Ember","vendor":"Goodfire","category":"development"},{"slug":"affinity-designer","name":"Affinity Designer","short":"Affinity Designer","vendor":"Canva","category":"design_cad"},{"slug":"final-cut-pro","name":"Final Cut Pro","short":"Final Cut Pro","vendor":"Apple","category":"video"},{"slug":"kugou-music","name":"Kugou Music","short":"Kugou Music","vendor":"Tencent Music Entertainment","category":"audio"},{"slug":"kuwo-music","name":"Kuwo Music","short":"Kuwo Music","vendor":"Tencent Music Entertainment","category":"audio"},{"slug":"migu-music","name":"Migu Music","short":"Migu Music","vendor":"China Mobile","category":"audio"},{"slug":"ida-pro","name":"IDA Pro","short":"IDA Pro","vendor":"Hex-Rays","category":"development"},{"slug":"surge","name":"Surge","short":"Surge","vendor":"Surge Networks","category":"network_system"},{"slug":"proxifier","name":"Proxifier","short":"Proxifier","vendor":"Initex","category":"network_system"},{"slug":"grok-bot","name":"Grok Bot","short":"Grok Bot","vendor":"xAI","category":"ai_agents","ai":true},{"slug":"manus","name":"Manus","short":"Manus","vendor":"Manus","category":"ai_agents","ai":true},{"slug":"abbyy-finereader","name":"ABBYY FineReader","short":"FineReader","vendor":"ABBYY","category":"office"},{"slug":"mathpix","name":"Mathpix","short":"Mathpix","vendor":"Mathpix","category":"office"},{"slug":"aqua-platform","name":"Aqua Platform","short":"Aqua Platform","vendor":"Aqua Security","category":"development"},{"slug":"dbeaver-pro","name":"DBeaver PRO","short":"PRO","vendor":"DBeaver","category":"development"},{"slug":"aseprite","name":"Aseprite","short":"Aseprite","vendor":"Igara Studio","category":"image"},{"slug":"supermemory","name":"Supermemory","short":"Supermemory","vendor":"Supermemory","category":"development"},{"slug":"graphify-cloud","name":"Graphify Cloud","short":"Graphify Cloud","vendor":"Graphify Labs","category":"development"},{"slug":"jobscan","name":"Jobscan","short":"Jobscan","vendor":"Jobscan","category":"productivity"},{"slug":"teal","name":"Teal","short":"Teal","vendor":"Teal","category":"productivity"},{"slug":"kickresume","name":"Kickresume","short":"Kickresume","vendor":"Kickresume","category":"productivity"},{"slug":"icloud-photos","name":"iCloud Photos","short":"iCloud Photos","vendor":"Apple","category":"image"},{"slug":"meta-muse","name":"Meta Muse","short":"Muse","vendor":"Meta","category":"ai_agents","ai":true},{"slug":"openai-dots","name":"OpenAI dots","short":"dots","vendor":"OpenAI","category":"ai_agents","ai":true},{"slug":"conductor","name":"Conductor","short":"Conductor","vendor":"Melty Labs","category":"development","ai":true},{"slug":"houdahspot","name":"HoudahSpot","short":"HoudahSpot","vendor":"Houdah Software","category":"productivity"},{"slug":"cube-acr","name":"Cube ACR","short":"Cube ACR","vendor":"Catalina Group","category":"audio"},{"slug":"browser-use-cloud","name":"Browser Use Cloud","short":"Cloud","vendor":"Browser Use","category":"ai_agents","ai":true},{"slug":"hone","name":"Hone","short":"Hone","vendor":"Hone","category":"games"},{"slug":"highcharts","name":"Highcharts","short":"Highcharts","vendor":"Highsoft","category":"development"},{"slug":"amcharts","name":"amCharts","short":"amCharts","vendor":"amCharts","category":"development"},{"slug":"warp","name":"Warp","short":"Warp","vendor":"Warp","category":"network_system"},{"slug":"easeus-partition-master","name":"EaseUS Partition Master","short":"Partition Master","vendor":"EaseUS","category":"network_system"},{"slug":"quillbot","name":"QuillBot","short":"QuillBot","vendor":"QuillBot","category":"productivity","ai":true},{"slug":"grammarly","name":"Grammarly","short":"Grammarly","vendor":"Grammarly","category":"productivity","ai":true},{"slug":"eraser","name":"Eraser","short":"Eraser","vendor":"Eraser","category":"design_cad"},{"slug":"prisma","name":"Prisma","short":"Prisma","vendor":"Prisma","category":"development"},{"slug":"suno","name":"Suno","short":"Suno","vendor":"Suno","category":"audio","ai":true},{"slug":"udio","name":"Udio","short":"Udio","vendor":"Udio","category":"audio","ai":true},{"slug":"semrush","name":"Semrush","short":"Semrush","vendor":"Semrush","category":"social_marketing"},{"slug":"ahrefs","name":"Ahrefs","short":"Ahrefs","vendor":"Ahrefs","category":"social_marketing"},{"slug":"screaming-frog-seo-spider","name":"Screaming Frog SEO Spider","short":"SEO Spider","vendor":"Screaming Frog","category":"social_marketing"},{"slug":"plex","name":"Plex","short":"Plex","vendor":"Plex","category":"video"},{"slug":"moz-pro","name":"Moz Pro","short":"Pro","vendor":"Moz","category":"social_marketing"},{"slug":"epidemic-sound","name":"Epidemic Sound","short":"Epidemic Sound","vendor":"Epidemic Sound","category":"audio"},{"slug":"phantombuster","name":"PhantomBuster","short":"PhantomBuster","vendor":"PhantomBuster","category":"social_marketing"},{"slug":"expandi","name":"Expandi","short":"Expandi","vendor":"Expandi.io","category":"social_marketing"},{"slug":"linked-helper","name":"Linked Helper","short":"Linked Helper","vendor":"Linked Helper","category":"social_marketing"},{"slug":"compresr","name":"Compresr","short":"Compresr","vendor":"Compresr","category":"development"},{"slug":"the-token-company","name":"The Token Company","short":"The Token Company","vendor":"The Token Company","category":"development"},{"slug":"paperpal","name":"Paperpal","short":"Paperpal","vendor":"Cactus Communications","category":"office","ai":true},{"slug":"elicit","name":"Elicit","short":"Elicit","vendor":"Elicit","category":"productivity","ai":true},{"slug":"emby","name":"Emby","short":"Emby","vendor":"Emby","category":"video"},{"slug":"google-cloud-speech-to-text","name":"Google Cloud Speech-to-Text","short":"Cloud Speech-to-Text","vendor":"Google","category":"audio"},{"slug":"manychat","name":"ManyChat","short":"ManyChat","vendor":"ManyChat","category":"social_marketing"},{"slug":"chatfuel","name":"Chatfuel","short":"Chatfuel","vendor":"Chatfuel","category":"social_marketing"},{"slug":"sendpulse","name":"SendPulse","short":"SendPulse","vendor":"SendPulse","category":"social_marketing"},{"slug":"jasper","name":"Jasper","short":"Jasper","vendor":"Jasper","category":"social_marketing","ai":true},{"slug":"speechify","name":"Speechify","short":"Speechify","vendor":"Speechify","category":"audio","ai":true},{"slug":"audible","name":"Audible","short":"Audible","vendor":"Amazon","category":"audio"},{"slug":"undetectable-ai","name":"Undetectable AI","short":"Undetectable AI","vendor":"Undetectable AI","category":"productivity","ai":true},{"slug":"adobe-animate","name":"Adobe Animate","short":"Animate","vendor":"Adobe","category":"video"},{"slug":"solid-explorer","name":"Solid Explorer","short":"Solid Explorer","vendor":"NeatBytes","category":"network_system"},{"slug":"canva","name":"Canva","short":"Canva","vendor":"Canva","category":"design_cad"},{"slug":"claude-design","name":"Claude Design","short":"Claude Design","vendor":"Anthropic","category":"design_cad","ai":true},{"slug":"deleteme","name":"DeleteMe","short":"DeleteMe","vendor":"Abine","category":"productivity"},{"slug":"incogni","name":"Incogni","short":"Incogni","vendor":"Surfshark","category":"productivity"},{"slug":"genymotion","name":"Genymotion","short":"Genymotion","vendor":"Genymobile","category":"development"},{"slug":"browserstack-app-automate","name":"BrowserStack App Automate","short":"App Automate","vendor":"BrowserStack","category":"development"},{"slug":"prerender-io","name":"Prerender.io","short":"Prerender.io","vendor":"Prerender.io","category":"development"},{"slug":"mobaxterm","name":"MobaXterm","short":"MobaXterm","vendor":"Mobatek","category":"network_system"},{"slug":"kolide","name":"Kolide","short":"Kolide","vendor":"Kolide","category":"network_system"},{"slug":"twilio-whatsapp-api","name":"Twilio WhatsApp API","short":"WhatsApp API","vendor":"Twilio","category":"social_marketing"},{"slug":"localstack","name":"LocalStack","short":"LocalStack","vendor":"LocalStack","category":"development"},{"slug":"amazon-web-services","name":"Amazon Web Services","short":"Web Services","vendor":"Amazon","category":"development"},{"slug":"ltx-studio","name":"LTX Studio","short":"LTX Studio","vendor":"Lightricks","category":"video","ai":true},{"slug":"google-cloud-mcp-servers","name":"Google Cloud MCP Servers","short":"Cloud MCP Servers","vendor":"Google","category":"development"},{"slug":"parallels-desktop","name":"Parallels Desktop","short":"Parallels Desktop","vendor":"Alludo","category":"network_system"},{"slug":"claude-cowork","name":"Claude Cowork","short":"Claude Cowork","vendor":"Anthropic","category":"ai_agents","ai":true},{"slug":"cpanel","name":"cPanel","short":"cPanel","vendor":"WebPros","category":"network_system"},{"slug":"plesk","name":"Plesk","short":"Plesk","vendor":"WebPros","category":"network_system"},{"slug":"aapanel","name":"aaPanel","short":"aaPanel","vendor":"aaPanel","category":"network_system"},{"slug":"tikhub-io","name":"TikHub.io","short":"TikHub.io","vendor":"TikHub","category":"social_marketing"},{"slug":"chatbase","name":"Chatbase","short":"Chatbase","vendor":"Chatbase","category":"ai_agents","ai":true},{"slug":"intercom","name":"Intercom","short":"Intercom","vendor":"Intercom","category":"social_marketing"},{"slug":"replyguy","name":"ReplyGuy","short":"ReplyGuy","vendor":"ReplyGuy","category":"social_marketing","ai":true},{"slug":"n8n-cloud","name":"n8n Cloud","short":"Cloud","vendor":"n8n","category":"automation"},{"slug":"vidiq","name":"vidIQ","short":"vidIQ","vendor":"vidIQ","category":"social_marketing"},{"slug":"tubebuddy","name":"TubeBuddy","short":"TubeBuddy","vendor":"TubeBuddy","category":"social_marketing"},{"slug":"lovable","name":"Lovable","short":"Lovable","vendor":"Lovable","category":"development","ai":true},{"slug":"bolt-new","name":"Bolt.new","short":"Bolt.new","vendor":"StackBlitz","category":"development","ai":true},{"slug":"replit","name":"Replit","short":"Replit","vendor":"Replit","category":"development","ai":true},{"slug":"streamladder","name":"StreamLadder","short":"StreamLadder","vendor":"StreamLadder","category":"video"},{"slug":"murf","name":"Murf","short":"Murf","vendor":"Murf","category":"audio","ai":true},{"slug":"fish-audio","name":"Fish Audio","short":"Fish Audio","vendor":"Fish Audio","category":"audio","ai":true},{"slug":"blink-shell","name":"Blink Shell","short":"Blink Shell","vendor":"Blink Shell","category":"network_system"},{"slug":"novinhub","name":"Novinhub","short":"Novinhub","vendor":"Novinhub","category":"social_marketing"},{"slug":"rask-ai","name":"Rask AI","short":"Rask AI","vendor":"Rask AI","category":"video","ai":true},{"slug":"respeecher","name":"Respeecher","short":"Respeecher","vendor":"Respeecher","category":"audio","ai":true},{"slug":"kits-ai","name":"Kits AI","short":"Kits AI","vendor":"Kits AI","category":"audio","ai":true},{"slug":"perplexity","name":"Perplexity","short":"Perplexity","vendor":"Perplexity AI","category":"productivity","ai":true},{"slug":"notion-ai","name":"Notion AI","short":"Notion AI","vendor":"Notion Labs","category":"productivity","ai":true},{"slug":"tivimate","name":"TiviMate","short":"TiviMate","vendor":"TiviMate","category":"video"},{"slug":"deepwiki","name":"DeepWiki","short":"DeepWiki","vendor":"Cognition","category":"development","ai":true},{"slug":"mintlify","name":"Mintlify","short":"Mintlify","vendor":"Mintlify","category":"development"},{"slug":"roboflow","name":"roboflow","short":"roboflow","vendor":"Roboflow","category":"development","ai":true},{"slug":"opusclip","name":"OpusClip","short":"OpusClip","vendor":"Opus Clip","category":"video","ai":true},{"slug":"icons8","name":"Icons8","short":"Icons8","vendor":"Icons8","category":"design_cad"},{"slug":"iconfinder","name":"Iconfinder","short":"Iconfinder","vendor":"Iconfinder","category":"design_cad"},{"slug":"cleanmymac","name":"CleanMyMac","short":"CleanMyMac","vendor":"MacPaw","category":"network_system"},{"slug":"d-id","name":"D-ID","short":"D-ID","vendor":"D-ID","category":"video","ai":true},{"slug":"perplexity-comet","name":"Perplexity Comet","short":"Perplexity Comet","vendor":"Perplexity AI","category":"ai_agents","ai":true},{"slug":"uipath","name":"UiPath","short":"UiPath","vendor":"UiPath","category":"automation"},{"slug":"tripit","name":"TripIt","short":"TripIt","vendor":"SAP Concur","category":"productivity"},{"slug":"wanderlog","name":"Wanderlog","short":"Wanderlog","vendor":"Wanderlog","category":"productivity"},{"slug":"splitwise","name":"Splitwise","short":"Splitwise","vendor":"Splitwise","category":"productivity"},{"slug":"pika","name":"Pika","short":"Pika","vendor":"Pika","category":"video","ai":true},{"slug":"reallusion-character-creator","name":"Reallusion Character Creator","short":"Character Creator","vendor":"Reallusion","category":"design_cad"},{"slug":"socialrails","name":"SocialRails","short":"SocialRails","vendor":"SocialRails","category":"social_marketing"},{"slug":"lynote-ai","name":"Lynote.ai","short":"Lynote.ai","vendor":"Lynote.ai","category":"productivity","ai":true},{"slug":"youtube-premium","name":"YouTube Premium","short":"YouTube Premium","vendor":"Google","category":"video"},{"slug":"world-of-warcraft","name":"World of Warcraft","short":"World of Warcraft","vendor":"Blizzard Entertainment","category":"games"},{"slug":"qustodio","name":"Qustodio","short":"Qustodio","vendor":"Qustodio","category":"productivity"},{"slug":"nero-burning-rom","name":"Nero Burning ROM","short":"Burning ROM","vendor":"Nero","category":"network_system"},{"slug":"character-ai","name":"Character.AI","short":"Character.AI","vendor":"Character.AI","category":"ai_agents","ai":true},{"slug":"replika","name":"Replika","short":"Replika","vendor":"Luka","category":"ai_agents","ai":true},{"slug":"optmyzr","name":"Optmyzr","short":"Optmyzr","vendor":"Optmyzr","category":"social_marketing"},{"slug":"semgrep","name":"Semgrep","short":"Semgrep","vendor":"Semgrep","category":"development"},{"slug":"deepl-pro","name":"DeepL Pro","short":"Pro","vendor":"DeepL","category":"office"},{"slug":"immersive-translate","name":"Immersive Translate","short":"Immersive Translate","vendor":"Immersive Translate","category":"office"},{"slug":"infuse","name":"Infuse","short":"Infuse","vendor":"Firecore","category":"video"},{"slug":"live2d-cubism","name":"Live2D Cubism","short":"Cubism","vendor":"Live2D","category":"video"},{"slug":"kling","name":"Kling","short":"Kling","vendor":"Kuaishou","category":"video","ai":true},{"slug":"hunter-io","name":"Hunter.io","short":"Hunter.io","vendor":"Hunter","category":"social_marketing"},{"slug":"playstation-plus-premium","name":"PlayStation Plus Premium","short":"PlayStation Plus Premium","vendor":"Sony Interactive Entertainment","category":"games"},{"slug":"internet-download-manager","name":"Internet Download Manager","short":"Internet Download Manager","vendor":"Tonec","category":"network_system"},{"slug":"google-veo-3","name":"Google Veo 3","short":"Veo 3","vendor":"Google","category":"video","ai":true},{"slug":"tableplus","name":"TablePlus","short":"TablePlus","vendor":"TablePlus","category":"development"},{"slug":"viggle","name":"Viggle","short":"Viggle","vendor":"Viggle AI","category":"video","ai":true},{"slug":"termius","name":"Termius","short":"Termius","vendor":"Termius","category":"network_system"},{"slug":"qodo","name":"Qodo","short":"Qodo","vendor":"Qodo","category":"development","ai":true},{"slug":"surfer-seo","name":"Surfer SEO","short":"SEO","vendor":"Surfer","category":"social_marketing","ai":true},{"slug":"jscrambler","name":"Jscrambler","short":"Jscrambler","vendor":"Jscrambler","category":"development"},{"slug":"resolve-ai","name":"Resolve.ai","short":"Resolve.ai","vendor":"Resolve.ai","category":"development","ai":true},{"slug":"mx-player-pro","name":"MX Player Pro","short":"MX Player Pro","vendor":"J2 Interactive","category":"video"}],"rankings":{"growth":[{"list":"growth","position":1,"slug":"morluto__rea","value":22677,"windowDays":7},{"list":"growth","position":2,"slug":"pingdotgg__t3code","value":11234,"windowDays":7},{"list":"growth","position":3,"slug":"ccch1mneyyy__dsh-go","value":11204,"windowDays":7},{"list":"growth","position":4,"slug":"cloudflare__security-audit-skill","value":10927,"windowDays":7},{"list":"growth","position":5,"slug":"mattpocock__skills","value":9359,"windowDays":7},{"list":"growth","position":6,"slug":"milind-soni__openmausbot","value":8888,"windowDays":7},{"list":"growth","position":7,"slug":"debpalash__voicestudio","value":7098,"windowDays":7},{"list":"growth","position":8,"slug":"deepseek-ai__deepseek-harness","value":7064,"windowDays":7},{"list":"growth","position":9,"slug":"stablyai__orca","value":6830,"windowDays":7},{"list":"growth","position":10,"slug":"heydamir__unslop","value":6104,"windowDays":7}],"downloads":[{"list":"downloads","position":1,"slug":"astral-sh__uv","value":39153795,"windowDays":7},{"list":"downloads","position":2,"slug":"stablyai__orca","value":21501148,"windowDays":7},{"list":"downloads","position":3,"slug":"jdx__mise","value":8272910,"windowDays":7},{"list":"downloads","position":4,"slug":"pingdotgg__t3code","value":5954290,"windowDays":7},{"list":"downloads","position":5,"slug":"cli__cli","value":4029437,"windowDays":7},{"list":"downloads","position":6,"slug":"ollama__ollama","value":3721029,"windowDays":7},{"list":"downloads","position":7,"slug":"aquasecurity__trivy","value":3652929,"windowDays":7},{"list":"downloads","position":8,"slug":"burntsushi__ripgrep","value":2895927,"windowDays":7},{"list":"downloads","position":9,"slug":"anomalyco__opencode","value":2731763,"windowDays":7},{"list":"downloads","position":10,"slug":"clash-verge-rev__clash-verge-rev","value":2531299,"windowDays":7}],"new":[{"list":"new","position":1,"slug":"anefzaoui__misterdownloadmanager","value":0},{"list":"new","position":2,"slug":"invoke-ai__invokeai","value":0},{"list":"new","position":3,"slug":"juneyaooo__rea-video","value":0},{"list":"new","position":4,"slug":"pkhungurn__talking-head-anime-2-demo","value":0},{"list":"new","position":5,"slug":"pkhungurn__talking-head-anime-3-demo","value":0},{"list":"new","position":6,"slug":"pkhungurn__talking-head-anime-demo","value":0},{"list":"new","position":7,"slug":"pku-yuangroup__helios","value":0},{"list":"new","position":8,"slug":"playvoice__whisper-vits-svc","value":0},{"list":"new","position":9,"slug":"pnnbao97__vieneu-tts","value":0},{"list":"new","position":10,"slug":"pot-app__pot-desktop","value":0}]},"projects":[{"id":"MDEwOlJlcG9zaXRvcnk5MTMzNjc3","slug":"1hakr__anexplorer","name":"AnExplorer","owner":"1hakr","fullName":"1hakr/AnExplorer","type":"ready_app","status":"active","grade":"A","easeByPlatform":{},"risk":"attention","signals":[{"code":"ads_in_free_version","severity":"attention","evidence":"README.md: \"The free version of AnExplorer has ads. You can buy the pro version for an adfree version which is less than a beer!\" 广告代码通常会附带追踪,而仓库对此没有说明。","source":"ai","title":"免费版带广告"},{"code":"ftp_server_shares_phone_files_on_the_network","severity":"attention","evidence":"README.md, Network Manager: \"Transfer files with FTP Server from phone to PC. All you have to do is open the IP address in browser and all your files on the phone are easily accessible.\" README 没有提到密码或将其限制在自己的网络内,所以同一 Wi-Fi 下的其他人在该功能开启期间也可能读取到你的文件。","source":"ai","title":"ftp 服务器把手机文件分享到网络"},{"code":"root_features_can_break_the_phone","severity":"info","evidence":"README.md: \"Root File Manager (File Explorer) for advanced users to explore, edit, copy, paste and delete files in the root partition of phone storage... Explore root folders of the system like data, cache. You phone needs to be rooted to use this feature.\" 在那里删错文件可能导致手机无法启动。","source":"ai","title":"root 功能可能损坏手机"},{"code":"no_download_in_this_repository","severity":"info","evidence":"该仓库有 0 个 GitHub release；README.md 中的每个下载链接都指向一个应用商店（Google Play、Amazon、三星 Galaxy Store、Aptoide）。所以你安装的应用是作者自行构建的,无法与这份源代码对照核实。","source":"ai","title":"本仓库没有下载包"},{"code":"readme_out_of_date","severity":"info","evidence":"README.md 仍描述支持 \"all Android versions from Jellybean, Kitkat, Marshmallow to Nougat\"（2012 到 2016 年间发布的 Android 版本）以及 3 MB 的应用体积,而受检快照中的最后一次提交是在 2026-06-02,距检查时间 155 天前。代码仍在维护,文字没有。","source":"ai","title":"readme 内容过时"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["network_system","productivity"],"alternativeTo":[{"slug":"solid-explorer","name":"Solid Explorer","confirmed":true}],"declaredAlternatives":[],"stars":2054,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":4,"growthDays":6,"history":[{"date":"2026-11-05","stars":2050},{"date":"2026-11-06","stars":2050},{"date":"2026-11-07","stars":2050},{"date":"2026-11-08","stars":2052},{"date":"2026-11-09","stars":2053},{"date":"2026-11-10","stars":2054}],"firstPublishedAt":"2026-11-04T22:32:54.762Z","commitSha":"e856926de57c4e403b8678bf3b289ac618065648","verifiedAt":"2026-11-04T22:17:57.131Z","lastCommitAt":"2026-06-02T03:49:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Android file manager for phones, tablets and TV: moves files, opens USB drives and shares them over FTP.","summary":"A file manager for Android, made since 2013 by one developer. It copies, moves, renames, compresses and extracts files, sorts them into libraries (photos, videos, music, documents, archives, apps), reads SD cards and USB sticks through USB OTG, manages installed apps, edits text files and can turn the phone into an FTP server so a computer on the same network can fetch the files. It also works on Android TV, Wear OS and Chromebooks. You install it from Google Play, Amazon, Samsung Galaxy Store or Aptoide; the free version shows ads.","easeRationale":"Android is the only system it runs on, and there it is an A: you install it from Google Play, Amazon, the Samsung Galaxy Store or Aptoide and open it, with nothing else to set up. Computers get no grade; there is no desktop version.","costNote":"README: \"The free version of AnExplorer has ads. You can buy the pro version for an adfree version which is less than a beer!\" The repository gives no exact price.","notVerified":["Whether the app in the stores is built from this source code","What the ads and any tracking inside the free version collect","Which Android permissions it asks for in practice","Whether the FTP server can be protected with a password","Whether the root features still work on recent Android versions","The real price of the pro version"]},"zh-Hans":{"summaryShort":"适用于手机、平板和电视的 Android 文件管理器：移动文件、打开 USB 存储，并通过 FTP 分享。","summary":"一款 Android 文件管理器，由一名开发者从 2013 年起制作。它可以复制、移动、重命名、压缩和解压文件，把文件整理到各类库中（照片、视频、音乐、文档、压缩包、应用），通过 USB OTG 读取 SD 卡和 U 盘，管理已安装的应用，编辑文本文件，还能把手机变成 FTP 服务器，让同一网络下的电脑取用文件。它也能在 Android TV、Wear OS 和 Chromebook 上运行。可从 Google Play、Amazon、三星 Galaxy Store 或 Aptoide 安装；免费版带广告。","easeRationale":"Android 是它唯一支持的系统,在那里评级是 A：从 Google Play、Amazon、三星 Galaxy Store 或 Aptoide 安装后打开即可，没有其他需要设置的东西。电脑端没有评级,因为没有桌面版本。","costNote":"README：“AnExplorer 的免费版带广告。你可以购买专业版来去掉广告，价格比一杯啤酒还便宜！”仓库里没有给出确切价格。","notVerified":["商店里的应用是否基于这份源代码构建","免费版里的广告以及任何追踪手段具体收集了什么","实际使用中它会请求哪些 Android 权限","FTP 服务器是否可以设置密码保护","root 相关功能在较新的 Android 版本上是否仍能使用","专业版的真实价格"]}}},{"id":"R_kgDOHrwnHA","slug":"1panel-dev__1panel","name":"1Panel","owner":"1Panel-dev","fullName":"1Panel-dev/1Panel","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"program_downloaded_from_vendor_site_not_github","severity":"attention","evidence":"README.md 中唯一的安装说明是 `bash -c \"$(curl -sSL https://resource.1panel.pro/v2/quick_start.sh)\"`，直接从公司自己的域名运行脚本，而最近三个版本（v2.4.2、v2.4.1、v2.4.0）都没有附带各平台的二进制文件，所以装到你服务器上的东西并非来自 GitHub 的发布页面。","source":"ai","title":"程序从厂商网站下载，而非 github"},{"code":"admin_panel_reachable_over_the_network","severity":"attention","evidence":"README.md: \"After installation, open `http://<your-server-ip>:<port>/<security-path>` in your browser.\" 该面板能安装软件、编辑文件、读取数据库和运行容器，所以任何用密码访问该地址的人都能控制整台服务器，而示例地址用的是普通 http 而不是 https。","source":"ai","title":"管理面板可通过网络访问"},{"code":"open_core_limits_free_edition","severity":"info","evidence":"README 自己的表格限制了免费版：\"AI agents (OpenClaw)\" 在 OSS 中是 \"5 agent\"，Pro 版则无限制；\"WAF & advanced security\" 在免费版中标为 \"Basic\"；而网站防篡改、可用性监控、多节点管理、自定义 logo、KVM 网页界面和 AI 网关在 OSS 中均标为不可用。","source":"ai","title":"开源核心模式限制了免费版"},{"code":"owner_written_comparison_and_user_numbers","severity":"info","evidence":"将其与 cPanel、Plesk、aaPanel 和 Webmin 对比的 \"Why 1Panel?\" 表格及其中的数字，包括 \"Loved by a global community of 2.5M+ self-hosters\" 和 \"165+ apps\"，都是项目所有者自己写的，没有其他人核实。","source":"ai","title":"对比数据和用户数字由项目方自己撰写"}],"platforms":["linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"cpanel","name":"cPanel","confirmed":true},{"slug":"plesk","name":"Plesk","confirmed":true},{"slug":"aapanel","name":"aaPanel","confirmed":true}],"declaredAlternatives":[],"stars":37426,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":75,"growthDays":7,"history":[{"date":"2026-11-05","stars":37370},{"date":"2026-11-06","stars":37375},{"date":"2026-11-07","stars":37383},{"date":"2026-11-08","stars":37394},{"date":"2026-11-09","stars":37416},{"date":"2026-11-10","stars":37426}],"firstPublishedAt":"2026-11-04T22:38:15.138Z","commitSha":"730dbbab5b47097e67fca43935e979cdc08231ab","verifiedAt":"2026-11-04T22:19:13.769Z","lastCommitAt":"2026-11-10T12:56:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Puts a web control panel on a Linux server so you can set up websites, apps, databases and backups by clicking.","summary":"A control panel for a Linux server, used from a browser. After one command over SSH you get pages for watching the machine, managing files, databases and Docker containers, an app store of ready open-source programs (WordPress, Halo and others) with domain and SSL set-up, a basic web firewall and one-click backups to cloud storage. Newer versions also manage AI models and agents. You need your own Linux server or VPS, and some features are kept for the paid Pro and Ent editions.","easeRationale":"Linux only, and only as a server: you need your own machine or VPS, you connect over SSH, run the installer script, then read the login details with `1pctl user-info` before opening the panel in a browser. There is no version for Windows or Mac.","costNote":"README: \"1Panel OSS is free forever\", with Pro and Ent editions adding multi-node management, website tamper protection, uptime monitoring, custom branding, a KVM web interface, an AI gateway and priority support. No prices in the repository.","notVerified":["What the install script downloads and changes on your server","How much memory and disk the panel and its containers use","Whether the programs in the built-in app store are kept up to date","How well the basic firewall and security features actually protect the machine","The price of the Pro and Ent editions","Whether the AI model and agent features work with the providers you use"]},"zh-Hans":{"summaryShort":"在 Linux 服务器上装一个网页控制面板，点点鼠标就能配置网站、应用、数据库和备份。","summary":"一款通过浏览器使用的 Linux 服务器控制面板。通过 SSH 执行一条命令后,就能看到监控机器状态、管理文件、数据库和 Docker 容器的页面,一个现成开源程序的应用商店（WordPress、Halo 等）并带域名和 SSL 设置,一个基础的网页防火墙,以及一键备份到云存储。较新版本还能管理 AI 模型和智能体。你需要自己的 Linux 服务器或 VPS,部分功能只保留给付费的 Pro 和 Ent 版本。","easeRationale":"只支持 Linux,而且只能作为服务器使用：你需要自己的机器或 VPS,通过 SSH 连接,运行安装脚本,再用 `1pctl user-info` 读取登录信息,然后在浏览器中打开面板。没有 Windows 或 Mac 版本。","costNote":"README：“1Panel 开源版永久免费”，Pro 和 Ent 版增加了多节点管理、网站防篡改、可用性监控、自定义品牌、KVM 网页界面、AI 网关和优先支持。仓库里没有给出价格。","notVerified":["安装脚本在你的服务器上下载和更改了什么","面板及其容器占用多少内存和磁盘空间","内置应用商店里的程序是否保持更新","基础防火墙和防护功能实际能在多大程度上保护机器","Pro 和 Ent 版的价格","AI 模型和智能体功能是否能配合你使用的服务商"]}}},{"id":"R_kgDOKTUe1A","slug":"1panel-dev__maxkb","name":"MaxKB","owner":"1Panel-dev","fullName":"1Panel-dev/MaxKB","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"default_admin_password_in_readme","severity":"attention","evidence":"README 的快速开始公开了首次登录方式——\"username: admin\"、\"password: MaxKB@123..\"——同一条命令还打开了 8080 端口，所以任何在你修改密码之前就访问到这个端口的人都能登录。","source":"ai","title":"readme 中的默认管理员密码"},{"code":"no_ready_binaries_in_releases","severity":"info","evidence":"最近三个版本（v2.10.8-lts、v2.10.7-lts、v2.10.6-lts）都标注\"no platform binaries\"；唯一现成可用的是 Docker 镜像 1panel/maxkb。","source":"ai","title":"发布版中没有现成的安装包"},{"code":"sends_your_documents_to_the_model_you_choose","severity":"info","evidence":"README：\"Model-Agnostic: Supports various large models, including private models (such as DeepSeek, Llama, Qwen, etc.) and public models (like OpenAI, Claude, Gemini, MiniMax, etc.)\"——使用公开模型时，从你文档中检索出的文本会发送到那家公司的服务器。","source":"ai","title":"会把你的文档发给你选择的模型"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["python","gpu_any"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":["A model provider such as OpenAI, Claude, Gemini or DeepSeek (not needed if you use a local model with Ollama)"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["ai_agents","local_ai","automation"],"alternativeTo":[{"slug":"chatbase","name":"Chatbase","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true},{"slug":"intercom","name":"Intercom","confirmed":true}],"declaredAlternatives":[],"stars":23138,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":26,"growthDays":6,"history":[{"date":"2026-11-06","stars":23117},{"date":"2026-11-07","stars":23116},{"date":"2026-11-08","stars":23121},{"date":"2026-11-09","stars":23129},{"date":"2026-11-10","stars":23138}],"firstPublishedAt":"2026-11-05T22:43:48.221Z","commitSha":"0c7dd350a17fe3e885f30fbfb274badef08ac37c","verifiedAt":"2026-11-05T22:26:30.141Z","lastCommitAt":"2026-11-10T09:58:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs your own AI assistant over your documents, with workflows and tool use, from a single Docker container.","summary":"MaxKB is a web platform you run on your own machine or server. You upload documents or let it crawl pages, and it builds a question-and-answer assistant that reads them before answering. It also has a visual workflow editor and can call outside tools through MCP. It works with online models (OpenAI, Claude, Gemini, DeepSeek) and with models running on your own computer through Ollama, and the chat can be embedded in another website. Made for company support desks and internal knowledge bases. You start it with one Docker command and open it in the browser.","easeRationale":"The same route on all three systems, so the same grade: install Docker, run the single `docker run ... -p 8080:8080 1panel/maxkb` line from the README and open port 8080 in the browser. There is no installer to double-click, and the checks point to 900 MB to 4.3 GB of disk because the Python side pulls in torch and sentence-transformers.","costNote":"Free under the GPL-3.0 license. You only pay if you connect a paid model, such as OpenAI or Claude.","notVerified":["How much disk and memory it really takes after the container downloads its embedding model","Whether the Docker command alone is enough or extra settings are needed for PostgreSQL and pgvector in real use","How well answers hold up on your own documents","Whether the version offered on the maxkb.cn site has features that are not in this open-source code"]},"zh-Hans":{"summaryShort":"在你自己的文档上运行专属 AI 助手，支持工作流和工具调用，只需一个 Docker 容器即可启动。","summary":"MaxKB 是一个可以在你自己的电脑或服务器上运行的网页平台。你上传文档，或者让它抓取网页，它就会构建一个问答助手，先阅读这些内容再回答问题。它还有可视化工作流编辑器，并能通过 MCP 调用外部工具。它支持在线模型（OpenAI、Claude、Gemini、DeepSeek），也支持通过 Ollama 在你自己电脑上运行的模型，聊天窗口还可以嵌入到其他网站中。它是为企业客服台和内部知识库设计的。只需一条 Docker 命令即可启动，然后在浏览器中打开。","easeRationale":"三个系统走的是同一条路，所以评分相同：安装 Docker，运行 README 里那一行 `docker run ... -p 8080:8080 1panel/maxkb`，然后在浏览器中打开 8080 端口。没有可以双击的安装程序，检测结果显示磁盘占用在 900 MB 到 4.3 GB 之间，因为 Python 那一侧引入了 torch 和 sentence-transformers。","costNote":"基于 GPL-3.0 许可证免费使用。只有在你接入付费模型（例如 OpenAI 或 Claude）时才需要付费。","notVerified":["容器下载完嵌入模型之后，实际占用的磁盘和内存有多少","单靠这条 Docker 命令是否够用，还是在实际使用中还需要为 PostgreSQL 和 pgvector 做额外设置","在你自己的文档上，回答的质量能保持到什么程度","maxkb.cn 网站提供的版本是否有这份开源代码里没有的功能"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxOTk1NzAwNzE=","slug":"2dust__v2rayn","name":"v2rayN","owner":"2dust","fullName":"2dust/v2rayN","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"proxy_tool_restricted_in_some_countries","severity":"attention","evidence":"项目自身的标签是 \"proxy, shadowsocks, socks5, trojan, v2fly, v2ray, vless, vmess, xray, xtls\"，README 也强调签名有助于 \"prevent mirror, ISP, or CDN hijacking\"。这些协议本就是为了让流量绕过网络过滤而设计的，在一些国家，运行或传播这类工具会受到限制甚至被禁止，具体要看你所在地区的法律规定。","source":"ai","title":"代理工具在部分国家受到限制"},{"code":"needs_a_proxy_server_you_provide","severity":"info","evidence":"README 只把它描述为 \"A GUI client for Windows, Linux and macOS, support Xray and sing-box and others\"，并让你去 wiki 查看 \"usage guides and configuration details\"。它不附带任何服务器，所以在你自己填入地址或订阅链接之前，应用什么也做不了。","source":"ai","title":"需要你自己提供代理服务器"},{"code":"traffic_passes_through_whoever_owns_the_server","severity":"attention","evidence":"这个应用的全部作用就是把你的连接通过你或订阅链接选定的远程服务器转发出去。运营那台服务器的人能看到这条连接，仓库对大家使用的服务商也没有任何说明。","source":"ai","title":"流量会经过服务器所有者"},{"code":"release_files_are_gpg_signed","severity":"info","evidence":"README 的 \"GPG Verification\" 部分写道：\"Release files are signed with GPG to verify authenticity and integrity\"，并在 README 中公布了指纹。核对签名是你需要自己动手完成的步骤。","source":"ai","title":"发布文件带有 gpg 签名"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":161,"maxMB":483,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system"],"alternativeTo":[{"slug":"proxifier","name":"Proxifier","confirmed":true},{"slug":"surge","name":"Surge","confirmed":true}],"declaredAlternatives":[],"stars":120344,"downloads":70139428,"downloadsWeek":521569,"downloadsDays":7,"growth":473,"growthDays":7,"history":[{"date":"2026-10-26","stars":119169},{"date":"2026-10-27","stars":119261},{"date":"2026-10-28","stars":119370},{"date":"2026-10-29","stars":119460},{"date":"2026-10-30","stars":119568},{"date":"2026-10-31","stars":119653},{"date":"2026-11-01","stars":119738},{"date":"2026-11-02","stars":119807},{"date":"2026-11-03","stars":119871},{"date":"2026-11-04","stars":119946},{"date":"2026-11-05","stars":120013},{"date":"2026-11-06","stars":120077},{"date":"2026-11-07","stars":120151},{"date":"2026-11-08","stars":120201},{"date":"2026-11-09","stars":120281},{"date":"2026-11-10","stars":120344}],"firstPublishedAt":"2026-10-27T01:10:12.578Z","commitSha":"57f17d5939e88c8e643ab49b90869d637765e75a","verifiedAt":"2026-10-31T01:38:31.614Z","lastCommitAt":"2026-11-10T07:03:01Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop app that sends your internet traffic through proxy servers you add, on Windows, Linux or macOS.","summary":"v2rayN is a window-based client for proxy protocols such as VMess, VLESS, Trojan and Shadowsocks. You install it, paste the server details or a subscription link your provider gave you, pick a server and the app routes your traffic through it. It does not include any servers: you rent them or run your own. Ready downloads exist for Windows, Linux (.deb and .rpm) and macOS (.dmg), each between 70 and 161 MB because the connection programs (Xray, sing-box) travel inside. Configuration is explained in the project wiki, not in the README.","easeRationale":"On Windows you unzip the ready v2rayN-windows-64.zip and open the program, and the one extra step is pasting the server details or the subscription link from your provider before anything works. macOS has a .dmg and Linux has .deb and .rpm packages, so the install is just as direct there; on every system the server setup is explained in the wiki and not in the README.","costNote":"The app costs nothing and the code is GPL-3.0. The proxy servers it connects to are a separate expense: you rent them from a provider or host your own.","notVerified":["Whether it really connects on each of Windows, macOS and Linux with your own server","Which versions of the Xray and sing-box connection programs are inside the download, and whether the app updates them by itself from the internet","What the owner of the proxy server you add does with your traffic or logs","Real disk use after first run, estimated at 161 to 483 MB","Whether the file you downloaded matches the GPG signature, unless you check it"]},"zh-Hans":{"summaryShort":"一款桌面应用，可将你的网络流量通过你自行添加的代理服务器发送，支持 Windows、Linux 或 macOS。","summary":"v2rayN 是一个支持 VMess、VLESS、Trojan 和 Shadowsocks 等代理协议的窗口化客户端。你安装它后，粘贴服务商提供的服务器信息或订阅链接，选择一个服务器，应用就会把你的流量通过它转发出去。它不包含任何服务器：你需要自己租用或自行搭建。Windows、Linux（.deb 和 .rpm）和 macOS（.dmg）都有现成的下载包，每个文件在 70 到 161 MB 之间，因为连接程序（Xray、sing-box）打包在里面。配置方法写在项目 wiki 里，不在 README 中。","easeRationale":"在 Windows 上，你解压现成的 v2rayN-windows-64.zip 并打开程序即可，唯一额外的步骤是在一切开始工作之前粘贴服务商给的服务器信息或订阅链接。macOS 提供 .dmg，Linux 提供 .deb 和 .rpm 包，安装同样直接；在所有系统上，服务器设置都写在 wiki 里，而不是 README 中。","costNote":"这个应用本身不花钱，代码采用 GPL-3.0 许可。它连接的代理服务器是另一笔开销：你可以向服务商租用，也可以自己搭建。","notVerified":["它是否真的能在 Windows、macOS 和 Linux 上用你自己的服务器连接成功","下载包里到底是哪个版本的 Xray 和 sing-box 连接程序，以及应用是否会自己从网上更新它们","你添加的代理服务器的所有者会如何处理你的流量或日志","首次运行后的实际磁盘占用，估计在 161 到 483 MB 之间","除非你自己核对，否则无法确认下载的文件是否与 GPG 签名相符"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxODY3MzMwOTU=","slug":"2dust__v2rayng","name":"v2rayNG","owner":"2dust","fullName":"2dust/v2rayNG","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"legal_status_varies_by_country","severity":"attention","evidence":"项目标签是 \"proxy, shadowsocks, socks5, trojan, vless, vmess, vpn, xray, xtls\"——这些协议本就是为了让流量绕过网络封锁而设计的。这款应用本身在大多数地方是合法的，但在一些国家使用这类工具会受到限制甚至可能被追责。","source":"ai","title":"法律地位因国家而异"},{"code":"traffic_goes_through_third_party_server","severity":"attention","evidence":"README：\"A V2Ray client for Android, support Xray core and v2fly core\"。应用不附带任何服务器，所以所有经过代理的流量都会通过你导入的服务器或订阅，运营方可以看到你的连接。","source":"ai","title":"流量经过第三方服务器"},{"code":"downloads_rule_files_at_run_time","severity":"info","evidence":"README（Geoip 和 Geosite 部分）：下载功能 \"will get enhanced version in this repo\"（github.com/Loyalsoldier/v2ray-rules-dat），\"it needs a working proxy\"；也 \"possible to use a third-party dat file in the same folder\"。所以路由规则可以在安装之后从应用外部获取。","source":"ai","title":"运行时会下载规则文件"},{"code":"no_binaries_detected_in_releases","severity":"info","evidence":"最近三个版本（2.3.14、2.3.13、2.3.12）在 Radar 的检查中显示 \"no platform binaries\"，因为这些文件是 Android APK，而不是桌面安装程序。README 补充说发布文件带有 GPG 签名并给出了密钥指纹，所以下载内容是可以核实的。","source":"ai","title":"发布版本中未检测到可执行文件"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system"],"alternativeTo":[{"slug":"surge","name":"Surge","confirmed":true}],"declaredAlternatives":[],"stars":65672,"downloads":47777500,"downloadsWeek":409032,"downloadsDays":7,"growth":425,"growthDays":7,"history":[{"date":"2026-10-26","stars":64742},{"date":"2026-10-27","stars":64819},{"date":"2026-10-28","stars":64882},{"date":"2026-10-29","stars":64940},{"date":"2026-10-30","stars":65015},{"date":"2026-10-31","stars":65078},{"date":"2026-11-01","stars":65137},{"date":"2026-11-02","stars":65203},{"date":"2026-11-03","stars":65247},{"date":"2026-11-04","stars":65314},{"date":"2026-11-05","stars":65375},{"date":"2026-11-06","stars":65436},{"date":"2026-11-07","stars":65503},{"date":"2026-11-08","stars":65554},{"date":"2026-11-09","stars":65602},{"date":"2026-11-10","stars":65672}],"firstPublishedAt":"2026-10-27T01:10:12.594Z","commitSha":"aa4b05503a504a3f2e43818e6ee8ebf1caaa3df0","verifiedAt":"2026-10-31T01:41:43.052Z","lastCommitAt":"2026-11-10T11:23:01Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android app that sends your phone's internet traffic through a proxy server you supply, to reach blocked sites.","summary":"v2rayNG is an Android app that routes your phone's internet traffic through a proxy server, which is mostly used to reach websites and services that are blocked on a network or in a country. It speaks the VMess, VLESS, Shadowsocks and Trojan protocols using the Xray or v2fly engine. The app brings no servers of its own: you paste in a server or subscription link from a provider, or from a server you rent yourself. It does not run on a computer; the README points desktop users to v2rayN instead.","easeRationale":"It is an Android-only app, so none of the desktop platforms get a grade. On Android you install a ready APK from the GitHub releases page, but nothing works until you import a server or a subscription link from somewhere else.","costNote":"The app is free (GPL-3.0). It only works with a proxy server, which usually means paying a provider or renting your own server.","notVerified":["Which APK to pick for a given phone (arm64-v8a, armeabi-v7a, universal or x86) and how the regular and F-Droid builds differ.","Whether the GPG signature on the releases matches the fingerprint printed in the README.","How trustworthy any given server provider is; that depends on the provider, not on this app.","Battery and speed impact of keeping the VPN mode on.","Whether the engine bundled in the app is up to date, since the README says the core inside the aar is \"(probably) outdated\" for self-built versions."]},"zh-Hans":{"summaryShort":"一款 Android 应用，通过你自己提供的代理服务器转发手机的网络流量，以访问被屏蔽的网站。","summary":"v2rayNG 是一款 Android 应用，把你手机的网络流量通过一个代理服务器转发出去，主要用于访问在某个网络或某个国家被屏蔽的网站和服务。它使用 Xray 或 v2fly 引擎，支持 VMess、VLESS、Shadowsocks 和 Trojan 协议。应用本身不带任何服务器：你需要粘贴服务商提供的服务器或订阅链接，或者自己租用的服务器。它不能在电脑上运行；README 把桌面用户指向了 v2rayN。","easeRationale":"这是一款仅支持 Android 的应用，所以没有任何桌面系统获得评分。在 Android 上，你从 GitHub 发布页安装一个现成的 APK，但在你从别处导入服务器或订阅链接之前，什么都用不了。","costNote":"这个应用是免费的（GPL-3.0）。它只能配合代理服务器使用，这通常意味着要向服务商付费，或者自己租一台服务器。","notVerified":["对某款手机该选哪个 APK（arm64-v8a、armeabi-v7a、universal 还是 x86），以及常规版和 F-Droid 版有什么区别","发布版本上的 GPG 签名是否与 README 中公布的指纹相符","某个服务器服务商有多靠得住；这取决于服务商本身，与这款应用无关","一直开着 VPN 模式对电池和速度的影响","应用自带的引擎是否是最新的，因为 README 说自行编译版本里 aar 内的核心 \"(probably) outdated\"（可能已过时）"]}}},{"id":"R_kgDOMBVqQg","slug":"2noise__chattts","name":"ChatTTS","owner":"2noise","fullName":"2noise/ChatTTS","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"C"},"risk":"attention","signals":[{"code":"model_license_forbids_commercial_use","severity":"attention","evidence":"README.md: \"The released model is for academic purposes only\" 以及 \"The model is published under `CC BY-NC 4.0` license. It is intended for educational and research use, and should not be used for any commercial or illegal purposes.\" 代码本身是 AGPLv3+ 许可，与你下载的模型许可不同。","source":"ai","title":"模型许可禁止商业用途"},{"code":"released_model_deliberately_degraded","severity":"info","evidence":"README.md 声明: \"we added a small amount of high-frequency noise during the training of the 40,000-hour model, and compressed the audio quality as much as possible using MP3 format, to prevent malicious actors from potentially using it for criminal purposes.\" README 还说公开模型是一个未经微调的 4 万小时版本，而主模型是用超过 10 万小时数据训练的。","source":"ai","title":"发布的模型被故意降质"},{"code":"output_quality_varies","severity":"info","evidence":"README 的 FAQ: \"Model stability is not good enough, with issues such as multi speakers or poor audio quality... One can try multiple samples to find a suitable result.\" 只有 `[laugh]`、`[uv_break]` 和 `[lbreak]` 能被直接控制，示例中英文被标注为实验性，而且只支持英文和中文。","source":"ai","title":"输出质量不稳定"},{"code":"voice_imitation_possible","severity":"info","evidence":"路线图把 \"Open-source DVAE encoder and zero shot inferring code\" 标记为已完成，代码示例中会采样一个可以保存和复用的说话人向量，所以该项目能够生成模仿某个样本的声音。作者写道他们 \"internally trained a detection model and plan to open-source it in the future\"。","source":"ai","title":"可以模仿声音"},{"code":"development_slowed","severity":"info","evidence":"受检快照中的最后一次提交和最后一个发布版本（v0.2.5）都是 2026-04-10，距 2026-11-04 的检查日已有 208 天，而路线图中的多情绪控制和 ChatTTS.cpp 等条目仍未完成，还有 63 个未关闭的 issue。","source":"ai","title":"开发已经放缓"}],"platforms":["windows","mac","linux","web"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true}],"declaredAlternatives":[],"stars":39998,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":23,"growthDays":7,"history":[{"date":"2026-11-05","stars":39987},{"date":"2026-11-06","stars":39987},{"date":"2026-11-07","stars":39994},{"date":"2026-11-08","stars":39997},{"date":"2026-11-09","stars":39997},{"date":"2026-11-10","stars":39998}],"firstPublishedAt":"2026-11-04T22:32:54.782Z","commitSha":"77b89ee281cd479f5b1a787ada330dc975ca1f2a","verifiedAt":"2026-11-04T22:20:22.057Z","lastCommitAt":"2026-04-10T16:31:18Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns typed text into conversational speech on your own computer, with laughs and pauses, in English or Chinese.","summary":"A text-to-speech model made for dialogue: you type text and it reads it aloud with pauses, laughter and several different voices, in English or Chinese. It comes with a simple web page you start yourself and a command line example, and it is also usable as a Python library. The project calls itself the algorithm plus simple examples and sends end users to community projects built on it. You need Python, a graphics card with at least 4 GB of memory for a 30-second clip, and the published model is for study and research, not commercial work.","easeRationale":"The easiest way in is the web: the repository ships a Google Colab notebook you open in the browser and run cell by cell, so no install, which is a C. On Windows, Mac and Linux it is a D: you clone the repository, build a Python environment, install the requirements and run `python examples/web/webui.py`, and the FAQ asks for a graphics card with at least 4 GB of memory; some extra packages are Linux-only and Macs have no documented graphics support.","costNote":"No price. The code is AGPLv3+, but the model files are CC BY-NC 4.0: the README says they are \"for academic purposes only\" and \"should not be used for any commercial or illegal purposes\".","notVerified":["How the voices really sound to you, especially in English","Whether it runs on a Mac or on a computer without a dedicated graphics card, and how slowly","Real disk space taken by the downloaded model files","Whether the Google Colab notebook still works with current versions of the libraries","How the non-commercial model licence applies to what you want to do with the audio"]},"zh-Hans":{"summaryShort":"在你自己的电脑上把文字变成带笑声和停顿的对话式语音，支持英文或中文。","summary":"一个为对话场景打造的文字转语音模型：输入文字,它就用带停顿、笑声和多种不同声音朗读出来,支持英文或中文。它附带一个你自己启动的简单网页,以及一个命令行示例,也可以当作 Python 库使用。项目把自己定位为“算法加简单示例”,并把终端用户导向基于它搭建的社区项目。你需要 Python,一张至少 4GB 显存的显卡来生成 30 秒片段,而公开发布的模型仅用于学习和研究,不可商用。","easeRationale":"最省事的入口是网页：仓库自带一个 Google Colab 笔记本,在浏览器里逐个单元格运行即可,不需要安装,评级为 C。在 Windows、Mac 和 Linux 上是 D：需要克隆仓库、搭建 Python 环境、安装依赖并运行 `python examples/web/webui.py`，FAQ 要求显卡至少有 4GB 显存；部分额外包仅支持 Linux，Mac 没有文档记录的图形支持。","costNote":"没有价格。代码是 AGPLv3+ 许可,但模型文件是 CC BY-NC 4.0：README 说它们“仅用于学术目的”,“不应用于任何商业或非法用途”。","notVerified":["这些声音对你来说听起来究竟如何，尤其是英文","能否在 Mac 或没有独立显卡的电脑上运行，以及运行有多慢","下载的模型文件实际占用的磁盘空间","Google Colab 笔记本是否仍能配合当前版本的库运行","模型的非商业许可如何适用于你想对音频做的事"]}}},{"id":"R_kgDOMce-kQ","slug":"abus-aikorea__voice-pro","name":"Voice-Pro","owner":"abus-aikorea","fullName":"abus-aikorea/voice-pro","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"ships_celebrity_voice_samples","severity":"attention","evidence":"README 中有用于克隆的参考声音表格，取自真实公众人物，文件如 `celebrities30sREADME/English/Donald Trump.jpg`、`Elon Musk.jpg`、`Joe Rogan.jpg`、`celebrities30sREADME/Korean/IU.jpg` 和 `celebrities30sREADME/Japanese/Ayase Haruka.jpg`，还有“名人声音播客”这一功能描述。让真人说出他们从未说过的话，可能侵犯声音和肖像权，在不少国家还涉及冒充他人的法律问题。","source":"ai","title":"附带名人声音样本"},{"code":"downloads_videos_from_youtube","severity":"attention","evidence":"Dubbing Studio 标签页的第一个功能是使用 yt-dlp 进行“YouTube 视频下载与音频提取”（pyproject.toml 中为 `\"yt-dlp>=2025.9.26\"`）。从 YouTube 下载违反其服务条款，在大多数国家，为他人视频重新配音也涉及版权问题。","source":"ai","title":"从 youtube 下载视频"},{"code":"reads_your_browser_cookies","severity":"attention","evidence":"pyproject.toml 将 `browser_cookie3` 列为核心依赖。该库会读取电脑上已安装浏览器的 cookie 数据库，下载工具正是借此访问需要登录才能观看的视频；README 中从未提及这一点。","source":"ai","title":"读取你的浏览器 cookie"},{"code":"author_paused_development","severity":"info","evidence":"README 中两次提到：“由于 [WeConnect](https://www.wctokyoseoul.com) 的开发工作，Voice-Pro 目前暂时无法开发和更新。”快照中的最后一次提交和最后一个发布版本（v4.0.0）都是 2026-07-13，距检查时间 114 天前。","source":"ai","title":"作者暂停了开发"},{"code":"readme_contradicts_itself_on_systems","severity":"info","evidence":"“系统要求”写的是“操作系统：Windows 10/11（64位）、Linux、Mac（Apple Silicon）”，而顶部的说明却写着“在搭配 NVIDIA GPU 的 Windows 上运行良好。Mac 和 Linux 上的运行情况尚未验证。”在 pyproject.toml 中，GPU 相关包也只为 Linux 和 Windows 选取。","source":"ai","title":"readme 在系统支持上自相矛盾"},{"code":"readme_carries_marketing_and_old_paid_plans","severity":"info","evidence":"README 自称为“最好的 AI 语音识别、翻译和多语言配音方案”，以及“**ElevenLabs** 的有力替代品”，还保留着一份“截至 2025 年 4 月 15 日”的八个付费服务价格表，2.0 版本的更新历史里仍然描述着 60 秒免费试用和用于“无限使用”的 Shopify 订阅，而这些已经不再适用。","source":"ai","title":"readme 带有营销内容和过时的付费方案"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":1250,"maxMB":5250,"basis":"heuristic"},"apiKeys":["Microsoft Azure Speech (optional, better voices)","Microsoft Azure Translator (optional, instead of the free Google endpoint)"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["audio","video","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true},{"slug":"rask-ai","name":"Rask AI","confirmed":true},{"slug":"4k-video-downloader","name":"4K Video Downloader","confirmed":true}],"declaredAlternatives":[{"target":"invideo-ai","stance":"claims","evidence":"keywords: AI voice conversion, YouTube translation, subtitle generation, speech-to-text, text-to-speech, voice cloning, multilingual translation, ElevenLabs Alternative"},{"target":"elevenlabs","stance":"claims","evidence":"keywords: AI voice conversion, YouTube translation, subtitle generation, speech-to-text, text-to-speech, voice cloning, multilingual translation, ElevenLabs Alternative"},{"target":"benzinga-pro","stance":"claims","evidence":"A robust alternative to ElevenLabs, Voice-Pro empowers podcasters, developers, and creators with advanced voice solutions"},{"target":"dbeaver-pro","stance":"claims","evidence":"A robust alternative to ElevenLabs, Voice-Pro empowers podcasters, developers, and creators with advanced voice solutions"},{"target":"moz-pro","stance":"claims","evidence":"A robust alternative to ElevenLabs, Voice-Pro empowers podcasters, developers, and creators with advanced voice solutions"},{"target":"deepl-pro","stance":"claims","evidence":"A robust alternative to ElevenLabs, Voice-Pro empowers podcasters, developers, and creators with advanced voice solutions"}],"stars":13421,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":20,"growthDays":6,"history":[{"date":"2026-11-05","stars":13408},{"date":"2026-11-06","stars":13408},{"date":"2026-11-07","stars":13411},{"date":"2026-11-08","stars":13415},{"date":"2026-11-09","stars":13419},{"date":"2026-11-10","stars":13421}],"firstPublishedAt":"2026-11-04T22:32:54.791Z","commitSha":"7231384f797091dc0d6d0470360fc6fcc2095d0c","verifiedAt":"2026-11-04T22:21:31.919Z","lastCommitAt":"2026-07-13T01:28:05Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Takes a video, writes subtitles, translates them and dubs it with a synthetic or cloned voice, all on your own PC.","summary":"A dubbing workbench that opens in your browser but runs on your machine. It downloads a YouTube video or takes your own file, separates the voice from the music, transcribes it with Whisper, translates the subtitles into one of 100+ languages and reads the result aloud with Edge-TTS, kokoro, F5-TTS, E2-TTS or CosyVoice, including cloning a voice from a short sample. You start it with a script that installs Python and the models by itself (about 10 GB), and you need an NVIDIA graphics card with 4 GB or more.","easeRationale":"Windows is the only system the author says he has tested, and it is a D: you clone or unzip the source, run `start.bat`, and the first run downloads Python, the dependencies and about 10 GB of models, with an NVIDIA card of 4 GB or more and 20 GB of free disk space. Mac and Linux have the same `configure.sh` and `start.sh` scripts, so the same grade, but the README says their operation has not been verified.","costNote":"README: \"We have made all Voice-Pro code open source and completely free.\" Older versions had a 60-second limit with a paid subscription; that is gone. Optional Microsoft Azure translation and speech use your own paid Azure account.","notVerified":["Whether it really runs on Mac or Linux","Real disk use after the roughly 10 GB of models are downloaded","How good the subtitles, translations and cloned voices sound in your languages","Whether the free Google translation endpoint keeps working without being blocked","Whether `browser_cookie3` is actually used to read your browser cookies and when","Whether the celebrity reference clips are still included in the repository you download"]},"zh-Hans":{"summaryShort":"拿一段视频，写出字幕、翻译并用合成或克隆的声音配音，全部在你自己的电脑上完成。","summary":"一个在浏览器中打开、但在你自己机器上运行的配音工作台。它可以下载一段 YouTube 视频或使用你自己的文件，把人声和音乐分离，用 Whisper 转写，把字幕翻译成 100 多种语言中的一种，再用 Edge-TTS、kokoro、F5-TTS、E2-TTS 或 CosyVoice 朗读出来，还能根据一段简短样本克隆声音。你用一个脚本启动它，脚本会自行安装 Python 和模型（约 10GB），并且需要一张 4GB 以上显存的 NVIDIA 显卡。","easeRationale":"Windows 是作者说自己测试过的唯一系统，评级为 D：你克隆或解压源码，运行 `start.bat`，首次运行会下载 Python、依赖项和约 10GB 的模型，需要 4GB 以上显存的 NVIDIA 显卡和 20GB 的空闲磁盘空间。Mac 和 Linux 有相同的 `configure.sh` 和 `start.sh` 脚本，评级相同，但 README 说在这些系统上的运行情况尚未验证。","costNote":"README：“我们已将 Voice-Pro 的全部代码开源并完全免费。”旧版本曾有 60 秒时长限制并配合付费订阅，这一限制已取消。可选的微软 Azure 翻译和语音功能使用你自己的付费 Azure 账户。","notVerified":["它是否真的能在 Mac 或 Linux 上运行","下载约 10GB 模型后实际占用的磁盘空间","字幕、翻译和克隆声音在你所用语言中听起来究竟如何","免费的 Google 翻译接口是否一直能用而不被屏蔽","`browser_cookie3` 是否真的被用来读取浏览器 cookie，以及何时读取","你下载的仓库中是否仍包含名人参考音频片段"]}}},{"id":"R_kgDOOg6t_g","slug":"ace-step__ace-step","name":"ACE-Step","owner":"ace-step","fullName":"ace-step/ACE-Step","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"D","mac":"C","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞 (OSV)：datasets: GHSA-379c-qx7v-6h59, PYSEC-2026-3716; pytorch_lightning: GHSA-75m9-98v2-hjpm, PYSEC-2026-3043, PYSEC-2026-3967; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-489j-g2vx-39wf; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804","source":"check","title":"已知漏洞"},{"code":"copyright_and_voice_cloning","severity":"attention","evidence":"项目自己的免责声明提到\"因风格相似而意外侵犯版权、不当混用文化元素，以及被滥用于生成有害内容等潜在风险\"，摘要还把\"声音克隆、歌词编辑、混音\"列为它支持的功能之一。复制真实艺人的声音或风格可能引来法律麻烦。","source":"ai","title":"版权与声音克隆"},{"code":"fake_websites_warning","severity":"info","evidence":"README 中有一条\"重要提示\"，说 GitHub Pages 网站是唯一官方网站，列出了经过掩码处理的虚假域名（\"ac**p.com, a**p.org, a***c.org\"），并提醒不要\"在这些网站上付款\"。只从 GitHub 页面下载。","source":"ai","title":"虚假网站提醒"},{"code":"development_moved_on","severity":"info","evidence":"这里最后一次提交是在 2026-02-15，路线图上的 \"Release ControlNet training code\" 和 \"Release Singing2Accompaniment ControlNet\" 仍未勾选，而新闻区指向另一处在 2026-01-28 发布的更新版本 ACE-Step v1.5。这个仓库是旧版本。","source":"ai","title":"开发重心已转移到别处"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","docker","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":5045,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":19,"growthDays":6,"history":[{"date":"2026-11-04","stars":5026},{"date":"2026-11-05","stars":5026},{"date":"2026-11-06","stars":5032},{"date":"2026-11-07","stars":5034},{"date":"2026-11-08","stars":5037},{"date":"2026-11-09","stars":5044},{"date":"2026-11-10","stars":5045}],"firstPublishedAt":"2026-11-04T01:36:50.185Z","commitSha":"1bee4c9f5b43e30995f8d4d33b3919197ce1bd68","verifiedAt":"2026-11-04T01:24:52.881Z","lastCommitAt":"2026-02-15T04:57:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes complete songs, with singing and lyrics, from a text description, on your own computer.","summary":"ACE-Step turns a few style tags and your lyrics into a finished piece of music with vocals. Besides writing a new song it can redo a part of one, change the lyrics while keeping the melody, make the track longer or remix audio you upload. It opens as a page in your browser after you install it with Python commands, and it works best with an NVIDIA graphics card with 8 GB of video memory; a MacBook M2 Max also works, more slowly. The model files download by themselves on the first run.","easeRationale":"The quickest look is the maintainers' hosted \"Space Demo\" page linked at the top of the README, which needs nothing installed. To run it yourself the README gives only command-line steps: on Mac it is Python, a virtual environment and \"pip install -e .\" (a C), while Windows and Linux also mean matching PyTorch to your CUDA version for an NVIDIA card, and sometimes \"pip install triton-windows\", which puts them at D.","costNote":"Apache-2.0 licence, no account or payment.","notVerified":["How much disk space the model files really take after the automatic download","Whether the hosted demo page stays available and how long you wait in its queue","How slow it is on a computer with no graphics card, since the README only measured NVIDIA cards and a MacBook M2 Max","Whether the songs sound as good as the examples on the project page","Whether the install still works today, given the code has not changed since February 2026"]},"zh-Hans":{"summaryShort":"根据一段文字描述，在你自己的电脑上生成带演唱和歌词的完整歌曲。","summary":"ACE-Step 把几个风格标签和你写的歌词，变成一段带人声的完整音乐作品。除了写新歌，它还能重做歌曲的一部分、在保留旋律的同时更改歌词、把曲目延长，或者重新混音你上传的音频。安装时用 Python 命令，安装后在浏览器里打开一个页面；搭配显存 8 GB 的 NVIDIA 显卡效果最好，MacBook M2 Max 也能用，只是速度更慢。模型文件会在第一次运行时自动下载。","easeRationale":"最快的体验方式是维护者托管的 \"Space Demo\" 页面，链接在 README 顶部，不需要安装任何东西。要自己运行的话，README 只给出命令行步骤：在 Mac 上是 Python、一个虚拟环境和 \"pip install -e .\"（评级 C），而 Windows 和 Linux 还需要让 PyTorch 版本匹配你的 CUDA 版本以适配 NVIDIA 显卡，有时还要 \"pip install triton-windows\"，这让它们被评为 D。","costNote":"Apache-2.0 许可证，不需要账号也不需要付款。","notVerified":["模型文件自动下载后实际占用多少磁盘空间","托管的演示页面是否一直可用，以及排队要等多久","在没有显卡的电脑上有多慢，因为 README 只测过 NVIDIA 显卡和 MacBook M2 Max","生成的歌曲是否像项目页面上的示例那样好听","既然代码自 2026 年 2 月以来没有变化，安装流程今天是否仍然有效"]}}},{"id":"R_kgDOPpe8ZA","slug":"ace-step__ace-step-1.5","name":"ACE-Step 1.5","owner":"ace-step","fullName":"ace-step/ACE-Step-1.5","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"B","mac":"B","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接通过管道传给 shell 执行：curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"curl 管道直接执行到 shell"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞 (OSV)：gradio: GHSA-39mp-8hj3-5c49, GHSA-6655-8ph2-63j3, GHSA-7hp7-4p35-3cx2; gradio: GHSA-39mp-8hj3-5c49, GHSA-6655-8ph2-63j3, GHSA-7hp7-4p35-3cx2; torch: GHSA-3749-ghw9-m3mg, GHSA-47fc-vmwq-366v, GHSA-53q9-r3pm-6pq6; gradio: GHSA-39mp-8hj3-5c49, GHSA-6655-8ph2-63j3, GHSA-7hp7-4p35-3cx2; torch: GHSA-887c-mr87-cxwp, GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47; torch: GHSA-3749-ghw9-m3mg, GHSA-47fc-vmwq-366v, GHSA-53q9-r3pm-6pq6; torch: GHSA-rrmf-rvhw-rf47; torch: GHSA-rrmf-rvhw-rf47; gradio: GHSA-39mp-8hj3-5c49, GHSA-6655-8ph2-63j3, GHSA-7hp7-4p35-3cx2; mlx: GHSA-j842-xgm4-wf88, GHSA-w6vg-jg77-2qg6, PYSEC-2025-138; flash-attn: GHSA-7g5w-pq96-8c5w, PYSEC-2026-2477; flash-attn: GHSA-7g5w-pq96-8c5w, PYSEC-2026-2477","source":"check","title":"已知漏洞"},{"code":"copies_existing_music","severity":"attention","evidence":"功能表中列出了“Cover Generation — Create covers from existing audio”，语言模型对比表里有一列 \"Copy Melody\"，最高评级为 \"Strong\"。项目自己的免责声明也提醒“因风格相似而意外侵犯版权”。翻唱你并不拥有版权的歌曲可能触犯版权法。","source":"ai","title":"复制已有音乐"},{"code":"portable_package_hosted_offsite","severity":"info","evidence":"现成可用的安装包并不是 GitHub 发行版：README 让 Windows 用户前往 https://files.acemusic.ai/acemusic/win/ACE-Step-1.5.7z，让 Mac 用户前往 https://files.acemusic.ai/acemusic/mac/ACE-Step-1.5.zip，这是项目自己的服务器，自动检查无法看到里面的内容。","source":"ai","title":"便携包托管在仓库之外"},{"code":"fake_websites_warning","severity":"info","evidence":"README 中有一条“重要提示”，说 GitHub Pages 网站是唯一官方网站，列出了经过掩码处理的虚假域名（“ac**p.com, a**p.org, a***c.org”），并提醒不要“在这些网站上付款”。","source":"ai","title":"虚假网站提醒"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","node","docker","gpu_nvidia"],"disk":{"minMB":1050,"maxMB":4950,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[{"target":"browser-use-cloud","stance":"claims","evidence":"ACE-Step](https://github.com/ace-step/awesome-ace-step) — a curated list of community projects, alternative UIs, ComfyUI nodes, cloud deployments, training tools, and more built around ACE-Step"},{"target":"n8n-cloud","stance":"claims","evidence":"ACE-Step](https://github.com/ace-step/awesome-ace-step) — a curated list of community projects, alternative UIs, ComfyUI nodes, cloud deployments, training tools, and more built around ACE-Step"}],"stars":14603,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":224,"growthDays":6,"history":[{"date":"2026-11-04","stars":14379},{"date":"2026-11-05","stars":14412},{"date":"2026-11-06","stars":14455},{"date":"2026-11-07","stars":14491},{"date":"2026-11-08","stars":14526},{"date":"2026-11-09","stars":14572},{"date":"2026-11-10","stars":14603}],"firstPublishedAt":"2026-11-04T01:36:50.223Z","commitSha":"ca1e85fe9430179831e6bc6be790c332190a3866","verifiedAt":"2026-11-04T01:25:43.380Z","lastCommitAt":"2026-08-29T08:25:18Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes full songs with singing from a description or your lyrics, on your own computer or on the project's free website.","summary":"ACE-Step 1.5 writes music with vocals from a short description or your own lyrics, in lengths from 10 seconds to 10 minutes and in over 50 languages. It can also make a cover of a track you upload, redo a section, split a song into separate instrument tracks, add accompaniment to a voice recording and learn your style from eight songs. The owners say the quality sits between two paid Suno versions. It runs in your browser after you install it, on an NVIDIA, AMD, Intel or Apple chip; 6 GB of video memory is enough for the small model.","easeRationale":"Nothing to install if you use the maintainers' own pages: the README links a Hugging Face \"Space Demo\" and acemusic.ai. On Windows and Mac the README links a \"portable package with pre-installed dependencies\" and ready start scripts (start_gradio_ui.bat, start_gradio_ui_macos.sh), with the models downloading on the first run, so B. On Linux you clone the repository and run uv commands in a terminal, a C.","costNote":"MIT licence. The README also offers acemusic.ai, which it describes as \"100% free, no GPU required\".","notVerified":["Whether the music really sounds as good as the paid services the README compares itself to","How much disk space the model files take, since several models download on the first run","Whether it works on your graphics card, as the README splits features by video memory from 6 GB to 24 GB and up","How slow it is on a computer with no graphics card","What the portable Windows and Mac bundles on files.acemusic.ai contain"]},"zh-Hans":{"summaryShort":"根据描述或你的歌词制作带演唱的完整歌曲，可以在你自己的电脑上，也可以在项目的免费网站上。","summary":"ACE-Step 1.5 能根据一段简短描述或你自己写的歌词创作带人声的音乐，时长从 10 秒到 10 分钟，支持 50 多种语言。它还能为你上传的曲目制作翻唱、重做某一段、把一首歌拆分成各个乐器的独立音轨、为一段人声录音添加伴奏，并从八首歌中学习你的风格。开发方表示质量介于两个付费版 Suno 之间。安装后在浏览器里运行，支持 NVIDIA、AMD、Intel 或 Apple 芯片；小模型只需要 6 GB 显存。","easeRationale":"如果使用维护者自己的页面，就什么都不用安装：README 链接了一个 Hugging Face 上的 \"Space Demo\" 和 acemusic.ai。在 Windows 和 Mac 上，README 链接了一个“预装好依赖的便携包”和现成的启动脚本（start_gradio_ui.bat、start_gradio_ui_macos.sh），模型会在首次运行时下载，所以评为 B。在 Linux 上，你需要克隆仓库并在终端里运行 uv 命令，评为 C。","costNote":"MIT 许可证。README 还提供了 acemusic.ai，它被描述为“100% 免费，不需要 GPU”。","notVerified":["音乐是否真的像 README 拿来对比的付费服务那样好听","模型文件占用多少磁盘空间，因为有几个模型会在首次运行时下载","是否能在你的显卡上运行，因为 README 按显存从 6 GB 到 24 GB 以上划分了不同功能","在没有显卡的电脑上有多慢","files.acemusic.ai 上的 Windows 和 Mac 便携包里到底是什么"]}}},{"id":"R_kgDOIjFiOQ","slug":"activepieces__activepieces","name":"Activepieces","owner":"activepieces","fullName":"activepieces/activepieces","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"example_settings_run_code_unsandboxed","severity":"attention","evidence":".env.example 自带 \"AP_EXECUTION_MODE=UNSANDBOXED\"，构建器功能里还包括 \"Code with NPM\" 和 \"ASK AI in Code Piece\"。因此流程里的代码，包括你导入的流程模板中的代码，运行时周围没有任何沙箱。","source":"ai","title":"示例配置下的代码没有沙箱隔离运行"},{"code":"part_of_the_code_is_not_open_source","severity":"attention","evidence":"README 的许可证部分写道：“Activepieces 的社区版以 MIT 许可证开源发布，企业功能则在商业许可下发布”，并指向 packages/ee/LICENSE。GitHub 本身把这个仓库的许可证报告为 NOASSERTION，所以要留意自己能用哪些部分。","source":"ai","title":"部分代码不是开源的"},{"code":"connectors_come_from_npm","severity":"attention","evidence":"README 说“所有组件都是开源的，并发布在 npmjs.com 上，60% 的组件由社区贡献”，而且组件“在贡献后会直接发布到 npmjs.com”。保存你账号密码的那个连接器，可能是维护者之外的人写的。","source":"ai","title":"连接器来自 npm"},{"code":"usage_reporting_on_by_default","severity":"info","evidence":".env.example 设置了 \"AP_TELEMETRY_ENABLED=true\"，现成的流程模板也是从该公司的云端获取的：AP_TEMPLATES_SOURCE_URL=\"https://cloud.activepieces.com/api/v1/flow-templates\"。","source":"ai","title":"使用情况报告默认开启"}],"platforms":["windows","mac","linux"],"tools":["docker","node"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["OpenAI (only for the optional tool-search feature)"],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["automation","ai_agents"],"alternativeTo":[{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"make","name":"Make","confirmed":true},{"slug":"n8n-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[{"target":"zapier","stance":"claims","evidence":"An open source replacement for Zapier"}],"stars":25466,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":95,"growthDays":6,"history":[{"date":"2026-11-04","stars":25371},{"date":"2026-11-05","stars":25384},{"date":"2026-11-06","stars":25400},{"date":"2026-11-07","stars":25417},{"date":"2026-11-08","stars":25435},{"date":"2026-11-09","stars":25449},{"date":"2026-11-10","stars":25466}],"firstPublishedAt":"2026-11-04T01:36:50.232Z","commitSha":"f81ae8e341e80c48e47b783fb741ed9d5079def9","verifiedAt":"2026-11-04T01:27:08.902Z","lastCommitAt":"2026-11-10T21:27:07Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Builds automations that join your apps and AI steps, running on your own server instead of a paid service.","summary":"Activepieces joins apps together: you build a flow in a visual editor in your browser, with steps, branches, loops and AI pieces, so something like \"when a form arrives, add a row and send a message\" runs by itself. The README calls it \"An open source replacement for Zapier\" and lists over 200 connectors, most contributed by the community, which also work as MCP tools for assistants like Claude Desktop or Cursor. You run it yourself with Docker, Postgres and Redis and open it at http://localhost:8080.","easeRationale":"All three systems take the same route and get the same grade: install Docker, then start the four containers in docker-compose.yml (the app, five workers, a Postgres database and Redis) and open http://localhost:8080. Before the first start you fill in a .env file with your own keys, and the file itself says it is \"advisable to consult the documentation and use the tools/deploy.sh to generate the passwords, keys\". There is nothing to double-click.","costNote":"Running the Community Edition yourself costs nothing, but the README says \"enterprise features are released under Commercial License\" and points to a pricing page; the company also sells a hosted cloud. Any paid service your flows call is billed separately.","notVerified":["Whether the editor at http://localhost:8080 asks for a password on a first start with the example settings","Real disk space and memory once Postgres, Redis and five workers are running with your history","Whether the connectors you need are among the 200 listed and whether they still work","What your flows would cost on the paid cloud plan instead","What the usage reporting sends when it is left on"]},"zh-Hans":{"summaryShort":"搭建自动化流程，把你的应用和 AI 步骤连接起来，运行在你自己的服务器上，而不是付费服务上。","summary":"Activepieces 把各种应用连接在一起：你在浏览器里的可视化编辑器中搭建一个流程，包含步骤、分支、循环和 AI 组件，这样像“收到表单后，添加一行并发送一条消息”这样的事情就能自动完成。README 称它为“Zapier 的开源替代品”，列出了 200 多个连接器，大多数由社区贡献，这些连接器也可以作为 MCP 工具供 Claude Desktop 或 Cursor 等助手使用。你自己用 Docker、Postgres 和 Redis 来运行它，并在 http://localhost:8080 打开。","easeRationale":"三个系统走的是同一条路径，评级也相同：安装 Docker，然后启动 docker-compose.yml 里的四个容器（应用本身、五个 worker、一个 Postgres 数据库和 Redis），再打开 http://localhost:8080。第一次启动前你要填写一个 .env 文件，填入自己的密钥，文件本身也说“建议查阅文档，并使用 tools/deploy.sh 来生成密码和密钥”。没有什么文件可以直接双击打开。","costNote":"自己运行社区版不花钱，但 README 说“企业功能在商业许可下发布”，并指向一个价格页面；该公司还出售托管云服务。你的流程调用的任何付费服务都会单独计费。","notVerified":["用示例配置第一次启动时，http://localhost:8080 的编辑器是否会要求输入密码","Postgres、Redis 和五个 worker 带着你的历史数据运行起来后，实际占用的磁盘空间和内存","你需要的连接器是否在列出的 200 个之中，以及它们是否还能用","换成付费云方案的话，你的流程大概要花多少钱","使用情况报告在开启时到底发送了什么内容"]}}},{"id":"MDEwOlJlcG9zaXRvcnk4NDQ0OTA5MQ==","slug":"adam0white__social-media-monitor","name":"Social Media Monitor","owner":"adam0white","fullName":"adam0white/Social-Media-Monitor","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"web":"D"},"risk":"attention","signals":[{"code":"archived","severity":"info","evidence":"该仓库已被归档(只读,不再维护)","source":"check","title":"已归档"},{"code":"sources_from_2018_that_no_longer_exist","severity":"attention","evidence":"README 的计数器列表包括「Followers count from any Google+ page」和「Current international and local page rank from Alexa」。Google+ 于 2019 年关闭,Alexa 的排名服务于 2022 年关闭,而最后一次提交是在 2018-06-08,所以这些部分无法工作。README 还提到部分数字来自「URL parsing」,一旦网站改版就会失效。","source":"ai","title":"2018 年的数据来源如今已不存在"},{"code":"social_apis_changed_since_the_script_was_written","severity":"attention","evidence":"这个脚本最后一次修改是在 2018 年 6 月,早于 Twitter 开始对 API 访问收费,也早于 Facebook 和 Instagram 收紧主页或个人资料查询返回的内容。README 本身也警告:「It fails intermittently because of social medias' some limits」。预计你需要自己修复代码。","source":"ai","title":"脚本编写之后社交平台的 API 发生了变化"},{"code":"runs_inside_your_google_account_with_your_keys","severity":"attention","evidence":"使用说明要求你把脚本复制到自己的 Google 账号里,把 API 密钥粘贴进去(「Populate needed API_KEYS」),并设置每天运行的触发器。Apps Script 代码以你的 Google 权限运行,密钥就存放在脚本里,所以你分享这个脚本或表格给谁,谁也会拿到这些密钥。","source":"ai","title":"在你的 Google 账号里用你的密钥运行"},{"code":"copy_offered_from_a_google_hosted_link","severity":"info","evidence":"README.md 在 script.google.com 的「/copy」链接上提供脚本,在 docs.google.com 的「/copy」链接上提供表格。这些副本存在于已归档仓库之外,可能与这里的 Main.gs 不同,也没有历史记录可供比对。","source":"ai","title":"副本来自 Google 托管的链接"}],"platforms":["web"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Facebook Graph API","Twitter (X) API","YouTube Data API","OneSignal"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"buffer","name":"Buffer","confirmed":true}],"declaredAlternatives":[],"stars":68,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":68},{"date":"2026-11-08","stars":68},{"date":"2026-11-09","stars":68},{"date":"2026-11-10","stars":68}],"firstPublishedAt":"2026-11-06T22:59:23.695Z","commitSha":"ebba2de59f70204a3858e2c91abc82ec1985fe6f","verifiedAt":"2026-11-06T22:41:28.285Z","lastCommitAt":"2018-06-08T17:03:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Google Sheets script that writes down follower and like counts from social networks once a day.","summary":"This is a single Google Apps Script you copy into your own Google account. Once a day it asks the social networks how many likes or followers a page has - Facebook, Twitter, Instagram, YouTube, Google+, OneSignal and Alexa page rank - and writes the numbers into a Google Spreadsheet, so you can chart how they grow. Nothing is installed on your computer: you paste the code in the Google editor, delete the parts you do not want, add the API keys for each network and set a daily trigger. The author stopped work in 2018 and the repository is archived.","easeRationale":"It only runs in the browser, inside Google Apps Script, so there is nothing to install - but you have to edit the JavaScript yourself (\"Remove (or duplicate) the functions as needed. Also remove them from main function!\"), get an API key from each network you want to track, run a function by hand to test it and create the daily trigger.","costNote":"The script is free and Google Apps Script has a free quota, but reading Twitter/X numbers now needs a paid X API plan.","notVerified":["Which of the counters still return data today, since each network has changed its rules since 2018","Whether the code behind the script.google.com copy link matches Main.gs in the repository","Whether the daily run stays inside Google's free Apps Script quota","What the public example spreadsheet linked in the README contains or shares"]},"zh-Hans":{"summaryShort":"一个 Google Sheets 脚本,每天记录社交网络的粉丝数和点赞数。","summary":"这是一个 Google Apps Script 脚本,你把它复制到自己的 Google 账号里。它每天向社交网络查询一个主页的点赞数或粉丝数——Facebook、Twitter、Instagram、YouTube、Google+、OneSignal 和 Alexa 网站排名——并把数字写入一个 Google 表格,方便你查看增长情况。你的电脑上不会安装任何东西:你把代码粘贴到 Google 编辑器里,删掉不需要的部分,为每个网络添加 API 密钥,并设置每天运行的触发器。作者在 2018 年停止了维护,该仓库已被归档。","easeRationale":"它只在浏览器里运行,在 Google Apps Script 内部,所以不需要安装任何东西——但你得自己修改 JavaScript 代码(「Remove (or duplicate) the functions as needed. Also remove them from main function!」),从每个想追踪的网络获取 API 密钥,手动运行一个函数来测试,并创建每日触发器。","costNote":"这个脚本是免费的,Google Apps Script 也有免费额度,但现在读取 Twitter/X 的数据需要付费的 X API 套餐。","notVerified":["如今这些计数器中哪些还能返回数据,因为每个网络自 2018 年以来都改过自己的规则","script.google.com 复制链接背后的代码是否与仓库中的 Main.gs 一致","每日运行是否仍在 Google Apps Script 的免费额度之内","README 中链接的公开示例表格里包含或分享了什么内容"]}}},{"id":"R_kgDORRCyRw","slug":"addyosmani__agent-skills","name":"Agent Skills (Addy Osmani)","owner":"addyosmani","fullName":"addyosmani/agent-skills","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"ships_lifecycle_hooks","severity":"attention","evidence":"仓库根目录有一个 \"hooks/\" 文件夹，README.md 把它列为 \"Claude Code adapter\" 的一部分：\"Slash-command wrappers, marketplace metadata, and lifecycle hooks\"。这些钩子会在你的助手运行期间自动执行，而 README 没有说明它们具体做什么。","source":"ai","title":"附带生命周期钩子"},{"code":"readme_changes_global_git_setting","severity":"attention","evidence":"为了绕过插件安装时的一个错误，README.md 让你运行 \"git config --global url.\\\"https://github.com/\\\".insteadOf git@github.com:\"，这会改变 Git 在你电脑上所有仓库中的行为，而不只是这一个。","source":"ai","title":"README 要求修改 Git 的全局设置"},{"code":"autonomous_build_mode","severity":"info","evidence":"README.md 介绍了 \"/build auto\"：它 \"generates the plan and implements every task in a single approved pass - you approve the plan once, then it runs autonomously\"，只会在 \"on failures or risky steps\" 时暂停。助手会在你不看的情况下自行写代码并提交。","source":"ai","title":"自主构建模式"},{"code":"partial_install_breaks_references","severity":"info","evidence":"README.md：\"A per-skill npx install copies only skills/<name>/, not the repo-level references/ directory. The skill still works, but paths to supplementary shared checklists are unavailable.\" 已记录为 issue #361。","source":"ai","title":"部分安装会让引用失效"},{"code":"host_support_uneven","severity":"info","evidence":"README.md 逐个工具指出了缺口：在 Copilot 中 \"the lifecycle slash commands ... are Claude Code commands and do not appear\"，在某些版本的 Antigravity CLI 中 \"legacy command TOMLs are reported as converted but their wrapper commands are not discoverable\"。","source":"ai","title":"各工具的支持程度不一"},{"code":"instructions_become_agent_prompt","severity":"info","evidence":"README.md：\"Skills are plain Markdown - they work with any agent that accepts system prompts or instruction files.\" 这 25 个文件里写的任何内容都会作为指令交给你的助手，所以仓库一更新，助手的行为也会随之改变。","source":"ai","title":"这些指令会变成助手的提示词"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"coderabbit","name":"CodeRabbit","confirmed":true}],"declaredAlternatives":[],"stars":115059,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":2009,"growthDays":7,"history":[{"date":"2026-10-26","stars":110816},{"date":"2026-10-27","stars":111143},{"date":"2026-10-28","stars":111468},{"date":"2026-10-29","stars":111752},{"date":"2026-10-30","stars":112007},{"date":"2026-10-31","stars":112257},{"date":"2026-11-01","stars":112572},{"date":"2026-11-02","stars":112835},{"date":"2026-11-03","stars":113050},{"date":"2026-11-04","stars":113257},{"date":"2026-11-05","stars":113484},{"date":"2026-11-06","stars":113714},{"date":"2026-11-07","stars":113887},{"date":"2026-11-08","stars":114093},{"date":"2026-11-09","stars":114525},{"date":"2026-11-10","stars":115059}],"firstPublishedAt":"2026-10-27T01:10:12.595Z","commitSha":"301b02b12bff98e4688e6984af6556802322bbde","verifiedAt":"2026-11-02T02:53:56.883Z","lastCommitAt":"2026-11-01T19:24:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A pack of 25 written workflows you install into an AI coding assistant so it follows a stricter way of building software.","summary":"Agent Skills is a set of 25 instruction files (plus 4 reviewer personas and 9 slash commands) that you install into an AI coding assistant such as Claude Code, Cursor, Codex, Gemini CLI or Copilot. Each file is a step-by-step workflow with checkpoints the agent is told to follow: write a spec, break the work into small tasks, write tests first, review, then ship. There is no program to open: the files only do something inside an AI assistant you already use, and most of those are paid. It is aimed at people who write software with an AI agent.","easeRationale":"On Windows, macOS and Linux alike you must already have an AI coding assistant installed and then run a command: either \"npx skills add addyosmani/agent-skills\", which needs Node, or two slash commands inside Claude Code. There is nothing to download and open, and some hosts (Cursor, Windsurf, Copilot) need you to copy the files into the right folder by hand.","costNote":"The skills are free (MIT). They only work inside an AI coding assistant, and the ones listed (Claude Code, Cursor, Codex, Copilot) are paid or have paid plans.","notVerified":["Whether code written with these workflows is actually better than without them","What the included Claude Code hooks run on your computer","How much the longer prompts add to your AI assistant's usage or bill","Whether each of the listed assistants installs the pack as described, since the README itself lists several host-specific problems"]},"zh-Hans":{"summaryShort":"一套 25 个书面工作流程，装进 AI 编程助手后，让它按更严格的方式开发软件。","summary":"Agent Skills 是一组 25 个指令文件（另有 4 个审阅角色和 9 个斜杠命令），可安装到 Claude Code、Cursor、Codex、Gemini CLI 或 Copilot 等 AI 编程助手中。每个文件都是带检查点的分步工作流程，助手会被要求照着做：先写规格说明，把工作拆成小任务，先写测试，再审查，最后发布。这里没有可以打开的程序：这些文件只有在你已经在用的 AI 助手里才起作用，而这类助手大多要付费。它面向借助 AI 助手写软件的人。","easeRationale":"在 Windows、macOS 和 Linux 上都一样：你得先装好一个 AI 编程助手，再运行一条命令，要么是需要 Node 的 \"npx skills add addyosmani/agent-skills\"，要么是在 Claude Code 里输入两条斜杠命令。没有可下载后直接打开的东西，在某些宿主工具（Cursor、Windsurf、Copilot）里，还得手动把文件复制到正确的文件夹。","costNote":"这些 skill 免费（MIT）。它们只能在 AI 编程助手里使用，而文中列出的（Claude Code、Cursor、Codex、Copilot）都要付费，或设有付费套餐。","notVerified":["用这些工作流程写出的代码，是否真的比不用时更好","随附的 Claude Code 钩子（hooks）会在你的电脑上运行什么","较长的提示词会让你的 AI 助手多用多少额度，或多花多少钱","所列的各个助手是否都能按说明安装这套文件，因为 README 本身就列出了好几个与具体宿主工具有关的问题"]}}},{"id":"R_kgDOQ77-1A","slug":"affaan-m__ecc","name":"ECC","owner":"affaan-m","fullName":"affaan-m/ECC","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"284 天内获得 291779 个星标（约每天 1027 个）","source":"check","title":"星标增长速度异常"},{"code":"background_scripts_run_during_your_ai_sessions","severity":"attention","evidence":"安装配置可以生成一个 'hook runtime'，README 将其描述为 'Enforcement, session summaries, continuous learning, instincts, and context controls'。其中写道：'Any install whose profile or modules would materialize the hook runtime requires an explicit decision. Without --enable-hooks or --no-hooks, the installer prints what the hooks can do and stops before writing anything.' 一旦启用，这些脚本会在你的电脑上随会话事件自动运行。","source":"ai","title":"AI 会话期间在后台运行的脚本"},{"code":"one_command_sends_your_code_to_another_ai_provider","severity":"attention","evidence":"README，Astra review：'/astra-review sends Claude-written code to GPT-6-Astra for an independent cross-provider review. It runs the locally installed Codex CLI with your ChatGPT login (codex login) (...) Invoking the command is your consent to send that diff to OpenAI.'","source":"ai","title":"有一条命令会把你的代码发送给另一家 AI 服务商"},{"code":"sponsored_provider_links_with_referral_codes","severity":"info","evidence":".env.example 自称是 'Canonical list of required environment variables'，其中把 ATLAS_API_KEY 与 'https://www.atlascloud.ai/?utm_source=github&utm_medium=link&utm_campaign=ECC' 列在一起；README 的赞助商一行链接到 'https://platform.kimi.ai?aff=ecc'、'https://www.greptile.com/go/ecc' 和 'https://www.kimi.ai/code?aff=ecc'。Atlas Cloud 被列为过往赞助商。","source":"ai","title":"带推荐码的赞助服务商链接"},{"code":"some_commands_need_code_from_a_third_party_repository","severity":"info","evidence":"README：`multi-*` 命令需要另一个项目的 `ccg-workflow` 运行环境，而 ECC 自己也说明 'ECC does not bundle CCG or attest to a compatible, audited CCG release'。它要求你自行审查并安装上游项目的某个确切版本。","source":"ai","title":"部分命令需要第三方仓库的代码"},{"code":"always_loaded_rules_and_a_very_large_catalog","severity":"info","evidence":"README：'Access to 71 agents, 311 skills, and 97 legacy command shims, plus hooks, rules, memory, continuous learning'，以及 'Rules are always-loaded context, so begin with common and one pack for the stack you actually use.' 所有被加载的内容都会占用助手的上下文空间，并增加每次会话的成本。","source":"ai","title":"始终加载的规则和庞大的目录"},{"code":"ships_an_auto_update_script","severity":"info","evidence":"package.json 中发布的文件列表包含 `scripts/auto-update.js`，与 `scripts/install-apply.js` 和各个 hook 脚本放在一起，所以这个软件包自带更新程序；README 只声明 'ECC never uploads diagnostics automatically'。","source":"ai","title":"自带自动更新脚本"},{"code":"project_warns_about_fake_copies_of_itself","severity":"info","evidence":"README 的警告块：'Install ECC only from verified channels (...) Third-party re-uploads and unofficial mirrors are not maintained or reviewed by the project and may contain malware.'","source":"ai","title":"项目警告存在它的假冒副本"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["python","node"],"disk":{"minMB":200,"maxMB":800,"basis":"heuristic"},"apiKeys":["Anthropic (or the account of your AI coding assistant)","GitHub token (for the optional GitHub connector)","Astraflow / UModelVerse (optional model provider)","Astraflow China endpoint (optional model provider)","Atlas Cloud (optional model provider)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","automation"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":false}],"declaredAlternatives":[{"target":"claude-code","stance":"claims","evidence":"It delegates to the same safe setup flow, but it is available only after the plugin is installed and cannot replace Claude Code's built-in `/plugin` command during a first install"}],"stars":300724,"downloads":99858,"downloadsWeek":0,"downloadsDays":7,"growth":4177,"growthDays":7,"history":[{"date":"2026-10-25","stars":288904},{"date":"2026-10-26","stars":289846},{"date":"2026-10-27","stars":290896},{"date":"2026-10-28","stars":291774},{"date":"2026-10-29","stars":292720},{"date":"2026-10-30","stars":293419},{"date":"2026-10-31","stars":294074},{"date":"2026-11-01","stars":294844},{"date":"2026-11-02","stars":295674},{"date":"2026-11-03","stars":296547},{"date":"2026-11-04","stars":297162},{"date":"2026-11-05","stars":297796},{"date":"2026-11-06","stars":298500},{"date":"2026-11-07","stars":299098},{"date":"2026-11-08","stars":299640},{"date":"2026-11-09","stars":300168},{"date":"2026-11-10","stars":300724}],"firstPublishedAt":"2026-10-25T23:18:55.736Z","commitSha":"25278e7d7b6cf6e6a4e7a6c8e10f14ff22f0eb5b","verifiedAt":"2026-10-29T01:41:24.512Z","lastCommitAt":"2026-11-06T21:21:50Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A large add-on pack of instructions, skills, agents and automatic checks for AI coding assistants such as Claude Code and Codex.","summary":"ECC plugs 71 agents, 311 skills, rules, memory and background scripts into an AI coding assistant so it plans, tests, reviews and remembers in a fixed routine. It is for people who already write software with Claude Code, Codex, Cursor or a similar assistant. It does not replace the assistant: it runs inside one, and the README says it cannot even replace Claude Code's own plugin command. You need Node.js 18 or newer, Git and the assistant, then setup commands in a terminal. Optional extras need keys from other services.","easeRationale":"Same work on Windows, macOS and Linux and there is no installer program: you first install Node.js 18 or newer, Git and an AI coding assistant, then run the guided setup in a terminal (`npx ecc-universal@2.2.3 setup`) and pick a scope and a hook profile. The 19 releases carry no platform files, and the README warns that stacking two install paths in one assistant duplicates skills, hooks and settings.","costNote":"ECC itself is free (MIT). It only works inside an AI coding assistant, which normally needs a paid plan or paid API use. The README also sells a hosted 'ECC Pro' GitHub App for private repositories from $19 per seat per month, plus sponsorship tiers.","notVerified":["What the hook scripts actually run on your computer during each session once enabled","Whether 311 skills and 71 agents improve results or mainly add context and cost","How much extra AI usage, and money, one session with ECC consumes","Whether the auto-update script fetches new code after installation","Whether the Windows setup works with the current version of each assistant","Whether uninstall really removes everything it wrote across several assistants","Real disk use after a full install plus the Node dependencies"]},"zh-Hans":{"summaryShort":"一个为 Claude Code、Codex 等 AI 编程助手提供指令、技能、代理和自动检查的大型扩展包。","summary":"ECC 为 AI 编程助手接入 71 个代理、311 个技能、规则、记忆和后台脚本,让它按固定流程进行规划、测试、审查和记忆。它适合已经在用 Claude Code、Codex、Cursor 或类似助手写代码的人。它不会取代助手本身:它运行在助手内部,README 甚至说它无法替代 Claude Code 自带的插件命令。你需要 Node.js 18 或更高版本、Git 以及该助手,然后在终端里运行安装命令。可选的扩展功能需要其他服务的密钥。","easeRationale":"在 Windows、macOS 和 Linux 上操作方式相同,也没有安装程序:先安装 Node.js 18 或更高版本、Git 和一个 AI 编程助手,然后在终端运行引导式安装(`npx ecc-universal@2.2.3 setup`),并选择范围和钩子配置。19 个发行版本都不含任何平台专属文件,README 还提醒,在同一个助手里叠加两种安装方式会导致技能、钩子和设置重复。","costNote":"ECC 本身是免费的(MIT 许可)。它只能在 AI 编程助手内部运行,而助手通常需要付费套餐或付费 API 使用。README 还出售托管版的“ECC Pro” GitHub 应用,面向私有仓库,每个席位每月 19 美元起,另有赞助等级。","notVerified":["启用后,钩子脚本在你电脑上每次会话实际执行了什么","311 个技能和 71 个代理是否真的提升效果,还是主要增加上下文和成本","使用 ECC 的一次会话会多消耗多少 AI 用量和费用","自动更新脚本是否会在安装后获取新代码","Windows 上的安装是否适用于各助手的当前版本","卸载是否真的清除了它在多个助手中写入的所有内容","完整安装加上 Node 依赖项后的实际磁盘占用"]}}},{"id":"R_kgDORKpZAg","slug":"agricidaniel__claude-seo","name":"Claude SEO","owner":"AgriciDaniel","fullName":"AgriciDaniel/claude-seo","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"pages_are_sent_to_the_ai","severity":"info","evidence":"常见问题中写道“它分析的页面和数据会发送给 Claude，也就是 Anthropic 的模型，就像你在 Claude Code 里处理的其他任何内容一样”，而且部分检查还会读取公开来源，例如 Wikipedia、Reddit 和 Common Crawl。你审查的网站会被一个外部服务读取。","source":"ai","title":"页面内容会发送给 ai"},{"code":"setup_installs_its_own_python_and_browser","severity":"info","evidence":"README 把 \"/seo setup\" 描述为会创建“自己的 Python 环境和一个无头浏览器”（Chromium），顶层文件里还包括 install.sh、install.ps1 和一个 hooks/ 文件夹，所以这个插件也能自行运行命令。","source":"ai","title":"设置过程会安装自己的 python 和浏览器"},{"code":"paid_community_gets_a_private_version","severity":"info","evidence":"README 写道“这个公开版本采用 MIT 许可证，对所有人免费。AI Marketing Hub Pro 的会员还能获得一个私有版本，可以提前使用新功能”，仓库主页是 skool.com/ai-marketing-hub-pro。一些功能会先到达付费社区。","source":"ai","title":"付费社区能提前拿到私有版本"},{"code":"results_claimed_by_the_author","severity":"info","evidence":"“真实成果”部分展示了作者自己网站的数据，“在最初 12 周获得了 10.8K 次 Search Console 点击”，还有一张成本表，最高到“$106,500”的代理商报价，标注为“典型市场价格，仅供参考”。这些是作者自己给出的数字，没有经过其他人核实。","source":"ai","title":"成果由作者本人提出"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["Google APIs (PageSpeed Insights, Chrome UX Report, Search Console, Analytics 4)","SE Ranking, Profound or Bing Webmaster Tools, for the optional extensions"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","ai_skills"],"alternativeTo":[{"slug":"semrush","name":"Semrush","confirmed":true},{"slug":"ahrefs","name":"Ahrefs","confirmed":true},{"slug":"screaming-frog-seo-spider","name":"SEO Spider","confirmed":true},{"slug":"surfer-seo","name":"SEO","confirmed":true}],"declaredAlternatives":[],"stars":27899,"downloads":1957,"downloadsWeek":0,"downloadsDays":6,"growth":1279,"growthDays":6,"history":[{"date":"2026-11-04","stars":26620},{"date":"2026-11-05","stars":26908},{"date":"2026-11-06","stars":27037},{"date":"2026-11-07","stars":27233},{"date":"2026-11-08","stars":27454},{"date":"2026-11-09","stars":27682},{"date":"2026-11-10","stars":27899}],"firstPublishedAt":"2026-11-04T01:36:50.243Z","commitSha":"6be93f45064ffbadced572b5d5c95f55fd263af0","verifiedAt":"2026-11-04T01:28:16.471Z","lastCommitAt":"2026-11-10T13:43:46Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Audits a website's SEO inside Claude Code and hands you a sorted list of fixes, with reports you keep.","summary":"Claude SEO is a plugin for Claude Code. You give it a web address and it checks the things a search consultant would look at: crawling and indexing, page speed, content quality, Schema.org markup, local and shop pages, and how likely AI answers are to quote the site. The result is a sorted action plan saved as Markdown, PDF or JSON on your machine. It is made of 26 skills and 19 agents. The README says full crawlers like Screaming Frog go faster and deeper and that \"Many people use both\".","easeRationale":"The same on all three systems: you must already have Claude Code 1.0.33 or newer and Python 3.11 to 3.14, then type three commands inside it (\"/plugin marketplace add AgriciDaniel/claude-seo\", \"/plugin install ...\" and \"/seo setup\"). There is nothing to double-click, and the manual route is a git clone plus a shell or PowerShell script.","costNote":"The plugin is free (MIT). It only works inside Claude Code, which needs a paid plan, and the README says a full audit \"uses more of your Claude plan\" because five agents run on the more expensive model.","notVerified":["How much of your Claude plan a full audit really uses","Whether the findings match what a paid SEO suite reports for the same site","Whether the automatic setup of Python and the headless browser works on your machine","What the hooks in the plugin run and when","Whether the optional extensions need paid accounts of their own"]},"zh-Hans":{"summaryShort":"在 Claude Code 里审查网站的 SEO，给你一份排好序的修复清单，还有可以保留的报告。","summary":"Claude SEO 是 Claude Code 的一个插件。你给它一个网址，它就会检查 SEO 顾问会关注的那些方面：爬取与收录、页面速度、内容质量、Schema.org 标记、本地信息和店铺页面，以及 AI 回答引用该网站的可能性。结果是一份按优先级排序的行动计划，以 Markdown、PDF 或 JSON 的形式保存在你的电脑上。它由 26 个 skill 和 19 个 agent 组成。README 说像 Screaming Frog 这样的完整爬虫跑得更快、挖得更深，而“很多人两者都用”。","easeRationale":"三个系统上都一样：你需要已经安装了 Claude Code 1.0.33 或更新版本，以及 Python 3.11 到 3.14，然后在里面输入三个命令（\"/plugin marketplace add AgriciDaniel/claude-seo\"、\"/plugin install ...\" 和 \"/seo setup\"）。没有什么文件可以直接双击，手动安装的方式则是一次 git clone 加一个 shell 或 PowerShell 脚本。","costNote":"这个插件是免费的 (MIT)。它只能在 Claude Code 里运行，而 Claude Code 需要付费计划，README 说一次完整审查“会用掉你更多的 Claude 计划额度”，因为有五个 agent 跑在更贵的模型上。","notVerified":["一次完整审查实际会用掉多少你的 Claude 计划额度","审查结果是否和付费 SEO 套件对同一网站给出的结果一致","Python 和无头浏览器的自动安装是否能在你的电脑上正常运行","插件里的 hooks 具体执行了什么，什么时候执行","那些可选的扩展是否需要自己另外付费开通账号"]}}},{"id":"R_kgDORqec6w","slug":"ai-job-search-dev__ai-job-search","name":"AI Job Search","owner":"ai-job-search-dev","fullName":"ai-job-search-dev/ai-job-search","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号是 12 天前创建的","source":"check","title":"所有者账号注册时间很短"},{"code":"personal_data_in_a_public_copy","severity":"attention","evidence":"README.md 提醒：\"A fork of this repo is always public — GitHub does not allow private forks of public repositories — and /setup (step 3 below) writes your personal data (name, contact details, employment history, salary expectations) into tracked files.\" 完全按照快速开始的步骤操作（`gh repo fork`）可能会把你的简历数据公开出去。","source":"ai","title":"公开副本中含有个人数据"},{"code":"job_board_terms_of_service","severity":"attention","evidence":"`linkedin-search` 技能是 \"built on LinkedIn's public, unauthenticated jobs-guest endpoints\"，README 说它 \"Intended for personal use only — automated access is against LinkedIn's Terms of Service, so keep volume low.\"。/add-portal 生成器也会拒绝需要登录验证的网站，并把限制较严格的网站标记为仅供个人使用。","source":"ai","title":"招聘网站的服务条款"},{"code":"job_postings_can_try_to_hijack_the_agent","severity":"attention","evidence":"README.md 警告：\"Postings are treated as untrusted input (the workflow follows no instructions embedded in them and fetches no links from their body), but agentic defenses are instruction-level, not a sandbox - on an unfamiliar job board, skim what was fetched and written before you hit send.\"","source":"ai","title":"招聘信息可能试图劫持代理"},{"code":"third_party_skills_run_pre_approved","severity":"attention","evidence":"关于从别人的 fork 借用门户技能：\"Read the code. All of it - these CLIs run pre-approved on your machine (`.claude/settings.json` allowlists them) against your career data.\" 因此从陌生人 fork 复制来的代码会在不每次询问的情况下运行。","source":"ai","title":"第三方技能被预先批准运行"},{"code":"reads_your_email_and_publishes_your_pipeline","severity":"info","evidence":"`/gmail-sync` \"reads your Gmail (via the Gmail connector) for status signals on your open applications\"，`/notion-sync` 会把你的投递情况以只读视图发布到一个 Notion 数据库中。两者都是可选的，需要你自己连接账号。","source":"ai","title":"读取你的邮箱并公开你的求职进度"},{"code":"repository_moved_to_a_new_account","severity":"info","evidence":"README.md 中的 Trendshift 徽章仍然写着 \"MadsLorentzen%2Fai-job-search\"，但仓库现在归属于 `ai-job-search-dev` 组织，该账号注册仅 12 天。LICENSE 中写的名字是 Mads Lorentzen。这就解释了为什么一个创建于 2026 年 3 月的项目会出现新账号信号。","source":"ai","title":"仓库迁移到了一个新账号下"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Anthropic (Claude Code subscription or API credits)","Gmail connector (optional, for /gmail-sync)","Notion (optional, OAuth through the Notion MCP server, for /notion-sync)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["productivity","ai_skills","automation"],"alternativeTo":[{"slug":"jobscan","name":"Jobscan","confirmed":true},{"slug":"teal","name":"Teal","confirmed":true},{"slug":"kickresume","name":"Kickresume","confirmed":true}],"declaredAlternatives":[],"stars":71084,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1703,"growthDays":7,"history":[{"date":"2026-11-01","stars":68743},{"date":"2026-11-02","stars":69089},{"date":"2026-11-03","stars":69381},{"date":"2026-11-04","stars":69589},{"date":"2026-11-05","stars":69847},{"date":"2026-11-06","stars":70070},{"date":"2026-11-07","stars":70254},{"date":"2026-11-08","stars":70519},{"date":"2026-11-09","stars":70849},{"date":"2026-11-10","stars":71084}],"firstPublishedAt":"2026-11-01T01:55:44.403Z","commitSha":"a3ba937be8b01b9414b09f2f85912139dd72be32","verifiedAt":"2026-11-03T02:42:50.441Z","lastCommitAt":"2026-11-10T19:57:03Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs your job hunt inside Claude Code: finds postings, scores the fit, tailors your CV and cover letter, preps interviews.","summary":"AI Job Search is a set of commands you add to Claude Code to run a job hunt from your own computer. You fill in a profile from your CV and documents, then /scrape searches job boards, /rank shortlists them, /apply writes a tailored CV and cover letter as PDFs and checks how a recruiting system would read them, and /interview prepares you for each round. The shipped job boards are Danish plus LinkedIn and one tech aggregator; others you add yourself. You need a paid Claude plan, Python, Bun and a LaTeX install.","easeRationale":"All three platforms need the same long list before the first application: Claude Code with a paid plan, Python 3.10+, Bun and a full LaTeX install with lualatex and xelatex, then cloning the repository, running an installer script and typing commands inside Claude Code. The README notes extra per-platform fiddling, such as different LaTeX engines and installing Poppler on Windows, macOS or Linux for the CV text check.","costNote":"The framework is free (MIT). The README states Claude Code has \"No free tier: a Claude Pro/Max/Team subscription or Anthropic API credits\", so the AI is what you pay for. Donations via Ko-fi are optional.","notVerified":["Whether it improves your results; the 69 applications, 20 interviews and one contract are the author's own account of his search","What a week of searching and applying costs in Claude usage","Whether the job boards keep working, since the README itself says some portals block automated access","How well the CV and cover letter PDFs survive a real recruiting system","Whether the built-in defences hold against a job posting written to trick the agent"]},"zh-Hans":{"summaryShort":"在 Claude Code 里完成求职：查找职位、评估匹配度、定制简历和求职信、并准备面试。","summary":"AI Job Search 是一组你添加到 Claude Code 里的命令，用来在自己的电脑上完成求职流程。你根据简历和相关材料填写一份档案，然后 /scrape 搜索招聘网站，/rank 对结果排序筛选，/apply 生成定制的简历和求职信 PDF，并检查招聘系统会如何读取它们，/interview 帮你准备每一轮面试。内置的招聘网站包括丹麦的网站，加上 LinkedIn 和一个科技类聚合网站；其他网站需要你自己添加。你需要一个付费的 Claude 方案、Python、Bun 和一套 LaTeX 安装。","easeRationale":"三个平台在第一次投递前都需要同样长的准备清单：带付费方案的 Claude Code、Python 3.10+、Bun，以及包含 lualatex 和 xelatex 的完整 LaTeX 安装，还要克隆仓库、运行安装脚本，并在 Claude Code 里输入命令。README 还提到各平台会有额外的小麻烦，比如使用不同的 LaTeX 引擎，以及在 Windows、macOS 或 Linux 上安装 Poppler 用于简历文本检查。","costNote":"该框架是免费的（MIT 许可）。README 写道 Claude Code \"No free tier: a Claude Pro/Max/Team subscription or Anthropic API credits\"，也就是说你花钱买的是 AI 本身。通过 Ko-fi 捐款是可选的。","notVerified":["是否真的能提升结果；文中提到的 69 次投递、20 次面试和一份合同只是作者本人对自己求职经历的描述","一周的搜索和投递在 Claude 用量上花费多少","招聘网站是否能一直正常使用，因为 README 自己也说有些网站会屏蔽自动化访问","简历和求职信 PDF 在真实招聘系统里表现如何","内置防护能否抵御专门设计来欺骗代理的招聘信息"]}}},{"id":"R_kgDOVM98OA","slug":"aicodeking__leanrea","name":"LeanREA","owner":"aicodeking","fullName":"aicodeking/leanrea","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"reverse_engineering_may_be_restricted","severity":"attention","evidence":"该项目的全部目的就是逆向工程：描述中说它 \"replaces REA (Reverse Engineer Anything)\"，基准测试任务会追踪 \"a real Electron app's app.asar\" 内部的功能以及一款桌面应用的原生插件。拆解不属于你的软件可能违反其许可条款，在一些国家还受到法律限制。与 REA 的 README 不同，这个项目没有任何关于合法使用的声明。","source":"ai","title":"逆向工程可能受到限制"},{"code":"the_speed_numbers_come_from_the_author","severity":"info","evidence":"标题数字 \"38% faster, 35% cheaper\" 来自作者自己针对其对标项目所做的基准测试，他在 Caveats 中写道 \"Only 3 reps per cell, one model, one machine\"，并说明 Ghidra、Hopper 和 jadx 没有安装。他还指出，相比完全不使用 skill，提升幅度更小，\"about 19% faster\"。","source":"ai","title":"速度数据来自作者本人"},{"code":"the_assistant_runs_shell_commands_on_your_machine","severity":"info","evidence":"对比表中写着 \"Tools added to the agent: 0 (uses Bash)\"，也就是说这个 skill 的工作方式是让助手以你的权限直接运行命令，包括对被检查程序的 \"runtime tracing\"。","source":"ai","title":"助手会在你的电脑上运行 shell 命令"},{"code":"very_new_and_idle_since_launch","severity":"info","evidence":"仓库创建于 2026-10-17，已有 3233 个星标，没有发布版本，最后一次提交是 2026-10-17（距本次检查 10 天前），其中只有 .gitignore、LICENSE、README.md、bench/ 和 leanrea/。","source":"ai","title":"非常新，发布后一直没有更新"}],"platforms":["mac","linux"],"tools":["node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["AI coding assistant (Claude Code, Codex, Cursor, Gemini CLI and similar)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"ida-pro","name":"IDA Pro","confirmed":true},{"slug":"hopper","name":"Hopper","confirmed":true}],"declaredAlternatives":[],"stars":4185,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":212,"growthDays":7,"history":[{"date":"2026-10-26","stars":2990},{"date":"2026-10-27","stars":3232},{"date":"2026-10-28","stars":3444},{"date":"2026-10-29","stars":3619},{"date":"2026-10-30","stars":3747},{"date":"2026-10-31","stars":3799},{"date":"2026-11-01","stars":3852},{"date":"2026-11-02","stars":3929},{"date":"2026-11-03","stars":3973},{"date":"2026-11-04","stars":3997},{"date":"2026-11-05","stars":4020},{"date":"2026-11-06","stars":4042},{"date":"2026-11-07","stars":4074},{"date":"2026-11-08","stars":4111},{"date":"2026-11-09","stars":4130},{"date":"2026-11-10","stars":4185}],"firstPublishedAt":"2026-10-27T01:10:12.596Z","commitSha":"09ba4aa0217f82cd880f855a60b877f59f39b22b","verifiedAt":"2026-10-28T01:54:12.894Z","lastCommitAt":"2026-10-17T03:26:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A short instruction file that teaches an AI coding assistant to take apart programs on your computer with plain command-line tools.","summary":"leanrea is a single 8 KB instruction file (a \"skill\") for AI coding assistants such as Claude Code, Codex, Cursor or Gemini CLI. Instead of adding a server and hundreds of extra tools, it tells the assistant which ordinary command to use for each kind of file (Mac and Linux binaries, Electron apps, Android APKs, Go and .NET programs, WASM, SQLite, network captures) and warns it about common traps. The assistant then runs those commands itself. You install it with one npx command, and you supply the inspection tools and a paid AI assistant.","easeRationale":"On macOS, the system the author tested on, it is a C: `npx skills add aicodeking/leanrea` needs Node and a terminal, and the skill only pays off once the command-line inspection tools it routes to are present. Linux is the same grade for the Linux-side tools, but the author's examples lean on macOS commands such as otool and dyld_info. Windows is not covered: the manual install and the commands assume a Unix shell.","costNote":"The skill is free (MIT), but it only does anything inside a paid AI coding assistant. The author's own benchmark run of 36 tasks cost about 8 US dollars in model usage.","notVerified":["Whether the benchmark reproduces on another machine, model or app version; the author says the answer keys only match the versions he used.","Which inspection tools you have to install yourself (asar, jadx, apktool, Ghidra, tshark, binwalk and others are not included).","Whether the routing commands work on Linux and in agents other than Claude Code.","What the skill actually tells the assistant to do, since the SKILL.md text is not shown in the README."]},"zh-Hans":{"summaryShort":"一份简短的指令文件，教 AI 编程助手用普通的命令行工具拆解你电脑上的程序。","summary":"leanrea 是一个仅 8 KB 的单一指令文件（一种\"技能\"），供 Claude Code、Codex、Cursor 或 Gemini CLI 等 AI 编程助手使用。它不是添加一个服务器和成百上千的额外工具，而是告诉助手针对每种文件该用哪个常见命令（Mac 和 Linux 二进制文件、Electron 应用、安卓 APK、Go 和 .NET 程序、WASM、SQLite、网络抓包），并提醒它常见的坑。助手之后会自己运行这些命令。你用一条 npx 命令安装它，并自备检查工具和一个付费的 AI 助手。","easeRationale":"在作者实际测试过的 macOS 上，它得到 C 级：`npx skills add aicodeking/leanrea` 需要 Node 和终端，而这个技能只有在它所指向的命令行检查工具已经就位时才有用。Linux 在其自身工具上得到同样的评级，但作者的示例依赖于 otool 和 dyld_info 等 macOS 命令。Windows 完全没有被覆盖：手动安装步骤和命令都假定使用 Unix shell。","costNote":"这个技能是免费的（MIT），但它只能在付费的 AI 编程助手内部起作用。作者自己跑的 36 项任务基准测试花费了大约 8 美元的模型使用费用。","notVerified":["基准测试结果能否在另一台机器、模型或应用版本上重现；作者说标准答案只匹配他用过的那些版本。","你需要自己安装哪些检查工具（asar、jadx、apktool、Ghidra、tshark、binwalk 等均未包含）。","这些路由命令是否在 Linux 上以及 Claude Code 之外的助手中也能用。","这个技能到底告诉助手去做什么，因为 README 里并没有展示 SKILL.md 的文本内容。"]}}},{"id":"R_kgDOLswaBQ","slug":"aigc-apps__easyanimate","name":"EasyAnimate","owner":"aigc-apps","fullName":"aigc-apps/EasyAnimate","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在2024-05-29(892天前)","source":"check","title":"版本更新停滞"},{"code":"no_updates_for_over_a_year","severity":"attention","evidence":"最近一次push在2025-03-06,611天没有提交,只有一个来自2024-05-29的发布。README最新条目说模型已迁移到diffusers库(\"EasyAnimate-V5.1 is now supported in diffusers\", 2025.03.06),说明仓库本身似乎已停滞,工作转到了别处。","source":"ai","title":"超过一年没有更新"},{"code":"heavy_gpu_and_disk_use","severity":"info","evidence":"README.md: \"We need about 60GB available on disk (for saving weights)\",模型文件有30到39 GB,表格显示768x1344的视频只能在24 GB或更高显存的显卡上运行。确定性磁盘估算(1到4.5 GB)只涵盖代码和依赖包,不包括模型。","source":"ai","title":"GPU和磁盘占用大"},{"code":"docker_image_from_vendor_registry","severity":"info","evidence":"README.md Quick Start: \"docker pull mybigpai-public-registry.cn-beijing.cr.aliyuncs.com/easycv/torch_cuda:easyanimate\"。现成镜像来自阿里巴巴自己的仓库,仓库里没有公开的构建方法。","source":"ai","title":"Docker镜像来自厂商仓库"},{"code":"model_files_hosted_elsewhere","severity":"info","evidence":"Model zoo部分把每个权重都链接到huggingface.co/alibaba-pai或modelscope.cn。仓库里的Apache-2.0许可只覆盖代码;这些下载的模型文件随附的条款这里并未显示。","source":"ai","title":"模型文件托管在别处"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":1010,"maxMB":4570,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"kling","name":"Kling","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":2265,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":2265},{"date":"2026-11-09","stars":2265},{"date":"2026-11-10","stars":2265}],"firstPublishedAt":"2026-11-07T22:56:16.646Z","commitSha":"c2a70d17daa1549a22129d0cd0e02053ecccd75c","verifiedAt":"2026-11-07T22:40:47.887Z","lastCommitAt":"2025-03-06T11:41:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes about 6-second AI videos from text, a photo or another video on your own NVIDIA computer.","summary":"An Alibaba research project that generates videos of roughly 6 seconds at up to 1024x1024 from a text prompt, a starting image or a reference video, with camera and pose controls, and can also train your own style models. It has a browser page (Gradio), ComfyUI nodes, a Docker image and scripts you edit and run. You need Python 3.10 or 3.11, an NVIDIA card (12 GB works for small videos, 24 GB or more for the bigger ones) and about 60 GB of disk for the model files. It is a project for people comfortable with the command line.","easeRationale":"Both Windows and Linux need Python 3.10 or 3.11 with CUDA, a pip install of a long list of packages, a 30 to 39 GB model download and editing a Python file or starting the Gradio page by hand; the README says it was tested with an NVIDIA 3060 12 GB on Windows and V100/A10/A100 cards on Linux. Linux also offers a ready Docker image, but Docker plus an NVIDIA driver is no simpler.","notVerified":["Whether it still installs today, since the package list has not been touched in over a year and torch, diffusers and Gradio have moved on.","Real disk use after downloading models, beyond the 60 GB the README asks for.","How good the videos are on a 12 GB card with the memory-saving options, and how long each one takes.","Whether the Hugging Face and ModelScope online demos linked in the README are still running.","What licence the downloaded model files carry and whether they allow commercial use."]},"zh-Hans":{"summaryShort":"在你自己的NVIDIA电脑上,用文字、一张照片或另一段视频,生成约6秒的AI视频。","summary":"这是阿里巴巴的一个研究项目,可以根据文字提示、起始图片或参考视频生成最长约6秒、最高1024x1024分辨率的视频,支持镜头和姿态控制,还能训练你自己的风格模型。它有浏览器页面(Gradio)、ComfyUI节点、Docker镜像,以及需要你自行编辑和运行的脚本。你需要Python 3.10或3.11、一块NVIDIA显卡(12 GB可生成小视频,24 GB或更高可生成更大的视频),以及约60 GB磁盘空间存放模型文件。这是面向熟悉命令行的人的项目。","easeRationale":"无论Windows还是Linux,都需要带CUDA的Python 3.10或3.11、用pip安装一长串依赖包、下载30到39 GB的模型,并手动编辑Python文件或启动Gradio页面;README说测试环境是Windows上的NVIDIA 3060(12 GB)和Linux上的V100/A10/A100显卡。Linux还提供现成的Docker镜像,但Docker加上NVIDIA驱动并不会更简单。","notVerified":["今天是否还能安装成功,因为依赖列表已经一年多没更新,而torch、diffusers和Gradio都已经更新换代。","下载模型后的实际磁盘占用,是否超过README所说的60 GB。","在12 GB显卡上配合省显存选项时视频效果如何,每段要花多长时间。","README中链接的Hugging Face和ModelScope在线演示是否还能用。","下载的模型文件采用什么许可,是否允许商业使用。"]}}},{"id":"R_kgDOJKPe4Q","slug":"aigc-audio__audiogpt","name":"AudioGPT","owner":"AIGC-Audio","fullName":"AIGC-Audio/AudioGPT","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞 (OSV)：basicsr: GHSA-86w8-vhw6-q9qq, PYSEC-2026-1215; langchain: GHSA-2qmj-7962-cjq8, GHSA-3644-q5cj-c5c7, GHSA-3hjh-jh2h-vrg6; opencv-contrib-python: GHSA-cxjf-x6jp-p7mc, PYSEC-2023-181; pytorch-lightning: GHSA-2vj5-px25-gjrp, GHSA-4cv3-v7pv-rfhf, GHSA-75m9-98v2-hjpm; soupsieve: GHSA-2wc2-fm75-p42x, GHSA-836r-79rf-4m37, GHSA-gjv8-xp57-g29c; streamlit: GHSA-7p48-42j8-8846, GHSA-8qw9-gf7w-42x5, GHSA-rxff-vr5r-8cj5; torch: GHSA-3749-ghw9-m3mg, GHSA-47fc-vmwq-366v, GHSA-53q9-r3pm-6pq6; transformers: GHSA-282v-666c-3fvg, GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm","source":"check","title":"已知漏洞"},{"code":"no_licence_granted","severity":"attention","evidence":"仓库里的 LICENSE 文件是空的，GitHub 也报告该仓库没有许可证，但 README 却说“我们以开源方式提供我们的实现和预训练模型”。没有许可证文本，就没有人授权你使用或再分发这份代码。","source":"ai","title":"没有获得任何许可"},{"code":"unfinished_and_left_alone","severity":"info","evidence":"能力表里很多任务标记为 \"Yes (WIP)\" 或 \"WIP\"，README 自己也说“目前并非每个模型都有对应仓库”，最后一次提交是在 2023-05-05。可以预期会有缺失的部分，也不会再修复。","source":"ai","title":"未完成且无人维护"},{"code":"pulls_code_and_models_from_elsewhere","severity":"info","evidence":"requirements.txt 开头是 \"--extra-index-url https://download.pytorch.org/whl/cu113\"，结尾是 \"git+https://github.com/openai/CLIP.git\"，仓库里还有一个 download.sh 用来获取模型文件。安装过程会连接好几个外部地方。","source":"ai","title":"代码和模型都从别处拉取"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["OpenAI"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["audio","ai_agents"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true}],"declaredAlternatives":[],"stars":10167,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":1,"growthDays":6,"history":[{"date":"2026-11-04","stars":10166},{"date":"2026-11-05","stars":10164},{"date":"2026-11-06","stars":10164},{"date":"2026-11-07","stars":10165},{"date":"2026-11-08","stars":10166},{"date":"2026-11-09","stars":10166},{"date":"2026-11-10","stars":10167}],"firstPublishedAt":"2026-11-04T01:36:50.253Z","commitSha":"a674543c537bb3a60b8f521df76b97a627c1c379","verifiedAt":"2026-11-04T01:29:19.588Z","lastCommitAt":"2023-05-05T21:42:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that takes a written request and runs audio models for speech, singing, sound effects or a talking head.","summary":"AudioGPT is the code behind a research paper. You write what you want in plain words and a chat model picks which audio model to run: reading text aloud, changing a voice's style, transcribing speech, separating or cleaning up sound, making a sound effect from text or a picture, singing, and a talking head video. Many of the tasks are marked work in progress and the repository stopped in May 2023. It is a Python project for a Linux machine with an NVIDIA card, and the models are downloaded separately.","easeRationale":"The README's whole \"Get Started\" section is \"Please refer to run.md\", there are no releases and no installer. requirements.txt pins around 70 packages against the CUDA 11.3 build of PyTorch (torch==1.12.1, mmcv==1.5.0) and pulls one from GitHub, and a download.sh fetches the models, so getting it running is developer work on a Linux machine with an NVIDIA card.","costNote":"No price for the code, but it is built around a chat model: requirements.txt pins the \"openai\" and \"langchain\" packages and the main file is audio-chatgpt.py, so expect to pay for the OpenAI usage.","notVerified":["Whether the install still works at all, since the pinned versions are from 2023","Whether the Hugging Face demo page listed as the project homepage still runs","How much disk space the downloaded models take","Which tasks actually work, given how many are marked work in progress","How much the OpenAI usage costs for a session"]},"zh-Hans":{"summaryShort":"一份研究代码，接受文字请求后运行语音、歌唱、音效或说话人头像等音频模型。","summary":"AudioGPT 是一篇研究论文背后的代码。你用简单的文字写出想要什么，一个对话模型会选择运行哪个音频模型：朗读文字、改变声音风格、转写语音、分离或清理声音、根据文字或图片生成音效、歌唱，以及生成说话人头像视频。很多任务都标记为开发中，仓库在 2023 年 5 月停止更新。这是一个运行在带 NVIDIA 显卡的 Linux 机器上的 Python 项目，模型需要单独下载。","easeRationale":"README 的“Get Started”部分整段只写着 \"Please refer to run.md\"，没有发行版也没有安装程序。requirements.txt 固定了大约 70 个包，对应 PyTorch 的 CUDA 11.3 版本（torch==1.12.1、mmcv==1.5.0），还有一个包直接从 GitHub 拉取，另有一个 download.sh 用来获取模型，所以要让它跑起来，需要在带 NVIDIA 显卡的 Linux 机器上做开发者级别的工作。","costNote":"代码本身不收费，但它是围绕一个对话模型搭建的：requirements.txt 固定了 \"openai\" 和 \"langchain\" 这两个包，主文件是 audio-chatgpt.py，所以要准备为 OpenAI 的使用付费。","notVerified":["由于固定的版本来自 2023 年，安装流程是否还能完全正常运作","作为项目主页列出的 Hugging Face 演示页面是否还能用","下载的模型占用多少磁盘空间","鉴于这么多任务都标记为开发中，哪些任务真的能用","每次会话使用 OpenAI 要花多少钱"]}}},{"id":"R_kgDOG-nOqw","slug":"aiyakuaile__easy_tv_live","name":"Easy TV Live","owner":"aiyakuaile","fullName":"aiyakuaile/easy_tv_live","type":"ready_app","status":"alert","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"high","signals":[{"code":"supports_private_subscription_sources","severity":"attention","evidence":"README「TV源」部分：应用自带的频道列表来自 IPTV-ORG 项目，每天同步两次，作者写道\"本应用不对第三方订阅源的可访问性进行保证，推荐自行添加订阅源\"（应用不保证第三方订阅源一定能用，建议自行添加）。README 还记录了两种访问非公开源的方式：自定义请求头，\"#UA-Hint: 请将 User-Agent 设置为 okHttp/Mod-1.0.0 ，否则无法观看\"（把 User-Agent 设置为 okHttp/Mod-1.0.0，否则无法观看），以及写在链接里的基本身份认证，\"https://username:password@xxxxxxxxx.m3u\"。这些正是访问付费和私人 IPTV 订阅的方式。仓库没有说明这些列表来自哪里、由谁维护。付费电视频道带有频道所有者制定的内容授权规则，这些规则在不同国家之间也不一样。","source":"ai","title":"支持私人订阅源"},{"code":"license_is_non_commercial_creative_commons","severity":"info","evidence":"LICENSE 文件是\"Creative Commons Attribution-NonCommercial-ShareAlike 4.0\"：\"NonCommercial: You may not use the material for commercial purposes\"——这是一种内容许可证，不是软件许可证，所以 GitHub 显示没有标准许可证。","source":"ai","title":"许可证是非商业性的知识共享协议"},{"code":"readme_describes_a_version_that_is_not_in_releases","severity":"info","evidence":"README 里有一条关于\"v3.0\"及其新 logo 的重大变更说明，但 GitHub 上最新的发布版本是 2025-09-23 的 2.9.9，所以说明里描述的版本并不在这里的下载列表中。","source":"ai","title":"readme 描述的版本不在发布列表中"},{"code":"app_can_be_set_up_from_a_remote_file","severity":"info","evidence":"README「远程配置功能」：粘贴一个 JSON 文件的链接（在电视上还可以扫描二维码），应用就会自动套用这些设置，所以谁掌握这个文件，谁就掌握了应用的源和设置。","source":"ai","title":"应用可以通过远程文件进行配置"},{"code":"plays_channel_lists_from_outside_sources","severity":"high","evidence":"项目简介：\"Easy TV Live App, a lightweight IPTV player, supports all platforms and Android TV big screens\"。自带的列表来自 IPTV-ORG 项目，README「TV源」部分写道\"本应用不对第三方订阅源的可访问性进行保证，推荐自行添加订阅源\"（应用不保证第三方订阅源一定能用，建议自行添加）。README 记录了两种访问非公开源的方式：自定义请求头，\"#UA-Hint: 请将 User-Agent 设置为 okHttp/Mod-1.0.0 ，否则无法观看\"（把 User-Agent 设置为 okHttp/Mod-1.0.0，否则无法观看），以及写在链接里的凭据，\"https://username:password@xxxxxxxxx.m3u\"。仓库没有说明这些列表来自哪里、由谁维护。在许多国家，转播电视频道需要频道所有者的授权，付费电视频道还带有所有者制定的内容授权规则。","source":"ai","title":"播放来自外部来源的频道列表"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":19,"maxMB":58,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["video","hardware"],"alternativeTo":[{"slug":"tivimate","name":"TiviMate","confirmed":true},{"slug":"plex","name":"Plex","confirmed":false},{"slug":"emby","name":"Emby","confirmed":false}],"declaredAlternatives":[],"stars":1162,"downloads":37760,"downloadsWeek":190,"downloadsDays":4,"growth":1,"growthDays":6,"history":[{"date":"2026-11-06","stars":1163},{"date":"2026-11-07","stars":1163},{"date":"2026-11-08","stars":1163},{"date":"2026-11-09","stars":1162},{"date":"2026-11-10","stars":1162}],"firstPublishedAt":"2026-11-05T22:53:18.468Z","commitSha":"52bd5cb444e376652d25b9f7e3553ceccf4a5470","verifiedAt":"2026-11-05T22:46:08.150Z","lastCommitAt":"2026-11-09T23:04:50Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A simple IPTV player for Windows, Mac, Linux and Android TV: add an m3u channel list and watch with the remote.","summary":"Easy TV Live (极简TV) is a small IPTV player built with Flutter. It opens .m3u and .txt channel lists, groups and merges repeated channels, lets you change channel by typing a number and works with a TV remote on Android TV. It comes with the free channel list from the IPTV-ORG project, refreshed twice a day, and you can add your own lists, including local files, links with a password and links that need a custom User-Agent. Ready downloads exist for Windows, macOS and Linux. The README is in Chinese, with an English version in README-en.md.","easeRationale":"Release 2.9.9 has ready files for the three systems — easyTV-2.9.9-windows.zip (19 MB), easyTV-2.9.9-macos.zip (84 MB) and .deb, .rpm, .AppImage or .tar.gz for Linux — so you download, open and watch the channels that already come with it; nothing to install first.","costNote":"Free; the README has a WeChat donation code. The license allows only non-commercial use.","notVerified":["Whether the channels in the bundled IPTV-ORG list actually play, since the README says it does not guarantee any source works","How much of the interface is translated to English","Whether the Windows, macOS and Linux builds open without a warning from the system, as they are zips without a signed installer","What the app sends to the internet besides the channel lists and the Bing daily image"]},"zh-Hans":{"summaryShort":"一款简洁的 IPTV 播放器，支持 Windows、Mac、Linux 和 Android TV：添加 m3u 频道列表，用遥控器就能观看。","summary":"Easy TV Live（极简TV）是一款用 Flutter 开发的小型 IPTV 播放器。它能打开 .m3u 和 .txt 频道列表，对重复频道进行分组和合并，支持输入数字换台，并能在 Android TV 上配合遥控器使用。它自带来自 IPTV-ORG 项目的免费频道列表，每天更新两次，你也可以添加自己的列表，包括本地文件、带密码的链接，以及需要自定义 User-Agent 的链接。Windows、macOS 和 Linux 都有现成的下载包。README 是中文的，英文版在 README-en.md 中。","easeRationale":"2.9.9 版本为三个系统都提供了现成文件——easyTV-2.9.9-windows.zip（19 MB）、easyTV-2.9.9-macos.zip（84 MB），以及 Linux 的 .deb、.rpm、.AppImage 或 .tar.gz——所以你只需下载、打开，就能看到自带的频道；不需要先安装任何东西。","costNote":"免费；README 里有一个微信赞赏码。许可证只允许非商业用途。","notVerified":["内置的 IPTV-ORG 列表里的频道是否真的能播放，因为 README 说它不保证任何源都能用","界面里有多少内容被翻译成了英文","Windows、macOS 和 Linux 的安装包打开时是否会被系统弹出警告，因为它们是没有签名安装程序的压缩包","除了频道列表和必应每日图片之外，这个应用还向互联网发送了什么"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1MTk4MDQ1NQ==","slug":"alacritty__alacritty","name":"Alacritty","owner":"alacritty","fullName":"alacritty/alacritty","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"B"},"risk":"low","signals":[{"code":"authors_call_it_beta","severity":"info","evidence":"README：“这个软件被认为处于 beta 级别的完备程度；还缺少一些功能，也有一些要修复的问题，但已经有很多人把它当作日常使用的终端”。","source":"ai","title":"作者把它称为 beta 版"},{"code":"no_tabs_splits_or_settings_screen","severity":"info","evidence":"常见问题中写道“你不会找到标签页或分屏之类的东西（这些最好交给窗口管理器或终端多路复用器），也没有像图形化配置编辑器那样的便利功能”，README 还补充说“Alacritty 不会替你创建配置文件”。你需要手动编辑一个 alacritty.toml 文件。","source":"ai","title":"没有标签页、分屏或设置界面"},{"code":"needs_opengl_and_recent_windows","severity":"info","evidence":"需求部分要求“至少 OpenGL ES 2.0”，在 Windows 上还要求“支持 ConPTY（Windows 10 版本 1809 或更高）”。非常旧的机器或没有额外配置的虚拟机可能无法运行它。","source":"ai","title":"需要 opengl 和较新的 windows"},{"code":"releases_slower_than_the_code","severity":"info","evidence":"最新发行版 v0.17.0 是 2026-04-06 发布的，而代码最后一次更改是在 2026-11-02。现成文件比仓库落后了好几个月。","source":"ai","title":"发行版落后于代码"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":6,"maxMB":17,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":66162,"downloads":2978496,"downloadsWeek":11473,"downloadsDays":6,"growth":49,"growthDays":7,"history":[{"date":"2026-11-04","stars":66113},{"date":"2026-11-05","stars":66120},{"date":"2026-11-06","stars":66123},{"date":"2026-11-07","stars":66123},{"date":"2026-11-08","stars":66136},{"date":"2026-11-09","stars":66145},{"date":"2026-11-10","stars":66162}],"firstPublishedAt":"2026-11-04T01:36:50.262Z","commitSha":"aecf0f600fa840a415b55da5cf34fbeaec39ac32","verifiedAt":"2026-11-04T01:29:58.160Z","lastCommitAt":"2026-11-06T21:34:25Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Alacritty-v0.17.0-portable.exe","sizeBytes":6003200,"url":"https://github.com/alacritty/alacritty/releases/download/v0.17.0/Alacritty-v0.17.0-portable.exe","releaseUrl":"https://github.com/alacritty/alacritty/releases/tag/v0.17.0"},{"os":"mac","file":"Alacritty-v0.17.0.dmg","sizeBytes":6136240,"url":"https://github.com/alacritty/alacritty/releases/download/v0.17.0/Alacritty-v0.17.0.dmg","releaseUrl":"https://github.com/alacritty/alacritty/releases/tag/v0.17.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A fast, plain terminal window for typing commands, set up through a text file instead of menus.","summary":"Alacritty is a terminal: the window where you type commands. It draws the text with your graphics card to stay quick even with lots of output, and keeps itself deliberately simple. There are no tabs, no split panes and no settings screen; you change colours, font and keys by writing a TOML file yourself, and the program does not create that file for you. Ready installers exist for Windows and Mac, and Linux and BSD users install it from their package manager. The authors call it beta.","easeRationale":"Windows is the easiest, with a 2 MB Alacritty-v0.17.0-installer.msi (or a portable .exe) to double-click, and Mac has the same with a 6 MB .dmg. Linux gets a B because the releases carry no Linux file: the README says it \"can be installed by using various package managers on Linux\", so it is one command in your distribution instead of a download.","costNote":"Apache-2.0 licence, nothing to pay.","notVerified":["Whether it really feels faster than your current terminal on your own machine","Whether your Linux distribution's package manager has a recent version","How well it works over remote desktop or in a virtual machine without graphics acceleration","What the beta label means in practice for day-to-day use"]},"zh-Hans":{"summaryShort":"一个快速、简洁的终端窗口，用来输入命令，通过文本文件而不是菜单来设置。","summary":"Alacritty 是一个终端：你用来输入命令的窗口。它用你的显卡来绘制文字，即使输出很多也能保持流畅，而且刻意保持简单。它没有标签页、没有分屏，也没有设置界面；你需要自己编写一个 TOML 文件来修改颜色、字体和按键，程序本身不会替你创建这个文件。Windows 和 Mac 上有现成的安装程序，Linux 和 BSD 用户则通过各自的包管理器安装。作者把它称为 beta 版。","easeRationale":"Windows 上最简单，有一个 2 MB 的 Alacritty-v0.17.0-installer.msi（或便携版 .exe）可以直接双击，Mac 上同样有一个 6 MB 的 .dmg。Linux 评为 B，因为发行版里没有附带 Linux 文件：README 说它“可以在 Linux 上通过各种包管理器安装”，所以是在你的发行版里输入一条命令，而不是下载文件。","costNote":"Apache-2.0 许可证，不需要付费。","notVerified":["在你自己的机器上是否真的比你现在用的终端更快","你的 Linux 发行版的包管理器里是否有比较新的版本","在远程桌面或没有图形加速的虚拟机里表现如何","beta 这个标签在日常使用中实际意味着什么"]}}},{"id":"R_kgDOU9iibA","slug":"alchaincyf__huashu-art-motion","name":"Huashu Art Motion","owner":"alchaincyf","fullName":"alchaincyf/huashu-art-motion","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"26 天内获得 5933 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"content_in_chinese","severity":"info","evidence":"README.en.md 写明 \"The skill content is in Chinese\"：53 张风格配方卡、9 张语法卡和 12 份方法文档都是中文写的，尽管英文简介不是中文。","source":"ai","title":"内容是中文"},{"code":"parts_not_covered_by_the_mit_licence","severity":"attention","evidence":"README.md 的许可部分把几个随附文件排除在 MIT 之外：笔画数据沿用 Arphic Public License，字体沿用各自的 OFL 许可，作者的卡通角色画面和 Clawd 影片则 \"are for demo use only\" 且 \"not covered by MIT\"。","source":"ai","title":"部分内容不在 MIT 许可范围内"},{"code":"uses_another_company_mascot","severity":"attention","evidence":"README.md：\"Clawd is Anthropic's Claude Code mascot; the Clawd: Night at the Museum clips are an unofficial fan demo, not affiliated with Anthropic\"。样片还模仿了一些知名频道和节目的风格（Kurzgesagt、Vox、3Blue1Brown，以及一份 \"Rick & Morty style\" 配方），项目称只保存测得的参数，不保存它们的画面。","source":"ai","title":"使用了另一家公司的吉祥物"},{"code":"built_on_someone_elses_work","severity":"info","evidence":"README.md：项目起初是对 Tak 那部 15 秒的 \"Art History Speedrun\" 的重做，\"the scene layouts and the girl-and-cat premise follow his original\"，每一帧都用代码重新绘制，不含任何原始画面或音频。","source":"ai","title":"建立在他人作品之上"},{"code":"install_tested_only_partly","severity":"info","evidence":"README.md 这样概括 references/compatibility.md：在三个宿主工具上安装和重装都通过，\"Codex online regression passes, Claude Code / Kimi online regression blocked by test-account conditions, pending\"。","source":"ai","title":"安装只测试了一部分"},{"code":"people_need_an_image_model","severity":"info","evidence":"README.md：\"characters are handed to an image model to produce frames; the code only composes, swaps frames and handles texture\"，第九套叙事语法 \"requires your own character assets\"。有几个完整影片示例被描述为 \"code snapshots, not ready-to-render projects\"。","source":"ai","title":"需要自备图像模型"}],"platforms":["windows","mac","linux"],"tools":["python","node","ffmpeg","other"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","video","image"],"alternativeTo":[{"slug":"after-effects","name":"After Effects","confirmed":true},{"slug":"remotion","name":"Remotion","confirmed":true}],"declaredAlternatives":[],"stars":6696,"downloads":3369,"downloadsWeek":0,"downloadsDays":7,"growth":577,"growthDays":7,"history":[{"date":"2026-10-26","stars":5311},{"date":"2026-10-27","stars":5538},{"date":"2026-10-28","stars":5674},{"date":"2026-10-29","stars":5762},{"date":"2026-10-30","stars":5826},{"date":"2026-10-31","stars":5862},{"date":"2026-11-01","stars":5932},{"date":"2026-11-02","stars":6011},{"date":"2026-11-03","stars":6119},{"date":"2026-11-04","stars":6237},{"date":"2026-11-05","stars":6317},{"date":"2026-11-06","stars":6396},{"date":"2026-11-07","stars":6458},{"date":"2026-11-08","stars":6535},{"date":"2026-11-09","stars":6620},{"date":"2026-11-10","stars":6696}],"firstPublishedAt":"2026-10-27T01:10:12.597Z","commitSha":"8a0d7d72ca65fcce6f8ba82c4a6c8dfe35b8422c","verifiedAt":"2026-11-02T03:06:08.531Z","lastCommitAt":"2026-11-10T01:38:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill for AI coding assistants that writes code to turn art styles into moving pictures and explainer-video clips.","summary":"huashu-art-motion is an add-on for an AI coding assistant (Claude Code, Codex and similar): you describe a video and it writes code that draws and animates each scene. It carries recipes for 36 art styles (cave painting, Van Gogh, Bauhaus, 8-bit and more) and 9 narration grammars measured from well-known explainer channels (the repository keeps only the measured parameters, not their frames), plus scene code and quality checks. You run it from the terminal with uv (Python), ffmpeg and a headless Chromium. Almost all the written material is in Chinese.","easeRationale":"On all three systems it is terminal work with several tools: README.md lists uv, ffmpeg and Playwright Chromium (\"uv run --with playwright playwright install chromium\"), plus an AI coding assistant to drive the skill, and you render with commands like \"uv run --with playwright python scripts/engine/render.py --spec ... --out finance.mp4\". Windows is the least tested of the three: the thanks section records a Windows-only bug in the render service's parallel loading.","costNote":"The skill is free (MIT). It runs inside an AI coding assistant, most of which are paid, and the optional voice and image features use outside services with your own account.","notVerified":["Whether the videos you get look like the samples in the README, which were made by the author with his own assets","How long a render takes and how much it costs in AI assistant usage for a film of a few minutes","Whether the whole chain (uv, ffmpeg, headless Chromium) installs cleanly on Windows","What the optional voice and image features send to outside services, and what they cost","Disk use, which the checks could not measure because the releases carry only sample videos"]},"zh-Hans":{"summaryShort":"面向 AI 编程助手的一个 skill，能写代码把艺术风格变成动画和讲解视频片段。","summary":"huashu-art-motion 是 AI 编程助手（Claude Code、Codex 等）的一个扩展：你描述一段视频，它就写代码来绘制并让每个场景动起来。它自带 36 种艺术风格的配方（岩洞壁画、梵高、包豪斯、8 位像素等）和 9 套从知名讲解类频道测得的叙事语法（仓库只保存测得的参数，不含它们的画面），另有场景代码和质量检查。你通过终端运行它，需要 uv（Python）、ffmpeg 和无界面的 Chromium。几乎所有文字材料都是中文。","easeRationale":"在三个系统上都是要用到多个工具的终端操作：README.md 列出了 uv、ffmpeg 和 Playwright Chromium（\"uv run --with playwright playwright install chromium\"），还需要一个 AI 编程助手来驱动这个 skill，渲染则用类似 \"uv run --with playwright python scripts/engine/render.py --spec ... --out finance.mp4\" 的命令。Windows 是三者中测试最少的：致谢部分记录了一个只在 Windows 上出现的、与渲染服务并行加载有关的错误。","costNote":"这个 skill 免费（MIT）。它在 AI 编程助手里运行，而这类助手大多要付费；可选的语音和图像功能会用到外部服务，需要你自己的账号。","notVerified":["你得到的视频是否和 README 里的样片一样，那些样片是作者用自己的素材做的","渲染要花多长时间，以及做一部几分钟的片子会消耗多少 AI 助手的用量","整条工具链（uv、ffmpeg、无界面 Chromium）能否在 Windows 上顺利安装","可选的语音和图像功能会向外部服务发送什么，以及费用多少","磁盘占用，检查时无法测量，因为已发布的版本只带样片视频"]}}},{"id":"R_kgDOSG2mmQ","slug":"alchaincyf__huashu-design","name":"Huashu Design","owner":"alchaincyf","fullName":"alchaincyf/huashu-design","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"low","signals":[{"code":"agent_runs_scripts_and_fetches_web_assets","severity":"info","evidence":"该技能自带由助手在你机器上运行的脚本：`scripts/ref_board.py` 对参考网站截图，`scripts/render-video.js` 驱动浏览器渲染 MP4，`scripts/learn.py` 记录你的偏好笔记，`scripts/design_lint.py` 检查结果。品牌协议会从某公司官方页面下载 logo 和色彩，原型还会“真图从 Wikimedia/Met/Unsplash 取”（从 Wikimedia、Met 和 Unsplash 取真实图片），所以这些图片和 logo 能否使用由你自己判断。","source":"ai","title":"智能体运行脚本并获取网络资源"},{"code":"optional_cloud_keys_send_content_to_bytedance","severity":"info","evidence":".env.example 要求填写 `DOUBAO_TTS_API_KEY` 和 `DOUBAO_TTS_VOICE_ID`，用于字节跳动 openspeech 语音服务（包括一个克隆语音 ID），以及用于火山引擎 Ark 的 `ARK_API_KEY`，供 `scripts/cloud/ai-review-video.py` 使用。README 说这些云端部分是可选的，首次使用需要显式加 `--yes`，主流程在本地运行不需要任何密钥。","source":"ai","title":"可选的云端密钥会把内容发给字节跳动"},{"code":"some_bundled_assets_not_mit","severity":"info","evidence":"README 的许可部分说明：`assets/style-specimens/xhs/_kit/stickers/` 中的演示贴纸来自 huashu-bird-sticker，使用 PolyForm Shield 1.0.0 许可，“只用来演示版式”，在你的作品中必须替换成自己的素材。","source":"ai","title":"部分内置素材不是 mit 许可"},{"code":"quality_claims_tested_only_by_the_author","severity":"info","evidence":"这些说法都是项目方自己的测量结果：“不是「AI 做的还行」那种水平——是看起来像大厂设计团队做的”，一次 A/B 测试显示“v2 的稳定性方差比 v1 低 5 倍”，以及三轮被称为“内部测试”的盲选。README 在最后更坦诚：“这是一个 80 分的 skill”，没有可逐层编辑导出到 Figma 或 Keynote 的功能，在没有品牌素材时质量会降到“60–65 分”。","source":"ai","title":"质量说法只经作者自己测试"},{"code":"built_from_another_product_prompt","severity":"info","evidence":"起源部分说作者让一个智能体拆解了 Anthropic 的 Claude Design，“包括社区流传的系统提示词”，并将其提炼成这个技能，品牌素材协议也被描述为借鉴自那里。","source":"ai","title":"基于另一产品的提示词构建"},{"code":"chinese_first_documentation","severity":"info","evidence":"README.md 默认是中文，英文版是 README.en.md，包描述和更新日志都是中文，部分风格库专门针对中国和东亚设计（还有八款小红书排版）。社区翻译版本列明由其他人维护，并有各自的许可条款。","source":"ai","title":"文档以中文为主"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg","other"],"buildTools":["node","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Volcengine Doubao TTS (optional voice-over)","Volcengine Ark (optional AI review of the rendered video)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["design_cad","ai_skills","office"],"alternativeTo":[{"slug":"gamma","name":"Gamma","confirmed":true},{"slug":"figma","name":"Figma","confirmed":true},{"slug":"canva","name":"Canva","confirmed":true}],"declaredAlternatives":[],"stars":29125,"downloads":5948652,"downloadsWeek":0,"downloadsDays":5,"growth":454,"growthDays":6,"history":[{"date":"2026-11-05","stars":28795},{"date":"2026-11-06","stars":28890},{"date":"2026-11-07","stars":28944},{"date":"2026-11-08","stars":28977},{"date":"2026-11-09","stars":29044},{"date":"2026-11-10","stars":29125}],"firstPublishedAt":"2026-11-04T22:32:54.801Z","commitSha":"b2460b031c33ff32d2bf3b97263527daa231e4de","verifiedAt":"2026-11-04T22:22:53.270Z","lastCommitAt":"2026-11-10T00:15:18Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill that lets an AI coding assistant design web pages, app prototypes, slides, infographics and short animations.","summary":"A design skill you add to an AI coding assistant (Claude Code, Codex, Cursor, DeepSeek Harness and others) with one command. You describe what you need in a sentence and it builds the result in HTML: a clickable web page or phone prototype, a slide deck it can export as an editable PPTX or PDF, an infographic, a set of vertical social images, or a short MP4 or GIF. It offers three visual directions to choose from, drawing on a library of 92 styles with samples, and keeps notes about your taste in a folder on your computer. You need Node, a browser and, for video, FFmpeg.","easeRationale":"There is no program to open: on any of the three systems you need an AI coding assistant already working in a terminal, then `npx skills add alchaincyf/huashu-design` (which needs Node), and the scripts the skill runs want a Chrome or Edge on the machine plus FFmpeg for video and GIF export. The README also warns that older versions of the `skills` CLI copy only one file and you have to upgrade or clone the repository by hand.","costNote":"The skill is MIT and free for commercial use since 2026-05-14. The work itself is done by the AI assistant you already pay for, and the README's comparison with Claude Design says this route spends API credit instead of a subscription quota. The optional voice-over and video review use your own Volcengine keys.","notVerified":["Whether your assistant produces results like the examples in the README","How much the AI assistant charges for one design job","Whether the exported PPTX stays editable in your version of PowerPoint","What the scripts write into ~/.huashu-design/ and whether anything leaves your machine","Whether the images and logos the skill downloads may be used in your work","Whether it works on Windows, where the shell scripts in the toolchain were written for Unix shells"]},"zh-Hans":{"summaryShort":"一个让 AI 编程助手设计网页、应用原型、幻灯片、信息图和短动画的技能。","summary":"一个一条命令就能添加到 AI 编程助手（Claude Code、Codex、Cursor、DeepSeek Harness 等）上的设计技能。你用一句话描述需求，它就用 HTML 构建结果：可点击的网页或手机原型、可导出为可编辑 PPTX 或 PDF 的幻灯片、信息图、一组竖版社交媒体图片，或一段短的 MP4 或 GIF。它提供三种可选的视觉方向，背后是一个有 92 种风格和样例的库，并在你电脑上的一个文件夹里记录你的偏好。你需要 Node、浏览器，以及用于视频的 FFmpeg。","easeRationale":"没有程序可以直接打开：在三种系统上都需要一个已经在终端里工作的 AI 编程助手，然后运行 `npx skills add alchaincyf/huashu-design`（需要 Node），技能运行的脚本还需要机器上装有 Chrome 或 Edge，以及用于视频和 GIF 导出的 FFmpeg。README 还提醒，旧版本的 `skills` CLI 只会复制一个文件，需要手动升级或克隆整个仓库。","costNote":"该技能采用 MIT 许可，自 2026-05-14 起可免费商用。实际工作由你已经付费使用的 AI 助手完成，README 与 Claude Design 的对比说这条路线消耗的是 API 额度而不是订阅配额。可选的语音旁白和视频审核功能使用你自己的 Volcengine 密钥。","notVerified":["你的助手产出的结果是否和 README 中的示例一样好","AI 助手完成一次设计任务收费多少","导出的 PPTX 在你的 PowerPoint 版本中是否仍可编辑","脚本往 ~/.huashu-design/ 里写了什么，以及是否有内容离开你的电脑","技能下载的图片和 logo 是否可以用在你的作品里","是否能在 Windows 上运行，因为工具链里的 shell 脚本是为 Unix shell 编写的"]}}},{"id":"R_kgDOVOW1Yw","slug":"alchaincyf__huashu-motion","name":"Huashu Motion","owner":"alchaincyf","fullName":"alchaincyf/huashu-motion","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"mac":"D"},"risk":"low","signals":[{"code":"very_new_repository_fast_star_growth","severity":"info","evidence":"仓库创建于 2026-10-18，已有 1801 个星标，最近一次推送在 2026-10-22；可供判断的历史记录很少。","source":"ai","title":"非常新的仓库，星标增长很快"},{"code":"adds_watermark_to_exports","severity":"info","evidence":"README：对于超过 4 秒的视频，最后 1.2 秒会在空白角落显示一行小字 \"Made with huashu-motion\"；README 说明你可以让智能体把它关掉。","source":"ai","title":"导出内容会加水印"},{"code":"downloads_extra_files_at_run_time","severity":"info","evidence":"README \"需要什么\"：`om samples build` 会下载约 55MB 的乐器采样，Kokoro 语音首次使用时需要一个约 325MB 的模型。","source":"ai","title":"运行时会下载额外文件"},{"code":"third_party_license_carve_outs","severity":"info","evidence":"README 的许可部分：代码和文档采用 MIT，但随附的 GSAP 保持 GSAP Standard License，字体为 SIL OFL 1.1，runtime/clip/lib/toon.js 中的 \"豆子花叔\" 角色绘图不在 MIT 授权范围内。","source":"ai","title":"第三方许可的例外"}],"platforms":["mac"],"tools":["node","ffmpeg","other"],"buildTools":["node","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["AI coding assistant or model API (Claude Code, Codex, DeepSeek and similar)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","video","design_cad"],"alternativeTo":[{"slug":"after-effects","name":"After Effects","confirmed":true},{"slug":"claude-motion","name":"Claude Motion","confirmed":true},{"slug":"premiere-pro","name":"Premiere Pro","confirmed":true}],"declaredAlternatives":[],"stars":2607,"downloads":916,"downloadsWeek":0,"downloadsDays":7,"growth":362,"growthDays":7,"history":[{"date":"2026-10-27","stars":1801},{"date":"2026-10-28","stars":1853},{"date":"2026-10-29","stars":1893},{"date":"2026-10-30","stars":1924},{"date":"2026-10-31","stars":1940},{"date":"2026-11-01","stars":2015},{"date":"2026-11-02","stars":2122},{"date":"2026-11-03","stars":2245},{"date":"2026-11-04","stars":2346},{"date":"2026-11-05","stars":2410},{"date":"2026-11-06","stars":2444},{"date":"2026-11-07","stars":2478},{"date":"2026-11-08","stars":2520},{"date":"2026-11-09","stars":2566},{"date":"2026-11-10","stars":2607}],"firstPublishedAt":"2026-10-27T18:40:27.758Z","commitSha":"52467f99c602b76ba86d511d37acdfe93fa1e16e","verifiedAt":"2026-10-28T01:36:10.481Z","lastCommitAt":"2026-11-10T00:15:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ask your AI coding assistant to make a short product launch video with music, then click on the picture to change parts of it.","summary":"A skill pack plus command-line tool for AI coding assistants such as Claude Code or Codex. You describe the product you want a launch video for; the assistant writes the animation as a web page, writes a music score that is played locally from instrument samples, and opens an editor in your browser where you click on the frame and type what to change. Exports MP4, MOV with transparency, WebM or GIF. You need a coding assistant, Node 22, Chrome and FFmpeg, and you work through the terminal. The author says it was only tested on macOS.","easeRationale":"On macOS, the platform the author says he actually tested, you install it with one command but you also need an AI coding assistant, Node 22 or newer, Chrome and FFmpeg, and the whole job runs through terminal commands, so the grade is D. There are no ready-made binaries in the 5 releases.","costNote":"The skill itself is free and MIT-licensed, but it only works through an AI coding assistant you pay for. The author reports one 22-second video cost about CNY 1.95 in DeepSeek model usage.","notVerified":["Whether it runs on Windows or Linux: the README says it was only tested on macOS so far.","Whether the example videos shown in the README were produced exactly as described.","Real disk use after downloading the instrument samples and the voice models.","The real cost per video, which depends on the AI model you point it at.","Whether a video made from a single sentence reaches the quality of the samples, which the README says often took several rounds of feedback."]},"zh-Hans":{"summaryShort":"让 AI 编程助手制作一段带音乐的产品发布短视频，然后点击画面来修改其中的部分内容。","summary":"这是一个技能包加命令行工具，供 Claude Code 或 Codex 等 AI 编程助手使用。你描述想要做发布视频的产品；助手会把动画写成一个网页，用乐器采样在本地生成并播放一段配乐，并在你的浏览器里打开一个编辑器，你点击某一帧并输入要修改的内容。可导出 MP4、带透明通道的 MOV、WebM 或 GIF。你需要一个编程助手、Node 22、Chrome 和 FFmpeg，整个过程都在终端里操作。作者说目前只在 macOS 上测试过。","easeRationale":"在作者说自己真正测试过的 macOS 上，一条命令就能安装，但你还需要一个 AI 编程助手、Node 22 或更新版本、Chrome 和 FFmpeg，而且整个流程都要靠终端命令完成，所以评级是 D。已发布的 5 个版本里没有现成的二进制文件。","costNote":"这个技能本身是免费的，采用 MIT 许可，但它只能通过你付费使用的 AI 编程助手运行。作者报告说，一段 22 秒的视频花费了大约 1.95 元人民币的 DeepSeek 模型使用费。","notVerified":["是否能在 Windows 或 Linux 上运行：README 说目前只在 macOS 上测试过。","README 里展示的示例视频是否就是按描述的方式制作出来的。","下载乐器采样和语音模型之后实际占用的磁盘空间。","每段视频的实际花费，这取决于你所使用的 AI 模型。","仅凭一句话生成的视频能否达到示例的质量，README 说这些示例常常经过了好几轮反馈调整。"]}}},{"id":"R_kgDOSjGjSg","slug":"aldoprianandi__ig-autodm-worker","name":"ig-autodm-worker","owner":"aldoprianandi","fullName":"aldoprianandi/ig-autodm-worker","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"web":"E"},"risk":"attention","signals":[{"code":"needs_meta_app_approval","severity":"attention","evidence":"README.md Get started: \"Remote sending also needs a Cloudflare account, a Meta developer app, and suitable Instagram API access for a Business or Creator account\",\"Not included\"一栏列出了\"Guaranteed delivery, App Review approval, or unlimited free usage\"。没有Meta的审核通过,自动化就无法给其他人发送消息。","source":"ai","title":"需要Meta应用审核通过"},{"code":"stores_instagram_access_token","severity":"info","evidence":"README.md在运营者可控项中列出了\"encrypted stored tokens, and scheduled refresh\"。你Instagram账户的一个长期有效的Meta令牌会存放在你部署的Cloudflare项目里,维护好这个账户和它的密钥就是你自己的责任。","source":"ai","title":"保存Instagram访问令牌"},{"code":"platform_rules_apply_to_automated_dms","severity":"info","evidence":"README.md说明它不适用于:\"Scraping, cold DMs, or private Instagram APIs\",以及\"The service does not cold-message arbitrary users\"。自动发送的消息仍然要遵守Meta的消息规则,而你如何配置活动决定了你的账户是否守规矩。","source":"ai","title":"平台规则适用于自动化私信"},{"code":"small_single_maintainer_project","severity":"info","evidence":"6个star、5个fork、10个未解决issue,两个发布版本(v0.1.0和v0.2.0),package.json标记为\"private\": true,并使用了一个预发布依赖(\"miniflare\": \"5.20261006.0-alpha\")。这是一个人做的模板,还处于早期阶段。","source":"ai","title":"单人维护的小型项目"}],"platforms":["web"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram Graph API (Business or Creator account)","Cloudflare (Workers, D1, Queues)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":6,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":6},{"date":"2026-11-09","stars":6},{"date":"2026-11-10","stars":6}],"firstPublishedAt":"2026-11-07T22:56:16.667Z","commitSha":"e639b023d31adc1c94df2dacecd776542e47037d","verifiedAt":"2026-11-07T22:41:41.812Z","lastCommitAt":"2026-10-08T07:50:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Sends an Instagram DM with your link to anyone who comments a chosen word on your post, on your own servers.","summary":"A template you deploy to your own Cloudflare account that watches one Instagram Business or Creator account: when someone comments a keyword on a post you chose, it replies privately with a button that delivers your link or guide. It has an admin console for campaigns, keyword matching, a rate limiter and a kill switch, and uses only Meta's official API, with no Instagram password or scraping. To set it up you need Node.js 22, Git, a Cloudflare account with D1 and Queues, and a Meta developer app with Instagram access.","easeRationale":"There is nothing to download and open: the console only exists after you deploy it yourself. The README asks for Node.js 22, npm, Git, copying and editing two config files, creating D1 databases and queues with wrangler, and setting up a Meta developer app with Instagram API access, which the project itself describes as a fit only for people who \"can manage a Meta app and Cloudflare deployment\".","costNote":"No fee for the software; you pay for whatever your Cloudflare usage goes over the free tier.","notVerified":["Whether Meta grants the Instagram access the project needs for your account, and how long that takes.","What it really costs on Cloudflare once comments, queue jobs and retries pile up.","Whether comments arrive reliably by webhook, and how much the fallback polling (10 media per minute) misses.","Whether the admin console and the campaign features work as the screenshot shows, since there is no hosted demo.","How the project behaves when Meta changes its API or your token expires."]},"zh-Hans":{"summaryShort":"在你自己的服务器上,给在帖子下评论指定关键词的人发送带你链接的Instagram私信。","summary":"一个你部署到自己Cloudflare账户的模板,监控一个Instagram商业版或创作者账户:当有人在你选定的帖子下评论某个关键词时,它会私下回复一个按钮,发送你的链接或指南。它有活动管理后台、关键词匹配、限流器和紧急停止开关,只使用Meta官方API,不用Instagram密码,也不做爬取。搭建它需要Node.js 22、Git、一个带D1和Queues的Cloudflare账户,以及一个有Instagram权限的Meta开发者应用。","easeRationale":"没有可以直接下载打开的东西:管理后台只有在你自己部署之后才存在。README要求Node.js 22、npm、Git,复制并编辑两个配置文件,用wrangler创建D1数据库和队列,还要设置一个有Instagram API权限的Meta开发者应用,项目自己也说这只适合能\"can manage a Meta app and Cloudflare deployment\"的人。","costNote":"软件本身免费;你只为超出Cloudflare免费额度的用量付费。","notVerified":["Meta是否会为你的账户授予项目所需的Instagram权限,以及需要多久。","一旦评论、队列任务和重试积累起来,在Cloudflare上实际要花多少钱。","评论是否能通过webhook可靠送达,备用轮询(每分钟10条内容)会漏掉多少。","管理后台和活动功能是否像截图展示的那样工作,因为没有在线演示。","当Meta更改其API或你的令牌过期时,项目会如何表现。"]}}},{"id":"R_kgDOK-ZNOA","slug":"ali-vilab__dreamtalk","name":"DreamTalk","owner":"ali-vilab","fullName":"ali-vilab/dreamtalk","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：scipy: PYSEC-2023-102, PYSEC-2023-114；scikit-learn: GHSA-jw8x-6495-233v, PYSEC-2024-110；pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h；numpy: GHSA-fpfv-jqm9-f5jm；opencv-python: GHSA-qr4w-53vh-m672, PYSEC-2023-183","source":"check","title":"已知漏洞"},{"code":"model_files_not_publicly_available","severity":"attention","evidence":"README「Download Checkpoints」：\"In light of the social impact, we have ceased public download access to checkpoints. If you want to obtain the checkpoints, please request it by emailing mayf18@mails.tsinghua.edu.cn\"——没有这些文件，代码什么也生成不了。","source":"ai","title":"模型文件未公开提供"},{"code":"research_and_non_commercial_use_only","severity":"attention","evidence":"README 的 Disclaimer：\"This method is intended for RESEARCH/NON-COMMERCIAL USE ONLY\"，邮件申请\"implies your consent to use the provided method solely for academic research purposes\"，尽管代码文件里写的是 MIT。","source":"ai","title":"仅限研究和非商业用途"},{"code":"makes_a_real_persons_face_say_anything","severity":"attention","evidence":"这个工具可以根据任意音频为\"out-of-domain portraits\"生成动画，包括\"speech in multiple languages\"，所以一张照片加一段语音录音，就能生成一个人说出他从未说过的话的视频；作者以\"social impact\"为由停止了模型下载。","source":"ai","title":"能让真实人物的脸说出任何话"},{"code":"abandoned_and_built_on_end_of_life_python","severity":"info","evidence":"最后一次提交是 2024-01-06（距检查时 1034 天前），还有 46 个未关闭的 issue，安装命令是\"conda create -n dreamtalk python=3.7.0\"，搭配 PyTorch 1.8——这些都是已经不再获得支持的版本。","source":"ai","title":"已被放弃，且基于已停止维护的 python 版本构建"},{"code":"extra_steps_to_use_your_own_video_as_reference","severity":"info","evidence":"README：风格和姿态文件是\"3DMM parameter sequences extracted from reference videos\"，如果要用自己的视频，必须按照 PIRenderer 和 FOMM 的视频预处理脚本操作，视频需为 25 FPS 并裁剪为 256x256。","source":"ai","title":"使用自己的视频作为参考需要额外步骤"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg"],"buildTools":["python","ffmpeg"],"disk":{"minMB":160,"maxMB":420,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":false}],"declaredAlternatives":[],"stars":1788,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":1788},{"date":"2026-11-07","stars":1788},{"date":"2026-11-08","stars":1788},{"date":"2026-11-09","stars":1788},{"date":"2026-11-10","stars":1788}],"firstPublishedAt":"2026-11-05T22:43:48.251Z","commitSha":"ba798b0e586d96a63f61050f7e8e17287ff3c859","verifiedAt":"2026-11-05T22:28:05.879Z","lastCommitAt":"2024-01-06T05:24:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that makes a photo talk with expressions from an audio clip; the model files are not public.","summary":"DreamTalk is the code from an Alibaba research paper. Given one front-facing photo and an audio file, it generates a video of that face speaking, copying the speaking style and head movement of a reference clip. It runs from the command line and the result is a 256x256 video. The model files are no longer offered for download: the README asks you to email the author and says the method is for academic research only. You build a conda environment with Python 3.7, PyTorch, ffmpeg and dlib; a graphics card is advised, with a slower CPU option.","easeRationale":"The same E everywhere, and Linux is the realistic route: you create a conda environment with Python 3.7, install PyTorch 1.8 with CUDA 11.1 plus dlib and ffmpeg, then run a long python command with paths to audio, a 3DMM style file and a pose file. On top of that the model files are not downloadable, so even a developer cannot get a result without first emailing the author.","costNote":"No price. The code is MIT, but the README's disclaimer says the method is for \"RESEARCH/NON-COMMERCIAL USE ONLY\".","notVerified":["Whether the author still answers the email and sends the model files","How the result looks with your own photo and voice, and whether the mouth really matches","Whether the Python 3.7 and PyTorch 1.8 environment still installs on a current computer","How long one video takes on a CPU, the only option without an NVIDIA card","Real disk use, since the model files are not published and cannot be measured"]},"zh-Hans":{"summaryShort":"这是研究代码，能让一张照片根据音频说话并带上相应表情；模型文件没有公开。","summary":"DreamTalk 是阿里巴巴一篇研究论文附带的代码。给定一张正面照片和一段音频文件，它会生成一段这张脸说话的视频，并复刻参考视频里的说话风格和头部动作。它通过命令行运行，输出的是 256x256 的视频。模型文件已经不再提供下载：README 要求你给作者发邮件，并说明这个方法仅用于学术研究。你需要搭建一个带 Python 3.7、PyTorch、ffmpeg 和 dlib 的 conda 环境；建议使用显卡，也有速度较慢的 CPU 选项。","easeRationale":"三个系统的评分都一样，是 E，而 Linux 是现实中唯一可行的路径：你需要创建一个带 Python 3.7 的 conda 环境，安装带 CUDA 11.1 的 PyTorch 1.8，再加上 dlib 和 ffmpeg，然后运行一条很长的 python 命令，传入音频路径、一个 3DMM 风格文件和一个姿态文件。除此之外，模型文件本身无法下载，所以即便是开发者，在给作者发邮件之前也得不到任何结果。","costNote":"没有定价。代码是 MIT 许可，但 README 的免责声明说这个方法仅限于\"RESEARCH/NON-COMMERCIAL USE ONLY\"。","notVerified":["作者是否仍会回复邮件并发送模型文件","用自己的照片和声音生成的效果如何，嘴型是否真的对得上","Python 3.7 和 PyTorch 1.8 的环境在现在的电脑上是否还能安装","在唯一不需要 NVIDIA 显卡的 CPU 选项下，生成一段视频要花多长时间","真实的磁盘占用，因为模型文件没有公开，也就无法测量"]}}},{"id":"R_kgDOKp0fCQ","slug":"ali-vilab__vgen","name":"VGen","owner":"ali-vilab","fullName":"ali-vilab/VGen","type":"framework_infra","status":"active","grade":"A","easeByPlatform":{"linux":"D","web":"A"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub 没有检测到任何许可证，许可证文件、LICENSES 文件夹和 winget manifest 里也都没有。","source":"check","title":"没有许可证"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：transformers: GHSA-282v-666c-3fvg, GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm；opencv-python: GHSA-qr4w-53vh-m672, PYSEC-2023-183；opencv-python-headless: GHSA-jh2j-j4j9-crg3, PYSEC-2023-184；pytorch-lightning: GHSA-2vj5-px25-gjrp, GHSA-4cv3-v7pv-rfhf, GHSA-75m9-98v2-hjpm；gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2","source":"check","title":"已知漏洞"},{"code":"models_and_packages_come_from_outside_github","severity":"info","evidence":"README 里所有的 checkpoint 都是从 ModelScope 下载的（`snapshot_download('damo/I2VGen-XL', ...)`、`iic/HiGen`、`iic/tf-t2v`、`iic/InstructVideo`），依赖包也是从镜像源安装的：\"pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple\"。","source":"ai","title":"模型和依赖包来自 github 之外"},{"code":"stopped_being_updated","severity":"info","evidence":"最后一次提交是 2025-01-10（距检查时 664 天前），还有 116 个未关闭的 issue，固定的版本也都很旧：torch 1.12 搭配 CUDA 11.3、xformers 0.0.13、pytorch-lightning 1.4.2、gradio 3.39，以及一个过时的 triton 构建。","source":"ai","title":"已经停止更新"},{"code":"planned_features_never_shipped","severity":"info","evidence":"TODO 列表里仍有未勾选的项目——\"Release the code and pretrained models of HumanDiff\"、\"Release models optimized specifically for the human body and faces\"——\"(8) Other methods\"一节只写了\"In preparation!!\"。","source":"ai","title":"计划中的功能从未发布"},{"code":"author_states_weak_spots_of_the_model","severity":"info","evidence":"README：\"we find that the current model performs inadequately on anime images and images with a black background due to the lack of relevant training data\"，超分辨率模型\"only supports 32-frame input\"。","source":"ai","title":"作者说明了模型的弱点"}],"platforms":["linux","web"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":310,"maxMB":870,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["video","local_ai","development"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"ltx-studio","name":"LTX Studio","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":3152,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":3153},{"date":"2026-11-07","stars":3153},{"date":"2026-11-08","stars":3153},{"date":"2026-11-09","stars":3152},{"date":"2026-11-10","stars":3152}],"firstPublishedAt":"2026-11-05T22:43:48.260Z","commitSha":"4014d1fae4ac4a35e8e8442123611e913001399e","verifiedAt":"2026-11-05T22:29:07.914Z","lastCommitAt":"2025-01-10T09:09:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Alibaba's research code for turning a text prompt or a photo into a short video, with models you download yourself.","summary":"VGen is the research codebase of Alibaba's Tongyi Lab. It gathers several video generation methods in one place — I2VGen-XL (photo to video), HiGen and TF-T2V (text to video), DreamVideo (your own subject or motion), VideoLCM and others — and you drive each one by editing a YAML file and running inference.py, or by opening a local Gradio page. It can also train and fine-tune models. For a first look without installing anything, the authors keep browser demos on HuggingFace and ModelScope. The repository has no license file.","easeRationale":"The best way in is the browser: the README links demos of I2VGen-XL run by the authors on HuggingFace Spaces and ModelScope, where you upload a picture and get a video without installing anything. Doing it on your own machine is a D and really a Linux job: a conda environment with Python 3.8, PyTorch 1.12 built for CUDA 11.3, ffmpeg installed with apt-get, model files pulled from ModelScope and a YAML config per model.","costNote":"No price. There is also no license file, so the terms for using the code and the videos it makes are not stated.","notVerified":["Whether the HuggingFace and ModelScope demos are still online and how long the queue is","How much graphics memory each model needs, which the README never states","The real disk space taken by the checkpoints downloaded from ModelScope","Whether the Python 3.8 and CUDA 11.3 environment still installs on a current machine","What you are allowed to do with the generated videos, since there is no license file"]},"zh-Hans":{"summaryShort":"阿里巴巴的研究代码，能把一段文字提示或一张照片变成一段短视频，模型需要自己下载。","summary":"VGen 是阿里巴巴通义实验室的研究代码库。它把多种视频生成方法集中在一处——I2VGen-XL（照片生成视频）、HiGen 和 TF-T2V（文字生成视频）、DreamVideo（用你自己的主体或动作）、VideoLCM 等——你通过编辑一个 YAML 文件并运行 inference.py，或者打开本地的 Gradio 页面来操作每一种方法。它也可以用来训练和微调模型。如果只想先看看效果而不想安装任何东西，作者在 HuggingFace 和 ModelScope 上提供了网页端演示。这个仓库没有许可证文件。","easeRationale":"最好的入门方式是浏览器：README 链接了作者在 HuggingFace Spaces 和 ModelScope 上运行的 I2VGen-XL 演示，你只需上传一张图片就能得到视频，不用安装任何东西。在自己电脑上运行的评分是 D，而且基本上只能在 Linux 上做：需要带 Python 3.8 的 conda 环境、为 CUDA 11.3 编译的 PyTorch 1.12、用 apt-get 安装的 ffmpeg、从 ModelScope 下载的模型文件，以及每个模型对应的一个 YAML 配置。","costNote":"没有定价。也没有许可证文件，所以没有说明使用这段代码和它生成的视频的条款。","notVerified":["HuggingFace 和 ModelScope 上的演示是否仍在线，排队要等多久","每个模型需要多少显存，README 从来没有说明","从 ModelScope 下载的模型文件实际占用多少磁盘空间","Python 3.8 和 CUDA 11.3 的环境在现在的电脑上是否还能安装","生成的视频你能做什么用，因为没有许可证文件说明"]}}},{"id":"R_kgDOVMseiw","slug":"amosblomqvist__learnharness","name":"LearnHarness","owner":"amosblomqvist","fullName":"amosblomqvist/learnharness","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"18 天内获得 5194 个星标","source":"check","title":"年轻的仓库却有很多星标"},{"code":"windows_installer_unsigned","severity":"attention","evidence":"README.md：\"Windows (x64)：运行安装程序。目前它未签名；Windows 可能会显示'未知发布者'或 SmartScreen 警告。\" macOS 版本则被描述为已签名并经过公证。","source":"ai","title":"windows 安装程序未签名"},{"code":"your_answers_go_to_the_model_provider","severity":"info","evidence":"README.md：\"你的聊天记录、图谱、设置和 API 密钥都保存在你的机器上\"，但同时也说\"模型请求会发送该任务所需的上下文；网络搜索和语音输入功能也会联系外部服务。本地存储并不意味着离线处理。\" 你边想边说输入的内容，会发送给你选择的那个提供方。","source":"ai","title":"你的回答会发给模型提供方"},{"code":"cost_depends_on_your_provider","severity":"info","evidence":"README.md：\"实时费用表只估算已知的 API 费用，不包括订阅、CLI 或自定义端点的费用\"，更长的提示缓存保留时间\"会增加缓存写入的费用\"。一次长时间的学习会话可能会花掉真金白银。","source":"ai","title":"费用取决于你的提供方"},{"code":"early_version","severity":"info","evidence":"这个仓库创建于 2026-10-16，最新发行版是 v0.4.1，有 375 个未关闭的 issue，大约两周内发布了 10 个版本：这个应用还在 1.0 版本之前，而且变化很快。","source":"ai","title":"早期版本"},{"code":"updates_itself","severity":"info","evidence":"README.md：\"三个平台都支持应用内更新\"，package.json 中也包含 electron-updater，所以这个应用可以自行下载并安装新版本。","source":"ai","title":"会自动更新"},{"code":"mac_intel_not_covered","severity":"info","evidence":"发行版里只有标记为 arm64 的 Mac 文件（LearnHarness-0.4.1-arm64.dmg、-arm64-mac.zip），README.md 也写着 \"macOS (Apple Silicon)\"，所以较旧的 Intel Mac 没有对应的构建版本。","source":"ai","title":"不支持 mac 的 intel 机型"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":157,"maxMB":472,"basis":"installer"},"apiKeys":["An AI model provider (your own key, a subscription connection, Claude Code, or an OpenAI-compatible endpoint)","Deepgram (only for speaking instead of typing)","A web search service (only for background research)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["productivity","local_ai"],"alternativeTo":[{"slug":"quizlet","name":"Quizlet","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[],"stars":8234,"downloads":161884,"downloadsWeek":57283,"downloadsDays":7,"growth":3042,"growthDays":7,"history":[{"date":"2026-10-27","stars":3102},{"date":"2026-10-28","stars":3620},{"date":"2026-10-29","stars":4020},{"date":"2026-10-30","stars":4300},{"date":"2026-10-31","stars":4541},{"date":"2026-11-01","stars":4740},{"date":"2026-11-02","stars":4973},{"date":"2026-11-03","stars":5192},{"date":"2026-11-04","stars":5351},{"date":"2026-11-05","stars":5534},{"date":"2026-11-06","stars":5690},{"date":"2026-11-07","stars":6154},{"date":"2026-11-08","stars":6435},{"date":"2026-11-09","stars":7575},{"date":"2026-11-10","stars":8234}],"firstPublishedAt":"2026-10-27T18:40:27.769Z","commitSha":"e69e85269c064c6c9c49a8b06ae78c06e79c4a63","verifiedAt":"2026-11-04T01:43:16.295Z","lastCommitAt":"2026-11-10T19:53:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop app where AI tutors you on a hard subject, checks what you already understand and keeps notes between sessions.","summary":"LearnHarness is a desktop program for studying something difficult with AI. It first asks questions to find where your understanding stops, plans a path to your goal, then teaches one reasoning step at a time, asking you to predict and to explain your thinking in your own words. Separate practice sessions pose problems, and what you learn is kept as a knowledge map you can look at and reuse later. Installers exist for Windows, macOS (Apple Silicon) and Linux, but you must plug in your own AI model: a provider key, a Claude Code install or a local model server.","easeRationale":"All three systems have a ready installer (a 157 MB Windows setup, a signed macOS DMG for Apple Silicon, an AppImage or .deb for Linux), and the one extra step everywhere is choosing your models and entering a key, which is why it is a B and not an A. On Linux you also have to make the AppImage executable first, and the Windows installer is unsigned.","costNote":"The app is free (AGPL-3.0) with no subscription of its own, but you pay your AI provider for every session. README: \"No LearnHarness subscription or markup. Billing depends on your provider\".","notVerified":["Whether the teaching actually works better than asking a chatbot directly","How much a session costs with each model provider","Whether everything it says it keeps on your machine really stays there","Whether the knowledge map keeps being useful after many sessions","Real disk use after installing (the installers are 133-174 MB; the checks estimate 157-472 MB)"]},"zh-Hans":{"summaryShort":"一款桌面应用：由 AI 辅导你学习难学的科目，先检查你已经懂了多少，并在多次学习之间保留笔记。","summary":"LearnHarness 是一款借助 AI 学习难题的桌面程序。它先提问，找出你的理解止步于何处，再规划一条通往目标的路径，然后一次只讲一个推理步骤，要你先做预测，并用自己的话解释思路。另有独立的练习环节出题，你学到的内容会保存为一张知识地图，之后可以查看和重复使用。它提供 Windows、macOS（Apple Silicon）和 Linux 的安装包，但你必须接入自己的 AI 模型：服务商的密钥、已安装的 Claude Code，或本地模型服务器。","easeRationale":"三个系统都有现成的安装包（157 MB 的 Windows 安装程序、面向 Apple Silicon 的已签名 macOS DMG、Linux 的 AppImage 或 .deb），各平台唯一多出来的一步是选择模型并填入密钥，所以评级是 B 而不是 A。在 Linux 上还得先把 AppImage 设为可执行，而 Windows 安装程序没有签名。","costNote":"这款应用免费（AGPL-3.0），自身没有订阅，但每次学习你都要向自己的 AI 服务商付费。README：\"No LearnHarness subscription or markup. Billing depends on your provider\"。","notVerified":["这种教学方式是否真的比直接问聊天机器人效果更好","用各家模型服务商进行一次学习要花多少钱","它声称保存在你电脑上的内容是否真的都留在本机","知识地图在经过多次学习之后是否仍然有用","安装后的实际磁盘占用（安装包为 133-174 MB；检查估计为 157-472 MB）"]}}},{"id":"R_kgDOVXel6Q","slug":"andyhuo520__digital-human-koubo","name":"Digital Human Koubo","owner":"andyhuo520","fullName":"andyhuo520/digital-human-koubo","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"makes_talking_videos_of_a_real_face","severity":"attention","evidence":"这个工具能用一张照片造出一位可以念任何脚本的主播（\"Digital human from one photo\"、\"Pick a photo: frontal face, even light\"）。README.md 的 \"Responsible use\" 一节要求你 \"Use only your own face/voice or ones you have explicit permission to use. Never impersonate a real person or creator\"。很多国家对未经许可使用他人的面孔或声音都有相关规定。","source":"ai","title":"用真人面孔生成说话视频"},{"code":"paid_credits_for_every_video","severity":"attention","evidence":"README.md 的费用表：27 秒广告 482 点，4:36 的教程约 3,250 点，按 \"Grok Video 1.5 - 720p 17 credits/s\" 计。助手会先报价并等你同意，但没有 Renoise 的付费余额就什么也生成不了。","source":"ai","title":"每条视频都要花付费点数"},{"code":"editing_engine_needs_a_company_licence","severity":"attention","evidence":"README.md：\"Remotion is used for rendering and has its own license: free for individuals and companies of up to 3 people; larger companies need a company license\"。本仓库的 MIT 许可并不涵盖这一点。","source":"ai","title":"剪辑引擎对公司需要单独许可"},{"code":"tested_with_one_agent_only","severity":"info","evidence":"README.md 的 \"Supported agents\" 表里只有 Claude Code 被标为 \"Tested - this repo was built with it\"；Codex 是 \"Expected to work\"，OpenClaw 和 WorkBuddy 是 \"Should work\"，而 Claude desktop 在剪辑步骤上只是 \"sandbox only\"。","source":"ai","title":"只用一个助手测试过"},{"code":"imitates_known_creators_formats","severity":"info","evidence":"48 个包装组件是 \"measured frame-by-frame from 25 videos by top creators (MrBeast, Mark Rober, Ryan Trahan, MKBHD, ... Vox, Kurzgesagt)\"。README.md 称风格 A \"learns only the structure and editing grammar\"，并且 \"never uses a real creator's face, voice, set or lines\"。","source":"ai","title":"模仿知名创作者的形式"},{"code":"sample_ids_belong_to_the_author","severity":"info","evidence":"README.md：\"The material: IDs in this repo belong to the author's account and won't work on yours - that's expected.\" 在做第一条视频之前，你必须先创建自己的主播并上传自己的参考图片。","source":"ai","title":"示例 ID 属于作者本人"},{"code":"known_quality_problems","severity":"info","evidence":"README.md 的 \"Known pitfalls\" 列表包括：视频模型自带的硬字幕、声音落后于嘴型（\"0.38 s late by 4:43 over 44 clips\"），以及换用其他声音时约 100 毫秒的口型偏差。","source":"ai","title":"存在已知的质量问题"}],"platforms":["windows","mac","linux"],"tools":["python","node","ffmpeg","other"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Renoise (video and image generation, paid credits, browser login)","Xiaomi MiMo TTS (only for a designed or cloned voice)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","ai_skills","social_marketing"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":false}],"declaredAlternatives":[{"target":"resilio-sync","stance":"claims","evidence":"pause-split clauses without the script; `patchwords.py` replaces just that clause with a voice-clone take stretched to the original length, so lip sync stays intact (don't re-voice the whole film)"}],"stars":1915,"downloads":8157,"downloadsWeek":1980,"downloadsDays":7,"growth":157,"growthDays":7,"history":[{"date":"2026-10-27","stars":1081},{"date":"2026-10-28","stars":1309},{"date":"2026-10-29","stars":1464},{"date":"2026-10-30","stars":1554},{"date":"2026-10-31","stars":1588},{"date":"2026-11-01","stars":1653},{"date":"2026-11-02","stars":1726},{"date":"2026-11-03","stars":1758},{"date":"2026-11-04","stars":1776},{"date":"2026-11-05","stars":1807},{"date":"2026-11-06","stars":1851},{"date":"2026-11-07","stars":1857},{"date":"2026-11-08","stars":1875},{"date":"2026-11-09","stars":1897},{"date":"2026-11-10","stars":1915}],"firstPublishedAt":"2026-10-27T18:40:27.779Z","commitSha":"f661a6c047bb74ea306ef58aeb1d461a52c01871","verifiedAt":"2026-11-02T03:05:51.505Z","lastCommitAt":"2026-10-31T12:05:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns one photo and one sentence into a finished talking-head video with an AI presenter, captions and editing.","summary":"Digital Human Koubo is an add-on for an AI coding assistant (Claude Code, Codex and similar) that produces short videos with an AI presenter made from a single photo. The assistant writes the script, generates the lip-synced clips through the paid Renoise service, checks them against the script with speech recognition, evens out the voice, then edits everything in Remotion with one of five styles, captions, a cover frame and 48 ready-made on-screen effects. You need an AI assistant that can run commands, a Renoise account with credits, and ffmpeg, Python, Node and whisper.cpp installed.","easeRationale":"On Windows, macOS and Linux alike the install table in README.md asks for ffmpeg, Python with numpy and Pillow, Playwright with Chromium, Node 18+, whisper.cpp and a 500 MB-class speech model, plus an AI assistant that can run commands and a connected Renoise account, which is D on every system. macOS gets one extra feature: \"The text-behind-presenter effect uses Apple Vision and needs macOS\".","costNote":"The skill is free (MIT), but the video generation is billed in Renoise credits: the README's own runs cost 482 credits for a 27-second ad and about 3,250 for a 4-minute 36-second tutorial. Remotion, used for the editing, is free only for individuals and companies of up to 3 people.","notVerified":["What a video of your own will really cost in Renoise credits and money","Whether the result looks like the demo videos, which were made by the author with his own presenter","What Renoise does with the photo of your face and the clips it generates","Whether the whole chain (ffmpeg, Python, Playwright, Node, whisper.cpp) installs cleanly on Windows, since only Claude Code on the author's machine is reported as tested","Disk use, including the speech model you have to download separately"]},"zh-Hans":{"summaryShort":"用一张照片和一句话，生成带 AI 主播、字幕和剪辑的完整口播视频。","summary":"Digital Human Koubo 是 AI 编程助手（Claude Code、Codex 等）的一个扩展，能用一张照片生成的 AI 主播制作短视频。助手先写脚本，通过付费服务 Renoise 生成口型同步的片段，用语音识别对照脚本检查，调平音量，再用 Remotion 完成剪辑，可选五种风格，并带字幕、封面帧和 48 种现成的屏幕特效。你需要一个能运行命令的 AI 助手、一个有余额的 Renoise 账号，并装好 ffmpeg、Python、Node 和 whisper.cpp。","easeRationale":"在 Windows、macOS 和 Linux 上都一样：README.md 的安装表要求 ffmpeg、带 numpy 和 Pillow 的 Python、带 Chromium 的 Playwright、Node 18+、whisper.cpp 和一个约 500 MB 的语音识别模型，还要有能运行命令的 AI 助手和已连接的 Renoise 账号，所以各系统都是 D。macOS 多一项功能：\"The text-behind-presenter effect uses Apple Vision and needs macOS\"。","costNote":"这个 skill 免费（MIT），但视频生成按 Renoise 点数计费：README 自己的实测，一条 27 秒广告花了 482 点，一个 4 分 36 秒的教程约 3,250 点。用于剪辑的 Remotion 仅对个人和不超过 3 人的公司免费。","notVerified":["你自己做一条视频实际要花多少 Renoise 点数和多少钱","成品是否和演示视频一样，那些演示是作者用自己的主播做的","Renoise 会怎么处理你脸部的照片和它生成的片段","整条工具链（ffmpeg、Python、Playwright、Node、whisper.cpp）能否在 Windows 上顺利安装，因为只有作者电脑上的 Claude Code 被报告为已测试","磁盘占用，包括需要另外下载的语音识别模型"]}}},{"id":"R_kgDORzWbfQ","slug":"andyshaman__bydmate","name":"BYDMate","owner":"AndyShaman","fullName":"AndyShaman/BYDMate","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"licence_allows_personal_use_only","severity":"attention","evidence":"LICENSE 文件写道“基于 PolyForm Noncommercial License 1.0.0 授权”，只允许非商业用途使用，GitHub 也报告该仓库没有标准许可证。把它用在出租车、租车或任何商业场景中，都不被这份许可证允许。","source":"ai","title":"许可证只允许个人使用"},{"code":"app_controls_the_car","severity":"attention","evidence":"README 列出了一份“110 条指令的目录”，涵盖车窗、天窗、后备箱和前备箱、驾驶模式和车门锁（“Замки | запереть, отпереть”），可以通过语音或自动规则触发。它自己的提示写道，车速超过 120 km/h 车窗会被锁定，超过 80 km/h 天窗会被锁定，超过 30 km/h 解锁会被锁定，所以其余情况下它可以在你开车时自行操作。","source":"ai","title":"应用能控制汽车"},{"code":"uses_the_accessibility_service","severity":"attention","evidence":"README 说拦截方向盘按键“требует службы специальных возможностей, она включается автоматически”（需要无障碍服务，该服务会自动开启），同样的服务还用来操作 Yandex Navigator。这个 Android 权限能让一个应用读取并操作屏幕上的任何内容。","source":"ai","title":"使用了无障碍服务"},{"code":"voice_requests_go_to_an_ai_provider","severity":"info","evidence":"语音识别在设备本地完成，但 README 说固定短语列表之外的内容“уходит облачному AI-агенту”（会发送给云端 AI 代理），通过 OpenRouter、z.ai 或你自己的服务器。它还说云端功能默认是关闭的，密钥“保存在本地，代码和仓库里都没有任何密钥”。","source":"ai","title":"语音请求会发送给 ai 提供方"},{"code":"tested_mainly_on_one_car","severity":"info","evidence":"README 反复提到“Проверено на Leopard 3”（在豹 3 上测试过），涉及诊断功能和方向盘按键，并列出其他支持的车型（海狮 07、宋、元 Plus、海豹、汉，以及 DM-i 和 DM-o 混动车型）。有些功能，比如电池健康状态，只标注支持特定车型。","source":"ai","title":"主要只在一款车上测试过"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenRouter, z.ai or your own OpenAI-compatible server, for the voice assistant","Exa, for better web search","MiniMax, for an online voice"],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["hardware","automation","ai_agents"],"alternativeTo":[],"declaredAlternatives":[],"stars":500,"downloads":130142,"downloadsWeek":7522,"downloadsDays":6,"growth":24,"growthDays":6,"history":[{"date":"2026-11-04","stars":476},{"date":"2026-11-05","stars":483},{"date":"2026-11-06","stars":486},{"date":"2026-11-07","stars":488},{"date":"2026-11-08","stars":493},{"date":"2026-11-09","stars":497},{"date":"2026-11-10","stars":500}],"firstPublishedAt":"2026-11-04T01:36:50.278Z","commitSha":"16f9f2c03cbc5e70e11bb8da21a74b2229fa4100","verifiedAt":"2026-11-04T01:31:21.125Z","lastCommitAt":"2026-11-06T11:05:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An app for BYD car screens: real battery use, a trip and charge log, car automation and a Russian voice assistant.","summary":"BYDMate is an app for the screen of a BYD car (DiLink 3.0 to 5.1 and the UI7 firmware). It reads the battery system directly, so it shows real consumption instead of the car's own optimistic figure, keeps a log of trips and charges, splits the screen, puts the navigator on the instrument cluster and shows the blind-spot camera when you indicate. It also has rules that act on the car by itself and a voice assistant that speaks Russian only. The trip log and the quick commands work with no internet; the full assistant needs a key from an AI provider.","easeRationale":"It only runs on a BYD car's Android screen, so no desktop system gets a grade. There it is a B: you install the app file from the releases page yourself, since it is in no store, and then go through several steps in the settings, such as downloading the 226 MB speech model over Wi-Fi, pasting an AI provider key and teaching it a steering-wheel button on cars other than the Leopard 3.","costNote":"No price for the app, and the licence allows personal use only. The voice assistant needs a key from an AI provider: the README recommends OpenRouter (\"paid, fractions of a cent per question\") and mentions a free but slower z.ai option.","notVerified":["Whether it works on your exact car model and firmware version","How you get the app onto the car's screen, since the install section is not in the package read here","Whether installing it affects your warranty or the car's own software","How much the AI provider costs for everyday use of the assistant","Whether the car commands behave the same on models other than the Leopard 3","What the releases page actually offers for download, as the file list is not shown here"]},"zh-Hans":{"summaryShort":"一个用于比亚迪汽车屏幕的应用：真实电量消耗、行程和充电记录、汽车自动化，以及一个俄语语音助手。","summary":"BYDMate 是一个安装在比亚迪汽车屏幕上的应用（DiLink 3.0 到 5.1，以及 UI7 固件）。它直接读取电池系统，因此显示的是真实能耗，而不是汽车自带的偏乐观数字；它还会记录行程和充电日志，可以分屏，把导航放到仪表盘上，打转向灯时显示盲区摄像头画面。它还有能自行操作汽车的规则，以及一个只支持俄语的语音助手。行程记录和快捷命令不需要联网；完整的语音助手功能需要一个 AI 提供方的密钥。","easeRationale":"它只能在比亚迪汽车的 Android 屏幕上运行，所以没有桌面系统获得评级。在那里它是 B 级：你需要自己从发行版页面安装应用文件，因为它不在任何商店里，然后在设置里完成几个步骤，比如通过 Wi-Fi 下载 226 MB 的语音模型，粘贴 AI 提供方的密钥，并在豹 3 以外的车型上训练方向盘按键。","costNote":"这个应用不收费，许可证只允许个人使用。语音助手需要一个 AI 提供方的密钥：README 推荐 OpenRouter（“付费，每个问题几分钱”），并提到一个免费但更慢的 z.ai 选项。","notVerified":["是否适用于你的具体车型和固件版本","怎样把应用装到车机屏幕上，因为这里读取的材料里没有安装部分","安装它是否会影响保修或汽车自带的软件","助手日常使用时 AI 提供方要花多少钱","在豹 3 以外的车型上，车辆指令表现是否一样","发行版页面实际提供了哪些下载内容，因为这里没有显示文件列表"]}}},{"id":"R_kgDOVbNNlg","slug":"anefzaoui__misterdownloadmanager","name":"MisterDownloadManager","owner":"anefzaoui","fullName":"anefzaoui/misterdownloadmanager","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"12 天内获得 5494 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"installers_not_code_signed","severity":"attention","evidence":"README 的 Download 一节：\"Windows may say 'Windows protected your PC' the first time, because this early version isn't code-signed yet. Click More info, then Run anyway.\" Status 一节又写了 \"Not yet: code signing\"，macOS 的说明则让你右键点击程序再选择 Open。","source":"ai","title":"安装包没有代码签名"},{"code":"legal_status_varies_by_country","severity":"attention","evidence":"项目自己的标签和功能列表里有 \"bittorrent-client\"、\"debrid\"、\"real-debrid\"、\"yt-dlp\" 和 \"youtube-downloader\"：它会从网页下载视频，并通过 BitTorrent 分享文件，这违反部分网站的条款，在一些国家还受到限制或被禁止。","source":"ai","title":"各国法律情况不同"},{"code":"mac_build_untested_by_author","severity":"attention","evidence":"README 的 \"On a Mac? Be the first to try it\"：macOS 文件标有 -experimental，并写着 \"we haven't tried them on a Mac ourselves yet\"。Status 一节补充道 \"The macOS builds are untested\"。","source":"ai","title":"作者没有测试过 Mac 版本"},{"code":"opens_a_network_service_on_your_computer","severity":"info","evidence":"README 的功能列表和 Dockerfile：程序会启动一个 HTTPS 远程控制服务（EXPOSE 8463），供家庭网络内配对的手机使用，另外还有本地 API 和 MCP 服务器，让 Claude、Cursor 或 VS Code 添加、暂停和删除下载任务。","source":"ai","title":"会在你的电脑上开启网络服务"},{"code":"translations_machine_made","severity":"info","evidence":"README 的 Status 一节：\"Translations are machine-made. Rows marked `#, machine` in i18n/ wait for a native speaker\"；在 86 种语言中，只有马来语的菜单、工具栏和列表列名经过母语者校对。","source":"ai","title":"界面翻译由机器生成"},{"code":"speed_comparison_run_by_the_author","severity":"info","evidence":"README 的 \"The benchmark in detail\"：\"5x faster than IDM\" 这个数字来自作者自己的测试服务器（crates/labserver）和测试工具（tools/bench），不是独立机构的测量结果。","source":"ai","title":"速度对比由作者自己完成"},{"code":"very_young_project","severity":"info","evidence":"仓库创建于 2026-10-29，版本号为 0.5.2，12 天内发布了 8 次，README 写着 \"MDM is young\" 以及 \"some corners still good enough for now\"；当前有 140 个未关闭的问题。","source":"ai","title":"项目非常年轻"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","rust","ffmpeg"],"disk":{"minMB":16,"maxMB":49,"basis":"installer"},"apiKeys":["Debrid account (Real-Debrid, AllDebrid or Premiumize)"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system","productivity"],"alternativeTo":[{"slug":"internet-download-manager","name":"Internet Download Manager","confirmed":true},{"slug":"4k-video-downloader","name":"4K Video Downloader","confirmed":true},{"slug":null,"name":"Internet Download Manager (IDM)","confirmed":false}],"declaredAlternatives":[],"stars":5494,"downloads":0,"downloadsWeek":0,"downloadsDays":0,"growth":4175,"growthDays":2,"history":[],"firstPublishedAt":"2026-11-10T23:52:43.860Z","commitSha":"70599ffc0a2ff2703bab7d1d52b749c91582b471","verifiedAt":"2026-11-10T23:34:05.268Z","lastCommitAt":"2026-11-10T22:54:14Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"MisterDownloadManager-0.5.2-setup.exe","sizeBytes":17065873,"url":"https://github.com/anefzaoui/misterdownloadmanager/releases/download/v0.5.2/MisterDownloadManager-0.5.2-setup.exe","releaseUrl":"https://github.com/anefzaoui/misterdownloadmanager/releases/tag/v0.5.2"},{"os":"mac","file":"MisterDownloadManager-0.5.2-macos-arm64-experimental.dmg","sizeBytes":15510960,"url":"https://github.com/anefzaoui/misterdownloadmanager/releases/download/v0.5.2/MisterDownloadManager-0.5.2-macos-arm64-experimental.dmg","releaseUrl":"https://github.com/anefzaoui/misterdownloadmanager/releases/tag/v0.5.2"},{"os":"linux","file":"MisterDownloadManager-0.5.2-linux-x86_64.AppImage","sizeBytes":20777464,"url":"https://github.com/anefzaoui/misterdownloadmanager/releases/download/v0.5.2/MisterDownloadManager-0.5.2-linux-x86_64.AppImage","releaseUrl":"https://github.com/anefzaoui/misterdownloadmanager/releases/tag/v0.5.2"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Free download manager for Windows, Mac and Linux: resumable downloads, browser button, videos, image galleries and torrents.","summary":"Mister Download Manager is a desktop download manager written in Rust, built to look and work like Internet Download Manager. It splits each file into several connections, resumes after a drop, catches downloads from Chrome, Firefox, Edge, Brave, Opera and Vivaldi through an extension, and also handles web videos, image galleries, torrents and magnet links. It has queues, a scheduler, a VPN-only mode, a command line, an MCP server for AI assistants and a remote control for your phone. Installers are ready for Windows and Linux; the Mac builds are marked experimental.","easeRationale":"On Windows you run the setup.exe from the releases page and it installs the app, the browser link and your language, so that is the best grade; Linux has ready .deb, .rpm and AppImage files. Mac gets a .dmg you drag to Applications, but the builds are named -experimental and the author says they have not been tried on a Mac. Nothing is code-signed yet, so Windows and macOS show a warning the first time.","costNote":"No paid edition, no trial and no account; donations through GitHub Sponsors are optional. Debrid file-host accounts it can connect to (Real-Debrid, AllDebrid, Premiumize) are paid services of your own.","notVerified":["whether yt-dlp, ffmpeg and 7-Zip come inside the installer or are downloaded the first time you use the video, gallery or archive features","whether the download speeds in the README's comparison hold on a normal internet connection rather than the author's local test server","whether the Mac builds start and work at all, since the author has not tested them","how much disk space the app uses after its lists, logs, crash reports and the IP country table grow","what the browser extension and the remote control send over your network while they are running"]},"zh-Hans":{"summaryShort":"适用于 Windows、Mac 和 Linux 的免费下载管理器：断点续传、浏览器按钮、网页视频、图集和种子下载。","summary":"Mister Download Manager 是一款用 Rust 编写的桌面下载管理器，外观和用法都模仿 Internet Download Manager。它把每个文件拆成多条连接，断线后可以续传，通过扩展接管 Chrome、Firefox、Edge、Brave、Opera 和 Vivaldi 的下载，还能处理网页视频、图片画廊、种子和磁力链接。它带有下载队列、定时计划、只在 VPN 下下载的模式、命令行、供 AI 助手使用的 MCP 服务器，以及用手机远程控制的功能。Windows 和 Linux 的安装包已经就绪；Mac 版本被标记为实验性。","easeRationale":"在 Windows 上，从发布页下载 setup.exe 运行即可装好程序、浏览器联动和界面语言，所以给出最好的等级；Linux 有现成的 .deb、.rpm 和 AppImage 文件。Mac 提供 .dmg，拖进「应用程序」即可，但这些文件名里带 -experimental，作者说自己没有在 Mac 上试过。目前还没有任何代码签名，因此 Windows 和 macOS 第一次运行时都会弹出警告。","costNote":"没有付费版本、试用期或账号；通过 GitHub Sponsors 捐助是自愿的。它可以连接的 debrid 网盘账号（Real-Debrid、AllDebrid、Premiumize）是你自己付费的服务。","notVerified":["yt-dlp、ffmpeg 和 7-Zip 是随安装包一起提供，还是在你第一次使用视频、图集或压缩包功能时才下载","README 对比中的下载速度在普通网络环境下是否也成立，而不是只在作者本地的测试服务器上成立","Mac 版本能否正常启动和使用，因为作者没有测试过","当软件的列表、日志、崩溃报告和 IP 国家对照表增长后，会占用多少磁盘空间","浏览器扩展和远程控制功能运行时，会在你的网络上发送哪些内容"]}}},{"id":"R_kgDOM-HaTw","slug":"anliyuan__ultralight-digital-human","name":"Ultralight Digital Human","owner":"anliyuan","fullName":"anliyuan/Ultralight-Digital-Human","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/file/d/1e4Z9zS053JEWl6Mj3W9Lbc9GDtzHIg6b/view?usp=drive_link","source":"check","title":"外部下载链接"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹或 winget 清单均未检测到许可证","source":"check","title":"没有许可证"},{"code":"makes_a_real_person_say_anything","severity":"attention","evidence":"整个项目的目的就是用一个真人的视频训练，然后用任意音频文件驱动那张脸（`python inference.py --asr hubert --dataset ... --audio_feat your_test_audio_hu.npy`）。README 甚至说演示视频是未经许可使用的一段中国电视主持人的播报：“我会提供一个demo视频，来自康辉老师的口播，侵删”。让真人看起来说出他们从未说过的话，可能侵犯肖像权并触犯冒充他人相关的法律，而项目中没有任何环节询问该人是否同意。","source":"ai","title":"可以让真人说出任何话"},{"code":"license_badge_without_license_file","severity":"attention","evidence":"README 显示一个写着 `license-Apache 2` 并链接到 `./LICENSE` 的徽章，但仓库顶层只有 .gitignore、README.md、data_utils/、datasetsss.py、demo/、dihuman_run.py、face_utils.py、inference.py、pth2onnx.py、train.py 和 unet.py——并没有 LICENSE 文件，所以这个徽章承诺的条款其实并不存在于仓库中。","source":"ai","title":"有许可证徽章却没有许可证文件"},{"code":"author_points_to_a_successor_project","severity":"info","evidence":"README 现在开篇就写道：“如果你想要一个更新、更易部署的版本，请查看 FeatherTalk。它是 Ultralight Digital Human 的一个精简后续版本。”快照中的最后一次提交是 2026-07-22，距检查时间 105 天前，还有 119 个未关闭的 issue。","source":"ai","title":"作者指向一个后续项目"},{"code":"research_code_not_finished","severity":"info","evidence":"README 说早期代码有几个 bug 最近才修复，流式代码“代码还没有完善”，若要实时使用“重构代码是必须的”，而作者手里更快的一种方法目前保持闭源，因为“有人在商用，暂时不做开源”。","source":"ai","title":"研究代码尚未完成"},{"code":"strict_requirements_for_the_input_video","severity":"info","evidence":"README 要求：视频时长 3 到 5 分钟，每一帧都要全脸可见，使用 wenet 编码器时帧率需恰好为 20fps，使用 hubert 时为 25fps，测试音频需为 16000Hz，作者还提醒噪音、空房间的回声或不清晰的说话很可能毁掉结果，建议使用外接麦克风。","source":"ai","title":"对输入视频有严格要求"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["video","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":2670,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":4,"growthDays":6,"history":[{"date":"2026-11-05","stars":2667},{"date":"2026-11-06","stars":2667},{"date":"2026-11-07","stars":2668},{"date":"2026-11-08","stars":2668},{"date":"2026-11-09","stars":2669},{"date":"2026-11-10","stars":2670}],"firstPublishedAt":"2026-11-04T22:32:54.809Z","commitSha":"5aa8b5a8aae4b003b7de5750abeac5343a7b38e0","verifiedAt":"2026-11-04T22:23:49.925Z","lastCommitAt":"2026-07-22T12:29:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that learns one person's face from a short video, then makes them lip-sync any audio you give it.","summary":"Python training and inference scripts for a very small talking-head model. You film one person for three to five minutes with their whole face visible and clean sound, run the preprocessing, train a model for that person on an NVIDIA graphics card, and afterwards you can feed in any recording and get a video where that person's mouth follows the words. The author says it is light enough to run in real time on phones. There is no program to install, no ready model and no licence file, and the author now points to a newer project, FeatherTalk.","easeRationale":"Both Windows and Linux are a D and need the same work: a conda environment with Python 3.10 and PyTorch with CUDA, a file downloaded by hand from Google Drive into data_utils/, then separate commands for preprocessing, training and inference plus an ffmpeg command to put the sound back. Macs are not covered, since the setup is built around an NVIDIA card.","costNote":"Nothing is charged, but no licence file was found, so the terms for using the code and anything you make with it are unclear. You pay in computer time for training.","notVerified":["How good the resulting face and mouth look with your own video","How long training takes and how much graphics memory it needs","Whether the file on Google Drive is still available and what exactly it contains","Whether it really runs in real time on a phone, since no mobile code is published here","What you are allowed to do with the code and the videos, given the missing licence"]},"zh-Hans":{"summaryShort":"研究代码，从一段短视频学习一个人的脸，然后让这张脸对任意音频做出对口型效果。","summary":"用于训练和推理一个非常小的说话人头像模型的 Python 脚本。你拍摄一个人 3 到 5 分钟、全脸可见且音质干净的视频，运行预处理，在一张 NVIDIA 显卡上为这个人训练一个模型，之后就可以输入任意录音，得到一个嘴型跟随这些话语的视频。作者说它足够轻量，可以在手机上实时运行。这里没有可安装的程序、没有现成模型，也没有许可证文件，作者现在把大家指向一个更新的项目 FeatherTalk。","easeRationale":"Windows 和 Linux 评级都是 D，且需要同样的工作：搭建带 Python 3.10 和支持 CUDA 的 PyTorch 的 conda 环境，手动从 Google Drive 下载一个文件放到 data_utils/ 中，再分别运行预处理、训练和推理命令，最后用一条 ffmpeg 命令把声音放回去。由于整套配置是围绕 NVIDIA 显卡设计的，Mac 没有覆盖。","costNote":"没有收费，但没有找到许可证文件，所以使用这段代码及其产出物的条款并不明确。你付出的代价是训练所需的计算时间。","notVerified":["用你自己的视频生成的脸部和嘴型效果究竟有多好","训练需要多长时间，需要多少显存","Google Drive 上的那个文件是否仍然可用，其中具体包含什么","它是否真的能在手机上实时运行，因为这里没有发布任何移动端代码","鉴于许可证缺失，你能对代码和视频做什么尚不明确"]}}},{"id":"R_kgDOOiiGLw","slug":"anomalyco__opencode","name":"opencode","owner":"anomalyco","fullName":"anomalyco/opencode","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"low","signals":[{"code":"full_access_agent_by_default","severity":"info","evidence":"README.md：「**build** - Default, full-access agent for development work」；只有可选的「plan」代理会「Denies file edits by default」并「Asks permission before running bash commands」。","source":"ai","title":"默认就是完全访问权限的代理"},{"code":"desktop_app_is_beta","severity":"info","evidence":"README.md 标题：「### Desktop App (BETA)」。项目自身仍在测试这个带窗口的版本。","source":"ai","title":"桌面应用处于测试阶段"},{"code":"install_script_runs_from_website","severity":"info","evidence":"README.md 安装部分：「# YOLO」，后跟 `curl -fsSL https://opencode.ai/v2/install | bash`。仓库中的 `install` 文件会向 https://opencode.ai/update/api/latest/cli/npm 询问要获取哪个版本，并从 registry.npmjs.org 下载，因此这条路径运行的是安装时获取的代码，而不是这份快照里的文件。","source":"ai","title":"安装脚本从网站运行代码"},{"code":"sends_your_code_to_outside_services","severity":"info","evidence":"package.json 列出了多个 AI 提供商的包，例如 `@ai-sdk/openai-compatible`、`@ai-sdk/google` 和 `@ai-sdk/mistral`。助手读取的任何文件都会发送给你所连接的模型服务。","source":"ai","title":"会把你的代码发送给外部服务"},{"code":"very_fast_release_pace","severity":"info","evidence":"项目数据：自 2025-04-30 以来在 GitHub 上已有 892 次发布，v2.2.3、v2.2.4 和 v2.2.5 均标注日期为 2026-10-30。版本每天多次变动，所以你今天安装的内容明天的表现可能会不同。","source":"ai","title":"发布节奏非常快"},{"code":"updates_itself_from_the_project_servers","severity":"info","evidence":"package.json 对 `electron-updater@6.8.9` 和 `app-builder-lib@26.15.7` 做了补丁，仓库中的 `install` 脚本会读取 https://opencode.ai/update/api/latest/cli/npm。因此桌面应用和命令行都能自行获取新版本。","source":"ai","title":"会从项目自己的服务器自动更新"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":252,"maxMB":755,"basis":"installer"},"apiKeys":["AI model provider account or key (for example OpenAI-compatible services, Google, Mistral)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents","local_ai"],"alternativeTo":[{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true},{"slug":"claude-code","name":"Claude Code","confirmed":true},{"slug":"qodo","name":"Qodo","confirmed":true}],"declaredAlternatives":[],"stars":225884,"downloads":142116693,"downloadsWeek":2731763,"downloadsDays":7,"growth":1530,"growthDays":7,"history":[{"date":"2026-10-25","stars":221557},{"date":"2026-10-26","stars":221967},{"date":"2026-10-27","stars":222385},{"date":"2026-10-28","stars":222762},{"date":"2026-10-29","stars":223069},{"date":"2026-10-30","stars":223336},{"date":"2026-10-31","stars":223602},{"date":"2026-11-01","stars":223864},{"date":"2026-11-02","stars":224116},{"date":"2026-11-03","stars":224354},{"date":"2026-11-04","stars":224578},{"date":"2026-11-05","stars":224780},{"date":"2026-11-06","stars":225033},{"date":"2026-11-07","stars":225267},{"date":"2026-11-08","stars":225503},{"date":"2026-11-09","stars":225652},{"date":"2026-11-10","stars":225884}],"firstPublishedAt":"2026-10-25T23:18:56.612Z","commitSha":"591690c375df2ff6306101fbd2fe70cfbdd8f27e","verifiedAt":"2026-10-31T01:40:26.496Z","lastCommitAt":"2026-11-10T21:23:29Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An AI assistant for programmers that reads and edits the code in your projects, in a terminal or a desktop app.","summary":"OpenCode is an AI coding assistant for people who write software. You describe a change in plain words and it reads your project files, edits them and can run commands. A second, read-only \"plan\" mode only looks and asks before running commands, and a \"general\" helper handles long searches. It runs in a terminal, and the desktop app, marked beta, has ready installers for Windows, Mac and Linux. You still need an account or key for an AI model service, which may charge for use.","easeRationale":"Release v2.2.5 carries ready desktop installers for all three systems (`opencode-desktop-win-x64.exe` at 252 MB, `opencode-desktop-mac-arm64.dmg`, and .deb/.rpm/.AppImage for Linux), all marked beta; after installing you still have to connect an AI model account or key. The terminal version is a separate download or one install command.","costNote":"The program is free (MIT license). The AI model service you connect it to may charge for use; the package states no prices.","notVerified":["Whether the beta desktop app works reliably on each system","How much the connected AI model service will actually cost for your use","Exactly which files and project content are sent to the model service","Whether any free model option works without a key of your own","What the app downloads when it updates itself while running","Real disk use, since the Windows desktop installer alone is 252 MB"]},"zh-Hans":{"summaryShort":"一款面向程序员的 AI 助手，能在终端或桌面应用中读取并编辑你项目里的代码。","summary":"OpenCode 是一款面向软件开发者的 AI 编程助手。你用简单的话描述需要的改动，它就会读取项目文件、进行编辑，并可以运行命令。另外还有一个只读的「plan」模式，只会查看和提问，不会在未经确认前运行命令；「general」助手则负责处理长时间的搜索。它可以在终端中运行，标记为 beta 版的桌面应用则提供了 Windows、Mac 和 Linux 的现成安装包。你仍然需要一个 AI 模型服务的账号或密钥，该服务可能会按使用量收费。","easeRationale":"v2.2.5 版本为三大系统都提供了现成的桌面安装包（`opencode-desktop-win-x64.exe`，252 MB；`opencode-desktop-mac-arm64.dmg`；以及适用于 Linux 的 .deb/.rpm/.AppImage），均标记为 beta 版；安装后你仍需连接一个 AI 模型的账号或密钥。终端版本是单独的下载包，或通过一条安装命令获取。","costNote":"该程序本身是免费的（MIT 许可证）。你连接的 AI 模型服务可能会按使用量收费，资料中没有列出具体价格。","notVerified":["测试版桌面应用在各系统上是否都能稳定运行","连接的 AI 模型服务按你的实际使用量会产生多少费用","究竟哪些文件和项目内容会被发送给模型服务","是否有无需自备密钥就能使用的免费模型选项","应用在运行中自我更新时会下载哪些内容","实际占用的磁盘空间，因为仅 Windows 桌面安装包就有 252 MB"]}}},{"id":"R_kgDOVOGXPQ","slug":"ansub__tv-detox","name":"TV Detox","owner":"Ansub","fullName":"Ansub/tv-detox","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{},"risk":"high","signals":[{"code":"can_leave_the_tv_stuck_restarting","severity":"high","evidence":"README.md 自己的警告：\"This turns off system apps. A few TVs got stuck at the logo or kept restarting\"，对于 Google TV 和 Chromecast 则是 \"On some of these devices that left them stuck restarting, and the only fix was a factory reset\"。它还记录了有电视在 install.sh 运行途中忘掉了 Wi-Fi 网络。","source":"ai","title":"可能让电视卡在反复重启"},{"code":"ai_agent_changes_your_tv","severity":"attention","evidence":"主要的操作方式是把一段提示词粘贴到 Claude Code 或 Codex 里，由它连接电视并停用应用。提示词本身写着 \"you're an AI, so you can make mistakes\"，README.md 还补充道 \"Do it at your own risk ... This is what worked for us, not professional advice\"。","source":"ai","title":"AI 助手会改动你的电视"},{"code":"debugging_must_be_turned_on","severity":"attention","evidence":"README.md：\"While USB, Network or Wireless debugging is on, other devices on your network can ask to control the TV. Turn it off once setup is done\"。每次想撤销或更新时，你也得再次把它打开。","source":"ai","title":"必须打开调试功能"},{"code":"features_stop_working","severity":"info","evidence":"README.md 按品牌列出了副作用：\"The YouTube button on the remote can stop working, and some lose casting\"，在两台 TCL 电视上 \"the app buttons on the remote (YouTube, Netflix and others) stopped working\"，而关闭手机遥控服务会 \"also stops your phone from working as a remote\"。","source":"ai","title":"部分功能会失效"},{"code":"accessibility_service_and_play_protect","severity":"info","evidence":"在 Google TV 上，主屏幕通过无障碍服务接管主页键：README.md 提到 \"Play Protect may warn about the app because it has an accessibility service\"，而且侧载的应用 \"may first need App info > ... > Allow restricted settings\"。","source":"ai","title":"无障碍服务与 Play Protect"},{"code":"app_file_fetched_by_the_script","severity":"info","evidence":"install.sh 不自带应用，而是去下载：\"curl -fsSL -o \\\"$APK\\\" https://github.com/Ansub/tv-detox/releases/latest/download/tvhome.apk\"，除非你自己编译过。你也可以用 Java 17 和 Android SDK 从源码编译。","source":"ai","title":"应用文件由脚本下载"},{"code":"links_to_the_author_community","severity":"info","evidence":"仓库的主页字段和 README.md 里的第一个徽章都指向作者的 Skool 社群（\"Join the community\"），旁边还有他的 YouTube 频道，所以这个项目同时也是一种推广。","source":"ai","title":"链接到作者的社群"}],"platforms":[],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[{"slug":"projectivy-launcher","name":"Launcher","confirmed":true}],"declaredAlternatives":[],"stars":2313,"downloads":24368,"downloadsWeek":3431,"downloadsDays":7,"growth":187,"growthDays":7,"history":[{"date":"2026-10-27","stars":1684},{"date":"2026-10-28","stars":1720},{"date":"2026-10-29","stars":1763},{"date":"2026-10-30","stars":1903},{"date":"2026-10-31","stars":1979},{"date":"2026-11-01","stars":2038},{"date":"2026-11-02","stars":2083},{"date":"2026-11-03","stars":2126},{"date":"2026-11-04","stars":2168},{"date":"2026-11-05","stars":2194},{"date":"2026-11-06","stars":2222},{"date":"2026-11-07","stars":2245},{"date":"2026-11-08","stars":2269},{"date":"2026-11-09","stars":2296},{"date":"2026-11-10","stars":2313}],"firstPublishedAt":"2026-10-27T18:40:27.791Z","commitSha":"b22a1e368ee56d670d11115d8408e9025488240b","verifiedAt":"2026-11-02T02:57:47.502Z","lastCommitAt":"2026-10-30T07:16:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns off the junk apps on an old Android TV and puts a plain home screen with just your apps, inputs and a clock.","summary":"TV Detox has two halves: a very small home screen app for Android TV or Google TV (your apps, your HDMI inputs, a clock, no ads or recommendation rows, about 30 MB of memory), and a ready-made prompt that tells Claude Code or Codex on your laptop to switch off the maker's junk apps one by one, asking you before each change. You can also skip the AI and run the install script yourself. You need to turn on debugging on the TV and have adb on your laptop. Apps are turned off, not deleted, and undo.sh puts them back.","easeRationale":"The app runs on a TV, so no desktop system gets a grade. From your laptop you turn on hidden developer options on the TV, install Google's adb tool and run shell scripts (on Windows through Git Bash); newer Google TVs need a pairing code too. The deciding part is judgement: which of the TV's apps to switch off is decided app by app, by you or by an AI agent.","costNote":"The app and the scripts are free (MIT). The recommended path uses Claude Code or Codex, which are paid, but the README says you can skip the AI and run install.sh with only adb.","notVerified":["Whether it works on your exact TV, since the tested list is short: Android TV 9, 11 and 12 and Google TV 14 on a few TCL and OnePlus sets, plus emulators","Whether undo.sh really puts everything back on your TV","How much faster the TV actually feels: the README's own memory numbers are partly explained by the restart","Whether the tvhome.apk the script downloads is built from the code in this repository","What the AI agent decides to turn off on your TV, which differs every time"]},"zh-Hans":{"summaryShort":"关掉旧安卓电视上的垃圾应用，换成只有你的应用、输入源和时钟的简洁主屏幕。","summary":"TV Detox 分为两部分：一个给 Android TV 或 Google TV 用的极小主屏幕应用（你的应用、你的 HDMI 输入源、一个时钟，没有广告和推荐栏，约占 30 MB 内存），以及一段现成的提示词，让你笔记本上的 Claude Code 或 Codex 逐个关闭厂商预装的垃圾应用，每次改动前都会先问你。你也可以不用 AI，自己运行安装脚本。你需要在电视上打开调试，并在笔记本上装好 adb。应用只是被停用，并没有删除，undo.sh 可以把它们恢复。","easeRationale":"这个应用在电视上运行，所以没有任何桌面系统获得评级。你要在笔记本上打开电视里隐藏的开发者选项，安装 Google 的 adb 工具，并运行 shell 脚本（在 Windows 上要通过 Git Bash）；较新的 Google TV 还需要配对码。真正起决定作用的是判断力：电视上哪些应用该关，要逐个应用来定，由你或 AI 助手决定。","costNote":"应用和脚本免费（MIT）。推荐的方式要用 Claude Code 或 Codex，它们是付费的，但 README 说你可以不用 AI，只靠 adb 运行 install.sh。","notVerified":["它能否在你的这台电视上正常工作，因为测试过的名单很短：几台 TCL 和 OnePlus 电视上的 Android TV 9、11、12 和 Google TV 14，外加模拟器","undo.sh 是否真的能把你电视上的一切恢复原样","电视实际能快多少：README 自己给出的内存数字，有一部分是重启造成的","脚本下载的 tvhome.apk 是否由本仓库里的代码编译而来","AI 助手会决定在你的电视上关掉什么，而且每次都不一样"]}}},{"id":"R_kgDONoVX3w","slug":"antgroup__ditto-talkinghead","name":"Ditto","owner":"antgroup","fullName":"antgroup/ditto-talkinghead","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"can_make_a_real_person_appear_to_speak","severity":"attention","evidence":"README.md描述了由\"--audio_path\"和\"--source_path\"(任意图片)驱动的\"Controllable Realtime Talking Head Synthesis\"。同一条能给你自己肖像配音动画的命令,也能用任意音频给别人的照片配上动画,仓库里除了Apache-2.0许可证之外,没有看到同意确认、水印或使用限制。","source":"ai","title":"能让真人看起来在说话"},{"code":"no_updates_for_a_year","severity":"attention","evidence":"最近一次push在2025-11-12,360天没有提交,也完全没有GitHub发布。README最新的条目是2025.11.12发布的训练代码。","source":"ai","title":"一年没有更新"},{"code":"ready_models_only_for_newer_nvidia_cards","severity":"info","evidence":"README.md: \"We have provided the tensorRT model with hardware-compatibility-level=Ampere_Plus (checkpoints/ditto_trt_Ampere_Plus/). If your GPU does not support it, please execute the cvt_onnx_to_trt.py script to convert from the general onnx model\"。在更旧的显卡上,你得自己构建引擎。","source":"ai","title":"现成模型只支持更新的NVIDIA显卡"},{"code":"model_files_hosted_elsewhere","severity":"info","evidence":"README.md Download Checkpoints: \"git clone https://huggingface.co/digital-avatar/ditto-talkinghead checkpoints\"。模型文件,包括一个编译好的插件(libgrid_sample_3d_plugin.so),都来自Hugging Face而不是这个仓库;确定性磁盘估算(100到300 MB)只覆盖代码。","source":"ai","title":"模型文件托管在别处"},{"code":"pinned_old_dependencies","severity":"info","evidence":"environment.yaml锁定了python=3.10.16、带cuda 12.1的pytorch=2.5.1、tensorrt==8.6.1和numpy==2.0.1,全都来自2024年末,可能与当前驱动或更新的CUDA安装冲突。","source":"ai","title":"依赖版本锁定在旧版本"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":932,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":3,"growthDays":6,"history":[{"date":"2026-11-08","stars":930},{"date":"2026-11-09","stars":931},{"date":"2026-11-10","stars":932}],"firstPublishedAt":"2026-11-07T22:56:16.677Z","commitSha":"c3e47eee2e626500017a0556b470d6d4182f85e8","verifiedAt":"2026-11-07T22:42:27.332Z","lastCommitAt":"2025-11-12T13:58:42Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns one photo of a face plus an audio file into a video of that face speaking, on your own NVIDIA computer.","summary":"Research code from Ant Group that animates a still portrait so it lip-syncs and moves along with an audio file you provide, saving the result as an MP4. You run it with one command after creating a conda environment, installing TensorRT and downloading about a dozen model files from Hugging Face. It was tested on Linux (CentOS 7.2) with an A100 card, and the ready-made fast models only cover NVIDIA Ampere cards or newer; older cards need a conversion step. There is also an official Colab notebook to try it in the browser.","easeRationale":"On Linux you clone the project, build a conda environment, install TensorRT 8.6.1, fetch the model files with git lfs and then type a command with five paths; there is no window or page to click. Windows and macOS are not mentioned at all, and the tested setup was CentOS 7.2 with an NVIDIA A100.","notVerified":["Whether it installs and runs on anything other than the tested CentOS plus A100 setup, and whether it works on Windows at all.","Real disk use after downloading the model files from Hugging Face.","How fast it is on a consumer card, and whether the \"realtime\" claim holds outside an A100.","How good the result looks with your own photo and audio.","Whether the official Colab notebook still runs."]},"zh-Hans":{"summaryShort":"在你自己的NVIDIA电脑上,把一张人脸照片加一段音频变成这张脸说话的视频。","summary":"蚂蚁集团的研究代码,能让一张静态肖像根据你提供的音频对口型并做出相应动作,把结果保存为MP4。创建conda环境、安装TensorRT并从Hugging Face下载约十几个模型文件后,只需一条命令就能运行。它在Linux(CentOS 7.2)配A100显卡上测试过,现成的快速模型只支持NVIDIA Ampere或更新架构的显卡;更旧的显卡需要额外的转换步骤。官方还提供一个Colab笔记本,可以在浏览器里试用。","easeRationale":"在Linux上,你需要克隆项目、搭建conda环境、安装TensorRT 8.6.1、用git lfs获取模型文件,然后手动输入一条带五个路径的命令;没有窗口或页面可点。README完全没提Windows和macOS,测试环境是CentOS 7.2配NVIDIA A100。","notVerified":["除了测试过的CentOS加A100环境外,是否能在其他环境安装运行,以及在Windows上是否能用。","从Hugging Face下载模型文件后的实际磁盘占用。","在消费级显卡上速度如何,\"realtime\"的说法离开A100是否依然成立。","用你自己的照片和音频生成的效果如何。","官方的Colab笔记本是否还能运行。"]}}},{"id":"R_kgDON91aYw","slug":"anthropics__claude-code","name":"Claude Code","owner":"anthropics","fullName":"anthropics/claude-code","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道直接传给 shell 执行：curl -fsSL https://claude.ai/install.sh | bash","source":"check","title":"curl 管道直接传给 shell 执行"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"not_open_source_and_source_code_not_here","severity":"attention","evidence":"LICENSE.md 只写着 \"© Anthropic PBC. All rights reserved. Use is subject to Anthropic's Commercial Terms of Service\"。仓库里只有文档、示例、插件和 issue；程序本体是以发布版里现成的二进制文件形式提供的。","source":"ai","title":"不是开源软件，源代码也不在这里"},{"code":"usage_and_conversation_data_collected","severity":"attention","evidence":"README 的 Data collection 部分写道：\"When you use Claude Code, we collect feedback, which includes usage data (such as code acceptance or rejections), associated conversation data, and user feedback submitted via the /bug command.\" 同一部分还说有保留期限制，且反馈不会用于模型训练。","source":"ai","title":"会收集使用数据和对话数据"},{"code":"it_runs_commands_and_changes_your_files","severity":"attention","evidence":"仓库描述中写道，这是一个 \"agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.\"","source":"ai","title":"会运行命令并修改你的文件"},{"code":"old_install_route_dropped","severity":"info","evidence":"README 的 Get started 部分写道：\"Installation via npm is deprecated. Use one of the recommended methods below.\" npm 包仍在发布，但已不再是推荐的安装方式。","source":"ai","title":"旧的安装方式已被弃用"},{"code":"very_large_issue_backlog","severity":"info","evidence":"GitHub 的数据显示有 16215 个未关闭的 issue，所以你报告的问题可能要等很长时间。","source":"ai","title":"积压了大量未处理的 issue"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":148,"maxMB":443,"basis":"installer"},"apiKeys":["Anthropic (a Claude account you sign in to, or an API key)"],"cost":{"model":"paid"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["development","ai_agents"],"alternativeTo":[],"declaredAlternatives":[],"stars":155200,"downloads":1819392,"downloadsWeek":184496,"downloadsDays":7,"growth":769,"growthDays":7,"history":[{"date":"2026-10-26","stars":153478},{"date":"2026-10-27","stars":153607},{"date":"2026-10-28","stars":153747},{"date":"2026-10-29","stars":153872},{"date":"2026-10-30","stars":153984},{"date":"2026-10-31","stars":154079},{"date":"2026-11-01","stars":154176},{"date":"2026-11-02","stars":154303},{"date":"2026-11-03","stars":154431},{"date":"2026-11-04","stars":154536},{"date":"2026-11-05","stars":154659},{"date":"2026-11-06","stars":154790},{"date":"2026-11-07","stars":154905},{"date":"2026-11-08","stars":154987},{"date":"2026-11-09","stars":155077},{"date":"2026-11-10","stars":155200}],"firstPublishedAt":"2026-10-27T01:10:12.599Z","commitSha":"964d69565952a176ac1975ed218362991db71433","verifiedAt":"2026-11-03T02:41:16.354Z","lastCommitAt":"2026-11-10T21:21:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A terminal assistant that reads your project and makes the changes you ask for in plain words.","summary":"Claude Code is a program you run in the terminal. You describe what you want in plain words and it reads your project, edits files, runs commands and handles git work for you. It also plugs into editors and can be called on GitHub. It works on Windows, macOS and Linux, and is installed with a single command. It is made by Anthropic: the code here is not open source, and you need an Anthropic account that is paid for to use it.","easeRationale":"On all three systems the install is one terminal or PowerShell line (or brew/winget), and the released files are ready binaries with nothing else to install. It stays at C because everything happens in a terminal: there is no window to click, and you have to sign in to a paid account before the first run.","costNote":"The program is downloaded at no charge, but it only works with an Anthropic account you pay for (a Claude plan or API credits). Neither the README nor the repository states prices; LICENSE.md points to Anthropic's Commercial Terms of Service.","notVerified":["What the program actually does on your machine, since the source code is not published here.","What a month of real use costs, because the repository lists no prices.","Which parts of a conversation are kept and for how long, beyond what the data-usage pages say.","Whether the Windows, macOS and Linux builds behave the same, since releases come out almost daily.","How much disk it ends up using, as each release file is between 148 and 443 MB."]},"zh-Hans":{"summaryShort":"一个终端助手，能读取你的项目，并按你用简单话语提出的要求进行修改。","summary":"Claude Code 是一个在终端里运行的程序。你用简单的话描述你想要的结果，它就会读取你的项目、编辑文件、运行命令，并替你处理 git 相关的工作。它还能接入编辑器，也可以在 GitHub 上被调用。它可以在 Windows、macOS 和 Linux 上运行，只需一条命令就能安装。它由 Anthropic 制作：这里的代码不是开源的，使用它需要一个付费的 Anthropic 账号。","easeRationale":"三个系统的安装都只需要在终端或 PowerShell 里输入一行命令（或用 brew/winget），发布的文件是现成的二进制文件，不用再安装别的东西。它停留在 C 级，是因为一切都在终端里完成：没有窗口可以点击，而且第一次运行前必须登录一个付费账号。","costNote":"程序本身可以免费下载，但只能配合你付费购买的 Anthropic 账号使用（Claude 方案或 API 额度）。README 和仓库都没有列出价格；LICENSE.md 指向 Anthropic 的商业服务条款。","notVerified":["程序在你电脑上到底做了什么，因为源代码并没有发布在这里","一个月的实际使用要花多少钱，因为仓库里没有列出价格","一段对话的哪些部分会被保留、保留多久，数据使用说明页面之外的情况并不清楚","Windows、macOS 和 Linux 版本的表现是否一致，因为几乎每天都有新版本发布","最终会占用多少磁盘空间，因为每个发布文件的大小在 148 到 443 MB 之间"]}}},{"id":"R_kgDOP0wfhg","slug":"anthropics__skills","name":"Agent Skills","owner":"anthropics","fullName":"anthropics/skills","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"low","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单中都没有检测到许可证","source":"check","title":"没有许可证"},{"code":"document_skills_source_available","severity":"info","evidence":"README.md：\"Many skills in this repo are open source (Apache 2.0)\"，但 skills/docx、skills/pdf、skills/pptx 和 skills/xlsx 中的 skill \"are source-available, not open source\"。重复使用这四个会受到限制，这是顶层缺少许可证之外的额外限制。","source":"ai","title":"文档类 skill 仅源码可见，并非开源"},{"code":"demo_only_disclaimer","severity":"info","evidence":"README.md：\"These skills are provided for demonstration and educational purposes only.\" 以及 \"the implementations and behaviors you receive from Claude may differ from what is shown in these skills.\"","source":"ai","title":"声明仅用于演示"},{"code":"skills_include_scripts","severity":"info","evidence":"README.md：\"Skills are folders of instructions, scripts, and resources that Claude loads dynamically\"。在 Claude Code 中，skill 里的脚本可以在你的电脑上运行。","source":"ai","title":"skill 中包含脚本"}],"platforms":["windows","mac","linux","web"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Claude (Claude.ai paid plan, Claude Code login or Anthropic API)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["ai_skills","office","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":192180,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":5074,"growthDays":7,"history":[{"date":"2026-10-26","stars":183991},{"date":"2026-10-27","stars":184255},{"date":"2026-10-28","stars":184518},{"date":"2026-10-29","stars":184802},{"date":"2026-10-30","stars":185324},{"date":"2026-10-31","stars":185896},{"date":"2026-11-01","stars":186393},{"date":"2026-11-02","stars":186774},{"date":"2026-11-03","stars":187106},{"date":"2026-11-04","stars":187383},{"date":"2026-11-05","stars":187713},{"date":"2026-11-06","stars":188128},{"date":"2026-11-07","stars":188672},{"date":"2026-11-08","stars":189502},{"date":"2026-11-09","stars":190785},{"date":"2026-11-10","stars":192180}],"firstPublishedAt":"2026-10-27T01:10:12.600Z","commitSha":"ca45fc496a3436251f191354091d557da31c15b5","verifiedAt":"2026-10-30T01:40:46.229Z","lastCommitAt":"2026-11-06T22:32:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Anthropic's official pack of Claude skills, including the ones that make Word, PDF, PowerPoint and Excel files.","summary":"This is Anthropic's public collection of Agent Skills: folders of instructions, scripts and resources that Claude loads to do a task the same way every time. It has examples for art, music and design, testing web apps, building MCP servers and company communications, plus the document skills behind Claude's Word, PDF, PowerPoint and Excel files. It also holds the Agent Skills specification and a template for your own skills. On Claude.ai paid plans the example skills are already there; in Claude Code you add them with two commands. The README calls them demos for learning.","easeRationale":"Easiest on the web: on Claude.ai paid plans these example skills are already enabled, so you only need the paid account. On Windows, Mac and Linux you first install Claude Code and then run two commands, such as /plugin marketplace add anthropics/skills and /plugin install document-skills@anthropic-agent-skills.","costNote":"The files are free to read. Using them needs Claude: the README says they are already available on Claude.ai paid plans, and Claude Code or the Claude API are paid.","notVerified":["Whether the document skills' scripts need Python or other libraries installed on Windows when used in Claude Code.","How far these versions differ from what Claude.ai actually uses today.","The exact terms for reusing the source-available document skills.","Whether the skills work in AI assistants other than Claude.","How much extra paid AI usage each skill consumes."]},"zh-Hans":{"summaryShort":"Anthropic 官方的 Claude 技能包,包括用于生成 Word、PDF、PowerPoint 和 Excel 文件的技能。","summary":"这是 Anthropic 公开发布的 Agent Skills(智能体技能)合集:这些文件夹里包含指令、脚本和资源,Claude 加载后能以同样的方式反复完成某项任务。其中有艺术、音乐与设计、网页应用测试、构建 MCP 服务器以及公司沟通方面的示例,还有支撑 Claude 生成 Word、PDF、PowerPoint 和 Excel 文件的文档技能。仓库中也包含 Agent Skills 规范,以及供你创建自己技能的模板。在 Claude.ai 的付费计划中,这些示例技能已经内置;在 Claude Code 中,你可以用两条命令添加它们。README 中称它们是用于学习的演示。","easeRationale":"在网页端最简单:在 Claude.ai 的付费计划中,这些示例技能已经启用,你只需要有付费账户。在 Windows、Mac 和 Linux 上,你需要先安装 Claude Code,然后运行两条命令,例如 /plugin marketplace add anthropics/skills 和 /plugin install document-skills@anthropic-agent-skills。","costNote":"这些文件本身免费可读。使用它们需要 Claude:README 说明它们已经在 Claude.ai 付费计划中提供,而 Claude Code 或 Claude API 都是收费的。","notVerified":["在 Claude Code 中使用时,文档技能的脚本是否需要在 Windows 上安装 Python 或其他库。","这些版本与 Claude.ai 当前实际使用的版本相差多少。","重复使用这些源码可见的文档技能的确切使用条款。","这些技能是否能在 Claude 以外的其他 AI 助手中使用。","每个技能会额外消耗多少付费 AI 用量。"]}}},{"id":"R_kgDOT3jedQ","slug":"anywhere-labs__dsh-desktop","name":"DSH Desktop","owner":"anywhere-labs","fullName":"anywhere-labs/dsh-desktop","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"lan_access_without_a_password","severity":"attention","evidence":"README.md 自己对可选局域网设置给出的危险提示：\"opening it to the local network provides no authentication; everyone on the same network can open DSH and operate your computer. Only turn it on in networks you fully trust.\" 该选项默认关闭，网页服务只在本机监听。","source":"ai","title":"局域网访问无需密码"},{"code":"ai_agent_acts_on_your_computer","severity":"attention","evidence":"这个产品是给助手框架套的一层桌面外壳：自带终端，package.json 还引入了上游名为 dsh-sandbox、dsh-sandbox-policy 和 dsh-user-approval 的部件。能编辑文件、运行命令的助手，如果你批准了错误的一步，就可能造成破坏。","source":"ai","title":"AI 助手会在你的电脑上执行操作"},{"code":"not_the_official_deepseek_app","severity":"attention","evidence":"README.md 反复声明这是 \"an independent community open-source project with no affiliation, cooperation, authorisation or endorsement from DeepSeek\"，DeepSeek 和上游团队都没有成员参与，GitHub 贡献者页面上出现的上游名字只是来自继承的分叉历史。\"DeepSeek Harness\" 是 DeepSeek 的注册商标。","source":"ai","title":"不是 DeepSeek 官方应用"},{"code":"prebuilt_packages_committed_to_the_repo","severity":"attention","evidence":"package.json 把二十多个上游包换成了存放在仓库内的压缩包，例如 \"@agents-anywhere/dsh-bridge-next\": \"file:vendor/agents-anywhere/agents-anywhere-dsh-bridge-next-2.1.0-desktop...tgz\"，以及整个 @deepseek-ai/dsh-* 运行环境来自 \"file:vendor/dsh-runtime/0.2.1-rc.3/...\"。因此，运行的一部分是以现成压缩包的形式提供的，而不是在这里由可读源码构建出来的。","source":"ai","title":"预编译包直接放在仓库里"},{"code":"phone_remote_control_built_in","severity":"info","evidence":"README.md：\"Remote control from a phone - built in\"：在 iOS 或 Android 上可以 \"connect remotely to Desktop, start tasks, watch the agent's progress\"。从 2.0.9 版起通过作者自己的 Agents-Anywhere 项目内置。","source":"ai","title":"内置手机远程控制"},{"code":"sends_an_installation_id_on_update_checks","severity":"info","evidence":"README.md：检查更新时会在 X-DSH-Desktop-Installation-Id 请求头里带上已安装的版本、渠道（stable 或 beta）和 \"a random UUID generated and kept on this machine\"；项目称它不是由硬件信息得出的。","source":"ai","title":"检查更新时会发送安装 ID"},{"code":"affiliate_and_sponsor_links","severity":"info","evidence":"README.md 的赞助商表格链接到带有追踪或推荐码的 AI 令牌转售商，例如 \"https://88api.ai/sign-up?aff=VnEb\" 和 \"https://onliqiu.com/r/ds2h\"，所以项目会从它引去的注册中获利。","source":"ai","title":"含推广和赞助商链接"},{"code":"pre_release_builds_on_top","severity":"info","evidence":"最新的三个版本全部来自预发布渠道（v2.0.21-next、v2.0.21-beta.1、v2.0.20-next），插件市场和插件也来自社区来源，所以你安装到的东西变化很快。","source":"ai","title":"最新的版本都是预发布版"},{"code":"mostly_chinese_documentation","severity":"info","evidence":"README、用户指南链接、社群（微信、QQ）以及大部分文档都是中文；虽然有英文 README（README.en.md），但项目是面向说中文的用户写的。","source":"ai","title":"文档大多是中文"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":309,"maxMB":926,"basis":"installer"},"apiKeys":["An AI model provider for the agent (DeepSeek or another model the upstream harness supports)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","development"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true},{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"conductor","name":"Conductor","confirmed":true}],"declaredAlternatives":[],"stars":35628,"downloads":288317,"downloadsWeek":26539,"downloadsDays":7,"growth":915,"growthDays":7,"history":[{"date":"2026-10-26","stars":33383},{"date":"2026-10-27","stars":33613},{"date":"2026-10-28","stars":33836},{"date":"2026-10-29","stars":34028},{"date":"2026-10-30","stars":34174},{"date":"2026-10-31","stars":34281},{"date":"2026-11-01","stars":34416},{"date":"2026-11-02","stars":34542},{"date":"2026-11-03","stars":34713},{"date":"2026-11-04","stars":34861},{"date":"2026-11-05","stars":34996},{"date":"2026-11-06","stars":35154},{"date":"2026-11-07","stars":35277},{"date":"2026-11-08","stars":35388},{"date":"2026-11-09","stars":35514},{"date":"2026-11-10","stars":35628}],"firstPublishedAt":"2026-10-27T01:10:12.601Z","commitSha":"31a96e1750aab260cee87a501e39d78482db3218","verifiedAt":"2026-11-02T02:58:50.182Z","lastCommitAt":"2026-11-06T16:29:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs the DeepSeek Harness AI coding agent as a normal desktop app, with a plugin market and control from your phone.","summary":"DSH Desktop packs DeepSeek Harness, an open AI coding agent that usually runs from the command line, into a normal desktop program. It starts and manages the local service for you, adds a window, a system tray icon, a terminal, automatic updates and a first-run wizard, so you do not have to install Node.js or type commands. It has a built-in plugin market, and from version 2.0.9 you can also connect from an iOS or Android phone to start tasks and watch the agent work. Windows and macOS installers are the official ones; Linux packages appear in the beta and next channels.","easeRationale":"On Windows you run a 309 MB installer and on macOS you drag a universal DMG into Applications, with \"no extra environment needed\" as the README puts it; the extra step on both is the first-run wizard and pointing the agent at a model, so it is a B. Linux is the same grade but less official: the README lists only Windows and macOS, while the releases carry .deb, AppImage and tar.gz files under the beta and next channels.","costNote":"The app is free and open source (MIT) and the README asks you to refuse anyone selling it. The agent inside it needs access to AI models, which you pay for separately; the project's own sponsors are model API resellers.","notVerified":["Which model account the agent needs and what a day of use costs","Whether the archive files in vendor/ match the public source of the upstream project","What the plugins in the built-in market do, since anyone can supply a source","Whether the Linux packages work, since the project documents only Windows and macOS","Real disk use after installing (the installers are 309-926 MB) and what the app sends while it runs"]},"zh-Hans":{"summaryShort":"把 DeepSeek Harness AI 编程助手做成普通桌面应用，带插件市场，还能用手机控制。","summary":"DSH Desktop 把 DeepSeek Harness（一个通常在命令行里运行的开源 AI 编程助手）打包成一个普通的桌面程序。它会替你启动和管理本地服务，并增加窗口、系统托盘图标、终端、自动更新和首次运行向导，所以你不用安装 Node.js，也不用敲命令。内置插件市场，从 2.0.9 版起还可以用 iOS 或 Android 手机连接，发起任务并观看助手工作。Windows 和 macOS 的安装包是正式版；Linux 安装包出现在 beta 和 next 渠道里。","easeRationale":"在 Windows 上运行一个 309 MB 的安装程序，在 macOS 上把通用 DMG 拖进“应用程序”，正如 README 所说 \"no extra environment needed\"；两边多出来的一步是首次运行向导，以及让助手指向某个模型，所以评级是 B。Linux 评级相同，但没那么正式：README 只列出 Windows 和 macOS，而发布的版本在 beta 和 next 渠道里带有 .deb、AppImage 和 tar.gz 文件。","costNote":"这款应用免费且开源（MIT），README 还要求你拒绝任何倒卖它的人。里面的助手需要接入 AI 模型，这部分你要另外付费；项目自己的赞助商都是模型 API 的转售商。","notVerified":["助手需要哪种模型账号，以及用一天要花多少钱","vendor/ 里的压缩包是否与上游项目的公开源码一致","内置市场里的插件会做什么，因为任何人都可以提供来源","Linux 安装包能否正常工作，因为项目只记录了 Windows 和 macOS","安装后的实际磁盘占用（安装包为 309-926 MB），以及应用运行时会发送什么"]}}},{"id":"R_kgDONlCshw","slug":"apify__apify-mcp-server","name":"Apify MCP Server","owner":"apify","fullName":"apify/apify-mcp-server","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"attention","signals":[{"code":"telemetry_on_by_default","severity":"attention","evidence":"README 的遥测部分写道：“Apify MCP 服务器会收集关于工具调用的遥测数据……遥测功能默认是**开启**的”，其中包括作为 `_traceId` 的会话 ID，以及模型为每次搜索写的一句话理由；“stdio 传输方式也会用到 Sentry”。你需要设置 `TELEMETRY_ENABLED=false` 或 `--telemetry-enabled=false` 才能关闭它。package.json 里也把 @segment/analytics-node 和 @sentry/node 列为依赖。","source":"ai","title":"遥测功能默认开启"},{"code":"agent_can_change_or_delete_your_apify_account_data","severity":"attention","evidence":"工具列表里包括 `delete-actor`（“永久从你的账户中删除一个 Actor”）、`delete-schedule`、`delete-actor-version` 和 `apify-api-write`（“向任意 Apify API 路径发送带 JSON 内容的 POST、PUT、PATCH 或 DELETE 请求”）。它们默认不会被加载，但一旦被选中，助手就会用你的令牌来执行操作。","source":"ai","title":"助手可以更改或删除你 apify 账户的数据"},{"code":"collects_personal_data_from_social_networks","severity":"attention","evidence":"README 里的示例提示包括“获取这 20 个 LinkedIn 档案的当前职位、公司和工作履历”、“获取 @nasa 在 Instagram 上最新的 50 条帖子，包括点赞和评论”，以及“在 Google 地图上查找奥斯汀（德克萨斯州）的牙医并附上电话号码”。抓取他人的数据和网站内容，会受到各网站条款的限制，在一些国家还受隐私法律的限制。","source":"ai","title":"会从社交网络收集个人数据"},{"code":"hosted_service_code_not_in_this_repository","severity":"info","evidence":"README 的开发部分写道：“mcp.apify.com 上的托管服务运行的是这套核心逻辑，但代码存放在另一个私有仓库里。”所以推荐的这个选项，其确切代码你在这里是看不到的。","source":"ai","title":"托管服务的代码不在这个仓库里"},{"code":"runs_and_data_happen_on_apify_servers","severity":"info","evidence":"README：“在两种模式下，Actor 仍然都在 Apify 平台上运行并存储结果”，以及“请求和 Actor 的输入会发送到 Apify API，用来运行你调用的 Actor。”","source":"ai","title":"运行和数据都发生在 apify 的服务器上"}],"platforms":["windows","mac","linux","web"],"tools":["node"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Apify (APIFY_TOKEN, or sign-in with OAuth on the hosted server)","Anthropic (ANTHROPIC_API_KEY, only for the project's own test runs)","OpenRouter (OPENROUTER_API_KEY, only for the project's own test runs)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","ai_agents","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":15611,"downloads":11893,"downloadsWeek":129,"downloadsDays":6,"growth":840,"growthDays":6,"history":[{"date":"2026-11-04","stars":14771},{"date":"2026-11-05","stars":14921},{"date":"2026-11-06","stars":15068},{"date":"2026-11-07","stars":15182},{"date":"2026-11-08","stars":15306},{"date":"2026-11-09","stars":15454},{"date":"2026-11-10","stars":15611}],"firstPublishedAt":"2026-11-04T01:36:50.288Z","commitSha":"17e1e61cc3368c5d04ba0274c6da8adc317229d6","verifiedAt":"2026-11-04T01:21:07.412Z","lastCommitAt":"2026-11-10T22:05:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant like Claude or ChatGPT pull data from websites, maps and social media with ready-made scrapers.","summary":"This is an MCP server: a connector you add to an AI assistant such as Claude, ChatGPT, Cursor or VS Code. Once connected, the assistant can search the Apify Store, pick one of thousands of ready-made scrapers (called Actors), run it and read the results, for jobs like Google Maps listings, search results, online shops or social media posts. The scrapers run on Apify's servers, not on your computer. It is for people who want an assistant that fetches web data. You sign in with an Apify account; connecting is free and runs are billed by Apify.","easeRationale":"The easiest route is the hosted server (web): you open mcp.apify.com, copy the generated line or click the one-click link for your assistant, and sign in to Apify in the browser, with nothing to install. Running it on your own machine on Windows, macOS or Linux is harder: the README requires Node.js 22 or later, an `npx @apify/actors-mcp-server` command in the client's configuration file, and your API token pasted in by hand.","costNote":"The server itself is free (MIT). The README says connecting is free and that Actor runs are billed to your Apify account at each Actor's price, with $5 of monthly usage included in the free plan.","notVerified":["Whether the hosted server at mcp.apify.com runs the same code as this repository, which the README says lives in a separate private repository","What an Actor run really costs for your task, and how fast the $5 monthly free credit runs out","Whether the one-click setup actually works in each assistant, since the README says support for OAuth, dynamic tool lists and widgets differs between clients","What the telemetry and Sentry reports contain in practice","How much disk and memory the local server uses, since the figure here is an estimate"]},"zh-Hans":{"summaryShort":"让 Claude 或 ChatGPT 这样的 AI 助手用现成的抓取工具，从网站、地图和社交媒体获取数据。","summary":"这是一个 MCP 服务器：一个你添加到 Claude、ChatGPT、Cursor 或 VS Code 等 AI 助手上的连接器。连接后，助手可以搜索 Apify Store，从数千个现成的抓取工具（叫作 Actor）里选一个，运行它并读取结果，用于诸如 Google 地图信息、搜索结果、网店或社交媒体帖子之类的任务。这些抓取工具运行在 Apify 的服务器上，而不是你的电脑上。它适合想要一个能获取网页数据的助手的人。你需要用 Apify 账号登录；连接是免费的，运行由 Apify 计费。","easeRationale":"最简单的方式是使用托管服务器（网页版）：打开 mcp.apify.com，复制生成的那一行，或者点击你的助手对应的一键链接，然后在浏览器里登录 Apify，不需要安装任何东西。要在你自己的电脑上运行，无论是 Windows、macOS 还是 Linux，都更难一些：README 要求 Node.js 22 或更新版本，要在客户端的配置文件里写一条 `npx @apify/actors-mcp-server` 命令，还要手动粘贴你的 API 令牌。","costNote":"服务器本身是免费的 (MIT)。README 说连接是免费的，Actor 的运行会按每个 Actor 的价格计入你的 Apify 账户，免费计划里包含每月 $5 的用量。","notVerified":["mcp.apify.com 上的托管服务器是否运行着和这个仓库相同的代码，因为 README 说它存放在另一个私有仓库里","对你的任务来说，运行一个 Actor 实际要花多少钱，以及每月 $5 的免费额度多快会用完","一键设置在每个助手里是否真的能用，因为 README 说不同客户端对 OAuth、动态工具列表和小组件的支持不一样","遥测和 Sentry 报告在实际中具体包含什么内容","本地服务器占用多少磁盘和内存，因为这里给出的数字只是估算"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxODA2ODc2MjQ=","slug":"aquasecurity__trivy","name":"Trivy","owner":"aquasecurity","fullName":"aquasecurity/trivy","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"past_supply_chain_compromise","severity":"attention","evidence":"2026 年的事件（由 Wiz 和 Snyk 披露）：2026-02-28 一个组织令牌被盗，2026-03-19 TeamPCP 团伙在 GitHub releases、Docker Hub、GHCR 和 ECR 中发布了被篡改的二进制文件，并重写了 trivy-action 的 76 个标签中的 75 个。维护者已经移除了这些恶意文件。请只下载事件之后发布的版本，并在使用前核对官方公布的 checksum 和签名。","source":"ai","title":"过去发生的供应链被篡改事件"},{"code":"commercial_upsell_in_readme","severity":"info","evidence":"README.md: \"If you liked Trivy, you will love Aqua which builds on top of Trivy to provide even more enhanced capabilities for a complete security management offering.\"","source":"ai","title":"Readme 中的商业推广"},{"code":"canary_builds_offered","severity":"info","evidence":"README.md: \"There are canary builds ... generated with every push to the main branch. Please be aware: canary builds might have critical bugs, so they are not recommended for use in production.\"","source":"ai","title":"提供 canary 构建版本"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","go"],"disk":{"minMB":51,"maxMB":154,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","automation"],"alternativeTo":[{"slug":"snyk","name":"Snyk","confirmed":true},{"slug":"aqua-platform","name":"Aqua Platform","confirmed":true}],"declaredAlternatives":[],"stars":39209,"downloads":66648179,"downloadsWeek":3652929,"downloadsDays":7,"growth":265,"growthDays":7,"history":[{"date":"2026-10-26","stars":38745},{"date":"2026-10-27","stars":38784},{"date":"2026-10-28","stars":38815},{"date":"2026-10-29","stars":38840},{"date":"2026-10-30","stars":38867},{"date":"2026-10-31","stars":38882},{"date":"2026-11-01","stars":38902},{"date":"2026-11-02","stars":38918},{"date":"2026-11-03","stars":38944},{"date":"2026-11-04","stars":38983},{"date":"2026-11-05","stars":39032},{"date":"2026-11-06","stars":39070},{"date":"2026-11-07","stars":39102},{"date":"2026-11-08","stars":39126},{"date":"2026-11-09","stars":39167},{"date":"2026-11-10","stars":39209}],"firstPublishedAt":"2026-10-27T01:10:12.602Z","commitSha":"e2c05f7ae66bedacd4a51594a71cbe95566cc838","verifiedAt":"2026-10-31T01:38:33.980Z","lastCommitAt":"2026-11-10T08:07:08Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Scans software, containers and cloud setups for known security holes, leaked passwords and risky settings.","summary":"Trivy is a security scanner you use by typing commands. It checks container images, folders of code, Git repositories, virtual machine images and Kubernetes clusters for known vulnerabilities, risky configuration, exposed secrets such as passwords or keys, and software licenses, and it can list every component inside (an SBOM). It is for developers and IT or security teams. Windows, Mac and Linux all get a ready program with nothing else to install. It is made by Aqua Security, which also sells a paid platform built on it.","easeRationale":"On Windows the release is a ready program in a zip (trivy_0.76.1_windows-64bit.zip) with no other tools to install, but you unzip it and use it by typing commands in a terminal. Mac and Linux are the same idea, with .deb and .rpm packages on Linux.","costNote":"Free (Apache-2.0). Aqua Security sells a commercial platform built on top of Trivy; the package does not state prices.","notVerified":["How much data it downloads on first run and how often (scanners like this usually fetch a vulnerability database).","Whether scanning container images on Windows needs Docker or another container tool for your case.","How many false alarms or missed problems it reports on your own projects.","Whether the release files you download match the published checksums and signatures.","Whether any copies of the tampered 2026 versions are still circulating in mirrors, caches or third-party channels."]},"zh-Hans":{"summaryShort":"扫描软件、容器和云端配置，查找已知漏洞、泄露的密码和风险设置。","summary":"Trivy 是一款通过输入命令使用的扫描工具。它可以检查容器镜像、代码文件夹、Git 仓库、虚拟机镜像和 Kubernetes 集群，查找已知漏洞、有风险的配置、暴露的密码或密钥等敏感信息，以及软件许可证，还能列出其中的每个组件（即 SBOM）。它面向开发者以及 IT 或风险防护团队。Windows、Mac 和 Linux 都有现成的程序，无需安装其他工具。它由 Aqua Security 开发，该公司还销售基于它构建的付费平台。","easeRationale":"在 Windows 上，发布的版本是压缩包里的现成程序（trivy_0.76.1_windows-64bit.zip），不需要安装其他工具，但你需要解压后在终端里输入命令来使用它。Mac 和 Linux 的方式相同，Linux 上还提供 .deb 和 .rpm 安装包。","costNote":"免费（Apache-2.0 许可）。Aqua Security 销售一款基于 Trivy 构建的商业平台；该软件包未说明价格。","notVerified":["首次运行时会下载多少数据，以及下载频率如何（这类工具通常会获取一个漏洞数据库）。","在 Windows 上扫描容器镜像是否需要 Docker 或其他容器工具，具体取决于你的情况。","在你自己的项目上，它会报告多少错误警报或漏报的问题。","你下载的发布文件是否与官方公布的 checksum 和签名一致。","2026 年遭篡改的版本是否仍有副本在镜像站、缓存或第三方渠道中流传。"]}}},{"id":"R_kgDOVBkRbA","slug":"arahim3__digup","name":"DigUp","owner":"ARahim3","fullName":"ARahim3/DigUp","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"24 天内获得 6093 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"unsigned_windows_installer","severity":"attention","evidence":"README，Install 部分：\"Windows may say it doesn't recognize the app, because the installer isn't signed yet: choose More info, then Run anyway.\"","source":"ai","title":"Windows 安装程序未签名"},{"code":"beta_on_windows_and_linux","severity":"info","evidence":"README 把 Windows 和 Linux 指向 \"beta release\" v0.7.0-beta.3，而最新的稳定版（v0.6.2）只有 Mac 的 DMG。","source":"ai","title":"Windows 和 Linux 仍是 beta"},{"code":"large_model_download_on_first_run","severity":"info","evidence":"README：\"While the model downloads (865 MB, once)\" — 安装包只有 8-140 MB，所以大部分磁盘占用是后来才出现的，再加上索引（一个大型代码项目约 130 MB）。","source":"ai","title":"首次运行要下载大模型"},{"code":"reads_your_personal_files","severity":"info","evidence":"按设计，它会为你选择的文件夹建立索引：\"pictures and screenshots... PDFs, page by page... documents... audio of any length\"。README 称它从不读取密钥、密码文件或 .ssh 这类隐藏文件夹，并且只读取文件，不会修改或上传。","source":"ai","title":"会读取你的个人文件"},{"code":"mac_only_features","severity":"info","evidence":"README 常见问题：在 Windows 和 Linux 上 \"the text inside screenshots isn't read yet\"，而旧的 Word .doc 文件、RAW 照片和 SVG 图片 \"are only read on a Mac\"。","source":"ai","title":"部分功能仅限 Mac"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":39,"maxMB":116,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["productivity","network_system"],"alternativeTo":[{"slug":"houdahspot","name":"HoudahSpot","confirmed":true}],"declaredAlternatives":[],"stars":6360,"downloads":21897,"downloadsWeek":1605,"downloadsDays":7,"growth":174,"growthDays":7,"history":[{"date":"2026-10-26","stars":5869},{"date":"2026-10-27","stars":5922},{"date":"2026-10-28","stars":5977},{"date":"2026-10-29","stars":6013},{"date":"2026-10-30","stars":6048},{"date":"2026-10-31","stars":6065},{"date":"2026-11-01","stars":6093},{"date":"2026-11-02","stars":6138},{"date":"2026-11-03","stars":6186},{"date":"2026-11-04","stars":6236},{"date":"2026-11-05","stars":6261},{"date":"2026-11-06","stars":6286},{"date":"2026-11-07","stars":6307},{"date":"2026-11-08","stars":6323},{"date":"2026-11-09","stars":6341},{"date":"2026-11-10","stars":6360}],"firstPublishedAt":"2026-10-27T01:10:12.603Z","commitSha":"dc2651ad8299160bbe91e972f857e1a1599a8419","verifiedAt":"2026-11-02T02:50:44.096Z","lastCommitAt":"2026-10-26T20:44:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Searches your own files by what is inside them: photos, PDFs, documents, recordings, videos and code.","summary":"DigUp puts a search panel on your Mac, Windows or Linux computer. You pick folders, and it reads the pictures, PDFs, documents, spreadsheets, recordings and videos inside. Then you describe what you want in your own words, and a result opens at the PDF page, the minute of a recording or the line of code. It is free and open source and handles more than 100 languages. The README says indexing stays on your computer. After installing, the app downloads an 865 MB model once. Mac needs an Apple Silicon Mac with macOS 14 or later; the Windows and Linux builds are labelled beta.","easeRationale":"On a Mac you drag the DMG into Applications (or use Homebrew) and the app downloads the 865 MB model by itself at first run, so it is an installer plus one step. Windows and Linux are the same with the setup .exe, .deb or AppImage, except that those builds are still labelled beta and Windows shows an unknown-publisher warning because the installer is not signed.","costNote":"Free and open source (MIT). The author has an optional donation link.","notVerified":["Whether nothing really leaves the computer apart from the one-time model download and the daily update check.","Real disk use after the 865 MB model plus the index for your own folders.","How long the first pass actually takes on your machine, especially on Windows or Linux without a supported graphics card.","Whether the beta Windows and Linux builds work as well as the Mac version.","Whether the folder and file-type exclusions (keys, .ssh, cloud-only files, code projects) hold in practice."]},"zh-Hans":{"summaryShort":"按内容搜索你自己的文件：照片、PDF、文档、录音、视频和代码。","summary":"DigUp 在你的 Mac、Windows 或 Linux 电脑上放一个搜索面板。你选好文件夹，它就会读取里面的图片、PDF、文档、电子表格、录音和视频。然后你用自己的话描述想找的东西，结果会直接打开到 PDF 的那一页、录音的那一分钟或代码的那一行。它免费开源，支持 100 多种语言。README 说索引留在你的电脑上。安装后，应用会一次性下载一个 865 MB 的模型。Mac 需要搭载 Apple Silicon 并运行 macOS 14 或更高版本；Windows 和 Linux 版本标注为 beta。","easeRationale":"在 Mac 上把 DMG 拖进“应用程序”（或用 Homebrew），应用首次运行时会自己下载 865 MB 的模型，所以是安装包加一步操作。Windows 和 Linux 也一样，分别用安装用的 .exe、.deb 或 AppImage，只不过这些版本仍标注为 beta，而且 Windows 会因为安装程序没有签名而提示发布者未知。","costNote":"免费开源（MIT）。作者设有可选的捐赠链接。","notVerified":["除了一次性的模型下载和每天的更新检查，是否真的没有任何内容离开你的电脑。","加上 865 MB 的模型和你自己文件夹的索引之后，实际的磁盘占用。","在你的电脑上第一遍处理实际要花多久，尤其是在没有受支持显卡的 Windows 或 Linux 上。","Windows 和 Linux 的 beta 版本是否和 Mac 版一样好用。","对文件夹和文件类型的排除规则（密钥、.ssh、仅在云端的文件、代码项目）在实际使用中是否有效。"]}}},{"id":"R_kgDOVdca1A","slug":"arjun-ingole__hop","name":"Hop","owner":"Arjun-Ingole","fullName":"Arjun-Ingole/hop","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"repartitions_your_system_disk","severity":"attention","evidence":"hop 会在 Windows 启动所在的磁盘上缩小分区并创建新分区（README.md：\"Shrink C: or another drive, use free space, or use a second disk inside the PC\"，还会 \"Add a startup entry with the ISO's signed boot loader\"）。它说 \"nothing on the disk changes until those checks pass\"，但任何会移动分区和启动项的工具，都可能让电脑无法启动。","source":"ai","title":"会对系统磁盘重新分区"},{"code":"can_erase_windows","severity":"attention","evidence":"README.md：\"Can replace Windows entirely, for the systems where hop has seen it work... It's a separate choice with its own warning, and you have to type ERASE.\" 这条路径会删除磁盘上的所有内容。","source":"ai","title":"可以完全抹掉 Windows"},{"code":"asks_you_to_weaken_boot_protections","severity":"attention","evidence":"README.md：\"CachyOS, Arch, Omarchy and Pop!_OS need Secure Boot off\"，以及 \"Omarchy also needs BitLocker fully off on its disk, not paused\"。关闭 Secure Boot 和 BitLocker 会降低电脑自身的保护，而 hop 不会替你完成这一步。","source":"ai","title":"要求你削弱系统的启动保护"},{"code":"unsigned_program_needing_administrator","severity":"attention","evidence":"README.md：\"Windows asks for administrator access, and SmartScreen may warn because hop isn't code-signed yet: click More info → Run anyway.\" 也就是说，一个能完全访问你磁盘的程序，却没有能说明是谁制作的签名。","source":"ai","title":"未签名的程序却要求管理员权限"},{"code":"downloaded_iso_not_always_checksum_verified","severity":"info","evidence":"README.md：\"It gets the ISO from the project's own server and verifies its checksum where the project publishes one.\" 当某个 Linux 项目没有发布校验值时，下载内容就不会用这种方式被检查。","source":"ai","title":"下载的 ISO 并非总是经过校验"},{"code":"very_new_project_early_version","severity":"info","evidence":"该仓库创建于 2026-10-30，最新版本是 v0.3.2，几天内发布了 6 个版本，已有 1,390 个星标。README 还列出了目前还不能用的功能（不支持 Kubuntu；Debian、CachyOS、Linux Mint 和 Zorin OS \"can't go on a second disk yet\"）。","source":"ai","title":"项目非常新，版本也还很早期"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":1,"maxMB":2,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system"],"alternativeTo":[{"slug":"easeus-partition-master","name":"Partition Master","confirmed":true}],"declaredAlternatives":[],"stars":1639,"downloads":3243,"downloadsWeek":1797,"downloadsDays":7,"growth":185,"growthDays":7,"history":[{"date":"2026-11-01","stars":1225},{"date":"2026-11-02","stars":1387},{"date":"2026-11-03","stars":1454},{"date":"2026-11-04","stars":1503},{"date":"2026-11-05","stars":1526},{"date":"2026-11-06","stars":1561},{"date":"2026-11-07","stars":1594},{"date":"2026-11-08","stars":1613},{"date":"2026-11-09","stars":1626},{"date":"2026-11-10","stars":1639}],"firstPublishedAt":"2026-11-01T01:55:44.432Z","commitSha":"d3ad3f1a883f162c0acae30052aa666c1c731407","verifiedAt":"2026-11-03T02:43:32.962Z","lastCommitAt":"2026-11-09T09:24:28Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"hop.exe","sizeBytes":768512,"url":"https://github.com/Arjun-Ingole/hop/releases/download/v0.3.3/hop.exe","releaseUrl":"https://github.com/Arjun-Ingole/hop/releases/tag/v0.3.3"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Installs Linux next to Windows with no USB stick: it makes room on your disk and starts the Linux installer.","summary":"hop is a single Windows program that sets up a Linux system beside Windows without a USB stick. You pick a Linux system from a list, choose a drive, drag a slider for how much space it gets and click Install: hop downloads the ISO, frees space on the disk, copies the installer onto a small partition and restarts the PC into it. It can also undo everything and give the space back, or wipe Windows completely if you type ERASE. It needs Windows 10 (1803 or newer) or 11 on a UEFI PC with a GPT disk.","easeRationale":"On Windows you download one 1 MB `hop.exe` from the releases page and run it, with nothing to install first; the app asks for administrator access and, because it is not code-signed yet, you have to click past the SmartScreen warning.","costNote":"Free and MIT licensed; the Linux systems it installs are free too.","notVerified":["Whether it works on your particular PC, firmware and disk layout","Whether Undo really gives all the space back after a failed attempt","What happens to a BitLocker-encrypted drive or to Windows recovery after the disk is resized","Whether each listed Linux system installs cleanly, since the list is based on what the author has seen work","Whether the downloaded ISO is the genuine one when the Linux project publishes no checksum"]},"zh-Hans":{"summaryShort":"无需 U 盘即可在 Windows 旁安装 Linux：它会在磁盘上腾出空间，并启动 Linux 安装程序。","summary":"hop 是一个单独的 Windows 程序，可以在不使用 U 盘的情况下，在 Windows 旁边安装一个 Linux 系统。你从列表中选择一个 Linux 系统，选择一个磁盘，拖动滑块决定分配多少空间，然后点击安装：hop 会下载 ISO、在磁盘上腾出空间、把安装程序复制到一个小分区，并重启电脑进入该分区。它也可以撤销所有操作并把空间还回去，或者在你输入 ERASE 后完全抹掉 Windows。它需要运行在采用 GPT 磁盘的 UEFI 电脑上的 Windows 10（1803 或更新版本）或 Windows 11。","easeRationale":"在 Windows 上，你只需从发布页面下载一个 1 MB 的 `hop.exe` 并运行，不需要先安装任何东西；该应用会请求管理员权限，而且因为还没有代码签名，你需要点过 SmartScreen 的警告才能继续。","costNote":"免费，采用 MIT 许可；它安装的 Linux 系统也都是免费的。","notVerified":["它能否在你具体的电脑、固件和磁盘布局上正常运行","撤销操作在失败尝试之后是否真的能把空间全部还回来","调整磁盘大小后，使用 BitLocker 加密的磁盘或 Windows 恢复分区会发生什么","列表中的每个 Linux 系统是否都能顺利安装，因为该列表只是基于作者看到的能跑通的情况","当某个 Linux 项目没有发布校验值时，下载的 ISO 是否就是原始文件"]}}},{"id":"R_kgDOPFB9Lw","slug":"armanshirzad__socialmediacontentcreationandpostingautomationplatform","name":"Social Media Content Automation Platform","owner":"ArmanShirzad","fullName":"ArmanShirzad/SocialMediaContentCreationAndPostingAutomationPlatform","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞(OSV):aiohttp: GHSA-27mf-ghqm-j3j8, GHSA-2fqr-mr3j-6wp8, GHSA-2g87-pp6c-29x5; anyio: GHSA-5p39-cfhj-2xmp, GHSA-82r6-8w77-94w6, PYSEC-2026-4024; certifi: GHSA-6q48-6c7w-9gw2, PYSEC-2026-4320; filelock: GHSA-69hj-f68j-2879, GHSA-8mx5-7m5m-r77g, GHSA-qmgc-5h2g-mvrw; h11: GHSA-vqfr-h8mv-ghfj, PYSEC-2026-348; idna: GHSA-65pc-fj4g-8rjx, PYSEC-2026-215; instagrapi: GHSA-ggxf-37hm-9wqf; jinja2: GHSA-cpwx-vrp4-4pq7, GHSA-gmj6-6f8f-6699, GHSA-q2x7-8rv6-6q7h; oauthlib: GHSA-7984-qgvg-3j9c, GHSA-hj66-6f7g-4r5v, GHSA-xpv3-w29h-x7cv; pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h; pyasn1: GHSA-2qm8-x737-gmw5, GHSA-63vm-454h-vhhq, GHSA-8ppf-4f7h-5ppj; pycryptodomex: GHSA-j225-cvw7-qrx7, PYSEC-2024-3; pydantic: GHSA-mr82-8j83-vxmv, PYSEC-2026-1812; pytest: GHSA-6w46-j5rx-g56g, PYSEC-2026-1845; python-dotenv: GHSA-mf9w-mj56-hr94, PYSEC-2026-2270; requests: GHSA-9hjg-9r4m-mvj7, GHSA-gc5v-m9x4-r6x2, PYSEC-2026-1872; urllib3: GHSA-2xpw-w6gg-jr37, GHSA-38jv-5279-wg99, GHSA-48p4-8xcf-vxj5","source":"check","title":"存在已知漏洞"},{"code":"unofficial_instagram_login","severity":"attention","evidence":"requirements.txt锁定了\"instagrapi==2.0.0\"和\"instabot==0.117.0\",这些库通过Instagram的私有应用接口,用你的用户名和密码登录,而不是Meta官方API。README.md说这个队列的作用是\"to comply with Instagram's strict rate limits and avoid account bans\",说明作者自己也预期账号会有风险。","source":"ai","title":"使用非官方方式登录Instagram"},{"code":"republishes_other_peoples_articles","severity":"attention","evidence":"README.md: \"Fetches the latest articles from a source in this case is the beloved renowned TechCrunch\",然后自动翻译、总结并发布,结果出现在一个公开频道上(\"https://www.youtube.com/@techcrunchfarsi\")。未经许可转载他人的文章和图片,责任在使用者自己。","source":"ai","title":"转载他人文章"},{"code":"abandoned_after_one_day","severity":"attention","evidence":"仓库创建于2025-07-01,当天就有push;494天后仍然没有新的提交,也没有发布版本。","source":"ai","title":"一天后就被放弃"},{"code":"secrets_kept_in_project_files","severity":"attention","evidence":"README.md Setup: \"Place your Google API client secret JSON in the root directory\"、\"Edit files in the config/ directory as needed (e.g., for blacklists, whitelists, credentials)\",以及\"For Instagram, set credentials in config.py\"。五个服务的登录信息,包括一个Instagram密码,最终都以明文形式存放在项目文件夹里。","source":"ai","title":"密钥保存在项目文件里"},{"code":"readme_points_to_files_that_are_missing","severity":"info","evidence":"README.md让你去修改\"groqtranslator.py\"里的提示词,并在\"config.py\"里设置凭据,但这两个文件在仓库里都不存在;发布逻辑其实写在名为test_x_posting.py和test_youtube_posting.py的文件里,README自己也把它们列为X和YouTube的模块。","source":"ai","title":"README指向缺失的文件"},{"code":"republishes_scraped_news_articles","severity":"high","evidence":"README.md: \"Fetches the latest articles from a source in this case is the beloved renowned TechCrunch\",然后\"Translates articles to Persian using LLM-powered translation\"、\"Creates videos from article content using stock images or footage\"以及\"Posts to YouTube, Telegram, X (Twitter), and optionally Instagram\",每6小时一次,结果发布在公开频道\"https://www.youtube.com/@techcrunchfarsi\"上;项目还附带了一个专门针对该网站的抓取器(\"scraper.py # TechCrunch scraper\")。文章和图片属于这家新闻网站,版权法和该网站的条款决定了哪些内容可以被复制和转载。Instagram这一步使用\"instagrapi==2.0.0\"和\"instabot==0.117.0\",它们通过Instagram的私有应用接口,用你的用户名和密码登录,而不是Meta官方API,Instagram的条款禁止通过非官方接口访问;README说这个队列的作用是\"to comply with Instagram's strict rate limits and avoid account bans\"。","source":"ai","title":"转载抓取来的新闻文章"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg"],"buildTools":["python","ffmpeg"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["Groq (translation and summaries)","Pexels (stock images and footage)","Google / YouTube Data API (OAuth client secret)","Telegram Bot API","X (Twitter) API"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation","video"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"pictory","name":"Pictory","confirmed":true}],"declaredAlternatives":[],"stars":51,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":1,"growthDays":6,"history":[{"date":"2026-11-08","stars":51},{"date":"2026-11-09","stars":51},{"date":"2026-11-10","stars":51}],"firstPublishedAt":"2026-11-07T22:56:16.685Z","commitSha":"b7e757c4608bdba7bcb51b2eb200c56e171c738d","verifiedAt":"2026-11-07T22:43:24.364Z","lastCommitAt":"2025-07-01T16:08:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Scripts that grab news articles, translate and summarize them, build short videos and post them to your social accounts.","summary":"A set of Python scripts that read the newest TechCrunch articles, translate and summarize them with an AI service, build a short video from stock footage and captions, and then post the result to Telegram, X, YouTube and, if you want, Instagram, every six hours. It was written for one Persian-language channel and is not a finished product: you clone it, install the packages, and set up keys for Groq, Pexels, Google/YouTube, Telegram and X yourself. The voiceover step is described as a placeholder.","easeRationale":"On all three systems the steps are the same: install Python, create a virtual environment, install a long list of packages, drop a Google client-secret file in the folder and fill in keys for Groq, Pexels, Telegram, X and Instagram before running main.py from the terminal. There is no window to click, and the README's setup section points to configuration files that are not in the repository.","costNote":"The code is free, but posting to X needs X's paid API access, and the AI and stock-footage services have their own free limits.","notVerified":["Whether the pipeline still runs, since the pinned packages are from late 2024 and the APIs for X, Instagram and YouTube have changed.","What it costs each month once X API access and the AI service are counted.","Whether the Instagram part works without getting the account blocked.","Whether the voiceover step does anything, since the README calls it a placeholder.","How the videos it makes actually look, beyond the author's own channel."]},"zh-Hans":{"summaryShort":"脚本自动抓取新闻文章,翻译并总结,制作成短视频后发布到你的社交账号。","summary":"一组Python脚本,读取TechCrunch最新文章,用AI服务翻译并总结,用素材库画面和字幕制作成短视频,然后每六小时发布到Telegram、X、YouTube,以及你想要的话还有Instagram。它是为一个波斯语频道写的,并不是一个成品:你需要自己克隆代码、安装依赖包,并配置Groq、Pexels、Google/YouTube、Telegram和X的密钥。配音这一步在说明里被称为占位功能。","easeRationale":"在三个系统上步骤都一样:安装Python、创建虚拟环境、安装一长串依赖包、把Google的client-secret文件放进文件夹,并在终端运行main.py之前填好Groq、Pexels、Telegram、X和Instagram的密钥。没有窗口可点,README的设置部分还指向了仓库里并不存在的配置文件。","costNote":"代码本身免费,但发布到X需要X的付费API权限,AI服务和素材库服务也各有自己的免费额度上限。","notVerified":["这套流程是否还能运行,因为锁定的依赖包来自2024年末,而X、Instagram和YouTube的API都已经变化。","算上X的API权限和AI服务后,每月实际花费是多少。","Instagram那部分是否能在不被封号的情况下运作。","配音这一步是否真的起作用,因为README把它称为占位功能。","除了作者自己的频道之外,它生成的视频实际效果如何。"]}}},{"id":"R_kgDOVea27Q","slug":"arpanjot-coding__avault","name":"AVault","owner":"arpanjot-coding","fullName":"arpanjot-coding/AVault","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{},"risk":"high","signals":[{"code":"call_recording_legality_varies_by_country","severity":"attention","evidence":"LICENSE 文件中有一段免责声明，提到 \"Violations of national, federal, state, provincial, or international telecommunications and wiretapping laws (including single-party or all-party consent statutes)\"，README 说使用目的 \"primarily for lawful research, education and consent-based evaluation\"。在许多国家，未经对方同意录音属于违法行为；该应用不会询问或记录对方的同意。","source":"ai","title":"录制电话在不同国家的合法性不同"},{"code":"license_notices_contradict_each_other","severity":"attention","evidence":"README 的 \"Licensing notes\" 部分写道：\"The repository currently includes an MIT LICENSE file, while many source files contain GPL-3.0-or-later notices\"，以及 \"Maintainer clarification of the repository-wide licensing presentation remains necessary before representing the whole project under a single license.\" 自动检测只读取了那份 MIT 文件。","source":"ai","title":"许可证说明相互矛盾"},{"code":"code_appears_taken_from_another_project","severity":"attention","evidence":"README 中所有的代码路径都在 \"app/src/main/java/com/kitsumed/shizucallrecorder/\" 下，这个包名属于另一位作者和另一个项目，而 LICENSE 写的是 \"Copyright (c) 2026 AVault Project\"，GitHub 也没有把该仓库标记为 fork。这些文件里的 GPL-3.0-or-later 声明也指向同样的结论。","source":"ai","title":"代码看起来是从别的项目拿来的"},{"code":"release_apk_is_a_debug_build","severity":"attention","evidence":"README 的 \"Validation and compatibility\" 部分写道：\"The downloadable APK is explicitly a debug-signed development build.\" 调试版本使用的是开发密钥签名，而不是项目的发布密钥。","source":"ai","title":"发布的 APK 是调试版本"},{"code":"main_feature_not_tested_on_a_real_call","severity":"attention","evidence":"README 的 \"Validation and compatibility\" 部分写道：\"A live-call audio capture test was not performed during this refresh.\" 列出的检查只是用示例应用和生成的静音做的界面检查。","source":"ai","title":"主要功能没有在真实通话中测试过"},{"code":"works_around_android_call_recording_block","severity":"info","evidence":"应用描述写着 \"Rootless call recording via Shizuku\"，README 解释录音是 \"through the Shizuku shell service\" 完成的。Shizuku 通过调试权限赋予应用特权级 shell 能力，这正是它能获取 Android 不向普通应用公开的音频的方式。","source":"ai","title":"绕过 Android 对通话录音的限制"},{"code":"very_new_repository","severity":"info","evidence":"该仓库创建于 2026-10-31，已经有 541 个星标，两个版本（v2.0.0 和 v2.0.1）都在三天内发布。目前还没有可供参考的历史记录。","source":"ai","title":"仓库非常新"},{"code":"records_calls_without_a_consent_step","severity":"high","evidence":"应用描述写着 \"Rootless call recording via Shizuku\"。README 说 AVault \"does not obtain consent or determine legal permission for you\"，LICENSE 对 \"Interception, recording, storage, decryption, or transmission of telephone, VoIP, or cellular communications without appropriate legal consent\" 和 \"Violations of national, federal, state, provincial, or international telecommunications and wiretapping laws (including single-party or all-party consent statutes)\" 都不承担责任。许多国家的法律要求通话的一方或所有参与方同意后才能录音，而这款应用没有任何询问对方的步骤。","source":"ai","title":"录音没有征得对方同意的步骤"}],"platforms":[],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["audio","network_system"],"alternativeTo":[{"slug":"cube-acr","name":"Cube ACR","confirmed":true}],"declaredAlternatives":[],"stars":621,"downloads":484,"downloadsWeek":281,"downloadsDays":7,"growth":58,"growthDays":7,"history":[{"date":"2026-11-02","stars":540},{"date":"2026-11-03","stars":563},{"date":"2026-11-04","stars":581},{"date":"2026-11-05","stars":599},{"date":"2026-11-06","stars":599},{"date":"2026-11-07","stars":620},{"date":"2026-11-08","stars":619},{"date":"2026-11-09","stars":619},{"date":"2026-11-10","stars":621}],"firstPublishedAt":"2026-11-02T03:04:39.235Z","commitSha":"ae54151f0889b900e62c859f0255e88d55f4a208","verifiedAt":"2026-11-03T02:37:08.352Z","lastCommitAt":"2026-11-02T11:12:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android app that records your phone calls without root and keeps the audio in a PIN-locked, encrypted vault on the phone.","summary":"AVault is an Android app that records phone calls and stores them in an encrypted library on the phone. You can search recordings by contact, number or calling app, lock the app with a PIN or fingerprint, bookmark recordings and set an auto-delete period of 7 to 365 days. It needs Android 11 or later and the separate Shizuku app, which you start through wireless debugging or an ADB command from a computer. Recording calls needs the other person's consent in many countries. Whether both sides of a call are captured depends on your phone and calling app.","easeRationale":"It is an Android-only app, so none of the desktop systems get a grade. On Android you install the APK from the Releases page, then install the separate Shizuku app and start its service through wireless debugging or an ADB command typed on a computer, so there are several steps before the first recording.","notVerified":["Whether recording works on your phone, Android version and calling app — the project says it did not test a live call","Whether both sides of a conversation are captured, or only your side","Whether the encryption key is really held by your phone's security chip, which the README says depends on the device","How much phone storage a month of recordings uses","Whether the APK offered in Releases was built from the code shown in the repository"]},"zh-Hans":{"summaryShort":"一款无需 root 就能录制手机通话的 Android 应用，把录音保存在手机本地、用 PIN 锁定并加密的保险库里。","summary":"AVault 是一款 Android 应用，可以录制电话通话并存储在手机本地的加密资料库中。你可以按联系人、号码或通话应用搜索录音，用 PIN 或指纹锁定应用，为录音添加书签，并设置 7 到 365 天的自动删除周期。它需要 Android 11 或更新版本，以及一个独立的 Shizuku 应用，你可以通过无线调试或在电脑上运行 ADB 命令来启动它。在许多国家，录制通话需要对方同意。是否能同时录到通话双方，取决于你的手机和通话应用。","easeRationale":"这是一款仅支持 Android 的应用，所以桌面系统都没有评级。在 Android 上，你需要从 Releases 页面安装 APK，再安装独立的 Shizuku 应用，并通过无线调试或在电脑上输入 ADB 命令启动它的服务，所以在第一次录音之前要完成好几个步骤。","notVerified":["录音功能是否能在你的手机、Android 版本和通话应用上正常工作——项目方说他们没有测试过真实通话","是否能同时录到对话双方，还是只录到你自己这一侧","加密密钥是否真的由手机的硬件密钥芯片保管，README 说这取决于具体设备","一个月的录音会占用多少手机存储空间","Releases 中提供的 APK 是否是用仓库里展示的代码构建的"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzMTIyNDk1MA==","slug":"artesaos__seotools","name":"SEOTools","owner":"artesaos","fullName":"artesaos/seotools","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"low","signals":[{"code":"docs_reference_retired_tooling","severity":"info","evidence":"README.md里仍然挂着travis-ci.org的构建徽章,仓库里也还留着.travis.yml文件,尽管这项服务已经关闭;安装说明也是从Laravel 5.5和5.8讲起,Laravel 11+只是作为附注提到。","source":"ai","title":"文档引用了已停用的工具"},{"code":"copied_comments_in_the_docs","severity":"info","evidence":"README.md写道\"With Twitter you can create OpenGraph tags to the head\",JsonLd和JsonLdMulti章节把每个方法都描述成\"of twitter card tag\"(例如\"JsonLd::setTitle($type); // title of twitter card tag\"),说明文档有些部分是复制过来却没有改对。","source":"ai","title":"文档里有复制粘贴的注释"},{"code":"releases_slowed_down","severity":"info","evidence":"最近一次提交和最近一次发布都在2026-03-28,距今224天,再往前一次发布是2025-03-14,如今的更新大约一年才有一次,还有33个未解决的issue。","source":"ai","title":"发布节奏变慢"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","social_marketing"],"alternativeTo":[],"declaredAlternatives":[],"stars":3370,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":2,"growthDays":6,"history":[{"date":"2026-11-08","stars":3369},{"date":"2026-11-09","stars":3369},{"date":"2026-11-10","stars":3370}],"firstPublishedAt":"2026-11-07T22:56:16.696Z","commitSha":"c92d09f0ad4491181863850a3ab2e1534361409b","verifiedAt":"2026-11-07T22:43:55.121Z","lastCommitAt":"2026-03-28T12:37:33Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Laravel add-on that writes the title, description and link-preview tags that search engines and social networks read.","summary":"A PHP package for websites built with Laravel or Lumen. Instead of writing head tags by hand, you set the title, description, keywords, canonical link, Open Graph and Twitter Card properties and JSON-LD data in your code, and the package prints the right HTML. It is installed with Composer inside an existing project, so it is only useful to someone who writes the site's code. It has been maintained since 2015, with 35 releases, and does not cover sitemaps, RSS or trailing-slash URLs.","easeRationale":"There is nothing to install and open: on any system you need an existing Laravel or Lumen project, PHP and Composer, then \"composer require artesaos/seotools\", registering a service provider, publishing a config file with artisan and writing PHP in your controllers and views.","notVerified":["Whether it works with the newest Laravel and PHP versions, since the README's examples target older ones.","Whether the tags it prints match what Google, Facebook and X expect today.","How much of the open issue list affects everyday use."]},"zh-Hans":{"summaryShort":"一个Laravel扩展包,生成搜索引擎和社交网络读取的标题、描述和链接预览标签。","summary":"一个面向Laravel或Lumen构建的网站的PHP包。你不用手写head标签,只需在代码里设置标题、描述、关键词、规范链接(canonical)、Open Graph和Twitter Card属性以及JSON-LD数据,这个包就会输出正确的HTML。它通过Composer安装到已有项目里,所以只对编写网站代码的人有用。它从2015年起持续维护,有35个发布版本,但不涵盖站点地图(sitemap)、RSS或结尾斜杠的URL。","easeRationale":"没有可以直接安装打开的东西:任何系统上都需要一个已有的Laravel或Lumen项目、PHP和Composer,然后运行\"composer require artesaos/seotools\"、注册一个service provider、用artisan发布配置文件,并在控制器和视图里编写PHP代码。","notVerified":["是否兼容最新版本的Laravel和PHP,因为README的示例针对的是较旧的版本。","它输出的标签是否符合Google、Facebook和X如今的要求。","未解决的issue列表中有多少会影响日常使用。"]}}},{"id":"R_kgDOMHLb-A","slug":"artipub__artipub","name":"ArtiPub","owner":"artipub","fullName":"artipub/artipub","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"your_publishing_tokens_go_into_the_config","severity":"info","evidence":"这个工具会代表你发布到「Notion, Medium, Dev.to, etc.」(README.md 的 Features 部分),这意味着每个账号的令牌都要存放在工具能读取的地方。这些令牌能让这台电脑上的任何程序写入你的 Notion 工作区,或以你的名义发布文章。","source":"ai","title":"你的发布令牌会写进配置文件"},{"code":"no_changes_for_about_eleven_months","severity":"info","evidence":"检查结果:lastCommitAt 为 2025-12-07,距本次评审 334 天,最新版本 v1.1.10 也发布于同一天。Notion、Medium 和 Dev.to 的发布接口会按各自的节奏变化,某个平台可能会失效却等不到修复。","source":"ai","title":"大约十一个月没有更新"},{"code":"small_single_author_project","severity":"info","evidence":"项目数据:90 颗星和 11 次 fork,许可证和 package.json 都只列出一位作者(Potter.yan)。外部使用较少,意味着某个平台的问题可能还没被发现。","source":"ai","title":"由单一作者维护的小项目"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Notion","Dev.to","Medium","Image hosting service"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"zapier","name":"Zapier","confirmed":false}],"declaredAlternatives":[],"stars":90,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":90},{"date":"2026-11-08","stars":90},{"date":"2026-11-09","stars":90},{"date":"2026-11-10","stars":90}],"firstPublishedAt":"2026-11-06T22:59:23.718Z","commitSha":"06531e7f97df83d6bf2bc34bc6711d11e08e5ae2","verifiedAt":"2026-11-06T22:42:14.479Z","lastCommitAt":"2025-12-07T03:07:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Takes one Markdown article and posts it to Notion, Medium and Dev.to at once, handling the local images.","summary":"ArtiPub is a Node.js library with a command-line tool for people who write articles in Markdown and want them on several sites without copying and pasting. You point it at a Markdown file and it uploads the local images to an image host, swaps the links and publishes the article to platforms such as Notion, Medium and Dev.to; plugins and middleware let you change the text along the way. You need Node.js, a terminal and an access token for each platform you publish to. There is no window to click, and the docs live on a separate site.","easeRationale":"The same on Windows, Mac and Linux: install Node.js, install the @artipub/cli or @artipub/core package from npm and set up a config with a token for each platform, then run it in a terminal. No installer and no graphical interface; the repository itself is a pnpm workspace meant for developers.","notVerified":["Which platforms really work in the current version, since the README says \"including but not limited to Notion, Medium, Dev.to\" without stating what is finished","Whether publishing the same article twice updates the existing post or creates a copy","Where the images end up and what the image host keeps, as that depends on the service you configure","Whether the npm packages match the code in this repository"]},"zh-Hans":{"summaryShort":"把一篇 Markdown 文章一次性发布到 Notion、Medium 和 Dev.to,并自动处理本地图片。","summary":"ArtiPub 是一个 Node.js 库,带有命令行工具,适合用 Markdown 写文章并想发布到多个网站而不想手动复制粘贴的人。你指定一个 Markdown 文件,它会把本地图片上传到图床,替换链接,然后把文章发布到 Notion、Medium、Dev.to 等平台;插件和中间件可以在发布过程中修改文本。你需要 Node.js、一个终端,以及要发布的每个平台的访问令牌。没有可点击的窗口,文档也放在另一个网站上。","easeRationale":"在 Windows、Mac 和 Linux 上步骤相同:安装 Node.js,从 npm 安装 @artipub/cli 或 @artipub/core 包,为每个平台配置一个令牌,然后在终端里运行。没有安装程序,也没有图形界面;这个仓库本身就是一个面向开发者的 pnpm workspace。","notVerified":["当前版本里哪些平台真的能用,因为 README 只说「including but not limited to Notion, Medium, Dev.to」,没说哪些已经做完","重复发布同一篇文章是更新已有的帖子,还是会生成一份新的","图片最终存放在哪里、图床保留了什么,这取决于你配置的服务","npm 上的包是否与这个仓库里的代码一致"]}}},{"id":"R_kgDOVfieRQ","slug":"assassincreator__steam-stash-value-report-card","name":"Steam Stash Value Report Card","owner":"AssassinCreator","fullName":"AssassinCreator/steam-stash-value-report-card","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号是 80 天前创建的","source":"check","title":"所有者账号注册时间较短"},{"code":"contradictory_platform_claims","severity":"info","evidence":"README 的徽章写着 \"runs on Windows · macOS · Linux\"，\"Where to run it\" 部分也写着 \"home Linux and macOS work fine too\"，但 Install 部分却说 \"This is a Windows build... The CLI refuses to start on other systems\"。唯一发布的文件也只支持 Windows。","source":"ai","title":"关于支持平台的说法相互矛盾"},{"code":"unsigned_windows_executable","severity":"attention","evidence":"README 的下载说明写道：\"The build isn't code-signed yet, so SmartScreen may warn you the first time. Choose More info, then Run anyway.\" Windows 无法确认这个文件是谁制作的。","source":"ai","title":"未签名的 Windows 可执行文件"},{"code":"optional_steam_session_cookie","severity":"attention","evidence":"README 的 Privacy 部分写道，从 `STASHCARD_STEAM_LOGIN_SECURE` 读取的可选 cookie \"is a full Steam session: only add it if you understand that\"。它会被保存在 `secrets.json` 中。主要功能即使没有它也能使用。","source":"ai","title":"可选的 Steam 会话 cookie"},{"code":"very_new_repository","severity":"info","evidence":"项目数据显示，该仓库创建于 2026-11-02，与它唯一的版本 v1.0.0 同一天发布，目前已有 608 个星标。目前还没有可供参考的历史记录。","source":"ai","title":"仓库非常新"},{"code":"prices_may_be_wrong_or_stale","severity":"info","evidence":"README 的免责声明写道：\"Prices are informational estimates from public market data, not offers, appraisals or financial advice, and can be stale or wrong.\" 计算总价用到的市场手续费在 `markets.py` 中被标记为 \"TODO(verify)\"。该项目与 Valve 没有关联。","source":"ai","title":"价格可能有误或已过期"},{"code":"reads_data_about_other_people","severity":"info","evidence":"使用示例中包含 `stashcard diff yourname theirname`，用来比较两个账号。README 说别人的卡片默认会打上模糊（`--redact full`），但另一个人库存的数字仍会保留在与图片一起保存的 `card.json` 文件中。","source":"ai","title":"读取其他人的数据"}],"platforms":["windows"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":8,"maxMB":24,"basis":"installer"},"apiKeys":["Steam Web API key (optional, for library value only)","Steam login cookie (optional, for Steam sales history)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games","image"],"alternativeTo":[],"declaredAlternatives":[],"stars":789,"downloads":40,"downloadsWeek":39,"downloadsDays":7,"growth":88,"growthDays":7,"history":[{"date":"2026-11-03","stars":701},{"date":"2026-11-04","stars":744},{"date":"2026-11-05","stars":802},{"date":"2026-11-06","stars":786},{"date":"2026-11-07","stars":791},{"date":"2026-11-08","stars":788},{"date":"2026-11-09","stars":788},{"date":"2026-11-10","stars":789}],"firstPublishedAt":"2026-11-03T02:51:37.285Z","commitSha":"4964f49bf29ec608537f912a81cf1c0addb35a0f","verifiedAt":"2026-11-03T02:37:03.860Z","lastCommitAt":"2026-11-07T02:23:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds up what a Steam account's game items are worth and turns the result into a picture you can share.","summary":"A Windows terminal program that reads a public Steam profile and its item inventories (CS2, Dota 2, TF2, Rust and Steam community items), looks the items up on the Steam market and on several skin marketplaces, and saves the result as a PNG card: totals, value per game, top items, fees and where each item sells best. It needs no account, and an offline demo works with no internet. You paste a profile link and answer menu questions in the terminal; a one-file portable .exe is offered, and the README also documents installing it with Python 3.11+ and pipx.","easeRationale":"Windows is the only system it runs on: the README says \"The CLI refuses to start on other systems (`This build is Windows-only.`)\". There is a single portable `Stashcard-1.0.0-portable-x64.exe`, but the Install section of the same README says it needs Python 3.11+ and describes installing it with `pipx`, and the checks also found Python as a requirement, so plan on installing Python and typing commands in a terminal.","costNote":"No paid services: the README says every price source is free and needs no key.","notVerified":["Whether the one-file portable .exe really runs on a clean Windows machine without installing Python","Whether real scans finish at all, since the README says Steam limits how often it answers and the program may fall back to old cached prices","How close the totals are to what you could actually sell the items for, and whether the marketplace fees in the code are current","Real disk use after the program caches inventories, prices and whole marketplace catalogues under %LOCALAPPDATA%\\stashcard","Whether the optional Steam session cookie is only ever sent to steamcommunity.com, as the README states"]},"zh-Hans":{"summaryShort":"统计一个 Steam 账号游戏道具的总价值，并把结果做成一张可以分享的图片。","summary":"一个 Windows 终端程序，读取公开的 Steam 个人资料及其道具库存（CS2、Dota 2、TF2、Rust 和 Steam 社区物品），在 Steam 市场和多个饰品交易平台上查询这些物品，并把结果保存为一张 PNG 卡片：总价值、各游戏价值、热门物品、手续费，以及每件物品在哪里卖得最好。它不需要账号，离线演示模式无需联网也能使用。你粘贴一个个人资料链接，并在终端里回答菜单问题；项目提供一个单文件便携版 .exe，README 也说明了用 Python 3.11+ 和 pipx 安装的方法。","easeRationale":"它只能在 Windows 上运行：README 写道 \"The CLI refuses to start on other systems (`This build is Windows-only.`)\"。虽然提供了一个单文件便携版 `Stashcard-1.0.0-portable-x64.exe`，但同一份 README 的 Install 部分说需要 Python 3.11+，并说明了用 `pipx` 安装的方法，检查也发现 Python 是必需的，所以要准备好安装 Python 并在终端里输入命令。","costNote":"没有付费服务：README 说所有的价格来源都是免费的，也不需要密钥。","notVerified":["单文件便携版 .exe 是否真的能在一台干净的 Windows 电脑上运行，而不需要安装 Python","真实扫描是否能顺利完成，因为 README 说 Steam 会限制请求频率，程序可能会回退使用旧的缓存价格","统计总价与实际能卖出的价格有多接近，代码里的市场手续费是否是最新的","程序把库存、价格和完整的市场目录缓存到 %LOCALAPPDATA%\\stashcard 后，实际占用的磁盘空间","可选的 Steam 会话 cookie 是否真的只会发送给 steamcommunity.com，正如 README 所说"]}}},{"id":"R_kgDOHy0BzQ","slug":"astral-sh__ruff","name":"Ruff","owner":"astral-sh","fullName":"astral-sh/ruff","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"install_script_piped_from_web","severity":"info","evidence":"README.md 的安装部分建议在 macOS 和 Linux 上运行 `curl -LsSf https://astral.sh/ruff/install.sh | sh`，在 Windows 上运行 `powershell -c \"irm https://astral.sh/ruff/install.ps1 | iex\"`，这会直接运行从 astral.sh 下载的脚本，而不会先展示脚本内容。","source":"ai","title":"安装脚本从网络直接管道执行"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","rust"],"disk":{"minMB":10,"maxMB":29,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":51159,"downloads":39846463,"downloadsWeek":1060790,"downloadsDays":7,"growth":740,"growthDays":7,"history":[{"date":"2026-10-26","stars":50317},{"date":"2026-10-27","stars":50326},{"date":"2026-10-28","stars":50347},{"date":"2026-10-29","stars":50332},{"date":"2026-10-30","stars":50347},{"date":"2026-10-31","stars":50369},{"date":"2026-11-01","stars":50390},{"date":"2026-11-02","stars":50400},{"date":"2026-11-03","stars":50419},{"date":"2026-11-04","stars":50432},{"date":"2026-11-05","stars":50462},{"date":"2026-11-06","stars":50612},{"date":"2026-11-07","stars":50939},{"date":"2026-11-08","stars":51108},{"date":"2026-11-09","stars":51135},{"date":"2026-11-10","stars":51159}],"firstPublishedAt":"2026-10-27T01:10:12.604Z","commitSha":"68207fbea5f356462896db633257604e32eb2dba","verifiedAt":"2026-11-01T01:38:51.795Z","lastCommitAt":"2026-11-10T18:46:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A fast command-line tool that checks Python code for mistakes and reformats it in a consistent style.","summary":"Ruff is a terminal tool for people who write Python. It scans a project's Python files, points out over 900 kinds of problems such as unused imports, can fix many of them automatically, and reformats the code in a style close to Black's. It is written in Rust and ships as a ready binary for Windows, macOS and Linux (about 10 MB), installable with pip, uv, Homebrew or the project's own install script. It also plugs into VS Code and other editors.","easeRationale":"Equal on the three platforms: ready binaries are published for each one, but Ruff has no window or buttons, so you install it with a command (pip, uv, Homebrew or the install script) and then type commands such as `ruff check` in a terminal.","costNote":"MIT license; the company behind it, Astral, sells other products.","notVerified":["Whether the published binaries match the source code in this repository","How many of the fixes it applies automatically are the ones you actually want","Whether the checks behave the same on your Python version and project layout"]},"zh-Hans":{"summaryShort":"一个快速的命令行工具，检查 Python 代码中的问题并按统一风格重新格式化。","summary":"Ruff 是一个给写 Python 的人用的终端工具。它会扫描项目里的 Python 文件，指出 900 多种问题，比如没用到的导入，还能自动修复其中很多问题，并把代码重新格式化成接近 Black 风格的样子。它用 Rust 编写，为 Windows、macOS 和 Linux 提供现成的二进制文件（约 10 MB），可以用 pip、uv、Homebrew 或项目自带的安装脚本安装。它还能接入 VS Code 和其他编辑器。","easeRationale":"三个平台待遇相同：每个平台都发布了现成的二进制文件，但 Ruff 没有窗口也没有按钮，所以你要用一条命令安装它（pip、uv、Homebrew 或安装脚本），然后在终端里输入像 `ruff check` 这样的命令。","costNote":"MIT 许可证；项目背后的公司 Astral 还销售其他产品。","notVerified":["发布的二进制文件是否与这个仓库里的源代码一致","它自动应用的修复中有多少真的是你想要的","这些检查在你的 Python 版本和项目结构上是否表现一致"]}}},{"id":"R_kgDOKbIFZQ","slug":"astral-sh__uv","name":"uv","owner":"astral-sh","fullName":"astral-sh/uv","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"install_pipes_remote_script_to_shell","severity":"info","evidence":"README.md 建议在 Mac 和 Linux 上用 `curl -LsSf https://astral.sh/uv/install.sh | sh` 安装，在 Windows 上用 `powershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"` 安装，这会在你没有先阅读的情况下运行从 astral.sh 下载的脚本。相同的版本也以普通压缩包的形式提供在 GitHub 上。","source":"ai","title":"安装时会直接运行远程脚本"},{"code":"self_update_and_runtime_downloads","severity":"info","evidence":"README.md 记录了用于替换程序自身的 `uv self update`，以及 `uv python install 3.12 3.13 3.14` 和 `uv venv --python 3.12.0`，它们会在你使用时从互联网下载 Python 版本和软件包。","source":"ai","title":"会自我更新，使用时还会下载组件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","docker","rust"],"disk":{"minMB":15,"maxMB":46,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":91854,"downloads":1211740883,"downloadsWeek":39153795,"downloadsDays":7,"growth":337,"growthDays":7,"history":[{"date":"2026-10-25","stars":91174},{"date":"2026-10-26","stars":91228},{"date":"2026-10-27","stars":91263},{"date":"2026-10-28","stars":91321},{"date":"2026-10-29","stars":91343},{"date":"2026-10-30","stars":91376},{"date":"2026-10-31","stars":91403},{"date":"2026-11-01","stars":91440},{"date":"2026-11-02","stars":91481},{"date":"2026-11-03","stars":91517},{"date":"2026-11-04","stars":91542},{"date":"2026-11-05","stars":91604},{"date":"2026-11-06","stars":91678},{"date":"2026-11-07","stars":91737},{"date":"2026-11-08","stars":91774},{"date":"2026-11-09","stars":91815},{"date":"2026-11-10","stars":91854}],"firstPublishedAt":"2026-10-25T23:18:57.481Z","commitSha":"e4c55567ada966de4911513b5c526070e6cddd3f","verifiedAt":"2026-10-28T01:36:37.212Z","lastCommitAt":"2026-11-10T21:50:24Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Installs Python and Python packages from the terminal, and keeps a project's dependencies and versions organised.","summary":"uv is a command-line tool that does the jobs usually split between pip, pipx, poetry, pyenv, virtualenv and twine: it installs packages, creates isolated environments, locks exact dependency versions, installs and switches Python versions, runs single-file scripts and publishes packages. It is for people who write Python. There are ready binaries for Windows, Mac and Linux (about 15 MB) and a one-line install command; you do not need Python or Rust installed first, and uv can download Python for you.","easeRationale":"Every release ships a ready binary for Windows, Mac and Linux (for example uv-x86_64-pc-windows-msvc.zip, 15 MB), and the README gives a single install command per system; nothing else has to be installed first. After that you use it by typing commands in a terminal, so it is only useful to someone comfortable with a terminal.","costNote":"Free and open source (MIT or Apache-2.0, your choice). Astral, the company behind it, sells other products.","notVerified":["The 10-100x speed comparison with pip shown in the README, which was measured by the authors","How much disk the package cache and the downloaded Python versions grow to over time (the installed program itself is about 15-46 MB)","Whether a given project's dependencies resolve to exactly the same versions as with pip or poetry","Where the Python builds that uv downloads come from and how they are verified"]},"zh-Hans":{"summaryShort":"在终端里安装 Python 和 Python 包，并整理好一个项目的依赖和版本。","summary":"uv 是一个命令行工具，把通常由 pip、pipx、poetry、pyenv、virtualenv 和 twine 分担的工作合在一起：安装包、创建隔离环境、锁定精确的依赖版本、安装并切换 Python 版本、运行单文件脚本以及发布包。它适合写 Python 的人使用。Windows、Mac 和 Linux 都有现成的二进制文件（约 15 MB）和一条安装命令；你不需要事先安装 Python 或 Rust，uv 还能帮你下载 Python。","easeRationale":"每个发布版本都为 Windows、Mac 和 Linux 提供现成的二进制文件（例如 uv-x86_64-pc-windows-msvc.zip，15 MB），README 针对每个系统只给出一条安装命令；在此之前不需要先安装别的东西。之后你需要在终端里输入命令来使用它，所以只对习惯使用终端的人有用。","costNote":"免费且开源（可选 MIT 或 Apache-2.0）。背后的公司 Astral 还出售其他产品。","notVerified":["README 中展示的、与 pip 相比快 10 到 100 倍的速度对比，是由作者自己测量的。","包缓存和下载的 Python 版本随着时间会占用多少磁盘空间（安装的程序本身约为 15-46 MB）。","某个项目的依赖是否能解析出与 pip 或 poetry 完全相同的版本。","uv 下载的 Python 构建版本来自哪里，以及它们是如何被验证的。"]}}},{"id":"R_kgDOOiIsNw","slug":"asyncfuncai__deepwiki-open","name":"DeepWiki Open","owner":"AsyncFuncAI","fullName":"AsyncFuncAI/deepwiki-open","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"default_passwords_in_the_litellm_setup","severity":"attention","evidence":"docker-compose-litellm.yml 把 4000 和 5432 端口暴露给主机，如果你不自己设置，就会用固定的密钥作为后备：\"LITELLM_MASTER_KEY=${LITELLM_MASTER_KEY:-sk-1234}\" 和 \"POSTGRES_PASSWORD: ${LITELLM_DB_PASSWORD:-litellm_password}\"。","source":"ai","title":"litellm 配置里有默认密码"},{"code":"your_code_is_sent_to_the_model_provider","severity":"info","evidence":"容器会提示\"OPENAI_API_KEY and/or GOOGLE_API_KEY environment variables are not set. These are required for DeepWiki to function properly\"——为了编写百科，这个工具会把仓库的代码和文本发送给这些公司，这对私有仓库尤其重要，除非你改用本地的 Ollama 配置。","source":"ai","title":"你的代码会被发送给模型提供商"},{"code":"newer_version_is_offered_on_an_outside_site","severity":"info","evidence":"README 里的\"Deepwiki-Open 2.0 (Grok Wiki is now live)\"一节只写了\"Download at https://grok-wiki.com\"，这个链接同时也是仓库的主页；那个网站实际提供的内容并不在这个仓库里。","source":"ai","title":"外部网站提供了更新的版本"},{"code":"readme_dropped_the_local_install_steps","severity":"info","evidence":"README 里没有本地运行的快速开始说明——里面展示的命令只有 kubectl 和 helm——尽管仓库里其实有 docker-compose.yml、Dockerfile、Dockerfile-ollama-local 和 run.sh。","source":"ai","title":"readme 删掉了本地安装步骤"},{"code":"needs_a_lot_of_memory","severity":"info","evidence":"docker-compose.yml 为容器设置了\"mem_limit: 6g\"和\"mem_reservation: 2g\"，Next.js 的构建过程则使用 NODE_OPTIONS=\"--max-old-space-size=4096\"运行。","source":"ai","title":"需要很多内存"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["OpenAI","Google (Gemini)","OpenRouter or a local Ollama model as an alternative to the two above"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"deepwiki","name":"DeepWiki","confirmed":true},{"slug":"mintlify","name":"Mintlify","confirmed":true},{"slug":null,"name":"DeepWiki (Cognition)","confirmed":false}],"declaredAlternatives":[],"stars":18397,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":28,"growthDays":6,"history":[{"date":"2026-11-06","stars":18372},{"date":"2026-11-07","stars":18374},{"date":"2026-11-08","stars":18375},{"date":"2026-11-09","stars":18388},{"date":"2026-11-10","stars":18397}],"firstPublishedAt":"2026-11-05T22:53:18.498Z","commitSha":"b81b40e4500c09e82c8148f1880b5ed0261c1672","verifiedAt":"2026-11-05T22:46:39.051Z","lastCommitAt":"2026-11-10T08:13:03Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Point it at a GitHub, GitLab or Bitbucket repository and it writes an illustrated wiki explaining the code.","summary":"DeepWiki-Open reads a repository you name and uses an AI model to write documentation about it: a structured wiki, Mermaid diagrams of how the parts fit together and a code map for guided tours of the code. You run it on your own machine with Docker Compose (a web page on port 3000 plus an API on 8001) or on Kubernetes with the included Helm chart, and you supply your own model keys — OpenAI and Google by default, with OpenRouter or a local Ollama model as alternatives. The author calls it \"my own implementation attempt of DeepWiki\".","easeRationale":"Same route and same grade on the three systems: install Docker, write a .env file with your OpenAI and Google keys, build and start the container with docker-compose.yml and open port 3000. The compose file reserves up to 6 GB of memory, and the only step-by-step in the README is the Kubernetes one with kubectl and Helm.","costNote":"The code is MIT and free. By default you pay OpenAI and Google per repository you index; the repository also ships a Dockerfile and instructions for using a local Ollama model instead.","notVerified":["How much the OpenAI and Google calls cost for one repository, which depends on its size","How good and how correct the generated wiki and diagrams are","Whether the local Ollama setup gives usable results without the paid keys","Real disk use in ~/.adalflow after indexing a few repositories","Whether the program at grok-wiki.com is built from this same code"]},"zh-Hans":{"summaryShort":"把它指向一个 GitHub、GitLab 或 Bitbucket 仓库，它就会写出一份带插图的百科，解释这些代码。","summary":"DeepWiki-Open 会读取你指定的仓库，并用一个 AI 模型为它编写文档：一份结构化的百科、展示各部分如何配合的 Mermaid 图，以及一张用于导览代码的代码地图。你可以用 Docker Compose 在自己的电脑上运行它（网页在 3000 端口，API 在 8001 端口），也可以用自带的 Helm chart 在 Kubernetes 上运行，并提供你自己的模型密钥——默认是 OpenAI 和 Google，也可以用 OpenRouter 或本地的 Ollama 模型作为替代。作者把它称为\"my own implementation attempt of DeepWiki\"。","easeRationale":"三个系统走的是同一条路，评分也一样：安装 Docker，写一个包含你的 OpenAI 和 Google 密钥的 .env 文件，用 docker-compose.yml 构建并启动容器，然后打开 3000 端口。compose 文件预留了最多 6 GB 内存，README 里唯一的分步说明是 Kubernetes 的那一个，用到了 kubectl 和 Helm。","costNote":"代码是 MIT 许可，免费使用。默认情况下，你索引的每个仓库都要向 OpenAI 和 Google 付费；仓库里也附带了 Dockerfile 和使用本地 Ollama 模型作为替代的说明。","notVerified":["调用 OpenAI 和 Google 处理一个仓库要花多少钱，这取决于仓库的大小","生成的百科和图表有多好、有多准确","不用付费密钥、只用本地 Ollama 配置是否能得到可用的结果","索引几个仓库之后，~/.adalflow 里实际占用了多少磁盘空间","grok-wiki.com 上的程序是否就是用这同一份代码构建的"]}}},{"id":"R_kgDON5cH1w","slug":"ath-maas__comfyui-copilot","name":"ComfyUI-Copilot","owner":"ATH-MaaS","fullName":"ATH-MaaS/ComfyUI-Copilot","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"readme_suggests_running_comfyui_as_root","severity":"attention","evidence":"README.md 中的安装说明：“Manager 需要权限。为避免执行时出错，建议用 \\\"sudo python main.py\\\" 运行 ComfyUI。”以这种方式启动后，ComfyUI 及其中的每个插件都拥有对电脑的完全控制权。","source":"ai","title":"readme 建议以 root 身份运行 comfyui"},{"code":"advertised_features_being_switched_off","severity":"attention","evidence":"README.md 顶部的一则通知说，节点信息查询、任务推荐和工作流生成“即将不再可用”，并且“API 服务已暂停”——但下面的功能列表仍然把这些功能都描述在内。这里最后一次提交是在 2026-09-11，最后一个发布版本是在 2025-08-18。","source":"ai","title":"宣传的功能正在被下线"},{"code":"repository_owner_differs_from_readme_and_license","severity":"info","evidence":"这个仓库位于 ATH-MaaS 名下，但 pyproject.toml 指向别处（Repository = https://github.com/AIDC-AI/ComfyUI-Copilot），README 链接到同一个 AIDC-AI 地址，LICENSE 写着“Copyright (C) 2025 AIDC-AI”。GitHub 没有把这份拷贝标记为 fork，所以很可能是账号改了名，但文字内容没有同步更新。","source":"ai","title":"仓库所有者与 readme 和许可证不一致"},{"code":"downloads_model_files_while_you_use_it","severity":"info","evidence":"README.md：“如果发现缺少模型，会自动提示你下载该模型”，还有一个 Model Download 按钮，requirements.txt 中还引入了“modelscope>=1.28.0”。数 GB 大小的模型文件会从外部获取到你的 ComfyUI 文件夹中。","source":"ai","title":"使用过程中会下载模型文件"},{"code":"ships_prebuilt_javascript","severity":"info","evidence":"仓库顶层既有源代码界面（ui/），也有一个已经构建好的 dist/ 文件夹，而 ComfyUI 加载的正是后者。阅读源代码并不能告诉你那个打包文件里到底有什么。","source":"ai","title":"附带预编译的 javascript"},{"code":"your_email_is_sent_to_the_project_service","severity":"info","evidence":"README.md，API Key 生成部分：“在弹出窗口中输入你的邮箱地址，API Key 随后会自动发送到你的邮箱”。你的工作流描述内容同样会离开你的电脑，发往你配置的那个模型服务。","source":"ai","title":"你的邮箱会被发送给项目方的服务"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI or another OpenAI-compatible model service (you can point it at LM Studio on your own computer instead)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["image","ai_agents"],"alternativeTo":[],"declaredAlternatives":[],"stars":5581,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":5579},{"date":"2026-11-06","stars":5578},{"date":"2026-11-07","stars":5579},{"date":"2026-11-08","stars":5580},{"date":"2026-11-09","stars":5581},{"date":"2026-11-10","stars":5581}],"firstPublishedAt":"2026-11-04T22:38:15.160Z","commitSha":"a5aa6acd992fe3b42b9482d30899cdf314cdbe1c","verifiedAt":"2026-11-04T22:14:57.516Z","lastCommitAt":"2026-09-11T02:19:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds an AI chat panel inside ComfyUI that builds, fixes and tunes image-generation workflows for you.","summary":"ComfyUI-Copilot is an add-on for ComfyUI, the node editor used to generate images. It puts a chat panel in the ComfyUI screen: you describe what you want and it proposes a ready workflow, explains nodes, suggests models, looks for errors in the workflow on the canvas and can run the same workflow many times with different settings to compare results. It is for people who already use ComfyUI. You copy it into ComfyUI's custom_nodes folder, install its Python packages, then type in your own AI model key and address, or point it at LM Studio running on your computer.","easeRationale":"The same route on Windows, Mac and Linux, so the same grade: you must already have ComfyUI working, then clone this folder into ComfyUI/custom_nodes and run \"pip install -r requirements.txt\" (on Windows with ComfyUI's own python_embeded\\python.exe), restart ComfyUI and type your model key and address into the settings panel. The one-click route through ComfyUI Manager exists, but the README itself says it \"is prone to bugs, so it's recommended to use the git installation method above\".","costNote":"The add-on itself is free (MIT). The project's own free AI service is gone: the README says \"The API service has been suspended\" and that you must \"enter your own API Key and Base URL\", so you pay the model provider per use, unless you point it at a model running on your own computer with LM Studio.","notVerified":["Which features still work now that the project's own API service is suspended","What the chat sends to the model service besides your question: node list, workflow contents or file names","How much a month of workflow generating and debugging costs on a paid model key","Whether the LM Studio route described in HOW_TO_USE_LMSTUDIO.md works with a small local model","What is inside the prebuilt dist/ folder that ComfyUI loads into the browser","Real disk space after the suggested model downloads"]},"zh-Hans":{"summaryShort":"在 ComfyUI 内部加入一个 AI 聊天面板，帮你搭建、修复和调优图像生成工作流。","summary":"ComfyUI-Copilot 是 ComfyUI（用于生成图像的节点编辑器）的一个插件。它在 ComfyUI 界面里加入一个聊天面板：你描述想要的效果，它就提出一个现成的工作流，解释节点，推荐模型，在画布上的工作流中查找错误，还能用不同设置多次运行同一个工作流来比较结果。它面向已经在使用 ComfyUI 的人。你把它复制到 ComfyUI 的 custom_nodes 文件夹，安装其 Python 依赖包，然后输入自己的 AI 模型密钥和地址，或者指向本机运行的 LM Studio。","easeRationale":"Windows、Mac 和 Linux 的路径相同,因此评级相同：你必须已经让 ComfyUI 正常运行,然后把这个文件夹克隆到 ComfyUI/custom_nodes 中,运行 “pip install -r requirements.txt”（在 Windows 上用 ComfyUI 自带的 python_embeded\\python.exe）,重启 ComfyUI,再在设置面板里填入模型密钥和地址。通过 ComfyUI Manager 一键安装的方式也存在,但 README 自己说它“容易出 bug,所以建议使用上面的 git 安装方式”。","costNote":"插件本身是免费的（MIT 许可）。项目自带的免费 AI 服务已经关闭：README 说“API 服务已暂停”，必须“输入你自己的 API Key 和 Base URL”，因此除非你把它指向用 LM Studio 在本机运行的模型，否则需要按使用量向模型供应商付费。","notVerified":["在项目自己的 API 服务暂停后，哪些功能还能正常使用","聊天除了你的问题之外，还向模型服务发送了什么：节点列表、工作流内容还是文件名","用付费模型密钥一个月生成和调试工作流要花多少钱","HOW_TO_USE_LMSTUDIO.md 中描述的 LM Studio 方式能否配合小型本地模型使用","ComfyUI 加载到浏览器里的预编译 dist/ 文件夹里到底有什么","建议下载的模型文件之后实际占用的磁盘空间"]}}},{"id":"R_kgDOQQ9_ZA","slug":"ath-maas__pixelle-video","name":"Pixelle-Video","owner":"ATH-MaaS","fullName":"ATH-MaaS/Pixelle-Video","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"windows_bundle_older_than_the_code","severity":"info","evidence":"最新的 Windows zip 包是 2026-01-27 的 Pixelle-Video-v0.1.15-win64.zip，而 pyproject.toml 已经写着版本 0.2.0，这里最后一次提交是 2026-06-14。所以这个现成的打包版本并不包含 README 中列出的后续变化，比如 2026-06-01 新增的直接 API 图像和视频模型设置。","source":"ai","title":"windows 打包版本比代码旧"},{"code":"repository_owner_differs_from_readme","severity":"info","evidence":"这份拷贝位于 ATH-MaaS 名下，但 README.md 里的每一个链接都指向别处：徽章、issue 页面、aidc-ai.github.io 上的文档，以及克隆命令 “git clone https://github.com/AIDC-AI/Pixelle-Video.git”。GitHub 没有把它标记为 fork，所以很可能账号被改名了，但文字没有随之更新。","source":"ai","title":"仓库所有者与 readme 不一致"},{"code":"your_keys_sit_in_a_plain_config_file","severity":"info","evidence":"你在浏览器里输入的密钥会保存到 config.yaml，docker-compose.yml 把它以可读写方式挂载进两个容器（“./config.yaml:/app/config.yaml”）。README 还提供了“打印模型请求参数”的调试功能，会把提示词、模型名称和输入文件路径打印到终端。","source":"ai","title":"你的密钥存放在一个明文配置文件中"},{"code":"retries_around_provider_content_checks","severity":"info","evidence":"README.md，API 视频生成部分：“支持网络下载重试与内容审核失败后的提示词中性化重试”——当供应商的内容审核拒绝某个场景时，程序会把提示词改写得更中性，然后再次尝试。工具本身不会在你发布之前检查成品视频。","source":"ai","title":"绕过供应商内容审核的重试机制"},{"code":"whole_video_is_made_without_a_human_in_between","severity":"info","evidence":"功能列表宣称“输入主题，自动生成完整视频”，脚本由语言模型写成，没有事实核查环节。视频仍然要受到每个供应商条款中关于 AI 内容以及 bgm/ 中音乐的规定约束。","source":"ai","title":"整个视频制作过程没有人工把关"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","docker","ffmpeg"],"disk":{"minMB":380,"maxMB":1140,"basis":"installer"},"apiKeys":["An AI text model: Qwen/Tongyi, OpenAI, DeepSeek and similar, or Ollama on your own computer","Image and video models: DashScope/Wan, OpenAI GPT Image, Volcengine Seedream/Seedance, Kling (not needed if you run ComfyUI yourself)","RunningHub, a cloud ComfyUI used instead of a local one"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","social_marketing","audio"],"alternativeTo":[{"slug":"pictory","name":"Pictory","confirmed":true},{"slug":"invideo-ai","name":"AI","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":false}],"declaredAlternatives":[],"stars":29954,"downloads":87147,"downloadsWeek":891,"downloadsDays":5,"growth":193,"growthDays":6,"history":[{"date":"2026-11-05","stars":29791},{"date":"2026-11-06","stars":29826},{"date":"2026-11-07","stars":29853},{"date":"2026-11-08","stars":29878},{"date":"2026-11-09","stars":29926},{"date":"2026-11-10","stars":29954}],"firstPublishedAt":"2026-11-04T22:33:17.112Z","commitSha":"848b054e4fae40dabc62ec58e960b573e83793ac","verifiedAt":"2026-11-04T22:16:14.060Z","lastCommitAt":"2026-06-14T12:40:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Type a topic and it writes the script, makes the pictures and the narration, and assembles a short video.","summary":"Pixelle-Video makes vertical or horizontal short videos on its own. You type a topic in a page that opens in your browser; it writes the narration with an AI text model, creates one image or clip per sentence, reads the text out loud with a text-to-speech voice, adds background music and stitches everything together. You can also paste your own script, upload your own photos or a voice sample to clone. Windows users unzip a ready bundle and double-click start.bat; on Mac and Linux you install uv and ffmpeg and start it from the terminal.","easeRationale":"Windows is the easiest and sets the grade: the release page has a 380 MB zip, and the README says it needs no Python, uv or ffmpeg — you unzip, double-click start.bat and the browser opens at http://localhost:8501, where the one remaining step is pasting your model keys into the settings panel. Mac and Linux have no bundle: you install uv and ffmpeg, clone the project and run \"uv run streamlit run web/app.py\" before the same key setup. Paying nothing for the AI part means also installing ComfyUI and Ollama yourself, which is a much longer job on any of the three.","costNote":"The program is free (Apache-2.0), but it only works once you plug in model services: an AI text model plus image, video or voice generation. The README's own FAQ says a completely free route exists — \"LLM 使用 Ollama（本地运行）+ ComfyUI 本地部署 = 0 元\" (Ollama and ComfyUI on your own machine, which needs a graphics card) — and otherwise you pay per video to providers such as Qwen, OpenAI, Kling, Seedance or RunningHub.","notVerified":["Whether the Windows zip really runs with no extra installs, and what is inside those 380 MB","What a finished video costs with paid image or video models, since each scene is a separate call","How long one video takes, and whether a computer without a graphics card can use the free Ollama plus ComfyUI route","Whether the older Windows bundle still talks to the current provider APIs","Real disk space in the output/ folder after a few videos","Whether the voice cloning from an uploaded sample works outside Chinese"]},"zh-Hans":{"summaryShort":"输入一个主题，它就写脚本、生成图片和配音，并拼接成一段短视频。","summary":"Pixelle-Video 能自己制作竖版或横版短视频。你在浏览器打开的页面里输入一个主题；它用 AI 文本模型写好配音文案，为每句话生成一张图片或一段片段，用文字转语音朗读出来，加上背景音乐，再把一切拼接起来。你也可以粘贴自己的脚本，上传自己的照片或一段用于克隆的语音样本。Windows 用户解压一个现成的包，双击 start.bat 即可；Mac 和 Linux 用户需要安装 uv 和 ffmpeg，再从终端启动。","easeRationale":"Windows 最简单，决定了评级：发布页提供一个 380MB 的 zip 包，README 说不需要 Python、uv 或 ffmpeg——解压、双击 start.bat，浏览器就会打开 http://localhost:8501，剩下唯一的步骤是把模型密钥粘贴到设置面板里。Mac 和 Linux 没有打包版本：需要安装 uv 和 ffmpeg，克隆项目并运行 “uv run streamlit run web/app.py”，然后再做同样的密钥配置。如果想在 AI 部分完全不花钱，就得自己再装 ComfyUI 和 Ollama，这在三个系统上都是一项更长的工作。","costNote":"该程序免费（Apache-2.0 许可），但只有接入模型服务后才能使用：一个 AI 文本模型，加上图像、视频或语音生成。README 自己的 FAQ 说存在一条完全免费的路线——“LLM 使用 Ollama（本地运行）+ ComfyUI 本地部署 = 0 元”（需要显卡）——否则就要按视频向 Qwen、OpenAI、Kling、Seedance 或 RunningHub 等供应商付费。","notVerified":["Windows 的 zip 包是否真的不需要额外安装，那 380MB 里到底有什么","用付费图像或视频模型做完一个视频要花多少钱，因为每个场景是单独一次调用","制作一个视频要多久，没有显卡的电脑能否使用免费的 Ollama 加 ComfyUI 路线","旧的 Windows 打包版本是否仍能对接当前的供应商 API","output/ 文件夹里做几个视频之后实际占用多少磁盘空间","从上传样本克隆声音的功能在非中文场景下是否有效"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzMjI4NTA1","slug":"atom__atom","name":"Atom","owner":"atom","fullName":"atom/atom","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"C"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布是 2022-03-08（1708 天前）","source":"check","title":"发布早已停止"},{"code":"archived","severity":"info","evidence":"该仓库已归档（只读，不再维护）","source":"check","title":"仓库已归档"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：async: GHSA-fwr7-v2mv-hh25; postcss: GHSA-566m-qj78-rww5, GHSA-6g55-p6wh-862q, GHSA-7fh5-64p2-3v2j; postcss-selector-parser: GHSA-rj75-hqrm-r3gf; semver: GHSA-c2qf-rxjj-qqgw","source":"check","title":"存在已知漏洞"},{"code":"outdated_browser_engine","severity":"attention","evidence":"package.json：\"electronVersion\": \"11.5.0\"。Atom 运行在这个旧版 Electron（基于 Chromium）上，而且因为项目已归档，不会再升级。","source":"ai","title":"浏览器内核过旧"},{"code":"discontinued_by_owner","severity":"info","evidence":"README.md：\"Atom and all repositories under Atom will be archived on December 15, 2022.\" README 里仍写着 \"Atom will automatically update\"，但这一点已经不再成立。","source":"ai","title":"作者已停止该项目"},{"code":"usage_metrics_package","severity":"info","evidence":"package.json 打包了 \"metrics\": \"git+https://github.com/atom/metrics.git#d35d034\" 和 \"exception-reporting\"，这是上报使用情况和错误的内置包。","source":"ai","title":"内置使用数据上报包"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":["node","docker"],"disk":{"minMB":198,"maxMB":593,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"sublime-text","name":"Sublime Text","confirmed":true}],"declaredAlternatives":[],"stars":60692,"downloads":94766661,"downloadsWeek":17206,"downloadsDays":7,"growth":0,"growthDays":7,"history":[{"date":"2026-10-26","stars":60710},{"date":"2026-10-27","stars":60709},{"date":"2026-10-28","stars":60711},{"date":"2026-10-29","stars":60708},{"date":"2026-10-30","stars":60708},{"date":"2026-10-31","stars":60708},{"date":"2026-11-01","stars":60704},{"date":"2026-11-02","stars":60704},{"date":"2026-11-03","stars":60702},{"date":"2026-11-04","stars":60696},{"date":"2026-11-05","stars":60694},{"date":"2026-11-06","stars":60692},{"date":"2026-11-07","stars":60690},{"date":"2026-11-08","stars":60690},{"date":"2026-11-09","stars":60691},{"date":"2026-11-10","stars":60692}],"firstPublishedAt":"2026-10-27T01:10:12.605Z","commitSha":"1c3bd35ce238dc0491def9e1780d04748d8e18af","verifiedAt":"2026-11-10T23:29:33.743Z","lastCommitAt":"2022-11-22T16:11:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A customizable text and code editor from GitHub that was discontinued in 2022 and gets no more updates.","summary":"Atom is a desktop text and code editor made by GitHub, known for themes and add-on packages that change how it looks and works. It was meant for programmers and anyone who edits text files. GitHub stopped the project and archived it in December 2022, so there are no more fixes or security updates. The last Windows installers (AtomSetup-x64.exe for 64-bit) are still in the releases, along with a Mac zip and Linux packages, and the README lists Git as a prerequisite.","easeRationale":"Windows is the easiest: a ready installer (AtomSetup-x64.exe, 198 MB) in the last release, with Git listed as a prerequisite to install first; on Mac you unzip atom-mac.zip and run the app. Linux is harder: the README asks you to set up your package manager or install a list of system libraries with apt and then extract an archive by hand.","costNote":"Free (MIT).","notVerified":["Whether installing add-on packages from inside Atom still works now that the project's online services were retired.","Whether the old installers still run correctly on current Windows 11 versions.","Whether the built-in usage and error reporting still tries to send data anywhere.","Exposure to security problems fixed in newer Electron versions when opening files or packages from others."]},"zh-Hans":{"summaryShort":"一款来自 GitHub 的可定制文本和代码编辑器,已于 2022 年停止维护,不再有更新。","summary":"Atom 是 GitHub 制作的桌面文本和代码编辑器,以主题和可以改变外观与功能的附加包而闻名。它面向程序员和任何需要编辑文本文件的人。GitHub 于 2022 年 12 月终止并封存了这个项目,因此不会再有修复或漏洞修补更新。最后一版的 Windows 安装包(64 位的 AtomSetup-x64.exe)仍保留在发布记录中,还有 Mac 版压缩包和 Linux 版软件包,README 中把 Git 列为前置条件。","easeRationale":"Windows 上最简单:最后一次发布中有现成的安装包(AtomSetup-x64.exe,198 MB),需要先安装 Git 作为前置条件;Mac 上需要解压 atom-mac.zip 然后运行应用。Linux 较难:README 要求你先配置好包管理器,或用 apt 安装一系列系统库,然后手动解压一个压缩包。","costNote":"免费(MIT 许可证)。","notVerified":["在项目的在线服务已经停止之后,从 Atom 内部安装附加包是否还能正常使用。","旧版安装包在当前的 Windows 11 系统上是否仍能正常运行。","内置的使用与错误报告功能是否仍会尝试向外发送数据。","打开他人提供的文件或附加包时,是否会遇到新版 Electron 中已经修复的漏洞风险。"]}}},{"id":"R_kgDOH3JoLw","slug":"automatic1111__stable-diffusion-webui","name":"Stable Diffusion web UI","owner":"AUTOMATIC1111","fullName":"AUTOMATIC1111/stable-diffusion-webui","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"D","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖项存在已知漏洞（OSV）：gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8gq9-2x98-w8hf, GHSA-8qvm-5x2c-j2w7; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-37q5-v5qm-c9v8; setuptools: GHSA-5rjg-fvgr-3xxf, GHSA-cx63-2mw6-8hw5, GHSA-h35f-9h28-mq5c; gitpython: GHSA-239g-whfq-7xj9, GHSA-23mf-xhv8-69c2, GHSA-284h-m62q-gf8w; pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; diskcache: GHSA-w8v5-vhqr-4h9v, PYSEC-2026-2447; fastapi: PYSEC-2024-38; gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8gq9-2x98-w8hf, GHSA-8qvm-5x2c-j2w7; pytorch_lightning: GHSA-4cv3-v7pv-rfhf, GHSA-75m9-98v2-hjpm, GHSA-98fp-7v67-4v3q; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-37q5-v5qm-c9v8","source":"check","title":"存在已知漏洞"},{"code":"inactive_development","severity":"attention","evidence":"最后一次提交在 2024-07-27（距本次检查 824 天），最后一个版本 v1.10.1 发布于 2025-02-09，目前有 2510 个未解决的 issue；固定版本的依赖项没有在更新（见 known_vulnerabilities）。","source":"ai","title":"开发已停滞"},{"code":"old_python_version_required","severity":"info","evidence":"README.md，\"Automatic Installation on Windows\"：\"Install Python 3.10.6 (Newer version of Python does not support torch), checking \\\"Add Python to PATH\\\"\"。","source":"ai","title":"需要旧版 Python"},{"code":"third_party_extensions_run_code","severity":"info","evidence":"README.md：\"Custom scripts with many extensions from community\" 和 \"Running arbitrary python code from UI (must run with `--allow-code` to enable)\"。扩展是其他作者编写的代码，会在你的电脑上运行。","source":"ai","title":"第三方扩展会运行代码"},{"code":"anime_tagging_for_adult_boorus","severity":"info","evidence":"README.md：\"DeepDanbooru integration, creates danbooru style tags for anime prompts\"。人们配合它使用的模型和标签集来自本仓库之外，可能包含成人内容；仓库本身不附带任何模型。","source":"ai","title":"动漫标签功能面向成人图库"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_any","other"],"buildTools":["python","node","gpu_any"],"disk":{"minMB":1050,"maxMB":4950,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image","local_ai"],"alternativeTo":[{"slug":"midjourney","name":"Midjourney","confirmed":true},{"slug":"adobe-firefly","name":"Firefly","confirmed":true},{"slug":"photoshop","name":"Photoshop","confirmed":true}],"declaredAlternatives":[],"stars":165285,"downloads":5939904,"downloadsWeek":0,"downloadsDays":7,"growth":49,"growthDays":7,"history":[{"date":"2026-10-26","stars":165154},{"date":"2026-10-27","stars":165161},{"date":"2026-10-28","stars":165170},{"date":"2026-10-29","stars":165179},{"date":"2026-10-30","stars":165185},{"date":"2026-10-31","stars":165207},{"date":"2026-11-01","stars":165214},{"date":"2026-11-02","stars":165231},{"date":"2026-11-03","stars":165236},{"date":"2026-11-04","stars":165239},{"date":"2026-11-05","stars":165256},{"date":"2026-11-06","stars":165243},{"date":"2026-11-07","stars":165253},{"date":"2026-11-08","stars":165257},{"date":"2026-11-09","stars":165270},{"date":"2026-11-10","stars":165285}],"firstPublishedAt":"2026-10-27T01:10:12.606Z","commitSha":"82a973c04367123ae98bd9abdf80d9eda9b910e2","verifiedAt":"2026-10-30T01:43:00.785Z","lastCommitAt":"2024-07-27T12:49:39Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Create and edit images from text descriptions on your own computer, with Stable Diffusion models in a browser page.","summary":"A local web page for Stable Diffusion: you type a description and it makes images, repaints parts of a picture (inpainting and outpainting), enlarges photos and repairs faces, with many community extensions. It is for hobbyists and artists who want AI images without a subscription. Everything runs on your own PC and a capable graphics card helps a lot (NVIDIA recommended; AMD, Intel and Apple chips have separate guides). On Windows you install Python 3.10.6 and git, then run webui-user.bat. Models take several GB of disk.","easeRationale":"Easiest on Windows: install Python 3.10.6 and git, then double-click webui-user.bat (or unzip the older NVIDIA-only sd.webui.zip and run update.bat and run.bat). Linux is similar but you first install system packages with apt, dnf, zypper or pacman. On a Mac the README sends you to a separate Apple Silicon guide with more terminal steps, and there is no ready installer anywhere.","notVerified":["Whether it still installs cleanly today, since development stopped in 2024 and the dependencies are pinned to old versions.","Real disk use after downloading torch and image models (the checks estimate 1 to 5 GB before extra models).","Speed and memory use on your graphics card; the README cites 4 GB cards and reports of 2 GB working.","What code the installer and the community extensions download at run time.","Which models you end up using, since the repository ships none and each has its own licence."]},"zh-Hans":{"summaryShort":"在自己的电脑上,通过浏览器页面用 Stable Diffusion 模型根据文字描述生成和编辑图片。","summary":"一个本地网页版的 Stable Diffusion 工具:输入一段描述就能生成图片,还能重绘图片局部(局部重绘和扩展绘制)、放大照片、修复人脸,并支持大量社区扩展。它面向希望在不订阅的情况下生成 AI 图片的爱好者和艺术创作者。一切都在你自己的电脑上运行,一块性能不错的显卡会大有帮助(推荐 NVIDIA;AMD、Intel 和苹果芯片都有各自的使用指南)。在 Windows 上你需要安装 Python 3.10.6 和 git,然后运行 webui-user.bat。模型文件会占用好几 GB 的磁盘空间。","easeRationale":"在 Windows 上最简单:安装 Python 3.10.6 和 git,然后双击 webui-user.bat(或者解压仅支持 NVIDIA 的旧版 sd.webui.zip,运行 update.bat 和 run.bat)。Linux 上流程类似,但要先用 apt、dnf、zypper 或 pacman 安装系统软件包。Mac 上 README 会把你引导到另一份 Apple Silicon 专用指南,步骤更多、更依赖终端命令,目前没有任何现成的安装包。","notVerified":["如今是否还能顺利安装,因为项目在 2024 年就停止了开发,依赖项固定在较旧的版本上。","下载 torch 和图像模型之后的实际磁盘占用(检测估算在添加额外模型之前约为 1 到 5 GB)。","在你的显卡上运行的速度和显存占用情况;README 提到 4 GB 显卡可用,也有 2 GB 显卡能运行的报告。","安装程序和社区扩展在运行时会下载哪些代码。","你最终会使用哪些模型,因为仓库本身不附带任何模型,且每个模型都有自己的许可条款。"]}}},{"id":"R_kgDOGJHxiw","slug":"automatisch__automatisch","name":"Automatisch","owner":"automatisch","fullName":"automatisch/automatisch","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"default_login_published_in_readme","severity":"attention","evidence":"README 安装部分写道：“你可以用 `user@automatisch.io` 这个邮箱地址和 `sample` 密码登录”。任何能访问到你服务器的人，在你修改之前都可以用这组信息登录。docker-compose.yml 里还附带了一个固定的数据库密码（POSTGRES_PASSWORD=automatisch_password）。","source":"ai","title":"默认登录信息公开写在了 readme 里"},{"code":"part_of_the_code_needs_a_paid_licence","severity":"attention","evidence":"LICENSE 文件说 AGPL-3.0 适用于“文件名中不包含 \\\".ee.\\\" 的文件，包含该字样的文件则受 LICENSE.enterprise 约束”，而 LICENSE.enterprise 说那部分软件“只有在你（以及你所代表的任何实体）拥有与用户座席数量相符的有效 Automatisch 企业许可证时，才能用于生产环境”。GitHub 把这个仓库的许可证报告为 NOASSERTION。","source":"ai","title":"部分代码需要付费许可证"},{"code":"quiet_for_months","severity":"info","evidence":"最后一次提交是在 2026-01-15（在本次检查之前 292 天），最新的发行版 v0.15.0 是在 2025-08-08 发布的。这个项目没有被归档，但已经好几个月没有任何进展。","source":"ai","title":"好几个月没有动静"},{"code":"contributor_licence_agreement_required","severity":"info","evidence":"顶层目录里有一份 CONTRIBUTOR_LICENSE_AGREEMENT.md，所以贡献者在代码被接受之前，要把权利转让给这家公司（LICENSE.enterprise 中提到的 AB Software GmbH）。","source":"ai","title":"需要签署贡献者许可协议"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Each app you connect (for example Slack or Twitter) needs its own login or API credentials"],"cost":{"model":"freemium"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["automation","productivity"],"alternativeTo":[{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"make","name":"Make","confirmed":true}],"declaredAlternatives":[{"target":"zapier","stance":"claims","evidence":"# Automatisch - Open Source Zapier Alternative"}],"stars":14019,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":12,"growthDays":6,"history":[{"date":"2026-11-04","stars":14007},{"date":"2026-11-05","stars":14009},{"date":"2026-11-06","stars":14012},{"date":"2026-11-07","stars":14013},{"date":"2026-11-08","stars":14014},{"date":"2026-11-09","stars":14015},{"date":"2026-11-10","stars":14019}],"firstPublishedAt":"2026-11-04T01:41:48.225Z","commitSha":"41f3c56a163103cae761bb44f03638cb84e859ee","verifiedAt":"2026-11-04T01:21:50.185Z","lastCommitAt":"2026-01-15T09:57:39Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Connects your apps, like Slack or Twitter, so one action triggers another, running on your own server.","summary":"Automatisch is a workflow tool you install on your own computer or server. You pick a service such as Twitter or Slack, choose what starts the workflow and what happens next, and it runs by itself, without writing code. The point is that the data stays on your machine instead of a company's cloud. Installing it is a technical job: you clone the repository and start it with Docker, which also runs a database and a Redis server. The main version is free; parts of the code marked \".ee.\" need a paid business licence.","easeRationale":"There is no installer or ready-made file in any of the 15 releases. On all three systems the README's route is the same terminal work: install Docker, `git clone` the repository and run `docker compose up`, which starts the app plus a PostgreSQL 14.5 database and Redis 7.0.4. After that you use it in a browser at port 3000.","costNote":"The README says Automatisch Community Edition is open source under AGPL-3.0, and that Automatisch Enterprise Edition is \"a commercial offering\". No prices are given in the repository.","notVerified":["How much disk space and memory the three containers really use, since the checks could not estimate it","Which of the app connections still work with the current APIs of Slack, Twitter and the other services, after months without commits","Whether `docker compose up` still builds today, since the compose file pins PostgreSQL 14.5 and Redis 7.0.4","Which features are missing from the free edition compared with the paid Enterprise edition, which the repository does not list","What the paid Enterprise licence costs"]},"zh-Hans":{"summaryShort":"连接你的应用，比如 Slack 或 Twitter，让一个动作触发另一个，运行在你自己的服务器上。","summary":"Automatisch 是一个你安装在自己电脑或服务器上的工作流工具。你选择一个服务，比如 Twitter 或 Slack，设定什么触发工作流、接下来发生什么，然后它就会自动运行，不用写代码。它的要点是数据留在你自己的机器上，而不是进入某家公司的云端。安装它是一项技术活：你需要克隆仓库，并用 Docker 启动它，Docker 还会同时运行一个数据库和一个 Redis 服务器。主版本是免费的；代码里标记为 \".ee.\" 的部分需要付费的商业许可证。","easeRationale":"15 个发行版里都没有安装程序或现成文件。三个系统上 README 给出的路径是同样的终端操作：安装 Docker，`git clone` 仓库，然后运行 `docker compose up`，这会启动应用本身，还有一个 PostgreSQL 14.5 数据库和 Redis 7.0.4。之后你在浏览器里通过 3000 端口使用它。","costNote":"README 说 Automatisch 社区版以 AGPL-3.0 开源发布，而 Automatisch 企业版是“一项商业产品”。仓库里没有给出具体价格。","notVerified":["三个容器实际占用多少磁盘和内存，因为检查没能估算出来","在几个月没有提交之后，哪些应用连接仍然能配合 Slack、Twitter 和其他服务目前的 API 工作","`docker compose up` 今天是否还能构建成功，因为 compose 文件固定了 PostgreSQL 14.5 和 Redis 7.0.4","相比付费的企业版，免费版缺少哪些功能，仓库里没有列出","付费的企业版许可证要多少钱"]}}},{"id":"R_kgDOOMICbA","slug":"awslabs__mcp","name":"AWS MCP Servers","owner":"awslabs","fullName":"awslabs/mcp","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"aws_points_users_to_a_successor_project","severity":"attention","evidence":"README.md 顶部提示:「The Agent Toolkit for AWS is the successor to the MCP servers, plugins, and skills available on AWS Labs... If you're building production software using coding agents... we recommend Agent Toolkit for AWS. This repo continues to work and accept contributions. Over time, the most useful projects here will move into Agent Toolkit for AWS.」列表中已有多项被标记为「DEPRECATED」(Cloud Control API、Amazon Q Business)。你在这里配置好的东西以后可能需要更换。","source":"ai","title":"AWS 把用户指向一个后续项目"},{"code":"recommended_configs_give_the_assistant_write_access","severity":"attention","evidence":"多个服务器的安装按钮包含「--allow-write」和「--allow-sensitive-data-access」参数(例如 Amazon EKS MCP Server 和 AWS Serverless MCP Server),还有一些设置了「AWS_PROFILE」:「your-aws-profile」。开启这些之后,AI 助手就能以该配置文件的权限创建、修改或删除真实的云资源并读取数据。","source":"ai","title":"推荐配置给了助手写入权限"},{"code":"actions_can_create_real_aws_charges","severity":"info","evidence":"列表中包含会部署和管理付费服务的服务器——EKS 和 ECS 集群、Lambda 和 serverless 应用、Bedrock 知识库和自定义模型导入——还有一个「Cost & Operations」部分。助手创建的任何东西都会计入你的 AWS 账单,直到你删除为止。","source":"ai","title":"这些操作可能产生真实的 AWS 费用"},{"code":"one_repository_with_dozens_of_servers_at_different_stages","severity":"info","evidence":"README 按主题列出了几十个独立的服务器,有的处于预览阶段(「AWS MCP Server (in preview)」),有的已被弃用,仓库本身有 235 个发布版本和 262 个未关闭的 issue。它们的质量和维护水平并不一致,所以要评估你打算使用的那一个具体服务器。","source":"ai","title":"一个仓库里有处于不同阶段的几十个服务器"},{"code":"transport_change_already_broke_older_setups","severity":"info","evidence":"README.md:「On May 26th, 2025, Server Sent Events (SSE) support was removed from all MCP servers in their latest major versions」,服务器现在「support stdio only」。这里的更新已经迫使一些用户停留在旧的主版本上,所以要锁定你依赖的版本。","source":"ai","title":"传输方式的变更已经让旧的配置失效"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["AWS account credentials (AWS profile)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":9903,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":24,"growthDays":6,"history":[{"date":"2026-11-07","stars":9890},{"date":"2026-11-08","stars":9892},{"date":"2026-11-09","stars":9898},{"date":"2026-11-10","stars":9903}],"firstPublishedAt":"2026-11-06T22:59:23.727Z","commitSha":"894547a8d9315bca8beed8476524171e7324c6e2","verifiedAt":"2026-11-06T22:43:12.564Z","lastCommitAt":"2026-11-10T17:12:44Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Official AWS add-ons that let an AI assistant read AWS docs and work with your cloud resources.","summary":"A collection of MCP servers published by AWS Labs. You add one or more of them to an AI assistant such as Cursor, VS Code, Claude Code, Kiro, Cline or Windsurf, and the assistant gains tools for a specific part of AWS: reading the official documentation, writing CloudFormation or CDK, deploying to EKS, ECS, Lambda or serverless, querying databases, checking costs, opening support cases and more. It is for people who already build on AWS: you install the uv tool, paste a JSON config into your assistant and point it at an AWS profile.","easeRationale":"The same work on Windows, Mac and Linux: install the uv tool, add a JSON block to your AI assistant's MCP config (the README has one-click buttons for Kiro, Cursor and VS Code) and set up an AWS profile with the right permissions. There is no standalone app - the servers only do something inside an assistant that already speaks MCP, over stdio.","costNote":"The servers are free and open source (Apache-2.0). Anything they do in AWS - Bedrock calls, clusters, storage - is billed to your AWS account.","notVerified":["Which of the listed servers work with your assistant and in your AWS region today","What each server sends to AWS or to a remote managed endpoint while you use it","What the assistant will actually do with write access enabled, and what it would cost","Whether the published uvx packages match the code in this repository"]},"zh-Hans":{"summaryShort":"AWS 官方插件,让 AI 助手读取 AWS 文档并操作你的云资源。","summary":"这是 AWS Labs 发布的一组 MCP 服务器集合。你可以把其中一个或多个添加到 Cursor、VS Code、Claude Code、Kiro、Cline 或 Windsurf 等 AI 助手里,助手就能获得针对 AWS 某个具体领域的工具:阅读官方文档、编写 CloudFormation 或 CDK、部署到 EKS、ECS、Lambda 或 serverless、查询数据库、检查费用、开支持工单等等。它面向已经在 AWS 上构建的人:你安装 uv 工具,把一段 JSON 配置粘贴进助手,并指向一个 AWS 配置文件(profile)。","easeRationale":"在 Windows、Mac 和 Linux 上做法相同:安装 uv 工具,把一段 JSON 添加到你的 AI 助手的 MCP 配置里(README 为 Kiro、Cursor 和 VS Code 提供了一键按钮),并配置一个拥有正确权限的 AWS 配置文件。这不是一个独立应用——这些服务器只有在已经支持 MCP(通过 stdio)的助手内部才会发挥作用。","costNote":"这些服务器是免费开源的(Apache-2.0)。它们在 AWS 上做的任何事——调用 Bedrock、集群、存储——都会计入你的 AWS 账单。","notVerified":["列表中的哪些服务器今天能在你的助手和你所在的 AWS 区域正常工作","使用期间每个服务器会向 AWS 或远程托管端点发送什么内容","开启写入权限后,助手实际会做什么,又会花多少钱","发布的 uvx 包是否与这个仓库里的代码一致"]}}},{"id":"R_kgDORhV56Q","slug":"aymenhmaidiwastaken__unsender-for-instagram","name":"Unsender for Instagram","owner":"aymenhmaidiwastaken","fullName":"aymenhmaidiwastaken/unsender-for-instagram","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"automating_instagram_can_cost_you_the_account","severity":"attention","evidence":"README.md 的免责声明:「Using automation tools may violate Instagram's Terms of Service. Use this extension at your own risk」,以及「Instagram may impose rate limits on accounts that unsend messages too quickly」。功能列表里还提供「Configurable delays... to stay under Instagram's radar」,用来让自动化操作不被察觉。限制或封号的后果由你自己承担。","source":"ai","title":"自动化操作 Instagram 可能让你失去账号"},{"code":"deletions_cannot_be_undone","severity":"attention","evidence":"README.md 的 FAQ 写道:「the message will disappear from both sides of the conversation」,而这个工具会在无人看管的情况下以每分钟约 8 到 15 条消息的速度遍历整个历史记录。没有预览,也没有办法找回消息,无论是你这一侧还是对方那一侧。","source":"ai","title":"删除操作无法撤销"},{"code":"built_on_instagram_page_buttons_and_left_untouched","severity":"info","evidence":"这个扩展通过查找九种语言里的「Unsend」按钮文字来工作,README 本身也警告:「This extension interacts with Instagram's web interface, which may change at any time」。最后一次提交是在 2026-03-08,距本次评审 243 天,没有发布版本,16 颗星,只有一位作者。","source":"ai","title":"建立在 Instagram 页面按钮之上且已无人维护"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":16,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":16},{"date":"2026-11-08","stars":16},{"date":"2026-11-09","stars":16},{"date":"2026-11-10","stars":16}],"firstPublishedAt":"2026-11-06T22:59:23.735Z","commitSha":"0c5a5217a0e61b048ebfe3428a7cad400c03d79f","verifiedAt":"2026-11-06T22:43:51.092Z","lastCommitAt":"2026-03-08T07:45:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Chrome extension that goes through one Instagram DM chat and unsends every message you sent, one by one.","summary":"A small Chrome extension for people who want to wipe their own side of an Instagram conversation. You open the chat in Chrome, click the extension, set how many seconds to wait between deletions (3 to 7 by default) and press Start; it scrolls the whole history, finds the messages you sent and clicks Unsend on each, showing a counter on the page. The other person's messages are left alone. It is not in the Chrome Web Store: you download the folder and load it with developer mode on. Works in Chrome, Edge, Brave and other Chromium browsers.","easeRationale":"Same steps on Windows, Mac and Linux, and nothing to install beyond a Chromium browser - but there is no store listing or installer: you download the repository, open chrome://extensions, turn on Developer mode and load the unpacked folder by hand, which Chrome then warns about on every start.","notVerified":["Whether the code really keeps everything local, as the README claims, since a content script can read the page you are on","Whether it finds every message in a long or group conversation, or skips some","Whether Instagram limits or blocks the account after a long run","Whether it still works with the current Instagram web pages, given the last change was in March 2026"]},"zh-Hans":{"summaryShort":"一个 Chrome 扩展,遍历一个 Instagram 私信会话,逐条撤回你发送过的每一条消息。","summary":"这是一个小型 Chrome 扩展,适合想清空自己在某个 Instagram 对话里发送内容的人。你在 Chrome 里打开聊天,点击扩展,设置每次删除之间等待的秒数(默认 3 到 7 秒),然后点击 Start;它会滚动整个聊天记录,找到你发送的消息并逐条点击 Unsend,页面上会显示一个计数器。对方发送的消息不受影响。它不在 Chrome 网上应用店里:你需要下载这个文件夹,并打开开发者模式来加载。可在 Chrome、Edge、Brave 等基于 Chromium 的浏览器中使用。","easeRationale":"在 Windows、Mac 和 Linux 上步骤相同,除了一个 Chromium 浏览器之外不需要安装别的——但它没有商店页面,也没有安装程序:你需要下载这个仓库,打开 chrome://extensions,开启开发者模式,手动加载解压后的文件夹,之后 Chrome 每次启动都会对此发出警告。","notVerified":["代码是否真的像 README 所说的那样完全在本地运行,因为一个 content script 能读取你所在的页面","在很长或群组对话里,它是否能找到所有消息,还是会漏掉一些","长时间运行之后,Instagram 是否会限制或封禁这个账号","它是否还能适配 Instagram 当前的网页,因为最后一次改动是在 2026 年 3 月"]}}},{"id":"R_kgDOQ4PBWw","slug":"ayuuxh2__insta-p8","name":"Insta P8","owner":"ayuuxh2","fullName":"ayuuxh2/insta-p8","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞 (OSV)：next: GHSA-2464-fhr4-w9xw, GHSA-267c-6grr-h53f, GHSA-26hh-7cqf-hhc6; zod: GHSA-mrhx-c7xc-m3w5","source":"check","title":"已知漏洞"},{"code":"depends_on_instagram_rules_and_app_review","severity":"attention","evidence":"README 自己的“重要说明”写道“Instagram 的 API 需要正确的 Meta 应用权限，生产环境使用还需要审核”、“一些功能可能要求你的 Instagram 账号是商业或创作者类型”，以及“遵守 Instagram 平台条款和反垃圾信息政策”。功能列表里包括“关注后才能解锁内容的流程”和“回复前的随机延迟，让行为更自然”。Meta 是否批准你的应用，以及你的账号被允许进行哪种自动化消息发送，由 Meta 决定，不是这个项目能决定的。","source":"ai","title":"依赖于 instagram 的规则和应用审核"},{"code":"database_protection_depends_on_manual_steps","severity":"attention","evidence":"README 说运行时的迁移执行器不会运行 `CREATE POLICY` 或 `ALTER TABLE ... ENABLE ROW LEVEL SECURITY`，所以如果你跳过了把 `schema.sql` 粘贴到 SQL 编辑器这一步，表就可能在没有行级访问保护的情况下存在。架构列表里还包括 `lib/supabase-admin.ts`——“服务角色客户端（绕过 RLS）”——安装过程还要求你保存 Supabase 的服务角色密钥，而这个密钥对你的数据库拥有完全访问权限。","source":"ai","title":"数据库保护依赖人工步骤"},{"code":"handles_private_messages_of_other_people","severity":"attention","evidence":"功能里包括“自动把收到的私信保存到收件箱”、“存储收到和发出的 Instagram 消息”，以及一个实时对话面板，所以每个给这个账号写私信的人，他们的私信最终都会进入你的 Supabase 数据库，而且当 AI 回复读取“最近的对话历史”时，这些内容也会进入 AI 提供方的请求里。","source":"ai","title":"处理其他人的私信"},{"code":"young_project_built_with_ai_tooling","severity":"info","evidence":"这个仓库创建于 2026-01-12，package.json 里仍然写着 `\"name\": \"my-v0-project\"`、`\"version\": \"0.1.0\"`、`\"private\": true`——这是 v0 代码生成器的默认值。顶层目录里有三种不同的锁定文件（bun.lock、package-lock.json、pnpm-lock.yaml）、一个 `.codex/` 文件夹、`.mcp.json` 和 used-knowledge.md，README 里还有一份“SEO 使用场景”关键词列表，而“为 API 路由添加测试”被列为一个适合新手的 issue，这说明目前还没有测试。","source":"ai","title":"用 ai 工具搭建的年轻项目"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Supabase (project URL, anon key and service role key)","Meta / Instagram developer app (INSTAGRAM_APP_ID, INSTAGRAM_APP_SECRET, webhook token)","Groq, or any OpenAI-compatible provider, for the AI replies"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation","ai_agents"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[{"target":"invideo-ai","stance":"claims","evidence":"<strong>Free open-source ManyChat alternative — Instagram DM automation, comment-to-DM funnels, story replies, AI auto-replies, and a live inbox.</strong>"},{"target":"manychat","stance":"claims","evidence":"<strong>Free open-source ManyChat alternative — Instagram DM automation, comment-to-DM funnels, story replies, AI auto-replies, and a live inbox.</strong>"},{"target":"chatfuel","stance":"claims","evidence":"A self-hosted, AI-ready, developer-friendly alternative to paid Instagram automation tools like ManyChat, Chatfuel, and SendPulse"},{"target":"sendpulse","stance":"claims","evidence":"A self-hosted, AI-ready, developer-friendly alternative to paid Instagram automation tools like ManyChat, Chatfuel, and SendPulse"}],"stars":182,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":4,"growthDays":6,"history":[{"date":"2026-11-04","stars":178},{"date":"2026-11-05","stars":179},{"date":"2026-11-06","stars":180},{"date":"2026-11-07","stars":180},{"date":"2026-11-08","stars":180},{"date":"2026-11-09","stars":181},{"date":"2026-11-10","stars":182}],"firstPublishedAt":"2026-11-04T01:41:48.245Z","commitSha":"5f4945fb8714de9e9264d7e89e59534da061e759","verifiedAt":"2026-11-04T01:22:57.354Z","lastCommitAt":"2026-10-22T15:01:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Replies to your Instagram DMs, comments and story replies by itself, with keyword rules or an AI answer.","summary":"InstaAuto is a dashboard you install yourself to automate an Instagram Business or Creator account: keyword replies in DMs, a public comment reply plus a private message with your link, answers to story replies, and an AI reply when no rule matches. It also has an inbox for answering by hand. Setting it up is developer work: a Supabase database, a Meta developer app for Instagram, environment variables and a public web address for the webhooks. The code is MIT and the services it uses have free tiers.","easeRationale":"None of the 4 releases has a file to download: on Windows, macOS and Linux you do the same work. You clone the repository, run `npm install`, create a Supabase project, paste the whole of `schema.sql` into its SQL editor, enable the `pg_cron` extension, create a Meta developer app with Instagram Login and webhooks, fill in ten environment variables and publish the site (the README's route is a fork deployed to Vercel) so Instagram can reach the webhook. The README itself aims the self-hosted route at \"developers, agencies, privacy-focused teams, power users\".","costNote":"The code is MIT. The README says it runs on the Supabase and Vercel free tiers and points to a free Groq key for the AI replies; heavier use would mean paying those services.","notVerified":["Whether the hosted \"web tester\" option really exists and is open, since the README gives no address for it and says access \"may require manual approval\" from Meta","Whether Meta approves a new app for the Instagram permissions this needs, and how long that takes","Whether the automations work as described, which can only be seen by connecting a real Instagram Business account","Whether the Supabase and Vercel free tiers hold up for an account with many messages","What the AI replies really cost if you move off Groq's free key","Real disk use, since the figure here is an estimate"]},"zh-Hans":{"summaryShort":"自动回复你 Instagram 的私信、评论和快拍回复，用关键词规则或 AI 回答。","summary":"InstaAuto 是一个你自己安装的面板，用来自动化一个 Instagram 商业或创作者账号：私信里的关键词回复、公开评论回复加上带你链接的私信、对快拍回复的回答，以及在没有规则匹配时给出的 AI 回复。它还有一个收件箱可以手动回复。配置它是开发者的工作：一个 Supabase 数据库、一个用于 Instagram 的 Meta 开发者应用、环境变量，以及一个供 webhook 使用的公开网址。代码是 MIT 许可，它用到的服务都有免费额度。","easeRationale":"4 个发行版里都没有可下载的文件：在 Windows、macOS 和 Linux 上你要做同样的工作。你需要克隆仓库，运行 `npm install`，创建一个 Supabase 项目，把整份 `schema.sql` 粘贴到它的 SQL 编辑器里，启用 `pg_cron` 扩展，创建一个带 Instagram Login 和 webhook 的 Meta 开发者应用，填写十个环境变量，并发布网站（README 的路线是把一个 fork 部署到 Vercel），这样 Instagram 才能访问到 webhook。README 本身把自托管这条路线定位给“开发者、代理机构、注重隐私的团队、高级用户”。","costNote":"代码是 MIT 许可。README 说它运行在 Supabase 和 Vercel 的免费额度上，并指向一个免费的 Groq 密钥用于 AI 回复；用量更大的话就需要为这些服务付费了。","notVerified":["托管的“网页测试器”选项是否真的存在并且开放，因为 README 没有给出它的网址，而且说访问“可能需要 Meta 的人工审批”","Meta 是否会批准一个新应用使用这里需要的 Instagram 权限，以及需要多长时间","这些自动化功能是否真的像描述那样工作，这只能通过连接一个真实的 Instagram 商业账号才能看到","Supabase 和 Vercel 的免费额度能否支撑一个消息量很大的账号","如果不再用 Groq 的免费密钥，AI 回复实际要花多少钱","真实的磁盘占用，因为这里给出的数字只是估算"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2MTk4MDY1OA==","slug":"azerothcore__azerothcore-wotlk","name":"AzerothCore","owner":"azerothcore","fullName":"azerothcore/azerothcore-wotlk","type":"framework_infra","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"unofficial_server_for_a_commercial_game","severity":"attention","evidence":"该仓库把自己描述为针对 \"world-of-warcraft\" 的 \"emulator\"（标签：emulator、mmorpg-server、wow），docker-compose.yml 里有一个 `ac-tools` 服务，注释是 \"used for extracting maps from files shipped with game client\"。在非 Blizzard 的服务器上游玩违反该游戏的使用条款，并取代了其付费订阅，多个国家已经通过法律行动关闭过此类服务器的运营者。代码本身是该项目在 GPL-2.0 下的原创成果，不包含任何 Blizzard 的文件。","source":"ai","title":"面向商业游戏的非官方服务器"},{"code":"default_database_password_in_compose","severity":"attention","evidence":"docker-compose.yml 设置了 `MYSQL_ROOT_PASSWORD=${DOCKER_DB_ROOT_PASSWORD:-password}`，并把数据库发布到主机：`${DOCKER_DB_EXTERNAL_PORT:-3306}:3306`。按原样启动时，MySQL 的 root 账号密码是 \"password\"，并且监听在容器外部。","source":"ai","title":"compose 中的数据库默认密码"},{"code":"server_ports_published_by_default","severity":"info","evidence":"docker-compose.yml 把 3724（登录）、8085（世界）和 7878（SOAP 远程命令端口）发布到主机上。SOAP 端口接受管理命令，所以任何能连上它并知道一个账号的人都可以对服务器发起操作。","source":"ai","title":"默认发布的服务器端口"},{"code":"install_from_moving_branch","severity":"info","evidence":"最新的带标签发行版是 2024-11-16 的，而最新提交是 2026-11-08 的，Docker 镜像默认使用 `${DOCKER_IMAGE_TAG:-master}`，所以正常安装会跟随 master 分支，而不是固定版本。目前有 2300 个未关闭的 issue。","source":"ai","title":"安装来自不断变化的分支"},{"code":"unofficial_server_for_a_commercial_game","severity":"high","evidence":"描述：\"Complete Open Source and Modular solution for MMO\"；标签：\"emulator\"、\"mmorpg-server\"、\"world-of-warcraft\"、\"wow\"。docker-compose.yml 里有一个 `ac-tools` 服务，注释是 \"used for extracting maps from files shipped with game client\"，worldserver 挂载了提取出的这些 \"client data\"。World of Warcraft 的使用条款不允许把游戏连接到 Blizzard 以外的服务器，提取出的客户端数据是游戏公司自己的资料，这样的服务器取代了游戏的付费订阅；多个国家已经通过法律行动关闭过非官方服务器的运营者。服务器代码本身是该项目在 GPL-2.0 下的原创成果，仓库中不包含任何游戏文件。","source":"ai","title":"面向商业游戏的非官方服务器"}],"platforms":["linux","windows","mac"],"tools":["docker"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["games","network_system"],"alternativeTo":[{"slug":"world-of-warcraft","name":"World of Warcraft","confirmed":true},{"slug":null,"name":"World of Warcraft subscription","confirmed":false}],"declaredAlternatives":[],"stars":9168,"downloads":254,"downloadsWeek":0,"downloadsDays":1,"growth":23,"growthDays":6,"history":[{"date":"2026-11-09","stars":9164},{"date":"2026-11-10","stars":9168}],"firstPublishedAt":"2026-11-08T23:01:18.666Z","commitSha":"b070292f6a9bef091eeccaf644afb8076a0778ef","verifiedAt":"2026-11-08T22:45:37.827Z","lastCommitAt":"2026-11-10T18:58:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs your own World of Warcraft (Wrath of the Lich King) server, for you and your friends, on a PC you control.","summary":"AzerothCore is the server side of World of Warcraft, version 3.3.5a (Wrath of the Lich King), rewritten as open source. You run it yourself: a login server, a world server and a MySQL database, usually started with Docker, plus game data taken from a WoW client you already have. Hundreds of community modules change rates, add bots or new content. There are no ready downloads, so expect the terminal, a few gigabytes of data and some configuration. It is aimed at people who want to host a server, not at players looking for a one-click game.","easeRationale":"There are no binary downloads in any release: on Linux, the best-documented path, you install Docker, run `docker compose up -d --build`, which builds the server, imports three MySQL databases and needs game data extracted from a WoW client, then create accounts by hand in the server console. Windows and Mac follow the same Docker route, and the repository warns that the development container wants Linux or WSL2 to perform well.","costNote":"The server software is free (GPL-2.0). You still need a World of Warcraft 3.3.5a client of your own and a machine to run the server on.","notVerified":["Whether the Docker build finishes on a given machine and how long it takes","How much disk and memory the three databases and the extracted game data really use","Whether a WoW 3.3.5a client you own works with the version in master","How well the community modules behave, since they are kept in other repositories","Whether the server resists abuse once its ports are reachable from the internet"]},"zh-Hans":{"summaryShort":"在你自己掌控的电脑上运行属于你和朋友的魔兽世界（巫妖王之怒）服务器。","summary":"AzerothCore 是魔兽世界 3.3.5a 版（巫妖王之怒）的服务器端，以开源形式重写而成。你需要自己运行：一个登录服务器、一个世界服务器和一个 MySQL 数据库，通常用 Docker 启动，还需要从你已有的 WoW 客户端中取出游戏数据。数百个社区模块可以调整掉落率、加入机器人或新内容。没有现成的下载包，所以要准备好使用终端、几个 GB 的数据和一些配置工作。它面向想要搭建服务器的人，而不是想要一键游玩的玩家。","easeRationale":"所有发行版里都没有二进制下载：在文档最完善的 Linux 上，你要安装 Docker，运行 `docker compose up -d --build`，它会构建服务器、导入三个 MySQL 数据库，还需要从 WoW 客户端提取的游戏数据，然后在服务器控制台里手动创建账号。Windows 和 Mac 走的是同样的 Docker 路线，仓库也提醒开发容器在 Linux 或 WSL2 上才能表现良好。","costNote":"服务器软件本身是免费的（GPL-2.0）。你仍然需要自己的魔兽世界 3.3.5a 客户端，以及一台用来运行服务器的机器。","notVerified":["Docker 构建能否在某台机器上完成，以及需要多长时间","三个数据库和提取出的游戏数据实际占用多少磁盘和内存","你自己拥有的 WoW 3.3.5a 客户端能否配合 master 分支里的版本使用","社区模块的表现如何，因为它们保存在其他仓库里","服务器端口一旦能从互联网访问，是否能抵御滥用"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzOTM1NzE1OTk=","slug":"babysor__mockingbird","name":"MockingBird","owner":"babysor","fullName":"babysor/MockingBird","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"E","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布于 2021-11-07（1823 天前）","source":"check","title":"发布版本已过时"},{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/file/d/110A11KZoVe7vy6kXlLb6zVPLb_J91I_t/view?usp=sharing","source":"check","title":"外部下载链接"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：numpy: GHSA-6p56-wp2h-9hxr, GHSA-fpfv-jqm9-f5jm；numpy: GHSA-6p56-wp2h-9hxr, GHSA-fpfv-jqm9-f5jm","source":"check","title":"已知漏洞"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.5/10","source":"check","title":"scorecard 评分偏低"},{"code":"copies_a_real_persons_voice","severity":"attention","evidence":"项目自我描述为“5 秒克隆一个声音，实时生成任意语音”，gen_voice.py 接收一个文本文件加一个声音 .wav 文件。没有任何环节询问这个人是否同意，也没有把结果标记为生成内容。用别人的声音说出未经许可的话，在不少国家是受限甚至可被处罚的行为。","source":"ai","title":"复制真人的声音"},{"code":"voice_models_come_from_file_sharing_links","severity":"attention","evidence":"README.md 第 2.3 节列出的现成声音模型只是社区成员发布在 pan.baidu.com、aliyundrive.com 和个人 OneDrive 上的带密码链接（例如“https://pan.baidu.com/s/1iONvRxmkI-t1nHqxKytY3g 4j5d”）。其中三个写着“仅适用于 0.0.1 版本”。这些都是陌生人提供的 PyTorch 模型文件，程序会直接加载它们。","source":"ai","title":"声音模型来自文件分享链接"},{"code":"author_stopped_and_points_to_a_paid_service","severity":"attention","evidence":"README.md 第一行：“虽然我不再积极更新这个仓库，但你可以看到我一直在推动这项技术……我也在打造一个优化并托管在云端的版本：https://noiz.ai/”。最后一个发布版本是 2021 年的，还有 481 个未关闭的 issue。","source":"ai","title":"作者已停止维护并指向一个付费服务"},{"code":"only_works_with_old_versions","severity":"info","evidence":"README.md：“这里推荐的环境是仓库标签 0.0.1、Pytorch1.9.0 配 Torchvision0.10.0 和 cudatoolkit10.2……因为 requirements.txt 是几个月前导出的，所以不能配合更新的版本使用”。requirements.txt 在 Windows 上仍固定 numpy==1.19.3，其他系统上固定 numpy==1.20.3，正是检查标记出已知漏洞的那些版本。","source":"ai","title":"只能配合旧版本使用"},{"code":"chinese_only_and_cli_demo_broken","severity":"info","evidence":"功能列表只列出普通话，该项目是 Real-Time-Voice-Cloning 的一个分支，“只支持英文”，第 2 节提醒“demo_cli 目前无法工作，因此需要额外的合成器模型”。没有中文模型的话，就没有任何现成的东西可以试用。","source":"ai","title":"仅支持中文，且命令行演示已损坏"},{"code":"training_needs_a_big_machine","severity":"info","evidence":"训练使用 aidatatang_200zh 和 aishell3 等数据集（数十 GB 的音频），docker-compose.yml 预留了一个 Nvidia 设备，FAQ 第 6 条告诉 Windows 用户“把虚拟内存改为 100G（102400）”。","source":"ai","title":"训练需要一台强大的机器"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg","gpu_nvidia"],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":36876,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":7,"history":[{"date":"2026-11-05","stars":36880},{"date":"2026-11-06","stars":36872},{"date":"2026-11-07","stars":36871},{"date":"2026-11-08","stars":36872},{"date":"2026-11-09","stars":36873},{"date":"2026-11-10","stars":36876}],"firstPublishedAt":"2026-11-04T22:32:54.821Z","commitSha":"28dc5e14f12d7c754612af2fde8e78a4b03f8616","verifiedAt":"2026-11-04T22:17:09.400Z","lastCommitAt":"2026-03-03T14:59:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from a short recording and makes it read Chinese text out loud, on your own computer.","summary":"MockingBird copies the sound of a voice from a recording and then reads text out loud in that voice. It is built for Mandarin Chinese: the project says the original English model does not fit Chinese letters, so you either train your own voice model on large Chinese speech datasets or download one of the ready models that community members shared on file-sharing links. You run it from the terminal with Python, PyTorch and ffmpeg, and it then opens a page at http://localhost:8080 or a desktop window. An Nvidia graphics card is expected.","easeRationale":"Windows and Linux are the least hard and set the grade, but it is still a terminal job: install Python, PyTorch with the right CUDA version and ffmpeg, run \"pip install -r requirements.txt\", find a voice model, then start \"python web.py\" or \"python demo_toolbox.py\". On a Mac with Apple chip the README only offers a workaround: a Rosetta terminal, building pyworld and ctc-segmentation from source code and running Python through an x86 wrapper script, which is developer work.","notVerified":["How good the cloned voice really sounds, and whether any of the shared models still load with today's PyTorch","What is inside the model files on the Baidu and OneDrive links, since the project does not publish checksums","Whether the install works at all with current Python and PyTorch instead of the 2021 versions","Real disk space and training time, which depend on the Chinese datasets you download","Whether it can be made to work for languages other than Mandarin"]},"zh-Hans":{"summaryShort":"从一段简短录音中复制声音，让它在你自己的电脑上朗读中文文本。","summary":"MockingBird 从一段录音中复制声音的音色，然后用这个声音朗读文本。它专为普通话打造：项目说原来的英文模型不适合汉字，所以你要在大型中文语音数据集上训练自己的声音模型，或者下载社区成员通过文件分享链接提供的现成模型。你通过终端配合 Python、PyTorch 和 ffmpeg 运行它，之后它会打开一个 http://localhost:8080 页面或一个桌面窗口。通常需要一张 Nvidia 显卡。","easeRationale":"Windows 和 Linux 难度相对最低，决定了评级，但仍然是终端操作：安装 Python、匹配 CUDA 版本的 PyTorch 和 ffmpeg，运行 “pip install -r requirements.txt”，找到一个声音模型，再启动 “python web.py” 或 “python demo_toolbox.py”。在搭载 Apple 芯片的 Mac 上，README 只提供一个变通方案：使用 Rosetta 终端，从源码编译 pyworld 和 ctc-segmentation，并通过一个 x86 包装脚本运行 Python，这属于开发者级别的工作。","notVerified":["克隆的声音实际听起来如何，以及那些共享模型是否还能配合当前的 PyTorch 加载","百度网盘和 OneDrive 链接里的模型文件里到底有什么，因为项目没有发布校验值","在当前版本的 Python 和 PyTorch（而非 2021 年的版本）上能否正常安装","实际占用的磁盘空间和训练时间，这取决于你下载的中文数据集","能否改造成支持普通话以外的其他语言"]}}},{"id":"R_kgDOVfQKQQ","slug":"balanov__uncut","name":"uncut","owner":"Balanov","fullName":"Balanov/uncut","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"runs_commands_with_no_sandbox_and_no_confirmation","severity":"attention","evidence":"README 的 Safety 部分写道：\"UNCUT is not a sandbox. The model can execute shell commands with your operating-system user's permissions. Workspace checks on file tools do not constrain what a shell command can access\"，以及 \"There are no approval popups\"。模型的命令会在你看到之前就执行。","source":"ai","title":"没有沙箱也无需确认即可运行命令"},{"code":"instructions_from_the_project_folder_go_to_the_model","severity":"attention","evidence":"README 的 Configuration 部分写道：\"`AGENTS.md` is read in full from the workspace root and added to the system prompt.\" 如果你把它指向从别人那里下载的代码，那个文件里的文字就会引导模型，而模型还能运行命令。`--no-agents` 标志可以关闭这个功能。","source":"ai","title":"项目文件夹里的指令会传给模型"},{"code":"unsigned_downloads","severity":"attention","evidence":"README：\"The binaries are not code-signed yet, so SmartScreen or Gatekeeper may warn. Verify the SHA-256 checksum before allowing a downloaded executable.\" 每个发布版本都附带一个 `checksums.txt` 供你核对。","source":"ai","title":"下载文件没有签名"},{"code":"whole_conversation_sent_every_time","severity":"info","evidence":"README：\"The entire conversation is sent on every model call\"，不会做任何缩减。对于付费服务商来说，这会让每多一步的花费都比那些会精简历史记录的工具更高。","source":"ai","title":"每次都会发送整段对话"},{"code":"session_files_can_hold_secrets","severity":"info","evidence":"README 的 Sessions 部分写道：\"Session files can contain source code, command output, and secrets. Store them accordingly and do not commit them accidentally.\" 只有在使用 `--session` 时才会写入会话文件。","source":"ai","title":"会话文件里可能包含敏感信息"},{"code":"very_new_repository","severity":"info","evidence":"项目数据：创建于 2026-11-01，同一天发布了两个版本，最新标签是 v0.1.1，已有 574 个星标。README 把它描述为 837 行 Go 代码，规模很小，但目前还没有可供参考的历史记录。","source":"ai","title":"仓库非常新"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["go"],"disk":{"minMB":6,"maxMB":19,"basis":"installer"},"apiKeys":["OpenAI or another provider that uses the same format (not needed when the model runs on your own computer)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents","automation"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true}],"declaredAlternatives":[],"stars":263,"downloads":242,"downloadsWeek":83,"downloadsDays":7,"growth":0,"growthDays":7,"history":[{"date":"2026-11-03","stars":592},{"date":"2026-11-04","stars":600},{"date":"2026-11-05","stars":607},{"date":"2026-11-06","stars":251},{"date":"2026-11-07","stars":255},{"date":"2026-11-08","stars":257},{"date":"2026-11-09","stars":260},{"date":"2026-11-10","stars":263}],"firstPublishedAt":"2026-11-03T02:51:37.295Z","commitSha":"0beea8f7a25ba340512ed4f5ce3b80f6b8ffe063","verifiedAt":"2026-11-03T02:39:10.529Z","lastCommitAt":"2026-11-09T13:12:18Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A small command-line program that lets an AI model read, write and run commands inside one folder of code.","summary":"A single small program for the terminal that gives an AI model four abilities inside a folder you choose: read a file, write a file, replace one piece of text, and run a command. There is no window and no menus; you type your request, the answer comes back as text, and the whole conversation is sent to the model every time, with nothing shortened. It works with OpenAI and any service that speaks the same format, including a model running on your own computer. It is aimed at programmers and at scripts that run without a person watching.","easeRationale":"On all three systems you download one ready-made file from the releases page, so nothing has to be compiled, but there is no installer: the README's steps copy the file into a folder, add that folder to your PATH and set environment variables, and everything happens in a terminal. Windows uses PowerShell commands for the same steps.","costNote":"The program is free (MIT). You pay the model provider for every request, unless you point it at a free local model server such as Ollama.","notVerified":["Whether nothing is really ever shortened in daily use; the README offers a proof script, but we did not run it","Whether a chosen model actually calls the four tools correctly, which the README says depends on the model","What a real task costs at a paid provider, since the full conversation is resent on every step","Whether the downloaded files match the published checksums","How well it behaves on Windows, where the README warns that some commands produce output it refuses to read"]},"zh-Hans":{"summaryShort":"一个小巧的命令行程序，让 AI 模型在一个代码文件夹里读取、写入文件并运行命令。","summary":"一个供终端使用的小型独立程序，赋予 AI 模型在你选定的文件夹内的四种能力：读取文件、写入文件、替换一段文字，以及运行命令。没有窗口也没有菜单；你输入请求，答案以文本形式返回，而且每次都会把整段对话完整发送给模型，不会做任何缩减。它可以配合 OpenAI 以及任何使用同一格式的服务使用，包括在你自己电脑上运行的模型。它面向的是程序员，以及无需人工看守运行的脚本。","easeRationale":"三个系统都是从发布页面下载一个现成的文件，不需要编译任何东西，但没有安装程序：README 的步骤是把文件复制到一个文件夹、把该文件夹加入 PATH，并设置环境变量，整个过程都在终端里完成。Windows 用的是 PowerShell 命令来完成同样的步骤。","costNote":"该程序是免费的（MIT 许可）。你要为每次请求向模型提供商付费，除非把它指向一个免费的本地模型服务器，比如 Ollama。","notVerified":["日常使用中是否真的从不缩减内容；README 提供了一个验证脚本，但我们没有运行它","选定的模型是否真的能正确调用这四个工具，README 说这取决于具体模型","在付费服务商上完成一次真实任务要花多少钱，因为每一步都会重新发送完整对话","下载的文件是否与公布的校验值一致","它在 Windows 上表现如何，README 提到有些命令产生的输出它会拒绝读取"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0NTA1NTY5Mw==","slug":"balena-io__etcher","name":"balenaEtcher","owner":"balena-io","fullName":"balena-io/etcher","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖项存在已知漏洞（OSV）：electron-updater: GHSA-9jxc-qjr9-vjxq; immutable: GHSA-v56q-mh7h-f735, GHSA-wf6x-7x77-mvgw, GHSA-xvcm-6775-5m9r; lodash: GHSA-f23m-r3pf-42rh, GHSA-r5fr-rjxr-66jc, GHSA-xxjr-mmjv-4gpg; uuid: GHSA-w5hq-g745-h8pq","source":"check","title":"存在已知漏洞"},{"code":"crash_reports_to_sentry","severity":"info","evidence":"package.json 在运行时依赖中列出了 \"@sentry/electron\": \"^4.24.0\"，因此该应用会把崩溃和错误报告发送到发布者运营的 Sentry 服务。","source":"ai","title":"向 Sentry 发送崩溃报告"},{"code":"erases_the_chosen_drive_and_needs_admin_rights","severity":"info","evidence":"写入镜像会替换所选存储卡或 U 盘上的全部内容。该应用需要直接访问磁盘：package.json 依赖 @ronomon/direct-io、drivelist 和 mountutils，在 Debian 系统上需要 polkit 认证代理，在 Windows 上附带 winusb-driver-generator。项目称它会把你的内置硬盘从列表中排除。","source":"ai","title":"会清空所选磁盘，且需要管理员权限"},{"code":"updates_itself","severity":"info","evidence":"package.json 依赖 electron-updater 6.1.8，因此已安装的应用可以自行下载并应用更新。自动检查显示，这个固定的确切版本存在一条已知漏洞公告。","source":"ai","title":"会自我更新"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":262,"maxMB":785,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["network_system","hardware"],"alternativeTo":[{"slug":"wintousb","name":"WinToUSB","confirmed":true}],"declaredAlternatives":[],"stars":34680,"downloads":319023921,"downloadsWeek":475538,"downloadsDays":7,"growth":25,"growthDays":7,"history":[{"date":"2026-10-26","stars":34598},{"date":"2026-10-27","stars":34605},{"date":"2026-10-28","stars":34613},{"date":"2026-10-29","stars":34616},{"date":"2026-10-30","stars":34626},{"date":"2026-10-31","stars":34628},{"date":"2026-11-01","stars":34639},{"date":"2026-11-02","stars":34650},{"date":"2026-11-03","stars":34655},{"date":"2026-11-04","stars":34655},{"date":"2026-11-05","stars":34663},{"date":"2026-11-06","stars":34664},{"date":"2026-11-07","stars":34669},{"date":"2026-11-08","stars":34674},{"date":"2026-11-09","stars":34676},{"date":"2026-11-10","stars":34680}],"firstPublishedAt":"2026-10-27T01:10:12.608Z","commitSha":"dbd6aaed37c4e844f7aed978db01d9f38680eff8","verifiedAt":"2026-10-28T01:41:08.449Z","lastCommitAt":"2026-10-22T19:23:36Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"balenaEtcher-2.1.9.Setup.exe","sizeBytes":274253464,"url":"https://github.com/balena-io/etcher/releases/download/v2.1.9/balenaEtcher-2.1.9.Setup.exe","releaseUrl":"https://github.com/balena-io/etcher/releases/tag/v2.1.9"},{"os":"mac","file":"balenaEtcher-2.1.9-arm64.dmg","sizeBytes":201565645,"url":"https://github.com/balena-io/etcher/releases/download/v2.1.9/balenaEtcher-2.1.9-arm64.dmg","releaseUrl":"https://github.com/balena-io/etcher/releases/tag/v2.1.9"},{"os":"linux","file":"balena-etcher_2.1.9_amd64.deb","sizeBytes":151512800,"url":"https://github.com/balena-io/etcher/releases/download/v2.1.9/balena-etcher_2.1.9_amd64.deb","releaseUrl":"https://github.com/balena-io/etcher/releases/tag/v2.1.9"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Writes an operating system image onto a USB stick or SD card, and checks every byte afterwards.","summary":"balenaEtcher copies a downloaded operating system image (a .iso or .img file, such as Raspberry Pi OS or a Linux install disc) onto a USB stick or SD card. You pick the file, pick the drive and press Flash; it hides your internal disks from the list and verifies the written data afterwards. It can also write directly to Raspberry Pi boards that support USB device boot mode. It is for anyone, with ready installers for Windows 10 and later, macOS 10.13 and later and most 64-bit Linux distributions.","easeRationale":"Every release carries a ready installer for each system: balenaEtcher-2.1.9.Setup.exe (262 MB) for Windows, .dmg files for Intel and Apple Silicon Macs, and .deb, .rpm and zip builds for Linux, plus winget, Chocolatey and AUR packages. You open it and use the window; nothing else has to be installed.","costNote":"Free and open source (Apache-2.0). It is made by Balena Ltd, which sells separate paid cloud products; you do not need an account for Etcher.","notVerified":["Whether a given image file and a given card or card reader really work together","What the crash reports contain, for example file names or drive models, and whether they can be turned off","How long a write takes, which depends on the card and the USB port","Whether the Raspberry Pi USB device boot flashing works on your board","How much space the installed app takes after unpacking (the installers are 118-271 MB)"]},"zh-Hans":{"summaryShort":"把操作系统镜像写入 U 盘或 SD 卡，写完之后逐字节校验。","summary":"balenaEtcher 把下载好的操作系统镜像文件（.iso 或 .img，例如 Raspberry Pi OS 或某个 Linux 安装盘）写入 U 盘或 SD 卡。你选择文件、选择盘位，然后点 Flash；它会把列表中你的内部磁盘隐藏起来，写入完成后再校验写入的数据。它还可以直接写入支持 USB 设备启动模式的 Raspberry Pi 主板。适合任何人使用，提供适用于 Windows 10 及更新版本、macOS 10.13 及更新版本，以及大多数 64 位 Linux 发行版的现成安装包。","easeRationale":"每个发布版本都为各系统提供现成的安装包：Windows 用的 balenaEtcher-2.1.9.Setup.exe（262 MB），Intel 和 Apple Silicon Mac 用的 .dmg 文件，以及 Linux 用的 .deb、.rpm 和 zip 版本，还有 winget、Chocolatey 和 AUR 包。你打开它，用窗口操作即可；不需要再安装别的东西。","costNote":"免费且开源（Apache-2.0）。它由 Balena Ltd 制作，该公司另外出售付费云产品；使用 Etcher 不需要账号。","notVerified":["某个具体的镜像文件和某张卡或读卡器是否真的能配合使用。","崩溃报告里包含什么内容，例如文件名或驱动器型号，以及能否关闭上报。","写入需要多长时间，这取决于卡本身和 USB 接口。","Raspberry Pi 的 USB 设备启动写入方式在你的主板上是否可用。","安装的应用解压后实际占用多少空间（安装包本身为 118-271 MB）。"]}}},{"id":"R_kgDOO4LQPA","slug":"beehiveinnovations__pal-mcp-server","name":"PAL MCP Server","owner":"BeehiveInnovations","fullName":"BeehiveInnovations/pal-mcp-server","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"install_option_runs_the_latest_code_from_git","severity":"attention","evidence":"安装\"Option B\"会在你的助手配置里写入这一行:`uvx --from git+https://github.com/BeehiveInnovations/pal-mcp-server.git pal-mcp-server` ——没有锁定版本,所以每次启动都会下载并运行仓库主分支当时的内容,而你的 API 密钥就在同一个配置块里。","source":"ai","title":"安装选项运行的是 git 上的最新代码"},{"code":"your_files_go_to_several_ai_companies","severity":"info","evidence":"README 中写道:\"Smart file handling - Auto-expand directories\",在代码审查示例的第 6 步里也写着\"Shares the relevant files, findings, etc with Gemini Pro\"——你正在处理的代码会被发送给你启用的那些服务商。","source":"ai","title":"你的文件会发给好几家 ai 公司"},{"code":"no_updates_for_eleven_months","severity":"info","evidence":"最后一次提交和最后一次发布都是 2025-12-15,距检查时 325 天,还有 153 个未关闭的 issue,而 conf/ 和 .env.example 里的模型列表一直在变。","source":"ai","title":"十一个月没有更新"},{"code":"renamed_project","severity":"info","evidence":"README 写道\"Formerly known as Zen MCP\",并链接到 docs/name-change.md,所以你在别处找到的指南和配置可能仍在使用旧名字。","source":"ai","title":"项目改过名字"},{"code":"half_the_tools_are_off_by_default","severity":"info","evidence":"README 的 Tool Configuration 部分写道:\"DISABLED_TOOLS=analyze,refactor,testgen,secaudit,docgen,tracer\"——这六个工具只有在你编辑这份列表之后才会启用,因为\"Each tool adds to context window usage\"。","source":"ai","title":"一半的工具默认是关闭的"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":["python","docker"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["At least one model provider: Google Gemini, OpenAI, Azure OpenAI, X.AI Grok, DIAL or OpenRouter","A local Ollama or LM Studio model instead of a paid key"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"coderabbit","name":"CodeRabbit","confirmed":true}],"declaredAlternatives":[],"stars":11838,"downloads":1773,"downloadsWeek":65,"downloadsDays":4,"growth":14,"growthDays":6,"history":[{"date":"2026-11-06","stars":11825},{"date":"2026-11-07","stars":11828},{"date":"2026-11-08","stars":11831},{"date":"2026-11-09","stars":11832},{"date":"2026-11-10","stars":11838}],"firstPublishedAt":"2026-11-05T22:43:48.279Z","commitSha":"7afc7c1cc96e23992c8f105f960132c657883bb1","verifiedAt":"2026-11-05T22:31:14.363Z","lastCommitAt":"2025-12-15T17:07:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets Claude Code, Codex or Gemini CLI call other AI models for reviews, debates and second opinions.","summary":"PAL MCP (formerly Zen MCP) is an MCP server you plug into a coding assistant such as Claude Code, Codex CLI, Gemini CLI or Cursor. From inside your usual chat you can send a task to other models — Gemini, OpenAI, Grok, Azure, OpenRouter, DIAL or a local Ollama model — and it keeps the conversation going between them: code review by two models, a planner, a debug walkthrough, a pre-commit check, a consensus debate. You install it with git and uv, put at least one model key in a .env file and point your assistant's config at it.","easeRationale":"On Mac and Linux it is a C: the prerequisites are Python 3.10+, Git and uv, then you clone the repository and run ./run-server.sh, which writes the settings for Claude Code, Gemini CLI, Codex CLI and Qwen CLI by itself. Windows is a D because the README's commands are bash and the project keeps a separate \"WSL Setup - Windows users\" guide, so you first set up the Linux subsystem.","costNote":"The server is free (Apache 2.0), but it only does something with a model key you pay for; the README also lists Ollama local models as a free option.","notVerified":["How much the calls to Gemini, OpenAI or Grok cost in a normal day of work","Whether run-server.ps1 really sets everything up on Windows without the Linux subsystem","Whether the model names in conf/ still exist at the providers, after almost a year without updates","How well the local Ollama option works compared with the paid models","What exactly each tool sends to each provider on a real project"]},"zh-Hans":{"summaryShort":"让 Claude Code、Codex 或 Gemini CLI 调用其他 AI 模型进行代码审查、讨论和获取第二意见。","summary":"PAL MCP(原名 Zen MCP)是一个 MCP 服务器,你可以把它接入 Claude Code、Codex CLI、Gemini CLI 或 Cursor 等编码助手。在平常的对话里,你可以把任务发给其他模型——Gemini、OpenAI、Grok、Azure、OpenRouter、DIAL,或本地的 Ollama 模型——它会让这些模型之间保持对话:两个模型一起做代码审查、做规划、走一遍调试过程、提交前检查,或者进行共识讨论。安装时需要用 git 和 uv,在 .env 文件里至少填一个模型的密钥,再把助手的配置指向它。","easeRationale":"在 Mac 和 Linux 上是 C 级:前提条件是 Python 3.10+、Git 和 uv,然后克隆仓库并运行 ./run-server.sh,它会自动写好 Claude Code、Gemini CLI、Codex CLI 和 Qwen CLI 的配置。Windows 是 D 级,因为 README 里的命令都是 bash,项目还保留了一份单独的指南 \"WSL Setup - Windows users\",所以你得先配置 Linux 子系统。","costNote":"这个服务器是免费的(Apache 2.0),但只有配上你自己付费的模型密钥才能用;README 也把本地 Ollama 模型列为免费选项。","notVerified":["日常工作中调用 Gemini、OpenAI 或 Grok 的费用大概是多少","run-server.ps1 是否真的能在 Windows 上不借助 Linux 子系统完成全部配置","conf/ 里的模型名称在近一年未更新的情况下,是否仍在各家服务商那边存在","本地 Ollama 选项的效果和付费模型相比如何","在实际项目中,每个工具到底把什么内容发给了每个服务商"]}}},{"id":"R_kgDOQ8AnQQ","slug":"blader__humanizer","name":"Humanizer","owner":"blader","fullName":"blader/humanizer","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"attention","signals":[{"code":"hides_that_a_text_was_written_by_ai","severity":"attention","evidence":"该仓库把自己描述为一个 \"Agent skill that removes signs of AI-generated writing from text\"，标签包括 `ai-humanizer` 和 `humanize-ai-text`。把 AI 写的文字当作自己写的，违反了大多数学校、大学以及许多出版商的规定。README 确实说过 \"Getting past AI detectors is not a goal, and detectors still flag most of its output.\"","source":"ai","title":"掩盖文本是由 AI 写成的事实"},{"code":"rewrites_your_files_in_place","severity":"attention","evidence":"README 的 Usage 部分写道：\"Humanize the prose in docs/launch-post.md\" —— 你给出一个路径，助手就会编辑那个文件。README 说它 \"changes only the prose, leaving code, data, frontmatter, and link targets alone\"，这只是作者自己对规则的描述，不是保证。请保留一份副本或使用版本控制。","source":"ai","title":"直接修改你的文件"},{"code":"your_text_goes_to_the_ai_company","severity":"info","evidence":"这个技能本身只是一堆指令；真正的改写是由你安装它所在的那个助手完成的（Claude Code、Codex、Claude.ai、Copilot 等），所以你粘贴的文字，包括为匹配语气而添加的任何写作样本，都会被发送给那家公司。","source":"ai","title":"你的文字会被发送给 AI 公司"},{"code":"instructions_become_assistant_prompts","severity":"info","evidence":"该项目本质上是一个 `SKILL.md` 加上一些辅助文件，会作为指令提供给你的助手。你安装那天仓库里写的内容，就是你的助手会遵循的内容，之后的更新会改变这种行为。","source":"ai","title":"这些指令会变成助手的提示词"},{"code":"readme_is_sponsored","severity":"info","evidence":"README.md 的第二行写着 \"sponsored by [cfo.ai](https://cfo.ai)\"，所以这个页面除了是文档之外，也是一块付费推广空间。","source":"ai","title":"README 中含有付费推广内容"},{"code":"quality_claim_is_self_reported","severity":"info","evidence":"README 写道：\"In a blind test, judges preferred Humanizer's rewrite over the original AI text 16 times out of 16\"，链接到项目自己的 issue #229。没有项目之外的人核实过这项测试。","source":"ai","title":"质量方面的说法是自我宣称的"}],"platforms":["windows","mac","linux","web"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","office","productivity"],"alternativeTo":[{"slug":"quillbot","name":"QuillBot","confirmed":true},{"slug":"grammarly","name":"Grammarly","confirmed":true},{"slug":"undetectable-ai","name":"Undetectable AI","confirmed":true}],"declaredAlternatives":[],"stars":67517,"downloads":5342,"downloadsWeek":0,"downloadsDays":7,"growth":3117,"growthDays":7,"history":[{"date":"2026-11-03","stars":64400},{"date":"2026-11-04","stars":64799},{"date":"2026-11-05","stars":65219},{"date":"2026-11-06","stars":65601},{"date":"2026-11-07","stars":65957},{"date":"2026-11-08","stars":66416},{"date":"2026-11-09","stars":67051},{"date":"2026-11-10","stars":67517}],"firstPublishedAt":"2026-11-03T02:51:37.305Z","commitSha":"0ef9fe5bc3572734fd3ae6c448fe00a30cd5dcf0","verifiedAt":"2026-11-03T02:42:08.854Z","lastCommitAt":"2026-11-10T01:34:24Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An instruction pack for AI assistants that rewrites AI-sounding text so it reads like a person wrote it.","summary":"A set of written instructions you add to an AI assistant (Claude Code, Codex, Claude.ai, Copilot and others). You paste text or point at a file and it rewrites what gives away AI writing, following 26 patterns from Wikipedia's \"Signs of AI writing\", while keeping the facts. You can paste a sample of your own writing so the result sounds like you. The README says beating AI detectors is not a goal and that detectors still flag most of its output. There is no program to open: it works inside an assistant you already pay for.","easeRationale":"The easiest route is Claude.ai in the browser: download the repository as a ZIP from the green Code button and upload it as a skill in Settings, with nothing to install. On Windows, macOS and Linux you instead work in a terminal-style assistant: two slash commands inside Claude Code 2.1.142 or newer, or `npx skills add blader/humanizer --global --agent codex`, which needs Node.","costNote":"The instructions are free (MIT). They only do something inside an AI assistant, and the ones listed (Claude Code, Codex, Claude.ai, Copilot) are paid or have paid plans. Each rewrite uses that assistant's allowance.","notVerified":["Whether the rewrite keeps the meaning of your text, since the README's own rule is that no fact may be invented","Whether it installs and answers as described in each of the listed assistants","How much of your assistant's monthly allowance a long rewrite uses","Whether AI detectors flag the result, which the README says they usually still do","What the files in the `scripts/`, `agents/` and `.claude-plugin/` folders do when the plugin is installed"]},"zh-Hans":{"summaryShort":"一套给 AI 助手用的指令集，把带有 AI 痕迹的文字改写成像人写的一样。","summary":"这是一套写好的指令，你把它添加到 AI 助手里使用（Claude Code、Codex、Claude.ai、Copilot 等）。你粘贴文本或指向一个文件，它会按照维基百科 \"Signs of AI writing\" 中的 26 种模式，改写那些暴露 AI 写作痕迹的地方，同时保留事实内容。你也可以粘贴一段自己的文字样本，让结果听起来更像你本人的风格。README 说，打败 AI 检测工具并不是目标，而且检测工具仍然会标记它大部分的输出结果。没有程序需要打开：它运行在你已经付费使用的助手内部。","easeRationale":"最简单的方式是在浏览器里使用 Claude.ai：从绿色的 Code 按钮下载仓库的 ZIP 文件，然后在 Settings 里把它作为技能上传，不需要安装任何东西。在 Windows、macOS 和 Linux 上，你需要在一个终端风格的助手里操作：在 Claude Code 2.1.142 或更新版本中使用两个斜线命令，或者运行 `npx skills add blader/humanizer --global --agent codex`，这需要 Node。","costNote":"这套指令是免费的（MIT 许可）。它只能在 AI 助手内部发挥作用，而列出的助手（Claude Code、Codex、Claude.ai、Copilot）都是付费的或有付费方案。每次改写都会消耗该助手的额度。","notVerified":["改写是否保留了原文的意思，因为 README 自己的规则就是不能编造任何事实","它是否能在每一个列出的助手里按描述那样安装并响应","一次长篇改写会消耗你助手月度额度的多少","AI 检测工具是否会标记改写结果，README 说它们通常还是会这样做","插件安装后，`scripts/`、`agents/` 和 `.claude-plugin/` 文件夹里的文件都做了什么"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMTM5MTY4OTY=","slug":"blender__blender","name":"Blender","owner":"blender","fullName":"blender/blender","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"github_repo_is_only_a_mirror","severity":"info","evidence":"仓库描述：\"Official mirror of Blender\"，主页指向 developer.blender.org 上的 GitHub 镜像说明页，GitHub 上的发布版本为 0 个 — 程序本身要从 blender.org 下载，开发工作在 projects.blender.org 上进行。","source":"ai","title":"GitHub 仓库只是镜像"},{"code":"heavy_program","severity":"info","evidence":"安装大约占用 349 MB 到 1 GB 的磁盘，README 介绍了整套 3D 流程，包括渲染和模拟，这些对处理器和显卡的要求很高。","source":"ai","title":"程序体量大"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python"],"disk":{"minMB":349,"maxMB":1048,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0-or-later","category":"copyleft","detectedFrom":"winget"},"categories":["design_cad","video","image"],"alternativeTo":[{"slug":"autodesk-maya","name":"Maya","confirmed":true},{"slug":"autodesk-3ds-max","name":"3ds Max","confirmed":true},{"slug":"cinema-4d","name":"Cinema 4D","confirmed":true},{"slug":"after-effects","name":"After Effects","confirmed":false}],"declaredAlternatives":[],"stars":21846,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":242,"growthDays":7,"history":[{"date":"2026-10-25","stars":21260},{"date":"2026-10-26","stars":21305},{"date":"2026-10-27","stars":21349},{"date":"2026-10-28","stars":21394},{"date":"2026-10-29","stars":21420},{"date":"2026-10-30","stars":21447},{"date":"2026-10-31","stars":21485},{"date":"2026-11-01","stars":21510},{"date":"2026-11-02","stars":21553},{"date":"2026-11-03","stars":21604},{"date":"2026-11-04","stars":21637},{"date":"2026-11-05","stars":21666},{"date":"2026-11-06","stars":21691},{"date":"2026-11-07","stars":21729},{"date":"2026-11-08","stars":21767},{"date":"2026-11-09","stars":21812},{"date":"2026-11-10","stars":21846}],"firstPublishedAt":"2026-10-25T23:32:48.676Z","commitSha":"4c6a9739e8717a4dce80c0a793142ebc84fadb14","verifiedAt":"2026-11-02T02:51:52.630Z","lastCommitAt":"2026-11-10T21:41:37Z","isMirror":true,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Free 3D program for modelling, animating, rendering and editing video, with a ready installer.","summary":"Blender is a complete 3D program: you can model objects and characters, rig and animate them, run simulations, render finished images and film, and it also includes compositing, motion tracking and a video editor. It is used by hobbyists, studios and teachers, and is free with no account or licence. You download an installer for Windows, macOS or Linux from blender.org and open it. A modern graphics card makes rendering much faster, but is not required to start. The programming language inside it is Python, which comes with the app.","easeRationale":"On Windows there is a ready installer (also in winget) that you download and open, and the official site publishes the same kind of ready build for macOS and Linux. Nothing else has to be installed first; Python comes inside the app.","notVerified":["How well it performs on your own computer and graphics card, especially for rendering and simulation.","Whether a specific add-on you need works with the version you install.","Actual disk use after caches, add-ons and your own project files.","Whether the download from blender.org matches the code in this mirror."]},"zh-Hans":{"summaryShort":"免费的 3D 软件，可建模、做动画、渲染和剪辑视频，有现成的安装包。","summary":"Blender 是一款完整的 3D 软件：你可以给物体和角色建模、绑定骨骼并制作动画、运行模拟、渲染成品图像和影片，它还内置合成、运动跟踪和视频编辑器。业余爱好者、工作室和教师都在用，它免费，无需账号或许可证。你从 blender.org 下载适用于 Windows、macOS 或 Linux 的安装包，打开即可。较新的显卡能让渲染快很多，但并不是入门的必要条件。它内置的编程语言是 Python，随软件一起提供。","easeRationale":"在 Windows 上有现成的安装包（winget 里也有），下载后打开即可；官网也为 macOS 和 Linux 提供同样的现成版本。不用先装别的东西，Python 就在软件里面。","notVerified":["它在你自己的电脑和显卡上表现如何，尤其是渲染和模拟时。","你需要的某个插件能否在你安装的版本上使用。","加上缓存、插件和你自己的项目文件之后，实际的磁盘占用。","从 blender.org 下载的程序是否与这个镜像里的代码一致。"]}}},{"id":"R_kgDOTs_xqg","slug":"boykopovar__anyps5","name":"AnyPS5","owner":"boykopovar","fullName":"boykopovar/AnyPS5","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","linux":"C"},"risk":"high","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"90 天内获得 50474 个星标（约每天 561 个）","source":"check","title":"星标增长速度异常"},{"code":"needs_ps5_game_files_you_must_obtain","severity":"attention","evidence":"README，Disclaimer：\"It does not include, distribute, or require copyrighted software, firmware, cryptographic keys, or proprietary libraries. Users are responsible for ensuring that any binaries used with this project are obtained and used in accordance with applicable laws and their respective license terms.\" PS5 游戏并不以电脑文件的形式出售，所以程序只能从保护它们的主机里取出，而绕过这种保护在许多国家是违法的。","source":"ai","title":"需要自己获取 PS5 游戏文件"},{"code":"very_early_version","severity":"attention","evidence":"最新版本是 v0.1.4，有 2153 个未解决的 issue，README 写道 \"Unsupported or unexpected states throw a C++ exception... An exception that nothing catches prints its what() to stderr and ends the process\"，各游戏的运行情况只记录在 GitHub 讨论区里。","source":"ai","title":"版本非常早期"},{"code":"not_affiliated_with_sony","severity":"info","evidence":"README：\"AnyPS5 is an independent project and is not affiliated with, authorized or endorsed by Sony Interactive Entertainment.\"","source":"ai","title":"与索尼无关联"},{"code":"reimplements_console_system_libraries","severity":"info","evidence":"README：\"Includes a relinker that converts executable to the target system's native format and implementations of system prx libraries suitable for dynamic linking\"，而项目自己的进度页统计的是 \"percentage of the functions known to the project so far... not of every PS5 system function\"。","source":"ai","title":"重新实现了主机的系统库"},{"code":"needs_ps5_game_files_from_a_protected_console","severity":"high","evidence":"简介：\"Tool for automatic PS5 executables porting to Linux and Windows\"；README，Disclaimer：\"It does not include, distribute, or require copyrighted software, firmware, cryptographic keys, or proprietary libraries. Users are responsible for ensuring that any binaries used with this project are obtained and used in accordance with applicable laws and their respective license terms.\" PS5 游戏只为主机出售，主机会对其加密，而许多国家都有法律限制绕过复制保护；README 还称该项目 \"is not affiliated with, authorized or endorsed by Sony Interactive Entertainment\"。","source":"ai","title":"需要从受保护的主机里取得 PS5 游戏文件"}],"platforms":["windows","linux"],"tools":[],"buildTools":[],"disk":{"minMB":5,"maxMB":16,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":52949,"downloads":210545,"downloadsWeek":16976,"downloadsDays":7,"growth":2011,"growthDays":7,"history":[{"date":"2026-10-27","stars":49027},{"date":"2026-10-28","stars":49312},{"date":"2026-10-29","stars":49573},{"date":"2026-10-30","stars":49853},{"date":"2026-10-31","stars":50145},{"date":"2026-11-01","stars":50472},{"date":"2026-11-02","stars":50715},{"date":"2026-11-03","stars":50938},{"date":"2026-11-04","stars":51143},{"date":"2026-11-05","stars":51452},{"date":"2026-11-06","stars":51747},{"date":"2026-11-07","stars":52014},{"date":"2026-11-08","stars":52284},{"date":"2026-11-09","stars":52586},{"date":"2026-11-10","stars":52949}],"firstPublishedAt":"2026-10-27T02:46:41.232Z","commitSha":"93dbd0f82aea684e95f15f06e7d42c0064af18cf","verifiedAt":"2026-11-02T02:52:56.400Z","lastCommitAt":"2026-11-10T22:06:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Converts PS5 game programs so they run on a Windows or Linux PC, without an emulator.","summary":"AnyPS5 takes a program built for the PlayStation 5 and rewrites it into a Windows or Linux program, together with its own versions of the console's system libraries, so it runs directly instead of inside an emulator. You download the converter and the library pack, convert the file (there is also a converter page in the browser) and run the result; game controllers work through SDL and keyboard controls come from a text file. The project does not give you any game: you have to supply the PS5 files yourself. It is at version 0.1.4 and only some titles work.","easeRationale":"On Windows you download a ready relinker.exe plus a 91 MB library pack, unpack it and convert each program from the command line before running it, and you have to get the PS5 files yourself. Linux is the same work with .tar.gz or .zip files and no installer. On macOS only the converter part is supported and you have to compile it yourself.","notVerified":["Whether a game you own actually converts and runs, and how well it plays.","Whether the browser converter really keeps files on your computer, as the README says.","What graphics card and drivers you need in practice, since the project renders through Vulkan.","Whether the files in Releases were built from the published source.","The legal position where you live of taking programs off a PlayStation 5."]},"zh-Hans":{"summaryShort":"把 PS5 游戏程序转换后在 Windows 或 Linux 电脑上运行，无需模拟器。","summary":"AnyPS5 会把为 PlayStation 5 制作的程序改写成 Windows 或 Linux 程序，并附带它自己实现的主机系统库，使程序直接运行，而不是在模拟器里运行。你下载转换器和库包，转换文件（浏览器里也有一个转换页面），再运行结果；游戏手柄通过 SDL 工作，键盘操作由一个文本文件设定。项目本身不提供任何游戏：PS5 文件得由你自己准备。目前是 0.1.4 版，只有部分游戏能运行。","easeRationale":"在 Windows 上，你要下载现成的 relinker.exe 和一个 91 MB 的库包，解压后先用命令行转换每个程序再运行，而且 PS5 文件得自己想办法获得。Linux 的工作量相同，用的是 .tar.gz 或 .zip 文件，没有安装程序。在 macOS 上只支持转换器这一部分，而且得自己编译。","notVerified":["你拥有的游戏能否真的转换并运行，以及玩起来效果如何。","浏览器里的转换器是否真的像 README 所说，把文件留在你的电脑上。","实际需要什么显卡和驱动，因为项目是通过 Vulkan 渲染的。","Releases 里的文件是否由已发布的源码编译而来。","在你所在的地方，从 PlayStation 5 里提取程序的法律地位。"]}}},{"id":"R_kgDOT-bMvA","slug":"brightbeanxyz__brightbean-chat","name":"BrightBean Chat","owner":"brightbeanxyz","fullName":"brightbeanxyz/brightbean-chat","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"meta_channels_need_app_review_first","severity":"attention","evidence":"README 的“Meta 需要批准的内容”部分：一个新应用只有标准权限，“只能触达在该应用里拥有角色的用户的账号”；要用于别人的账号，就意味着每项权限都要经过应用审核，“需要一段从头到尾展示每项权限实际使用情况的录屏”，还要做企业验证，而且“要按周而不是按天来估算时间，并且预期至少会被拒一次”。Instagram 还要求把账号添加为 Instagram Tester。Telegram、短信和邮件不需要这一步。","source":"ai","title":"meta 的渠道需要先通过应用审核"},{"code":"no_releases_and_upgrades_are_rebuilds","severity":"info","evidence":"这个仓库创建于 2026-08-20，GitHub 发行版是 0 个，有 28 个未关闭的 issue。docker-compose.prod.yml 声明“这个仓库里没有任何工作流会把镜像推送到任何地方”，而且“docs/self-hosting.md 里的升级方式因此就是 git pull 加重新构建”，所以没有版本号可以固定，也没有镜像可以拉取。","source":"ai","title":"没有发行版，升级就是重新构建"},{"code":"you_hold_other_people_s_private_messages","severity":"attention","evidence":"这个产品把联系人、按线程组织的对话、自定义字段和上传的媒体文件都存在你自己的 PostgreSQL 里，平台凭据则用你自己生成的 ENCRYPTION_KEY_SALT 加密。docker-compose.prod.yml 指出“Postgres 保存着联系人和加密后的平台凭据”。备份、这个密钥丢失后的后果，以及这些数据的法律责任，都落在运行这台服务器的人身上。","source":"ai","title":"你保存着其他人的私信"},{"code":"bulk_messaging_is_regulated","severity":"info","evidence":"README 自己的渠道表格列出了这个工具必须遵守的规则：Instagram、Messenger 和 WhatsApp 上有 24 小时的消息窗口，窗口之外要用已批准的模板，要处理 STOP/HELP/START，美国的短信号码要完成 A2P 10DLC 注册，还要有 SPF/DKIM/DMARC 记录，因为“Gmail 和 Yahoo 会拒收没有这些记录的域名发出的群发邮件”。","source":"ai","title":"群发消息受到监管"},{"code":"payment_variable_in_the_environment_reference","severity":"info","evidence":"检查发现应用能识别的变量里有 STRIPE_WEBHOOK_SECRET（以及 TICK_TOKEN），而 README 说自托管产品“没有需要注册的支付服务商”。这个仓库同时也是作者付费托管版本运行的代码，这很可能就是这个变量出现的原因；自托管并不需要它。","source":"ai","title":"环境变量参考里出现了一个支付变量"}],"platforms":["windows","mac","linux","web"],"tools":["docker"],"buildTools":["python","node","docker"],"disk":{"minMB":200,"maxMB":800,"basis":"heuristic"},"apiKeys":["A token for at least one channel — a Telegram bot token from BotFather is the quickest","Meta developer app (client id, secret and verify token) for Instagram, Messenger and WhatsApp","Twilio (account SID and auth token) for the SMS channel","SMTP, Resend or Amazon SES for the email channel and for the sign-up and password e-mails","Google sign-in (GOOGLE_AUTH_CLIENT_ID and GOOGLE_AUTH_CLIENT_SECRET), optional login button"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[{"target":"manychat","stance":"claims","evidence":"<strong>The open-source ManyChat alternative"}],"stars":86,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":4,"growthDays":6,"history":[{"date":"2026-11-04","stars":82},{"date":"2026-11-05","stars":82},{"date":"2026-11-06","stars":83},{"date":"2026-11-07","stars":83},{"date":"2026-11-08","stars":83},{"date":"2026-11-09","stars":85},{"date":"2026-11-10","stars":86}],"firstPublishedAt":"2026-11-04T01:41:48.258Z","commitSha":"7bf959b30cc7f345f84d2be59e68a0a05c327e4f","verifiedAt":"2026-11-04T01:24:28.698Z","lastCommitAt":"2026-10-21T07:59:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Builds chat flows that answer your Telegram, Instagram, WhatsApp, Messenger, SMS and email messages by themselves.","summary":"BrightBean Chat is a chat-marketing tool you host yourself. You draw a flow in a visual builder (keyword replies, comment-to-DM, delays, conditions, drip sequences, broadcasts) and run it on Telegram, Instagram, Messenger, WhatsApp, SMS and email, with a shared inbox where a person can take over a conversation. Installing it means Docker, PostgreSQL and a public web address, plus a developer app of your own at Meta for the Instagram, Messenger and WhatsApp channels. The code is AGPL-3.0; the authors also run a hosted version.","easeRationale":"The easiest route is the authors' hosted version at brightbean.xyz/chat: you sign up and connect a channel, with nothing to install, so web is a B. Self-hosting is terminal work on all three systems and there is no release to download: you install Docker, clone the repository and run `docker compose up` for a local try, or copy `deploy/env.prod.example`, generate secrets with `make prod-secrets` and start a five-service production stack (PostgreSQL, migration, Gunicorn, a worker and Caddy) on a server with a public domain name.","costNote":"The code is AGPL-3.0 with, in the README's words, \"Unlimited contacts, every feature\". The authors' hosted version at brightbean.xyz/chat is described as \"free for your first 25 contacts a month\". The channels can still cost money at their providers: a Twilio number for SMS, an email provider, and a WhatsApp number metered by Meta.","notVerified":["Whether the hosted version at brightbean.xyz/chat really runs this same code, and what its paid plans cost beyond 25 contacts a month","Whether the security behaviour the README promises (tenant isolation, SSRF guard, encrypted credentials, webhook signature checks) holds in practice, which would need an audit","Whether each channel works end to end, which can only be seen with real Meta, Twilio and e-mail accounts","How the stack behaves on the 1 vCPU / 1 GB host the documentation names as workable, and how much disk the database and media grow to","Whether a brand-new project three months old keeps being maintained"]},"zh-Hans":{"summaryShort":"搭建聊天流程，自动回复你 Telegram、Instagram、WhatsApp、Messenger、短信和邮件里的消息。","summary":"BrightBean Chat 是一个你自己托管的聊天营销工具。你在一个可视化构建器里设计一个流程（关键词回复、评论转私信、延迟、条件分支、滴灌式序列、群发），然后在 Telegram、Instagram、Messenger、WhatsApp、短信和邮件上运行它，还有一个共享收件箱，让人工随时可以接管对话。安装它需要 Docker、PostgreSQL 和一个公开网址，还需要你自己在 Meta 上为 Instagram、Messenger 和 WhatsApp 渠道创建一个开发者应用。代码是 AGPL-3.0 许可；作者也运营着一个托管版本。","easeRationale":"最简单的方式是使用作者在 brightbean.xyz/chat 的托管版本：注册并连接一个渠道，不需要安装任何东西，所以网页版评为 B。自托管在三个系统上都是终端操作，而且没有发行版可以下载：你需要安装 Docker，克隆仓库，运行 `docker compose up` 做本地试用，或者复制 `deploy/env.prod.example`，用 `make prod-secrets` 生成密钥，再在一台有公开域名的服务器上启动一套包含五个服务的生产环境（PostgreSQL、迁移、Gunicorn、一个 worker 和 Caddy）。","costNote":"代码是 AGPL-3.0 许可，用 README 的话说，“联系人数量不限，所有功能都有”。作者在 brightbean.xyz/chat 的托管版本被描述为“每月前 25 个联系人免费”。各个渠道在它们自己的提供方那里仍然可能要花钱：短信要用 Twilio 号码，邮件需要邮件服务提供方，WhatsApp 号码则由 Meta 按量计费。","notVerified":["brightbean.xyz/chat 上的托管版本是否真的运行着这同一份代码，超过每月 25 个联系人之后的付费方案要多少钱","README 承诺的这些防护机制（租户隔离、SSRF 防护、加密的凭据、webhook 签名校验）在实际中是否真的成立，这需要一次审计才能确认","每个渠道是否能完整走通，这只能用真实的 Meta、Twilio 和邮件账号才能看到","在文档里说可行的 1 vCPU / 1 GB 主机上，这套系统表现如何，数据库和媒体文件会把磁盘占用撑大到什么程度","一个才三个月的新项目是否能持续得到维护"]}}},{"id":"R_kgDOR8sA6Q","slug":"brodyautomates__ig-setter","name":"IG Setter","owner":"brodyautomates","fullName":"brodyautomates/ig-setter","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"web":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖存在已知漏洞(OSV):next: GHSA-2464-fhr4-w9xw, GHSA-2xp9-vwfh-vxw4, GHSA-36qx-fr4f-26g5","source":"check","title":"存在已知漏洞"},{"code":"setup_guide_is_missing_from_the_repository","severity":"attention","evidence":"README 两次指向 \"docs/SETUP.md\" 作为\"Full instructions\",仓库结构里也写了一个 docs/ 文件夹,但顶层根本没有 docs/——只有一个叫 \"set up guide/\" 的文件夹。","source":"ai","title":"仓库里缺少安装指南"},{"code":"asks_for_the_database_master_key","severity":"attention","evidence":".env.example 在公开密钥旁边还要你填 \"SUPABASE_SERVICE_ROLE_KEY=your-service-role-key\";这个密钥能绕过数据库的访问规则,却会出现在发布到 Vercel 网站的环境变量里。","source":"ai","title":"索要数据库主密钥"},{"code":"browser_automation_library_not_explained","severity":"info","evidence":"package.json 的 devDependencies 里有 \"puppeteer\": \"^24.40.0\"(一个控制浏览器的工具),但 README 说 Instagram 消息走的是 \"Meta Graph API v19.0\",描述的结构里完全没提到要用浏览器。","source":"ai","title":"浏览器自动化库没有说明用途"},{"code":"replies_are_written_by_ai_without_telling_the_other_person","severity":"info","evidence":"README:\"Claude AI auto-drafts replies on every inbound DM\",还有 \"Thread status tracking: Active → Qualified → Booked → Closed\" 以及每日的 \"deals closed, revenue\"——给你账号发消息的人可能收到的是 AI 的回复,而 Instagram 自己的规则也限制自动化私信。","source":"ai","title":"ai 代写回复,却不告知对方"},{"code":"small_and_untouched_project","severity":"info","evidence":"创建和最后一次 push 都在 2026-04-08,距检查时 211 天,只有 28 个星标、10 个 fork,没有发布版本也没有许可证文件;依赖的是 Next.js 14.2.5,正是检查标记出已知漏洞的版本。","source":"ai","title":"项目规模小且长期无变化"}],"platforms":["web"],"tools":["other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Supabase (including the service role key)","Meta / Instagram Graph API access token","Anthropic (Claude), used from n8n","n8n, self-hosted or cloud"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","ai_agents","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":28,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":28},{"date":"2026-11-07","stars":28},{"date":"2026-11-08","stars":28},{"date":"2026-11-09","stars":28},{"date":"2026-11-10","stars":28}],"firstPublishedAt":"2026-11-05T22:43:48.288Z","commitSha":"1a44f4312eb01f1ba4ec8ed81b37e9a33edcefa1","verifiedAt":"2026-11-05T22:32:06.654Z","lastCommitAt":"2026-04-08T04:18:03Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A dashboard where Claude answers your Instagram direct messages by itself and you can step in to reply by hand.","summary":"ig-setter is a web dashboard for whoever sells through Instagram direct messages. Each message that arrives goes to an n8n workflow, Claude writes the reply, and you watch the conversation live and can send your own answer instead. It also tracks each contact as Active, Qualified, Booked or Closed and shows daily numbers. It is not a program you install: you create accounts on Supabase, n8n and the Meta developer portal, run an SQL file, import the workflow and publish the dashboard on Vercel with your keys.","easeRationale":"There is nothing to install on your computer — you use it in the browser — but getting there is a D: create a Supabase project and run an SQL migration, create an app in the Meta developer portal and set up the webhook, import a workflow into n8n, then click the Vercel button and fill in nine environment variables. The README also sends you to Meta's App Review for production access.","costNote":"The code is free. The README's own table: Supabase, Vercel and the Meta app free, n8n \"Free (self-host) or $20/mo (cloud)\", Claude API \"~$0.001/DM\".","notVerified":["Whether Meta approves the app for replying to other people's messages, which needs App Review","What is inside the \"set up guide/\" folder and whether it replaces the missing docs/SETUP.md","The real monthly cost once the message volume grows, beyond the README's ~$0.001 per message","Whether the n8n workflow in n8n/ig-setter-workflow.json still matches the current Meta API","What the Claude prompt tells the AI to say to your contacts"]},"zh-Hans":{"summaryShort":"一个仪表盘,Claude 自动回复你 Instagram 的私信,你也可以随时介入手动回复。","summary":"ig-setter 是为通过 Instagram 私信卖货的人做的网页仪表盘。每条收到的消息都会进入一个 n8n 工作流,由 Claude 写回复,你可以实时看到对话,并随时自己发消息代替它。它还会把每个联系人标记为 Active、Qualified、Booked 或 Closed,并显示每日数据。它不是一个可以直接安装的程序:你需要在 Supabase、n8n 和 Meta 开发者后台注册账号,运行一个 SQL 文件,导入工作流,再用自己的密钥把仪表盘发布到 Vercel。","easeRationale":"电脑上没有什么要安装的——全部在浏览器里用——但到达这一步是 D 级:要在 Supabase 建项目并运行一次 SQL 迁移,在 Meta 开发者后台建应用并配置 webhook,把一个工作流导入 n8n,然后点击 Vercel 的按钮并填好九个环境变量。README 还会让你去申请 Meta 的 App Review 才能用于生产环境。","costNote":"代码是免费的。README 自己的表格写着:Supabase、Vercel 和 Meta 应用免费,n8n \"Free (self-host) or $20/mo (cloud)\",Claude API \"~$0.001/DM\"。","notVerified":["Meta 是否会批准这个应用去回复别人的消息,这需要通过 App Review","\"set up guide/\" 文件夹里到底是什么,它能不能替代缺失的 docs/SETUP.md","消息量变大之后每月的真实花费,而不只是 README 里说的每条消息约 $0.001","n8n/ig-setter-workflow.json 里的工作流是否还跟 Meta 目前的 API 匹配","Claude 的提示词到底让 AI 跟你的联系人说了什么"]}}},{"id":"R_kgDONIn9tw","slug":"browser-use__browser-use","name":"Browser Use","owner":"browser-use","fullName":"browser-use/browser-use","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"sold_with_stealth_and_captcha_solving","severity":"attention","evidence":"README 第一行就宣传“stealth browsers from $0.02/hour, top score (90.5%) on Stealth Bench V2”，主打演示是“Find an available slot, pick a date and time, handle the CAPTCHA, and book a driving test”，FAQ 里说云端提供“stealth browsers and proxies designed to reduce bot detection and CAPTCHA challenges”。CAPTCHA 和机器人检测正是网站用来拦截自动化访问的手段，很多网站的条款禁止这样做；这个项目把这件事留给用户自己处理。","source":"ai","title":"搭配隐身和 captcha 破解功能出售"},{"code":"telemetry_and_version_check_on_by_default","severity":"attention","evidence":".env.example 里自带 \"ANONYMIZED_TELEMETRY=true\" 和 \"BROWSER_USE_VERSION_CHECK=true\"，pyproject.toml 把 posthog==7.7.0 列为必需依赖。使用数据会发送到项目自己的统计系统，智能体启动时这个库也会联网，除非你把两者都关掉。","source":"ai","title":"默认开启遥测和版本检查"},{"code":"agent_can_use_your_signed_in_browser_and_saved_runs_hold_secrets","severity":"attention","evidence":"FAQ 认证部分：“Local browser: Use Browser.from_system_chrome() to reuse a Chrome profile”，云端浏览器则有一种配置同步功能会“transfers cookies”。回放（replay）相关的 FAQ 警告：“Keep the saved file private. With sensitive_data, typed secrets are saved as <secret>key</secret> placeholders, but page snapshots and results are saved as they were and can still contain them.”也就是说，智能体可以在你已经登录的账号内操作，它保存下来的记录可能含有个人数据。","source":"ai","title":"智能体可使用你已登录的浏览器，保存的记录可能含有机密"},{"code":"open_library_in_front_of_a_paid_service","severity":"info","evidence":"项目描述是“Open-source Python library for AI browser agents, plus Browser Use Cloud: stealth browsers and a hosted agent API”，pyproject.toml 把这个包描述为“Open-source browser agent + hosted cloud browsers for AI agents, from $0.02/browser-hour”，并要求依赖 browser-use-sdk 和 browser-harness，FAQ 还推荐公司自己的 BU2 模型，需要用他们自己的密钥。这个库本身是 MIT 许可，但推荐的使用路径要经过一个有自己条款和隐私政策的付费服务。","source":"ai","title":"开源库前端对接的是付费服务"},{"code":"readme_contains_instructions_addressed_to_ai_assistants","severity":"info","evidence":"README 里有一段以“AI agents and crawlers: read browser-use.com/llms.txt”开头的内容，还有一个“Path 2: CLI”部分，整段内容就是一条要粘贴进 Claude Code 或其他助手里的提示语，让它安装这个包并“run `browser-use skill install` to register the skill”。运行它就等于把整个安装过程，包括一份外部安装指南，交给了一个 AI 助手去完成。","source":"ai","title":"readme 中包含专门写给 ai 助手的指令"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","ffmpeg"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI (OPENAI_API_KEY) or another model provider — Anthropic, Google, Groq, DeepSeek, or a local Ollama model instead","Browser Use Cloud (BROWSER_USE_API_KEY) — only for the BU2 model, cloud browsers, proxies and CAPTCHA handling"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","automation","development"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"bright-data","name":"Bright Data","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":120584,"downloads":2844,"downloadsWeek":0,"downloadsDays":7,"growth":624,"growthDays":7,"history":[{"date":"2026-10-26","stars":119223},{"date":"2026-10-27","stars":119317},{"date":"2026-10-28","stars":119401},{"date":"2026-10-29","stars":119501},{"date":"2026-10-30","stars":119600},{"date":"2026-10-31","stars":119679},{"date":"2026-11-01","stars":119776},{"date":"2026-11-02","stars":119872},{"date":"2026-11-03","stars":119960},{"date":"2026-11-04","stars":120032},{"date":"2026-11-05","stars":120123},{"date":"2026-11-06","stars":120205},{"date":"2026-11-07","stars":120260},{"date":"2026-11-08","stars":120339},{"date":"2026-11-09","stars":120442},{"date":"2026-11-10","stars":120584}],"firstPublishedAt":"2026-10-27T01:10:12.610Z","commitSha":"b49a68e4060acb226da8f784194fef301c53a6d7","verifiedAt":"2026-11-01T01:44:00.727Z","lastCommitAt":"2026-11-10T09:26:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Python library that lets an AI model drive a web browser for you, clicking, typing and filling forms on real sites.","summary":"Browser Use is a Python library (plus a command-line tool) that hands a web browser over to an AI model: you write the task in a sentence, and the agent opens pages, clicks, types, reads results and reports back. It is for developers who build automations; using it means installing Python, writing a small script and paying an AI provider per task. The same code can drive the browser on your own machine or Browser Use's paid cloud browsers, which the README prices at $0.02 per browser-hour and sells with stealth, residential proxies and CAPTCHA solving.","easeRationale":"The same on all three systems: install Python 3.11+ (the README uses the `uv` tool), run `uv add browser-use`, put a model key in a .env file, write a short Python file and run it; a local Chrome is needed unless you pay for a cloud browser. The project also ships a Dockerfile as an alternative path, and ffmpeg only arrives with the optional `video` extra. On Windows the extra Claude \"Bash\" integration needs WSL, which the README states.","costNote":"The library is free and MIT-licensed, but every task calls a paid AI model (the README's own example uses an OpenAI key, and the recommended BU2 model needs a Browser Use key). Cloud browsers are $0.02 per browser-hour plus traffic. A local model through Ollama is possible with no per-use cost.","notVerified":["How much a real task costs in model calls, which depends on the model and the number of steps","Whether the agent actually finishes a given task on a given site, and how often it gets stuck","Whether the stealth and CAPTCHA features work on the sites you care about, which the README itself does not guarantee","What the telemetry contains and what Browser Use keeps from cloud runs","Disk use, estimated at 100 to 300 MB by a rule of thumb, before Chrome and model downloads","Whether it behaves the same on Windows, where some parts need WSL"]},"zh-Hans":{"summaryShort":"一个 Python 库，让 AI 模型替你操作网页浏览器：点击、输入并在真实网站上填表单。","summary":"Browser Use 是一个 Python 库（外加一个命令行工具），把网页浏览器交给 AI 模型操作：你用一句话写出任务，智能体就会打开页面、点击、输入、读取结果并向你汇报。它适合构建自动化流程的开发者：使用它意味着要安装 Python、写一小段脚本，并按任务向 AI 供应商付费。同一套代码既可以在你自己的电脑上操作浏览器，也可以使用 Browser Use 的付费云端浏览器，README 给出的价格是每浏览器小时 0.02 美元，还附带隐身（stealth）、住宅代理和 CAPTCHA 破解功能。","easeRationale":"三个系统待遇相同：安装 Python 3.11+（README 使用 `uv` 工具），运行 `uv add browser-use`，把模型密钥放进 .env 文件，写一个简短的 Python 文件并运行；除非你付费用云端浏览器，否则需要本地的 Chrome。项目还提供了一个 Dockerfile 作为备选方案，ffmpeg 只有在安装可选的 `video` 扩展时才会出现。在 Windows 上，额外的 Claude “Bash”集成需要 WSL，这是 README 里写明的。","costNote":"这个库是免费的，采用 MIT 许可证，但每个任务都会调用付费的 AI 模型（README 自己的示例用的是 OpenAI 密钥，推荐的 BU2 模型需要 Browser Use 的密钥）。云端浏览器每浏览器小时收费 0.02 美元，另加流量费用。通过 Ollama 使用本地模型则不产生按次使用的费用。","notVerified":["一个真实任务在模型调用上实际花多少钱，这取决于模型和步骤数量","智能体是否真的能在某个网站上完成某个任务，以及卡住的频率有多高","隐身和 CAPTCHA 功能是否在你关心的网站上真的有效，README 本身并不保证这一点","遥测数据包含什么内容，以及 Browser Use 从云端运行中保留了什么","磁盘占用，经验估计在 Chrome 和模型下载之前为 100 到 300 MB","它在 Windows 上的表现是否一致，那里有些部分需要 WSL"]}}},{"id":"R_kgDOUeBQeg","slug":"browser-use__jev-ultrafast","name":"Jev Ultrafast","owner":"browser-use","fullName":"browser-use/jev-ultrafast","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"46 天内获得 24046 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"acts_in_your_logged_in_chrome","severity":"attention","evidence":"README，Evidence and limits：\"Owned tabs share the existing Chrome profile\"，以及 \"Chrome connects through Browser Harness... Allow remote debugging in Chrome when prompted\" — 助手会在一个你可能已经登录了各种账号的浏览器里点击和输入。","source":"ai","title":"在你已登录的 Chrome 里操作"},{"code":"page_text_sent_to_external_services","severity":"attention","evidence":"元素表和可见文字在每次决策时都会发给 TypeSafe，文本生成则发往 .env.example 里的地址（\"TEXT_MODEL_BASE_URL=https://openrouter.ai/api/v1\"）。README 里的 \"Send visible text\" 一项讲的是如何裁剪这些内容，而不是把它们留在本地。","source":"ai","title":"页面文字会发送给外部服务"},{"code":"speed_claims_from_one_task","severity":"info","evidence":"README：\"In six alternating runs with identical models and settings... Median task time went from 9.450 s → 7.092 s... This is three repeats of one task on one browser profile, not a general reliability benchmark.\"","source":"ai","title":"速度数据只来自一项任务"},{"code":"early_mvp_with_known_gaps","severity":"info","evidence":"pyproject.toml 中的版本是 0.1.0，GitHub 上没有发布版本，README 列出 \"Shadow roots, frames, canvas, uploads, pop-up tabs, nested scrolling, and arbitrary keyboard widgets remain outside this MVP.\"","source":"ai","title":"早期 MVP，已知存在缺口"},{"code":"readme_promotes_the_vendors_paid_cloud","severity":"info","evidence":"README 的第一个板块是该公司托管产品的通知：\"BU Ultrafast, Browser Use Cloud's own hosted agent, is live on every plan\"，并附有指向 browser-use.com 和 cloud.browser-use.com 的跟踪链接。","source":"ai","title":"README 在推广厂商的付费云服务"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["TypeSafe (Jev model)","OpenRouter or another OpenAI-compatible text model"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","automation","development"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":24373,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":235,"growthDays":7,"history":[{"date":"2026-10-26","stars":23699},{"date":"2026-10-27","stars":23762},{"date":"2026-10-28","stars":23830},{"date":"2026-10-29","stars":23892},{"date":"2026-10-30","stars":23946},{"date":"2026-10-31","stars":23991},{"date":"2026-11-01","stars":24045},{"date":"2026-11-02","stars":24098},{"date":"2026-11-03","stars":24138},{"date":"2026-11-04","stars":24176},{"date":"2026-11-05","stars":24209},{"date":"2026-11-06","stars":24237},{"date":"2026-11-07","stars":24257},{"date":"2026-11-08","stars":24274},{"date":"2026-11-09","stars":24309},{"date":"2026-11-10","stars":24373}],"firstPublishedAt":"2026-10-27T01:10:12.612Z","commitSha":"8251e272398011cdef648965204d1c20cd838333","verifiedAt":"2026-11-02T02:53:42.304Z","lastCommitAt":"2026-10-31T17:35:44Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Python code that lets an AI use your Chrome browser to carry out a goal you type, quickly.","summary":"Jev Ultrafast is Python code for programmers that drives your own Chrome browser: you give one goal in plain words, such as finding flights, and the AI picks which button or box to use, step by step. It reads the page's controls as a list instead of taking screenshots, which the authors measure as faster than their earlier version. You need Python 3.12 and the uv tool, a Chrome you let it control, and two paid API keys (TypeSafe's Jev model and an OpenRouter-style text model). It is an early MVP with no packaged release.","easeRationale":"On any of the three systems it is terminal work: install Python 3.12 and uv, clone the repository, run uv sync, copy .env.example and paste two paid API keys, then let Browser Harness connect to Chrome with remote debugging allowed. There is no installer and no release to download.","costNote":"The code is free (MIT), but each run calls TypeSafe's Jev model and a text model through OpenRouter, both paid.","notVerified":["What each run really costs in TypeSafe and text-model credits.","Whether the speed figures hold on sites other than the ones measured.","Whether it works the same on Windows and macOS, since the README only shows one set of commands.","Exactly what page content is sent to TypeSafe and to the text model.","How often the agent picks the wrong control or stops with BLOCKED on your own task."]},"zh-Hans":{"summaryShort":"一段 Python 代码，让 AI 使用你的 Chrome 浏览器，快速完成你输入的目标。","summary":"Jev Ultrafast 是写给程序员的 Python 代码，用来操控你自己的 Chrome 浏览器：你用平实的话给出一个目标，比如查找航班，AI 就逐步选择该用哪个按钮或输入框。它把页面上的控件当作一份列表来读取，而不是截图，作者测得这比他们之前的版本更快。你需要 Python 3.12 和 uv 工具、一个允许它控制的 Chrome，以及两个付费的 API 密钥（TypeSafe 的 Jev 模型和一个 OpenRouter 类的文本模型）。这是早期的 MVP，没有打包好的发布版本。","easeRationale":"在三个系统上都是终端操作：安装 Python 3.12 和 uv，克隆仓库，运行 uv sync，复制 .env.example 并填入两个付费 API 密钥，再让 Browser Harness 连接到已允许远程调试的 Chrome。没有安装程序，也没有可下载的发布版本。","costNote":"代码免费（MIT），但每次运行都会调用 TypeSafe 的 Jev 模型，以及通过 OpenRouter 调用一个文本模型，两者都要付费。","notVerified":["每次运行实际要消耗多少 TypeSafe 和文本模型的额度。","速度数据在被测网站之外的其他网站上是否依然成立。","它在 Windows 和 macOS 上是否表现一致，因为 README 只给出了一套命令。","究竟有哪些页面内容会发送给 TypeSafe 和文本模型。","在你自己的任务中，助手选错控件或以 BLOCKED 停下的频率有多高。"]}}},{"id":"R_kgDOOsK2AA","slug":"browseros-ai__browseros","name":"BrowserOS","owner":"browseros-ai","fullName":"browseros-ai/BrowserOS","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"agent_acts_in_your_logged_in_accounts","severity":"attention","evidence":"README:「Your logins. Agents automate your real work using your logged-in accounts, not a blank sandbox」,以及「One click brings over your logins, bookmarks and extensions」。列出的任务包括发布到 LinkedIn 和 Twitter/X、清空收件箱、更新 CRM,所以 AI 是用你真实的账号在行动,它在页面上读到的内容也可能影响它接下来的操作。","source":"ai","title":"代理会在你已登录的账号里行动"},{"code":"installer_hosted_outside_github","severity":"info","evidence":"README 里的下载链接指向 cdn.browseros.com(BrowserOS_neo_installer.exe、BrowserOS_neo.dmg、.AppImage、.deb),而这里列出的 GitHub 发布版本只包含 claw-server 的资源压缩包,所以浏览器安装程序本身并不来自 GitHub 的发布页面。","source":"ai","title":"安装程序托管在 GitHub 之外"},{"code":"usage_telemetry_enabled_by_default","severity":"info","evidence":"README 的 FAQ:「BrowserOS neo sends anonymous product-usage events (agent connect/disconnect, version, OS) to help us improve the app; it never sends URLs, page content, prompts, tool results, or screenshots. Off with one toggle in Settings.」","source":"ai","title":"默认启用使用情况遥测"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":15,"maxMB":45,"basis":"installer"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["ai_agents","automation","productivity"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true},{"slug":"perplexity-comet","name":"Perplexity Comet","confirmed":true}],"declaredAlternatives":[],"stars":14304,"downloads":158847,"downloadsWeek":1969,"downloadsDays":3,"growth":53,"growthDays":6,"history":[{"date":"2026-11-07","stars":14273},{"date":"2026-11-08","stars":14283},{"date":"2026-11-09","stars":14296},{"date":"2026-11-10","stars":14304}],"firstPublishedAt":"2026-11-06T22:59:23.743Z","commitSha":"1d3c3dfaa9bdec18efae4ed8fbf3606101531852","verifiedAt":"2026-11-06T22:36:31.884Z","lastCommitAt":"2026-11-10T04:45:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A second Chromium browser for your AI agents: connect Claude Code or Codex and watch them do web tasks in your accounts.","summary":"BrowserOS neo is a Chromium-based browser you install next to Chrome and hand over to AI agents. It imports your Chrome logins, bookmarks and extensions, then connects to an AI assistant you already use (Claude Code, Codex, Cursor and others speak the same protocol) so the assistant can browse for you: post to social media, clear an inbox, pull reports. Each task runs in its own tab, with a dashboard and a recorded replay of every session. There are installers for Windows, macOS and Linux. The browser is free; the AI assistant you connect is usually a paid plan.","easeRationale":"There is a ready installer for each system (.exe for Windows, .dmg or Homebrew for macOS, AppImage and .deb for Linux x64), so install is one download. It is a B rather than an A because the browser does nothing on its own: after installing you still have to import from Chrome and connect an AI assistant that lives outside this project.","costNote":"The browser is free and open source. It only does something once you connect an AI assistant such as Claude Code, Codex or Cursor, which normally needs a paid plan or API credit.","notVerified":["Whether the installers really work on each system and how well the Chrome import carries over logins","Real disk use of the browser: the 15-45 MB estimate comes from the claw-server zips on the release page, not from the Chromium-based installer, which is normally much larger","Whether the usage events stop when you flip the Settings toggle, and whether sessions and screenshots under ~/.browserclaw/ truly stay on your machine","What an AI agent will actually do inside your accounts, and how much your AI plan is billed for each task","The claim that it uses fewer tokens than Claude's Chrome extension or the Codex browser"]},"zh-Hans":{"summaryShort":"为你的 AI 代理准备的第二个 Chromium 浏览器:连接 Claude Code 或 Codex,看它们在你的账号里完成网页任务。","summary":"BrowserOS neo 是一个基于 Chromium 的浏览器,你可以把它安装在 Chrome 旁边,交给 AI 代理使用。它会导入你 Chrome 里的登录状态、书签和扩展,然后连接到你已经在用的 AI 助手(Claude Code、Codex、Cursor 等都使用同一套协议),这样助手就能替你浏览网页:发社交媒体帖子、清空收件箱、提取报告。每个任务都在自己的标签页里运行,带有一个仪表盘和每次会话的录制回放。提供 Windows、macOS 和 Linux 的安装程序。浏览器本身是免费的;你连接的 AI 助手通常需要付费套餐。","easeRationale":"每个系统都有现成的安装程序(Windows 用 .exe,macOS 用 .dmg 或 Homebrew,Linux x64 用 AppImage 和 .deb),所以安装只需下载一次。它是 B 而不是 A,因为浏览器本身什么也不做:安装之后你还得从 Chrome 导入数据,并连接一个在这个项目之外的 AI 助手。","costNote":"这个浏览器是免费开源的。只有在你连接了 Claude Code、Codex 或 Cursor 这类 AI 助手之后它才会发挥作用,而这些助手通常需要付费套餐或 API 额度。","notVerified":["安装程序在每个系统上是否真的能用,以及 Chrome 导入对登录状态的还原程度如何","浏览器实际占用的磁盘空间:15-45 MB 的估算来自发布页面上 claw-server 的压缩包,而不是基于 Chromium 的安装程序本身,后者通常要大得多","关闭 Settings 里的开关后使用事件是否真的停止,以及 ~/.browserclaw/ 下的会话和截图是否真的只留在你的机器上","AI 代理在你的账号里实际会做什么,每个任务又会让你的 AI 套餐产生多少费用","它比 Claude 的 Chrome 扩展或 Codex 浏览器使用更少 token 的说法是否属实"]}}},{"id":"MDEwOlJlcG9zaXRvcnk3NzE0NTA2Ng==","slug":"budtmo__docker-android","name":"Docker-Android","owner":"budtmo","fullName":"budtmo/docker-android","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"E","linux":"D"},"risk":"attention","signals":[{"code":"using_it_means_agreeing_to_data_collection","severity":"attention","evidence":"LICENSE.md：“使用本软件即表示你同意以下非个人身份信息……将由维护者收集、处理和使用”，并且你“授予我们一项非排他性、不可撤销、全球性、永久且免许可费的权限，用于使用、修改和发布这些数据”。该列表包括你的城市、地区和国家（“来自 https://ipinfo.io”）、你机器的操作系统字符串、屏幕分辨率以及你使用了哪些模拟器设置。Pro 版表格把用户行为分析在免费版标为“是”，在付费版标为“否”。","source":"ai","title":"使用即表示同意收集数据"},{"code":"license_adds_rules_on_top_of_apache","severity":"attention","evidence":"LICENSE.md 说“本程序遵循 Apache License 2.0 的条款，以及以下关于分支和数据处理的附加条款。因此，这是一个定制版 Apache 2.0 许可证，不是你可以任选其一的双许可模式”，分支项目必须“移除我们的 Google Form ID……或者彻底停止使用情况采集”。GitHub 自己的许可证检测器对该仓库给出的结果是 NOASSERTION（无法判定）。","source":"ai","title":"许可证在 apache 基础上附加了规则"},{"code":"advertised_features_are_sponsor_only","severity":"attention","evidence":"仓库简介宣称有“视频录制、mcp 服务器和 AI 智能体”，但 PRO VERSION 表格把 AI 智能体、视频录制、无头模式、root 权限、Google Play 商店和 Android 15 及更新版本在本版本中标为“否”，只有在需要活跃 GitHub 赞助的 budtmo2/docker-android-pro 中标为“是”（部分为“是（即将推出）”）。","source":"ai","title":"宣传的功能仅限赞助者使用"},{"code":"sponsor_block_sells_proxies","severity":"info","evidence":"README.md 顶部附近的 Sponsors 板块宣传 proxy-seller.com，称有“超过 4700 万住宅 IP，覆盖 220 多个地区”，还附带折扣码。这是项目页面里的付费广告，并非工具本身的一部分。","source":"ai","title":"赞助商板块在推销代理服务"},{"code":"phone_screen_is_served_on_a_local_web_page","severity":"info","evidence":"快速开始部分使用“docker run -d -p 6080:6080 ... -e WEB_VNC=true”，第 4 点说打开 http://localhost:6080；用例 3 是通过 adb connect “在主机上控制 Android 模拟器”。文档中没有提到密码，所以在共享机器或云服务器上，任何能访问该端口的人都能看到并控制这台手机。","source":"ai","title":"手机屏幕通过本地网页提供访问"},{"code":"android_versions_are_behind","severity":"info","evidence":"免费镜像的表格最高只到 2023 年发布的 Android 14（API 34），而这里最后一次提交是在 2026-11-02。更新的 Android 版本只出现在赞助者镜像中。","source":"ai","title":"android 版本落后"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Genymotion SaaS or Genymotion on AWS/GCP/Alibaba, only for the cloud device images"],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["development","hardware"],"alternativeTo":[{"slug":"genymotion","name":"Genymotion","confirmed":true},{"slug":"browserstack-app-automate","name":"App Automate","confirmed":true}],"declaredAlternatives":[],"stars":16113,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":25,"growthDays":6,"history":[{"date":"2026-11-05","stars":16091},{"date":"2026-11-06","stars":16098},{"date":"2026-11-07","stars":16100},{"date":"2026-11-08","stars":16107},{"date":"2026-11-09","stars":16108},{"date":"2026-11-10","stars":16113}],"firstPublishedAt":"2026-11-04T22:32:54.844Z","commitSha":"7d12a128e29f29fd0b666da376cb8ac00b23a76c","verifiedAt":"2026-11-04T22:18:23.238Z","lastCommitAt":"2026-11-10T15:39:05Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs Android phone emulators inside Docker and shows the phone screen in your browser, for app testing.","summary":"Docker-Android packs the Android emulator into ready Docker images. One command starts a container with a phone profile such as Samsung Galaxy S10 and Android 9 to 14, and you watch and tap the screen at http://localhost:6080 in your browser. It is made for app developers and testers: build an Android project, run Appium or Espresso tests, simulate SMS, plug it into Jenkins or a cloud server, or drive it from an assistant through the project's MCP server. The image runs on Ubuntu, with Docker and hardware virtualization (KVM).","easeRationale":"Ubuntu Linux is the home platform and sets the grade: install Docker, check that virtualization is on with kvm-ok, then run one long docker command with --device /dev/kvm and open http://localhost:6080. Windows 11 is about as hard but with extra plumbing: the README has you join the kvm group, write nestedVirtualization=true into .wslconfig and a chown line into /etc/wsl.conf, then restart WSL. On a Mac the README offers no real route, only \"you need to use Virtual Machine that support Virtualization with Ubuntu OS because the image can be run under Ubuntu OS only\".","costNote":"The images in this repository are free. A separate docker-android-pro \"can be pulled only by active sponsor\" through GitHub Sponsors, and its table keeps the AI agent, proxy and language settings, root privileges, headless mode, Android 15 to 17, Selenium 4 integration, Google Play Store and video recording for paying sponsors. Genymotion Cloud, which it can also drive, is a paid service.","notVerified":["How many gigabytes each emulator image takes once pulled, and how much memory one phone needs","Whether the WSL2 route really gives working KVM acceleration on a given Windows 11 machine","Whether anything useful runs on an Apple-chip Mac, where nested virtualization is normally missing","What the usage reporting sends in practice, and whether it can be switched off in the free image","Whether the MCP server and the Appium MCP, both marked beta, work with a given assistant","How well heavy apps and games behave inside the emulator"]},"zh-Hans":{"summaryShort":"在 Docker 中运行 Android 手机模拟器，并在浏览器里显示手机屏幕，用于测试应用。","summary":"Docker-Android 把 Android 模拟器打包成现成的 Docker 镜像。一条命令就能启动一个带有手机配置（例如三星 Galaxy S10，Android 9 到 14）的容器，你可以在浏览器里通过 http://localhost:6080 查看并操作屏幕。它面向应用开发者和测试人员：构建 Android 项目、运行 Appium 或 Espresso 测试、模拟短信、接入 Jenkins 或云服务器，或者通过项目的 MCP 服务器让智能体来操作它。镜像运行在 Ubuntu 上，需要 Docker 和硬件虚拟化（KVM）。","easeRationale":"Ubuntu Linux 是它的本家平台，决定了评级：安装 Docker，用 kvm-ok 确认虚拟化已开启，然后运行一条带 --device /dev/kvm 的长命令，再打开 http://localhost:6080。Windows 11 的难度差不多，但多了一些配置：README 要你加入 kvm 用户组，在 .wslconfig 里写入 nestedVirtualization=true，在 /etc/wsl.conf 里加一行 chown，然后重启 WSL。在 Mac 上 README 没有给出真正可行的路径，只说“你需要使用支持虚拟化的虚拟机并安装 Ubuntu 系统，因为该镜像只能在 Ubuntu 系统下运行”。","costNote":"本仓库里的镜像是免费的。一个独立的 docker-android-pro 版本“只能由活跃的赞助者下载”，通过 GitHub Sponsors 获取，其表格把 AI 智能体、代理和语言设置、root 权限、无头模式、Android 15 到 17、Selenium 4 集成、Google Play 商店和视频录制都保留给付费赞助者。它也能驱动的 Genymotion Cloud 是一项付费服务。","notVerified":["每个模拟器镜像下载后占用多少 GB，一台手机需要多少内存","WSL2 路线是否真的能在某台 Windows 11 机器上提供可用的 KVM 加速","搭载 Apple 芯片的 Mac 上能否运行任何有用的东西，因为那类机器通常缺少嵌套虚拟化","使用情况上报实际发送了什么内容，免费镜像中能否关闭它","都标记为测试版的 MCP 服务器和 Appium MCP 能否配合某个特定助手使用","重负载的应用和游戏在模拟器里表现如何"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1MzYzMTk0NQ==","slug":"burntsushi__ripgrep","name":"ripgrep","owner":"BurntSushi","fullName":"BurntSushi/ripgrep","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"terminal_only_no_window","severity":"info","evidence":"README, Installation 部分写道：\"The binary name for ripgrep is rg.\" README 里的每个示例都是命令行形式，例如 `rg -n -w '[A-Z]+_SUSPEND'`。它没有图形界面，只有一个从 README 链接过去的非官方网页版试用环境。","source":"ai","title":"只能在终端使用，没有窗口界面"},{"code":"can_run_other_programs_you_configure","severity":"info","evidence":"README 写道，ripgrep \"supports arbitrary input preprocessing filters which could be PDF text extraction, less supported decompression, decrypting, automatic encoding detection and so on\"。这意味着它会运行你写在配置文件或参数里的命令，所以在使用共享的配置文件之前，最好先读一下里面的内容。","source":"ai","title":"可以运行你自己配置的其他程序"},{"code":"latest_release_is_older_than_the_code","severity":"info","evidence":"最新发布版本是 2026-07-15 的 15.2.0，而最新的提交是在 2026-10-27，也就是说最近的改动只存在于源代码中。你的软件包管理器提供的版本可能还要更旧，README 里就提到了 Debian stable 的情况。","source":"ai","title":"最新发布版本比代码本身更旧"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":2,"maxMB":5,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Unlicense","category":"permissive","detectedFrom":"github"},"categories":["development","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":69970,"downloads":72670810,"downloadsWeek":2895927,"downloadsDays":7,"growth":289,"growthDays":7,"history":[{"date":"2026-10-26","stars":69477},{"date":"2026-10-27","stars":69513},{"date":"2026-10-28","stars":69545},{"date":"2026-10-29","stars":69551},{"date":"2026-10-30","stars":69583},{"date":"2026-10-31","stars":69598},{"date":"2026-11-01","stars":69627},{"date":"2026-11-02","stars":69662},{"date":"2026-11-03","stars":69681},{"date":"2026-11-04","stars":69698},{"date":"2026-11-05","stars":69736},{"date":"2026-11-06","stars":69808},{"date":"2026-11-07","stars":69873},{"date":"2026-11-08","stars":69904},{"date":"2026-11-09","stars":69937},{"date":"2026-11-10","stars":69970}],"firstPublishedAt":"2026-10-27T01:10:12.612Z","commitSha":"91f5d22ef48bde459d90a8c7dbeab416e97d0f2b","verifiedAt":"2026-10-31T01:39:55.958Z","lastCommitAt":"2026-11-09T20:53:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A terminal command that searches text inside all files of a folder very fast, skipping files git ignores.","summary":"ripgrep (the command is `rg`) looks for a word or a pattern inside every file of a folder and prints the lines it finds. It skips what your .gitignore lists, hidden files and binary files unless you ask otherwise, and it can limit the search to one kind of file, read inside compressed files and handle other text encodings. It is one small program of about 2 MB with ready downloads for Windows, macOS and Linux, and it is also in most package managers. There is no window: everything happens in the terminal with flags.","easeRationale":"On every system it is one 2 MB ready file with nothing else to install: on Windows `winget install BurntSushi.ripgrep.MSVC` or the zip from the releases page, on macOS `brew install ripgrep`, on Linux `sudo apt-get install ripgrep` or the .deb. The grade stays at C because it has no window at all, so you have to work in the terminal and learn flags such as `rg -tpy foo` or `rg -uuu`. Rust is only needed if you choose to compile it yourself with `cargo build --release`.","costNote":"Free; COPYING says the code is dual-licensed under the Unlicense and MIT, which puts almost no conditions on reuse.","notVerified":["The real speed on your computer and your files: the README's own tables show cases where it is only slightly faster, and one pattern that takes over 15 seconds","Whether the prebuilt file you download includes the optional PCRE2 engine needed for `-P`, since that is a build choice","Which version your package manager installs, which can be older than 15.2.0","Whether a configuration file you inherited from someone else changes the default search behaviour"]},"zh-Hans":{"summaryShort":"一个终端命令，能在文件夹里的所有文件中快速搜索文本，并跳过 git 忽略的文件。","summary":"ripgrep（命令名是 `rg`）会在一个文件夹里的每个文件中查找一个词或一个模式，并打印出找到的那些行。默认情况下它会跳过 .gitignore 里列出的内容、隐藏文件和二进制文件，除非你另行指定；它还可以把搜索限定在某一类文件上，读取压缩文件内部的内容，并处理其他文本编码。它是一个大约 2 MB 的小程序，提供适用于 Windows、macOS 和 Linux 的现成下载，并且也能在大多数软件包管理器中找到。它没有窗口界面：一切都在终端里通过参数（flags）完成。","easeRationale":"在所有系统上，它都是一个 2 MB 左右的现成文件，无需安装其他任何东西：在 Windows 上可以用 `winget install BurntSushi.ripgrep.MSVC`，或者从 releases 页面下载 zip；在 macOS 上用 `brew install ripgrep`；在 Linux 上用 `sudo apt-get install ripgrep` 或 .deb 包。评级维持在 C，因为它完全没有窗口界面，你必须在终端里操作，并学习像 `rg -tpy foo` 或 `rg -uuu` 这样的参数。只有当你想自己用 `cargo build --release` 编译时才需要 Rust。","costNote":"免费；COPYING 文件说明代码采用 Unlicense 和 MIT 双重许可，这几乎不对重用设置任何条件。","notVerified":["在你自己的电脑和文件上的实际速度：README 自带的表格就显示过只是略快一点的情况，以及有一种模式耗时超过 15 秒。","你下载的预编译文件是否包含 `-P` 所需的可选 PCRE2 引擎，因为这取决于构建方式。","你的软件包管理器安装的是哪个版本，它可能比 15.2.0 更旧。","你从别人那里继承来的配置文件是否会改变默认的搜索行为。"]}}},{"id":"R_kgDOL-f_zw","slug":"buxuku__smartsub","name":"SmartSub","owner":"buxuku","fullName":"buxuku/SmartSub","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：diff: GHSA-73rr-hh4g-fpgx","source":"check","title":"已知漏洞"},{"code":"downloads_videos_and_unlocks_paid_content_with_cookies","severity":"attention","evidence":"该应用内置下载器，提供“yt-dlp（YouTube 及 1800+ 站点）与 lux（B 站、抖音、小红书等国内平台）双引擎”，还可以导入浏览器 cookie 来“解锁登录后可见的高清画质与会员内容”。以这种方式保存视频违反 YouTube、哔哩哔哩等网站的使用条款，在一些国家还会受到更严格的限制。请只用于你自己的素材或你被允许复制的素材。","source":"ai","title":"下载视频并借助 cookie 解锁付费内容"},{"code":"copies_a_real_persons_voice","severity":"attention","evidence":"配音功能提供“声音克隆：本地 ZipVoice 零样本克隆（一段参考音频即建即用）”，加上 ElevenLabs 的即时克隆和火山引擎的声音复刻 2.0：一段简短录音就足以让视频用那个声音说话。没有任何环节核实声音主人是否同意，输出内容也没有被标记为生成内容。","source":"ai","title":"复制真人的声音"},{"code":"free_voice_service_is_a_reverse_engineered_endpoint","severity":"info","evidence":"在配音引擎表格中，Edge TTS 被描述为“免费、无需 Key；属逆向接口试用档，不承诺可用性，断供时建议切换其它引擎”——之所以免费且不需要密钥，是因为它对接的是通过逆向工程得到的微软接口，无法保证一直可用。","source":"ai","title":"免费语音服务是一个逆向得来的接口"},{"code":"the_page_earns_from_the_services_it_suggests","severity":"info","evidence":"README.md 开头的赞助板块宣传了一个带推荐链接的 API 代理商（aiyoyoo.com/register?aff=YWL9FL2XFPKM，“倍率低至0.05x”，“新用户注册赠送 $20”），翻译列表里链接到带 ?aff=QvHM 的 DeerAPI，赞助商部分还使用了 DigitalOcean 的推荐链接。所以项目页面里的一些推荐会给作者带来收益。","source":"ai","title":"项目页面从它推荐的服务中获利"},{"code":"mac_build_is_not_notarized","severity":"info","evidence":"只有 Windows 安装包经过签名：代码签名部分说明范围是“Windows 安装包（SmartSub_Windows_<版本>_x64.exe）”，通过 SignPath 完成，并且“签名从证书签发后的首个版本开始生效，此前发布的版本未签名”。Mac 的 FAQ 要求手动移除隔离标志，这正是未签名、未公证应用的典型表现。","source":"ai","title":"mac 版本没有经过公证"},{"code":"background_service_that_outside_assistants_can_drive","severity":"info","evidence":"自动化部分描述了一个“无窗口后台守护（Headless Daemon）”，当某个 AI 工具或 CLI 第一次调用它时会自行启动，README 提到的 111 个 MCP 工具可让 Cursor、Claude Code 或 Codex 运行整个流程。你连接的助手随后就能读写你的媒体文件以及应用的设置，包括存放在那里的密钥。","source":"ai","title":"外部助手可以操控的后台服务"},{"code":"models_come_from_several_download_sources","severity":"info","evidence":"模型是在应用内部从 Hugging Face、hf-mirror.com 镜像站、ModelScope 和 GitHub 获取的（“支持 ModelScope / GitHub 等多源”），下载引擎 yt-dlp 和 lux 也是在应用内部安装的。108MB 的安装包只是个开始，模型文件要大得多。","source":"ai","title":"模型来自多个下载源"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","gpu_nvidia","ffmpeg"],"disk":{"minMB":108,"maxMB":324,"basis":"installer"},"apiKeys":["AI models for translating, polishing and the built-in assistant: DeepSeek, Qwen, Gemini, SiliconFlow or any OpenAI-compatible endpoint (Ollama on your own computer works too)","Cloud speech-to-text, nine providers including OpenAI, Deepgram, ElevenLabs Scribe, Gladia, Tencent, Alibaba, iFlytek, Volcengine and Xiaomi MiMo","Cloud dubbing voices: Azure Speech, ElevenLabs, Volcengine Doubao, OpenAI-compatible endpoints"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","audio","local_ai"],"alternativeTo":[{"slug":"descript","name":"Descript","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"otter-ai","name":"Otter.ai","confirmed":true},{"slug":"4k-video-downloader","name":"4K Video Downloader","confirmed":false}],"declaredAlternatives":[],"stars":6000,"downloads":479411,"downloadsWeek":9778,"downloadsDays":5,"growth":57,"growthDays":6,"history":[{"date":"2026-11-05","stars":5955},{"date":"2026-11-06","stars":5960},{"date":"2026-11-07","stars":5967},{"date":"2026-11-08","stars":5980},{"date":"2026-11-09","stars":5993},{"date":"2026-11-10","stars":6000}],"firstPublishedAt":"2026-11-04T22:38:15.171Z","commitSha":"ac8d2f8ee76fe4d84f23c24a48474df19346bddd","verifiedAt":"2026-11-04T22:19:31.814Z","lastCommitAt":"2026-11-09T07:03:45Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns speech in videos into subtitles on your own computer, translates them, dubs them and burns them in.","summary":"SmartSub (妙幕) is a desktop app for subtitles. You drop in video or audio files, or paste a link from YouTube or Bilibili, and it writes down what is said using speech models that run on your machine, translates the subtitles, lets you correct them line by line against the video, can read the translation out loud in a synthetic or cloned voice, and finally burns or attaches the subtitles to the video. There are installers for Windows, Mac and Linux, and the first run asks you to download a speech model. Cloud services are optional extras.","easeRationale":"All three systems get B: there is a ready installer for each (a signed .exe for Windows, .dmg for Intel and Apple-chip Macs, .deb or AppImage for Linux, plus a Homebrew cask), and the one extra step is the first-run wizard asking you to download a speech model, with GPU acceleration packs also fetched inside the app. Windows is the smoothest. On a Mac the FAQ warns the app may be reported as damaged and tells you to run \"sudo xattr -dr com.apple.quarantine /Applications/SmartSub.app\" once in the terminal.","costNote":"The app is free and MIT licensed, and the README describes a route that costs nothing: local speech models, the built-in free translation sources, local voices and the built-in ffmpeg. The paid cloud options (OpenAI, ElevenLabs, Azure, Tencent, Alibaba and others) are billed by those companies if you turn them on.","notVerified":["How accurate the subtitles are for a given language, and how long one hour of video takes on your machine","Real disk space after the speech models, voice models and GPU acceleration packs are downloaded","Whether the GPU acceleration really works on your graphics card, since the FAQ already answers crashes after turning it on","The privacy claim that nothing is uploaded: only the parts you point at a cloud service can be checked from outside","What the cloud providers charge for long videos","Whether the cloned voice is good enough to pass for the original"]},"zh-Hans":{"summaryShort":"在你自己的电脑上把视频里的语音变成字幕，翻译、配音并压制进视频。","summary":"SmartSub（妙幕）是一款桌面字幕应用。你把视频或音频文件拖进去，或者粘贴一个 YouTube 或哔哩哔哩链接，它用在本机运行的语音模型把说话内容写下来，翻译字幕，让你逐行对照视频校对，可以用合成或克隆的声音朗读译文，最后把字幕压制或附加到视频里。它为 Windows、Mac 和 Linux 提供安装包，首次运行会要求下载一个语音模型。云端服务是可选的附加项。","easeRationale":"三个系统都是 B 级：每个系统都有现成的安装包（Windows 上是签名的 .exe，Mac 上为 Intel 和 Apple 芯片分别提供 .dmg，Linux 上有 .deb 或 AppImage，还有一个 Homebrew cask），唯一的额外步骤是首次运行的向导要求下载语音模型，GPU 加速包也是在应用内部获取。Windows 体验最顺畅。在 Mac 上，FAQ 提醒应用可能被报告为“已损坏”，需要在终端里运行一次“sudo xattr -dr com.apple.quarantine /Applications/SmartSub.app”。","costNote":"该应用免费并采用 MIT 许可，README 描述了一条不花钱的路线：本地语音模型、内置的免费翻译源、本地语音以及自带的 ffmpeg。付费的云端选项（OpenAI、ElevenLabs、Azure、腾讯、阿里云等）如果你启用，会由这些公司收费。","notVerified":["某种语言的字幕准确度如何，一小时视频在你机器上需要多久","下载语音模型、语音克隆模型和 GPU 加速包之后实际占用的磁盘空间","GPU 加速是否真的能在你的显卡上生效，因为 FAQ 已经在回答开启后崩溃的问题","“没有任何内容会上传”的隐私说法：只有你指向云端服务的那部分才能从外部核实","云端供应商对长视频收费多少","克隆的声音是否足够逼真，能以假乱真"]}}},{"id":"MDEwOlJlcG9zaXRvcnk3OTAwNjcyMw==","slug":"calclavia__deepj","name":"DeepJ","owner":"calclavia","fullName":"calclavia/DeepJ","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","linux":"E"},"risk":"attention","signals":[{"code":"obsolete_dependencies","severity":"attention","evidence":"README 的 Requirements 部分写着\"Python 3.5\"(自2020年起已不再受支持),requirements.txt 中列出的\"keras\"和\"tensorflow-gpu\"没有锁定版本,而后者是 TensorFlow 已停止发布的包名,因此在现代机器上直接安装很可能需要修改才能成功。","source":"ai","title":"依赖已过时"},{"code":"no_trained_model_or_dataset","severity":"info","evidence":"README写道:\"The dataset is not provided in this repository. To train a custom model, you will need to include a MIDI dataset in the data/ folder.\"顶层文件列表中没有发布文件,也没有权重文件,所以在训练之前无法生成任何内容。","source":"ai","title":"没有训练好的模型或数据集"},{"code":"research_prototype_unmaintained","severity":"info","evidence":"项目链接到论文 arxiv.org/abs/1801.00887,将风格控制称为\"a simple technique ... as a proof of concept\",最后一次提交在2018-03-05,还有17个未关闭的issue,也没有任何发布。","source":"ai","title":"研究原型已无人维护"},{"code":"third_party_dependency_from_git","severity":"info","evidence":"README要求克隆 github.com/vishnubob/python-midi 并运行\"python3 setup.py install\",也就是说部分安装来自 Python 包索引之外的另一个代码仓库。","source":"ai","title":"第三方依赖来自git"}],"platforms":["windows","linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai","development"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":744,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":744},{"date":"2026-11-08","stars":744},{"date":"2026-11-09","stars":744},{"date":"2026-11-10","stars":744}],"firstPublishedAt":"2026-11-06T22:59:23.752Z","commitSha":"e2058ca4c05d10e66a87c6c4205cd5495c846627","verifiedAt":"2026-11-06T22:37:09.270Z","lastCommitAt":"2018-03-05T19:00:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that trains a neural network to compose MIDI music in a chosen composer's style. You bring the music data.","summary":"DeepJ is the code behind a 2018 research paper on generating music in a specific mixture of composer styles. It is a set of Python scripts: one trains the model, another writes new MIDI files. There is no ready program, no trained model and no music data in the repository, so you have to collect your own MIDI collection and train it yourself on a machine with a graphics card. It is meant for people comfortable with Python and machine learning, and the code has not been touched since March 2018.","easeRationale":"On Linux or Windows this is developers-only work: the README asks for Python 3.5, a separate library (python-midi) installed from source with setup.py, then pip install of Keras and tensorflow-gpu, and finally your own MIDI dataset placed in a data/ folder before any training can start. There is no installer, no release file and no trained model to simply run.","costNote":"MIT licensed code, free to use. You pay in your own hardware or rented GPU time for training.","notVerified":["Whether the code still installs and runs with today's Python, Keras and TensorFlow versions","How long training takes, which graphics card it needs and how much memory and disk the dataset and checkpoints use (the 100-300 MB estimate is a guess, not measured)","How good the generated music actually sounds, since the paper's rating comes from the authors","Whether the external python-midi library still builds on a modern system"]},"zh-Hans":{"summaryShort":"训练神经网络以特定作曲家风格创作MIDI音乐的研究代码,需要你自己提供音乐数据。","summary":"DeepJ 是 2018 年一篇研究论文背后的代码,该论文研究如何以多位作曲家风格的混合方式生成音乐。它是一组 Python 脚本:一个用于训练模型,另一个用于生成新的 MIDI 文件。仓库中没有现成的程序、没有训练好的模型,也没有音乐数据,所以你需要自己收集 MIDI 数据集,并在配有显卡的机器上自行训练。它面向熟悉 Python 和机器学习的人群,代码自 2018 年 3 月起未再更新。","easeRationale":"在 Linux 或 Windows 上,这完全是开发者的工作:README 要求 Python 3.5,需要用 setup.py 从源码安装另一个库(python-midi),再用 pip 安装 Keras 和 tensorflow-gpu,最后还要把你自己的 MIDI 数据集放进 data/ 文件夹才能开始训练。没有安装程序,没有发布文件,也没有可以直接运行的训练好的模型。","costNote":"代码采用 MIT 许可,可免费使用。训练所需的硬件或租用的 GPU 时间费用由你自己承担。","notVerified":["该代码能否在当前版本的 Python、Keras 和 TensorFlow 下仍然正常安装和运行","训练需要多长时间、需要什么显卡,以及数据集和检查点占用多少内存和磁盘空间(100-300 MB 的估计只是猜测,并未实际测量)","生成的音乐实际听起来有多好,因为论文中的评分来自作者本人","外部库 python-midi 是否仍能在现代系统上构建"]}}},{"id":"R_kgDORz_BDQ","slug":"calesthio__openmontage","name":"OpenMontage","owner":"OpenMontageInc","fullName":"OpenMontageInc/OpenMontage","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号是 47 天前创建的","source":"check","title":"账号注册时间很短的所有者"},{"code":"paid_provider_spend_on_your_accounts","severity":"attention","evidence":".env.example 列出了 29 个服务商账号的密钥，所有工作都在你自己的账号上运行；README 的例子引用了一段 60 秒视频的 \"Total cost: $1.33\"，以及完整配置的 \"~$1–$3\"，并描述了默认的总预算上限为 $10。","source":"ai","title":"在你自己的账号上产生付费服务商开销"},{"code":"no_releases_and_early_version_number","severity":"info","evidence":"项目信息：GitHub 上 0 个发行版，323 个未关闭的 issue；setup.py 声明 version=\"0.1.0\"。你安装的是当天 main 分支上的内容。","source":"ai","title":"没有发行版且版本号很早"},{"code":"agpl_network_copyleft","severity":"info","evidence":"LICENSE 是 GNU Affero General Public License v3：如果你通过网络把修改后的版本提供给其他人使用，就必须把那个版本的源代码也给他们。","source":"ai","title":"AGPL 的网络传染性条款"},{"code":"code_fetched_from_npm_at_run_time","severity":"info","evidence":"README 在介绍 HyperFrames 渲染器时写道：\"Consumed via `npx hyperframes`; no monorepo checkout needed\" —— 这个包会在每次制作运行时从 npm 拉取。","source":"ai","title":"运行时从 npm 拉取代码"},{"code":"repository_owner_mismatch","severity":"info","evidence":"项目信息给出的仓库是 OpenMontageInc/OpenMontage，而 README 里的克隆命令和徽章仍然指向 github.com/calesthio/OpenMontage。","source":"ai","title":"仓库所有者不一致"}],"platforms":["windows","mac","linux"],"tools":["python","node","ffmpeg","other"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":["fal.ai (FAL_KEY / FAL_AI_API_KEY)","OpenAI","Google (Gemini, Imagen, Cloud TTS)","ElevenLabs","Replicate","Kling","Runway","HeyGen","Suno","Hugging Face","Pexels","Pixabay"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["ai_skills","video","ai_agents"],"alternativeTo":[{"slug":"invideo-ai","name":"AI","confirmed":true},{"slug":"pictory","name":"Pictory","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"opusclip","name":"OpusClip","confirmed":false},{"slug":"ltx-studio","name":"LTX Studio","confirmed":false}],"declaredAlternatives":[],"stars":79396,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":3423,"growthDays":7,"history":[{"date":"2026-11-03","stars":75973},{"date":"2026-11-04","stars":76449},{"date":"2026-11-05","stars":77088},{"date":"2026-11-06","stars":77638},{"date":"2026-11-07","stars":78217},{"date":"2026-11-08","stars":78628},{"date":"2026-11-09","stars":79069},{"date":"2026-11-10","stars":79396}],"firstPublishedAt":"2026-11-03T02:52:21.596Z","commitSha":"356b5fa92474998c9f9a0782fd5547d05837d284","verifiedAt":"2026-11-08T23:04:55.858Z","lastCommitAt":"2026-11-09T06:23:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your AI coding assistant into a video studio: script, narration, visuals, subtitles and the final render.","summary":"A set of Python tools, YAML pipelines and Markdown skill files that your AI coding assistant (Claude Code, Cursor, Copilot, Windsurf, Codex) reads to build a video end to end: research, script, narration, generated images or stock footage, subtitles and the final render. For video creators who already work inside one of those assistants. You clone the repo and run `make setup`, with Python 3.10+, FFmpeg and Node.js on a patched 22.x (22.23.4+), 24.x (24.22.1+) or 26.x (26.12.1+) release. Provider keys are optional.","easeRationale":"Grade D on Windows, macOS and Linux alike: there is nothing to download (0 GitHub releases), so you clone the repository and run `make setup` in a terminal, and you need Python 3.10+, FFmpeg and a patched Node.js release on the 22.x (22.23.4+), 24.x (24.22.1+) or 26.x (26.12.1+) line (`.node-version` selects 22.23.4), plus one of the supported AI coding assistants to drive it. The free local video generation path also asks for a graphics card.","costNote":"AGPL-3.0 code, no purchase. The README describes a path with no paid keys (Piper voices, open archives, Remotion, FFmpeg) and its own examples quote \"Total cost: $1.33\" for one 60-second video and \"~$1–$3\" for a full setup when paid providers are used.","notVerified":["Whether `make setup` finishes on each system; the README itself notes that on Windows `npm install` may fail with ERR_INVALID_ARG_TYPE and suggests `npx --yes npm install`.","What one video really costs on your provider accounts, and whether the budget cap and approval thresholds hold in practice.","What the AI coding assistant itself costs in tokens for a full production, which is on top of the provider keys.","Real disk use after the Python packages, the Node modules and any local video models are downloaded.","What the `npx hyperframes` package downloads and runs at production time.","Whether the no-paid-key path (Piper voices, open archives, Remotion) produces videos you would publish.","Whether the quality gates, self-review and decision log behave as the README describes."]},"zh-Hans":{"summaryShort":"把你的 AI 编程助手变成一个视频工作室：脚本、配音、画面、字幕和最终渲染。","summary":"这是一套 Python 工具、YAML 流水线和 Markdown 技能文件，供你的 AI 编程助手（Claude Code、Cursor、Copilot、Windsurf、Codex）读取，从头到尾制作一段视频：调研、脚本、配音、生成的图片或素材库画面、字幕和最终渲染。适合已经在用其中一种助手工作的视频创作者。你克隆仓库并运行 `make setup`，需要 Python 3.10+、FFmpeg，以及打过补丁的 Node.js 22.x（22.23.4+）、24.x（24.22.1+）或 26.x（26.12.1+）版本。服务商密钥是可选的。","easeRationale":"在 Windows、macOS 和 Linux 上都是 D 级：没有东西可以下载（GitHub 上 0 个发行版），所以你要克隆仓库并在终端里运行 `make setup`，还需要 Python 3.10+、FFmpeg，以及 22.x（22.23.4+）、24.x（24.22.1+）或 26.x（26.12.1+）系列里打过补丁的 Node.js 版本（`.node-version` 指定的是 22.23.4），再加上一个受支持的 AI 编程助手来驱动它。免费的本地视频生成路径还需要一块显卡。","costNote":"代码采用 AGPL-3.0 许可，无需购买。README 描述了一条不需要付费密钥的路径（Piper 语音、开放素材库、Remotion、FFmpeg），其自带的例子引用了一段 60 秒视频的 \"Total cost: $1.33\"，以及使用付费服务商时完整配置的 \"~$1–$3\"。","notVerified":["`make setup` 能否在各个系统上顺利完成；README 自己也提到在 Windows 上 `npm install` 可能会报 ERR_INVALID_ARG_TYPE 错误，并建议使用 `npx --yes npm install`。","在你自己的服务商账号上，一段视频实际花费多少，以及预算上限和审批阈值在实际使用中是否真的有效。","完整制作一段视频时，AI 编程助手本身在 token 上的花费是多少，这是在服务商密钥费用之外的额外开销。","下载完 Python 包、Node 模块和任何本地视频模型之后，实际占用的磁盘空间。","`npx hyperframes` 这个包在制作时到底会下载并运行什么内容。","不花钱买密钥的那条路径（Piper 语音、开放素材库、Remotion）产出的视频，是否达到你愿意发布的水平。","质量把关、自我审查和决策记录是否真的像 README 描述的那样运作。"]}}},{"id":"R_kgDOMHKOZQ","slug":"cameronking4__replyguy-clone","name":"ReplyGuy Clone","owner":"cameronking4","fullName":"cameronking4/ReplyGuy-clone","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖存在已知漏洞(OSV):next: GHSA-2464-fhr4-w9xw, GHSA-2xp9-vwfh-vxw4, GHSA-36qx-fr4f-26g5;next-auth: GHSA-5jpx-9hw9-2fx4, GHSA-7rqj-j65f-68wh, GHSA-x445-f3h2-j279","source":"check","title":"存在已知漏洞"},{"code":"posts_ai_replies_in_your_name","severity":"attention","evidence":"仓库描述写道,它能\"Gain organic sales by replying to posts on social media scraped using keywords on X, Linkedin & Reddit\",README 也说要\"respond organically while promoting whatever you want to promote\",按每日计划进行。做这件事的依赖都在:`twitter-api-v2`、`snoowrap`(Reddit)、`linkedin-api-client` 和 `openai`。X、LinkedIn 和 Reddit 的条款都限制批量自动回复以及未声明的付费推广,这样做的账号常常会被封。","source":"ai","title":"以你的名义发布 ai 写的回复"},{"code":"scrapes_social_networks_with_a_hidden_browser","severity":"attention","evidence":"`puppeteer`(一个被远程控制的 Chrome)和 `apify-client`(一个商业抓取服务)都是依赖项,README 自己也这么说:\"scrape posts across X, Linkedin & Reddit\"。这样收集帖子违反了这些网站的条款,在一些地方也违反数据保护规定,因为这些帖子带着别人的名字。","source":"ai","title":"用隐藏的浏览器抓取社交网络数据"},{"code":"readme_belongs_to_another_project","severity":"info","evidence":"安装部分让你从另一个仓库创建项目——`npx create-next-app my-saas-project --example \"https://github.com/mickasmt/next-saas-stripe-starter\"`——`package.json` 里的名字仍然是 `\"saas-starter\"`,作者是 `mickasmt`,整个 Tech Stack、Roadmap 和 Author 部分说的都是那个付费 SaaS 模板。第 2 步让你复制 `.env.example`,但文件列表里根本没有这个文件,所以需要的密钥都没有文档说明。","source":"ai","title":"readme 属于另一个项目"},{"code":"no_commits_for_almost_two_years","severity":"attention","evidence":"最后一次提交是在 2024-12-08,距这次检查 697 天,而锁定的版本 `next` 14.2.7 和 `next-auth` 4.24.7 正是漏洞检查标记出来的那些。仓库没有被标记为归档,但没人在更新它。","source":"ai","title":"近两年没有提交"},{"code":"built_to_charge_other_people","severity":"info","evidence":"`stripe`、一个 `prisma/` schema、通过 Auth.js 登录,以及 roadmap 里的一个计费表单(\"Add Server Actions on billing form (stripe)\")都说明这是一个付费订阅网站套件,不是给一个人用的程序;首页 `https://buzzdaddy.ai` 就是作者自己托管的版本。","source":"ai","title":"是用来向别人收费的"},{"code":"posts_promotional_replies_as_organic_content","severity":"high","evidence":"描述:\"Gain organic sales by replying to posts on social media scraped using keywords on X, Linkedin & Reddit. Use the AI agent to craft UGC buzz about your product / startup / channel etc. on a daily schedule.\";README:\"Set keywords to scrape posts across X, Linkedin & Reddit and respond organically while promoting whatever you want to promote. Use the AI agent to automate your UGC promotion daily. UGC converts better than paid ads!\" 做这件事的部分都在 package.json 里:\"twitter-api-v2\"、\"snoowrap\"、\"linkedin-api-client\"、\"puppeteer\"、\"apify-client\" 和 \"openai\"。很多国家的广告规定要求推广性内容必须标明是广告,而 X、LinkedIn 和 Reddit 的条款也限制自动发帖以及从这些网站收集帖子的行为。","source":"ai","title":"把推广回复包装成自然内容发布"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["OpenAI","Apify","X (Twitter) API","Reddit API","LinkedIn API","Stripe","Resend"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","ai_agents","automation"],"alternativeTo":[{"slug":"replyguy","name":"ReplyGuy","confirmed":true},{"slug":"phantombuster","name":"PhantomBuster","confirmed":true},{"slug":"jasper","name":"Jasper","confirmed":true}],"declaredAlternatives":[],"stars":121,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":1,"growthDays":6,"history":[{"date":"2026-11-06","stars":120},{"date":"2026-11-07","stars":121},{"date":"2026-11-08","stars":121},{"date":"2026-11-09","stars":121},{"date":"2026-11-10","stars":121}],"firstPublishedAt":"2026-11-05T22:43:48.298Z","commitSha":"391eefab4301d1cd71037714746fd43c98cca16a","verifiedAt":"2026-11-05T22:25:03.829Z","lastCommitAt":"2024-12-08T03:36:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted web app that finds posts on X, LinkedIn and Reddit by keyword and posts AI-written replies promoting your product.","summary":"BuzzDaddy is the source code of a web app you host yourself. You give it keywords, it collects matching posts from X, LinkedIn and Reddit, and an AI writes replies that bring up your product, on a daily schedule. It is a Next.js project: you clone it, install Node and pnpm, fill in a long environment file, create a database for Prisma and bring your own OpenAI, Apify and social network keys before anything works. Aimed at founders who do their own marketing. The README is still the one from the paid-SaaS template it was built on and never explains the reply features.","easeRationale":"There is nothing to download: no GitHub release and no installer. On Windows, Mac and Linux alike you clone the repository, install Node and pnpm, run `pnpm install`, copy an environment file, point Prisma at a database and paste in keys for OpenAI, Apify and the social networks, which puts every platform at D.","costNote":"The code costs nothing, but the dependencies point at paid services: OpenAI for the replies, Apify for the scraping and the X (Twitter) API, whose cheapest paid tier was 100 USD a month when this code was written.","notVerified":["Whether `pnpm install` and `next build` still work, since the README warns that updating the `resend` and `remark-gfm` packages breaks the build","Which environment variables are actually required, because the `.env.example` the README asks you to copy is not in the repository","What the replies cost per month in OpenAI, Apify and X API fees","Whether the X, LinkedIn and Reddit parts still work against those networks' current APIs, unchanged since December 2024","Whether an account used this way keeps running or gets suspended by the network"]},"zh-Hans":{"summaryShort":"自托管网页应用,按关键词在 X、LinkedIn 和 Reddit 上找帖子,并发布 AI 写的回复来推广你的产品。","summary":"BuzzDaddy 是一个需要你自己托管的网页应用的源代码。你给它关键词,它就会从 X、LinkedIn 和 Reddit 收集匹配的帖子,再由 AI 每天按计划写回复来带出你的产品。这是一个 Next.js 项目:你需要克隆代码、安装 Node 和 pnpm、填写一份很长的环境变量文件、为 Prisma 建一个数据库,并自备 OpenAI、Apify 和各社交网络的密钥才能让它跑起来。目标用户是自己做营销的创业者。README 还是它所基于的那个付费 SaaS 模板的原文,从没解释过这些回复功能。","easeRationale":"没有东西可以下载:没有 GitHub release,也没有安装程序。Windows、Mac、Linux 上都一样:克隆仓库、安装 Node 和 pnpm、运行 `pnpm install`、复制一份环境变量文件、把 Prisma 指向一个数据库,再粘贴 OpenAI、Apify 和各社交网络的密钥,所以所有平台都是 D 级。","costNote":"代码本身免费,但依赖的都是收费服务:OpenAI 用来写回复,Apify 用来抓取数据,还有 X(Twitter)的 API——写这份代码的时候,它最便宜的付费档每月要 100 美元。","notVerified":["`pnpm install` 和 `next build` 是否还能正常运行,因为 README 警告说更新 `resend` 和 `remark-gfm` 这两个包会让构建失败","到底需要哪些环境变量,因为 README 让你复制的 `.env.example` 根本不在仓库里","按月算,OpenAI、Apify 和 X API 的费用加起来是多少","针对 X、LinkedIn 和 Reddit 的部分,自 2024 年 12 月以来没有更新,是否还能对接这些网络现在的 API","这样用的账号会不会一直正常,还是会被平台封掉"]}}},{"id":"R_kgDOVevQxw","slug":"caru-ini__stack-diagram","name":"Stack Diagram","owner":"caru-ini","fullName":"caru-ini/stack-diagram","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"attention","signals":[{"code":"logos_downloaded_from_third_party_sites_at_run_time","severity":"attention","evidence":"README 的 \"Licenses and trademarks\" 部分写道：\"Logos are not bundled with the skill. They are downloaded at build time from svgl, devicon (MIT), gilbarbara/logos (CC0), AWS Architecture Icons and Lucide (ISC), or from the brand's own site or repository.\" 来源不是一份固定的清单，下载的 SVG 文件随后会被内嵌进你分享的那个 HTML 文件里。","source":"ai","title":"Logo 是运行时从第三方网站下载的"},{"code":"offers_to_post_your_diagram_publicly","severity":"info","evidence":"README 的 \"Share it to the gallery\" 部分写道：\"After drawing, the agent offers once to post the diagram to the public gallery at https://stacks.caru.moe.\" 它说会展示哪些内容会公开，并且 \"posts only after you say yes twice\"，\"Your prompt is never sent\"，但你架构的那张图确实会被发布出去。","source":"ai","title":"会提议把你的图公开发布"},{"code":"very_new_repository","severity":"info","evidence":"该仓库创建于 2026-11-01，已经有 1221 个星标，三个版本（v0.1.0 到 v0.3.0）都在最初两天内发布。目前还没有可供参考的历史记录。","source":"ai","title":"仓库非常新"},{"code":"diagrams_contain_other_companies_logos","severity":"info","evidence":"README 写道：\"All brand names and logos are trademarks of their respective owners; their use in a diagram does not imply endorsement.\" 在把生成的图表用于营销材料之前，值得考虑这一点。","source":"ai","title":"图表中包含其他公司的 Logo"}],"platforms":["windows","mac","linux","web"],"tools":["python","node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","design_cad","development"],"alternativeTo":[{"slug":"lucidchart","name":"Lucidchart","confirmed":true},{"slug":"microsoft-visio","name":"Visio","confirmed":true},{"slug":"eraser","name":"Eraser","confirmed":true}],"declaredAlternatives":[],"stars":2245,"downloads":362,"downloadsWeek":191,"downloadsDays":7,"growth":719,"growthDays":7,"history":[{"date":"2026-11-02","stars":1201},{"date":"2026-11-03","stars":1526},{"date":"2026-11-04","stars":1736},{"date":"2026-11-05","stars":1862},{"date":"2026-11-06","stars":1937},{"date":"2026-11-07","stars":2058},{"date":"2026-11-08","stars":2121},{"date":"2026-11-09","stars":2197},{"date":"2026-11-10","stars":2245}],"firstPublishedAt":"2026-11-02T03:04:39.266Z","commitSha":"9e90a3b286baa6eef9700ca01abb80dbf0ad1514","verifiedAt":"2026-11-03T02:38:09.026Z","lastCommitAt":"2026-11-09T06:00:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill for AI assistants that turns one sentence about your tech stack into an architecture diagram with official logos.","summary":"stack-diagram is a skill you add to an AI assistant such as Claude Code, Codex or Cursor. You describe your setup in a sentence (\"Next.js and a Hono API on Vercel, Postgres on Neon, Clerk for logins\") and it draws an architecture diagram using each product's official logo, in a light and a dark version. You get a single HTML file with the logos and fonts inside, a PNG, or a React component. It needs an AI assistant already installed, Python 3.9 or later, and internet access to fetch the logos; PNG export also downloads a Chromium browser.","easeRationale":"The easiest route is the Claude web and desktop apps: you upload stack-diagram.zip from the Releases page as a skill, one step on top of an account you already have. On Windows, Mac and Linux you first need an AI coding assistant installed, then run \"npx skills add caru-ini/stack-diagram\" (which needs Node.js) and have Python 3.9 or later available, plus a Chromium download if you want PNG files.","costNote":"The skill is free (MIT). It only works inside an AI assistant, which usually needs a paid plan or paid API usage.","notVerified":["Whether the diagram it draws actually matches the setup you described","Whether every logo site is reachable and still has the logo for each product you name","What ends up inside the generated HTML file once remote logo files are inlined into it","How much your AI assistant charges in tokens for one diagram","Whether PNG export works on your machine after Playwright downloads Chromium"]},"zh-Hans":{"summaryShort":"一个给 AI 助手用的技能，把一句描述你技术栈的话变成带有官方 Logo 的架构图。","summary":"stack-diagram 是一个你添加到 Claude Code、Codex 或 Cursor 这类 AI 助手里的技能。你用一句话描述你的技术栈（比如 \"Next.js and a Hono API on Vercel, Postgres on Neon, Clerk for logins\"），它就会用每个产品的官方 Logo 绘制一张架构图，并提供浅色和深色两个版本。你会得到一个内嵌了 Logo 和字体的单一 HTML 文件、一张 PNG 图片，或者一个 React 组件。它需要你已经安装好的 AI 助手、Python 3.9 或更新版本，以及用于获取 Logo 的网络连接；导出 PNG 还会下载一个 Chromium 浏览器。","easeRationale":"最简单的方式是 Claude 的网页版和桌面版应用：从 Releases 页面上传 stack-diagram.zip 作为技能，在你已有的账号基础上多做一步就行。在 Windows、Mac 和 Linux 上，你首先需要已经安装好的 AI 编程助手，然后运行 \"npx skills add caru-ini/stack-diagram\"（需要 Node.js），并准备好 Python 3.9 或更新版本，如果想要 PNG 文件还需要下载 Chromium。","costNote":"该技能是免费的（MIT 许可）。它只能在 AI 助手内部使用，而助手通常需要付费方案或付费的 API 用量。","notVerified":["它画出的图是否真的符合你描述的配置","每个 Logo 来源网站是否都能访问，是否还保存着你提到的每个产品的 Logo","远程 Logo 文件内嵌进去之后，生成的 HTML 文件里到底包含什么内容","画一张图会消耗你 AI 助手多少 token","Playwright 下载 Chromium 之后，PNG 导出在你的电脑上是否能正常工作"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2MTM0OTcyMw==","slug":"casey__just","name":"just","owner":"casey","fullName":"casey/just","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"runs_whatever_commands_the_justfile_contains","severity":"attention","evidence":"不带参数输入 `just` 会「runs the recipe with the [default] attribute, or the first recipe in the justfile」，README 中还写道「Recipes can be written in arbitrary languages, like Python or Node.js」。因此，你下载的项目自带的 justfile 可能在你的电脑上运行任何内容，运行前最好先读一读。","source":"ai","title":"会运行 justfile 中包含的任何命令"},{"code":"curl_pipe_shell","severity":"attention","evidence":"README 提供了这种安装方式：`curl --proto '=https' --tlsv1.2 -sSf https://just.systems/install.sh | bash -s -- --to DEST`，它会在你没有先阅读的情况下运行从网上获取的脚本。包管理器和发布压缩包则避免了这一点。","source":"ai","title":"curl 管道直接执行到 shell"},{"code":"releases_include_checksums","severity":"info","evidence":"README：「Releases include a SHA256SUM file which can be used to verify the integrity of pre-built binary archives」，可用 `shasum --algorithm 256 --ignore-missing --check SHA256SUMS` 检查。验证是一个手动步骤。","source":"ai","title":"发布版本附带校验和"},{"code":"promises_not_to_break_existing_files","severity":"info","evidence":"README 的 Backwards Compatibility 部分写道：「Future releases will not introduce backwards incompatible changes that make existing justfiles stop working」，以及「There will never be a just 2.0」。仍不稳定的功能放在 `--unstable` 或 `set unstable` 之后。该项目自 2016 年以来共有 147 次发布，最新版本是 2026-10-15 发布的 1.59.0。","source":"ai","title":"承诺不会破坏现有文件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":2,"maxMB":7,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"CC0-1.0","category":"permissive","detectedFrom":"github"},"categories":["development","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":36648,"downloads":69692115,"downloadsWeek":2215095,"downloadsDays":7,"growth":101,"growthDays":7,"history":[{"date":"2026-10-26","stars":36443},{"date":"2026-10-27","stars":36459},{"date":"2026-10-28","stars":36472},{"date":"2026-10-29","stars":36479},{"date":"2026-10-30","stars":36491},{"date":"2026-10-31","stars":36494},{"date":"2026-11-01","stars":36514},{"date":"2026-11-02","stars":36531},{"date":"2026-11-03","stars":36547},{"date":"2026-11-04","stars":36568},{"date":"2026-11-05","stars":36580},{"date":"2026-11-06","stars":36596},{"date":"2026-11-07","stars":36604},{"date":"2026-11-08","stars":36611},{"date":"2026-11-09","stars":36632},{"date":"2026-11-10","stars":36648}],"firstPublishedAt":"2026-10-27T01:10:12.613Z","commitSha":"c7317c85c21ebd83ea4bb8de26f8afec44c7e597","verifiedAt":"2026-10-31T01:42:34.964Z","lastCommitAt":"2026-11-02T05:24:07Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Saves the commands a project needs in one file, so you run them by name: `just test` instead of a long line.","summary":"just is a small terminal program that keeps a project's commands in a file called `justfile`. Each entry, called a recipe, has a name and the lines to run, so instead of remembering a long command you type `just test` or `just build`. Recipes take arguments, can depend on each other, can run in parallel and can even be written in Python or another language. It is one 2 MB file for Windows, macOS and Linux, is in nearly every package manager, and the author describes it as simpler than `make`.","easeRationale":"One ready 2 MB file on all three systems, by package manager (`winget install --id Casey.Just --exact`, `brew install just`, `apt install just`) or straight from the releases page. It stays at C because there is no window: you write a `justfile` by hand and run everything from the terminal. On Windows there is an extra catch: the README says recipes need an `sh` from Git for Windows, GitHub Desktop or Cygwin, unless you add `set shell := [\"powershell.exe\", \"-c\"]`.","costNote":"Free. The licence is CC0-1.0, a public-domain dedication, so the author gives up copyright claims; the text notes trademark and patent rights are not waived.","notVerified":["Whether your Windows setup has an `sh` on the PATH, which the README requires unless you change the shell setting","Which version your package manager installs, which can lag behind 1.59.0","What a `justfile` written by somebody else actually does when you run it","Whether recipes written for Unix shells behave the same under PowerShell, which the README warns about for positional arguments"]},"zh-Hans":{"summaryShort":"把项目需要的命令保存在一个文件里，让你按名字运行它们：用 `just test` 代替一长串命令。","summary":"just 是一个小巧的终端程序，把项目的命令保存在一个叫 `justfile` 的文件里。每一条记录称为一个 recipe（配方），有名字和要执行的内容，这样你不用记住一长串命令，只需输入 `just test` 或 `just build`。配方可以带参数，可以互相依赖，可以并行运行，甚至可以用 Python 或其他语言编写。它在 Windows、macOS 和 Linux 上都只是一个 2 MB 的文件，几乎每个包管理器里都能找到它，作者将它描述为比 `make` 更简单的工具。","easeRationale":"三大系统都有现成的 2 MB 文件，可通过包管理器安装（`winget install --id Casey.Just --exact`、`brew install just`、`apt install just`），也可以直接从发布页面下载。它停留在 C 档，因为没有窗口界面：你需要手动编写 `justfile`，并在终端里运行一切。在 Windows 上还有一个额外的说明：README 提到配方需要来自 Git for Windows、GitHub Desktop 或 Cygwin 的 `sh`，除非你添加 `set shell := [\"powershell.exe\", \"-c\"]`。","costNote":"免费。许可证是 CC0-1.0，一种公共领域授权，作者据此放弃了版权主张；文本说明商标权和专利权并未放弃。","notVerified":["你的 Windows 环境的 PATH 中是否有 `sh`，README 要求必须有，除非你修改 shell 设置","你的包管理器安装的是哪个版本，可能落后于 1.59.0","别人编写的 `justfile` 在你运行时实际会做什么","为 Unix shell 编写的配方在 PowerShell 下是否表现一致，README 对位置参数这一点提出了警告"]}}},{"id":"R_kgDOSEljpA","slug":"cathrynlavery__diagram-design","name":"Diagram Design","owner":"cathrynlavery","fullName":"cathrynlavery/diagram-design","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"reads_your_website_for_branding","severity":"info","evidence":"README 的上手流程：\"onboard diagram-design to https://yoursite.com\" → \"fetches the homepage → extracts the dominant palette + font stack\"。助手会访问你指定的网站，并把颜色写入 references/style-guide.md。README 指向 PRIVACY.md，说明这个 skill 会通过网络发送哪些内容。","source":"ai","title":"会读取你的网站来提取品牌风格"},{"code":"diagrams_fetch_google_fonts","severity":"info","evidence":"README，Offline fonts：\"By default a diagram loads Instrument Serif, Geist and Geist Mono from Google Fonts, the only network request it makes.\" 如果要求使用系统字体，文件就能完全离线使用。","source":"ai","title":"图表会从 Google Fonts 加载字体"},{"code":"copycat_listings_warning","severity":"info","evidence":"README Install：\"Official builds come only from this repository. LittleMight, Cathryn Lavery's company, publishes the plugin directory listings; a listing under any other name is an unofficial copy.\"","source":"ai","title":"关于仿冒列表的提醒"},{"code":"results_vary_between_runs","severity":"info","evidence":"README，How it compares：\"The downsides are real. Output quality depends on the model you run it with, and the same prompt can lay out differently twice. The result is not an editable source format: there is no text to diff in a pull request and no canvas to drag a box on.\"","source":"ai","title":"每次运行的结果会有差异"}],"platforms":["windows","mac","linux"],"tools":["other","node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","design_cad","development"],"alternativeTo":[{"slug":"lucidchart","name":"Lucidchart","confirmed":true},{"slug":"microsoft-visio","name":"Visio","confirmed":true},{"slug":"eraser","name":"Eraser","confirmed":true},{"slug":"figma","name":"Figma","confirmed":false}],"declaredAlternatives":[],"stars":66303,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1646,"growthDays":7,"history":[{"date":"2026-10-30","stars":63261},{"date":"2026-10-31","stars":63485},{"date":"2026-11-01","stars":63997},{"date":"2026-11-02","stars":64351},{"date":"2026-11-03","stars":64657},{"date":"2026-11-04","stars":64975},{"date":"2026-11-05","stars":65272},{"date":"2026-11-06","stars":65485},{"date":"2026-11-07","stars":65660},{"date":"2026-11-08","stars":65852},{"date":"2026-11-09","stars":66068},{"date":"2026-11-10","stars":66303}],"firstPublishedAt":"2026-10-30T01:57:54.257Z","commitSha":"0cf73c3fcc56667ad91a9a04c439ac6ccfabc9d1","verifiedAt":"2026-10-30T01:43:10.704Z","lastCommitAt":"2026-11-09T22:22:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An add-on for AI assistants that turns a plain request into a tidy diagram saved as one HTML file you can open.","summary":"A skill you install inside an AI coding assistant (Claude Code, Codex, Cursor, Copilot and others). You describe the diagram in your own words and the assistant writes one self-contained HTML file with the drawing inside, in one editorial style: 45 types, from architecture and flowcharts to Gantt, Sankey and org charts. It can copy the colours and fonts from your website so diagrams match your brand, redraw existing draw.io, Mermaid or Excalidraw files, and save the result as SVG or PNG. You need an AI assistant, which usually needs a paid plan.","easeRationale":"On any of the three systems you first need an AI assistant installed, then one or two commands: `npx skills add cathrynlavery/diagram-design` (which needs Node) or, in Claude Code, `/plugin marketplace add` plus `/plugin install`. Finished diagrams are plain HTML files that open with a double-click. On Windows the README notes that symlinks need Developer Mode, otherwise the installer makes copies, and PNG export asks for an extra install (`pip install playwright && playwright install chromium`).","costNote":"The skill is free (MIT). It only works inside an AI assistant, which usually needs a paid plan or paid API use, and each diagram spends some of that allowance.","notVerified":["How good the diagrams are with your own assistant and model, since the author says output depends on both","Whether the brand onboarding really picks the right colours and fonts from a given website","What the skill and the generated files send over the network beyond the Google Fonts request","Whether every listed host (Cursor, Copilot, Qoder, Pi, Kiro and the rest) installs it as described, especially on Windows","How much of your AI plan a diagram consumes"]},"zh-Hans":{"summaryShort":"一款 AI 助手插件,能把一句简单描述变成整洁的图表,保存为单个可打开的 HTML 文件。","summary":"这是一个安装在 AI 编程助手(Claude Code、Codex、Cursor、Copilot 等)内部的技能(skill)。你用自己的话描述想要的图表,助手就会生成一个包含该图形的独立 HTML 文件,风格统一:支持 45 种类型,从架构图、流程图到甘特图、桑基图和组织结构图。它可以从你的网站中提取颜色和字体,让图表与你的品牌风格一致,也能重新绘制已有的 draw.io、Mermaid 或 Excalidraw 文件,并把结果保存为 SVG 或 PNG。你需要一个 AI 助手,而这通常需要付费计划。","easeRationale":"在三大系统上,你都需要先安装好一个 AI 助手,然后运行一两条命令:`npx skills add cathrynlavery/diagram-design`(需要 Node),或者在 Claude Code 中依次运行 `/plugin marketplace add` 和 `/plugin install`。生成的图表是普通的 HTML 文件,双击即可打开。在 Windows 上,README 提到符号链接需要开启开发者模式,否则安装程序会改为复制文件;导出 PNG 则需要额外安装(`pip install playwright && playwright install chromium`)。","costNote":"这个技能本身免费(MIT 许可证)。它只能在 AI 助手内使用,通常需要付费计划或付费的 API 用量,每生成一张图表都会消耗一部分额度。","notVerified":["用你自己的助手和模型生成的图表质量如何,因为作者表示效果取决于两者。","品牌信息导入功能是否真的能从指定网站中提取出正确的颜色和字体。","除了向 Google Fonts 发出的请求之外,这个技能以及生成的文件还会通过网络发送哪些信息。","列出的每一个宿主环境(Cursor、Copilot、Qoder、Pi、Kiro 等)是否都能按说明安装成功,尤其是在 Windows 上。","生成一张图表会消耗多少 AI 使用额度。"]}}},{"id":"R_kgDOVWlZiw","slug":"ccch1mneyyy__dsh-go","name":"dsh-go","owner":"ccch1mneyyy","fullName":"ccch1mneyyy/dsh-Go","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"8 天内获得 5957 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"installed_from_an_apk_file_outside_an_app_store","severity":"attention","evidence":"README 说 Android 构建以 APK 侧载为目标，不是 Google Play 发布配置（\"当前 Android 构建以 APK 侧载为目标，不是 Google Play 发布配置\"），所以你需要允许安装未知来源的应用，也没有商店审核或自动更新。它还提醒，之前用不同密钥签名的测试版本无法直接覆盖安装，卸载会清除对话记录、Linux 环境和设置。","source":"ai","title":"通过商店外的 APK 文件安装"},{"code":"agent_can_read_your_screen_and_tap_for_you","severity":"attention","evidence":"\"手机操控\"功能需要 Android 无障碍权限，README 的数据部分说任务所需的屏幕文字可能会被发送给模型，而能读取图片的模型可能会收到截图。它说付款、密码、验证码和系统设置始终会交还给你自己处理，\"免打扰\"版本还需要 Android 14 或更新版本，加上 Shizuku 或自行连接的 ADB。","source":"ai","title":"代理可以读取你的屏幕并代替你点击"},{"code":"usage_statistics_on_by_default","severity":"attention","evidence":"README 的数据和权限部分写道：\"『发送匿名使用统计』默认开启\"，必须在首次使用引导或在 设置 → 隐私与分享 中手动关闭。","source":"ai","title":"使用统计默认是开启的"},{"code":"conversations_shared_through_the_authors_service","severity":"info","evidence":"README 说，整段对话可以转换成一个链接，对方在浏览器中回放，仓库里也附带了对应的服务代码，位于 `services/`、`site/` 和 `infra/`。README 说这些链接采用端到端加密，密钥就包含在链接里，并提醒要检查预览内容，因为拿到链接的人可以保存里面的内容。","source":"ai","title":"对话通过作者的服务被分享出去"},{"code":"unofficial_client_not_linked_to_deepseek","severity":"info","evidence":"README 两次声明这是一个与 DeepSeek 无关的社区客户端（\"DSH Go 是社区开发的非官方客户端，与 DeepSeek 没有关联\"）。应用名称、Logo 和两个吉祥物都不在开源许可范围内（ASSETS.md），尽管代码本身采用 AGPL-3.0 许可。","source":"ai","title":"非官方客户端，与 DeepSeek 无关"},{"code":"very_new_project_that_asks_for_stars","severity":"info","evidence":"该仓库创建于 2026-10-25，已经有 5957 个星标、259 个未关闭的 issue 和 10 个发布版本。README 请求读者点星标，好让 iOS 和 HarmonyOS 版本更快到来（\"大家 Star 点得越多，我适配得越快哦\"），所以星标数量是在被主动推动，而不只是自然积累的。","source":"ai","title":"项目非常新却主动要求点星标"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["A model provider of your choice, for example DeepSeek"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["ai_agents","development","local_ai"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true}],"declaredAlternatives":[],"stars":18631,"downloads":376370,"downloadsWeek":253778,"downloadsDays":7,"growth":11204,"growthDays":7,"history":[{"date":"2026-10-31","stars":3615},{"date":"2026-11-01","stars":4744},{"date":"2026-11-02","stars":5876},{"date":"2026-11-03","stars":7427},{"date":"2026-11-04","stars":9354},{"date":"2026-11-05","stars":11360},{"date":"2026-11-06","stars":12885},{"date":"2026-11-07","stars":14268},{"date":"2026-11-08","stars":15991},{"date":"2026-11-09","stars":17514},{"date":"2026-11-10","stars":18631}],"firstPublishedAt":"2026-10-31T01:54:32.571Z","commitSha":"4ac19eef0412f8e7138e1ddd18528912edba801c","verifiedAt":"2026-11-03T02:54:00.023Z","lastCommitAt":"2026-11-10T08:23:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android app that runs the DeepSeek coding agent on your phone, or connects to the one on your computer.","summary":"DSH Go is an unofficial Android app for DeepSeek Harness, a coding agent. It installs a small Ubuntu with Node.js inside the app, so the agent runs on the phone without root and without Termux; keep about 1 GB free for it. You can also skip that and connect over SSH to the agent on your computer or server. It chats, sends pictures, takes voice input, shows tool approvals, has two cartoon mascots and, if you allow it, can read your screen and tap for you. Android 8 or newer, installed from an APK file.","easeRationale":"It is an Android app, so none of the desktop systems get a grade. On Android you download the arm64 APK from the releases page and allow the install, then on first start you either let it install its own Linux environment (about 1 GB) or point it at a computer, and you still have to set up a model and your credentials before the first message.","costNote":"The app is free (AGPL-3.0; the sdk/ and examples/ folders are MIT). Model requests go to whichever provider you set up and are billed to your account there.","notVerified":["Whether the in-app Linux environment really installs and runs on your phone model, and how much battery and storage it uses beyond the stated 1 GB.","Whether phone control stops at payments, passwords and verification codes as the README claims.","What is sent with the usage statistics before you switch them off.","That the share links are end-to-end encrypted as described, and what the sharing service keeps.","What a month of model use costs at your provider.","Most of the documentation is in Chinese; the app interface offers English, which was not checked."]},"zh-Hans":{"summaryShort":"一款安卓应用，可在手机上运行 DeepSeek 编程智能体，或连接你电脑上的那一个。","summary":"DSH Go 是一款非官方的安卓应用，用于 DeepSeek Harness 这个编程智能体。它会在应用内安装一个带 Node.js 的小型 Ubuntu，因此智能体无需 root、无需 Termux 就能在手机上运行；请预留约 1 GB 空间。你也可以跳过这一步，通过 SSH 连接电脑或服务器上的智能体。它可以聊天、发送图片、语音输入、显示工具授权，有两个卡通吉祥物，并且在你允许的情况下，可以读取你的屏幕并代你点击。需要 Android 8 或更高版本，通过 APK 文件安装。","easeRationale":"这是一款安卓应用，所以桌面系统都不评分。在安卓上，你从发布页下载 arm64 APK 并允许安装；首次启动时，要么让它安装自带的 Linux 环境（约 1 GB），要么指向一台电脑，而且在发出第一条消息之前，你仍需配置模型和你的凭据。","costNote":"该应用免费（AGPL-3.0；sdk/ 和 examples/ 文件夹为 MIT）。对模型的请求会发往你所配置的服务商，并记在你在那里的账户上。","notVerified":["应用内的 Linux 环境能否真正在你的手机型号上安装并运行，以及除了所说的 1 GB 之外还会占用多少电量和存储。","手机操控是否像 README 所说的那样，在付款、密码和验证码处停下。","在你关闭使用统计之前，它会发送哪些内容。","分享链接是否如描述的那样做了端到端加密，以及分享服务会保存什么。","在你的服务商那里，使用模型一个月要花多少钱。","大部分文档是中文；应用界面提供英文，但这一点没有核实。"]}}},{"id":"R_kgDOH3RdSQ","slug":"cemu-project__cemu","name":"Cemu","owner":"cemu-project","fullName":"cemu-project/Cemu","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"needs_game_files_you_provide","severity":"attention","evidence":"顶层文件列表只有代码和构建文件(src/、dependencies/、CMakeLists.txt、vcpkg.json),没有任何游戏数据。README.md 说 Cemu \"is able to run most Wii U games and homebrew in a playable state\",但从未说明游戏从何而来;关于设置部分只链接了一个外部页面,\"[Setup Guide](https://cemu.cfw.guide)\"。因此用户必须自行获取 Wii U 游戏文件。版权和防拷贝相关法律限制了对游戏和主机文件的复制,各国允许的范围也不一样。","source":"ai","title":"需要你自行提供游戏文件"},{"code":"ready_download_older_than_the_code","severity":"info","evidence":"检查显示最近一次提交是在 2026-11-04,但最新发行版是 2025-02-06 的 v2.6。使用现成下载版本的人拿到的代码比仓库大约旧了 21 个月。","source":"ai","title":"现成下载版本比代码更旧"},{"code":"mac_build_described_as_experimental","severity":"info","evidence":"README.md: \"The native macOS build is currently purely experimental and should not be considered stable or ready for issue-free gameplay. There are also known issues with degraded performance due to the use of MoltenVK and Rosetta for ARM Macs.\"","source":"ai","title":"mac 版本被标注为实验性"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":25,"maxMB":75,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":10141,"downloads":11700478,"downloadsWeek":20867,"downloadsDays":2,"growth":48,"growthDays":6,"history":[{"date":"2026-11-08","stars":10115},{"date":"2026-11-09","stars":10124},{"date":"2026-11-10","stars":10141}],"firstPublishedAt":"2026-11-07T23:09:40.896Z","commitSha":"3b1031612eb21302a8032e2f7a6db31549bd269b","verifiedAt":"2026-11-07T23:03:39.075Z","lastCommitAt":"2026-11-10T07:57:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs Wii U games and homebrew on a Windows, Linux or Mac computer; you add the game files yourself.","summary":"Cemu is a Wii U emulator: it makes a Windows, Linux or Mac computer behave like a Nintendo Wii U so games and homebrew can run. Ready downloads exist for the three systems, and on Windows there is also a portable version you just unzip. The repository contains no games: you supply the game files, and the README only links an outside \"Setup Guide\" for that. The README calls the macOS build \"purely experimental\". Activity: the last commit is from 2026-11-04 and the project has 95 releases, the newest being v2.6 from 2025-02-06.","easeRationale":"On Windows you download a ready installer or a portable zip (25 MB) and open it, and Linux has an AppImage and a Flathub package, so the program itself needs nothing installed; the extra step is adding your own Wii U game files, which the README does not explain, pointing to an outside Setup Guide. The Mac .dmg installs the same way, but the README says that build is \"purely experimental\" and slower on ARM Macs.","notVerified":["Whether a specific Wii U game actually runs, and how well, on each system","How usable the experimental macOS build is in practice, especially on ARM Macs","Total disk space once game files, saves and graphic packs are added (the 25-75 MB measured is only the program)","Whether the downloaded builds behave the same as the current code in the repository"]},"zh-Hans":{"summaryShort":"在 Windows、Linux 或 Mac 电脑上运行 Wii U 游戏和同人软件;游戏文件需自行添加。","summary":"Cemu 是一款 Wii U 模拟器:它让 Windows、Linux 或 Mac 电脑表现得像一台任天堂 Wii U,从而运行游戏和同人软件。三大系统都提供现成的下载版本,Windows 上还有一个解压即用的便携版。仓库本身不包含任何游戏:需要你自己提供游戏文件,README 只链接了一个外部的 \"Setup Guide\" 来说明这点。README 把 macOS 版本称为 \"purely experimental\"。活跃度:最近一次提交是 2026-11-04,项目共有 95 个发行版本,最新的是 2025-02-06 发布的 v2.6。","easeRationale":"在 Windows 上,你下载一个现成的安装程序或便携版压缩包(25 MB)并打开即可,Linux 上有 AppImage 和 Flathub 软件包,所以程序本身不需要安装其他东西;额外的一步是添加你自己的 Wii U 游戏文件,README 对此没有说明,只是指向了一个外部的 Setup Guide。Mac 的 .dmg 安装方式相同,但 README 说该版本 \"purely experimental\",在 ARM Mac 上速度更慢。","notVerified":["某个具体的 Wii U 游戏在各个系统上是否真的能运行,运行得怎么样","实验性的 macOS 版本在实际使用中好不好用,尤其是在 ARM Mac 上","加上游戏文件、存档和材质包之后的总磁盘占用(测得的 25-75 MB 只是程序本身)","下载到的版本是否和仓库里当前的代码表现一致"]}}},{"id":"R_kgDOVfKPyQ","slug":"chadyi__studyontv","name":"StudyOnTV","owner":"chadyi","fullName":"chadyi/StudyOnTV","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"license_personal_use_only","severity":"attention","evidence":"LICENSE：\"Source Available License — Personal Use Only\"，\"Copyright (c) 2024 chadyi. All rights reserved.\" 它允许出于 \"personal, non-commercial, household use only\" 的目的运行和修改代码，并明确禁止 \"redistribution of this software or any derivative work, in source or binary form, to any third party\"。尽管仓库描述里用了 \"open-source\" 这个说法，这并不是一个开源许可证。","source":"ai","title":"仅限个人使用的许可证"},{"code":"third_party_analytics_in_the_app","severity":"attention","evidence":"README.md 常见问题：\"题库和记录保存在被控设备。开启 AI 批改后，题目和答案会发送到家长配置的服务；项目中已有的第三方统计配置另见 [统计说明](./docs/analytics.md)\" —— 题库留在设备本地，但项目里已经包含了一份第三方统计（analytics）配置，记录在 docs/analytics.md 中，而这个应用是装在孩子设备上的。","source":"ai","title":"应用内的第三方统计"},{"code":"requests_accessibility_and_device_admin","severity":"attention","evidence":"README.md：规则可以打开 \"管控期间拦截系统设置\"（在锁定期间拦截系统设置），这 \"需无障碍或悬浮窗配合使用情况访问\"（需要无障碍或悬浮窗配合使用情况访问权限），并且 \"支持设备管理员的系统可另行授权防卸载\"（在支持设备管理员的系统上可以另外授权来防止卸载）。这些权限让应用能够观察并封锁整台设备上发生的事情；README 说这些权限是可选的。","source":"ai","title":"请求无障碍权限和设备管理员权限"},{"code":"http_server_on_the_home_network","severity":"info","evidence":"README.md（\"它是怎么工作的\"）：应用在 \"局域网 8099 端口\" 上运行了一个内置的 HTTP 服务，家长通过 `http://<设备IP>:8099/admin/` 访问它。这是局域网上的普通 HTTP，只靠首次使用时设置的家长密码来保护。","source":"ai","title":"家庭网络上的 http 服务"},{"code":"download_mirror_outside_github","severity":"info","evidence":"README.md 为无法访问 GitHub 的用户提供了第三方站点的备用下载：\"手机 / 平板版 · 蒲公英备用下载\"（pgyer.com/StudyOnTVMobile），电视版在 pgyer.com/StudyOnTVTV。那里提供的文件并不是 GitHub release 里的资源。","source":"ai","title":"GitHub 之外的下载镜像"},{"code":"no_binaries_detected_in_releases","severity":"info","evidence":"最近的三个发行版（v0.3.21、v0.3.20、v0.3.19，都是 2026-11-07 的）在 Radar 的检查里显示 \"no platform binaries\"，因为这些文件是 Android APK，而不是桌面安装包。该仓库创建于 2026-11-01，已经有 25 个发行版，说明这款应用变化非常快。","source":"ai","title":"发行版中未检测到二进制文件"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["DeepSeek or another compatible AI service (Qwen, Zhipu, Kimi, SiliconFlow, local Ollama)"],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["productivity","hardware"],"alternativeTo":[{"slug":"qustodio","name":"Qustodio","confirmed":true},{"slug":"quizlet","name":"Quizlet","confirmed":true}],"declaredAlternatives":[],"stars":1320,"downloads":31178,"downloadsWeek":3501,"downloadsDays":1,"growth":626,"growthDays":4,"history":[{"date":"2026-11-09","stars":1155},{"date":"2026-11-10","stars":1320}],"firstPublishedAt":"2026-11-08T23:01:18.701Z","commitSha":"f76f3d5b8907f1ae28a68d4764599bb3ef51185d","verifiedAt":"2026-11-08T22:47:08.312Z","lastCommitAt":"2026-11-10T14:26:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Android TV, tablet and phone app where a child answers study questions to unlock screen time; parents manage it over Wi-Fi.","summary":"StudyOnTV (题了个题) locks an Android TV, tablet or phone until the child answers a set of study questions; answering right unlocks a few minutes of screen time, with daily limits and time windows. Parents scan a QR code and open a management page served by the device itself on the home Wi-Fi, where they set the rules, edit the question bank (50 questions are built in, more can be imported as JSON) and read seven days of history. Everything is stored on the device in SQLite, with no account and no cloud. Android 5.0 or later.","easeRationale":"It is an Android-only app, so none of the desktop platforms get a grade. On a phone or tablet you install the APK from the releases page and set a parent password on first start; on a TV you have to copy the APK over with a USB stick or ADB and then grant permissions such as overlay or accessibility for the locking to hold, which is the harder part.","costNote":"Free for personal family use only; the license forbids commercial use and redistribution. Optional AI grading uses an outside service the parent pays for.","notVerified":["What the third-party statistics configuration in the project actually sends, and from which device","Whether the locking really holds on a given TV or phone, since the README says vendor background limits and standby behaviour must be checked on the real device","Whether a child can get around it by force-stopping the app, revoking permissions or uninstalling it","What is sent to the AI grading service when a parent turns it on, including photos of handwritten answers, and what it costs","Whether the question bank and rules survive an app update or a device change beyond the manual JSON export"]},"zh-Hans":{"summaryShort":"一款 Android TV、平板和手机应用，孩子要回答学习问题才能解锁屏幕时间；家长通过 Wi-Fi 管理。","summary":"StudyOnTV（题了个题）会锁定 Android TV、平板或手机，直到孩子回答一组学习问题；答对才能解锁几分钟的屏幕时间，并带有每日上限和时间段限制。家长扫描二维码，打开设备自身在家庭 Wi-Fi 上提供的管理页面，在那里设置规则、编辑题库（内置 50 道题，还可以用 JSON 导入更多）并查看七天的历史记录。所有数据都以 SQLite 形式保存在设备本地，不需要账号，也不上云。支持 Android 5.0 及以上。","easeRationale":"这是一款仅支持 Android 的应用，所以桌面平台都没有评分。在手机或平板上，你从 releases 页面安装 APK，并在首次启动时设置家长密码；在电视上，你需要用 USB 存储设备或 ADB 把 APK 拷过去，然后授予悬浮窗或无障碍之类的权限才能让锁定生效，这是比较麻烦的部分。","costNote":"仅限个人家庭使用免费；许可证禁止商业使用和再分发。可选的 AI 判卷功能使用的是家长自付费用的外部服务。","notVerified":["项目里的第三方统计配置到底发送了什么内容，又是从哪台设备发出的","锁定在某台具体的电视或手机上能否真正生效，因为 README 说厂商的后台限制和待机行为需要在真实设备上验证","孩子能否通过强行停止应用、撤销权限或卸载应用来绕过限制","家长打开 AI 判卷功能后，会发送给判卷服务什么内容，包括手写答案的照片，以及这要花多少钱","题库和规则在应用更新或更换设备之后，除了手动的 JSON 导出之外能否保留下来"]}}},{"id":"R_kgDOT5nSrw","slug":"charanmn7__ig-comment-dm","name":"ig-comment-dm","owner":"CharanMN7","fullName":"CharanMN7/ig-comment-dm","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"ai_agent_setup_with_your_logins","severity":"info","evidence":"README 提供了另一种配置方式,让 AI 助手代劳:\"open the cloned repo in Claude Code and say 'set this up'. A bundled skill runs the terminal side and, with Claude in Chrome, clicks through Meta while you keep the logins and secrets.\"。这意味着让一个 AI 工具在你的 Meta 和 Cloudflare 账号里操作(仓库自带 .claude/、CLAUDE.md 和 AGENTS.md)。","source":"ai","title":"ai 助手用你的登录信息来配置"},{"code":"depends_on_instagram_rules_and_caps","severity":"info","evidence":"README 的 \"What can go wrong\" 一节说,这个工具只对专业版 Instagram 账号有效,允许 \"One private reply per comment, ever, within 7 days\",被 Meta 限制在 \"About 750 private replies per hour per account\",并固定使用 Graph API v23.0,\"which Meta will eventually retire\"。Instagram 随时可能更改或切断这个接口权限。","source":"ai","title":"依赖 instagram 的规则和限额"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta for Developers app (Instagram Graph API)","Cloudflare account (Workers and D1)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[],"stars":7,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":7},{"date":"2026-11-09","stars":7},{"date":"2026-11-10","stars":7}],"firstPublishedAt":"2026-11-07T22:56:16.713Z","commitSha":"e4feebf89dde918ec6412514afe348e2abdcb521","verifiedAt":"2026-11-07T22:37:49.272Z","lastCommitAt":"2026-10-27T13:46:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Sends an Instagram DM with your link when someone comments a keyword, running on your own free Cloudflare account.","summary":"A comment-to-DM tool for Instagram creators and small shops. You write keyword rules in a small admin page, and when someone comments the keyword on your post the tool sends them a private message with your link, plus an optional public reply. It runs on your own Cloudflare account using Meta's official Instagram API, so no password or scraping is involved. To use it you need Node.js 22.13 or newer, free Cloudflare and Meta developer accounts, an Instagram Creator or Business account, about five terminal commands and around 15 minutes in Meta's dashboard.","easeRationale":"There is no installer: on Windows, macOS and Linux alike you install Node.js 22.13+, clone the repository and run about five commands (npm install, wrangler login, npm run setup), which is a C. The commands are the same everywhere, but the README adds roughly 15 minutes of clicking in Meta's developer dashboard, creating an app and connecting your Instagram account as a tester.","costNote":"The README says $0 a month: Meta does not charge for the API and the project is sized to fit Cloudflare's free plan, whose limits the README lists (for example 100,000 Worker requests and 100,000 database row writes a day).","notVerified":["Whether the setup script really finishes without errors on each operating system, and whether Meta's dashboard still looks like the guide describes.","Whether DMs are actually delivered, and how many of them Instagram silently drops or sends to Message requests.","Whether a busy account stays inside Cloudflare's free-plan limits, or starts failing after the daily database write cap.","Whether the token encryption and the nightly token refresh work as described, since that can only be seen in a live deployment."]},"zh-Hans":{"summaryShort":"有人在评论里打出关键词时,用你自己的免费 Cloudflare 账号给对方发一条带链接的 Instagram 私信。","summary":"一款面向 Instagram 创作者和小店主的「评论转私信」工具。你在一个小型管理页面里写好关键词规则,当有人在你的帖子下评论该关键词时,工具会给对方发一条带你链接的私信,还可以选择附带一条公开回复。它运行在你自己的 Cloudflare 账号上,使用 Meta 官方的 Instagram API,因此不涉及密码或抓取数据。使用前你需要 Node.js 22.13 或更高版本、免费的 Cloudflare 和 Meta 开发者账号、一个 Instagram 创作者或企业账号、大约五条终端命令,以及在 Meta 后台花费约 15 分钟。","easeRationale":"没有安装程序:在 Windows、macOS 和 Linux 上都是先安装 Node.js 22.13+,克隆仓库,再运行大约五条命令(npm install、wrangler login、npm run setup),属于 C 级。这些命令到处都一样,但 README 还要求在 Meta 开发者后台再花大约 15 分钟,创建一个应用并把你的 Instagram 账号添加为测试者。","costNote":"README 说每月 0 美元:Meta 不对这个 API 收费,项目也是按 Cloudflare 免费套餐的限额设计的,README 列出了这些限额(例如每天 100,000 次 Worker 请求和 100,000 次数据库行写入)。","notVerified":["配置脚本在各个操作系统上是否真的能顺利跑完不出错,以及 Meta 后台现在是否还和指南描述的一样","私信是否真的能送达,其中有多少被 Instagram 悄悄丢弃,或者被转进「消息请求」","一个活跃账号是否会保持在 Cloudflare 免费套餐限额以内,还是在触及每日数据库写入上限后开始失败","令牌加密和夜间令牌刷新是否像说明的那样运作,因为这一点只有在实际部署中才能看到"]}}},{"id":"R_kgDOJF_VYg","slug":"chatboxai__chatbox","name":"Chatbox","owner":"chatboxai","fullName":"chatboxai/chatbox","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B","web":"B"},"risk":"attention","signals":[{"code":"this_is_the_community_edition_of_a_paid_product","severity":"attention","evidence":"README：“这是 Chatbox 社区版的仓库，以 GPLv3 许可证开源”，以及“我们会定期把代码从专业版仓库同步到这个仓库，反过来也一样。”付费版本在一个你看不到的地方开发，所以两者随时可能出现差异。","source":"ai","title":"这是一个付费产品的社区版"},{"code":"download_links_point_to_the_company_site","severity":"info","evidence":"README 的下载按钮链接到 https://chatboxai.app/?c=download-windows，以及 macOS 和 Linux 对应的地址，而这里列出的三个最新发行版（v1.23.5、v1.23.3 和 v1.23.2）都没有附带任何文件。所以你安装的东西来自公司自己的网站，而不是在这个仓库里公开构建出来的文件。","source":"ai","title":"下载链接指向公司自己的网站"},{"code":"error_reporting_built_in","severity":"info","evidence":"package.json 把 @sentry/react 列为依赖之一，所以这个程序自带一个错误报告服务，会把崩溃数据发给开发者。仓库内容没有说明这个功能能否关闭。","source":"ai","title":"内置错误报告"},{"code":"your_conversations_go_to_the_provider_you_choose","severity":"info","evidence":"README 里“本地数据存储”这一条说你的数据留在设备上，这指的是保存下来的历史记录：你问的任何内容仍然会发送给你输入密钥的那个提供方（OpenAI、Claude、Gemini 等），并按那家公司的条款处理。只有使用 Ollama 这条路径才会把文字留在你自己的机器上。","source":"ai","title":"你的对话会发送给你选择的提供方"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["node"],"disk":{"minMB":195,"maxMB":584,"basis":"installer"},"apiKeys":["A model provider of your choice — OpenAI, Azure OpenAI, Claude, Google Gemini, among others","None if you use Ollama with local models instead"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["local_ai","productivity"],"alternativeTo":[{"slug":"typingmind","name":"TypingMind","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[],"stars":42294,"downloads":518840,"downloadsWeek":0,"downloadsDays":6,"growth":47,"growthDays":7,"history":[{"date":"2026-11-04","stars":42253},{"date":"2026-11-05","stars":42256},{"date":"2026-11-06","stars":42258},{"date":"2026-11-07","stars":42269},{"date":"2026-11-08","stars":42278},{"date":"2026-11-09","stars":42290},{"date":"2026-11-10","stars":42294}],"firstPublishedAt":"2026-11-04T01:41:48.268Z","commitSha":"d17455c5ca8fd3ff44385c6d4d2118599e4c970f","verifiedAt":"2026-11-04T01:25:27.989Z","lastCommitAt":"2026-11-10T16:11:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop chat window for ChatGPT, Claude, Gemini or a model running on your own computer with Ollama.","summary":"Chatbox is a program you install on Windows, macOS or Linux (there are also phone apps and a web version) that gives you one chat window for many AI models. You paste the key of the service you want to use — OpenAI, Claude, Gemini, Azure and others — or point it at Ollama to talk to a model running on your own machine. It keeps the conversations on your device, writes code and maths nicely, saves prompts and can make images with DALL-E 3. This repository is the free Community Edition, under GPL-3.0.","easeRationale":"On Windows there is a Setup.exe, on macOS an Intel and an Apple Silicon build, and on Linux an AppImage, so there is nothing to compile; the web version runs in a browser. It is a B rather than an A everywhere because, as the README's Quick Start says, after installing you still have to \"Configure your AI provider (OpenAI, Claude, etc.) in settings\", which means pasting in a key or installing Ollama first.","costNote":"The Community Edition in this repository is free software (GPL-3.0) and the installers cost nothing, but you pay the AI provider whose key you paste in. The exception is Ollama: local models cost nothing to use. The repository does not state the prices of the company's own Chatbox AI service.","notVerified":["Whether the installers offered at chatboxai.app are built from this Community Edition code","What the company's paid Chatbox AI plans cost and which features they add, since the repository does not say","What the built-in error reporting sends and whether it can be turned off","How much you will spend at each AI provider, which depends on how much you use it","Whether the app really works on each system and version listed, and the real disk use after installing"]},"zh-Hans":{"summaryShort":"一个桌面聊天窗口，支持 ChatGPT、Claude、Gemini，或者用 Ollama 在你自己电脑上运行的模型。","summary":"Chatbox 是一个你安装在 Windows、macOS 或 Linux 上的程序（也有手机应用和网页版），给你一个窗口就能和很多 AI 模型聊天。你粘贴想用的服务的密钥——OpenAI、Claude、Gemini、Azure 等等——或者把它指向 Ollama，和运行在你自己机器上的模型对话。它把对话保存在你的设备上，代码和数学公式排版美观，能保存提示词，还能用 DALL-E 3 生成图片。这个仓库是免费的社区版，使用 GPL-3.0 许可证。","easeRationale":"Windows 上有一个 Setup.exe，macOS 上有 Intel 和 Apple Silicon 两个版本，Linux 上有一个 AppImage，所以不需要自己编译；网页版在浏览器里就能运行。它到处都评为 B 而不是 A，因为正如 README 的快速上手部分所说，安装之后你仍然需要“在设置里配置你的 AI 提供方（OpenAI、Claude 等）”，也就是说要先粘贴一个密钥或者先安装 Ollama。","costNote":"这个仓库里的社区版是自由软件（GPL-3.0），安装程序本身不收费，但你要向你粘贴密钥的那个 AI 提供方付费。例外是 Ollama：本地模型的使用不花钱。仓库里没有说明公司自己的 Chatbox AI 服务的价格。","notVerified":["chatboxai.app 上提供的安装程序是否是用这份社区版代码构建的","该公司自己的 Chatbox AI 付费方案要多少钱，增加了哪些功能，因为仓库里没有说明","内置的错误报告发送了什么内容，以及能否关闭它","你在每个 AI 提供方那里会花多少钱，这取决于你用得多不多","这个应用是否真的能在列出的每个系统和版本上运行，以及安装后实际占用多少磁盘空间"]}}},{"id":"R_kgDOJQ_nbA","slug":"chatchat-space__langchain-chatchat","name":"Langchain-Chatchat","owner":"chatchat-space","fullName":"chatchat-space/Langchain-Chatchat","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最新发行版发布于 2024-07-12（距今 844 天）","source":"check","title":"发行版陈旧"},{"code":"web_page_has_no_login_if_you_open_it_up","severity":"attention","evidence":"README 提醒默认监听地址 `DEFAULT_BIND_HOST` 是 127.0.0.1，\"所以无法通过其他 ip 进行访问\"，并告诉你要在 `basic_settings.yaml` 里把它改成 0.0.0.0 才能从别的机器访问。内容里没有任何地方提到密码或用户账号，所以任何能访问到那个地址的人，都可能读取和修改知识库。","source":"ai","title":"如果对外开放，网页没有登录保护"},{"code":"needs_an_older_python","severity":"info","evidence":"pyproject.toml 固定了 `python = \">=3.8.1,<3.12,!=3.9.7\"`，所以它无法在 Python 3.12 或更新版本上安装，而这正是当前系统自带的版本。README 列出的是 Python 3.8–3.11。","source":"ai","title":"需要较旧版本的 python"},{"code":"no_movement_for_about_a_year","severity":"info","evidence":"最后一次提交是在 2025-11-10，距本次检查 358 天，README 里仍然写着 \"🐳 Docker 镜像将会在近期更新\"，旁边的镜像标签却停留在 2024-08-29。里程碑列表止于 2024 年 6 月。","source":"ai","title":"大约一年没有任何进展"},{"code":"license_stated_twice_and_differently","severity":"info","evidence":"GitHub 和 LICENSE 文件都写的是 Apache-2.0，README 也重复了这一点，但 pyproject.toml 却声明 `license = \"MIT\"`。两者都是宽松许可证，所以这个差异主要影响想要复用代码的人。","source":"ai","title":"许可证被两次声明，而且不一致"},{"code":"dependencies_fetched_from_a_mirror_by_default","severity":"info","evidence":"pyproject.toml 设置了一个名为 \"tsinghua\" 的包源，地址是 `https://pypi.tuna.tsinghua.edu.cn/simple/`，并设为 `priority = \"primary\"`，所以任何从源码树安装的人，都会从中国的这个镜像站拉取 Python 库，而不是从 PyPI。","source":"ai","title":"依赖默认从一个镜像站获取"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["Online model services such as OpenAI, Azure OpenAI, Claude or Zhipu — only if you choose them instead of local models"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["local_ai","office","ai_agents"],"alternativeTo":[{"slug":"notebooklm","name":"NotebookLM","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[],"stars":38714,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":2,"growthDays":7,"history":[{"date":"2026-11-04","stars":38715},{"date":"2026-11-05","stars":38713},{"date":"2026-11-06","stars":38713},{"date":"2026-11-07","stars":38716},{"date":"2026-11-08","stars":38714},{"date":"2026-11-09","stars":38713},{"date":"2026-11-10","stars":38714}],"firstPublishedAt":"2026-11-04T01:41:48.277Z","commitSha":"49165d6af4438aa7e8a1f71ce276db55f4405151","verifiedAt":"2026-11-04T01:26:27.124Z","lastCommitAt":"2025-11-10T09:27:42Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ask questions about your own documents and get answers from an AI model running on your machine.","summary":"Langchain-Chatchat builds a question-and-answer service over your own files. You drop documents into a knowledge base, it turns them into a searchable index, and the answers are written by a language model you choose. It can run entirely offline: the models are loaded by a separate program (Ollama, Xinference, LocalAI or FastChat), and online services such as OpenAI or Claude can be plugged in instead. It is installed with pip or Docker and used in a browser page on your own computer. The documentation is mainly in Chinese.","easeRationale":"There is no file to download in any of the 31 releases; the work is the same on Windows, macOS and Linux. You need Python, then `pip install langchain-chatchat`, then a second program to serve the models which the README asks you to install in a separate virtual environment, then `chatchat init`, hand edits in `model_settings.yaml`, `chatchat kb -r` to index the knowledge base and `chatchat start -a` to open it. Docker images exist as an alternative, which is still terminal work.","costNote":"Apache-2.0 code, and free to run end to end with open models on your own machine. It can also be pointed at paid online services (OpenAI, Azure, Claude, Zhipu, Baichuan), which you then pay for.","notVerified":["Whether the install still works today on a supported Python version, after a year without commits","How well it answers in languages other than Chinese","How much disk and memory the whole thing takes once the model framework and the models are added, since the figure here covers only the project's own files","Whether the Docker images named in the README are still available","What an online provider would cost if you use one instead of local models"]},"zh-Hans":{"summaryShort":"就你自己的文档提问，由运行在你电脑上的 AI 模型给出答案。","summary":"Langchain-Chatchat 基于你自己的文件搭建一个问答服务。你把文档放进一个知识库，它会把文档转换成一个可检索的索引，答案由你选择的语言模型来写。它可以完全离线运行：模型由一个独立的程序（Ollama、Xinference、LocalAI 或 FastChat）加载，也可以换成接入 OpenAI 或 Claude 之类的在线服务。它通过 pip 或 Docker 安装，在你自己电脑上的浏览器页面里使用。文档主要是中文。","easeRationale":"31 个发行版里都没有可下载的文件；在 Windows、macOS 和 Linux 上要做的工作是一样的。你需要先装 Python，再执行 `pip install langchain-chatchat`，然后 README 要求你在一个单独的虚拟环境里安装第二个程序来提供模型服务，接着是 `chatchat init`、手动编辑 `model_settings.yaml`、用 `chatchat kb -r` 为知识库建立索引，以及用 `chatchat start -a` 打开它。也有 Docker 镜像可以作为替代方案，但那仍然是终端操作。","costNote":"代码采用 Apache-2.0 许可证，在你自己的机器上用开放模型从头跑到尾是免费的。它也可以接入付费的在线服务（OpenAI、Azure、Claude、智谱、百川），接入后由你来付费。","notVerified":["在停滞一年没有提交之后，安装流程今天在受支持的 Python 版本上是否还能用","除了中文以外的语言，它回答得怎么样","加上模型框架和模型本身之后，整套东西占用多少磁盘和内存，因为这里给出的数字只覆盖了项目自己的文件","README 里提到的 Docker 镜像是否还能用","如果改用在线服务而不是本地模型，大概要花多少钱"]}}},{"id":"R_kgDOJH-jmg","slug":"chatgptnextweb__nextchat","name":"NextChat","owner":"ChatGPTNextWeb","fullName":"ChatGPTNextWeb/NextChat","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B","web":"B"},"risk":"attention","signals":[{"code":"install_script_piped_from_the_internet","severity":"attention","evidence":"README 里的 Shell 安装方式是 `bash <(curl -s https://raw.githubusercontent.com/Yidadaa/ChatGPT-Next-Web/main/scripts/setup.sh)`，它会直接运行当下载下来的脚本，你没有机会先读一遍。","source":"ai","title":"安装脚本通过管道直接从网上执行"},{"code":"public_copy_without_a_password_spends_your_credit","severity":"attention","evidence":"README 说“这个项目提供的访问控制有限”，而 .env.template 里把 `CODE` 密码标记为可选，`OPENAI_API_KEY` 却是必填的。一个部署到 Vercel 或 Docker 上、填了密钥但没设 `CODE` 的副本，会让任何找到这个地址的人用你的额度聊天。`HIDE_USER_API_KEY` 和 `DISABLE_GPT4` 这两个选项也是因为同样的原因而存在的。","source":"ai","title":"没有密码的公开副本会消耗你的额度"},{"code":"documentation_routes_your_key_through_a_third_party_proxy","severity":"attention","evidence":"开发部分告诉那些连不上 OpenAI 的人把 `BASE_URL` 设成 `https://chatgpt1.nextweb.fun/api/proxy`，.env.template 也建议用 `BASE_URL`/`PROXY_URL` 来覆盖设置。按这种方式配置后，包括 API 密钥和你对话的文字内容，都会经过那台外部服务器。","source":"ai","title":"文档让你的密钥经过第三方代理"},{"code":"dependencies_not_all_from_the_normal_sources","severity":"info","evidence":"package.json 里有一个依赖直接从一个网址获取：`\"rt-client\": \"https://github.com/Azure-Samples/aoai-realtime-audio-sdk/releases/download/js/v0.5.0/rt-client-0.5.0.tgz\"`，Dockerfile 还设置了 `yarn config set registry 'https://registry.npmmirror.com/'`，所以构建时会从中国的一个镜像站获取这些库，而不是从 npm。","source":"ai","title":"依赖并非全部来自常规来源"},{"code":"desktop_files_and_releases_are_old","severity":"info","evidence":"最新的发行版是 2025-07-29 的 v2.16.1，没有附带任何文件；上一个带桌面安装程序的是 2024-11-11 的 v2.15.8。提交仍在持续（最近一次是 14 天前），所以你下载到的东西远远落后于代码。现成的 Docker 镜像是在一个个人账号 `yidadaa/chatgpt-next-web` 下发布的，下载链接也仍然指向旧仓库 `Yidadaa/ChatGPT-Next-Web`。","source":"ai","title":"桌面文件和发行版都很旧"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":7,"maxMB":20,"basis":"installer"},"apiKeys":["OpenAI (marked \"required\" in .env.template)","Anthropic (Claude), Google Gemini, DeepSeek or Azure OpenAI — as an alternative to OpenAI","Many others, each optional: xAI, Moonshot, SiliconFlow, 302.AI, Stability, Baidu, ByteDance, Alibaba, Tencent, iFlytek, ChatGLM","No key at all if you point it at your own model server, such as LocalAI or RWKV-Runner"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["productivity"],"alternativeTo":[{"slug":"typingmind","name":"TypingMind","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[],"stars":88948,"downloads":703767,"downloadsWeek":0,"downloadsDays":6,"growth":11,"growthDays":7,"history":[{"date":"2026-11-04","stars":88941},{"date":"2026-11-05","stars":88942},{"date":"2026-11-06","stars":88946},{"date":"2026-11-07","stars":88945},{"date":"2026-11-08","stars":88941},{"date":"2026-11-09","stars":88941},{"date":"2026-11-10","stars":88948}],"firstPublishedAt":"2026-11-04T01:41:48.285Z","commitSha":"1d3dc331512929a4b1d3e82354e72acce33b5c7a","verifiedAt":"2026-11-04T01:27:31.695Z","lastCommitAt":"2026-10-20T10:19:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A light chat app for GPT, Claude, Gemini and DeepSeek, using your own key, on the desktop or on your own site.","summary":"NextChat is a small chat program (about 5 MB on the desktop) that talks to many AI services with a key you provide: OpenAI, Claude, Gemini, DeepSeek, Azure and a long list of Chinese providers, or a model server you run yourself. It saves prompt templates, shows code, maths and diagrams, works offline as a web app and keeps the conversations in your browser. You can also publish your own copy with one click on Vercel or run it with Docker, in which case set the access password.","easeRationale":"Release v2.15.8 carries ready-made files for every desktop system (NextChat_2.15.8_x64-setup.exe and an .msi for Windows, a universal .dmg for macOS, an AppImage and a .deb for Linux), and there is a web app you can just open, so nothing has to be compiled. It is a B and not an A because the app does nothing until you paste in the key of an AI service, or point it at a model server of your own.","costNote":"The code is MIT and the program costs nothing, but you pay the AI service whose key you paste in; a self-deployed model server is the free route. The same team also sells a ready-made service at nextchat.club, an iOS app and an \"Enterprise Edition\" (contact business@nextchat.club); the repository gives no prices.","notVerified":["Whether the 2024 desktop installers still work with the current AI services","What the paid service at nextchat.club and the Enterprise Edition cost, and whether they run this same code","How much you will spend at the AI provider you connect","What the chat-log sync feature sends to the WebDAV or Upstash address you configure","Whether the MCP feature works, which needs the program rebuilt with ENABLE_MCP=true"]},"zh-Hans":{"summaryShort":"一个轻量聊天应用，支持 GPT、Claude、Gemini 和 DeepSeek，用你自己的密钥，可在桌面端或你自己的网站上使用。","summary":"NextChat 是一个小巧的聊天程序（桌面版大约 5 MB），用你提供的密钥和很多 AI 服务对话：OpenAI、Claude、Gemini、DeepSeek、Azure，以及一长串中国的服务商，或者你自己运行的模型服务器。它能保存提示词模板，展示代码、数学公式和图表，作为网页应用也能离线使用，并把对话保存在你的浏览器里。你还可以一键在 Vercel 上发布自己的一份，或者用 Docker 运行，这种情况下要设置访问密码。","easeRationale":"v2.15.8 这个发行版给每个桌面系统都带来了现成文件（Windows 的 NextChat_2.15.8_x64-setup.exe 和一个 .msi，macOS 的通用 .dmg，Linux 的 AppImage 和 .deb），还有一个网页版可以直接打开，所以不需要自己编译任何东西。它评为 B 而不是 A，因为在你粘贴一个 AI 服务的密钥，或者把它指向你自己的模型服务器之前，这个应用什么都做不了。","costNote":"代码是 MIT 许可，程序本身不收费，但你要向你粘贴密钥的那个 AI 服务付费；自己部署模型服务器是免费的路线。同一个团队还在 nextchat.club 出售一个现成的服务、一个 iOS 应用，以及一个“企业版”（联系 business@nextchat.club）；仓库里没有给出价格。","notVerified":["2024 年的桌面安装程序是否还能配合现在的 AI 服务使用","nextchat.club 的付费服务和企业版要多少钱，它们是否运行着这同一份代码","你连接的那个 AI 提供方会花掉你多少钱","聊天记录同步功能会把什么内容发送到你配置的 WebDAV 或 Upstash 地址","MCP 功能是否真的能用，因为它需要用 ENABLE_MCP=true 重新编译这个程序"]}}},{"id":"R_kgDOVKRzDQ","slug":"chenglou__handwired","name":"Handwired","owner":"chenglou","fullName":"chenglou/handwired","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"low","signals":[{"code":"experiment_not_a_finished_product","severity":"info","evidence":"package.json 里有 \"private\": true，没有发布版本或公开的软件包；README 说这个转换是 \"for fun\" 做的，把各种想法写成 \"what does that even do? Let's find out!\"，并把 lab/ 称为 \"a bunch of old experiments you can check or ignore\"。","source":"ai","title":"实验项目，不是成品"},{"code":"new_repo_fast_stars","severity":"info","evidence":"创建于 2026-10-15，已有 2090 个星标，最后一次提交在 2026-10-17 —— 大部分关注是在工作停止之后才到来的。","source":"ai","title":"新仓库，星标增长很快"},{"code":"gpu_demos_need_a_webgpu_browser","severity":"info","evidence":"README：\"We then collect the + and * and relu into a giant matmul, and either run it on JS or WebGPU\" —— 走显卡的路径只在支持 WebGPU 的浏览器中有效。","source":"ai","title":"显卡演示需要支持 WebGPU 的浏览器"},{"code":"very_new_build_tools","severity":"info","evidence":"package.json 固定了 \"typescript\": \"^7.0.2\"，外加第二份 \"typescript6\"，还有 \"oxlint\": \"latest\"，并且只把 Bun 作为唯一的运行器，所以本地环境依赖更新很快的工具。","source":"ai","title":"构建工具非常新"}],"platforms":["web","windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":2142,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":35,"growthDays":7,"history":[{"date":"2026-10-27","stars":2029},{"date":"2026-10-28","stars":2055},{"date":"2026-10-29","stars":2064},{"date":"2026-10-30","stars":2075},{"date":"2026-10-31","stars":2087},{"date":"2026-11-01","stars":2090},{"date":"2026-11-02","stars":2095},{"date":"2026-11-03","stars":2107},{"date":"2026-11-04","stars":2117},{"date":"2026-11-05","stars":2123},{"date":"2026-11-06","stars":2125},{"date":"2026-11-07","stars":2131},{"date":"2026-11-08","stars":2133},{"date":"2026-11-09","stars":2138},{"date":"2026-11-10","stars":2142}],"firstPublishedAt":"2026-10-27T18:40:27.813Z","commitSha":"257856e3b221140acb4cc4460626b850c0bd4ff1","verifiedAt":"2026-11-02T02:54:28.826Z","lastCommitAt":"2026-10-17T00:33:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Experiment that rewrites ordinary JavaScript into the maths a neural net uses, with live demos.","summary":"Handwired is an experiment for programmers. It takes ordinary JavaScript (if-else, loops, state) and rewrites it as the kind of multiplication-and-addition a neural net is made of, so the same program can run on the graphics card, be trained, or be bolted onto a real neural net. No training is needed because the numbers are written by hand. You can try the demos in a browser at chenglou.me/handwired; to use the code you clone the repository and install Bun. It is a personal project with no published package or release.","easeRationale":"The demos are already online at chenglou.me/handwired, so on the web you only open the page. To run or use the code on Windows, macOS or Linux you install Bun, clone the repository and run \"bun install\" and \"bun start\" in a terminal.","notVerified":["Whether the demos run in your browser, which depends on WebGPU support.","Whether the speed claim of searching 100k documents in milliseconds holds on your machine.","How much real JavaScript the transform handles beyond the examples in docs/ and template/.","Whether the project is still being worked on, since the last commit is from 2026-10-17 and there are no releases."]},"zh-Hans":{"summaryShort":"一个实验项目，把普通的 JavaScript 改写成神经网络所用的数学运算，并附有在线演示。","summary":"Handwired 是面向程序员的实验项目。它把普通的 JavaScript（if-else、循环、状态）改写成神经网络所由构成的那种乘法和加法，因此同一段程序可以在显卡上运行、被训练，或接到真正的神经网络上。因为数字是手写的，所以不需要训练。你可以在浏览器中试用演示：chenglou.me/handwired；要使用代码，则需克隆仓库并安装 Bun。这是个人项目，没有发布的软件包或版本。","easeRationale":"演示已经在 chenglou.me/handwired 上线，所以在网页上你只需打开页面。要在 Windows、macOS 或 Linux 上运行或使用代码，你需要安装 Bun、克隆仓库，并在终端里运行 \"bun install\" 和 \"bun start\"。","notVerified":["演示能否在你的浏览器中运行，这取决于浏览器是否支持 WebGPU。","在毫秒内搜索 100k 份文档的速度说法，在你的电脑上是否成立。","除了 docs/ 和 template/ 里的示例之外，这种转换能处理多少真实的 JavaScript。","项目是否仍在维护，因为最后一次提交是 2026-10-17，而且没有发布版本。"]}}},{"id":"R_kgDOL_2xwg","slug":"cherryhq__cherry-studio","name":"Cherry Studio","owner":"CherryHQ","fullName":"CherryHQ/cherry-studio","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"被锁定版本的依赖存在已知漏洞（OSV）：sharp：GHSA-rgj7-g3m4-5g8c、GHSA-wq5f-xc86-pv6w；systeminformation：GHSA-3f7j-7jwc-fxfj、GHSA-4vjg-vx72-qq4g、GHSA-7648-jmwf-vw7j","source":"check","title":"存在已知漏洞"},{"code":"agents_can_run_commands_on_your_computer","severity":"attention","evidence":"README“Key Features”部分：智能体“Work with files, run commands, and use browser and computer automation tools”，可以用 Skills 和 MCP 服务器扩展，还能“Run scheduled tasks and connect agents to messaging platforms through Channels”——这个应用可以在你的电脑上自主行动。","source":"ai","title":"智能体可以在你的电脑上运行命令"},{"code":"copyleft_license_agpl","severity":"info","evidence":"LICENSE 是 AGPL-3.0。README 补充说，只要“subject to full compliance with the terms”就允许商业使用，如果想要豁免，需要向 bd@cherry-ai.com 申请付费商业许可证，所以通过网络提供的修改版本必须公开其源代码。","source":"ai","title":"copyleft 的 agpl 许可证"},{"code":"crash_reports_to_sentry","severity":"info","evidence":".env.example：“Sentry uses a built-in public DSN; development is disabled and production uploads require user consent”，也就是说，一旦你同意，安装好的应用就可以把崩溃报告发送给外部服务。","source":"ai","title":"崩溃报告会发送到 sentry"},{"code":"vendor_cloud_and_regional_build","severity":"info","evidence":"这个仓库会构建两个版本（`CHERRY_EDITION=global` 和使用 `electron-builder.cn.config.cjs` 的 `CHERRY_EDITION=cn`），还有一个“Cherry Cloud”后端设置（.env.example 中的 `MAIN_VITE_CHERRY_CLOUD_API_ORIGIN`），所以某个构建版本会连接哪些服务器取决于你下载的是哪个版本。","source":"ai","title":"供应商云服务与地区专属构建版本"},{"code":"api_gateway_opens_local_access","severity":"info","evidence":"README“Key Features”部分：“use the API Gateway to connect other applications to your configured models”，也就是说，一旦你开启这个功能，应用就可以把你的供应商密钥提供给电脑上的其他程序使用。","source":"ai","title":"api 网关会开放本机访问"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":289,"maxMB":866,"basis":"installer"},"apiKeys":["A model provider such as OpenAI, Anthropic or Gemini (or a local model through Ollama or LM Studio instead)","Web search provider","WebDAV server for backups"],"cost":{"model":"freemium"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["local_ai","ai_agents","productivity"],"alternativeTo":[{"slug":"typingmind","name":"TypingMind","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[],"stars":53401,"downloads":36190815,"downloadsWeek":655709,"downloadsDays":7,"growth":166,"growthDays":7,"history":[{"date":"2026-10-26","stars":53007},{"date":"2026-10-27","stars":53033},{"date":"2026-10-28","stars":53052},{"date":"2026-10-29","stars":53086},{"date":"2026-10-30","stars":53108},{"date":"2026-10-31","stars":53131},{"date":"2026-11-01","stars":53169},{"date":"2026-11-02","stars":53202},{"date":"2026-11-03","stars":53235},{"date":"2026-11-04","stars":53256},{"date":"2026-11-05","stars":53288},{"date":"2026-11-06","stars":53306},{"date":"2026-11-07","stars":53323},{"date":"2026-11-08","stars":53342},{"date":"2026-11-09","stars":53380},{"date":"2026-11-10","stars":53401}],"firstPublishedAt":"2026-10-27T01:10:12.614Z","commitSha":"72041cd52010d19ea961dc93cf7bc1c71cc39d4e","verifiedAt":"2026-11-01T01:44:34.040Z","lastCommitAt":"2026-11-10T17:24:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"One desktop app to chat with many AI models, run agents, build a local knowledge base and create images.","summary":"Cherry Studio is an installable program for Windows, macOS and Linux that puts many AI services in one window. You add your own keys for OpenAI, Anthropic, Gemini and others, or use local models through Ollama or LM Studio, then chat, compare answers from several models, build knowledge bases from your files, translate, search the web, generate images and set up agents that use tools and MCP servers. Conversations and knowledge bases are kept on your device, with optional WebDAV backup. The installer is around 290 MB.","easeRationale":"B on all three: every release has a ready installer (win-x64-setup.exe or a portable .exe, a mac .dmg, and an AppImage, .deb or .rpm for Linux), but after opening it you still have to add an API key or install a local model before it can answer anything.","costNote":"The Community Edition is free under AGPL-3.0; there is a paid Enterprise Edition and a commercial license on request. Cloud models are billed by whichever provider you use.","notVerified":["Whether the dependencies flagged with known vulnerabilities (sharp, systeminformation) affect the shipped app","What the app sends to its own servers in each edition, which needs watching the traffic","How much your use of cloud models will cost per month","Whether the installers were built from this source code"]},"zh-Hans":{"summaryShort":"一个桌面应用，集合了与多种 AI 模型聊天、运行智能体、搭建本地知识库和生成图片的功能。","summary":"Cherry Studio 是一个可以安装在 Windows、macOS 和 Linux 上的程序，把许多 AI 服务集中在一个窗口里。你可以填入自己的 OpenAI、Anthropic、Gemini 等密钥，也可以通过 Ollama 或 LM Studio 使用本地模型，然后聊天、对比多个模型的回答、用你自己的文件搭建知识库、翻译、搜索网页、生成图片，并配置能使用工具和 MCP 服务器的智能体。对话记录和知识库都保存在你自己的设备上，可以选择用 WebDAV 做备份。安装包大约 290 MB。","easeRationale":"三个平台都是 B 级：每个发行版都有现成的安装程序（Windows 上是 win-x64-setup.exe 或便携版 .exe，macOS 上是 .dmg，Linux 上是 AppImage、.deb 或 .rpm），但打开之后你还得先添加一个 API 密钥或安装一个本地模型，它才能回答任何问题。","costNote":"社区版在 AGPL-3.0 许可下免费；还有一个付费的企业版，以及可按需申请的商业许可证。云端模型由你使用的供应商计费。","notVerified":["被标记存在已知漏洞的依赖（sharp、systeminformation）是否会影响发行的应用","每个版本会向自己的服务器发送什么内容，这需要观察网络流量才能确认","你每月使用云端模型大概会花多少钱","安装程序是否是基于这份源代码构建的"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1NDc1NjkyMQ==","slug":"chrisvfritz__prerender-spa-plugin","name":"prerender-spa-plugin","owner":"chrisvfritz","fullName":"chrisvfritz/prerender-spa-plugin","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布于 2018-09-01（2985 天前）","source":"check","title":"发布版本已过时"},{"code":"archived","severity":"info","evidence":"该仓库已被归档（只读，不再维护）","source":"check","title":"已归档"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.6/10","source":"check","title":"scorecard 评分偏低"},{"code":"author_marked_it_deprecated","severity":"attention","evidence":"README.md 的第一行就是“<h1 align=\\\"center\\\">[DEPRECATED]</h1>”，后面还跟着一个“Maintainers Wanted”（招募维护者）徽章。仓库已被归档，还有 204 个未关闭的 issue，所以在那里提交的任何内容都不会得到回应。README 指向了一些替代方案，比如 react-snap、presite，以及它本身基于构建的 prerenderer 库。","source":"ai","title":"作者将其标记为已废弃"},{"code":"built_for_very_old_tooling","severity":"info","evidence":"package.json 写着“engines”: { “node”: “>=4.0.0” }，依赖 html-minifier ^3.5.16 和 @prerenderer/renderer-puppeteer ^0.2.0，并使用 Babel 6 工具链构建。最后一个发布版本是 2018 年的，正处于 webpack 3、4 时代，在 webpack 5 之前。","source":"ai","title":"为非常老旧的工具链构建"},{"code":"installs_a_whole_browser","severity":"info","evidence":"默认的渲染器是 @prerenderer/renderer-puppeteer，它自带 Chrome：“如果你要预渲染最多几百页并希望结果准确，就用 @prerenderer/renderer-puppeteer（再见了，内存！）”。因此安装这个插件会下载一个几百 MB 大小的浏览器。","source":"ai","title":"会安装整个浏览器"},{"code":"runs_your_pages_during_the_build","severity":"info","evidence":"该插件会启动一个本地服务器（“8000 之后的第一个空闲端口”），并在你构建时在浏览器里加载每个列出的路由，所以你应用的代码以及它加载的任何外部脚本都会在构建机器上运行。屏蔽这一点是可选的：“skipThirdPartyRequests……默认值：false”。","source":"ai","title":"在构建期间运行你的页面"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"prerender-io","name":"Prerender.io","confirmed":true}],"declaredAlternatives":[],"stars":7258,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":7259},{"date":"2026-11-06","stars":7258},{"date":"2026-11-07","stars":7258},{"date":"2026-11-08","stars":7258},{"date":"2026-11-09","stars":7257},{"date":"2026-11-10","stars":7258}],"firstPublishedAt":"2026-11-04T22:32:54.854Z","commitSha":"60e400472a8dcc82dce1f39b724f949f99683e5f","verifiedAt":"2026-11-04T22:20:09.879Z","lastCommitAt":"2023-04-09T13:38:15Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A webpack add-on that saves ready HTML pages for a JavaScript site, so search engines see the content.","summary":"This is an add-on for webpack, the tool web developers use to package a site. During the build it starts your single-page app in a hidden Chrome, waits for each listed address to finish drawing and saves the result as a plain HTML file, so visitors and search engines get the finished page instead of an empty one. It is meant for developers who edit webpack.config.js; there is nothing to click. The README's own title says [DEPRECATED] and the repository is archived, so no fixes are coming.","easeRationale":"Developers only on every system, and the grade is the same for all three: it is an npm package you add to the plugins list in webpack.config.js, pointing it at your build folder and the list of addresses to render. It only has an effect inside a webpack build, and the README notes it works only for apps that use HTML5 history routing.","notVerified":["Whether it still builds with current Node and webpack 5, since the last release is from 2018","How long a build takes and how much memory it uses for a site with many addresses","Whether a given framework version hands over correctly from the saved HTML to the live app","Real disk space, which depends mostly on the Chrome that puppeteer downloads"]},"zh-Hans":{"summaryShort":"一个 webpack 插件，为 JavaScript 网站保存现成的 HTML 页面，让搜索引擎也能看到内容。","summary":"这是网页开发者用来打包网站的工具 webpack 的一个插件。在构建过程中，它在一个隐藏的 Chrome 中启动你的单页应用，等每个列出的地址绘制完成后，把结果保存为纯 HTML 文件，这样访客和搜索引擎得到的是绘制完成的页面，而不是一个空白页。它面向编辑 webpack.config.js 的开发者；没有任何东西可以点击。README 的标题本身就写着 [DEPRECATED]，仓库也已被归档，所以不会再有修复。","easeRationale":"在任何系统上都只针对开发者，三个系统的评级也相同：它是一个 npm 包，你把它加到 webpack.config.js 的 plugins 列表里，指定构建文件夹和要渲染的地址列表。它只在 webpack 构建过程中起作用，README 还提到它只适用于使用 HTML5 history 路由的应用。","notVerified":["它是否仍能配合当前的 Node 和 webpack 5 构建，因为最后一个发布版本是 2018 年的","对于地址很多的网站，一次构建要花多长时间、占用多少内存","某个特定框架版本能否正确地从保存的 HTML 过渡到实际运行的应用","实际占用的磁盘空间，这主要取决于 puppeteer 下载的 Chrome"]}}},{"id":"R_kgDOSEVU0g","slug":"citrolabs__ego-lite","name":"Ego Lite","owner":"citrolabs","fullName":"citrolabs/ego-lite","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B"},"risk":"attention","signals":[{"code":"browser_binary_not_in_this_repository","severity":"attention","evidence":"LICENSE 部分写道:\"The contents of this repository are released under the MIT License. The ego lite browser is a separate, free download.\"仓库里只有 skill 和文档,GitHub 发行版里没有任何平台的二进制文件,安装包是从公司自己的服务器获取的(类似 https://cdn.ego.app/setup/windows/x64/egoliteSetup-Y7MbxKIuhzFB.exe 这样的链接),所以你安装的浏览器代码在这里是看不到的。","source":"ai","title":"浏览器二进制文件不在这个仓库里"},{"code":"agent_gets_your_logged_in_sessions","severity":"attention","evidence":"README 说:\"On first launch, ego lite asks one question, whether to migrate your Chrome data. Say yes and your agent inherits your existing logins, cookies, extensions, and bookmarks\",以及 \"the agent can always reach your real logins and tabs through ego-browser\"。一个在你已登录账号里行动的 AI 助手,可以以你的名义发帖、购买或更改内容,给出的示例任务是 \"follow @ego_agent on x.com for me\"。","source":"ai","title":"agent 会获得你已登录的会话"},{"code":"scraping_and_captcha_claims","severity":"info","evidence":"对比表把 \"Fewer CAPTCHAs\" 和 \"Runs large scraping jobs\" 列为功能,README 还提到了 \"Codex scraping 5 competitor sites\"。自动化采集以及绕过网站的反机器人检测,可能违反该网站的使用条款。","source":"ai","title":"关于抓取和验证码的说法"},{"code":"telemetry_collected","severity":"info","evidence":"README 说:\"ego lite keeps its data collection deliberately narrow: simple product signals, like whether you've set ego lite as your default browser\",也就是说这个应用确实会向公司发送一些使用数据。","source":"ai","title":"会收集遥测数据"}],"platforms":["windows","mac"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","automation","productivity"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true},{"slug":"perplexity-comet","name":"Perplexity Comet","confirmed":true}],"declaredAlternatives":[],"stars":22436,"downloads":10225,"downloadsWeek":432,"downloadsDays":2,"growth":787,"growthDays":6,"history":[{"date":"2026-11-08","stars":22130},{"date":"2026-11-09","stars":22290},{"date":"2026-11-10","stars":22436}],"firstPublishedAt":"2026-11-07T22:56:16.720Z","commitSha":"bacaece43900e8ca68eea3b6f1663835ea38df9e","verifiedAt":"2026-11-07T22:38:37.703Z","lastCommitAt":"2026-11-04T09:11:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A browser that lets an AI assistant click through websites in its own window while you keep using your own tabs.","summary":"ego lite is a web browser built so an AI coding assistant such as Claude Code, Codex or Cursor can do web tasks for you: it opens pages in separate workspaces called Spaces, reads them as text and clicks, types and collects information while your own tabs stay untouched. On first launch it offers to copy your Chrome data, so the assistant can use your existing logins. Installers are offered for Windows and macOS (Linux is listed as a plan). You also need an AI assistant of your own, which has its own cost.","easeRationale":"On Windows and macOS you download a ready installer (.exe, or .dmg for Apple Silicon or Intel) and opening it also adds the ego-browser skill to your assistants, which is a B because the browser alone does nothing: you still have to have an AI assistant CLI installed and then start a task with /ego-browser. Linux is only described as a plan on the roadmap.","costNote":"The repository is MIT and the README says the browser is a separate free download. The AI assistant that drives it (Claude Code, Codex, Cursor or similar) is not included and usually has its own subscription or token cost.","notVerified":["Whether the installer downloaded from the company's server matches the code published in this repository, since the browser itself is not here.","The benchmark numbers in the README (47% to 63% faster, success rates and cost), which were measured and published by the authors.","What exactly the app sends back as \"product signals\", and what else the browser contacts while running.","How well the assistant really completes tasks on your own sites, and how much it spends in tokens doing so.","Disk space and memory used after the Chrome data migration."]},"zh-Hans":{"summaryShort":"一款浏览器,让 AI 助手在自己的窗口里点击浏览网站,你照常使用自己的标签页。","summary":"ego lite 是一款浏览器,专门让 Claude Code、Codex 或 Cursor 这类 AI 编程助手替你完成网页任务:它在叫作 Spaces 的独立工作区里打开页面,把内容当作文本来读取,并进行点击、输入和收集信息,而你自己的标签页不受影响。首次启动时,它会询问是否迁移你的 Chrome 数据,这样助手就能使用你已有的登录状态。目前提供 Windows 和 macOS 的安装包(Linux 列在计划中)。你还需要自备一个 AI 助手,它本身也有自己的费用。","easeRationale":"在 Windows 和 macOS 上,你下载一个现成的安装包(.exe,或面向 Apple 芯片/Intel 的 .dmg)并打开,打开的同时还会把 ego-browser 这个 skill 加到你的助手里,属于 B 级,因为浏览器本身什么也做不了:你仍然需要装好一个 AI 助手的命令行工具,然后用 /ego-browser 开始任务。Linux 只是在路线图里被列为一个计划。","costNote":"该仓库采用 MIT 许可,README 说浏览器本体是单独的免费下载项。驱动它的 AI 助手(Claude Code、Codex、Cursor 或类似工具)不包含在内,通常有自己的订阅费或 token 费用。","notVerified":["从公司服务器下载的安装包是否与这个仓库里发布的代码一致,因为浏览器本体并不在这里。","README 里的性能数据(快 47% 到 63%、成功率和成本),是作者自己测量并公布的。","应用究竟把哪些内容当作 \"product signals\" 发回去,运行时浏览器还会联系哪些地方。","助手在你自己的网站上完成任务的实际效果如何,又会花掉多少 token。","迁移 Chrome 数据之后占用的磁盘空间和内存。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk4NDI1NDQ3OA==","slug":"clankwright__spam-bot-3000","name":"Spam Bot 3000","owner":"clankwright","fullName":"clankwright/spam-bot-3000","type":"dev_tool","status":"alert","grade":"D","easeByPlatform":{"linux":"D"},"risk":"high","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"built_to_evade_platform_defences","severity":"high","evidence":"README 自己就是这么写的:\"The bot has some built in anti-spam filter avoidance features to help you remain undetected\"、\"optional bypassing of proprietary twitter APIs and their inherent limitations\",`-b, --browser` 选项是\"thwart api limits by mimicking human in browser!\",针对 Facebook 则是\"zero reliance on proprietary facebook APIs and their inherent limitations\"。绕开一个网站的反垃圾措施和速率限制,本来就是它宣称的功能之一。Reddit、X 和 Facebook 的服务条款都限制自动化访问和批量发帖。","source":"ai","title":"专门用来绕开平台的防护措施"},{"code":"mass_unsolicited_messages","severity":"high","evidence":"README 的使用说明里提供了 `-p, --promote`(\"favorite tweets and reply to tweeters in twit_scrape_dump.txt with random promo from twit_promos.txt\")、`-d, --direct-message`(\"direct message tweeters in twit_scrape_dump.txt with random promo from twit_promos.txt\"),还附带 `auto_spam.bash`,它会\"automatically scrape twitter for queries, filter out results to ignore, and spam remaining results\",Reddit 那边也是同样的回复加广告循环。作者自己也提醒\"your bot could find itself, along with any associated accounts, banned from platforms very quickly\"。很多国家都有关于向未同意的人发送推广消息的广告和数据保护规定。","source":"ai","title":"大量发送未经同意的消息"},{"code":"collects_other_people_s_private_data","severity":"attention","evidence":"Facebook 相关功能是用来\"scrape public and private user profiles for keywords\"和\"scrape public and private group feeds\",并注明\"access to private data requires login to authorized account with associated access\"。仓库里甚至还带着别人帖子的抓取样本(`twit_scrape_dump.txt`、`red_scrape_dump.txt`)。","source":"ai","title":"收集他人的私密数据"},{"code":"your_passwords_saved_in_plain_text","severity":"attention","evidence":"你需要把账号的登录信息写进项目文件夹里普通的文本和 JSON 文件:`credentials.txt` 末尾是 `your_twitter_username` 和 `your_twitter_password`,Facebook 的 `jobs.json` 里有 `\"fb_login\"` 和 `\"fb_password\"`。没有任何加密,一次 git 操作失误就可能把它们公开出去。","source":"ai","title":"密码以明文保存"},{"code":"no_commits_for_seven_years","severity":"attention","evidence":"最后一次提交是在 2019-04-18,距这次检查 2758 天,Instagram 支持还停留在\"TODO\",Twitter 相关部分是照着当年免费的旧 API 和 140 字符限制(\"each line must by <= 140 characters long\")写的,链接的 `dev.twitter.com` 页面也早已不存在。X 的 API 和 Facebook 的页面从那以后已经完全变了。","source":"ai","title":"七年没有提交代码"},{"code":"no_license","severity":"info","evidence":"没有许可证文件,也没有检测到任何许可证,所以默认情况下其他人没有权利复制、修改或再分发这份代码。","source":"ai","title":"没有许可证"}],"platforms":["linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Reddit API (praw app credentials)","X (Twitter) API (tweepy app credentials)"],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"phantombuster","name":"PhantomBuster","confirmed":true},{"slug":"replyguy","name":"ReplyGuy","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":false}],"declaredAlternatives":[],"stars":147,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":147},{"date":"2026-11-07","stars":147},{"date":"2026-11-08","stars":147},{"date":"2026-11-09","stars":147},{"date":"2026-11-10","stars":147}],"firstPublishedAt":"2026-11-05T22:53:18.512Z","commitSha":"205757abef294b4db339f2e679f19cd8271c30c1","verifiedAt":"2026-11-05T22:47:07.390Z","lastCommitAt":"2019-04-18T19:14:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Terminal bot that searches Reddit, Twitter and Facebook for keywords and then mass-replies, follows and messages with your ads.","summary":"spam-bot-3000 is a Python command-line bot for mass promotion on social media. It searches subreddits and Twitter for keyword combinations, saves the results in text files, and then replies to, favourites, follows or direct-messages every result with a random advert from your own list. A Firefox window driven by Selenium does the Facebook scraping and the Twitter posting, which is how it gets around the platforms' own limits. You need Linux, Python 3, four pip packages, Firefox and your account logins. Last touched in April 2019.","easeRationale":"There is no release to download. The only documented install is Debian/Ubuntu Linux: `sudo apt install python3-pip`, then `pip3 install --user tweepy bs4 praw selenium`, plus Firefox and geckodriver for the browser modes, and then you hand-write several files of credentials, keywords and adverts before the first command does anything.","costNote":"The code is free and runs on your own machine with your own social media accounts; the README names no paid service.","notVerified":["Whether anything still works against today's X, Reddit and Facebook, unchanged since April 2019","Whether the four pip packages still install and run on a current Python 3","Whether the Selenium parts find today's Facebook and X page layouts","How quickly an account used this way is suspended","What the shipped `auto_spam.bash` and the other bash scripts do, since they are not reproduced here"]},"zh-Hans":{"summaryShort":"终端机器人,在 Reddit、Twitter 和 Facebook 上按关键词搜索,然后用你的广告大量回复、关注和私信。","summary":"spam-bot-3000 是一个用 Python 写的命令行机器人,用来在社交媒体上大量推广。它在 subreddit 和 Twitter 上搜索关键词组合,把结果存成文本文件,然后对每条结果回复、收藏、关注或发私信,内容是从你自己的清单里随机挑出的广告。一个由 Selenium 控制的 Firefox 窗口负责 Facebook 的抓取和 Twitter 的发帖,也是靠它绕开平台自身的限制。你需要 Linux、Python 3、四个 pip 包、Firefox 以及你各账号的登录信息。最后一次改动是在 2019 年 4 月。","easeRationale":"没有版本可以下载。文档里唯一的安装方式是 Debian/Ubuntu Linux:`sudo apt install python3-pip`,然后 `pip3 install --user tweepy bs4 praw selenium`,浏览器模式还要装 Firefox 和 geckodriver,之后你还要手写好几个文件,填上凭证、关键词和广告,第一条命令才能跑起来。","costNote":"代码免费,在你自己的机器上用你自己的社交媒体账号运行;README 没提到任何收费服务。","notVerified":["这些功能针对今天的 X、Reddit 和 Facebook 是否还能用,毕竟自 2019 年 4 月起就没再改过","这四个 pip 包能否在现在的 Python 3 上正常安装和运行","用 Selenium 的那些部分能否识别现在 Facebook 和 X 的页面布局","这样用的账号多快会被封","附带的 `auto_spam.bash` 和其他 bash 脚本具体做了什么,因为这里没有把它们的内容列出来"]}}},{"id":"R_kgDOOGUp_A","slug":"claraverse-space__claraverse","name":"ClaraVerse","owner":"claraverse-space","fullName":"claraverse-space/ClaraVerse","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"mac":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"same_passwords_for_everyone","severity":"attention","evidence":"开箱即用的 `docker-compose.production.yml` 把 `MYSQL_ROOT_PASSWORD: claraverse`、`MYSQL_PASSWORD: claraverse` 和 `DATABASE_URL=mysql://claraverse:claraverse@mysql:3306` 都写死了,还设置了 `ALLOWED_ORIGINS=*`,并且在没配置时退回到公开的 `WATCHTOWER_HTTP_API_TOKEN`、也就是 `claraverse-watchtower-token`;`docker-compose.yml` 则退回到 `JWT_SECRET=local_test_jwt_secret_32chars_min` 和 `DEV_API_KEY=claraverse-dev-key-2024`,并把 MySQL(3306)、MongoDB(27017,无用户名密码)和 Redis(6379)直接发布在机器自己的端口上。任何能连上这台机器的人都知道这些值。","source":"ai","title":"所有人用的密码都一样"},{"code":"updates_itself_through_the_docker_socket","severity":"attention","evidence":"生产环境的 compose 文件加入了一个 Watchtower 容器,挂载了 `/var/run/docker.sock:/var/run/docker.sock`,设置 `WATCHTOWER_HTTP_API_UPDATE=true`,每小时轮询一次,应用本身也会调用它:\"Backend calls Watchtower's HTTP API to apply updates on demand from the admin dashboard\"。镜像是按 `latest` 标签拉取的,所以新代码随时可能替换正在运行的内容,谁拿到那个 token 就能触发更新;Docker socket 等于对整台机器的完全控制权。","source":"ai","title":"通过 docker socket 自我更新"},{"code":"install_script_piped_from_the_internet","severity":"attention","evidence":"README 里第一个安装选项是 `curl -fsSL https://raw.githubusercontent.com/claraverse-space/ClaraVerse/main/cli/install.sh | bash && claraverse init`,它会执行那个地址当时提供的任何内容,你完全没机会先看一眼。","source":"ai","title":"安装脚本直接从网上管道执行"},{"code":"an_assistant_that_runs_commands_on_your_computer","severity":"attention","evidence":"README 描述了 Clara Agent,用 `claraverse agent install` 安装:\"a coding agent that runs in your terminal and ships in this repo. It reads, writes, and edits files and runs commands on your machine\",Devices 功能还能让这个助手\"reach MCP on any of them remotely\"。","source":"ai","title":"一个能在你电脑上执行命令的助手"},{"code":"the_install_script_in_the_repository_is_a_leftover","severity":"info","evidence":"仓库顶层的 `install.sh` 里还留着别的项目的占位地址:`REPO_URL=\"https://raw.githubusercontent.com/yourusername/ClaraVerse-Scarlet-OSS/main\"`,它会下载 `docker-compose.prod.yml`,这个文件名在本仓库里并不存在。README 指向的其实是另一个脚本,`cli/install.sh`。","source":"ai","title":"仓库里的安装脚本是别的项目留下的"},{"code":"apps_promised_in_the_description_are_not_here","severity":"info","evidence":"仓库描述里承诺\"With desktop, IOS, Android Apps\",但这个仓库里只有一个 Go 后端、一个 React 前端和一些 Docker 文件;README 把你指向另一个仓库(ClaraVerse-Desktop)去找桌面应用,完全没提到任何 iOS 或 Android 应用。","source":"ai","title":"描述里承诺的应用程序这里都没有"},{"code":"license_has_extra_conditions","severity":"info","evidence":"GitHub 把这个许可证标记为 NOASSERTION,因为 LICENSE 文件在 AGPL-3.0 正文前面加了条款:\"Additional Terms (AGPL Section 7)\"要求署名,还有\"DUAL LICENSING OPTIONS\",为 1000 人以上的公司提供收费的商业许可证。copyleft 的义务依然存在:把修改版当作网络服务提供,就必须公开源代码。","source":"ai","title":"许可证带有额外条件"}],"platforms":["linux","mac","web"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["OpenAI, Anthropic or Google — not needed if you use Ollama or LM Studio models","E2B (sandbox that runs code; optional)","Composio (the 150+ integrations; optional)","Telegram bot (the Channels feature; optional)"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"file"},"categories":["local_ai","ai_agents","automation"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"typingmind","name":"TypingMind","confirmed":true},{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"n8n-cloud","name":"Cloud","confirmed":true},{"slug":"claude-code","name":"Claude Code","confirmed":false}],"declaredAlternatives":[],"stars":3902,"downloads":235,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":3904},{"date":"2026-11-07","stars":3904},{"date":"2026-11-08","stars":3902},{"date":"2026-11-09","stars":3902},{"date":"2026-11-10","stars":3902}],"firstPublishedAt":"2026-11-05T22:43:48.317Z","commitSha":"078d39859182a48d55d452544191989233c0d062","verifiedAt":"2026-11-05T22:28:09.577Z","lastCommitAt":"2026-08-03T16:56:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted AI workspace: chat with local or online models, teams of agents, memory, knowledge bases and visual workflows.","summary":"ClaraVerse is a private AI workspace you run on your own machine or server with Docker: chat, teams of agents that pass work through a review board, a memory that keeps facts about you, knowledge bases over your own PDFs, a drag-and-drop workflow builder and access from Telegram. It picks up models from Ollama or LM Studio on your computer, or uses your OpenAI, Anthropic or Google keys. Docker and about 4 GB of RAM are the requirements; the full set-up also starts MySQL, MongoDB, Redis, a search engine and a vector database. The makers also run a hosted version.","easeRationale":"The makers' hosted version is listed in the README as \"Free hosted version, no setup required\", so on the web it is a B: an account plus either a model key or a local model. Self-hosting on Linux or Mac is a D: the two documented routes are `curl -fsSL .../install.sh | bash && claraverse init` or `docker compose -f docker-compose.production.yml up -d`, which start the app next to MySQL, MongoDB, Redis, SearXNG, Qdrant and an embeddings container, on 4 GB of RAM or more. None of the 4 releases has a file to download, and there are no Windows instructions here; the desktop app is in a separate repository.","costNote":"The code is AGPL-3.0 and local models cost nothing to run. The LICENSE file adds a second option: a commercial license, \"FREE upon request\" for companies under 1000 employees and paid for 1000 or more. Online model keys are billed by whoever provides them.","notVerified":["Whether the hosted version at claraverse.space really is free, and what it does with your conversations","Whether the whole stack starts on 4 GB of RAM, and how much disk MySQL, MongoDB, Qdrant and the embeddings models take after the first boot","Whether the \"150+ integrations\" and \"200+ integrations\" counted in the README work without a Composio key","Whether the automatic detection of Ollama and LM Studio works on each system","Whether the memory encryption works as described, since only the README describes it","What the Clara Agent does on your machine, as its code was not read here"]},"zh-Hans":{"summaryShort":"自托管 AI 工作区:和本地或在线模型聊天,多个智能体协作,带记忆、知识库和可视化工作流。","summary":"ClaraVerse 是一个私有的 AI 工作区,你用 Docker 在自己的机器或服务器上运行它:聊天、一组会把工作交给审核环节的智能体团队、能记住你相关事实的记忆功能、基于你自己 PDF 的知识库、一个拖拽式工作流编辑器,还能从 Telegram 接入。它会自动使用你电脑上 Ollama 或 LM Studio 里的模型,或者用你的 OpenAI、Anthropic 或 Google 密钥。要求是 Docker 和大约 4 GB 内存;完整安装还会一起启动 MySQL、MongoDB、Redis、一个搜索引擎和一个向量数据库。开发者自己也运营着一个托管版本。","easeRationale":"README 里把开发者的托管版本列为\"Free hosted version, no setup required\",所以在网页端是 B 级:只要一个账号,再配一个模型密钥或本地模型。在 Linux 或 Mac 上自托管是 D 级:文档里的两条路径是 `curl -fsSL .../install.sh | bash && claraverse init`,或者 `docker compose -f docker-compose.production.yml up -d`,它们会把这个应用和 MySQL、MongoDB、Redis、SearXNG、Qdrant 以及一个 embeddings 容器一起启动,需要 4 GB 以上内存。4 个发布版本都没有可下载的文件,这里也没有 Windows 的说明;桌面版放在另一个仓库里。","costNote":"代码是 AGPL-3.0 协议,本地模型运行不花钱。LICENSE 文件还加了第二个选项:商业许可证,员工少于 1000 人的公司\"FREE upon request\",1000 人以上要付费。在线模型的密钥费用由对应的服务商收取。","notVerified":["claraverse.space 上的托管版本是否真的免费,它又是怎么处理你的对话内容的","整套服务是否真能在 4 GB 内存上跑起来,首次启动之后 MySQL、MongoDB、Qdrant 和 embeddings 模型各占多少磁盘空间","README 里提到的\"150+ integrations\"和\"200+ integrations\"在没有 Composio 密钥的情况下能不能用","Ollama 和 LM Studio 的自动检测在各种系统上是否都能正常工作","记忆加密是否真的像描述那样起作用,因为这一点只有 README 里提到过","Clara Agent 在你机器上到底做了什么,因为这里没有读过它的代码"]}}},{"id":"R_kgDOKwVKzA","slug":"clash-verge-rev__clash-verge-rev","name":"Clash Verge Rev","owner":"clash-verge-rev","fullName":"clash-verge-rev/clash-verge-rev","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"proxy_tool_restricted_in_some_countries","severity":"attention","evidence":"该仓库把自己描述为一个 \"modern GUI client ... for tailored proxy experience\"，内置 Clash.Meta（mihomo）引擎，并提供会捕获全部流量的 \"TUN（虚拟网卡）\" 模式。这类工具在大多数地方属于普通软件，但在中国、俄罗斯、伊朗、土耳其和阿联酋等国家，使用未经批准的代理或 VPN 是受限或被禁止的。","source":"ai","title":"代理工具在一些国家受到限制"},{"code":"used_to_get_around_regional_blocks","severity":"attention","evidence":"Cargo.toml 中包含工作区成员 \"crates/clash-verge-media-unlock\"，README 中推广的服务商宣传 \"全面支持流媒体及各AI访问\"。通过其他国家观看流媒体服务通常违反该服务自己的条款。","source":"ai","title":"被用来绕过地区限制"},{"code":"ads_for_a_paid_provider_with_a_referral_link","severity":"attention","evidence":"README 的 \"Promotion\" 部分推销了一个代理服务商和一个 AI 支持产品；前者写的 \"official site\" 是 https://www.claudeborder.com，但所有链接，包括那个，都指向 https://cruise.54678999.xyz/#/register?code=58q5UJZc，一个在完全不同域名下的推荐地址。","source":"ai","title":"带推荐链接的付费服务商广告"},{"code":"needs_administrator_rights_for_a_background_service","severity":"info","evidence":"README 列出了 \"系统代理和守卫、TUN(虚拟网卡)模式\"，代码会安装一个具有特权的辅助服务（package.json 中的 dev:service 用来 \"install or update the isolated development service\"；Cargo.toml 引入了仅限 Windows 的 \"deelevate\" crate）。","source":"ai","title":"需要管理员权限来运行一个后台服务"},{"code":"rolling_build_channel_may_be_unstable","severity":"info","evidence":"README 的版本表把 AutoBuild 标记为 \"滚动更新版，适合测试反馈，可能存在缺陷\"，Alpha 渠道则被标记为已放弃；只有 Release 页面才是稳定版本。","source":"ai","title":"滚动更新渠道可能不稳定"},{"code":"continuation_of_an_abandoned_project","severity":"info","evidence":"README 的标题写着 \"Continuation of Clash Verge\"（zzzgydi/clash-verge），说明这是另一个团队在延续一个更早的项目。","source":"ai","title":"延续一个已被放弃的项目"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","rust"],"disk":{"minMB":239,"maxMB":716,"basis":"installer"},"apiKeys":["A proxy provider's subscription link or configuration file"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system"],"alternativeTo":[{"slug":"surge","name":"Surge","confirmed":true},{"slug":"proxifier","name":"Proxifier","confirmed":true}],"declaredAlternatives":[],"stars":157627,"downloads":52876284,"downloadsWeek":2531299,"downloadsDays":7,"growth":1552,"growthDays":7,"history":[{"date":"2026-10-26","stars":154093},{"date":"2026-10-27","stars":154373},{"date":"2026-10-28","stars":154633},{"date":"2026-10-29","stars":154920},{"date":"2026-10-30","stars":155175},{"date":"2026-10-31","stars":155367},{"date":"2026-11-01","stars":155561},{"date":"2026-11-02","stars":155808},{"date":"2026-11-03","stars":156075},{"date":"2026-11-04","stars":156327},{"date":"2026-11-05","stars":156544},{"date":"2026-11-06","stars":156757},{"date":"2026-11-07","stars":156957},{"date":"2026-11-08","stars":157149},{"date":"2026-11-09","stars":157388},{"date":"2026-11-10","stars":157627}],"firstPublishedAt":"2026-10-27T01:10:12.615Z","commitSha":"5b7e876ab0332620e100bf826fc84f20f1658acd","verifiedAt":"2026-11-03T02:39:22.993Z","lastCommitAt":"2026-11-10T17:29:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop app that sends your internet traffic through proxy servers you add, with rules for each app or site.","summary":"Clash Verge Rev is a window-and-tray app that routes your internet traffic through proxy servers. You paste a subscription link or a configuration file from a proxy provider, pick a server from the list, and it can send either the whole computer or only some apps and sites through it. It has the mihomo engine built in, a rule and node editor, backup over WebDAV and a system-wide mode. Installers are ready for Windows, macOS 11+ and Linux; the servers themselves are not included.","easeRationale":"Every platform has a ready installer (a .exe setup for Windows, .dmg for macOS 11+, .deb and .rpm for Linux), and the engine comes inside. It is B, not A, because the app does nothing until you paste a subscription link from a proxy provider, and the system-wide and TUN modes ask for administrator rights.","costNote":"The app is free under GPL-3.0 and takes donations through GitHub Sponsors. The proxy servers it connects to are a separate service, usually paid: the README advertises a paid provider with a referral link and a 3-day trial.","notVerified":["Whether the installers really are the code published here, since only the project signs them.","Whether no data leaves the computer, as the privacy policy states.","Whether a connection keeps working, which depends on the proxy provider you pay and not on the app.","Whether using it where you live is allowed, which depends on your country's rules.","How much a proxy subscription costs over time, and whether the advertised provider behaves as its text promises."]},"zh-Hans":{"summaryShort":"一款桌面应用，把你的网络流量通过你添加的代理服务器转发，并可以为每个应用或网站设置规则。","summary":"Clash Verge Rev 是一个带窗口和系统托盘的应用，通过代理服务器转发你的网络流量。你粘贴一个订阅链接，或来自某个代理服务商的配置文件，从列表中选一个服务器，就可以让整台电脑、或者只让部分应用和网站通过这个代理。它内置了 mihomo 引擎、规则和节点编辑器、WebDAV 备份，还有全局模式。Windows、macOS 11+ 和 Linux 都有现成的安装包；代理服务器本身不包含在内。","easeRationale":"每个平台都有现成的安装包（Windows 的 .exe 安装程序，macOS 11+ 的 .dmg，Linux 的 .deb 和 .rpm），而且引擎已经内置在里面。评级是 B 而不是 A，是因为在你粘贴代理服务商的订阅链接之前，这个应用什么都做不了，而且全局模式和 TUN 模式都需要管理员权限。","costNote":"该应用基于 GPL-3.0 许可免费提供，并通过 GitHub Sponsors 接受捐赠。它连接的代理服务器是另一项独立的服务，通常是付费的：README 推广了一个带推荐链接和 3 天试用期的付费服务商。","notVerified":["安装包是否真的是这里发布的代码，因为只有项目方自己为它签名","是否真的没有数据离开你的电脑，正如隐私政策所说","连接能否一直正常使用，这取决于你付费的代理服务商，而不是这个应用本身","在你所在的地方使用它是否合法，这取决于你所在国家的规定","代理订阅长期下来要花多少钱，以及推广的那家服务商是否真如宣传所说"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMTI2MTMwNDk=","slug":"cli__cli","name":"GitHub CLI","owner":"cli","fullName":"cli/cli","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"low","signals":[{"code":"linux_repo_key_rotation","severity":"info","evidence":"README：\"GitHub CLI's PGP signing key rotation for Linux package repositories is effective September 5, 2026; users experiencing package installation or update problems can find solutions in the public announcement\"。","source":"ai","title":"Linux 软件源更换密钥"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["go"],"disk":{"minMB":15,"maxMB":44,"basis":"installer"},"apiKeys":["GitHub account sign-in"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":46955,"downloads":152948257,"downloadsWeek":4029437,"downloadsDays":7,"growth":96,"growthDays":7,"history":[{"date":"2026-10-26","stars":46753},{"date":"2026-10-27","stars":46767},{"date":"2026-10-28","stars":46785},{"date":"2026-10-29","stars":46799},{"date":"2026-10-30","stars":46813},{"date":"2026-10-31","stars":46826},{"date":"2026-11-01","stars":46841},{"date":"2026-11-02","stars":46846},{"date":"2026-11-03","stars":46859},{"date":"2026-11-04","stars":46873},{"date":"2026-11-05","stars":46889},{"date":"2026-11-06","stars":46897},{"date":"2026-11-07","stars":46907},{"date":"2026-11-08","stars":46923},{"date":"2026-11-09","stars":46936},{"date":"2026-11-10","stars":46955}],"firstPublishedAt":"2026-10-27T01:10:12.616Z","commitSha":"f7a06a18a91eef5c0b218b549dab64b86f36f8b9","verifiedAt":"2026-10-29T01:40:23.843Z","lastCommitAt":"2026-11-10T17:21:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Use GitHub from a terminal: manage pull requests, issues, repositories and more by typing commands.","summary":"GitHub CLI (the gh command) is GitHub's official tool for working with GitHub from a terminal. You can create and review pull requests, handle issues, clone repositories, check automated runs and call GitHub's API without opening the website. It is for programmers who already use git and GitHub. Windows, Mac and Linux get ready installers and packages; after installing you sign in to your GitHub account and type commands. Release files come with signed build records you can check.","easeRationale":"Every platform has a ready package: a Windows installer (gh_2.103.3_windows_amd64.msi), a Mac .pkg, and .deb/.rpm packages for Linux; after installing you sign in to GitHub once, then work by typing commands in a terminal.","costNote":"Free (MIT). It works with your GitHub account; paid GitHub plans are separate and not needed to use the tool.","notVerified":["Where and how your GitHub sign-in token is stored on your computer.","Whether optional extensions or agent skills you add with gh behave as described, since they come from other authors.","Whether the Linux package repositories install cleanly after the signing key change."]},"zh-Hans":{"summaryShort":"在终端里使用 GitHub:通过输入命令管理拉取请求、议题、仓库等。","summary":"GitHub CLI(gh 命令)是 GitHub 官方提供的、用于在终端中操作 GitHub 的工具。你可以创建和审查拉取请求、处理议题、克隆仓库、查看自动化运行结果,并调用 GitHub 的 API,而不必打开网站。它适合已经在用 git 和 GitHub 的程序员。Windows、Mac 和 Linux 都提供现成的安装程序和软件包;安装后登录你的 GitHub 账户即可输入命令使用。各版本文件都附有可供核验的签名构建记录。","easeRationale":"每个平台都有现成的软件包:Windows 安装程序(gh_2.103.3_windows_amd64.msi)、Mac 的 .pkg,以及 Linux 的 .deb/.rpm 软件包;安装后只需登录一次 GitHub,之后在终端输入命令即可使用。","costNote":"免费(MIT 许可)。它使用你的 GitHub 账户;GitHub 的付费套餐是另一回事,使用该工具并不需要。","notVerified":["你的 GitHub 登录令牌在电脑上存储在哪里、如何存储","通过 gh 添加的可选扩展或代理技能是否真的按说明运作,因为它们来自其他作者","签名密钥更换后,Linux 的软件包仓库是否能顺利安装"]}}},{"id":"R_kgDOS-b1TA","slug":"cloudflare__security-audit-skill","name":"Security Audit Skill","owner":"cloudflare","fullName":"cloudflare/security-audit-skill","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"runs_the_audited_project_code","severity":"attention","evidence":"README 的 Requirements 部分：\"An OS-enforced sandbox for target-controlled builds, tests, processes, browsers, emulators, fuzzers, and fixtures. It must disable external networking, use a sanitized allowlisted environment, enforce resource limits, and allow writes only to assigned scratch paths. Without these controls, the workflow keeps the lead as `needs_validation` instead of executing target code.\" 搭建这个沙箱由你自己负责。","source":"ai","title":"会运行被审计项目的代码"},{"code":"finds_weaknesses_in_any_codebase_you_point_it_at","severity":"info","evidence":"README 的 Usage 部分：\"Start your coding agent in (or pointed at) the codebase you want to audit\"，触发词列表中包含 \"pen-test the code\"。这些发现描述的是真实存在的弱点，因此报告属于敏感材料，把它用在不属于你的软件上可能越界。","source":"ai","title":"能在你指定的任何代码库中找出弱点"},{"code":"many_model_calls_per_run","severity":"info","evidence":"README 描述了六个阶段，包含 \"isolated hunters\"、\"coverage critics\"、为每个候选项配备的 \"fresh verifier\"，以及用于记录核验的 \"fresh agents\"，并表示 \"a single run found roughly half of the vulnerabilities that repeated runs found in total\"。这意味着每次审计都会消耗大量 AI 用量。","source":"ai","title":"每次运行都会多次调用模型"},{"code":"no_releases_and_quiet_for_weeks","severity":"info","evidence":"该仓库在 GitHub 上没有发布版本（release），本次快照中的最后一次提交已是 44 天前，所以你安装的是当时主分支上的任何内容。","source":"ai","title":"没有发布版本，且数周无更新"},{"code":"writes_files_outside_your_project","severity":"info","evidence":"README 的 Usage 部分：\"In full audit mode, an unspecified output directory defaults to `~/security-audit-skill/<repo-name>/run-<N>`.\" 带有漏洞发现的报告会落在你的主文件夹中，除非你另选路径。","source":"ai","title":"会在你的项目之外写入文件"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"snyk","name":"Snyk","confirmed":true},{"slug":"semgrep","name":"Semgrep","confirmed":true}],"declaredAlternatives":[],"stars":69112,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":10927,"growthDays":7,"history":[{"date":"2026-10-27","stars":50205},{"date":"2026-10-28","stars":51675},{"date":"2026-10-29","stars":52869},{"date":"2026-10-30","stars":54143},{"date":"2026-10-31","stars":55343},{"date":"2026-11-01","stars":56438},{"date":"2026-11-02","stars":57412},{"date":"2026-11-03","stars":58185},{"date":"2026-11-04","stars":59065},{"date":"2026-11-05","stars":61220},{"date":"2026-11-06","stars":63714},{"date":"2026-11-07","stars":65495},{"date":"2026-11-08","stars":66716},{"date":"2026-11-09","stars":67926},{"date":"2026-11-10","stars":69112}],"firstPublishedAt":"2026-10-27T18:40:27.822Z","commitSha":"c1c8a8c1471069fb0e188eeaff69b8e8db6564a8","verifiedAt":"2026-10-29T01:43:40.515Z","lastCommitAt":"2026-09-14T19:28:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill that turns an AI coding agent into a security auditor and writes a report of what it found.","summary":"This is an instruction pack for AI coding agents, published by Cloudflare. Once installed, you ask your agent to \"security audit this codebase\" and it runs a six-step review: it maps the code, sends separate agents to hunt for weaknesses, has a second agent try to disprove each one, then writes a report splitting findings into confirmed, needs checking and rejected. It is for people who already use an AI coding agent; you also need Node installed for the two checker scripts. The repository is only text files and those scripts.","easeRationale":"Same work on all three systems and no installer: you need Node, you run `npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit` in a terminal, and then ask your AI coding agent for an audit. The README also asks for an OS-enforced sandbox before the agent is allowed to run the audited project's own code, which is extra setup.","costNote":"MIT and free. Nothing runs without an AI coding agent: the README asks for \"a coding agent with a model that supports tool use and parallel sub-agents\", which you pay for separately, and a long audit uses many model calls.","notVerified":["Whether the findings are real: the project's own report splits them into confirmed, needs_validation and rejected","How much your AI agent will charge for a full audit of a real project","Whether it works on a given coding agent, since it needs parallel sub-agents","How you would actually set up the OS-enforced sandbox the README requires","Whether it misses important weaknesses, since the README says one run found about half of what repeated runs found"]},"zh-Hans":{"summaryShort":"一个将 AI 编程代理变成漏洞审计员并撰写发现报告的技能。","summary":"这是 Cloudflare 发布的一套面向 AI 编程代理的指令包。安装后,你让代理对这个代码库做“漏洞审计”,它会执行六步审查:先绘制代码结构,再派出多个独立代理去寻找薄弱点,然后让另一个代理去反驳每一条发现,最后写出一份报告,把发现分为已确认、待核实和已驳回三类。它适合已经在用 AI 编程代理的人;两个检查脚本还需要安装 Node。该仓库只包含文本文件和这两个脚本。","easeRationale":"在三个系统上操作方式相同,也没有安装程序:你需要 Node,在终端运行 `npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit`,然后让你的 AI 编程代理执行审计。README 还要求在允许代理运行被审计项目自身代码之前,先搭建一个由操作系统强制的隔离环境(sandbox),这需要额外的配置。","costNote":"MIT 许可,免费。没有 AI 编程代理什么都跑不起来:README 要求“一个支持工具调用和并行子代理的模型的编程代理”,这部分费用需要你自行另付,一次长时间的审计会用掉很多模型调用次数。","notVerified":["发现的问题是否属实:项目自己的报告把它们分为已确认、待核实和已驳回","你的 AI 代理对一个真实项目做完整审计会收多少费用","它是否适用于某个特定的编程代理,因为它需要并行子代理","你实际该如何搭建 README 要求的、由操作系统强制的隔离环境","它是否会漏掉重要的薄弱点,因为 README 提到单次运行只发现了多次重复运行所发现问题的大约一半"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTI3NjE0Nw==","slug":"cmderdev__cmder","name":"Cmder","owner":"cmderdev","fullName":"cmderdev/cmder","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B"},"risk":"attention","signals":[{"code":"listed_releases_without_downloadable_files","severity":"attention","evidence":"材料列出了最近三个发布版：v2.0.0-rc.1（2026-11-10）、v1.3.26（2026-11-10）和 v1.3.25（2024-05-31），每一个都标为 \"no platform binaries\"，检查结果里也是 \"platformsFromReleases\": []。而 README 仍然把下载指向发布页。","source":"ai","title":"列出的发布版没有可下载文件"},{"code":"manual_update_of_bundled_programs","severity":"info","evidence":"README 的 \"Upgrading\"：\"Delete the `[cmder_root]/vendor` folder\" 以及 \"Extract the new `cmder_full.zip` or `cmder_nogit.zip` into `[cmder_root]/` overwriting all files when prompted.\" 内置程序都放在这个 vendor 文件夹里，因此只有用户手动操作时才会更新。","source":"ai","title":"内置程序需要手动更新"},{"code":"runs_a_git_it_finds_on_your_pc","severity":"info","evidence":"README 的 \"Vendored Git\"：\"On each instance of launching Cmder, an attempt is made to locate any other user provided Git binaries. Upon finding a `git.exe` binary, Cmder further compares its version against the vendored one _by executing_ it.\"","source":"ai","title":"会运行它在你电脑上找到的 Git"},{"code":"unfinished_builds_offered_outside_releases","severity":"info","evidence":"README 的 \"Latest Builds\"：\"New and unfinished work continues through feature PRs to `main`\" 和 \"You can download the latest successful builds of `main` from the **GitHub Actions > Artifacts** section\"。本次材料中的 v2.0.0-rc.1 是候选版本，不是正式版。","source":"ai","title":"在发布版之外提供未完成的构建"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"mobaxterm","name":"MobaXterm","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":27008,"downloads":11632140,"downloadsWeek":3995,"downloadsDays":5,"growth":3,"growthDays":6,"history":[{"date":"2026-11-05","stars":27006},{"date":"2026-11-06","stars":27006},{"date":"2026-11-07","stars":27006},{"date":"2026-11-08","stars":27007},{"date":"2026-11-09","stars":27007},{"date":"2026-11-10","stars":27008}],"firstPublishedAt":"2026-11-04T22:32:54.861Z","commitSha":"5655c5eb90a075b8bcdd8617bd331bd3dd96c5c2","verifiedAt":"2026-11-10T23:55:08.972Z","lastCommitAt":"2026-11-10T02:49:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A portable console for Windows with tabs, a nicer prompt and bundled Git, that can run from a USB stick.","summary":"Cmder is a console package for Windows. It bundles a terminal (ConEmu or Windows Terminal), the clink command-line add-on, a Monokai colour scheme and a Git for Windows install, all in one folder you can copy to a USB stick or cloud drive. You can open cmd, PowerShell, bash and mintty tabs in the same window, with aliases and a Git-aware prompt. It is for people who use the Windows command line often. The README says to download the release archive, extract it somewhere that does not need Administrator rights, and run Cmder.exe.","easeRationale":"On Windows the README asks only for three steps: download the latest release, extract the archive outside Program Files, and run Cmder.exe, with nothing else to install. It is not graded A because the three releases listed in this package (v2.0.0-rc.1, v1.3.26 and v1.3.25) all appear with no platform files and the check found no platforms from releases, so it is not certain a ready-made archive is waiting on that page.","costNote":"MIT licence; the README says each bundled program keeps its own licence.","notVerified":["Whether a ready-to-use cmder_full.zip or cmder_nogit.zip is actually downloadable from the releases page, since the listed releases show no files","Which versions of the bundled programs (ConEmu, clink, Git for Windows) ship inside the vendor folder","How much disk space the extracted folder takes, as the package reports no size","Whether it runs correctly on each Windows version, and whether the Mini and no-Git editions behave as the README describes"]},"zh-Hans":{"summaryShort":"适用于 Windows 的便携控制台：多标签页、更好看的提示符，并内置 Git，可以放在 U 盘里运行。","summary":"Cmder 是一个 Windows 控制台套件。它把终端（ConEmu 或 Windows Terminal）、命令行增强插件 clink、Monokai 配色方案和一份 Git for Windows 打包进同一个文件夹，你可以把它复制到 U 盘或云盘里。你可以在同一个窗口里打开 cmd、PowerShell、bash 和 mintty 标签页，并使用别名和能显示 Git 状态的提示符。它适合经常使用 Windows 命令行的人。README 说，下载发布压缩包，解压到不需要管理员权限的位置，然后运行 Cmder.exe。","easeRationale":"在 Windows 上，README 只要求三步：下载最新发布版、把压缩包解压到 Program Files 之外、运行 Cmder.exe，不需要再安装别的东西。没有给 A 级，是因为本次材料里列出的三个发布版（v2.0.0-rc.1、v1.3.26 和 v1.3.25）都没有附带平台文件，检查结果也没有从发布版中识别出任何平台，所以无法确认那个页面上真的有现成的压缩包。","costNote":"MIT 许可证；README 说每个内置程序各自保留自己的许可证。","notVerified":["发布页上是否真的能下载到现成的 cmder_full.zip 或 cmder_nogit.zip，因为列出的发布版都没有显示文件","vendor 文件夹里内置的程序（ConEmu、clink、Git for Windows）具体是哪些版本","解压后的文件夹占用多少磁盘空间，材料中没有给出大小","在各个 Windows 版本上是否都能正常运行，以及 Mini 版和不含 Git 的版本是否与 README 描述一致"]}}},{"id":"R_kgDON6mdyw","slug":"coleam00__local-ai-packaged","name":"Local AI Packaged","owner":"coleam00","fullName":"coleam00/local-ai-packaged","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"example_passwords_that_look_ready_to_use","severity":"attention","evidence":"`.env.example` 自带的值原样就能用:`DASHBOARD_PASSWORD=this_password_is_insecure_and_should_be_updated`、`NEO4J_AUTH=neo4j/password`、`N8N_ENCRYPTION_KEY=super-secret-key`、一个已经填好的 `SECRET_KEY_BASE=UpNVntn3cDxHJpq99YMc1T1AQgQpc8kfYTuRgBiYa15BLrx8etQoXz3gZv1/u2oq`,以及 Supabase 公开演示令牌 `ANON_KEY` 和 `SERVICE_ROLE_KEY`。README 在 2026 年 2 月的说明甚至让大家都复制同一个 `S3_PROTOCOL_ACCESS_KEY_SECRET=850181e4652dd023b7a98c58ae0d2d34bd487ee0cc3254aed6eda37307425907`。谁跳过了 \"YOU MUST CHANGE THESE\" 这一步,运行的数据库就是人人都能用这些密钥打开的。","source":"ai","title":"看起来已经可以直接用的示例密码"},{"code":"docker_control_left_open_without_protection","severity":"attention","evidence":"故障排查部分写道:\"If using Docker Desktop: Go into the Docker settings and make sure 'Expose daemon on tcp://localhost:2375 without TLS' is turned on\"。这个设置会把 Docker 的完全控制权,也就是整台机器的控制权,交给一个没有密码、没有加密的网络端口。","source":"ai","title":"Docker 控制接口毫无防护地暴露出去"},{"code":"ports_the_firewall_does_not_cover","severity":"attention","evidence":"云端部署部分带有作者自己的警告:\"ufw does not shield ports published by docker, because the iptables rules configured by docker are analyzed before those configured by ufw\",`.env.example` 还补充说 \"you might not want to expose Ollama or SearXNG since they aren't secured by default\"。在租用的服务器上,即使设置了防火墙,这些服务也可能最终让任何人都能访问到。","source":"ai","title":"防火墙没有覆盖到的端口"},{"code":"containers_always_pull_the_newest_version","severity":"attention","evidence":"镜像都没有固定版本:`n8nio/n8n:latest`、`ollama/ollama:latest`、`neo4j:latest`、`flowiseai/flowise`、`qdrant/qdrant`、`ghcr.io/open-webui/open-webui:main`。你机器上运行的就是这些项目最新发布的内容,README 自己列出的破坏性变更(6 月和 2026 年 2 月新增的 Supabase 变量)就说明了这会带来什么。","source":"ai","title":"容器总是拉取最新版本"},{"code":"code_fetched_from_another_project_at_start","severity":"info","evidence":"`docker-compose.yml` 开头是 `include: - ./supabase/docker/docker-compose.yml`,这个文件夹并不在本仓库里:运行 `start_services.py` 时它会下载 Supabase 自己的仓库,故障排查说明也证实了这一点(\"you had a 'bad' pull of the Supabase GitHub repository when you ran the start_services.py script\")。","source":"ai","title":"启动时从另一个项目拉取代码"},{"code":"a_lot_of_computer_for_one_package","severity":"info","evidence":"一条命令就能启动大约二十个容器——n8n、Ollama、Open WebUI、Flowise、Qdrant、Neo4j、SearXNG、Caddy、带 ClickHouse、MinIO 和 Redis 的 Langfuse,再加上整套 Supabase(Postgres、Kong、Studio、storage、analytics、pooler)——首次启动时 Ollama 还会自己下载两个模型(`ollama pull qwen2.5:7b-instruct-q4_K_M; ollama pull nomic-embed-text`)。仓库没有给出任何磁盘或内存的数字。","source":"ai","title":"一个软件包用掉太多电脑资源"},{"code":"workflows_can_run_commands_on_the_machine","severity":"info","evidence":"README 解释了如何打开新版本默认关闭的 n8n 节点:\"Starting with n8n v2+, the `Local File Trigger` and `Execute Command` nodes are disabled by default for security reasons\",只需取消 `NODES_EXCLUDE=[]` 的注释。打开之后,你导入的任何工作流都可以在容器里执行命令,并读取从你硬盘挂载进来的共享文件夹。","source":"ai","title":"工作流可以在机器上执行命令"},{"code":"monitoring_sends_usage_data_out","severity":"info","evidence":"Langfuse 启动时带着 `TELEMETRY_ENABLED: ${TELEMETRY_ENABLED:-true}`,所以默认情况下,这个监控容器会把使用情况报告给它的开发者——而这是一个主打本地运行的软件包。相比之下,n8n 自己的 `N8N_DIAGNOSTICS_ENABLED=false` 是关闭的。","source":"ai","title":"监控组件会把使用数据发出去"},{"code":"author_says_it_is_not_ready_for_real_use","severity":"info","evidence":"README 的说明:\"While it's not fully optimized for production environments, it combines robust components that work well together for proof-of-concept projects\"。最后一次提交是在 2026-02-26,距本次检查 252 天,还有 61 个未解决的 issue。","source":"ai","title":"作者自己说还没准备好用于正式场景"}],"platforms":["windows","mac","linux"],"tools":["docker","python","gpu_nvidia"],"buildTools":["docker","gpu_nvidia"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenAI (only for Supabase's SQL editor helper)","Google sign-in for Supabase (optional)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["local_ai","automation","ai_agents"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"n8n-cloud","name":"Cloud","confirmed":true},{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"langsmith","name":"LangSmith","confirmed":false},{"slug":"make","name":"Make","confirmed":false}],"declaredAlternatives":[],"stars":3790,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":3790},{"date":"2026-11-07","stars":3790},{"date":"2026-11-08","stars":3790},{"date":"2026-11-09","stars":3790},{"date":"2026-11-10","stars":3790}],"firstPublishedAt":"2026-11-05T22:43:48.326Z","commitSha":"2c541913c0f97d7e2e4de3ee7e3f790fc63ce613","verifiedAt":"2026-11-05T22:29:27.244Z","lastCommitAt":"2026-02-26T16:28:45Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"One Docker package that starts n8n, Ollama, Open WebUI, Supabase and more together, to build AI agents on your own machine.","summary":"Self-hosted AI Package is a ready-made Docker Compose set-up that starts a whole local AI kit in one go: Ollama for models on your own machine, Open WebUI to chat with them, n8n for low-code automation with example agent workflows included, Supabase as the database, Qdrant and Neo4j for search and graphs, SearXNG for web search, Langfuse for monitoring and Caddy for HTTPS. You install Docker, Python and Git, fill in a long file of passwords, then run one script. A graphics card makes the models usable; there is also a slower processor-only option.","easeRationale":"Windows, Mac and Linux are all a D and there is no release to download. On each one you install Docker, Python and Git, clone the repository, copy `.env.example` to `.env` and generate about fifteen passwords and secret keys by hand, then run `python start_services.py --profile gpu-nvidia` (or `cpu`, `gpu-amd`, `none`). Macs with Apple chips cannot pass the graphics card to Docker, so they run on the processor or point at an Ollama installed outside Docker.","costNote":"Every piece is open-source software running on your own machine, so there is nothing to pay. The same tools (n8n, Supabase, Langfuse, Flowise) also sell hosted versions, which this package replaces.","notVerified":["How much disk and memory the twenty-odd containers plus the two downloaded models really take","Whether the whole stack still comes up today, since every image follows the moving `latest` tag and Supabase keeps adding required variables","Whether the example n8n workflows work after import, as each needs credentials created by hand","What the `start_services.py` script does, since its code was not read here","How the models perform without a graphics card, on the `cpu` profile"]},"zh-Hans":{"summaryShort":"一个 Docker 软件包,同时启动 n8n、Ollama、Open WebUI、Supabase 等,用于在自己的电脑上搭建 AI 智能体。","summary":"Self-hosted AI Package 是一套现成的 Docker Compose 配置,一次性启动整套本地 AI 工具:用于在自己电脑上运行模型的 Ollama、用于和模型聊天的 Open WebUI、带示例智能体工作流的低代码自动化工具 n8n、作为数据库的 Supabase、用于搜索和图谱的 Qdrant 与 Neo4j、用于网页搜索的 SearXNG、用于监控的 Langfuse,以及用于 HTTPS 的 Caddy。你需要安装 Docker、Python 和 Git,填写一个很长的密码文件,然后运行一个脚本。显卡能让模型可用;也有一个较慢、只用处理器的选项。","easeRationale":"Windows、Mac 和 Linux 的易用度评级都是 D,也没有可下载的发行版。在每个系统上,你都要安装 Docker、Python 和 Git,克隆仓库,把 `.env.example` 复制为 `.env`,手动生成大约十五个密码和密钥,然后运行 `python start_services.py --profile gpu-nvidia`(或 `cpu`、`gpu-amd`、`none`)。搭载 Apple 芯片的 Mac 无法把显卡传给 Docker,所以只能用处理器运行,或者指向一个安装在 Docker 之外的 Ollama。","costNote":"每个组件都是运行在你自己电脑上的开源软件,因此不需要付费。同样这些工具(n8n、Supabase、Langfuse、Flowise)也出售托管版本,而这个软件包可以替代它们。","notVerified":["二十多个容器加上两个下载的模型,实际占用多少磁盘和内存","整套系统今天是否还能正常启动,因为每个镜像都跟随会变化的 `latest` 标签,而 Supabase 一直在增加必需的变量","导入后 n8n 的示例工作流是否能用,因为每个都需要手动创建凭据","`start_services.py` 脚本做了什么,因为这里没有读过它的代码","在 `cpu` 配置下,没有显卡时模型表现如何"]}}},{"id":"R_kgDOVRjs8w","slug":"colinhacks__reylo","name":"Reylo","owner":"colinhacks","fullName":"colinhacks/reylo","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"needs_a_beta_version_of_typescript","severity":"attention","evidence":"README 的 \"Setup\" 部分写道：\"TypeScript 7.1 (currently in beta). The first release with content mappers.\" package.json 锁定了 \"typescript\": \"^7.1.0-dev.20260918.1\"。预发布版本的编译器在正式发布前还可能发生变化。","source":"ai","title":"需要一个 TypeScript 的 beta 版本"},{"code":"type_checking_runs_the_package_code","severity":"attention","evidence":"README 写道：\"Then type-check with --runExternalCode, which lets tsc run content mappers\"，package.json 中声明了 \"contentMapper\": { \"exec\": [\"node\", \"src/mapper.ts\"] }。在你的编辑器或 CI 里做类型检查，会运行这个包自己的程序，这不是类型检查通常的工作方式。","source":"ai","title":"类型检查会运行这个包自己的代码"},{"code":"editor_connects_to_your_database","severity":"info","evidence":"README 的 \"What Reylo infers\" 部分写道：\"To type a query, Reylo has your dev database prepare it, without running it, and reads back the type of every column and parameter.\" 它会从你环境变量或 .env 文件中的 DATABASE_URL 读取连接字符串，所以你在打字的时候，编辑器就会打开一个数据库连接。","source":"ai","title":"编辑器会连接到你的数据库"},{"code":"early_version_moving_fast","severity":"info","evidence":"该仓库创建于 2026-10-21，当前版本是 0.7.0，已发布 9 个版本，只有最新的那个版本附带平台二进制文件。README 把 DuckDB 支持标记为 \"(preview)\"，目前有 15 个未关闭的 issue。","source":"ai","title":"早期版本，迭代很快"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["node","rust"],"disk":{"minMB":7,"maxMB":22,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"prisma","name":"Prisma","confirmed":true}],"declaredAlternatives":[],"stars":1575,"downloads":611,"downloadsWeek":484,"downloadsDays":7,"growth":89,"growthDays":7,"history":[{"date":"2026-10-27","stars":1341},{"date":"2026-10-28","stars":1365},{"date":"2026-10-29","stars":1403},{"date":"2026-10-30","stars":1425},{"date":"2026-10-31","stars":1444},{"date":"2026-11-01","stars":1464},{"date":"2026-11-02","stars":1477},{"date":"2026-11-03","stars":1486},{"date":"2026-11-04","stars":1497},{"date":"2026-11-05","stars":1514},{"date":"2026-11-06","stars":1527},{"date":"2026-11-07","stars":1538},{"date":"2026-11-08","stars":1544},{"date":"2026-11-09","stars":1567},{"date":"2026-11-10","stars":1575}],"firstPublishedAt":"2026-10-27T18:40:27.831Z","commitSha":"f52c637ccfb40b9372d45131190fe8f418420936","verifiedAt":"2026-11-03T02:38:57.489Z","lastCommitAt":"2026-11-10T17:41:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A TypeScript tool that lets you import .sql files straight into your code, with the column types read from your own database.","summary":"Reylo is for programmers who write SQL by hand in a TypeScript project. You put a query in a .sql file, import it like any other module, and the editor knows exactly which columns and types come back, because Reylo asks your development database to prepare the query and reads the answer. It writes no generated files. It works with Postgres, MySQL, SQLite and DuckDB and with your existing database driver. You need Node 24 or later, a running development database, and the beta of TypeScript 7.1, which is the first version with the feature Reylo is built on.","easeRationale":"There is a ready engine binary for Windows, Mac and Linux, so no platform is harder than another, but this is a programmer's tool: you install it with npm, add a beta version of TypeScript, edit tsconfig.json and package.json, put a database address in a .env file and keep a development database running. There is no window to open.","notVerified":["Whether the types it works out are right for your own schema, database version and queries","Whether the package installed from npm contains the same engine binary as the GitHub release files","Whether it keeps working as the beta TypeScript changes before its final release","How fast the editor stays on a large schema, since every query is sent to the database to be prepared","What it does when the development database is unreachable"]},"zh-Hans":{"summaryShort":"一个 TypeScript 工具，让你直接在代码里导入 .sql 文件，列类型则从你自己的数据库中读取。","summary":"Reylo 面向那些在 TypeScript 项目里手写 SQL 的程序员。你把一条查询放进一个 .sql 文件，像导入其他模块一样导入它，编辑器就能准确知道返回哪些列以及它们的类型，因为 Reylo 会让你的开发数据库准备这条查询并读取结果。它不会生成任何文件。它支持 Postgres、MySQL、SQLite 和 DuckDB，并可以配合你现有的数据库驱动使用。你需要 Node 24 或更新版本、一个正在运行的开发数据库，以及 TypeScript 7.1 的 beta 版本，这是第一个支持 Reylo 所依赖功能的版本。","easeRationale":"Windows、Mac 和 Linux 都有现成的引擎二进制文件，所以没有哪个平台会比其他的更麻烦，但这是一个程序员工具：你用 npm 安装它，添加一个 TypeScript 的 beta 版本，编辑 tsconfig.json 和 package.json，在 .env 文件里填上数据库地址，并保持一个开发数据库在运行。没有窗口可以打开。","notVerified":["它推断出的类型是否适用于你自己的数据库结构、数据库版本和具体查询","从 npm 安装的包里包含的引擎二进制文件，是否与 GitHub 发布文件中的一致","在 TypeScript 的 beta 版发布正式版之前不断变化的情况下，它是否能继续正常工作","在一个大型数据库结构上，编辑器的响应速度如何，因为每条查询都要发送到数据库去准备","当开发数据库无法访问时，它会怎么处理"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMzYzNDk0NjI=","slug":"colombiapython__social-media-automation","name":"Social Media Automation","owner":"ColombiaPython","fullName":"ColombiaPython/social-media-automation","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub 没有检测到许可证,许可证文件、LICENSES 文件夹和 winget manifest 里也都没有","source":"check","title":"没有许可证"},{"code":"your_password_typed_into_the_code","severity":"attention","evidence":"对这三个平台,README 都写着:\"There is a main method in the script, you need to edit and provide your user and password\"。你的 Facebook、LinkedIn 和 Twitter 密码最终会以明文形式写在 `fbposter.py`、`linkedinpost.py` 和 `tweetpost.py` 里面,之后一次 `git commit` 就会把它们公开出去。","source":"ai","title":"密码直接写进代码里"},{"code":"posting_by_remote_controlling_the_browser","severity":"attention","evidence":"项目描述是 \"Selenium Scripts To Post In Facebook, Twitter And Linkedin\":这些脚本会用你的身份登录一个由它们操控的 Firefox 窗口,而不是通过平台自己的界面,发布内容时写的是 \"on the Facebook Groups you are member\"。Facebook、LinkedIn 和 X 都禁止这样自动化操作账号,通常的结果是账号被封。","source":"ai","title":"通过远程操控浏览器来发布内容"},{"code":"no_commits_for_eight_years","severity":"attention","evidence":"最后一次提交是在 2018-06-13,距本次检查 3066 天,没有未解决的 issue。说明要求 \"Python 3.6.x\",脚本点击操作依据的是 2018 年时 Facebook、LinkedIn 和 Twitter 页面的样子;这三个网站后来都已经改版,X 当时甚至还不叫这个名字。","source":"ai","title":"八年没有任何提交"},{"code":"instructions_that_do_not_add_up","severity":"info","evidence":"Twitter 部分写着 \"Download put in line 23 the path to this and put in line 22 the path to this\",没说 \"this\" 是什么,也没有链接;Facebook 和 LinkedIn 部分都要你编辑 \"line 23\",但没展示是哪个文件。仓库里没有 `requirements.txt`,也没有任何配置示例,只有三个 `.py` 文件。","source":"ai","title":"说明前后不一致"}],"platforms":["linux","mac"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true}],"declaredAlternatives":[],"stars":109,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":1,"growthDays":6,"history":[{"date":"2026-11-06","stars":108},{"date":"2026-11-07","stars":108},{"date":"2026-11-08","stars":109},{"date":"2026-11-09","stars":109},{"date":"2026-11-10","stars":109}],"firstPublishedAt":"2026-11-05T22:43:48.334Z","commitSha":"a7e3bb1f025a9ea3d678bd51f796f3d949796168","verifiedAt":"2026-11-05T22:30:04.264Z","lastCommitAt":"2018-06-13T23:45:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Three Python scripts that drive Firefox to log in and post an image with text to Facebook groups, LinkedIn and Twitter.","summary":"Three short Python scripts from 2018, one per network. Each opens Firefox through Selenium, signs in with the login you type into the script itself, and posts an image with text: to the Facebook groups you belong to, to LinkedIn groups, company pages you administer or your own profile, and to Twitter. There is no interface and no settings file: you edit the `main` method of each file to set your user, your password, the picture's path and the group links. For someone comfortable with Python; untouched since June 2018.","easeRationale":"A D on Linux and Mac, the two systems the instructions fit (`$ source venv/bin/activate`): you install Python 3.6, pip and virtualenv, then Selenium, download Mozilla's geckodriver separately, and open the files in an editor to write your login and the picture's path into the code before `python fbposter.py` does anything. There is nothing to download and no Windows instructions.","costNote":"Three script files with nothing to pay and no service in between; it uses your own accounts.","notVerified":["Whether any of the three scripts still finds the buttons on today's Facebook, LinkedIn and X pages","Whether they run on a current Python, since the instructions name Python 3.6","What the scripts do with the login once it is written into them, as their code was not read here","Whether sign-in works at all now that the three networks ask for a second step or a code"]},"zh-Hans":{"summaryShort":"三个 Python 脚本,操控 Firefox 登录并把带文字的图片发到 Facebook 群组、LinkedIn 和 Twitter。","summary":"三个 2018 年写的简短 Python 脚本,每个网络一个。每个脚本都通过 Selenium 打开 Firefox,用你直接写进脚本里的账号登录,然后发一张带文字的图片:发到你所属的 Facebook 群组、LinkedIn 群组、你管理的公司主页或自己的主页,以及 Twitter。没有界面,也没有配置文件:你需要编辑每个文件里的 `main` 方法,填入你的账号、密码、图片路径和群组链接。适合熟悉 Python 的人;自 2018 年 6 月以来没有再更新。","easeRationale":"在 Linux 和 Mac 上评级是 D,说明文档也只适合这两个系统(`$ source venv/bin/activate`):你要安装 Python 3.6、pip 和 virtualenv,再安装 Selenium,另外下载 Mozilla 的 geckodriver,然后在编辑器里打开文件,把你的账号和图片路径写进代码,之后才能运行 `python fbposter.py`。没有任何可下载的版本,也没有 Windows 的说明。","costNote":"三个脚本文件,不需要付费,中间也没有任何服务;用的是你自己的账号。","notVerified":["这三个脚本是否还能在今天的 Facebook、LinkedIn 和 X 页面上找到对应的按钮","它们能否在当前版本的 Python 上运行,因为说明里写的是 Python 3.6","账号密码写进脚本之后,脚本对它们做了什么,因为这里没有读过脚本的代码","既然这三个平台现在都要求第二步验证或验证码,登录是否还能成功"]}}},{"id":"R_kgDOIygeJg","slug":"comfy-org__comfyui","name":"ComfyUI","owner":"Comfy-Org","fullName":"Comfy-Org/ComfyUI","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"D"},"risk":"attention","signals":[{"code":"custom_nodes_are_other_peoples_code_running_on_your_computer","severity":"attention","evidence":"README宣传\"Extend ComfyUI with custom nodes\"和ComfyUI-Manager，\"an extension that allows you to easily install, update, and manage custom nodes\"，可用`python main.py --enable-manager`开启。自定义节点就是陌生人写的Python代码，运行时使用的是你自己的权限，而管理器自身的选项里还提到\"background features like security checks\"这类后台功能。","source":"ai","title":"自定义节点其实是别人的代码，会在你电脑上运行"},{"code":"a_shared_image_can_carry_a_whole_workflow","severity":"attention","evidence":"README的Notes部分写道：\"Dragging a generated png on the webpage or loading one will give you the full workflow including seeds that were used to create it.\" 这方便你直接复制别人的设置，但从网上下载的工作流可能指向你没有检查过的自定义节点和文件。","source":"ai","title":"分享出去的图片可能带着整个工作流"},{"code":"models_are_a_separate_multi_gigabyte_download","severity":"info","evidence":"在你添加模型之前，什么都做不出来：\"Put your SD checkpoints (the huge ckpt/safetensors files) in: models/checkpoints\"，\"Put your VAE in: models/vae\"，更大的模型\"have multiple files\"，还各有自己的子文件夹。光是安装本身就估计要1945到5836MB，这还不算任何模型。","source":"ai","title":"模型是单独的几GB大文件下载"},{"code":"it_is_a_web_server_on_your_machine","severity":"info","evidence":"你用`python main.py`启动它，然后在浏览器里使用；README说明了如何添加自签名证书（`--tls-keyfile key.pem --tls-certfile cert.pem`），并提醒这种证书\"not appropriate for shared/production use\"。它还对外提供\"a local API for integrating workflows into applications\"，所以最好不要把这个端口开放给其他电脑访问。","source":"ai","title":"它是在你电脑上跑的一个网页服务器"},{"code":"moves_fast_and_breaks_extensions","severity":"info","evidence":"README的Release Process部分写道：\"Commits outside of the stable release tags may be very unstable and break many custom nodes\"，发布周期是每周一次，已有164个release和5391个未关闭的issue。界面本身放在另一个独立仓库里，每两周合并一次。","source":"ai","title":"更新很快，也容易弄坏扩展"},{"code":"each_model_has_its_own_licence","severity":"info","evidence":"这个程序本身使用GPL-3.0许可证，但README列出了几十个可下载的第三方模型（Flux、Qwen Image、Wan、HunyuanVideo等），还有连接到封闭模型的合作节点，比如\"Nano Banana, Seedance, Hunyuan3D\"。你用生成的图片和视频能做什么，取决于每个模型自己的条款，和这份许可证无关。","source":"ai","title":"每个模型都有自己的许可条款"}],"platforms":["windows","mac","linux"],"tools":["gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":1945,"maxMB":5836,"basis":"installer"},"apiKeys":["Comfy API (only for the optional partner nodes that call closed models)"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image","video","audio"],"alternativeTo":[{"slug":"midjourney","name":"Midjourney","confirmed":true},{"slug":"runway","name":"Runway","confirmed":true},{"slug":"adobe-firefly","name":"Firefly","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true},{"slug":"photoshop","name":"Photoshop","confirmed":false}],"declaredAlternatives":[],"stars":146169,"downloads":10878255,"downloadsWeek":221567,"downloadsDays":7,"growth":1297,"growthDays":7,"history":[{"date":"2026-10-26","stars":143287},{"date":"2026-10-27","stars":143570},{"date":"2026-10-28","stars":143796},{"date":"2026-10-29","stars":143993},{"date":"2026-10-30","stars":144208},{"date":"2026-10-31","stars":144364},{"date":"2026-11-01","stars":144498},{"date":"2026-11-02","stars":144675},{"date":"2026-11-03","stars":144872},{"date":"2026-11-04","stars":145022},{"date":"2026-11-05","stars":145216},{"date":"2026-11-06","stars":145423},{"date":"2026-11-07","stars":145583},{"date":"2026-11-08","stars":145778},{"date":"2026-11-09","stars":145983},{"date":"2026-11-10","stars":146169}],"firstPublishedAt":"2026-10-27T01:10:12.617Z","commitSha":"fbca05a62e23c1b08b6e77f408db2742127a9929","verifiedAt":"2026-10-31T01:43:46.054Z","lastCommitAt":"2026-11-10T20:57:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Creates and edits images, video and audio with AI models on your own computer, by wiring boxes together on a canvas.","summary":"ComfyUI runs open AI models on your own machine through a canvas of connected boxes: each box is a step (load a model, write the prompt, save the picture) and you join them into a workflow you can save and reuse. It handles image, video, audio, 3D and text models, plus inpainting, upscaling and masks. It runs offline and downloads nothing unless you ask. You supply the models yourself, as multi-gigabyte files placed in the right folders, and you need a reasonably capable graphics card. Windows and macOS have a desktop installer; Linux is a manual install.","easeRationale":"Windows and macOS have the desktop installer the README calls \"the easiest and best way to use ComfyUI for new users\", so the work left is downloading model files into the right folders and having a graphics card that can cope. The Windows portable 7z in this repository is 1.4 to 1.9 GB and the README says it \"is not recommended for regular users\". Linux is harder: there is no installer, you clone the repository, choose and install the right PyTorch for NVIDIA, AMD ROCm or Intel, run `pip install -r requirements.txt` and start it with `python main.py`.","costNote":"The program is free and GPL-3.0. Optional paid extras exist: the \"API nodes\" that reach closed models (switched off with --offline) and the company's paid Comfy Cloud for people without a suitable computer. The models themselves are usually free downloads, each with its own licence.","notVerified":["Whether your graphics card is enough: the README claims the biggest models run on 4 GB of video memory plus 8 GB of RAM, which is the project's own measurement","Real disk use once you add models, which can be far beyond the 1.9 to 5.8 GB of the install itself","How fast generation is on your hardware, and whether the \"fastest local inference engine in the world\" claim holds for you","What any custom node or downloaded workflow does on your computer","Whether the macOS and Linux routes work as smoothly as Windows, since only Windows files are published in this repository","What the optional paid API nodes and Comfy Cloud cost, which this repository does not state"]},"zh-Hans":{"summaryShort":"在你自己的电脑上用AI模型创作和编辑图片、视频和音频，只需把方块拖动连接到画布上即可。","summary":"ComfyUI 让你在自己的电脑上运行开放的AI模型，操作方式是在一块画布上连接方块：每个方块代表一步（加载模型、输入提示词、保存图片），把它们连起来就组成一个工作流，可以保存并反复使用。它支持图片、视频、音频、3D和文本模型，还有局部重绘（inpainting）、放大（upscaling）和蒙版功能。它可以离线运行，不会主动下载任何东西，除非你自己要求。模型需要你自己准备，是放在指定文件夹里的几GB大文件，而且你需要一块性能还不错的显卡。Windows和macOS提供桌面安装程序；Linux则需要手动安装。","easeRationale":"Windows和macOS都有README所说的\"the easiest and best way to use ComfyUI for new users\"的桌面安装程序，剩下要做的就是把模型文件下载到正确的文件夹，并准备一块能应付的显卡。本仓库中Windows便携版的7z压缩包大小在1.4到1.9GB之间，README说它\"is not recommended for regular users\"。Linux更麻烦：没有安装程序，需要自己克隆仓库，根据NVIDIA、AMD ROCm或Intel选择并安装对应的PyTorch，运行`pip install -r requirements.txt`，再用`python main.py`启动。","costNote":"这个程序是免费的，使用GPL-3.0许可证。还有一些可选的付费功能：连接到封闭模型的\"API节点\"（可用--offline关闭），以及公司提供的付费Comfy Cloud服务，适合没有合适电脑的人使用。模型本身通常可以免费下载，但每个模型都有自己的许可条款。","notVerified":["你的显卡是否够用：README声称最大的模型只需要4GB显存加8GB内存就能运行，这是项目自己给出的数据","加上模型之后实际占用的硬盘空间，可能远超安装本身的1.9到5.8GB","在你自己的硬件上生成速度有多快，以及它所说的\"全球最快的本地推理引擎\"是否对你也成立","任何自定义节点或下载来的工作流在你电脑上到底会做什么","macOS和Linux的使用体验是否和Windows一样顺畅，因为这个仓库里只发布了Windows的文件","可选的付费API节点和Comfy Cloud具体收费多少，这个仓库没有说明"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNDE1ODIzNw==","slug":"copy__v86","name":"v86","owner":"copy","fullName":"copy/v86","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"web":"A"},"risk":"attention","signals":[{"code":"old_systems_need_images_you_may_not_own","severity":"attention","evidence":"演示启动了 Windows 1.01、95、98 和 2000，致谢名单里感谢了\"WinWorld\"、\"OS/2 Museum\"和\"ArchiveOS\"作为\"sources of some old operating systems\"。模拟器本身是一个普通程序，但这些微软系统仍受版权保护，而那些存档网站提供的是现成的拷贝：在没有许可的情况下运行它们涉及版权问题，在一些国家下载它们本身也是如此。仓库本身不包含这些镜像（\"The disk images for testing are not included in this repository\"）。","source":"ai","title":"老系统需要你可能无权使用的镜像"},{"code":"an_embedded_pc_can_reach_the_network","severity":"info","evidence":"被模拟的硬件包括\"An NE2000 (RTL8390) PCI network card\"、\"Various virtio devices: Filesystem, network\"和一个拨号调制解调器，示例里还有\"TCP Terminal (fetch-based networking)\"以及通过 Broadcast Channel API 在多个标签页之间联网。因此，嵌入 v86 的网站会在你的标签页里运行一整台机器，而这台机器也可以通过你的浏览器对外通信。","source":"ai","title":"嵌入式电脑也能连上网络"},{"code":"the_emulated_pc_is_incomplete","severity":"info","evidence":"README 列出了缺失的部分：\"Task gates, far calls in protected mode\"、\"Multicore\"、\"64-bit extensions\"，还有一些异常和三角函数精度方面的警告。兼容性列表对其余部分也很坦诚：\"Plan 9 doesn't work\"、\"OS/2 doesn't work\"，不支持 64 位 Linux 内核，\"Windows XP, Vista and 8 work under certain conditions\"，以及 Windows 9x 的几个未解决的启动问题。","source":"ai","title":"被模拟的电脑并不完整"},{"code":"one_rolling_release_published_automatically","severity":"info","evidence":"唯一的发布版本标记为\"latest\"（2026-10-30），没有版本号，package.json 写的是版本 0.5，Cargo.toml 写的是 0.1.0。这个 npm 包\"is published automatically from this repository via GitHub Actions ... on pushes to master\"，所以你安装的内容会随着项目 master 分支的变化而变化。","source":"ai","title":"只有一个自动发布的滚动版本"},{"code":"project_refuses_ai_written_contributions","severity":"info","evidence":"一个关于生成式 AI 的章节写道：\"At this time, v86 does not accept pull requests or issues written partially or entirely by generative AI tools. Exceptions may be made on a case-by-case basis.\"在报告错误或提交补丁之前，值得了解这一点。","source":"ai","title":"项目拒绝接受 AI 撰写的贡献"}],"platforms":["web"],"tools":[],"buildTools":["node","rust"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"BSD-2-Clause","category":"permissive","detectedFrom":"github"},"categories":["development","games"],"alternativeTo":[],"declaredAlternatives":[],"stars":23672,"downloads":424,"downloadsWeek":154,"downloadsDays":5,"growth":11,"growthDays":6,"history":[{"date":"2026-11-05","stars":23667},{"date":"2026-11-06","stars":23664},{"date":"2026-11-07","stars":23667},{"date":"2026-11-08","stars":23669},{"date":"2026-11-09","stars":23669},{"date":"2026-11-10","stars":23672}],"firstPublishedAt":"2026-11-04T22:32:54.868Z","commitSha":"315b2b0b758703ef0e6cc7b4088d6068f83f3c20","verifiedAt":"2026-11-04T22:21:52.123Z","lastCommitAt":"2026-10-30T21:24:44Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs a whole old PC — DOS, Windows 95, Linux — inside a browser tab, with nothing to install.","summary":"v86 is an emulator of an old PC written to run inside a web page: it imitates a Pentium-era processor, screen, keyboard, mouse, sound card, disks and network card, and turns the guest's machine code into WebAssembly while it runs. The project page has ready demos that boot FreeDOS, Windows 95, 98 and 2000, Linux, ReactOS, Haiku and many hobby systems in the browser. For developers it is also a library you add to a web page (npm install v86) to put a working computer inside your own site. You must supply the disk image of the system you want to run.","easeRationale":"On the web it is as easy as possible: the Demos list in the README are links on copy.sh/v86 that boot a system in the browser with nothing installed, so any computer with a current browser gets an A. Building it yourself or embedding it in your own page is a different job — the README asks for make, Rust with the wasm32 target, a matching clang, Java and a recent Node — but that is developer work, not what an end user does.","notVerified":["How fast a given system feels in the browser and how much memory one tab takes","Whether a particular disk image boots, since the compatibility list has many conditions and open issues","Whether saving and restoring the machine state works reliably across browser versions","What the demo page on copy.sh keeps or logs when you use it","Whether sound, mouse and networking work in the system you want to run"]},"zh-Hans":{"summaryShort":"在浏览器标签页里运行一整台老式电脑——DOS、Windows 95、Linux——无需安装任何东西。","summary":"v86 是一个为在网页中运行而编写的老式电脑模拟器：它模拟了 Pentium 时代的处理器、屏幕、键盘、鼠标、声卡、磁盘和网卡，并在运行时把访客系统的机器码转换成 WebAssembly。项目主页提供了现成的演示，可以在浏览器里启动 FreeDOS、Windows 95、98 和 2000、Linux、ReactOS、Haiku 以及许多爱好者系统。对开发者来说，它也是一个可以添加到网页中的库（npm install v86），把一台可用的电脑放进你自己的网站。你需要自己提供想要运行系统的磁盘镜像。","easeRationale":"在网页上这已经是最简单的方式：README 里的演示列表是 copy.sh/v86 上的链接，无需安装任何东西即可在浏览器里启动系统，所以任何装有现代浏览器的电脑都能拿到 A。自己编译或把它嵌入自己的页面是另一回事——README 要求 make、带 wasm32 目标的 Rust、匹配的 clang、Java 和较新的 Node——但那是开发者的工作，不是普通用户要做的事。","notVerified":["某个系统在浏览器里运行的实际速度，以及一个标签页占用多少内存","某个特定磁盘镜像是否能启动，因为兼容性列表里有很多条件和未解决的问题","保存和恢复虚拟机状态是否在不同浏览器版本之间都能可靠工作","你使用 copy.sh 上的演示页面时，它会保留或记录什么内容","声音、鼠标和网络功能是否能在你想运行的系统里正常工作"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyNjU2MTI0NDA=","slug":"coqui-ai__tts","name":"Coqui TTS","owner":"coqui-ai","fullName":"coqui-ai/TTS","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"E","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发行是在 2023-12-12（1057 天前）","source":"check","title":"发行版已经过时"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：pylint：GHSA-22qq-2fwx-7p67；bokeh：GHSA-793v-589g-574v, PYSEC-2026-1223","source":"check","title":"已知漏洞"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.9/10","source":"check","title":"scorecard 评分偏低"},{"code":"copying_someone_s_voice","severity":"attention","evidence":"README 自己的示例就是用一段音频文件克隆声音 —— `tts.tts_to_file(\"This is voice cloning.\", speaker_wav=\"my/cloning/audio.wav\", ...)` —— 新闻板块还宣传“🐶Bark ... with unconstrained voice cloning”，另外还有一个声音转换的例子，把一个人的录音变成另一个人的声音。一般来说，在没有本人同意的情况下制作听起来像真人的录音，在许多国家受到法律限制。","source":"ai","title":"克隆某个人的声音"},{"code":"no_commits_for_almost_three_years","severity":"attention","evidence":"最后一次提交是在 2024-02-10，距本次检查 997 天，最新版本 v0.22.0 发布于 2023-12-12，在 46,178 颗星的情况下只有 2 个未关闭的 issue，这正是议题追踪被关闭的样子。该项目没有被标记为已归档，但这里已经没有人在修复任何问题。","source":"ai","title":"将近三年没有提交记录"},{"code":"needs_an_older_python","severity":"info","evidence":"setup.py 会在 `python >= 3.9 and < 3.12` 之外的版本报错，requirements.txt 固定了旧版库（Python 3.10 及以下用 `numpy==1.22.0`、`bokeh==1.4.0`、`pylint==2.10.2`），所以需要比当前系统自带版本更旧的 Python。","source":"ai","title":"需要更旧版本的 python"},{"code":"models_are_downloaded_at_run_time","severity":"info","evidence":"代码按 id 命名模型（例如 `tts_models/multilingual/multi-dataset/xtts_v2`），并在第一次使用时获取它们，所以真正运行的内容是安装之后才从网上下载的，而且适用的是每个模型各自的许可证，而不是仓库的 MPL-2.0。","source":"ai","title":"模型在运行时下载"},{"code":"tracking_pixel_in_the_readme","severity":"info","evidence":"README 嵌入了 `https://static.scarf.sh/a.png?x-pxid=cf317fe7-2188-4721-bc01-124bb5d5dbb2`，这是 Scarf 分析服务的一张图片，所以仅仅打开项目页面就会向第三方报告一次访问。","source":"ai","title":"readme 中的追踪像素"}],"platforms":["windows","mac","linux"],"tools":["python","docker","gpu_any"],"buildTools":["python","docker","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["audio","local_ai","development"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"descript","name":"Descript","confirmed":false}],"declaredAlternatives":[],"stars":46208,"downloads":7019881,"downloadsWeek":0,"downloadsDays":6,"growth":29,"growthDays":7,"history":[{"date":"2026-11-04","stars":46181},{"date":"2026-11-05","stars":46183},{"date":"2026-11-06","stars":46185},{"date":"2026-11-07","stars":46189},{"date":"2026-11-08","stars":46193},{"date":"2026-11-09","stars":46195},{"date":"2026-11-10","stars":46208}],"firstPublishedAt":"2026-11-04T01:49:35.995Z","commitSha":"dbf1a08a0d4e47fdad6172e433eeb34bc6b13b4e","verifiedAt":"2026-11-04T01:45:30.655Z","lastCommitAt":"2024-02-10T14:20:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns written text into spoken audio on your own computer, and can copy a voice from a short recording.","summary":"Coqui TTS is a Python toolkit that reads text aloud with ready-made models in many languages, and can imitate a voice from a sample recording (the XTTS and YourTTS models) or convert one voice into another. It also has the training code for making your own voice models. You use it from the terminal with the `tts` command, from Python code, or through a small web page served by its Docker image. It is a tool for people comfortable with Python; there is no installer, and the project has not been updated since February 2024.","easeRationale":"None of the 98 releases has a file to download. On Linux, which the README says it is tested on (Ubuntu 18.04), it is `pip install TTS` on a Python between 3.9 and 3.11, then a command such as `tts --text \"...\" --out_path speech.wav` that downloads the model on first use; a Docker image with a small web page is the other route, and a graphics card is what makes it usable in practice. macOS follows the same pip route without a CUDA card. Windows is worse than a D because the README has no instructions of its own: it links to an answer someone wrote on StackOverflow.","costNote":"The code is free software under MPL-2.0 and runs on your own machine, so there is nothing to pay. The pretrained models are downloaded separately and each carries its own terms.","notVerified":["Whether `pip install TTS` still resolves today, with the old pinned libraries and a Python older than 3.12","The licence of each pretrained model, which is separate from the repository's MPL-2.0 and can forbid commercial use","How good the result sounds and how closely a cloned voice matches, which only listening can tell","Real disk and memory use after the models download, since the figure here is an estimate","Whether it runs usably without an NVIDIA graphics card"]},"zh-Hans":{"summaryShort":"在你自己的电脑上把文字转换成语音朗读，还能根据一段简短录音模仿某个人的声音。","summary":"Coqui TTS 是一个 Python 工具包，能用多种语言的现成模型朗读文字，还能根据一段样本录音模仿某个声音（XTTS 和 YourTTS 模型），或者把一个声音转换成另一个声音。它还包含训练代码，可以用来制作你自己的语音模型。你可以在终端里用 `tts` 命令使用它，也可以在 Python 代码中调用，或者通过它的 Docker 镜像提供的一个小型网页来使用。这是一个适合熟悉 Python 的人使用的工具；没有安装程序，而且该项目自 2024 年 2 月以来没有更新过。","easeRationale":"98 个发行版中没有一个附带可下载的文件。在 README 声称测试过的 Linux（Ubuntu 18.04）上，先在 3.9 到 3.11 之间的 Python 版本上执行 `pip install TTS`，再运行类似 `tts --text \"...\" --out_path speech.wav` 的命令，它会在第一次使用时下载模型；另一条路是带有小型网页的 Docker 镜像，而显卡才是让它在实际使用中可行的关键。macOS 走的是同样的 pip 路线，但没有 CUDA 显卡。Windows 比 D 还差，因为 README 没有专门的说明：它只链接到有人在 StackOverflow 上写的一个答案。","costNote":"代码是基于 MPL-2.0 的自由软件，运行在你自己的机器上，所以没有什么需要付费的。预训练模型是单独下载的，每个模型都有自己的条款。","notVerified":["`pip install TTS` 今天是否仍能正常安装，因为它固定使用旧版库，并且需要低于 3.12 的 Python","每个预训练模型自己的许可证，它与仓库的 MPL-2.0 是分开的，可能禁止商业用途","结果听起来有多好，克隆出的声音有多像原声，这只有听过才知道","模型下载完成后的真实磁盘和内存占用，因为这里给出的数字只是估算","没有 NVIDIA 显卡时是否还能正常使用"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxODg2NjA2NjM=","slug":"corentinj__real-time-voice-cloning","name":"Real-Time Voice Cloning","owner":"CorentinJ","fullName":"CorentinJ/Real-Time-Voice-Cloning","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道直接传给shell执行：curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"安装命令把远程脚本直接传给shell执行"},{"code":"built_to_copy_a_real_person_s_voice","severity":"attention","evidence":"项目自己的描述是\"Clone a voice in 5 seconds to generate arbitrary speech in real-time\"，README中也说明它会根据\"a digital representation of a voice from a few seconds of audio\"构建声音模型，再用它朗读任意文字。一般来说，在未经本人同意的情况下制作听起来像真人的录音，在许多国家是受法律限制的。","source":"ai","title":"项目本身就是为了复制真人的声音而设计"},{"code":"windows_install_bypasses_the_script_policy","severity":"attention","evidence":"Windows上的安装步骤是`powershell -ExecutionPolicy ByPass -c \"irm https://astral.sh/uv/install.ps1 | iex\"`，它会关闭PowerShell的脚本保护，并执行当时下载的脚本。（Linux那一行用curl做了同样的事。）README也提供了`pip install -U uv`作为替代方案。","source":"ai","title":"Windows安装步骤绕过了脚本执行策略"},{"code":"author_says_it_is_out_of_date","severity":"info","evidence":"README中的\"Heads up\"部分写道：\"Like everything else in Deep Learning, this repo has quickly gotten old. Many SaaS apps (often paying) will give you a better audio quality than this repository will\"，并指向另一个项目（Chatterbox）以获得更好的音质。项目没有发布版本，而且固定使用的库版本都比较旧（torch 1.10、Pillow 8.4.0、scikit-learn 1.0.2）。","source":"ai","title":"作者自己说项目已经过时"},{"code":"needs_exactly_python_3_9","severity":"info","evidence":"pyproject.toml中设定`requires-python = \">=3.9,<3.10\"`，所以只能在Python 3.9上运行。uv工具会自动帮你获取这个版本，但更新的版本都无法使用。","source":"ai","title":"必须使用Python 3.9这一个版本"},{"code":"models_downloaded_at_run_time","severity":"info","evidence":"README中写道：\"Pretrained models are now downloaded automatically\"，来自https://huggingface.co/CorentinJ/SV2TTS。真正用于生成语音的内容是安装之后才从网上下载的，torch相关的安装包也来自pyproject.toml中指定的download.pytorch.org索引。","source":"ai","title":"模型在运行时才会下载"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true}],"declaredAlternatives":[],"stars":60237,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":13,"growthDays":7,"history":[{"date":"2026-11-04","stars":60223},{"date":"2026-11-05","stars":60228},{"date":"2026-11-06","stars":60229},{"date":"2026-11-07","stars":60233},{"date":"2026-11-08","stars":60236},{"date":"2026-11-09","stars":60238},{"date":"2026-11-10","stars":60237}],"firstPublishedAt":"2026-11-04T01:49:36.007Z","commitSha":"890f3a03187195b9829db2079b75c2ba2ab0405c","verifiedAt":"2026-11-04T01:45:56.791Z","lastCommitAt":"2026-03-09T10:31:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from about five seconds of audio and then reads any text you type in that voice.","summary":"This is the code behind a master's thesis: you give it a few seconds of someone speaking, it builds a digital fingerprint of that voice, and then it reads text you type in a voice that resembles it. There is a small desktop window (the toolbox) and a command-line version, and the training scripts are included. It runs on Windows and Linux, needs ffmpeg plus Python set up through uv, and an NVIDIA card makes it usable. The author himself says the project has \"quickly gotten old\" and that paid services sound better.","easeRationale":"There is no release to download at all. On both Windows and Linux the README asks you to install ffmpeg, then install the uv tool, then run `uv run --extra cuda demo_toolbox.py` (or `--extra cpu` without an NVIDIA card), which sets up a Python environment and downloads the models on first use. It is terminal work on both, and the project only supports Windows and 64-bit Linux, not macOS.","costNote":"MIT code, pretrained models downloaded at no charge, and everything runs on your own machine.","notVerified":["Whether the install still completes today, given Python 3.9 and the old pinned libraries","How much the cloned voice really sounds like the original, which only listening can tell","Whether it is usable without an NVIDIA card, since a CPU option exists but the speed is not stated","Real disk use after the models and any dataset are downloaded, since the figure here is an estimate","Whether the automatic model download still works"]},"zh-Hans":{"summaryShort":"只需约五秒的录音样本就能复制一个人的声音，然后用这个声音朗读你输入的任何文字。","summary":"这是一篇硕士论文背后的代码：你提供几秒钟某人说话的录音，程序就会为这个声音建立一种\"数字指纹\"，然后用接近这个声音的语气朗读你输入的文字。项目提供一个小型桌面窗口（工具箱）和一个命令行版本，并附带训练脚本。它可以在Windows和Linux上运行，需要安装ffmpeg以及通过uv配置的Python环境，有NVIDIA显卡会更好用。作者自己说这个项目\"很快就过时了\"，而付费服务的音质更好。","easeRationale":"这里完全没有可直接下载的发布版本。在Windows和Linux上，README都要求先安装ffmpeg，再安装uv工具，然后运行`uv run --extra cuda demo_toolbox.py`（没有NVIDIA显卡则用`--extra cpu`），这会配置Python环境并在首次使用时下载模型。两个系统上都需要在终端里操作，而且项目只支持Windows和64位Linux，不支持macOS。","costNote":"代码采用MIT许可证，预训练模型可免费下载，所有东西都在你自己的电脑上运行。","notVerified":["现在是否仍能顺利完成安装，因为项目用的是Python 3.9和一批较旧的固定版本库","克隆出的声音到底有多像原声，这一点只能靠听来判断","没有NVIDIA显卡是否也能正常使用，虽然有CPU选项，但没有说明速度","下载模型和任何数据集之后实际占用的磁盘空间，因为这里给出的数字只是估算","模型的自动下载功能是否仍然可用"]}}},{"id":"R_kgDOQ6n2Jw","slug":"coreyhaines31__marketingskills","name":"Marketing Skills","owner":"coreyhaines31","fullName":"coreyhaines31/marketingskills","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"attention","signals":[{"code":"paid_partners_and_the_author_s_own_products_inside_the_library","severity":"attention","evidence":"README中写道\"[Verified Partners](tools/REGISTRY.md#verified-partners) fund the work\"，并列出了四个付费服务（Converly、Ploy、AIsa、Omnia），它们在仓库内都有自己的接入指南，另外还有一个\"Become a partner\"链接。开头的段落还推广了作者自己的代理公司、通讯订阅、培训课程和一款AI agent产品，每个都带有`?ref=marketingskills`链接。README声称合作伙伴不会影响核心技能给出的建议；但这只是作者自己的说法，你收到的建议仍可能指向这些付费工具。","source":"ai","title":"资源库里包含付费合作伙伴和作者自己的产品"},{"code":"skills_let_the_agent_touch_your_live_accounts","severity":"attention","evidence":"README提供了一个\"guided tool connection workflow (tools/SETUP.md) for scoped credentials\"，而`conversion-tracking`、`ads`、`analytics`和`marketing-loops`（\"a repeatable loop an AI agent runs on a\" schedule）这些技能负责修改真实的跟踪数据、广告投放和已发布的页面。按照这些文件执行的助手，会用你给它的凭证，在真正花钱的账户上进行操作。","source":"ai","title":"这些技能让AI代理可以直接操作你的真实账户"},{"code":"some_skills_cover_regulated_practices","severity":"info","evidence":"这个包里包含`cold-email`（\"Write and run B2B cold outbound\"）、`prospecting`（建立待联系人员名单）、`sms`营销和`programmatic-seo`（\"SEO-driven pages at scale\"）。在很多国家，未经同意发送邮件和短信是受法律限制的，而批量生成的页面也违反一些搜索引擎的规则。","source":"ai","title":"部分技能涉及受法律监管的做法"},{"code":"changes_fast_and_renames_things","severity":"info","evidence":"这个仓库创建于2026-01-15，已经发布了100个release，最新的一个是在本次检查前四天发布的。2.0版本重命名了17个技能并合并了另外两个，README还得专门说明要用`rm -rf`手动删除旧文件夹，因为新文件夹是和旧的并排安装的。","source":"ai","title":"更新很快，而且会重命名内容"},{"code":"anyone_can_propose_a_skill","severity":"info","evidence":"README中欢迎提交pull request（\"Contributions welcome!\"），目前有85个未关闭的issue。这些文件是你的助手要遵循的指令，所以这里合并的任何改动都会影响所有安装或更新这个包的人。","source":"ai","title":"任何人都可以提议新增技能"}],"platforms":["windows","mac","linux","web"],"tools":["node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["An AI assistant account — Claude, OpenAI Codex, Cursor or similar","Your own marketing accounts and tools (analytics, ads, SEO data, partner services) if you want the assistant to read them"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","social_marketing","productivity"],"alternativeTo":[{"slug":"jasper","name":"Jasper","confirmed":true},{"slug":"surfer-seo","name":"SEO","confirmed":true}],"declaredAlternatives":[],"stars":59580,"downloads":68504,"downloadsWeek":15984,"downloadsDays":6,"growth":1266,"growthDays":7,"history":[{"date":"2026-11-04","stars":58507},{"date":"2026-11-05","stars":58713},{"date":"2026-11-06","stars":58900},{"date":"2026-11-07","stars":59046},{"date":"2026-11-08","stars":59213},{"date":"2026-11-09","stars":59398},{"date":"2026-11-10","stars":59580}],"firstPublishedAt":"2026-11-04T01:41:48.311Z","commitSha":"88c79f6528a9347420e57ceb1fff9a60de651768","verifiedAt":"2026-11-04T01:30:30.000Z","lastCommitAt":"2026-11-10T06:12:20Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Teaches your AI assistant to do marketing work: landing pages, copy, SEO, ads, e-mails and analytics.","summary":"This is a pack of about 55 \"skills\": instruction files you add to an AI assistant such as Claude Code, Codex, Cursor or Windsurf. Once installed, asking \"optimise this landing page\" or \"write a welcome e-mail sequence\" makes the assistant follow a marketing playbook for conversion, copywriting, SEO, ads, pricing, churn and analytics. One skill holds the description of your product and the others read it first. The files are free and MIT-licensed; the assistant that runs them is what costs money.","easeRationale":"The easiest route is Claude on the web: download a `<name>-claude-web.zip` from the releases and upload it under Customize → Skills, one skill per file, with nothing installed. On Windows, macOS and Linux you first need an assistant such as Claude Code and then a terminal command — `npx skills add coreyhaines31/marketingskills`, or `/plugin install marketing-skills` inside Claude Code — which is why those are a C.","costNote":"The skills are free (MIT). What they need is an AI assistant you pay for, by plan or by API use, and the longer playbooks are long prompts, so each run consumes tokens. Some suggested tools and partner services charge separately.","notVerified":["Whether the advice produces results, which depends on your product and market","How carefully each of the 55-odd skills is reviewed before release","How many tokens, and therefore how much money, a long playbook consumes on each run","What an assistant actually does with the credentials from the tool-connection workflow","Whether the ZIP files really install in Claude on the web and in Perplexity, which the repository itself calls unverified for some clients"]},"zh-Hans":{"summaryShort":"教你的AI助手做营销工作：落地页、文案、SEO、广告、邮件和数据分析。","summary":"这是一套大约55个\"技能\"（skills）组成的工具包：是一些指令文件，你把它们加到AI助手里使用，比如Claude Code、Codex、Cursor或Windsurf。安装之后，只要说\"优化这个落地页\"或\"写一组欢迎邮件序列\"，助手就会按照营销手册来处理转化率、文案撰写、SEO、广告、定价、留存和数据分析等工作。其中一个技能保存你产品的描述信息，其他技能会先读取它。这些文件是免费的，采用MIT许可证；真正花钱的是运行它们的那个AI助手。","easeRationale":"最简单的方式是在网页版Claude中使用：从releases中下载一个`<name>-claude-web.zip`，然后在Customize → Skills里逐个上传每个技能文件，不需要安装任何东西。在Windows、macOS和Linux上，你需要先有一个像Claude Code这样的助手，然后再运行一条终端命令——`npx skills add coreyhaines31/marketingskills`，或者在Claude Code里执行`/plugin install marketing-skills`——这就是为什么这些平台的评级是C。","costNote":"这些技能文件本身是免费的（MIT许可证）。你需要的是一个你付费使用的AI助手，按套餐或按API用量计费，而且较长的手册本身就是很长的提示词，所以每次运行都会消耗token。一些推荐的工具和合作服务是单独收费的。","notVerified":["这些建议是否真的有效，这取决于你的产品和市场","这55个左右的技能在发布前到底经过了多严格的审查","一份长手册每次运行会消耗多少token，也就是花多少钱","助手拿到工具连接流程中获取的凭证后到底会做什么","这些ZIP文件是否真的能在网页版Claude和Perplexity中安装，仓库自己也说对某些客户端这一点尚未验证"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzMjE3MzE5NTE=","slug":"cptofevilminions__fleetdm-automation","name":"FleetDM-Automation","owner":"CptOfEvilMinions","fullName":"CptOfEvilMinions/FleetDM-Automation","type":"framework_infra","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布是在 2022-03-20（1690 天前）","source":"check","title":"发布版本已经过时"},{"code":"test_keys_and_certificate_are_in_the_repository","severity":"attention","evidence":"README 写道 \"This project has a pre-defined JWT key of `super_secret_key_here` which should ONLY BE used for testing\"，还有 \"This project contains with a self-signed OpenSSL ceretificate which should ONLY BE used for testing\"，compose 文件会直接从仓库里读取 conf/tls/fleet.crt 和 conf/tls/fleet.key。仓库顶层还提交了一个 .env 文件。照原样部署的话，登录密钥和私钥就等于公开了。","source":"ai","title":"仓库里放着测试用的密钥和证书"},{"code":"pinned_to_2022_versions_and_untouched_since","severity":"attention","evidence":"compose 文件把版本固定在 fleetdm/fleet:v4.11.0、mysql:8.0.27、redis:6.2.6-alpine 和 nginx:1.21.6-alpine，\"Versions supported\" 里列出的是 \"Fleet FleetDM v4.11.0+\"、\"Ansible v2.11+\" 和 \"Ubuntu server 20.04\"。最后一次提交是 2022-03-20，所以这里的一切都比它安装的软件落后了四年，包括其中的漏洞修复。","source":"ai","title":"固定在 2022 年的版本，之后再没更新"},{"code":"the_server_sees_everything_on_every_machine","severity":"info","evidence":"仓库里还有 deploy_osquery_linux.yml 和 deploy_osquery_windows.yml，用来在要监控的电脑上安装 osquery 代理。能进入 Fleet 控制台的人，就能向这些电脑询问已安装的软件、用户和正在运行的进程，所以这个控制台需要和任何管理工具一样小心对待。","source":"ai","title":"服务器能看到每台电脑上的一切"},{"code":"settings_look_like_a_home_lab","severity":"info","evidence":"Traefik 配置文件里的路由是 \"Host(`fleet.hackinglab.local`)\"，普通的 compose 文件把 8000:80 和 8443:443 对外发布，Fleet 容器的内存上限是 \"memory: 100M\"。这些都是作者自己实验室用的数值，不是为真正一批电脑设计的规模。","source":"ai","title":"配置看起来像是家庭实验室"},{"code":"instructions_live_on_a_blog","severity":"info","evidence":"README 写道 \"For a complete setup guide see my blog post here\"，指向一篇讲 FleetDM v3.6.0 的 2021 年文章——比这里固定使用的版本还旧。末尾的参考列表里还有好几个是空的（\"* []()\"），说明仓库里本身的文档内容很少。","source":"ai","title":"使用说明放在博客上"}],"platforms":["linux"],"tools":["docker","other"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system"],"alternativeTo":[{"slug":"kolide","name":"Kolide","confirmed":true}],"declaredAlternatives":[],"stars":60,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":60},{"date":"2026-11-06","stars":60},{"date":"2026-11-07","stars":60},{"date":"2026-11-08","stars":60},{"date":"2026-11-09","stars":60},{"date":"2026-11-10","stars":60}],"firstPublishedAt":"2026-11-04T22:32:54.875Z","commitSha":"0e4735dc84df7ab59d36fa283d613c1149f053fb","verifiedAt":"2026-11-04T22:22:44.514Z","lastCommitAt":"2022-03-20T17:43:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ready Docker and Ansible recipes that set up a Fleet server to keep an eye on the computers in a network.","summary":"This is not a program but a set of recipes written by one person: Docker Compose files and Ansible playbooks that install Fleet, the web console for osquery, together with MySQL, Redis and an NGINX front end. With it an administrator gets a server that asks every enrolled Windows or Linux machine questions about its software, settings and running processes, and there are extra playbooks to install the osquery agent on those machines. The README points to the author's blog post for the full guide. Built for Ubuntu Server 20.04.","easeRationale":"Administrators only, and Linux is the single target, so there is one grade: you install Docker (or Ansible 2.11 and an Ubuntu 20.04 server), generate your own TLS certificate with openssl, invent a JWT key and database passwords, edit hosts.ini and the group variables, then run docker-compose, a Docker Swarm stack or \"ansible-playbook -i hosts.ini deploy_fleetdm.yml\". There is no interface and no installer.","costNote":"These recipes are MIT licensed and free. Fleet, the software they install, is open source but its maker also sells a paid tier, so the newer paid features are not part of what you get here.","notVerified":["Whether any of it still works with a current Fleet release, a current MySQL or a newer Ubuntu","Whether the Ansible roles leave a working server, since there is nothing here to test them against","What the committed .env and config files contain besides the test keys named in the README","How much memory and disk a real fleet of machines needs, which the lab limits do not answer","Whether the osquery agent playbooks still match today's osquery packages"]},"zh-Hans":{"summaryShort":"现成的 Docker 和 Ansible 配方，用来搭建 Fleet 服务器，监控网络中的电脑。","summary":"这不是一个程序，而是一个人写的一套配方：Docker Compose 文件和 Ansible playbook，用来安装 Fleet（osquery 的网页控制台），还有 MySQL、Redis 和一个 NGINX 前端。装好之后，管理员就能得到一台服务器，它会向网络里每台已注册的 Windows 或 Linux 电脑询问软件、设置和正在运行的进程，另外还有额外的 playbook 用来在这些电脑上安装 osquery 代理。README 把完整教程的链接指向了作者的博客文章。是为 Ubuntu Server 20.04 做的。","easeRationale":"只给管理员用，而且目标系统只有 Linux，所以只有一个难度级别：你需要安装 Docker（或者 Ansible 2.11 和一台 Ubuntu 20.04 服务器），用 openssl 自己生成 TLS 证书，自己编一个 JWT 密钥和数据库密码，编辑 hosts.ini 和分组变量，然后运行 docker-compose、一个 Docker Swarm 堆栈，或者执行 \"ansible-playbook -i hosts.ini deploy_fleetdm.yml\"。没有界面，也没有安装程序。","costNote":"这些配方采用 MIT 许可证，免费使用。它们安装的 Fleet 本身是开源的，但它的开发方也卖付费版本，所以更新的付费功能不包含在这里。","notVerified":["这些东西能不能在现在的 Fleet 版本、现在的 MySQL 或更新的 Ubuntu 上正常工作","Ansible 角色是否真能搭出一台能用的服务器，因为这里没有任何东西可以用来测试它们","仓库里已经提交的 .env 和配置文件里，除了 README 提到的测试密钥之外还有什么内容","真正的一批电脑需要多少内存和磁盘空间，实验室里的限制数值回答不了这个问题","osquery 代理的 playbook 是否还跟今天的 osquery 软件包匹配"]}}},{"id":"R_kgDOVMLR1A","slug":"cranebeaker__coop-anything","name":"Co-op Anything","owner":"cranebeaker","fullName":"cranebeaker/coop-anything","type":"dev_tool","status":"alert","grade":"E","easeByPlatform":{"windows":"E","linux":"E"},"risk":"high","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 16 天前","source":"check","title":"所有者账号很新"},{"code":"reverse_engineering_of_games_you_bought","severity":"attention","evidence":"README，What you can co-op：\"Full reverse engineering of state, saves and levels; the hardest path\"，所列工具为 \"REA, GhidraMCP, ida-pro-mcp, Frida, Cheat Engine\" 和 \"DLL hooks\"。游戏的许可协议通常禁止拆解游戏，在一些国家法律也有限制，即使这款游戏是你自己的。","source":"ai","title":"对你买的游戏做逆向工程"},{"code":"uses_valves_test_app_for_networking","severity":"attention","evidence":"README，FAQ：\"The Steam relay uses Valve's free Spacewar test app, so nobody buys anything else.\" Spacewar 是 Valve 内部的测试条目，并不是给别人的游戏使用的应用。","source":"ai","title":"用 Valve 的测试应用来联网"},{"code":"agent_installs_and_runs_many_third_party_tools","severity":"attention","evidence":"该指南让智能体在你的电脑上获取并运行 BepInEx、MelonLoader、UE4SS、Dumper-7、dnSpyEx、Il2CppDumper、Ghidra、x64dbg、MinHook 等；FAQ 只说了 \"The agent asks before installing anything or using your accounts.\"","source":"ai","title":"智能体会安装并运行许多第三方工具"},{"code":"readme_promises_it_works_for_any_game","severity":"info","evidence":"README，FAQ \"Will it work for my game?\"：\"Yes. Some games need more agent hours than others, but it's definitely possible.\" 列出了五款已完成的游戏，而项目只有 16 天，有 169 个未解决的 issue，也没有发布版本。","source":"ai","title":"README 承诺对任何游戏都有效"},{"code":"console_games_through_emulators_and_decomps","severity":"info","evidence":"README：对于 \"Console only (PS1, PS2, N64, Dreamcast, GameCube, Wii, GBA, ...)\"，路径是 \"Decomp or recompiled PC port first; else emulator netplay, or a memory-sync mod between two emulators\" —— 这需要你自己提供游戏文件。","source":"ai","title":"主机游戏要通过模拟器和反编译项目"},{"code":"takes_apart_games_you_bought","severity":"high","evidence":"README，What you can co-op：\"Full reverse engineering of state, saves and levels; the hardest path\"，所用工具为 \"REA, GhidraMCP, ida-pro-mcp, Frida, Cheat Engine\"；README，FAQ：\"The Steam relay uses Valve's free Spacewar test app, so nobody buys anything else.\" 游戏的许可协议通常禁止拆解或修改游戏，一些国家有法律限制绕过技术保护措施，Steam 的条款也规定了一个产品可以使用哪些应用 ID。对于 \"Is this legal?\"，README 的回答是 \"Modding games for co-op with friends is what the modding community has always done. This repo doesn't host game files or assets... and it doesn't touch anti-cheat or online services.\"","source":"ai","title":"拆解你买的游戏"}],"platforms":["windows","linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games","ai_skills","ai_agents"],"alternativeTo":[],"declaredAlternatives":[],"stars":2040,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":182,"growthDays":7,"history":[{"date":"2026-10-27","stars":1491},{"date":"2026-10-28","stars":1546},{"date":"2026-10-29","stars":1587},{"date":"2026-10-30","stars":1652},{"date":"2026-10-31","stars":1720},{"date":"2026-11-01","stars":1787},{"date":"2026-11-02","stars":1823},{"date":"2026-11-03","stars":1858},{"date":"2026-11-04","stars":1883},{"date":"2026-11-05","stars":1911},{"date":"2026-11-06","stars":1927},{"date":"2026-11-07","stars":1941},{"date":"2026-11-08","stars":1973},{"date":"2026-11-09","stars":2012},{"date":"2026-11-10","stars":2040}],"firstPublishedAt":"2026-10-27T18:40:27.842Z","commitSha":"f55babd6b2280ba788f5fd5ddd8c0559f09cbd43","verifiedAt":"2026-11-02T02:56:04.926Z","lastCommitAt":"2026-11-02T10:16:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Playbook, skills and MCP server that let an AI coding agent add co-op multiplayer to single-player PC games.","summary":"Co-op Anything is a pack of instructions, skills and small Python tools for AI coding agents such as Claude Code, Codex or Cursor. You point the agent at the repository and name a game you own; it looks for an existing co-op mod, or takes the game apart to add one, tests it with bots driving real copies of the game, and packages it so friends can join over Steam. There is also a directory of 230+ co-op mods. You need the game, an AI agent that you pay for, Python for the MCP server, and a lot of patience: the work is reverse engineering.","easeRationale":"On Windows, where most of the work happens, you set up an AI coding agent, clone the repository, register the Python MCP server by hand (\"claude mcp add coop-anything -- python <clone>/tools/mcp/coop_mcp.py\"), and the agent then installs reverse-engineering tools, compilers and mod loaders and runs several copies of the game. Linux is the same, with PartyDeck instead of Nucleus Co-op for split screen. There is no finished program to install here.","costNote":"The repository is free (MIT), but the work is done by your own AI coding agent, which normally costs money, and the README says big games take many agent hours.","notVerified":["Whether an agent can really make your particular game co-op, and how many hours and how much money it would take.","Whether the finished mods work and stay stable in a real session with friends.","How accurate the 230+ entries in the directory are, since the project itself grades them from \"untested\" upwards.","What the agent will actually download and run on your computer during the work.","Whether using Valve's Spacewar app as a relay is allowed by Steam's terms."]},"zh-Hans":{"summaryShort":"一套指南、技能和 MCP 服务器，让 AI 编程智能体给单人 PC 游戏加上合作联机模式。","summary":"Co-op Anything 是一套给 AI 编程智能体（如 Claude Code、Codex 或 Cursor）使用的说明、技能和 Python 小工具。你让智能体指向这个仓库，并说出一款你拥有的游戏；它会寻找现成的合作模组，或者拆解游戏来添加一个，用操控游戏真实副本的机器人来测试，再打包好，让朋友通过 Steam 加入。另有一个收录 230+ 个合作模组的目录。你需要拥有这款游戏、一个你自己付费的 AI 智能体、用于 MCP 服务器的 Python，以及大量耐心：这项工作就是逆向工程。","easeRationale":"在 Windows 上（大部分工作都在这里），你先设置好 AI 编程智能体，克隆仓库，手动注册 Python 的 MCP 服务器 (\"claude mcp add coop-anything -- python <clone>/tools/mcp/coop_mcp.py\")，然后智能体会安装逆向工程工具、编译器和模组加载器，并运行多个游戏副本。Linux 上是一样的，分屏时用 PartyDeck 代替 Nucleus Co-op。这里没有现成的程序可以安装。","costNote":"仓库本身免费 (MIT)，但工作由你自己的 AI 编程智能体完成，通常要花钱，README 还说大型游戏需要智能体运行很多个小时。","notVerified":["智能体是否真的能让你的这款游戏支持合作联机，以及需要多少小时、多少钱。","做好的模组在与朋友的真实游戏中是否能正常运行并保持稳定。","目录中 230+ 个条目有多准确，因为项目自己把它们从 \"untested\" 起逐级评分。","智能体在工作期间实际会在你的电脑上下载并运行什么。","把 Valve 的 Spacewar 应用用作中继是否符合 Steam 的条款。"]}}},{"id":"R_kgDOOpAIvg","slug":"cursortouch__windows-mcp","name":"Windows-MCP","owner":"CursorTouch","fullName":"CursorTouch/Windows-MCP","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接传给 shell 执行:curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"curl 管道直接执行"},{"code":"full_control_of_your_pc","severity":"attention","evidence":"工具列表给 AI 提供了\"PowerShell: To execute PowerShell commands\"\"FileSystem: Read, write, copy, move, delete, list, search\"\"Process: List running processes or terminate them by PID or name\"\"Clipboard: Read or set Windows clipboard content\"以及注册表访问权限,README写道\"All tools are enabled by default\"(需要用 --exclude-tools 才能屏蔽某个工具)。","source":"ai","title":"完全控制你的电脑"},{"code":"can_be_exposed_over_the_network","severity":"attention","evidence":"文档展示了如何将其作为网络服务运行,例如\"windows-mcp serve --transport sse --host 0.0.0.0\";令牌、IP白名单和TLS证书(--auth-key、--ip-allowlist、--ssl-certfile)都是需要你自己添加的额外参数,因此一条不留意的命令就会让桌面控制权暴露在网络上。","source":"ai","title":"可以暴露在网络上"},{"code":"user_takeover_is_best_effort","severity":"info","evidence":"README 在谈到控制拦截时写道:\"User takeover is therefore best-effort, not a security boundary or a guarantee of exclusive control\",还说\"an unanswered countdown starts control, so unattended runs keep working\",并且当 consent_countdown_seconds = 0 时AI会\"start control without asking\"。","source":"ai","title":"用户接管只是尽力而为"},{"code":"usage_telemetry_enabled_by_default","severity":"info","evidence":"pyproject.toml 依赖\"posthog>=7.4.0\",环境变量表中 ANONYMIZED_TELEMETRY 的默认值是\"true\",除非你将其设为 false,否则会把匿名使用数据发送到 https://us.i.posthog.com。","source":"ai","title":"默认启用使用情况遥测"}],"platforms":["windows"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":0,"maxMB":1,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","ai_agents","automation"],"alternativeTo":[{"slug":"uipath","name":"UiPath","confirmed":true},{"slug":"manus","name":"Manus","confirmed":true}],"declaredAlternatives":[],"stars":11480,"downloads":5153,"downloadsWeek":228,"downloadsDays":3,"growth":415,"growthDays":6,"history":[{"date":"2026-11-07","stars":11258},{"date":"2026-11-08","stars":11342},{"date":"2026-11-09","stars":11415},{"date":"2026-11-10","stars":11480}],"firstPublishedAt":"2026-11-06T22:59:23.762Z","commitSha":"b69ec872623b403f647fb3777685d4cf2678d3a2","verifiedAt":"2026-11-06T22:38:16.526Z","lastCommitAt":"2026-11-10T06:53:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant such as Claude use your Windows PC: click, type, open apps, read files and run PowerShell.","summary":"Windows-MCP is a small server that connects an AI assistant to Windows 10 or 11. Once it is registered in Claude Desktop, Claude Code, Perplexity Desktop, Gemini CLI or Codex, the assistant can take screenshots, read the screen, click, type, use keyboard shortcuts, open and resize apps, handle files, read the clipboard, end processes and run PowerShell commands. It works with any model, with no special vision model needed. You need Windows, Python 3.14 or newer and the uv package manager, plus a few commands or a config file edit in your assistant.","easeRationale":"Windows only, and it is a C: you install Python 3.14 or newer and the uv package manager, then run a command such as \"claude mcp add --transport stdio windows-mcp -- uvx windows-mcp serve\" or paste a JSON block into your assistant's config file. There is no installer: the release files are Python packages (.whl and .tar.gz), and the README warns the first start can take a minute or two and may time out once.","costNote":"MIT licensed and free. The AI assistant that drives it is separate and may cost money (a paid plan or API credit), depending on which one you use.","notVerified":["Whether it installs cleanly on your machine, since it requires Python 3.14 or newer, a very recent release","Whether the desktop-control gate and the takeover hotkey really stop the AI mid-action, which the README itself calls best-effort","Whether the anonymous telemetry stops when ANONYMIZED_TELEMETRY is set to false","What an AI will actually do with PowerShell, file deletion and process killing on your computer, including when a web page it reads contains instructions","The stated 0.2 to 0.5 second latency between actions"]},"zh-Hans":{"summaryShort":"让像 Claude 这样的 AI 助手使用你的 Windows 电脑:点击、输入、打开应用、读取文件和运行 PowerShell。","summary":"Windows-MCP 是一个小型服务器,用于把 AI 助手连接到 Windows 10 或 11。在 Claude Desktop、Claude Code、Perplexity Desktop、Gemini CLI 或 Codex 中注册后,助手就能截取屏幕、读取屏幕内容、点击、输入、使用键盘快捷键、打开和调整应用窗口大小、处理文件、读取剪贴板、结束进程并运行 PowerShell 命令。它可以配合任何模型使用,不需要特殊的视觉模型。你需要 Windows、Python 3.14 或更新版本、uv 包管理器,以及在你的助手里执行几条命令或编辑一个配置文件。","easeRationale":"仅支持 Windows,它在这里是一个 C:你需要安装 Python 3.14 或更新版本和 uv 包管理器,然后运行类似 \"claude mcp add --transport stdio windows-mcp -- uvx windows-mcp serve\" 的命令,或者把一段 JSON 粘贴到你的助手的配置文件里。没有安装程序:发布文件是 Python 包(.whl 和 .tar.gz),README 提示首次启动可能需要一两分钟,偶尔还会超时一次。","costNote":"采用 MIT 许可,免费使用。驱动它的 AI 助手是独立的,可能需要付费(付费套餐或 API 额度),具体取决于你使用哪一个。","notVerified":["能否在你的机器上顺利安装,因为它要求 Python 3.14 或更新版本,是非常新的版本","桌面控制的拦截和接管快捷键是否真的能在AI操作过程中将其中止,README 自己也称这只是尽力而为,并无保证","当 ANONYMIZED_TELEMETRY 设为 false 时,匿名遥测是否真的会停止","AI 在你的电脑上实际会用 PowerShell、删除文件和终止进程做些什么,包括当它读取的网页中包含指令时","官方给出的0.2到0.5秒的操作延迟"]}}},{"id":"R_kgDOO7R-vw","slug":"cybertimon__rapidraw","name":"RapidRAW","owner":"CyberTimon","fullName":"CyberTimon/RapidRAW","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"sign_in_service_and_cloud_features_in_the_app","severity":"attention","evidence":"package.json 包含 \"@clerk/react\" 和 \"tauri-plugin-clerk\"（一个账号登录服务），更新日志中有 \"integrated cloud API\" (2026-10-05)、\"Initial backend implementation of the cloud service functionality\" (2026-05-16) 和 \"backend preparation for cloud inpainting\" —— 所以这款本来主要在本地运行的照片编辑器，也有账号和服务器端。","source":"ai","title":"应用内含登录服务和云功能"},{"code":"ai_tools_download_models_at_run_time","severity":"info","evidence":"更新日志提到 \"New onnxruntime download strategy\"、\"fixed AI model and LUT download issues on Android\" 和 \"new sky and foreground models\"，还有 2026-10-06 加入的 \"AI-free mode\" —— AI 蒙版、降噪和深度工具会在安装之后再获取模型，所以实际磁盘占用高于 41-123 MB 的安装程序。","source":"ai","title":"AI 工具在运行时下载模型"},{"code":"changes_very_fast_with_many_open_issues","severity":"info","evidence":"69 个发布版本，检查当天仍有提交，更新日志几乎每天都有条目（包括 \"fixed AI hangs on macOS\"），还有 483 个未解决的 issue。","source":"ai","title":"变化很快，未解决的 issue 很多"},{"code":"generative_ai_needs_a_separate_heavy_install","severity":"info","evidence":"更新日志描述了 \"Initial generative AI integration with ComfyUI\"、一个 \"python middleware client for external generative AI integration (ComfyUI)\" 和 \"User customisable ComfyUI workflow selection\" —— 这些功能需要你在自己的机器上配置好 ComfyUI。","source":"ai","title":"生成式 AI 需要另外安装体量很大的组件"},{"code":"one_person_project","severity":"info","evidence":"README，The Idea：\"I started developing this project as a personal challenge when I was 18\" 以及 \"I often found existing software to be sluggish... so I decided to build my own\"，其他贡献者只因个别功能而被署名。","source":"ai","title":"一个人的项目"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":41,"maxMB":123,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"lightroom","name":"Lightroom","confirmed":true},{"slug":"adobe-camera-raw","name":"Camera Raw","confirmed":true},{"slug":"capture-one","name":"Capture One","confirmed":true}],"declaredAlternatives":[{"target":"lightroom","stance":"claims","evidence":"RapidRAW is a modern, high-performance alternative to Adobe Lightroom"}],"stars":11050,"downloads":799645,"downloadsWeek":22226,"downloadsDays":7,"growth":108,"growthDays":7,"history":[{"date":"2026-10-25","stars":10694},{"date":"2026-10-26","stars":10730},{"date":"2026-10-27","stars":10746},{"date":"2026-10-28","stars":10761},{"date":"2026-10-29","stars":10769},{"date":"2026-10-30","stars":10834},{"date":"2026-10-31","stars":10874},{"date":"2026-11-01","stars":10910},{"date":"2026-11-02","stars":10930},{"date":"2026-11-03","stars":10942},{"date":"2026-11-04","stars":10950},{"date":"2026-11-05","stars":10965},{"date":"2026-11-06","stars":10978},{"date":"2026-11-07","stars":10995},{"date":"2026-11-08","stars":11014},{"date":"2026-11-09","stars":11035},{"date":"2026-11-10","stars":11050}],"firstPublishedAt":"2026-10-25T23:18:58.364Z","commitSha":"d1a9e2a5cc7dc53ab5579f244377671534100032","verifiedAt":"2026-11-02T02:56:53.724Z","lastCommitAt":"2026-11-10T21:38:51Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"01_RapidRAW_v1.6.9_windows_x64.exe","sizeBytes":33919943,"url":"https://github.com/CyberTimon/RapidRAW/releases/download/v1.6.9/01_RapidRAW_v1.6.9_windows_x64.exe","releaseUrl":"https://github.com/CyberTimon/RapidRAW/releases/tag/v1.6.9"},{"os":"mac","file":"02_RapidRAW_v1.6.9_macos-15_aarch64.dmg","sizeBytes":34493666,"url":"https://github.com/CyberTimon/RapidRAW/releases/download/v1.6.9/02_RapidRAW_v1.6.9_macos-15_aarch64.dmg","releaseUrl":"https://github.com/CyberTimon/RapidRAW/releases/tag/v1.6.9"},{"os":"linux","file":"03_RapidRAW_v1.6.9_ubuntu-24.04_amd64.AppImage","sizeBytes":111839736,"url":"https://github.com/CyberTimon/RapidRAW/releases/download/v1.6.9/03_RapidRAW_v1.6.9_ubuntu-24.04_amd64.AppImage","releaseUrl":"https://github.com/CyberTimon/RapidRAW/releases/tag/v1.6.9"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Develops RAW photos from your camera: exposure, colour, masks, retouching and export, from an installer.","summary":"RapidRAW is a photo editor for the RAW files a camera produces. You browse your folders, adjust exposure, colour, curves and lens corrections, paint or let the AI make masks, clean up spots, merge HDR or panoramas, and export. Your original file is never changed: the edits are kept in a sidecar file next to it. The author, a Swiss photographer, describes it as an alternative to Adobe Lightroom. Ready installers exist for Windows, macOS and Linux. The editing runs on your graphics card, and some AI tools download their own models the first time you use them.","easeRationale":"Every release ships ready files for all three systems: .exe and .msi for Windows (x64 and ARM), .dmg for Intel and Apple Silicon Macs, and .deb, .rpm and AppImage for Linux. You download one and open it. Camera tethering is a separate download and only for macOS and Linux.","costNote":"Free and open source under the AGPL-3.0. The code includes a sign-in service and a cloud connection for some AI features.","notVerified":["Whether your camera's RAW files are decoded correctly, since support comes from the rawler and Lensfun databases.","How fast it really is compared with Lightroom on your computer and graphics card.","What the cloud features and the sign-in service send, and whether any of them cost money.","Real disk use once thumbnails, caches and the AI models are on disk.","Whether the Android builds mentioned in the README work as well as the desktop ones."]},"zh-Hans":{"summaryShort":"冲洗相机的 RAW 照片：曝光、色彩、蒙版、修饰和导出，通过安装程序即可使用。","summary":"RapidRAW 是处理相机所生成 RAW 文件的照片编辑器。你浏览文件夹，调整曝光、色彩、曲线和镜头校正，手动绘制或让 AI 生成蒙版，清除污点，合并 HDR 或全景图，然后导出。你的原始文件从不被改动：编辑内容保存在它旁边的附属文件中。作者是一位瑞士摄影师，他把它描述为 Adobe Lightroom 的替代品。提供适用于 Windows、macOS 和 Linux 的现成安装程序。编辑运算在你的显卡上进行，部分 AI 工具会在你首次使用时下载各自的模型。","easeRationale":"每个版本都为三种系统提供现成文件：Windows 的 .exe 和 .msi（x64 和 ARM），Intel 和 Apple Silicon Mac 的 .dmg，以及 Linux 的 .deb、.rpm 和 AppImage。下载一个然后打开即可。相机联机拍摄是单独的下载，且仅限 macOS 和 Linux。","costNote":"免费，开源，采用 AGPL-3.0 许可。代码中包含一个登录服务，以及供部分 AI 功能使用的云连接。","notVerified":["你的相机的 RAW 文件能否被正确解码，因为支持来自 rawler 和 Lensfun 数据库。","在你的电脑和显卡上，它实际上比 Lightroom 快多少。","云功能和登录服务会发送什么内容，以及其中是否有收费的。","缩略图、缓存和 AI 模型都存到磁盘上之后的实际磁盘占用。","README 提到的 Android 版本是否和桌面版一样好用。"]}}},{"id":"R_kgDOO3qbaQ","slug":"czlonkowski__n8n-mcp","name":"n8n-MCP","owner":"czlonkowski","fullName":"czlonkowski/n8n-mcp","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖中存在已知漏洞（OSV）：@modelcontextprotocol/sdk: GHSA-22jm-h49p-29qw, GHSA-46jr-4wx8-v453, GHSA-6prh-2h8m-c8cw；zod: GHSA-mrhx-c7xc-m3w5","source":"check","title":"依赖项中存在已知漏洞"},{"code":"telemetry_on_by_default","severity":"attention","evidence":".env.example中写道：\"Telemetry # Anonymous usage statistics are enabled by default. Set to true to turn them off. See PRIVACY.md for what is collected. # N8N_MCP_TELEMETRY_DISABLED=true\"。项目里还有一个PRIVACY.md文件。","source":"ai","title":"遥测功能默认开启"},{"code":"ai_gets_write_access_to_your_automations","severity":"attention","evidence":"README.md列出了在你添加n8n API密钥之后AI可以调用的工具，其中包括\"n8n_delete_workflow - Delete workflows permanently\"和\"n8n_manage_credentials - Manage n8n credentials (list, get, create, update, delete, getSchema)\"。README自己也警告说：\"NEVER edit your production workflows directly with AI!\"","source":"ai","title":"AI获得了修改你自动化流程的写入权限"},{"code":"author_warns_about_unpredictable_results","severity":"info","evidence":"README.md中的\"Important Safety Warning\"部分写道：\"Make a copy of your workflow before using AI tools\"、\"Test in development environment first\"、\"AI results can be unpredictable.\"","source":"ai","title":"作者提醒结果可能难以预测"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["n8n instance API key (N8N_API_KEY), only to let the AI manage your workflows","Other (AUTH_TOKEN), a password you invent yourself when you run it in HTTP mode"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","automation","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":23239,"downloads":839,"downloadsWeek":214,"downloadsDays":6,"growth":31,"growthDays":6,"history":[{"date":"2026-11-04","stars":23208},{"date":"2026-11-05","stars":23216},{"date":"2026-11-06","stars":23223},{"date":"2026-11-07","stars":23228},{"date":"2026-11-08","stars":23228},{"date":"2026-11-09","stars":23232},{"date":"2026-11-10","stars":23239}],"firstPublishedAt":"2026-11-04T01:36:50.312Z","commitSha":"1c755c5c8d99eb71faade17753a7f24138f4212c","verifiedAt":"2026-11-04T01:20:20.956Z","lastCommitAt":"2026-11-09T12:08:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Teaches AI assistants like Claude how n8n works, so they can build and fix your automation workflows for you.","summary":"n8n-MCP is an MCP server: a small program you plug into an AI assistant (Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Codex). Once connected, the assistant can look up n8n's 3,400+ workflow building blocks and ready-made templates, and, if you give it your n8n address and API key, create, change, test and delete workflows in your own n8n. It is for people who already use n8n. You need Node.js (or Docker) and an AI assistant that supports MCP. The author also runs a hosted version with a free daily limit.","easeRationale":"On Windows, Mac and Linux alike it is the same work: install Node.js, then paste a small block of settings into your AI assistant's configuration file and restart it. There is no installer and no window to click; everything happens in a text file, and the optional Docker route needs a terminal.","costNote":"The software is free (MIT licence). The author's hosted service at dashboard.n8n-mcp.com has a free tier of 100 tool calls a day, and the author asks for sponsorship.","notVerified":["Whether the AI assistant actually produces working n8n workflows, and how often it gets node settings wrong.","What the anonymous usage statistics really send, since that can only be seen while it runs.","How much the hosted service costs above the free 100 tool calls a day.","Real disk use: the bundled node database and the 2,352 templates grow the install beyond the plain program.","Whether the known vulnerabilities found in its pinned dependencies can be reached in normal use."]},"zh-Hans":{"summaryShort":"教Claude这样的AI助手了解n8n的工作原理，让它们帮你搭建和修复自动化工作流。","summary":"n8n-MCP是一个MCP服务器：一个你接入AI助手（Claude Code、Claude Desktop、Cursor、VS Code、Windsurf、Codex）的小程序。连接之后，助手就能查询n8n超过3,400个工作流构建模块和现成的模板；如果你提供自己n8n的地址和API密钥，它还能在你自己的n8n里创建、修改、测试和删除工作流。这是给已经在用n8n的人准备的。你需要Node.js（或Docker）以及一个支持MCP的AI助手。作者还提供一个托管版本，每天有免费额度。","easeRationale":"无论是Windows、Mac还是Linux，操作都一样：安装Node.js，然后把一小段配置粘贴到你的AI助手的配置文件里并重启它。没有安装程序，也没有可以点击的窗口；一切都在一个文本文件里完成，可选的Docker方式则需要用到终端。","costNote":"这个软件是免费的（MIT许可证）。作者在dashboard.n8n-mcp.com提供的托管服务有一个免费额度，每天100次工具调用，作者也在寻求赞助支持。","notVerified":["AI助手是否真的能生成可用的n8n工作流，以及它配置节点出错的频率有多高","匿名使用统计数据到底发送了什么内容，这只能在程序运行时才能看到","超出每天100次免费工具调用之后，托管服务的费用是多少","实际占用的磁盘空间：内置的节点数据库和2,352个模板会让安装占用的空间超过程序本身","固定版本依赖中已知的漏洞在正常使用中是否真的可被利用"]}}},{"id":"R_kgDOQFmuYA","slug":"czlonkowski__n8n-skills","name":"n8n-skills","owner":"czlonkowski","fullName":"czlonkowski/n8n-skills","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"C"},"risk":"attention","signals":[{"code":"the_assistant_can_change_your_live_automations","severity":"attention","evidence":"这些技能文档都是围绕能创建和更新工作流的 n8n-mcp 工具搭建的，涉及 \"Calls to n8n_instances and n8n_manage_credentials\" 和 \"Naming the instance on credential writes\"。多实例那份技能文档警告说 \"`switch` as the shared default for unnamed calls, and why it moves every chat on a hosted connection\"，还提到可能出现指向错误，也就是说助手可能会操作错实例——本该是预发布环境，结果动了生产环境。","source":"ai","title":"助手可以改动你正在用的自动化流程"},{"code":"one_skill_sets_up_a_server_for_you","severity":"attention","evidence":"第 14 份技能 n8n Self-Hosting 承诺要 \"Deploy a production self-hosted n8n end-to-end to a fresh Linux VM\"，用的是 \"Docker Compose behind Caddy\"，并且 \"Generates fresh secrets on the box\"。照着做，就意味着让助手在你自己的服务器上执行管理员级别的命令，并且相信它给的加密密钥、防火墙和备份的默认设置。","source":"ai","title":"有一份技能会帮你搭服务器"},{"code":"hooks_run_in_every_session","severity":"info","evidence":"这个插件带有一个 \"always-on enforcement layer\"（常驻检查层）：一份路由技能文档会 \"loaded into every session by a SessionStart hook\"，并且 \"re-fires on resume/clear/compact\"，另外还有 PreToolUse hook 会检查创建和更新的调用，以及一个在 validate_workflow 之后触发的 PostToolUse hook。README 写道 \"Every hook fails open and never blocks a tool call\"，而且这些 hook 只在以插件方式安装时运行，但它们确实会往每次会话里加文字。","source":"ai","title":"每次会话都会跑 hook"},{"code":"tied_to_the_authors_other_project_and_services","severity":"info","evidence":"这份指南假定你用的是 \"Version 2.97.0+\" 的 n8n-mcp，还直接写出了它具体的工具名称，所以一旦那个服务器或者 n8n 本身发生变化，内容就会跟不上。README 里还两次打广告，推广作者的付费服务：\"Need it built for you? Work with AiAdvisors — n8n automation audits, builds, and operations, run by the author of n8n-mcp and n8n-skills\"。","source":"ai","title":"依赖作者的另一个项目和服务"},{"code":"the_numbers_are_the_authors_own","severity":"info","evidence":"像 \"525+ n8n nodes supported\"、\"2,653+ workflow templates\"、\"Top 5 error patterns with solutions (covering 62%+ of failures)\" 和 \"Iterative: Test against evaluations, iterate until 100% pass\" 这些说法都来自 README。仓库里确实有 evaluations 文件夹，但没有公开具体的结果。","source":"ai","title":"数据都是作者自己给出的"}],"platforms":["windows","mac","linux","web"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["An API key for your own n8n instance, used by the n8n-mcp server when the assistant creates or changes workflows"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":6624,"downloads":9088,"downloadsWeek":87,"downloadsDays":5,"growth":36,"growthDays":6,"history":[{"date":"2026-11-05","stars":6593},{"date":"2026-11-06","stars":6598},{"date":"2026-11-07","stars":6602},{"date":"2026-11-08","stars":6606},{"date":"2026-11-09","stars":6613},{"date":"2026-11-10","stars":6624}],"firstPublishedAt":"2026-11-04T22:38:15.180Z","commitSha":"e037563e101749e630e6688d121d04529fe6ee7b","verifiedAt":"2026-11-04T22:23:58.973Z","lastCommitAt":"2026-10-24T11:43:44Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A pack of written guides that teach Claude Code how to build, fix and host n8n automations.","summary":"This is a set of 14 written skills for AI assistants, not a program you open. Installed into Claude Code (or uploaded to Claude.ai or another assistant that reads skill files), they tell the assistant how n8n automations are put together: expression syntax, node settings, error handling, files, sub-workflows, AI agents, and one skill for setting up your own n8n server with Docker and HTTPS. They are written for the author's n8n-mcp server, which the assistant uses to search nodes and write workflows, so you need that connected first.","easeRationale":"The same grade everywhere, because the work is the same: you must already have an AI assistant and, as the Prerequisites say, the \"n8n-mcp MCP server installed and configured\" with a .mcp.json entry. After that it is one command in Claude Code (\"/plugin install czlonkowski/n8n-skills\") on Windows, Mac and Linux, or on the web you download the zip from the releases page and upload the skill files in Claude.ai or ChatGPT settings. The web route is a little less work but still needs the same MCP server behind it.","costNote":"The skills are free (MIT). They only do something inside an AI assistant, which normally needs a paid plan or paid API use, and they are aimed at n8n, which is free when you host it yourself and paid on n8n Cloud.","notVerified":["Whether the written guidance still matches the current n8n and n8n-mcp versions","Whether the workflows an assistant builds with it actually run on your instance","How much the assistant plan or API use costs for a session of workflow building","What the hooks add to each session and how much context they take","Whether the self-hosting skill leaves a server set up safely, which can only be judged by running it"]},"zh-Hans":{"summaryShort":"一套写好的指南，教 Claude Code 怎么搭建、修复和托管 n8n 自动化流程。","summary":"这是为 AI 助手写的 14 份技能文档（skill），不是一个要打开运行的程序。把它们装进 Claude Code（或者上传到 Claude.ai 或其他能读取 skill 文件的助手），就能告诉助手 n8n 自动化流程是怎么搭起来的：表达式语法、节点设置、错误处理、文件、子工作流、AI 代理，还有一份教你用 Docker 和 HTTPS 搭建自己的 n8n 服务器的技能。这些文档是为作者自己做的 n8n-mcp 服务器写的，助手要靠它来搜索节点和编写工作流，所以你得先把它接好。","easeRationale":"难度到处都一样，因为要做的事都一样：你得先有一个 AI 助手，而且像前提条件里写的那样，要先装好并配置 \"n8n-mcp MCP server installed and configured\"，在 .mcp.json 里加上一条记录。之后在 Windows、Mac 和 Linux 上，在 Claude Code 里只需要一条命令（\"/plugin install czlonkowski/n8n-skills\"）；在网页端则是从 releases 页面下载 zip 包，再到 Claude.ai 或 ChatGPT 的设置里上传这些 skill 文件。网页这条路稍微省点事，但背后同样需要那个 MCP 服务器。","costNote":"这些技能文档是免费的（MIT 许可证）。它们只在 AI 助手里起作用，而用助手通常需要付费套餐或付费的 API 调用；它们是为 n8n 服务的，自己搭建 n8n 是免费的，用 n8n Cloud 则要付费。","notVerified":["这些文字指南是否还跟现在的 n8n 和 n8n-mcp 版本相符","助手用它搭出来的工作流是否真能在你自己的实例上跑起来","一次搭建工作流的会话，助手套餐或 API 用量到底要花多少钱","这些 hook 给每次会话加了什么内容，占用了多少上下文","自助搭建服务器的那份技能是否能把服务器配置得妥当，这一点只能靠实际运行来判断"]}}},{"id":"MDEwOlJlcG9zaXRvcnk5NDMxNDk=","slug":"d3__d3","name":"D3.js","owner":"d3","fullName":"d3/d3","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"web":"E"},"risk":"low","signals":[{"code":"no_charts_included","severity":"info","evidence":"README.md将其描述为一种\"low-level approach built on web standards\"，是\"a foundational building block of higher-level chart libraries\"：你需要自己用零件搭出每一个图表，这比直接用现成的图表库要花费更多功夫。","source":"ai","title":"不包含现成的图表"},{"code":"many_sub_packages","severity":"info","evidence":"package.json中列出了30个独立的d3-*依赖项（d3-array、d3-scale、d3-geo等），安装时每一个都会被下载。","source":"ai","title":"依赖许多子包"}],"platforms":["web"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"ISC","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"highcharts","name":"Highcharts","confirmed":true},{"slug":"amcharts","name":"amCharts","confirmed":true}],"declaredAlternatives":[],"stars":113900,"downloads":2206747,"downloadsWeek":0,"downloadsDays":7,"growth":35,"growthDays":7,"history":[{"date":"2026-10-26","stars":113803},{"date":"2026-10-27","stars":113811},{"date":"2026-10-28","stars":113817},{"date":"2026-10-29","stars":113819},{"date":"2026-10-30","stars":113829},{"date":"2026-10-31","stars":113832},{"date":"2026-11-01","stars":113847},{"date":"2026-11-02","stars":113858},{"date":"2026-11-03","stars":113865},{"date":"2026-11-04","stars":113867},{"date":"2026-11-05","stars":113875},{"date":"2026-11-06","stars":113872},{"date":"2026-11-07","stars":113878},{"date":"2026-11-08","stars":113883},{"date":"2026-11-09","stars":113890},{"date":"2026-11-10","stars":113900}],"firstPublishedAt":"2026-10-27T01:10:12.618Z","commitSha":"db118c93a429e82b83f091dbaa18fd0a82c06130","verifiedAt":"2026-11-04T01:43:36.484Z","lastCommitAt":"2026-11-03T19:42:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A JavaScript toolkit that programmers use to draw charts and data graphics on a web page.","summary":"D3 is a free JavaScript library for turning data into graphics on a web page, drawn with SVG, Canvas or plain HTML. It gives small building blocks (scales, axes, shapes, colours, animation, zoom) instead of ready-made charts, so you decide how every part looks. It is meant for people who write code: you add it to a website or install it with Node and write JavaScript. There is nothing to open and click, so it is not useful on its own if you do not program.","easeRationale":"The only way to use it is to write JavaScript for a web page, either by installing the package with Node or by linking the file from a CDN, so on the web it is an E: for developers only, with no program an ordinary user can open.","notVerified":["How much work it takes to build a given chart, which depends entirely on what you write","Whether the 30 d3-* packages installed with it are all still maintained","Real disk use after installing with Node (the 100-500 MB figure in the checks is an estimate, not a measurement)"]},"zh-Hans":{"summaryShort":"一套JavaScript工具包，程序员用它在网页上绘制图表和数据可视化内容。","summary":"D3是一个免费的JavaScript库，用来把数据变成网页上的图形，可以用SVG、Canvas或纯HTML绘制。它提供的是一些小的构建模块（比例尺、坐标轴、形状、颜色、动画、缩放），而不是现成的图表，所以每一部分长什么样都由你自己决定。它是给会写代码的人用的：你把它加到网站里，或者用Node安装后自己写JavaScript。它没有可以打开点击的界面，所以如果你不会编程，它本身对你没什么用。","easeRationale":"使用它的唯一方式就是为网页编写JavaScript代码，无论是用Node安装这个包，还是通过CDN链接这个文件，所以在网页端它的评级是E：只适合开发者使用，没有普通用户可以直接打开的程序。","notVerified":["做出某个图表需要花多少工作量，这完全取决于你自己写的代码","随它一起安装的30个d3-*子包是否都还在维护","用Node安装之后实际占用的磁盘空间（检查结果里给出的100-500 MB只是估算值，不是实测值）"]}}},{"id":"R_kgDOM_t66Q","slug":"d4vinci__scrapling","name":"Scrapling","owner":"D4Vinci","fullName":"D4Vinci/Scrapling","type":"dev_tool","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"built_to_get_around_website_bot_protection","severity":"attention","evidence":"README 把绕过防护当作一项主要功能来宣传：\"Anti-bot Bypass: Advanced stealth capabilities with `StealthyFetcher` and fingerprint spoofing. Can easily bypass all types of Cloudflare's Turnstile/Interstitial with automation\"，代码示例使用了 `solve_cloudflare=True`，MCP 部分说它能让 AI 模型 \"bypass any CAPTCHA is faces\"，`pyproject.toml` 把项目描述为 \"undetectable\"。绕过网站用来阻止自动化访问的系统，违反了大多数网站的使用条款，在一些国家还可能违法。README 自己的提醒也说这个库 \"for educational and research purposes only\"，并要求 \"respect the terms of service of websites and robots.txt files\"。","source":"ai","title":"专门设计用来绕过网站的反爬虫防护"},{"code":"robots_txt_is_off_by_default","severity":"attention","evidence":"README 的 Spiders 列表写道：\"Robots.txt Compliance: Optional `robots_txt_obey` flag that respects `Disallow`, `Crawl-delay`, and `Request-rate`\"。一个网站为爬虫发布的规则，只有在你打开这个选项之后才会被遵守。","source":"ai","title":"robots.txt 规则默认是关闭的"},{"code":"can_copy_whole_catalogues_and_hide_your_address","severity":"info","evidence":"现成的模板中包括 `ShopifySpider`，用来 \"collect products from public Shopify stores, one item per variant\"，而抓取器还提供代理轮换、TLS 指纹伪造（`impersonate='chrome'`）和通过 Cloudflare DNS-over-HTTPS 实现的 \"DNS Leak Prevention\"。用这种方式收集个人数据，也会带来隐私法规方面的义务。","source":"ai","title":"可以复制整个商品目录并隐藏你的地址"},{"code":"readme_is_paid_advertising_space","severity":"info","evidence":"README 中有很大一部分是代理服务和 CAPTCHA 破解公司的 \"Platinum Sponsors\" 和 \"Sponsors\" 板块，附带折扣码（\"Use code SCRAPLING35 for 35% discount\"、\"Code SCRAPEANDTAKE: 25% off\"），还有一行写着 \"Do you want to show your ad here?\"。那里的推荐应该当作广告，而不是建议。","source":"ai","title":"README 里有一大块是付费广告位"},{"code":"ai_server_feeds_outside_web_pages_to_your_assistant","severity":"info","evidence":"MCP 服务器 \"Let[s] AI chatbots and agents (Claude/Cursor/etc) scrape through Scrapling with HTTP requests, browser actions, screenshots, saved login state\"。README 说它 \"protect[s] against Prompt-Injection\"，但这只是作者自己的说法；它访问的页面上的文字依然会传到你的助手那里。","source":"ai","title":"AI 服务器会把外部网页内容喂给你的助手"},{"code":"saved_logins_and_captured_traffic_land_on_disk","severity":"info","evidence":"浏览器会话会保留 cookie 和状态（`StealthySession`，\"saved login state\"），`record_requests=True` 会 \"saves text response bodies\" 以便之后搜索。这些文件可能保存着你的登录信息，以及你访问过的网站的数据。","source":"ai","title":"保存的登录状态和抓取的流量会留在磁盘上"},{"code":"gets_around_website_bot_protection","severity":"high","evidence":"README 把这个列为主要功能：\"Anti-bot Bypass: Advanced stealth capabilities with `StealthyFetcher` and fingerprint spoofing. Can easily bypass all types of Cloudflare's Turnstile/Interstitial with automation\"；示例代码使用了 `solve_cloudflare=True`，MCP 服务器部分说它使用 \"our stealthy browser with the right tools to enable suitable AI models to bypass any CAPTCHA is faces while working (not to mention our Cloudflare solver)\"。`pyproject.toml` 把项目描述为 \"undetectable\"，并在关键词中重复了这个词。绕过一个网站用来阻止自动化访问的系统，违反了大多数网站的使用条款，许多国家也有关于规避技术访问保护的法律。README 自己的提醒也说这个库 \"provided for educational and research purposes only\"，并要求 \"Always respect the terms of service of websites and robots.txt files\"，而 spiders 列表显示，遵守 robots.txt 只是一个 \"Optional `robots_txt_obey` flag\"。","source":"ai","title":"绕过网站的反爬虫防护"}],"platforms":["windows","mac","linux"],"tools":["python","docker"],"buildTools":["python","docker"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"BSD-3-Clause","category":"permissive","detectedFrom":"github"},"categories":["development","automation","ai_agents"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"bright-data","name":"Bright Data","confirmed":true}],"declaredAlternatives":[],"stars":101070,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":6278,"growthDays":7,"history":[{"date":"2026-11-03","stars":94792},{"date":"2026-11-04","stars":96350},{"date":"2026-11-05","stars":97830},{"date":"2026-11-06","stars":98776},{"date":"2026-11-07","stars":99269},{"date":"2026-11-08","stars":99799},{"date":"2026-11-09","stars":100480},{"date":"2026-11-10","stars":101070}],"firstPublishedAt":"2026-11-03T02:51:37.357Z","commitSha":"95cf9019ab5a5d6f3e9654b7ef1c6a26e29dc5c0","verifiedAt":"2026-11-03T02:40:16.501Z","lastCommitAt":"2026-11-09T10:36:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Python toolkit that collects data from websites, from one page to a whole site, and saves it as JSON, CSV or Markdown.","summary":"A Python library and terminal command for collecting data from websites. It can make plain web requests, drive a hidden browser for pages that need JavaScript, follow links across a whole site with several pages at a time, pause and resume, and save the results as JSON, JSONL, CSV, XML or clean Markdown. It also ships a server so AI assistants can browse and pull data through it. It is made for programmers, although the `scrapling extract` command saves a page without writing code. You need Python 3.10+ and, for the browser part, a one-time download of a browser.","easeRationale":"The same on Windows, macOS and Linux: install Python 3.10+, then `pip install \"scrapling[fetchers]\"` and `scrapling install` to download the browser, and use it from a terminal or from your own Python code. A ready Docker image with the browsers inside exists as well, which is one tool more to install.","costNote":"The library is free (BSD-3-Clause). Large jobs usually mean paying a proxy company, and the README is sponsored by several of them.","notVerified":["Whether the Cloudflare and CAPTCHA handling still works today, since sites change their protection constantly","The speed numbers in the README, which were measured by the project itself","Real disk use after `scrapling install` downloads a full browser and its system files","Whether a crawl stays polite enough not to get your address blocked by the target site","Whether the claim of protection against prompt injection in the AI server holds up"]},"zh-Hans":{"summaryShort":"一个 Python 工具包，可以从网站采集数据，从单个页面到整个网站都行，并把结果保存为 JSON、CSV 或 Markdown。","summary":"一个用于从网站采集数据的 Python 库和终端命令。它可以发送普通网络请求，为需要 JavaScript 的页面驱动一个隐藏浏览器，一次跟踪整个网站的多个页面链接，支持暂停和继续，并把结果保存为 JSON、JSONL、CSV、XML 或整洁的 Markdown。它还附带一个服务器，让 AI 助手可以通过它浏览并提取数据。它是为程序员设计的，不过 `scrapling extract` 命令不写代码也能保存一个页面。你需要 Python 3.10+，浏览器部分还需要一次性下载一个浏览器。","easeRationale":"Windows、macOS 和 Linux 的步骤相同：安装 Python 3.10+，然后运行 `pip install \"scrapling[fetchers]\"` 和 `scrapling install` 来下载浏览器，再在终端或自己的 Python 代码里使用它。还有一个内置浏览器的现成 Docker 镜像，那又是多一个需要安装的工具。","costNote":"该库是免费的（BSD-3-Clause 许可）。大规模的任务通常意味着要向某家代理服务公司付费，而 README 本身也接受了多家这类公司的赞助。","notVerified":["Cloudflare 和 CAPTCHA 的处理方式今天是否依然有效，因为网站的防护一直在变化","README 中的速度数据，是项目方自己测得的","`scrapling install` 下载完整浏览器及其系统文件之后，实际占用的磁盘空间","一次爬取是否足够克制，不会导致目标网站封锁你的地址","AI 服务器中关于防范提示注入的说法是否真的站得住"]}}},{"id":"R_kgDOPzwRpg","slug":"damontecres__wholphin","name":"Wholphin","owner":"damontecres","fullName":"damontecres/Wholphin","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"needs_your_own_media_server","severity":"info","evidence":"README的Compatibility部分写道：\"Requires Android 6+ (or Fire TV OS 6+) and Jellyfin server 10.10.x or 10.11.x\"。这款应用本身不带任何内容，单独使用什么也做不了。","source":"ai","title":"需要你自己的媒体服务器"},{"code":"sideloading_asks_to_allow_unknown_apps","severity":"attention","evidence":"README的Installation第1步写道：\"Enable side-loading 'unknown' apps\"，并附有在Google TV和Fire TV上如何允许未知来源的链接。这个设置之后会对所有应用一直生效，所以走应用商店这条路更温和。","source":"ai","title":"旁加载需要允许未知应用"},{"code":"development_builds_sit_next_to_the_stable_one","severity":"info","evidence":"最近的三个发布版本是\"v1.1.0\"、\"develop-jellyfin12\"和\"develop\"（都是2026-11-03）。在发布页面上，很容易一不小心选到开发版本。","source":"ai","title":"开发版和稳定版放在一起"},{"code":"one_feature_depends_on_where_you_installed_it","severity":"info","evidence":"README的Features部分提到，用于发现电影和剧集的Seerr整合功能\"only available when installed from GitHub ... or the Play store\"——所以Amazon AppStore版本没有这个功能。","source":"ai","title":"有一个功能取决于你从哪里安装"},{"code":"no_binaries_detected_in_releases","severity":"info","evidence":"Radar的检查对最近三个发布版本报告\"no platform binaries\"，因为这些文件是面向电视设备的Android APK，而不是Windows、Mac或Linux的安装程序。","source":"ai","title":"发布版本中未检测到二进制文件"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["video","hardware"],"alternativeTo":[{"slug":"plex","name":"Plex","confirmed":true},{"slug":"infuse","name":"Infuse","confirmed":true}],"declaredAlternatives":[],"stars":3290,"downloads":82468,"downloadsWeek":3220,"downloadsDays":6,"growth":43,"growthDays":6,"history":[{"date":"2026-11-04","stars":3247},{"date":"2026-11-05","stars":3259},{"date":"2026-11-06","stars":3266},{"date":"2026-11-07","stars":3275},{"date":"2026-11-08","stars":3280},{"date":"2026-11-09","stars":3286},{"date":"2026-11-10","stars":3290}],"firstPublishedAt":"2026-11-04T01:36:50.330Z","commitSha":"2bb662fd603143753493d13e2b91ac6d18bdb8f5","verifiedAt":"2026-11-04T01:21:32.810Z","lastCommitAt":"2026-11-09T19:59:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free Android TV and Fire TV app to watch the films, series and music kept on your own Jellyfin media server.","summary":"Wholphin is a TV app for watching your own media library. It does not hold any films itself: it connects to a Jellyfin server that you (or someone you know) already run, and shows the library on the television. The home screen can be rearranged, there are two playback engines (ExoPlayer and MPV), Live TV and DVR, music, subtitles, chapters and theme music. The look is inspired by Plex, for people moving from Plex to Jellyfin. It runs on Android TV or Fire TV (Android 6 or newer) with a Jellyfin server 10.10 or 10.11.","easeRationale":"It is a TV app for Android TV and Fire TV, so none of the computer platforms get a grade. Installing is easy: it is in Google Play and the Amazon AppStore, and after that you only type your Jellyfin server address and log in. Without an app store you have to turn on installing from unknown sources and copy the APK file to the TV, which is harder.","costNote":"The app is free (GPL-2.0) on Google Play, the Amazon AppStore and GitHub. You still need a Jellyfin server, which is free software but runs on a computer or a box you pay for.","notVerified":["Whether it works smoothly on your particular TV or stick; the author says it is tested on many Android TV and Fire TV devices but asks for issue reports.","How much space it takes on the TV and how it behaves on older, slower sticks.","Whether the MPV engine plays your files without the server re-encoding them, which depends on your files and your TV.","What the app sends to the Jellyfin server and whether the automatic update check contacts anything else.","Whether subtitle download, Live TV, cinema mode and the Seerr discovery work, since each needs an extra server plugin or server."]},"zh-Hans":{"summaryShort":"一款免费的Android TV和Fire TV应用，用来观看你自己Jellyfin媒体服务器上保存的电影、剧集和音乐。","summary":"Wholphin是一款用来观看你自己媒体库的电视应用。它本身不存放任何影片：它连接到你（或你认识的人）已经运行的Jellyfin服务器，并把媒体库显示在电视上。主屏幕可以重新排列，有两种播放引擎（ExoPlayer和MPV）、直播电视和DVR、音乐、字幕、章节和主题音乐。界面风格受Plex启发，适合从Plex转到Jellyfin的人。它可以在Android TV或Fire TV（Android 6或更新版本）上运行，需要配合10.10或10.11版本的Jellyfin服务器。","easeRationale":"这是一款面向Android TV和Fire TV的电视应用，所以电脑平台都没有评分。安装很简单：它在Google Play和Amazon AppStore都有上架，安装后你只需输入Jellyfin服务器地址并登录。如果没有应用商店，你需要打开允许安装未知来源应用的设置，并把APK文件拷贝到电视上，这样做更麻烦。","costNote":"这款应用是免费的（GPL-2.0），可以在Google Play、Amazon AppStore和GitHub上获取。你仍然需要一台Jellyfin服务器，它是免费软件，但要运行在你自己购买的电脑或盒子上。","notVerified":["它在你具体的电视或电视棒上是否运行顺畅；作者说已经在很多Android TV和Fire TV设备上测试过，但希望大家反馈问题。","它在电视上占用多少存储空间，以及在较旧、较慢的电视棒上表现如何。","MPV引擎是否能直接播放你的文件而不需要服务器重新编码，这取决于你的文件和电视。","这款应用向Jellyfin服务器发送了什么内容，以及自动更新检查是否还联系了其他地方。","字幕下载、直播电视、影院模式和Seerr发现功能是否能用，因为每一个都需要额外的服务器插件或服务器。"]}}},{"id":"R_kgDOKVTz6A","slug":"danni-cool__wechatbot-webhook","name":"wechatbot-webhook","owner":"danni-cool","fullName":"danni-cool/wechatbot-webhook","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"archived","severity":"info","evidence":"该仓库已被归档(只读,不再维护)","source":"check","title":"已归档"},{"code":"account_restriction_risk_stated_by_author","severity":"attention","evidence":"README 的注意事项写道:\"项目目前基于web微信，其本身就有被限制风险，另外大概两天一掉线\" ——项目使用的是微信网页版,作者表示这本身就有账号被限制的风险,而且大约每两天就会掉一次线。作者还表示不再接受新的功能请求。","source":"ai","title":"作者说明存在账号受限风险"},{"code":"unofficial_wechat_interface","severity":"attention","evidence":"它通过带有本地补丁的第三方库驱动微信:package.json 中列出了\"wechaty\": \"^1.20.2\",以及针对 wechat4u@0.7.14、wechaty-puppet-wechat4u@1.14.13 和 wechaty-puppet@1.20.2 的 patchedDependencies。这并不是微信的官方接口,以这种方式自动化操作个人账号违反了平台规则;功能表也显示不支持企业微信消息(#142)。","source":"ai","title":"非官方的微信接口"},{"code":"service_holds_your_logged_in_account","severity":"attention","evidence":"你扫描二维码把自己的微信登录到该服务中(\"/login?token=[YOUR_PERSONAL_TOKEN]\"),默认情况下它会记住这个会话(DISABLE_AUTO_LOGIN),唯一的保护措施是一个令牌,如果你不自己设置,它就会被生成写入 .env 文件。任何拿着这个令牌访问 3001 端口的人都能读取收到的消息、通过 /resouces 获取头像,并以你的身份发送消息。","source":"ai","title":"服务掌握着你已登录的账号"},{"code":"documentation_in_chinese_only","severity":"info","evidence":"整个 README、功能表和 .env.example 的注释都是用中文写的(例如\"启动服务的端口\"\"扫码登陆\"),仓库里没有英文版本。","source":"ai","title":"文档只有中文"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Other (LOCAL_LOGIN_API_TOKEN)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"twilio-whatsapp-api","name":"WhatsApp API","confirmed":true}],"declaredAlternatives":[],"stars":2164,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":2165},{"date":"2026-11-08","stars":2165},{"date":"2026-11-09","stars":2165},{"date":"2026-11-10","stars":2164}],"firstPublishedAt":"2026-11-06T22:59:23.770Z","commitSha":"fcd7b47687e7a3a5ef935701b60b4160e76c40c2","verifiedAt":"2026-11-06T22:39:03.588Z","lastCommitAt":"2024-12-01T15:56:43Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Small server that sends and receives WeChat messages over simple web calls, so tools like n8n can chat for you.","summary":"This project turns a personal WeChat account into a web service. You start it with one command or a Docker container, scan a QR code with your phone, and then any tool that can call a URL (n8n, Coze, your own script, an AI app) can send text, images, videos and files to people or groups and receive incoming messages on an address you choose. It runs on Windows, macOS, Linux and ARM boards. The repository was archived in December 2024 and the documentation is in Chinese only.","easeRationale":"On any of the three systems it is a C: with Node 18.14.1 or newer you run \"npx wechatbot-webhook\", then open a local address and scan a QR code with WeChat; the recommended route is a Docker container (\"docker run -d --name wxBotWebhook -p 3001:3001 dannicool/docker-wechatbot-webhook\") plus reading the container log to find the login link. There is no installer and every later step is an HTTP call you build yourself.","costNote":"MIT licensed and free. You need a machine or small server to keep it running.","notVerified":["Whether the QR-code login still works today, since the repository was archived in December 2024 and the web version of WeChat changes often","How long a session stays connected in practice and how often you must scan again","Whether WeChat limits or blocks an account used this way","Real disk and memory use of the container (the 100-500 MB figure is an estimate, not measured)","What the 'windows protocol' branch announced in the README actually does, since the table marks it as not available"]},"zh-Hans":{"summaryShort":"小型服务器通过简单的网页调用收发微信消息,让 n8n 这类工具可以替你聊天。","summary":"这个项目把个人微信账号变成一个网络服务。你用一条命令或一个 Docker 容器启动它,用手机扫一个二维码,之后任何能调用 URL 的工具(n8n、Coze、你自己写的脚本、某个 AI 应用)都能向好友或群组发送文字、图片、视频和文件,并在你指定的地址上接收收到的消息。它可以在 Windows、macOS、Linux 和 ARM 开发板上运行。该仓库已在 2024 年 12 月被归档,文档只有中文版本。","easeRationale":"在这三种系统上都只能通过命令行安装:在 Node 18.14.1 或更新版本下运行 \"npx wechatbot-webhook\",然后打开一个本地地址,用微信扫码;推荐的方式是使用 Docker 容器(\"docker run -d --name wxBotWebhook -p 3001:3001 dannicool/docker-wechatbot-webhook\"),再查看容器日志找到登录链接。没有安装程序,之后的每一步都要靠你自己构建 HTTP 调用。","costNote":"采用 MIT 许可,免费使用。你需要一台机器或一个小型服务器来让它持续运行。","notVerified":["二维码登录今天是否还能用,因为该仓库已在2024年12月归档,而微信网页版经常变动","一个会话在实际使用中能保持连接多久,以及大概多久需要重新扫码一次","微信是否会限制或封禁以这种方式使用的账号","容器实际占用的磁盘和内存(100-500 MB 这个数字只是估计,并未实际测量)","README 中提到的'windows protocol'分支实际做了什么,因为表格把它标记为尚不可用"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzNzkxODM1","slug":"darktable-org__darktable","name":"darktable","owner":"darktable-org","fullName":"darktable-org/darktable","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"database_upgrade_cannot_be_undone","severity":"attention","evidence":"README，Updating from older versions：\"Migration to a newer database structure/newer release means that your edits (both new and old) will no longer be compatible with older versions of darktable. Upgrades are definitive.\" 程序在升级前确实会自动备份数据库。","source":"ai","title":"数据库升级无法撤销"},{"code":"newest_github_release_is_a_nightly_build","severity":"info","evidence":"GitHub 上最近的发布版标记为 \"nightly\" (2026-11-02)，而稳定版是 2026-10-04 的 release-5.6.2；README 说开发快照 \"is intended for testing and is generally not safe\"，并说 master 分支 \"might corrupt your database and XMP files, result in loss of data and edit history\"。","source":"ai","title":"GitHub 上最新的发布版是每夜构建"},{"code":"windows_and_mac_get_less_testing","severity":"info","evidence":"README，Supported platforms：\"other platforms MAY have additional bugs that are not present in the Linux version. In addition, for example, on the Windows platform, printing support is not currently implemented. So we recommend, if you have a choice of platform, to use darktable on Linux.\"","source":"ai","title":"Windows 和 Mac 得到的测试较少"},{"code":"mac_needs_apple_silicon_and_macos_15","severity":"info","evidence":"README 只列出 \"Apple Silicon Macs running macOS 15 and later\"，最新的发布版也只提供 arm64 的 .dmg。","source":"ai","title":"Mac 需要 Apple Silicon 和 macOS 15"},{"code":"optional_ai_features_need_extra_steps","severity":"info","evidence":"README，AI features (optional)：\"require building with -DUSE_AI=ON (off by default)\"，模型要从另一个仓库下载，显卡加速则需要把项目仓库中的脚本通过管道交给你的 shell 来设置 (Linux 上是 \"curl -fsSL ... | bash\"，Windows 上是 \"irm ... | iex\")，另外还要 CUDA、ROCm 或 OpenVINO。","source":"ai","title":"可选的 AI 功能需要额外步骤"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":141,"maxMB":424,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"lightroom","name":"Lightroom","confirmed":true},{"slug":"adobe-camera-raw","name":"Camera Raw","confirmed":true},{"slug":"capture-one","name":"Capture One","confirmed":true}],"declaredAlternatives":[],"stars":13434,"downloads":9190087,"downloadsWeek":40693,"downloadsDays":7,"growth":41,"growthDays":7,"history":[{"date":"2026-10-25","stars":13328},{"date":"2026-10-26","stars":13335},{"date":"2026-10-27","stars":13346},{"date":"2026-10-28","stars":13355},{"date":"2026-10-29","stars":13357},{"date":"2026-10-30","stars":13362},{"date":"2026-10-31","stars":13368},{"date":"2026-11-01","stars":13376},{"date":"2026-11-02","stars":13384},{"date":"2026-11-03","stars":13393},{"date":"2026-11-04","stars":13396},{"date":"2026-11-05","stars":13403},{"date":"2026-11-06","stars":13405},{"date":"2026-11-07","stars":13406},{"date":"2026-11-08","stars":13416},{"date":"2026-11-09","stars":13428},{"date":"2026-11-10","stars":13434}],"firstPublishedAt":"2026-10-25T23:18:59.247Z","commitSha":"0c67b465033205c46735c5a194f2b23d55e68363","verifiedAt":"2026-11-02T02:57:40.369Z","lastCommitAt":"2026-11-10T19:22:06Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"darktable-5.6.2-win64.exe","sizeBytes":138844821,"url":"https://github.com/darktable-org/darktable/releases/download/release-5.6.2/darktable-5.6.2-win64.exe","releaseUrl":"https://github.com/darktable-org/darktable/releases/tag/release-5.6.2"},{"os":"mac","file":"darktable-5.6.2-arm64.dmg","sizeBytes":88884896,"url":"https://github.com/darktable-org/darktable/releases/download/release-5.6.2/darktable-5.6.2-arm64.dmg","releaseUrl":"https://github.com/darktable-org/darktable/releases/tag/release-5.6.2"},{"os":"linux","file":"Darktable-5.6.2-x86_64.AppImage","sizeBytes":178346488,"url":"https://github.com/darktable-org/darktable/releases/download/release-5.6.2/Darktable-5.6.2-x86_64.AppImage","releaseUrl":"https://github.com/darktable-org/darktable/releases/tag/release-5.6.2"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Organises and develops RAW photos from your camera, with a ready installer for Windows, Mac and Linux.","summary":"darktable manages a photo collection in a database and develops the RAW files a camera produces: you sort and rate pictures on a light table, then adjust exposure, colour, noise, lens faults and local areas in the darkroom, and export. Your files are never changed; the edits live in the database and in sidecar files. It has been going since 2012, with two main releases a year and ready installers for Windows 10 or later, Apple Silicon Macs on macOS 15 or later, and Linux. The project states plainly that it is not a free replacement for Adobe Lightroom.","easeRationale":"Each release has a ready file for every system: a .exe installer for Windows (also for Windows on ARM), a .dmg for Apple Silicon Macs, and an AppImage, Flatpak or distribution package for Linux. You download it and open it. The project recommends Linux because that is where it is developed.","notVerified":["Whether the pre-built package you download includes Lua, OpenCL, OpenMP and Colord, which the README says improve the experience.","How usable it feels on a computer without a graphics card, since the project warns that denoise, local contrast, retouch and liquify can be \"slow beyond usable\".","Whether your camera and lens are covered by RawSpeed, LibRaw and the Lensfun database.","Real disk space once the library database, thumbnails and optional AI models are there.","Whether the optional AI features are enabled in the official builds, since the build flag is off by default."]},"zh-Hans":{"summaryShort":"整理并冲洗相机的 RAW 照片，提供适用于 Windows、Mac 和 Linux 的现成安装程序。","summary":"darktable 用数据库管理照片收藏，并冲洗相机生成的 RAW 文件：你在灯箱里整理和评分照片，然后在暗房里调整曝光、色彩、噪点、镜头缺陷和局部区域，最后导出。你的文件不会被改动；编辑内容保存在数据库和附属文件中。它从 2012 年起一直在开发，每年有两次主要版本发布，并提供适用于 Windows 10 或更高版本、macOS 15 或更高版本的 Apple Silicon Mac 以及 Linux 的现成安装程序。项目明确表示，它并不是 Adobe Lightroom 的免费替代品。","easeRationale":"每个版本都为每种系统提供现成文件：Windows 的 .exe 安装程序（也支持 ARM 版 Windows），Apple Silicon Mac 的 .dmg，以及 Linux 的 AppImage、Flatpak 或发行版软件包。下载后打开即可。项目推荐使用 Linux，因为开发就是在 Linux 上进行的。","notVerified":["你下载的预编译包是否包含 README 所说能改善体验的 Lua、OpenCL、OpenMP 和 Colord。","在没有显卡的电脑上用起来如何，因为项目警告说降噪、局部对比度、修饰和液化可能 \"slow beyond usable\"。","你的相机和镜头是否被 RawSpeed、LibRaw 和 Lensfun 数据库所覆盖。","资料库数据库、缩略图和可选 AI 模型都存到磁盘后的实际磁盘空间。","官方构建中是否启用了可选的 AI 功能，因为该构建开关默认是关闭的。"]}}},{"id":"R_kgDOPeGy7g","slug":"darkzogx__youtube-automation-agent","name":"YouTube Automation Agent","owner":"darkzOGx","fullName":"darkzOGx/youtube-automation-agent","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"a_channel_that_publishes_by_itself","severity":"attention","evidence":"README 里的日程安排是 \"Daily at 06:00 Content generation\"、每 \"Every 15 minutes\" 跑一次的发布队列,还有一个用 \"Activate & run now\" 启动的 \"Autonomous operator\",`.env.example` 把它限制在 `GLOBAL_RATE_LIMIT_PER_HOUR=50`。YouTube 对批量生产和重复内容的规定,让这样发布的频道面临失去收益资格甚至被下架的风险,README 结尾那行也提醒你要 \"Comply with YouTube's Terms of Service and Community Guidelines\"。","source":"ai","title":"一个自己发布内容的频道"},{"code":"dashboard_can_be_left_open_on_the_network","severity":"attention","evidence":"`.env.example` 写道:\"The dashboard listens on this machine only by default. Use HOST=0.0.0.0 to reach it from other devices, and set API_KEY below when you do\",而且 `API_KEY=` 默认是空的,这层保护被描述为可选:\"If set, mutating API routes require this value in the x-api-key header\"。谁要是开放了端口却没填这个密钥,就等于把生成视频和上传到频道的接口交给了同一网络里的任何人。","source":"ai","title":"控制台可能毫无保护地暴露在网络上"},{"code":"permission_to_upload_to_your_channel_sits_on_the_machine","severity":"info","evidence":"设置过程需要一个 \"Desktop app\" 类型的 OAuth 2.0 客户端,保存为 `config/credentials.json`,并开通 YouTube Data API v3 和 YouTube Analytics API。从那之后,这个程序就一直拥有在你频道发布内容和读取数据的权限,而依赖库 `@google-cloud/local-auth` 会把登录状态保存在硬盘上。","source":"ai","title":"上传到你频道的权限留在这台机器上"},{"code":"it_answers_your_viewers","severity":"info","evidence":"Audience Engagement Studio 提供 \"reply drafts that post only after you approve them\" 这样的功能,`.env.example` 把上限设为每天 `ENGAGEMENT_DAILY_REPLY_CAP=50` 条回复。也就是说,你视频下面的评论可能是 AI 写的,读者却看不到任何提示。","source":"ai","title":"它会回复你的观众"},{"code":"paid_seconds_add_up_quietly","severity":"info","evidence":"生成的视频片段是由服务商计费的,这也是为什么会有这些设置:`VIDEO_MAX_GENERATED_SECONDS=60`、`VIDEO_CLIP_DURATION=8`、`MAX_CONCURRENT_JOBS=1`,注释写着 \"Optional. Generation jobs run one at a time by default to control API spend\",还有 \"Paid image and video probes are separate opt-ins\"。仓库里任何地方都没有给出每个视频的价格。","source":"ai","title":"按秒计费的开销会悄悄累积"},{"code":"the_install_brings_a_whole_browser","severity":"info","evidence":"`npx playwright install chromium` 会下载一个完整的 Chromium 来渲染幻灯片视频,`ffmpeg-static` 还会附带一个打包好的 FFmpeg;生成的 MP4 和上传内容之后会堆积在 `./uploads` 里(`MAX_FILE_SIZE=52428800`)。检查里给出的磁盘估算只是猜测。","source":"ai","title":"安装过程会带来一整个浏览器"},{"code":"can_call_a_script_from_another_repository","severity":"info","evidence":"`.env.example` 里有 `DARKZSEO_PYTHON=python` 和 `DARKZSEO_PATH=../darkzseo/darkzseo.py`,用于 \"Discoverability preflight\",描述为 \"bundled and advisory; external checkout is explicit and falls back safely\"。如果你把它指向那里,Node 就会运行作者另一个独立项目 darkzseo 里的 Python 文件,这需要你安装好 Python。","source":"ai","title":"可以调用另一个仓库里的脚本"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg"],"buildTools":["node","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["One text AI service: Gemini, OpenAI, OpenRouter, Kimi, MiMo, GLM, DeepSeek or MiniMax","YouTube Data API v3 and YouTube Analytics API, as a Google Cloud \"Desktop app\" OAuth client","ElevenLabs or Azure Speech for better narration (optional)","Replicate, MiniMax, Kling or Alibaba DashScope for generated video clips (optional)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","social_marketing","ai_agents"],"alternativeTo":[{"slug":"invideo-ai","name":"AI","confirmed":true},{"slug":"pictory","name":"Pictory","confirmed":true},{"slug":"vidiq","name":"vidIQ","confirmed":true},{"slug":"tubebuddy","name":"TubeBuddy","confirmed":false},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":false}],"declaredAlternatives":[],"stars":5126,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":131,"growthDays":6,"history":[{"date":"2026-11-06","stars":5038},{"date":"2026-11-07","stars":5058},{"date":"2026-11-08","stars":5075},{"date":"2026-11-09","stars":5096},{"date":"2026-11-10","stars":5126}],"firstPublishedAt":"2026-11-05T22:43:48.343Z","commitSha":"e973d515aef90c467da2289144df737a3bd03859","verifiedAt":"2026-11-05T22:31:10.972Z","lastCommitAt":"2026-10-27T00:49:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs a YouTube channel on your computer: picks topics, writes the script, makes narration and video, and uploads what you approve.","summary":"YouTube Automation Agent is a program you run on your own machine with a web dashboard. It researches topics, writes the script, generates the narration and the pictures, assembles an MP4, writes the title, description and tags, and leaves each video waiting for your approval before it is scheduled and uploaded to your channel. Afterwards it reads your YouTube statistics and comments and suggests changes. You need Node.js, a Google Cloud project with the YouTube API switched on, and a key for at least one AI service.","easeRationale":"A C on Windows, Mac and Linux: none of the 5 releases has a file to download, so you install Node.js 18.17 or newer, then `git clone`, `npm install`, `npx playwright install chromium` (on Linux with `--with-deps`), `npm run walkthrough` and `npm start`, and open `http://localhost:3456`. FFmpeg comes with `npm install`. The slow part is outside the terminal: creating a project in Google Cloud Console, switching on two YouTube APIs, making a Desktop-app OAuth client, saving it as `config/credentials.json` and adding your own Google account as a test user, after which Google expires the sign-in every 7 days.","costNote":"The code is MIT and nothing is charged for it, but it does not work without a key for at least one AI service. The README says Gemini has free tiers for text and narration, and the default video is a slideshow built on your own machine; ElevenLabs narration and the five video generators (Seedance, MiniMax H3, Gemini Omni, Kling, Wan) are paid by the second, which is why there are settings to cap them.","notVerified":["Whether the finished videos are watchable, since the screenshots \"use sample data\"","What a video really costs in text, narration, image and clip fees","Whether YouTube keeps accepting a channel fed this way, and what it does to monetisation","Whether the whole chain works on each system, including the `playwright install chromium` step","How much disk the browser, FFmpeg, the generated media and the analytics database take"]},"zh-Hans":{"summaryShort":"在你的电脑上运营一个 YouTube 频道:选题、写脚本、生成旁白和视频,并上传你批准的内容。","summary":"YouTube Automation Agent 是一个在你自己电脑上运行的程序,带有一个网页控制台。它会研究选题、写脚本、生成旁白和图片、合成 MP4、写标题、描述和标签,并让每个视频在排期和上传到你的频道之前都等待你的批准。之后它还会读取你的 YouTube 数据和评论,并给出改进建议。你需要 Node.js、一个开通了 YouTube API 的 Google Cloud 项目,以及至少一个 AI 服务的密钥。","easeRationale":"在 Windows、Mac 和 Linux 上评级是 C:5 个发行版都没有可下载的文件,所以你要安装 Node.js 18.17 或更新版本,然后依次运行 `git clone`、`npm install`、`npx playwright install chromium`(在 Linux 上加 `--with-deps`)、`npm run walkthrough` 和 `npm start`,再打开 `http://localhost:3456`。FFmpeg 会随 `npm install` 一起装好。慢的部分是在终端之外:在 Google Cloud Console 创建一个项目,开通两个 YouTube API,创建一个“桌面应用”类型的 OAuth 客户端,保存为 `config/credentials.json`,再把你自己的 Google 账号加为测试用户,之后 Google 会让这个登录每 7 天就过期一次。","costNote":"代码是 MIT 许可,本身不收费,但没有至少一个 AI 服务的密钥就无法运行。README 说 Gemini 在文本和旁白上有免费额度,默认视频是在你自己电脑上生成的幻灯片;而 ElevenLabs 的旁白和五个视频生成器(Seedance、MiniMax H3、Gemini Omni、Kling、Wan)都是按秒计费,这也是为什么有设置可以限制它们用量的原因。","notVerified":["做出来的视频是否真的能看,因为截图 \"use sample data\"","一个视频在文本、旁白、图片和片段费用上到底要花多少钱","YouTube 是否还会接受这样生产出来的频道,这对收益化有什么影响","整条链路在每个系统上是否都能跑通,包括 `playwright install chromium` 这一步","浏览器、FFmpeg、生成的素材和分析数据库一共占用多少磁盘空间"]}}},{"id":"R_kgDOT3Z-Mw","slug":"dataelement__dsh-desktop","name":"DSHDesktop","owner":"dataelement","fullName":"dataelement/dsh-desktop","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B"},"risk":"attention","signals":[{"code":"early_preview_on_alpha_building_blocks","severity":"attention","evidence":"README 的 IMPORTANT 提示框写道：\"DSH Desktop is an early preview built on the rapidly evolving @deepseek-ai/dsh@0.2.2-alpha.2.\" package.json 中几乎所有依赖都是 alpha 版本，比如 \"@deepseek-ai/dsh-agent\": \"0.2.2-alpha.2\" 和 \"@deepseek-ai/cordis\": \"4.0.6-alpha.1\"。应用本身的版本是 0.13.1。","source":"ai","title":"建立在 alpha 阶段组件上的早期预览版"},{"code":"agent_can_run_commands_and_change_your_files","severity":"attention","evidence":"package.json 依赖 \"@deepseek-ai/dsh-bash-local\" 和 \"@deepseek-ai/dsh-bash-sandbox\"，README 说该应用 \"Uses the native system directory picker to add and manage project workspaces\"。这个 AI 会在你给它的文件夹里运行命令并编辑文件。","source":"ai","title":"代理可以运行命令并修改你的文件"},{"code":"can_open_a_public_internet_tunnel","severity":"attention","evidence":"README 的 \"Phone access\" 部分写道：\"Cloudflare Quick Tunnel is the usual long-lived remote path. Free Pinggy is the fallback when Cloudflare is unavailable.\" 通过这个隧道，你的手机会经由一个第三方服务连接到你电脑上的代理，由一个可以设为长期有效的 6 位配对密码保护。","source":"ai","title":"可以打开一个公开的互联网隧道"},{"code":"installs_plugins_and_presets_from_outside","severity":"info","evidence":"README 提到了 \"dsh-market, the community plugin market\"，并说预设是以 \"portable .dshpreset packages, with conflict checks and a trust warning before installation\" 的方式导入的。之所以存在一个 \"Safe Mode\"（保护模式），正是因为 \"a third-party plugin interferes with startup or rendering\"。","source":"ai","title":"会安装来自外部的插件和预设"},{"code":"anonymous_identifier_in_the_app","severity":"info","evidence":"package.json 依赖 \"@deepseek-ai/dsh-anonymous-user-id\"。README 没有解释这个标识符的用途，但提到 \"Installed builds check for updates shortly after startup and every six hours.\"","source":"ai","title":"应用内部有一个匿名标识符"},{"code":"not_made_by_deepseek","severity":"info","evidence":"README 的 License 部分写道：\"DSH Desktop is an independent community desktop application\"，以及 \"DeepSeek Harness and its dependencies remain subject to their respective upstream licenses and trademark policies.\" 这个名字、Logo 以及 @deepseek-ai 系列包，都可能让人以为它是 DeepSeek 自家出品的应用。","source":"ai","title":"并非 DeepSeek 官方制作"}],"platforms":["windows","mac"],"tools":[],"buildTools":["node"],"disk":{"minMB":268,"maxMB":803,"basis":"installer"},"apiKeys":["DeepSeek","Other model providers"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","local_ai","office"],"alternativeTo":[{"slug":"gamma","name":"Gamma","confirmed":true},{"slug":"manus","name":"Manus","confirmed":true},{"slug":"powerpoint","name":"PowerPoint","confirmed":false}],"declaredAlternatives":[],"stars":17900,"downloads":144185,"downloadsWeek":6110,"downloadsDays":7,"growth":708,"growthDays":7,"history":[{"date":"2026-10-26","stars":17031},{"date":"2026-10-27","stars":17149},{"date":"2026-10-28","stars":16875},{"date":"2026-10-29","stars":16975},{"date":"2026-10-30","stars":17043},{"date":"2026-10-31","stars":17004},{"date":"2026-11-01","stars":17058},{"date":"2026-11-02","stars":17108},{"date":"2026-11-03","stars":17192},{"date":"2026-11-04","stars":17271},{"date":"2026-11-05","stars":17433},{"date":"2026-11-06","stars":17489},{"date":"2026-11-07","stars":17546},{"date":"2026-11-08","stars":17592},{"date":"2026-11-09","stars":17655},{"date":"2026-11-10","stars":17900}],"firstPublishedAt":"2026-10-27T01:10:12.619Z","commitSha":"3e8db1f600ee1b893098965d06aa5a0168692d9b","verifiedAt":"2026-11-03T02:54:29.094Z","lastCommitAt":"2026-11-06T11:54:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop app for Windows and Mac that runs a DeepSeek AI agent on your computer, including a mode that builds editable slide decks.","summary":"DSH Desktop is an installable app that runs the DeepSeek Harness AI agent on your own computer, so you do not need a terminal or a browser tab. You pick folders for the agent to work in, choose DeepSeek's models or another provider, and keep your sessions and settings between updates. A PPT mode turns your material into editable PPTX decks from 16 templates, and Word and Excel skills work offline. You can pair a phone to continue a session. It needs a model account or key to do anything, and the installer is 268 MB on Windows and about 350 MB on Mac.","easeRationale":"Both Windows and Mac get a ready installer that bundles everything, including the Python libraries for the Office skills, so the only extra step is signing in to DeepSeek or entering another provider's key. Linux and Windows on ARM have no grade because the project says they are not supported.","costNote":"The app is free (MIT). The AI models are not: you pay DeepSeek or another provider for what the agent uses.","notVerified":["Whether the installers on the GitHub releases page were built from the code in this repository","What the DeepSeek models cost in practice for a day of agent work","How much disk the app uses after it stores sessions, workspaces and plugins, beyond the installer size","What leaves your computer when the phone tunnel is open, and what the tunnel provider can see","Whether the slide decks it produces come out usable without hand editing"]},"zh-Hans":{"summaryShort":"一款面向 Windows 和 Mac 的桌面应用，在你的电脑上运行 DeepSeek AI 代理，还包含一个能生成可编辑幻灯片的模式。","summary":"DSH Desktop 是一款可安装的应用，在你自己的电脑上运行 DeepSeek Harness AI 代理，这样你就不需要终端或浏览器标签页。你可以选择让代理工作的文件夹，选择 DeepSeek 的模型或其他服务商，并在更新之间保留你的会话和设置。PPT 模式可以把你的材料变成可编辑的 PPTX 幻灯片，提供 16 种模板，Word 和 Excel 技能可以离线使用。你还可以配对手机来继续一个会话。它需要一个模型账号或密钥才能做任何事，安装包在 Windows 上是 268 MB，在 Mac 上大约是 350 MB。","easeRationale":"Windows 和 Mac 都有现成的安装包，里面打包了所有东西，包括用于 Office 技能的 Python 库，所以唯一额外的步骤就是登录 DeepSeek 或输入另一个服务商的密钥。Linux 和基于 ARM 的 Windows 没有评级，因为项目方说这两者不受支持。","costNote":"该应用是免费的（MIT 许可）。AI 模型不是：你需要为代理使用的部分向 DeepSeek 或其他服务商付费。","notVerified":["GitHub 发布页面上的安装包是否是用这个仓库里的代码构建的","DeepSeek 模型在实际使用中，一天的代理工作要花多少钱","应用在保存会话、工作区和插件之后，除安装包大小之外实际占用多少磁盘空间","手机隧道打开时，有什么内容会离开你的电脑，隧道服务商又能看到什么","它生成的幻灯片是否不需要手动编辑就能直接使用"]}}},{"id":"R_kgDOSD6ylA","slug":"datascale-ai__opentalking","name":"OpenTalking","owner":"datascale-ai","fullName":"datascale-ai/opentalking","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"face_and_voice_cloning","severity":"attention","evidence":"README 里有一节 \"C. Video Clone\"，包含 \"Realtime camera imitation\" 和 \"Uploaded video imitation\"，另外还有 \"Cloned voice driven\" 的视频生成功能，以及 .env.example 里小米 MiMo 的 \"voice clone profiles\"。同样的功能也能被用来让一个真实的人看起来说了他从没说过的话；路线图里还把 \"licensed voices, and synthetic-content labeling\" 列为尚未完成。","source":"ai","title":"能克隆人脸和声音"},{"code":"packages_from_third_party_mirror","severity":"attention","evidence":"pyproject.toml 把一个非官方的软件包源设成了默认源：\"[[tool.uv.index]] url = \\\"https://pypi.tuna.tsinghua.edu.cn/simple\\\" default = true\"，所以除非你自己改掉，安装时拉取的 Python 包来自中国的一个镜像站，而不是官方 PyPI。","source":"ai","title":"软件包来自第三方镜像源"},{"code":"service_open_to_local_network","severity":"info","evidence":".env.example 里自带 OPENTALKING_API_HOST=0.0.0.0 和 OPENTALKING_WEB_HOST=0.0.0.0，docker-compose.yml 还对外发布了 8000 和 5173 端口，所以只要能连到你的机器，任何人都能访问这个 API 和网页；文档里也没提到网页有登录验证。","source":"ai","title":"服务对局域网完全开放"},{"code":"imports_personal_chat_history","severity":"info","evidence":"README 的 \"Recent Progress\" 里 2026-06-25 那条写着：\"Added WeChat memory persona import\"，也就是说可以把你的微信私人聊天记录喂给这个虚拟形象当记忆。","source":"ai","title":"会导入个人聊天记录"},{"code":"very_new_project","severity":"info","evidence":"这个仓库创建于 2026-04-16，只有一个发布版本 v0.1.0（2026-06-15），没有任何平台的二进制文件；pyproject.toml 里写的版本号也是 0.1.0。","source":"ai","title":"项目非常新"}],"platforms":["windows","mac","linux"],"tools":["python","node","docker","gpu_nvidia","ffmpeg"],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":1410,"maxMB":5670,"basis":"heuristic"},"apiKeys":["Chat model (OpenAI-compatible, default DashScope) — OPENTALKING_LLM_API_KEY","DashScope speech recognition — OPENTALKING_STT_DASHSCOPE_API_KEY","Xiaomi MiMo speech recognition — OPENTALKING_STT_XIAOMI_API_KEY"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","audio","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":3344,"downloads":164,"downloadsWeek":5,"downloadsDays":5,"growth":35,"growthDays":6,"history":[{"date":"2026-11-05","stars":3315},{"date":"2026-11-06","stars":3319},{"date":"2026-11-07","stars":3324},{"date":"2026-11-08","stars":3328},{"date":"2026-11-09","stars":3332},{"date":"2026-11-10","stars":3344}],"firstPublishedAt":"2026-11-04T22:47:04.133Z","commitSha":"8c739a5a6f114daf71aeace832a668c3ad60f536","verifiedAt":"2026-11-04T22:39:39.877Z","lastCommitAt":"2026-09-04T01:49:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Run a talking AI avatar on your own machine: it listens, answers out loud and moves its lips in real time.","summary":"OpenTalking puts together the parts of a talking digital person: a web page where you pick an avatar and a voice, speech recognition, a chat model, text to speech and lip-sync video sent to the browser. It also makes videos from text or audio and can copy a face from a camera or an uploaded video. It is aimed at people comfortable with a terminal: you clone the repository, install Python, Node.js and FFmpeg, fill in a settings file and start the services. A test mode runs without a graphics card; real lip-sync needs an NVIDIA GPU and downloaded model files.","easeRationale":"Linux is the smoothest path and is still a D: there is no installer, you run shell scripts or Docker Compose, install Python, Node.js and FFmpeg, edit a settings file, and need an NVIDIA GPU plus downloaded weights for real lip-sync. Windows is described through WSL2, and on Mac the docs only cover an Apple Silicon CPU/MPS path for one model.","costNote":"The code is free (Apache-2.0). The example settings point the chat model and speech recognition at paid cloud services (DashScope, Atlas Cloud), and the project also documents local speech and lip-sync options, so the running cost depends on which ones you choose.","notVerified":["Whether real-time conversation actually keeps up on the listed graphics cards, or on Windows via WSL2 and on Apple Silicon.","How much disk the model weights really take after download (the checks only estimate 1.4-5.7 GB).","How much the cloud chat and speech services cost per hour of conversation.","What the third-party Compshare image linked in the README contains.","Whether the lip-sync and voice quality in the demo videos matches what you get on your own machine."]},"zh-Hans":{"summaryShort":"在自己的电脑上跑一个会说话的 AI 虚拟形象：它能听、能出声回答，还能实时动嘴唇。","summary":"OpenTalking 把一个会说话的数字人所需的各个部件拼在一起：一个网页用来选虚拟形象和声音，加上语音识别、聊天模型、文字转语音，以及发送到浏览器的唇形同步视频。它还能用文字或音频生成视频，也能从摄像头或上传的视频里复制一张脸。它是给熟悉命令行的人用的：你需要克隆仓库，安装 Python、Node.js 和 FFmpeg，填好配置文件，然后启动各个服务。测试模式不用显卡就能跑；真正的唇形同步需要一块 NVIDIA 显卡和下载好的模型文件。","easeRationale":"Linux 是最顺畅的路线，但依然是 D 级：没有安装程序，你要自己跑 shell 脚本或 Docker Compose，安装 Python、Node.js 和 FFmpeg，编辑配置文件，而且真正的唇形同步还需要一块 NVIDIA 显卡加上下载好的权重文件。Windows 这边是靠 WSL2 来实现的，而 Mac 上的文档只覆盖了某一个模型在 Apple Silicon 的 CPU/MPS 路线。","costNote":"代码本身是免费的（Apache-2.0 许可证）。示例配置把聊天模型和语音识别指向了付费的云服务（DashScope、Atlas Cloud），项目也写了本地语音和唇形同步的方案，所以实际花多少钱取决于你选哪一种。","notVerified":["实时对话在列出的那些显卡上，以及在 Windows 通过 WSL2、在 Apple Silicon 上，是否真的能跟得上节奏。","模型权重下载完之后实际占用多少磁盘空间（检查结果只是估算了 1.4 到 5.7 GB）。","云端聊天和语音服务每小时对话大概要花多少钱。","README 里链接的第三方 Compshare 镜像里到底装了什么。","演示视频里的唇形同步和语音质量，和你在自己机器上跑出来的效果是否一样。"]}}},{"id":"R_kgDORtI5eg","slug":"davidona__streamvault-iptv","name":"StreamVault","owner":"Davidona","fullName":"Davidona/StreamVault-IPTV","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{},"risk":"high","signals":[{"code":"sources_you_bring_yourself","severity":"attention","evidence":"这个应用本身不附带任何频道。README 的 \"Notes\" 部分写着:\"StreamVault is an IPTV client, not a content provider\" 和 \"Use only playlists, streams, and guide sources you are authorized to access.\"。可以添加的来源包括 `Xtream Codes`、`Stalker Portal`、`Jellyfin` 媒体服务器,以及来自网址或本地文件的 `M3U` 播放列表,还有兼容 Stremio 的插件,可以 \"Play direct or debrid streams\"。功能列表里还包括录制直播频道:\"Built-in DVR with scheduled recording, background capture, recording playback, and app-managed default storage\"。付费电视频道和电影都带有其所有者制定的内容授权规则,而这些规则在不同国家并不相同。","source":"ai","title":"播放源由你自己提供"},{"code":"backups_hold_your_passwords_in_readable_form","severity":"attention","evidence":"来自 README:\"New backup exports can include provider credentials so restored providers can sync without entering passwords again. Exported backup files contain provider passwords in readable form; keep them private.\"。这些文件还可以发送到 Google Drive(\"local Android TV backups or Google Drive snapshots\")。","source":"ai","title":"备份里的密码是可读的明文"},{"code":"installs_code_from_outside_any_store","severity":"attention","evidence":"没有应用商店的上架页面:你是从 GitHub Releases 或通过 Downloader 手动安装 `StreamVault.apk` 的,而且 \"The app can also detect and download newer releases in-app through GitHub Releases\",也就是说应用会自己替换自己。除此之外,\"StreamVault can be extended with companion Android APK plugins\",这些插件可以改变播放源、播放方式和 \"Cast URL rewrite\",也就是说会有其他人写的 APK 在它旁边运行。","source":"ai","title":"安装的代码不经过任何应用商店"},{"code":"license_forbids_commercial_use_and_removing_links","severity":"attention","evidence":"LICENSE 文件的标题是 \"StreamVault Source-Available License (Non-Commercial)\",在常见的自由软件权限之上又加了一些条件:禁止任何形式的商业使用,包括 \"Monetizing the Software via advertisements, sponsorships, or subscriptions\";应用内必须保留对 \"David Nashash\" 的署名;仓库链接和捐赠链接 \"https://ko-fi.com/davidona\" \"may not be removed, hidden, or replaced\";衍生作品必须使用同样的许可证并公开源代码;而且 \"Any violation of the terms of this license will result in the immediate termination of the rights granted\"。GitHub 把这个许可证标记为 NOASSERTION,README 也提到这个项目在 2026 年 4 月之前 \"was originally released without an explicit license\"。","source":"ai","title":"许可证禁止商业用途和移除链接"},{"code":"adult_channels_come_with_the_playlists","severity":"info","evidence":"应用自带 \"Strong parental controls with PIN-protected categories and automatic adult-category detection\",依靠的是 \"provider flags and category naming heuristics\",因为人们加载的播放列表通常都包含成人分类。兼容 Stremio 的插件默认是 \"open by default\",要靠一个单独的设置(Settings → Privacy)隐藏,而不是靠家长 PIN。","source":"ai","title":"成人频道会跟着播放列表一起出现"},{"code":"very_young_project_moving_fast","severity":"info","evidence":"这个仓库在 2026-03-21 创建,现在已经有 124 个发行版,最新的是 `v1.2.4`,同一天还发布了 `v1.2.4-beta.000097.01`,目前有 54 个未解决的 issue,距最后一次提交只有 3 天。这是一个人(`ko-fi.com/davidona`)的项目,几乎每天都在变化。","source":"ai","title":"一个变化很快的年轻项目"},{"code":"plays_streams_from_providers_you_bring","severity":"high","evidence":"项目描述:\"StreamVault is a TV-first free IPTV player for Android TV\"。README 写道它 \"supports `M3U` playlists, `Xtream Codes`, `Stalker Portal`, and `Jellyfin` providers, plus Stremio-compatible add-ons\",这些插件可以 \"Play direct or debrid streams\",播放器还能 \"Play supported adaptive and DRM-protected streams from M3U playlists\",支持 \"Provider archive or catch-up support when the source exposes replay streams\",以及 \"Built-in DVR with scheduled recording, background capture, recording playback\"。README 的 Notes 部分自己也写着 \"StreamVault is an IPTV client, not a content provider\" 和 \"Use only playlists, streams, and guide sources you are authorized to access\",仓库并没有说这些播放列表来自哪里、由谁运营。付费电视频道、电影和剧集都带有其所有者制定的内容授权规则,在很多国家,转播或录制这些内容需要版权方的授权。","source":"ai","title":"播放你自己带来的播放源内容"}],"platforms":[],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Your IPTV provider (Xtream Codes or Stalker Portal user and password, or an M3U address)","Trakt, to sync what you have watched (optional)","OpenSubtitles, for subtitle search (optional)","Google Drive, for backups (optional)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["video","hardware"],"alternativeTo":[{"slug":"tivimate","name":"TiviMate","confirmed":true},{"slug":"plex","name":"Plex","confirmed":true},{"slug":"emby","name":"Emby","confirmed":false}],"declaredAlternatives":[],"stars":1028,"downloads":153402,"downloadsWeek":4454,"downloadsDays":4,"growth":38,"growthDays":6,"history":[{"date":"2026-11-06","stars":1014},{"date":"2026-11-07","stars":1021},{"date":"2026-11-08","stars":1027},{"date":"2026-11-09","stars":1028},{"date":"2026-11-10","stars":1028}],"firstPublishedAt":"2026-11-05T22:53:18.525Z","commitSha":"bacaae7972f4ebe322b12e38ea57cfc31ff48552","verifiedAt":"2026-11-05T22:47:39.350Z","lastCommitAt":"2026-11-09T20:22:29Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Free IPTV player for Android TV: add your provider's playlist and watch live channels, films and series, with a guide and recording.","summary":"StreamVault is a television player for Android TV, made to be used with the remote control. You add your own source — an M3U playlist, Xtream Codes, Stalker Portal or a Jellyfin server — and it organises live channels, films and series, with a programme guide, favourites, search, split screen, recording of live channels, downloads for watching offline, parental control with a PIN, subtitles and 37 languages. It installs as an APK from GitHub or with the Downloader code on the TV. It does not come with any channels.","easeRationale":"Android TV is the only system it runs on, and there it is a B: you install the ready APK (`releases/latest/download/StreamVault.apk`, or the Downloader code `6443624` on the TV) and open it, but the app shows nothing until you type in the address and password of a playlist or server of your own. Phones and tablets also run it. Computers get no grade; there is no desktop version, and the FFmpeg part comes inside the app (the Android NDK is only for rebuilding it).","costNote":"The app asks for nothing and has a donation page (ko-fi.com/davidona). What you watch is separate: the IPTV subscription, Jellyfin server or add-ons are yours to bring and may be paid.","notVerified":["Whether the APK offered for download is built from this source code","What the app sends out, and to whom, while you watch","Which Android permissions it asks for, and how much space recordings and downloads take","Whether it works with each kind of provider, which depends on the server on the other side","What the companion APK plugins do, since they come from other people"]},"zh-Hans":{"summaryShort":"面向 Android TV 的免费 IPTV 播放器:添加你自己的播放源,观看直播频道、电影和剧集,还带节目单和录制功能。","summary":"StreamVault 是一款面向 Android TV 的电视播放器,专为配合遥控器使用而设计。你添加自己的播放源——M3U 播放列表、Xtream Codes、Stalker Portal 或 Jellyfin 服务器——它就会整理出直播频道、电影和剧集,附带节目单、收藏、搜索、分屏、直播频道录制、离线下载、带 PIN 的家长控制、字幕和 37 种语言。安装方式是从 GitHub 获取 APK,或者在电视上用 Downloader 代码安装。它本身不附带任何频道。","easeRationale":"Android TV 是它唯一支持运行的系统,在那里评级是 B:你安装现成的 APK(`releases/latest/download/StreamVault.apk`,或者在电视上用 Downloader 代码 `6443624`)然后打开,但在你输入自己播放列表或服务器的地址和密码之前,应用里什么都不会显示。手机和平板也能运行。电脑没有评级;没有桌面版,FFmpeg 部分已经内置在应用里(Android NDK 只是用来重新编译它的)。","costNote":"这个应用本身不收费,有一个捐赠页面(ko-fi.com/davidona)。你观看的内容是另一件事:IPTV 订阅、Jellyfin 服务器或插件都是你自己提供的,可能是付费的。","notVerified":["提供下载的 APK 是否真的是由这份源代码编译的","你观看的时候,应用向外发送了什么数据,发给了谁","它请求了哪些 Android 权限,录制和下载又占用多少空间","它是否能配合每一种播放源使用,这取决于另一端的服务器","配套的 APK 插件做了什么,因为它们来自其他人"]}}},{"id":"R_kgDOPGh7XA","slug":"davila7__claude-code-templates","name":"Claude Code Templates","owner":"davila7","fullName":"davila7/claude-code-templates","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"installs_other_people_s_code_into_your_assistant","severity":"attention","evidence":"这个目录收录了 \"about 1,900 components\",来自很多仓库(Attribution 部分列出了 K-Dense-AI、anthropics/skills、obra/superpowers、wshobson/agents、awesome-claude-code 等),其中好几类组件装上之后会自己行动:\"Hooks (60+) Automation triggers\"、\"Mods (45+) Plugins that change Claude Code's own interface and behavior\"、\"Loops (18) Autonomous workflows with a goal, an interval and a stop condition\" 和 \"MCPs (100+) External service integrations\"。README 的快速安装部分里,每个安装示例都带着 `--yes`,比如 \"npx claude-code-templates@latest --skill creative-design/frontend-design --yes\";唯一没带的是 \"Browse and install interactively\"(`npx claude-code-templates@latest`)。README 提到的文件预览其实是网站上的功能:\"search the catalog, preview every file a component writes\"。这个项目确实会审查贡献内容(\"checked by Claude Code's plugin validator and scanned by NVIDIA SkillSpector\"),但你安装的终究是别人写的指令,运行在你的编程助手里面。","source":"ai","title":"把别人写的代码安装进你的助手里"},{"code":"every_run_fetches_the_newest_version","severity":"attention","evidence":"所有文档里的命令都是 `npx claude-code-templates@latest`,所以每次运行时,npm 都会下载并执行最后发布的那个版本——`package.json` 里的版本号 1.32.1 已经领先于最新的发行标签 v1.29.7。没有任何版本是固定的。","source":"ai","title":"每次运行都会拉取最新版本"},{"code":"local_pages_show_your_conversations","severity":"info","evidence":"`--chats` 会打开一个网页,地址是 `http://localhost:9876`,用来 \"Read and search your Claude Code conversations\",`--analytics` 则打开 `http://localhost:3333`。README 说这两个都 \"listen on 127.0.0.1 only and answer only to `localhost`, because they serve your conversation history\",并且面向公网的 `--tunnel` 选项已经在 1.31.0 中移除。","source":"ai","title":"本地页面会显示你的对话内容"},{"code":"error_reports_and_counts_sent_out","severity":"info","evidence":"`.env.example` 解释说 \"the aitmpl-cli DSN is a public Sentry DSN (send-only, not a secret) and ships hardcoded as the default in the source\",上报功能是选择性开启的(`CCT_ERROR_REPORTING=true`),而 `CCT_NO_TRACKING`/`CCT_NO_ANALYTICS`/`CI` 始终优先;下载统计数据会发到一个 Supabase 数据库,`package.json` 的关键词里也包含 \"tracking\" 和 \"analytics\"。","source":"ai","title":"错误报告和统计数据会被发送出去"},{"code":"each_component_carries_its_own_license","severity":"info","evidence":"这个仓库本身是 MIT 许可,但 Attribution 部分混杂着不同的许可证——MIT 的组件,\"awesome-claude-code ... CC0 1.0 Universal\",\"awesome-claude-skills - Apache 2.0\",\"cocoindex-claude - Apache 2.0\"——并写明 \"Each of these resources retains its original license and attribution\"。你安装的内容并不全部遵循仓库自己的许可证。","source":"ai","title":"每个组件都有自己的许可证"},{"code":"the_tool_changes_shape_quickly","severity":"info","evidence":"1.31.0 版本一次性移除了十七个选项(`--plugins`、`--skills-manager`、`--agents`、`--teams`、`--studio`、`--sandbox`、`--tunnel`、`--clone-session` 等),同时有 467 个未关闭的 issue,并且在本次检查当天还有提交。那次发布之前写的教程和视频,已经和现在的命令对不上了。","source":"ai","title":"这个工具变化得很快"},{"code":"readme_sells_a_proxy_service","severity":"info","evidence":"README 顶部的赞助区块在宣传 Swiftproxy:\"Connect your AI-powered workflows to reliable residential proxies for web data collection, browser automation, and AI-driven research\",\"90M+ residential IPs across 220+ locations, rotating and static residential proxies, with HTTP(S) and SOCKS5 support\",还带着徽章 \"Get 10% off - code PROXY90\"。Z.AI 的徽章链接指向 `z.ai/subscribe?ic=8JVLJQFSKB`,是一个推荐链接。","source":"ai","title":"readme 在推销一个代理服务"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["None to install components","Serper, SerpAPI or RapidAPI, GitHub token, Google AI, Supabase and a Discord bot — only for the maintainer's own scripts in `.env.example`"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"tailwind-plus","name":"Tailwind Plus","confirmed":false}],"declaredAlternatives":[],"stars":33268,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":104,"growthDays":4,"history":[{"date":"2026-11-06","stars":33164},{"date":"2026-11-07","stars":33188},{"date":"2026-11-08","stars":33213},{"date":"2026-11-09","stars":33240},{"date":"2026-11-10","stars":33268}],"firstPublishedAt":"2026-11-05T22:53:18.537Z","commitSha":"3b2daa1a0177f5f0bd0c6b1402c574f1ecd9ca37","verifiedAt":"2026-11-05T22:48:21.628Z","lastCommitAt":"2026-11-10T17:14:05Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"One command installs ready-made skills, agents, commands and service connections into Claude Code, from a catalogue of about 1,900.","summary":"Claude Code Templates is a catalogue plus a command-line tool for Anthropic's Claude Code. With one line it installs ready-made pieces into your project or your assistant: skills, specialist agents, slash commands, MCP connections to services such as GitHub and PostgreSQL, settings, hooks that fire on their own, mods that change Claude Code's own interface, and whole project templates — about 1,900 of them, browsable at aitmpl.com. The same tool opens two local pages, one with statistics of your sessions and one to read your past conversations. Needs Node.js.","easeRationale":"A C on Windows, Mac and Linux: install Node.js and run one line, `npx claude-code-templates@latest`, which opens a menu, or add a flag such as `--skill creative-design/frontend-design --yes` to install a single piece. None of the 22 releases has a file to download, and it is only of use on a machine that already has Claude Code.","costNote":"MIT code and a free catalogue, paid for by sponsors: the README carries a Swiftproxy banner with the discount code PROXY90, a Z.AI subscription link with a referral code, GitHub Sponsors and Buy Me a Coffee.","notVerified":["What each of the roughly 1,900 components actually does once installed, since only a few are shown","Whether the counts in the README (890+ skills, 420+ agents) match the catalogue","What the MCP connections send to the outside, and which keys each one asks for","Whether the components still fit the current version of Claude Code","What the two local pages expose about your conversations while they are open"]},"zh-Hans":{"summaryShort":"一条命令就能把现成的技能、智能体、命令和服务连接安装进 Claude Code,目录里大约有 1900 个。","summary":"Claude Code Templates 是 Anthropic 旗下 Claude Code 的一个目录加命令行工具。只需一行命令,就能把现成的组件安装进你的项目或助手里:技能(skills)、专业智能体、斜杠命令、连接 GitHub 和 PostgreSQL 等服务的 MCP 连接、各种设置、会自动触发的钩子(hooks)、改变 Claude Code 自身界面的 mods,以及完整的项目模板——一共大约 1900 个,可以在 aitmpl.com 浏览。同一个工具还会打开两个本地页面,一个显示你会话的统计数据,另一个可以阅读你过去的对话。需要 Node.js。","easeRationale":"在 Windows、Mac 和 Linux 上评级是 C:安装 Node.js,运行一行命令 `npx claude-code-templates@latest` 就会打开一个菜单,或者加上像 `--skill creative-design/frontend-design --yes` 这样的参数来只安装一个组件。22 个发行版都没有可下载的文件,而且只有在已经装了 Claude Code 的机器上才有用。","costNote":"代码是 MIT 许可,目录免费,由赞助商付费支持:README 里有一个 Swiftproxy 的横幅,带着折扣码 PROXY90,还有一个带推荐码的 Z.AI 订阅链接、GitHub Sponsors 和 Buy Me a Coffee。","notVerified":["这大约 1900 个组件装上之后到底各自做什么,因为只展示了其中几个","README 里的数字(890+ 技能、420+ 智能体)是否和目录一致","MCP 连接会往外发送什么数据,每一个又要求什么密钥","这些组件是否还适配当前版本的 Claude Code","这两个本地页面在打开期间会暴露关于你对话的什么内容"]}}},{"id":"R_kgDOPcxPow","slug":"daymoondevelopment__post-for-me","name":"Post for Me","owner":"DayMoonDevelopment","fullName":"DayMoonDevelopment/post-for-me","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"license_statement_conflicts_with_readme","severity":"attention","evidence":"LICENSE 文件是 GNU Affero 通用公共许可证第三版,但 README 的许可部分写着\"This project is owned fully by Day Moon Development LLC — see the package.json files for details\",而且根目录的 package.json 里完全没有许可字段(它被标记为\"private\": true)。因此,单看这个仓库本身,并不清楚你被允许用这段代码做什么。","source":"ai","title":"许可声明与readme不一致"},{"code":"no_self_hosting_instructions","severity":"info","evidence":"README 只为每个文件夹写了开发命令,并提醒\"Do not run bun install at the repo root\"。没有环境变量示例,没有 Docker 文件,没有部署指南,也没有 GitHub 发布版本,所以运行自己的副本就得靠自己从源码摸索。","source":"ai","title":"没有自行部署说明"},{"code":"depends_on_several_outside_services","severity":"info","evidence":"技术栈部分写明:数据库是\"Supabase (PostgreSQL)\",身份验证是\"Unkey API key management\",后台任务运行在\"Trigger.dev v3\"上,使用\"bun run deploy # trigger.dev deploy (push to trigger.dev cloud)\",而且 API 还有一个 Stripe 相关的模式(\"bun run kanel:typegen # Regenerate Kysely types (Stripe schema)\")。","source":"ai","title":"依赖多个外部服务"},{"code":"very_early_stage","severity":"info","evidence":"根目录 package.json 写着\"version\": \"0.0.1\",没有任何发布版本,仓库自2025年8月以来有86个星标和35个未关闭的issue。","source":"ai","title":"处于非常早期的阶段"}],"platforms":["linux","mac"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Supabase","Trigger.dev","Unkey","Developer app credentials for each social network (X, Instagram, TikTok, LinkedIn and the others)"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","automation","development"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"later","name":"Later","confirmed":true}],"declaredAlternatives":[],"stars":86,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":86},{"date":"2026-11-08","stars":86},{"date":"2026-11-09","stars":86},{"date":"2026-11-10","stars":86}],"firstPublishedAt":"2026-11-06T22:59:23.778Z","commitSha":"729582a204b2eaf700adc1c96c12713d0fd5e237","verifiedAt":"2026-11-06T22:40:06.227Z","lastCommitAt":"2026-11-10T21:10:08Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Source code of a hosted service that schedules and publishes posts to X, Instagram, TikTok, LinkedIn and more.","summary":"Post for Me is the full source code of the commercial service at postforme.dev: an API plus a dashboard that schedules and publishes posts to several social networks (the project lists Twitter/X, Bluesky, Instagram, Facebook, Threads, TikTok, LinkedIn, Pinterest and YouTube). It is aimed at developers who want posting built into their own app. The repository holds four separate apps (API, background jobs, dashboard, marketing site) and the README only explains how to run each one in development, using Bun, Supabase, Trigger.dev and Unkey.","easeRationale":"This is an E on Linux and macOS, the systems the tooling targets: there is no installer, no release and no deployment guide, only developer commands per folder (\"cd api && bun install && bun run start:dev\", \"bun run supabase:start\", \"trigger.dev dev\"). You need Bun 1.3.3 or newer, a Supabase database, a Trigger.dev project and Unkey for API keys before anything posts anywhere.","costNote":"The code is under AGPL-3.0, so it costs nothing to download. The same product is sold as a hosted service at postforme.dev, and running it yourself means accounts on outside services (Supabase, Trigger.dev, Unkey) that charge above their free tiers.","notVerified":["Whether anyone outside the company can actually deploy and run it, since there is no setup guide","Which developer credentials each social network needs and whether posting works for all the networks listed in the topics","What the hosted service at postforme.dev charges","Real disk, database and hosting cost of the four apps (the 100-500 MB figure is an estimate, not measured)","What the code sends to the company's own servers when you run your copy"]},"zh-Hans":{"summaryShort":"一个托管服务的源代码,用于安排和发布帖子到 X、Instagram、TikTok、LinkedIn 等平台。","summary":"Post for Me 是 postforme.dev 这个商业服务的完整源代码:一个 API 加一个仪表盘,用于向多个社交网络安排和发布帖子(项目列出了 Twitter/X、Bluesky、Instagram、Facebook、Threads、TikTok、LinkedIn、Pinterest 和 YouTube)。它面向想把发布功能集成到自己应用里的开发者。仓库包含四个独立的应用(API、后台任务、仪表盘、营销网站),README 只说明了如何在开发环境中分别运行每一个,用到了 Bun、Supabase、Trigger.dev 和 Unkey。","easeRationale":"在 Linux 和 macOS 上(也就是这套工具瞄准的系统)它被评为 E 级:没有安装程序,没有发布版本,也没有部署指南,每个文件夹只有开发者命令(\"cd api && bun install && bun run start:dev\"、\"bun run supabase:start\"、\"trigger.dev dev\")。在任何地方发布任何内容之前,你都需要 Bun 1.3.3 或更新版本、一个 Supabase 数据库、一个 Trigger.dev 项目,以及用于 API 密钥的 Unkey。","costNote":"代码采用 AGPL-3.0 许可,下载不花钱。同一个产品也在 postforme.dev 上作为托管服务出售,自己运行它需要外部服务(Supabase、Trigger.dev、Unkey)的账号,超出免费额度后这些服务会收费。","notVerified":["公司之外的人是否真的能部署并运行它,因为没有安装指南","每个社交网络各自需要什么开发者凭证,以及发布功能是否对主题中列出的所有网络都有效","postforme.dev 上的托管服务收费是多少","四个应用实际占用的磁盘、数据库和托管成本(100-500 MB 这个数字只是估计,并未实际测量)","当你运行自己的副本时,代码会向公司自己的服务器发送什么内容"]}}},{"id":"R_kgDOONWTjQ","slug":"dayuanjiang__next-ai-draw-io","name":"Next AI Draw.io","owner":"DayuanJiang","fullName":"DayuanJiang/next-ai-draw-io","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"B","mac":"B","linux":"B","web":"A"},"risk":"attention","signals":[{"code":"your_diagrams_and_prompts_leave_your_computer","severity":"attention","evidence":"README写道：\"Next AI Draw.io sends the current canvas, including your hand edits, with every message\"。在演示站点上，这些内容会发送到作者的服务器，再转发给选定的模型服务商，所以私密的图表不应该在那里画。","source":"ai","title":"你的图表和提示会离开你的电脑"},{"code":"tracing_libraries_bundled","severity":"attention","evidence":"package.json的依赖里包含\"@langfuse/client\"、\"@langfuse/otel\"、\"@langfuse/tracing\"、\"@opentelemetry/exporter-trace-otlp-http\"和\"@next/third-parties\"；README中的管理面板负责管理\"observability and quotas\"。因此运营该站点的人可以打开对提示内容和使用情况的追踪。","source":"ai","title":"内置了追踪库"},{"code":"mcp_server_downloaded_fresh_at_every_start","severity":"info","evidence":"README中的MCP设置使用了\"npx\"和\"@next-ai-drawio/mcp-server@latest\"。@latest标签意味着你的AI助手每次启动时都会从npm仓库获取当前最新的版本，而不是你检查过的某个固定版本。","source":"ai","title":"每次启动都会重新下载MCP服务器"},{"code":"large_download_and_disk_use","severity":"info","evidence":"v0.5.1发布版本的文件：\"Next.AI.Draw.io.Setup.0.5.1.exe (windows, 442 MB)\"、\".dmg (mac, 287 MB)\"、\"amd64.deb (linux, 221 MB)\"。这款桌面应用打包了一整套浏览器引擎和一份draw.io编辑器。","source":"ai","title":"下载包很大，占用磁盘空间也多"},{"code":"small_models_produce_broken_diagrams","severity":"info","evidence":"README中\"Which model\"一节写道：\"the task is to produce long text in a strict format (draw.io XML), so pick a capable model; small models tend to produce broken diagrams.\"最容易让人失望的，正是免费本地模型这条路线。","source":"ai","title":"小模型容易画出错乱的图表"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":442,"maxMB":1325,"basis":"installer"},"apiKeys":["An AI model provider of your choice (25 are supported: OpenAI, Anthropic, Google, AWS Bedrock, OpenRouter, DeepSeek and others)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["design_cad","productivity"],"alternativeTo":[{"slug":"lucidchart","name":"Lucidchart","confirmed":true},{"slug":"microsoft-visio","name":"Visio","confirmed":true},{"slug":"eraser","name":"Eraser","confirmed":true},{"slug":"miro","name":"Miro","confirmed":false}],"declaredAlternatives":[],"stars":36865,"downloads":162743,"downloadsWeek":3130,"downloadsDays":6,"growth":216,"growthDays":7,"history":[{"date":"2026-11-04","stars":36683},{"date":"2026-11-05","stars":36713},{"date":"2026-11-06","stars":36741},{"date":"2026-11-07","stars":36774},{"date":"2026-11-08","stars":36793},{"date":"2026-11-09","stars":36826},{"date":"2026-11-10","stars":36865}],"firstPublishedAt":"2026-11-04T01:36:50.337Z","commitSha":"c9dc5f5cc62e7e5a327a62b39027d80033e1548c","verifiedAt":"2026-11-04T01:22:26.921Z","lastCommitAt":"2026-11-07T07:12:19Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Describe a diagram in one sentence and an AI draws it on a real draw.io canvas that you can keep editing by hand.","summary":"Next AI Draw.io turns a sentence into a diagram: flowcharts, cloud architecture, ER and BPMN diagrams, with ready icon sets for AWS, Azure, Google Cloud and Kubernetes. You can upload a screenshot, a PDF or notes and ask for a diagram of them, select shapes and ask for changes, and every AI change becomes a version you can compare, restore or undo. The result is an ordinary draw.io file you can export as .drawio, .png or .svg. Use it on the author's website, install it on Windows, Mac or Linux, or plug it into Claude Code, Cursor or VS Code.","easeRationale":"On the web it is the easiest: open next-ai-drawio.jiang.jp and start drawing, with nothing to install, though the free demo has a daily limit. On Windows, Mac and Linux there are ready installers on the releases page (a 442 MB setup for Windows, .dmg for Mac, .deb, .rpm and AppImage for Linux), and then one extra step: open settings and paste the API key of an AI provider.","costNote":"The program is free (Apache-2.0). The AI drawing needs a model provider: the author's demo site has a daily limit, and beyond that you enter your own API key and pay that provider per use. A free local model through Ollama or LM Studio is also supported.","notVerified":["Whether the API key you type really stays in your browser and is never stored on the server, as the README says.","How much a provider charges in practice: each diagram sends the whole canvas again, so a long editing session can cost more than expected.","What the daily limit on the demo site is, and what happens when you reach it.","Whether the desktop app works offline once a local model is configured.","Whether the diagrams come out correct enough to use without hand fixing; that depends on the model you pick."]},"zh-Hans":{"summaryShort":"用一句话描述图表，AI就会把它画在真正的draw.io画布上，你可以继续手动编辑。","summary":"Next AI Draw.io能把一句话变成图表：流程图、云架构图、ER图和BPMN图，还有AWS、Azure、Google Cloud和Kubernetes的现成图标集。你可以上传一张截图、一份PDF或笔记，让它据此画出图表，也可以选中图形并要求修改，AI做出的每一次改动都会变成一个可以比较、还原或撤销的版本。最终得到的是一个普通的draw.io文件，可以导出为.drawio、.png或.svg。你可以在作者的网站上使用，也可以安装到Windows、Mac或Linux，或者接入Claude Code、Cursor或VS Code。","easeRationale":"在网页上使用最简单：打开next-ai-drawio.jiang.jp就能直接开始画图，不需要安装任何东西，不过免费演示版有每日次数限制。在Windows、Mac和Linux上，发布页面提供了现成的安装程序（Windows版安装包442 MB，Mac版是.dmg，Linux版有.deb、.rpm和AppImage），之后还要多一步：打开设置并粘贴某个AI服务商的API密钥。","costNote":"这款程序是免费的（Apache-2.0）。AI画图需要一个模型服务商：作者的演示站点有每日次数限制，超出之后你需要填入自己的API密钥，并按使用量向该服务商付费。也支持通过Ollama或LM Studio使用免费的本地模型。","notVerified":["你输入的API密钥是否真的只留在浏览器里，从不保存到服务器上，就像README里说的那样。","服务商在实际使用中会收多少费用：每次画图都会重新发送整个画布，所以一次较长的编辑过程可能比预期花费更多。","演示站点的每日限额具体是多少，以及达到限额后会发生什么。","配置好本地模型之后，桌面应用是否能离线使用。","画出来的图表是否足够准确，可以不用手动修正就使用；这取决于你选择的模型。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0NDY2MjY2OQ==","slug":"dbeaver__dbeaver","name":"DBeaver","owner":"dbeaver","fullName":"dbeaver/dbeaver","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"ai_chat_shares_database_structure","severity":"info","evidence":"README.md (AI integration)：\"We use smart chat context which provides LLMs details about database structure, SQL dialect, etc.\" 如果你开启 AI 聊天，关于你数据库的细节就会发送给你所配置的 AI 服务商。","source":"ai","title":"AI 聊天会共享数据库结构"},{"code":"features_reserved_for_paid_version","severity":"info","evidence":"README.md 列出了一长串只在 \"PRO versions\" 中才有的驱动（MongoDB、Cassandra、Redis、DynamoDB、Neo4j、ODBC，以及把 CSV/JSON/Parquet/XLSX 当作数据库），并说 \"Pro versions provide additional AI tools + native support of Anthropic/Grok/Azure/Bedrock/Gemini/Ollama providers.\"","source":"ai","title":"部分功能只留给付费版"},{"code":"large_open_issue_backlog","severity":"info","evidence":"3,341 个未解决的 issue，对应 52,266 个星标，该仓库自 2015-10-21 起运行，共有 259 个发布版本（2026-11-01 发布了 26.2.4）。它有规律的发布节奏，开发很活跃，但同时有很多问题一直开着，README 还请用户用 👍 投票来推动某个工单。","source":"ai","title":"积压的未解决 issue 很多"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":110,"maxMB":329,"basis":"installer"},"apiKeys":["OpenAI or an OpenAI-compatible endpoint (optional AI chat)","GitHub Copilot (optional AI chat)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","productivity"],"alternativeTo":[{"slug":"datagrip","name":"DataGrip","confirmed":true},{"slug":"navicat","name":"Navicat","confirmed":true},{"slug":"dbeaver-pro","name":"PRO","confirmed":true},{"slug":"tableplus","name":"TablePlus","confirmed":true}],"declaredAlternatives":[],"stars":52388,"downloads":47959009,"downloadsWeek":392104,"downloadsDays":7,"growth":95,"growthDays":7,"history":[{"date":"2026-10-26","stars":52220},{"date":"2026-10-27","stars":52224},{"date":"2026-10-28","stars":52233},{"date":"2026-10-29","stars":52244},{"date":"2026-10-30","stars":52251},{"date":"2026-10-31","stars":52255},{"date":"2026-11-01","stars":52266},{"date":"2026-11-02","stars":52276},{"date":"2026-11-03","stars":52293},{"date":"2026-11-04","stars":52297},{"date":"2026-11-05","stars":52317},{"date":"2026-11-06","stars":52327},{"date":"2026-11-07","stars":52332},{"date":"2026-11-08","stars":52354},{"date":"2026-11-09","stars":52370},{"date":"2026-11-10","stars":52388}],"firstPublishedAt":"2026-10-27T01:10:12.620Z","commitSha":"6d368ff8aedd9a9be85c7adbb5d8819674866e67","verifiedAt":"2026-11-02T02:57:32.823Z","lastCommitAt":"2026-11-10T19:11:01Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"dbeaver-ce-26.2.4-windows-x86_64.exe","sizeBytes":115077816,"url":"https://github.com/dbeaver/dbeaver/releases/download/26.2.4/dbeaver-ce-26.2.4-windows-x86_64.exe","releaseUrl":"https://github.com/dbeaver/dbeaver/releases/tag/26.2.4"},{"os":"mac","file":"dbeaver-ce-26.2.4-macos-aarch64.dmg","sizeBytes":124646758,"url":"https://github.com/dbeaver/dbeaver/releases/download/26.2.4/dbeaver-ce-26.2.4-macos-aarch64.dmg","releaseUrl":"https://github.com/dbeaver/dbeaver/releases/tag/26.2.4"},{"os":"linux","file":"dbeaver-ce-26.2.4-linux-x86_64.deb","sizeBytes":129167416,"url":"https://github.com/dbeaver/dbeaver/releases/download/26.2.4/dbeaver-ce-26.2.4-linux-x86_64.deb","releaseUrl":"https://github.com/dbeaver/dbeaver/releases/tag/26.2.4"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Open, browse and edit almost any database from one desktop app, run SQL queries and draw diagrams of your tables.","summary":"DBeaver is a desktop program for working with databases. You connect to MySQL, PostgreSQL, SQLite, Oracle, SQL Server, Snowflake and more than 100 other kinds, then browse and edit the data in a spreadsheet-like grid, write and run SQL, import and export data, and draw diagrams of how the tables relate. There is also an AI chat that can write queries for you if you add an OpenAI or Copilot key. It is made for developers, database administrators and analysts. Installers for Windows, Mac and Linux include Java, so nothing else is needed.","easeRationale":"Every platform has a ready installer in the releases (dbeaver-ce-26.2.4-windows-x86_64.exe, -macos-aarch64.dmg, and .deb/.rpm for Linux, with ARM builds too) and the README says OpenJDK 25 is included in all distributions, so you install it and click the app icon.","costNote":"The Community edition in this repository is free. Separate paid PRO versions add NoSQL databases, files as databases, more AI providers and official support.","notVerified":["Exactly what database details the AI chat sends to the AI provider.","Whether database drivers are downloaded from the internet the first time you connect to a new type of database.","How it performs with very large databases on your computer.","Which of the features you need are only in the paid PRO versions."]},"zh-Hans":{"summaryShort":"用一个桌面应用打开、浏览和编辑几乎任何数据库，运行 SQL 查询，并绘制表的关系图。","summary":"DBeaver 是用于操作数据库的桌面程序。你可以连接 MySQL、PostgreSQL、SQLite、Oracle、SQL Server、Snowflake 以及 100 多种其他类型的数据库，然后在类似电子表格的网格中浏览和编辑数据，编写并运行 SQL，导入导出数据，并绘制表之间关系的图表。如果你添加 OpenAI 或 Copilot 的密钥，还有一个可以替你写查询的 AI 聊天。它面向开发者、数据库管理员和分析师。适用于 Windows、Mac 和 Linux 的安装程序都自带 Java，所以不需要再装别的东西。","easeRationale":"每个平台在发布页都有现成的安装程序（dbeaver-ce-26.2.4-windows-x86_64.exe、-macos-aarch64.dmg，以及 Linux 的 .deb/.rpm，也有 ARM 版本），README 说所有发行版都包含 OpenJDK 25，所以安装后点击应用图标即可。","costNote":"本仓库中的社区版（Community）是免费的。另有付费的 PRO 版本，增加了 NoSQL 数据库、把文件当作数据库、更多 AI 服务商和官方支持。","notVerified":["AI 聊天究竟会把哪些数据库细节发送给 AI 服务商。","首次连接一种新类型的数据库时，数据库驱动是否会从互联网下载。","在你的电脑上处理非常大的数据库时表现如何。","你需要的功能中，哪些只在付费的 PRO 版本里才有。"]}}},{"id":"R_kgDORR71uQ","slug":"ddalcu__mlx-serve","name":"MLX Serve","owner":"ddalcu","fullName":"ddalcu/mlx-serve","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"D","mac":"B","linux":"D"},"risk":"low","signals":[{"code":"license_file_mixes_mit_and_apache","severity":"info","evidence":"LICENSE 文件开头是 MIT 协议文本(\"Copyright (c) 2026 David Dalcu\"),接着写道:\"This distribution also includes third-party software under its own licenses, including the Apache License 2.0, the BSD 3-Clause License, and further MIT-licensed components\"。正是这种混合导致许可证检测工具意见不一致(GitHub 没有为该仓库给出单一许可证);不过这些许可证都是宽松型的。","source":"ai","title":"许可证文件混合了 mit 和 apache"},{"code":"app_updates_itself","severity":"info","evidence":"安装部分写道,应用 \"suggests a model that fits your Mac, and updates itself\",也就是说安装之后它会自行拉取新代码,而不经过 GitHub 的发行页面。","source":"ai","title":"应用会自我更新"},{"code":"large_model_downloads_at_run_time","severity":"info","evidence":"应用和命令会在你需要时下载模型(\"mlx-serve run gemma4 — downloads Gemma 4 E4B\"),推荐表里的模型最大到 125B 和 744B 参数,\"streamed from the SSD\"。实际占用的磁盘空间完全取决于你拉取了哪些模型,远超安装包本身的 228 MB 到 683 MB。","source":"ai","title":"运行时会下载大体积模型"},{"code":"homebrew_tap_asks_for_trust","severity":"info","evidence":"通过 Homebrew 安装的路径是 \"brew tap ddalcu/mlx-serve https://github.com/ddalcu/mlx-serve\",接着是 \"brew trust ddalcu/mlx-serve\",这会让 Homebrew 运行这位作者自己 tap 里的代码,而不是官方 tap 的代码。","source":"ai","title":"homebrew tap 要求你给予信任"}],"platforms":["mac","windows","linux"],"tools":[],"buildTools":[],"disk":{"minMB":228,"maxMB":683,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["local_ai","development","image"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"midjourney","name":"Midjourney","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"otter-ai","name":"Otter.ai","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":2420,"downloads":68383,"downloadsWeek":1506,"downloadsDays":2,"growth":53,"growthDays":6,"history":[{"date":"2026-11-08","stars":2405},{"date":"2026-11-09","stars":2414},{"date":"2026-11-10","stars":2420}],"firstPublishedAt":"2026-11-07T22:56:16.729Z","commitSha":"c74444fc2302cce004d147f62c0cd195102190b2","verifiedAt":"2026-11-07T22:39:30.362Z","lastCommitAt":"2026-11-10T06:54:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs open AI models on your own Mac: chat, coding helpers, pictures, video, music and voice, with no account.","summary":"A Mac app plus a local server that runs open AI models on your own machine. You can chat with a model, generate pictures, video, music, speech and 3D, transcribe calls and write meeting notes, and point coding tools like Claude Code or Codex at the local server instead of a paid online service. The Mac app has an installer and suggests a first model for your amount of memory; it needs macOS 26.2 or later on Apple Silicon. Linux and Windows downloads exist but are command-line only and expect an NVIDIA graphics card.","easeRationale":"On a Mac you download MLX-Serve.dmg, drag it to Applications and open it, and the app then suggests and downloads a model that fits your memory, which is the one extra step that makes it a B; it needs macOS 26.2 or later on Apple Silicon. Linux and Windows are harder (D): the downloads are command-line servers built for NVIDIA cards (CUDA 13), and the README says Windows needs a build script.","costNote":"The README says nothing is paywalled and asks only for optional support through GitHub Sponsors or Buy Me a Coffee. The separate iPhone version is listed on the App Store.","notVerified":["Whether it installs and runs on each Mac, and whether the macOS 26.2 and Apple Silicon requirement rules out your machine.","The speed table in the README, measured by the author on an M5 Ultra with 256 GB.","Whether very large models really work by reading from the SSD, and how usable they feel.","Total disk space and memory after downloading the models you want.","What the self-update and the app contact over the network, since the README says it can run offline but does not list the addresses.","Whether the Windows and Linux builds work out of the box on a given NVIDIA card."]},"zh-Hans":{"summaryShort":"在自己的 Mac 上运行开放 AI 模型:聊天、编程助手、图片、视频、音乐和语音,无需账号。","summary":"一款 Mac 应用加本地服务器,在你自己的电脑上运行开放的 AI 模型。你可以和模型聊天,生成图片、视频、音乐、语音和 3D 内容,转录通话并写会议纪要,还可以把 Claude Code 或 Codex 这类编程工具指向这个本地服务器,而不是用付费的在线服务。Mac 应用自带安装程序,会根据你的内存大小推荐第一个模型;在 Apple 芯片上需要 macOS 26.2 或更高版本。Linux 和 Windows 也有下载版本,但只是命令行形式,并且需要 NVIDIA 显卡。","easeRationale":"在 Mac 上,你下载 MLX-Serve.dmg,把它拖进 Applications 并打开,之后应用会根据你的内存推荐并下载一个合适的模型,这是让它归为 B 级的唯一额外步骤;在 Apple 芯片上需要 macOS 26.2 或更高版本。Linux 和 Windows 更麻烦(D 级):下载的是为 NVIDIA 显卡(CUDA 13)打造的命令行服务器,README 还说 Windows 需要一个构建脚本。","costNote":"README 说没有任何内容需要付费解锁,只是希望通过 GitHub Sponsors 或 Buy Me a Coffee 获得可选的支持。另有一个 iPhone 版本在 App Store 上架。","notVerified":["它是否能在各款 Mac 上安装和运行,以及 macOS 26.2 和 Apple 芯片的要求是否会把你的机器排除在外。","README 里的速度对比表,是作者在一台 256 GB 内存的 M5 Ultra 上测得的。","超大模型是否真的能靠从 SSD 读取来正常工作,用起来体验如何。","下载完你想要的模型之后,总共占用多少磁盘空间和内存。","自动更新和这个应用在联网时具体联系了什么地址,因为 README 说它可以离线运行,但没有列出具体地址。","Windows 和 Linux 版本在某款 NVIDIA 显卡上是否真的开箱即用。"]}}},{"id":"R_kgDOR-gPKw","slug":"debpalash__voicestudio","name":"VoiceStudio","owner":"debpalash","fullName":"debpalash/VoiceStudio","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"voice_cloning_can_impersonate_real_people","severity":"attention","evidence":"仓库简介提供 \"voice cloning, voice design, video dubbing\"，README 的 \"License & responsible use\" 部分写道 \"Clone a voice only with its owner's permission.\" 克隆出的声音可能被用来在未经本人同意的情况下模仿真人，一些国家对此有限制。","source":"ai","title":"声音克隆可能被用来冒充真人"},{"code":"macos_gatekeeper_blocks_the_app","severity":"info","evidence":"README 中的说明：\"On macOS, 'VoiceStudio is damaged and can't be opened' means Gatekeeper blocked the app, not that the download is broken\"，并指向一个 \"Gatekeeper fix\" 页面。","source":"ai","title":"macOS Gatekeeper 会拦截该应用"},{"code":"model_weights_and_output_under_separate_terms","severity":"info","evidence":"README：\"Model weights, voices and generated output have their own licences, which the Model Manager shows for each model\"，而且该应用的许可证并不涵盖它们。","source":"ai","title":"模型权重与生成内容适用各自的条款"},{"code":"optional_analytics_off_by_default","severity":"info","evidence":"pyproject.toml 中关于 posthog 依赖的注释：\"Opt-in product analytics (core/analytics.py). Inert unless the build ships a POSTHOG_PROJECT_TOKEN *and* the user opts in — default OFF.\"","source":"ai","title":"可选的使用统计，默认关闭"},{"code":"dependency_installed_from_a_direct_url","severity":"info","evidence":"pyproject.toml：\"kittentts @ https://github.com/KittenML/KittenTTS/releases/download/0.8.1/kittentts-0.8.1-py3-none-any.whl\"——有一个语音引擎是直接从 GitHub 发布文件拉取的，而不是从常规的包索引。","source":"ai","title":"依赖从直接网址安装"},{"code":"bundles_a_video_downloader","severity":"info","evidence":"pyproject.toml 将 \"yt-dlp>=2026.7.4\" 列为依赖，因此该应用可以从网站拉取视频来进行配音或转写。是否允许这样做，取决于具体网站和当地法律。","source":"ai","title":"内置视频下载器"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","node","gpu_nvidia","ffmpeg"],"disk":{"minMB":212,"maxMB":636,"basis":"installer"},"apiKeys":["An OpenAI-compatible AI model, for cinematic dubbing and AI translation","Hugging Face token, for downloading some models"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["audio","video","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true},{"slug":"otter-ai","name":"Otter.ai","confirmed":true}],"declaredAlternatives":[],"stars":107142,"downloads":2610889,"downloadsWeek":449009,"downloadsDays":7,"growth":7098,"growthDays":7,"history":[{"date":"2026-10-27","stars":91348},{"date":"2026-10-28","stars":93015},{"date":"2026-10-29","stars":94328},{"date":"2026-10-30","stars":95755},{"date":"2026-10-31","stars":96872},{"date":"2026-11-01","stars":98004},{"date":"2026-11-02","stars":99072},{"date":"2026-11-03","stars":100044},{"date":"2026-11-04","stars":100899},{"date":"2026-11-05","stars":101814},{"date":"2026-11-06","stars":102769},{"date":"2026-11-07","stars":103742},{"date":"2026-11-08","stars":104819},{"date":"2026-11-09","stars":106033},{"date":"2026-11-10","stars":107142}],"firstPublishedAt":"2026-10-27T18:40:27.852Z","commitSha":"ac821cb60c1349ac8ea630c10319a6300a68b67d","verifiedAt":"2026-10-29T01:40:48.044Z","lastCommitAt":"2026-11-10T20:33:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop app that clones voices, dubs videos, dictates, transcribes and makes audiobooks on your own computer.","summary":"A desktop studio for AI speech that runs on your own machine. It clones a voice from a recording, designs new voices, dubs videos with timed speech, dictates with a floating widget, transcribes audio and builds audiobooks, in many languages. The models run locally; an NVIDIA GPU or Apple Silicon speeds it up, and the README says it also works on the CPU, just slower. There are installers for Windows, macOS and Linux, plus preview apps for Android and iPhone. The first time you use a feature the app asks before downloading the model it needs.","easeRationale":"Each system has a ready installer in the latest release (VoiceStudio-0.5.7-win-x64.exe, the .dmg for Apple Silicon, .AppImage/.deb/snap on Linux); after installing you still download each speech model inside the app the first time you use it. The README marks Intel Macs as \"App UI only, connects to a remote backend\", so an older Mac is harder.","costNote":"AGPL-3.0 and free to use. pyproject.toml says a separate commercial licence is sold for closed-source use (contact hi@voicestudio.sh), and model weights, voices and generated output have their own terms.","notVerified":["How much disk space the speech models add on top of the installer, since they download on first use","Whether the app runs at a usable speed on a PC with no dedicated graphics card","How good the voice cloning, dubbing and transcription quality is in practice","Whether the Windows-on-ARM, Intel Mac and phone builds work, since the README marks them experimental, limited or preview","Whether anything really leaves your computer, beyond the model downloads the app asks about"]},"zh-Hans":{"summaryShort":"一款桌面应用,可在你自己的电脑上克隆声音、为视频配音、语音输入、转录和制作有声书。","summary":"这是一个在你自己电脑上运行的 AI 语音桌面工作室。它可以从一段录音克隆声音、设计新的声音、为视频配上同步的配音、用悬浮小组件进行语音输入、转录音频,并制作多语言有声书。模型都在本地运行;NVIDIA 显卡或 Apple Silicon 芯片可以加速,README 说在 CPU 上也能跑,只是速度更慢。该应用为 Windows、macOS 和 Linux 提供安装程序,另外还有 Android 和 iPhone 的预览版应用。首次使用某个功能时,应用会先询问,再下载所需的模型。","easeRationale":"最新版本为每个系统都提供了现成的安装程序(Windows 的 VoiceStudio-0.5.7-win-x64.exe、Apple Silicon 的 .dmg、Linux 的 .AppImage/.deb/snap);安装完成后,你仍需在首次使用时在应用内下载每个语音模型。README 把 Intel 版 Mac 标记为“仅应用界面,连接远程后端”,所以较旧的 Mac 使用起来更困难。","costNote":"AGPL-3.0 许可,使用免费。pyproject.toml 说明另有商业许可出售,供闭源用途使用(联系 hi@voicestudio.sh),模型权重、声音和生成结果各有自己的使用条款。","notVerified":["语音模型在安装程序之外会占用多少磁盘空间,因为它们是首次使用时才下载的","在没有独立显卡的 PC 上,应用能否以可用的速度运行","实际使用中,声音克隆、配音和转录的质量如何","Windows on ARM、Intel 版 Mac 和手机版本是否能用,因为 README 把它们标记为实验性、受限或预览版","除了应用主动询问的模型下载之外,是否真的有数据离开你的电脑"]}}},{"id":"R_kgDOT3T1gw","slug":"deepseek-ai__deepseek-harness","name":"DeepSeek Harness","owner":"deepseek-ai","fullName":"deepseek-ai/deepseek-harness","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"76 天内获得 264824 颗星（约每天 3485 颗）","source":"check","title":"星标增长速度异常"},{"code":"project_asks_you_to_read_a_safety_notice_first","severity":"attention","evidence":"README 的 Developer preview 部分：'DeepSeek Harness is in developer preview and iterating rapidly. THERE WILL BE COMPATIBILITY-BREAKING CHANGES. Review the safety notice (SAFETY.md) before running the project.' 该仓库提供三种语言版本的 SAFETY.md。","source":"ai","title":"项目要求先阅读一份注意事项"},{"code":"pre_release_versions_only","severity":"attention","evidence":"最新的三个版本是 dsh-v0.2.2-alpha.2、dsh-v0.2.1-rc.4 和 dsh-v0.2.1-rc.3（2026-10-23 至 2026-10-28），package.json 中的版本号为 '0.2.2-alpha.2'。目前没有稳定的 1.0 版本。","source":"ai","title":"只有预发布版本"},{"code":"an_agent_that_acts_on_your_computer","severity":"attention","evidence":"该项目自称是 'an open-source agent harness'，采用 'everything-is-a-plugin' 架构，运行命令会启动一个服务器，并 'opens it in the default browser for a local launch'，地址为 http://127.0.0.1:3080。代理及其插件所做的一切，都是以你的用户身份访问你的文件来完成的。","source":"ai","title":"会在你电脑上执行操作的代理"},{"code":"plugins_come_from_anyone","severity":"attention","evidence":"README 的社区部分：'Add the `dsh-plugin` topic to your plugin repository for discoverability.' 各项功能预计以第三方插件的形式提供，通过 GitHub topic 查找，DeepSeek 不会审核这些插件。","source":"ai","title":"插件可能来自任何人"},{"code":"no_issue_tracker_in_use","severity":"info","evidence":"项目信息显示，这个拥有 264824 颗星和 32075 个 fork 的仓库没有未解决的 issue；README 把所有反馈引导到别处：'Submit feedback or bug reports through GitHub Discussions.' 因此很难查阅公开的错误报告。","source":"ai","title":"没有在使用 issue 跟踪"},{"code":"needs_a_very_recent_node_version","severity":"info","evidence":"package.json 设置了 'engines': { 'node': '>=24.9.0' } 和 'packageManager': 'pnpm@11.28.5'，因此较旧的 Node.js 无法运行它。","source":"ai","title":"需要非常新的 Node 版本"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","development"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true},{"slug":"cursor","name":"Cursor","confirmed":true}],"declaredAlternatives":[],"stars":278144,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":7064,"growthDays":7,"history":[{"date":"2026-10-25","stars":261471},{"date":"2026-10-26","stars":262584},{"date":"2026-10-27","stars":263688},{"date":"2026-10-28","stars":264805},{"date":"2026-10-29","stars":265802},{"date":"2026-10-30","stars":266735},{"date":"2026-10-31","stars":267448},{"date":"2026-11-01","stars":268435},{"date":"2026-11-02","stars":269615},{"date":"2026-11-03","stars":271080},{"date":"2026-11-04","stars":272433},{"date":"2026-11-05","stars":273719},{"date":"2026-11-06","stars":274848},{"date":"2026-11-07","stars":275694},{"date":"2026-11-08","stars":276413},{"date":"2026-11-09","stars":277313},{"date":"2026-11-10","stars":278144}],"firstPublishedAt":"2026-10-25T23:19:00.166Z","commitSha":"f3803c96974ad4a94195fb97e5c637e9c2a3afe3","verifiedAt":"2026-10-29T01:43:53.019Z","lastCommitAt":"2026-11-04T08:40:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"DeepSeek's own program for running an AI agent on your computer, opened in a browser page and extended through plugins.","summary":"DeepSeek Harness (`dsh`) is DeepSeek AI's open-source harness for running an AI agent on your own machine. Everything in it is a plugin, so other people can add capabilities. You install Node.js 24.9 or newer and run one command, which starts a local web interface at 127.0.0.1:3080 and opens your browser. It is aimed at developers: the project calls itself a developer preview, warns in capitals that there will be compatibility-breaking changes, and asks you to read its safety notice before running it.","easeRationale":"Same on all three and there is no installer: you install Node.js (package.json requires version 24.9.0 or newer, which is newer than most systems ship) and run `npx @deepseek-ai/dsh web` in a terminal, which opens the interface in your browser. None of the 34 releases carries a program file, so the desktop build must be compiled from source.","costNote":"The code is free under MIT. It is a harness for AI models, and model use is normally paid; the repository content does not say which provider or key you must configure.","notVerified":["Which model provider and API key you must configure, since the README does not say","What the agent and its plugins are allowed to do on your files and network","What the safety notice asks of you, since only its file name is in this package","Whether the npx command works on Windows as described","Real disk use after the install and the built artifacts","How much paid model use a normal session costs"]},"zh-Hans":{"summaryShort":"DeepSeek 自家用于在电脑上运行 AI 代理的程序,通过浏览器页面打开,并可通过插件扩展。","summary":"DeepSeek Harness(`dsh`)是 DeepSeek AI 开源的、用于在自己电脑上运行 AI 代理的运行框架。其中的一切都是插件,所以其他人也能为它添加功能。你安装 Node.js 24.9 或更高版本并运行一条命令,它会在 127.0.0.1:3080 启动本地网页界面并打开你的浏览器。它面向开发者:项目自称是开发者预览版,大写警告会有破坏兼容性的变更,并要求你在运行前阅读它的风险提示。","easeRationale":"三个系统上操作方式相同,也没有安装程序:你需要安装 Node.js(package.json 要求 24.9.0 或更高版本,比大多数系统自带的版本更新),然后在终端运行 `npx @deepseek-ai/dsh web`,它会在浏览器中打开界面。34 个发行版本中没有一个包含程序文件,所以桌面版必须从源码自行编译。","costNote":"代码在 MIT 许可下免费。它是面向 AI 模型的运行框架,模型使用通常需要付费;仓库内容没有说明你需要配置哪个服务商或密钥。","notVerified":["你需要配置哪个模型服务商和哪个 API 密钥,因为 README 没有说明","代理及其插件被允许对你的文件和网络做什么","风险提示具体要求你做什么,因为这个安装包里只有该文件的文件名","npx 命令是否真的能在 Windows 上按说明运行","安装加上编译产物之后的实际磁盘占用","正常一次会话的付费模型使用会花多少钱"]}}},{"id":"R_kgDOVWq3rA","slug":"denissergeevitch__ex-ussr-kinship-research","name":"Ex-USSR Kinship Research","owner":"DenisSergeevitch","fullName":"DenisSergeevitch/ex-ussr-kinship-research","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"handles_sensitive_data_about_real_people","severity":"attention","evidence":"参考资料文件夹中包含 \"privacy-and-safety.md — living people, minors, DNA, sensitive findings\"，README 也提醒 \"Keep private case data outside the skill's Git checkout.\" 案例文件以普通文件的形式保存在你的电脑上，里面有你亲属（包括在世亲属）的姓名、日期和文件资料。","source":"ai","title":"处理关于真实人物的敏感数据"},{"code":"agent_drives_your_logged_in_archive_account","severity":"attention","evidence":"README 写道：\"Find relevant FamilySearch books by place or DGS through an existing lawful browser session, view specific records, and retain only evidence fragments permitted by the source.\" 自动化使用一个你已登录的账号，可能违反网站条款，导致账号被限制。","source":"ai","title":"代理会操作你已登录的档案账号"},{"code":"agent_can_send_email_in_your_name","severity":"attention","evidence":"README 写道：\"Draft and, when authorized, send focused requests to archives, libraries, universities and museums using available email or official web tools\"，并且 \"it does not ask again for an already authorized send.\" 一旦你授权了发送，之后的消息就会在不再次确认的情况下发出。","source":"ai","title":"代理可以用你的名义发送邮件"},{"code":"instructions_are_written_in_russian","severity":"info","evidence":"README 写道：\"The skill's instructions are written in Russian, but the agent talks to you in your language.\" SKILL.md 是真正引导助手的文件，所以除非你会读俄语，否则无法核实自己安装的到底是什么。","source":"ai","title":"指令是用俄语写的"},{"code":"very_new_and_changing_daily","severity":"info","evidence":"该仓库创建于 2026-10-25，已经发布了 29 个版本，最新的是 2026-11-02 发布的 v1.32.0。变化这么快的指令，可能在不同的日子表现得不一样。","source":"ai","title":"项目非常新，而且每天都在变化"},{"code":"tries_a_captcha_once","severity":"info","evidence":"README 的 \"Responsible use\" 部分写道：\"The agent may try an actual CAPTCHA once through normal controls in its internal browser when supported and allowed\"，紧接着是 \"No external solvers, protection bypass, paywall, geo-block or rate-limit evasion, or leaked databases.\"","source":"ai","title":"会尝试一次验证码"}],"platforms":["windows","mac","linux"],"tools":["python","node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["FamilySearch"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","productivity"],"alternativeTo":[],"declaredAlternatives":[],"stars":2319,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":548,"growthDays":7,"history":[{"date":"2026-10-27","stars":1057},{"date":"2026-10-28","stars":1182},{"date":"2026-10-29","stars":1266},{"date":"2026-10-30","stars":1345},{"date":"2026-10-31","stars":1404},{"date":"2026-11-01","stars":1457},{"date":"2026-11-02","stars":1513},{"date":"2026-11-03","stars":1771},{"date":"2026-11-04","stars":1865},{"date":"2026-11-05","stars":1972},{"date":"2026-11-06","stars":2052},{"date":"2026-11-07","stars":2134},{"date":"2026-11-08","stars":2197},{"date":"2026-11-09","stars":2262},{"date":"2026-11-10","stars":2319}],"firstPublishedAt":"2026-10-27T18:40:27.861Z","commitSha":"5525e28ecc2e87bf3190c0ca5a9e254c0a965be2","verifiedAt":"2026-11-03T02:40:53.628Z","lastCommitAt":"2026-11-09T21:14:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill that turns an AI assistant into a family-history researcher for the former Soviet Union, with a source for every claim.","summary":"This is an instruction pack you install into an AI assistant such as Claude Code or Codex. You describe a family question and the assistant works like a genealogist for the former Soviet Union and the Russian Empire: church and civil records, military, deportation and repression files, and archive scans. It keeps lists of people, documents and searches, labels each claim as proven or not, and writes an HTML report with a family tree. It answers in your language even though its own instructions are in Russian. You need an AI assistant, Python 3, and patience with archives.","easeRationale":"Windows, Mac and Linux are the same: you need an AI assistant already installed, then either run \"npx skills add\" (which needs Node.js) or clone the folder into the assistant's skills directory, and have Python 3 available for the helper scripts. The optional PDF export also needs Chrome, Chromium or Edge.","costNote":"The skill is free (MIT). It runs inside an AI assistant, which usually needs a paid plan or paid API usage, and some archives charge for copies.","notVerified":["Whether the archives, catalogs and record sets it points to are still reachable and hold what it expects","Whether the family links it concludes are correct — the report labels claims, but nobody checked them for you","What the Russian instruction file actually tells the assistant to do","Whether FamilySearch and the WWI record site keep working with this kind of automated reading","How much your AI assistant charges in tokens for a long research case"]},"zh-Hans":{"summaryShort":"一个让 AI 助手变成前苏联家族史研究员的技能，每一条结论都有来源支撑。","summary":"这是一套你安装到 Claude Code 或 Codex 这类 AI 助手里的指令包。你描述一个家族问题，助手就会像前苏联和俄罗斯帝国的家谱研究员那样工作：教会和民事记录、军事、流放和镇压档案，以及档案扫描件。它会维护人员、文件和搜索记录的清单，为每个结论标注是否已被证实，并生成一份带家族树的 HTML 报告。即使它自己的指令是用俄语写的，它也会用你的语言回答。你需要一个 AI 助手、Python 3，以及面对档案资料的耐心。","easeRationale":"Windows、Mac 和 Linux 的要求相同：你需要已经安装好的 AI 助手，然后运行 \"npx skills add\"（需要 Node.js），或把文件夹克隆到助手的技能目录里，并准备好 Python 3 以运行辅助脚本。可选的 PDF 导出还需要 Chrome、Chromium 或 Edge。","costNote":"该技能是免费的（MIT 许可）。它运行在 AI 助手内部，而助手通常需要付费方案或付费的 API 用量，有些档案馆对复印件也会收费。","notVerified":["它指向的档案馆、目录和记录集是否仍然可以访问，并且确实包含它所预期的内容","它得出的亲属关系结论是否正确——报告会给结论打标签，但没有人替你核实过","那份俄语指令文件到底告诉助手要做什么","FamilySearch 和一战记录网站是否能一直配合这种自动化读取方式正常工作","一次长篇研究会消耗你 AI 助手多少 token"]}}},{"id":"R_kgDOVWmLuw","slug":"designfuchen__word-atlas","name":"Word Atlas","owner":"designfuchen","fullName":"designfuchen/word-atlas","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"web":"A"},"risk":"attention","signals":[{"code":"ai_generated_learning_content_not_reviewed","severity":"attention","evidence":"README 的 '内容来源与局限' 部分：释义、例句和插图由生成式 AI 制作，语音由 ElevenLabs 生成；内容通过了自动结构检查，但 '尚未逐条人工审校'（尚未逐条经过人工校对），可能包含错误的词义、句子、图片或发音。","source":"ai","title":"AI 生成的学习内容未经审校"},{"code":"repository_differs_from_hosted_website","severity":"info","evidence":"README 的 '版本说明'：仓库包含开放的基础版本和数据集，而线上网站 '独立迭代，功能与仓库可能不同'（自行迭代，功能可能与仓库不同）。","source":"ai","title":"仓库与线上网站不一致"},{"code":"content_under_separate_license","severity":"info","evidence":"LICENSE 和 CONTENT_LICENSE.md：MIT 仅涵盖代码和界面美术；词表、插图和录音采用 CC BY 4.0，如果你再次使用，需要注明 'Word Atlas contributors'。","source":"ai","title":"内容适用单独的许可证"},{"code":"very_new_repository_fast_growth","severity":"info","evidence":"GitHub 信息：仓库创建于 2026-10-25，到 2026-10-27 已有 2506 颗星和 587 个 fork，目前还没有打标签的发布版本。","source":"ai","title":"仓库很新，增长很快"}],"platforms":["web"],"tools":[],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["productivity"],"alternativeTo":[{"slug":"quizlet","name":"Quizlet","confirmed":true},{"slug":"maimemo","name":"Maimemo","confirmed":true}],"declaredAlternatives":[],"stars":4626,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":572,"growthDays":7,"history":[{"date":"2026-10-27","stars":2496},{"date":"2026-10-28","stars":3049},{"date":"2026-10-29","stars":3230},{"date":"2026-10-30","stars":3363},{"date":"2026-10-31","stars":3557},{"date":"2026-11-01","stars":3721},{"date":"2026-11-02","stars":3904},{"date":"2026-11-03","stars":4054},{"date":"2026-11-04","stars":4180},{"date":"2026-11-05","stars":4244},{"date":"2026-11-06","stars":4315},{"date":"2026-11-07","stars":4393},{"date":"2026-11-08","stars":4467},{"date":"2026-11-09","stars":4559},{"date":"2026-11-10","stars":4626}],"firstPublishedAt":"2026-10-27T18:40:27.870Z","commitSha":"99c007673d38f02c418d8e4cd2e9ec90b5d40be5","verifiedAt":"2026-10-28T01:41:44.664Z","lastCommitAt":"2026-11-08T16:06:33Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Illustrated IELTS vocabulary app: 7,298 words with pictures, UK/US audio and spaced-repetition review, free in the browser.","summary":"An English vocabulary app for IELTS students. Each of the 7,298 words comes with a drawing, Chinese and English meanings, two example sentences, British and American audio, and exercises; a review planner (FSRS) decides what to show you each day. You open it in a browser, no account needed, and progress stays on your device. The interface and most documentation are in Chinese. The repository is the open base version plus the word dataset; the author's website iterates separately and may have other features. Running the repository version yourself needs Node.js 20+.","easeRationale":"The web is the only place an end user runs it, and there it is an A: the author's site opens the app in any browser with no install and no account. Only people who want to host their own copy need Node.js 20+ and npm commands, and there are no desktop or mobile builds in the repository.","costNote":"App code is MIT and the dataset is CC BY 4.0. An iOS version is in closed beta; the README gives no price for it.","notVerified":["Whether the meanings, example sentences and recordings are correct, since they were generated and not proofread one by one","How much disk space a full clone takes, as the repository ships 7,298 illustrations and 22,905 audio files","Whether offline caching and the export/restore of your progress work as described","Whether the announced iOS app will stay free and whether it will match the open version"]},"zh-Hans":{"summaryShort":"配图版雅思词汇应用：7298 个单词配图片、英式/美式发音和间隔重复复习，浏览器里免费使用。","summary":"一款面向雅思考生的英语词汇应用。7298 个单词中的每一个都配有插画、中英文释义、两个例句、英式和美式发音，以及练习题；一个复习计划器（FSRS）决定每天给你看什么。你在浏览器里打开即可使用，不需要账号，进度保存在你的设备上。界面和大部分文档是中文。仓库里是开放的基础版加上词汇数据集；作者的网站在单独迭代，可能有其他功能。要自己运行仓库版本，需要 Node.js 20+。","easeRationale":"浏览器是终端用户唯一的使用场所，在这里它得到 A 级：作者的网站可以在任何浏览器里打开应用，无需安装、无需账号。只有想自己搭建的人才需要 Node.js 20+ 和 npm 命令，仓库里没有桌面端或移动端的构建版本。","costNote":"应用代码采用 MIT 许可，数据集采用 CC BY 4.0 许可。有一个 iOS 版本正在封闭测试中；README 没有给出它的价格。","notVerified":["释义、例句和录音是否准确，因为它们是生成出来的，没有逐一校对。","完整克隆占用多少磁盘空间，因为仓库里有 7298 张插画和 22905 个音频文件。","离线缓存以及进度的导出/恢复功能是否真的像描述那样运作。","官方宣布的 iOS 应用是否会一直免费，以及是否会和开放版本保持一致。"]}}},{"id":"R_kgDOOztdqQ","slug":"devnen__chatterbox-tts-server","name":"Chatterbox TTS Server","owner":"devnen","fullName":"devnen/Chatterbox-TTS-Server","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"D","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖中存在已知漏洞(OSV):torch: GHSA-63cw-57p8-fm3p, GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; torch: GHSA-rrmf-rvhw-rf47, PYSEC-2026-139; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74","source":"check","title":"已知漏洞"},{"code":"engine_installed_from_a_git_fork","severity":"attention","evidence":"每一条安装路径最后都是\"pip install --no-deps git+https://github.com/devnen/chatterbox-v2.git@master\"(Dockerfile 和所有依赖文件都是如此)。语音引擎取自维护者自己的 fork,内容就是当天 master 分支上有什么,而 --no-deps 跳过了依赖检查,所以落到你机器上的版本并没有被锁定。","source":"ai","title":"引擎从一个git fork安装"},{"code":"voice_cloning_of_real_people","severity":"attention","evidence":"\"Voice Cloning: Supports voice cloning using a reference audio file (.wav or .mp3)\",但 README 中任何地方都没有同意或可接受使用方面的说明。只要一段某人声音的简短录音,就足以让他们看起来说了任何话。该引擎确实包含了 Resemble 的水印库(依赖文件中的 resemble-perth==1.0.1)。","source":"ai","title":"克隆真实人物的声音"},{"code":"web_server_without_a_password","severity":"attention","evidence":"服务器开放 8004 端口,提供一个网页界面和一个兼容 OpenAI 的 API,界面还能\"view/edit config.yaml settings (server, model, paths)\";README 没有描述任何登录或令牌机制。v2.0.0 的更新说明也记录了一个过去的漏洞:\"Security: CWE-22 path traversal fixed on /tts and /v1/audio/speech voice file parameters\"。请把它留在你自己的机器上,或者放在需要密码才能访问的环境后面。","source":"ai","title":"没有密码保护的网页服务器"},{"code":"python_3_10_only","severity":"info","evidence":"README 的前提条件写道:\"Python: Version 3.10 required ... Python 3.11+ is not supported — key dependencies lack pre-built wheels, causing build failures.\"在 Windows 上,Portable Mode 会自带下载一个内嵌的 Python 3.10。","source":"ai","title":"仅支持python 3.10"},{"code":"amd_docker_relaxes_container_isolation","severity":"info","evidence":"docker-compose-rocm.yml、-rdna4.yml 和 -strixhalo.yml 设置了\"security_opt: seccomp=unconfined\"\"ipc: host\",并允许访问 /dev/kfd 和 /dev/dri 设备,还建议在 GPU 访问失败时取消注释\"privileged: true\"和\"/dev/mem\",这会让容器获得对主机极为广泛的访问权限。","source":"ai","title":"amd的docker降低了容器隔离度"}],"platforms":["windows","linux","mac"],"tools":["python","docker","gpu_nvidia","ffmpeg"],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["Hugging Face"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"speechify","name":"Speechify","confirmed":true}],"declaredAlternatives":[{"target":"projectivy-launcher","stance":"claims","evidence":"This guide provides multiple methods: using the automated launcher, the recommended `git stash` workflow, and a manual backup alternative"}],"stars":1499,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":4,"growthDays":6,"history":[{"date":"2026-11-07","stars":1497},{"date":"2026-11-08","stars":1497},{"date":"2026-11-09","stars":1499},{"date":"2026-11-10","stars":1499}],"firstPublishedAt":"2026-11-06T22:57:16.578Z","commitSha":"915ae289340e10c6047f27f47e22eae9bf350c32","verifiedAt":"2026-11-06T22:41:17.427Z","lastCommitAt":"2026-05-26T19:49:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns written text into speech on your own computer: pick a ready voice or copy one from a sample, even full audiobooks.","summary":"This project wraps Resemble AI's Chatterbox speech models in a server with a web page and an API. You paste text, choose one of the included voices or upload a short recording to copy a voice, and it writes out audio; long texts are split automatically, so a whole book becomes one file. Three models are offered (English, 23 languages, and a faster Turbo one) and you switch between them in a dropdown. It runs on Windows, Linux and Apple Silicon with Python 3.10, faster with an NVIDIA or AMD graphics card, and there is also a Docker route and a Google Colab demo.","easeRationale":"Windows is the easiest: install Python 3.10, clone the repository and double-click start.bat, and the launcher detects your card, installs everything and opens the web page (its Portable Mode then makes the folder movable). Linux is the same work with ./start.sh. macOS is harder (a D): the hardware table lists Apple Silicon as \"Manual install\", so you build the environment and install the packages by hand.","costNote":"The server and the models are free to download and run on your own machine; you pay in hardware and electricity, or in rented GPU time.","notVerified":["Whether it installs and uses your particular graphics card, given the many separate requirement files per card family","Real disk use: the README itself recommends 10 GB or more for dependencies and model cache, and the models alone are about 2 GB","How good the cloned voices sound and how fast generation is on your hardware","Whether every generated file carries the Perth watermark","Whether the Google Colab demo still runs"]},"zh-Hans":{"summaryShort":"在你自己的电脑上把文字转换成语音:选一个现成的声音,或从样本中复制一个声音,甚至能生成完整的有声书。","summary":"这个项目把 Resemble AI 的 Chatterbox 语音模型封装成一个带网页和 API 的服务器。你粘贴文本,选择内置的某个声音,或上传一段简短录音来复刻某个声音,它就会生成音频;长文本会自动拆分,所以整本书会变成一个文件。项目提供三个模型(英语、23种语言,以及更快的 Turbo 版),你可以在下拉菜单中切换。它可以在 Windows、Linux 和 Apple Silicon 上运行,需要 Python 3.10,配备 NVIDIA 或 AMD 显卡会更快,此外还有 Docker 方式和 Google Colab 演示。","easeRationale":"Windows 上最简单:安装 Python 3.10,克隆仓库,双击 start.bat,启动器会检测你的显卡、安装所有依赖并打开网页(之后它的 Portable Mode 能让这个文件夹随意移动)。Linux 上的工作量相同,用 ./start.sh。macOS 更难(评级为 D):硬件表把 Apple Silicon 标为\"Manual install\",也就是说你要自己搭建环境、手动安装各个包。","costNote":"服务器和模型都可以免费下载并在你自己的机器上运行;你付出的是硬件和电费,或者租用 GPU 的时间成本。","notVerified":["能否安装并使用你那张具体的显卡,因为不同显卡系列各有一套单独的依赖文件","实际的磁盘占用:README 本身建议为依赖项和模型缓存预留 10 GB 以上空间,模型本身大约就有 2 GB","克隆出来的声音听起来有多好,以及在你的硬件上生成速度有多快","每个生成的文件是否都带有 Perth 水印","Google Colab 演示是否还能运行"]}}},{"id":"R_kgDOS4HNrw","slug":"dietrichgebert__ponytail","name":"Ponytail","owner":"DietrichGebert","fullName":"DietrichGebert/ponytail","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"140 天内获得 180209 颗星（约每天 1287 颗）","source":"check","title":"星标增长速度异常"},{"code":"runs_hooks_on_agent_events","severity":"info","evidence":"INSTALL.md：\"The Claude Code and Codex plugins (and the Cursor hooks) run two tiny Node.js lifecycle hooks, so `node` needs to be on your PATH\"；在 Codex 中你必须 \"review and trust its two lifecycle hooks\"。","source":"ai","title":"在代理事件上运行钩子（hooks）"},{"code":"self_reported_benchmarks","severity":"info","evidence":"README.md 中醒目的数字 \"-53% code · -41% time · -26% cost · -45% tokens\" 来自作者自己的基准测试（\"39 tasks ... Opus 5.5, 5 runs each\"，benchmarks/results/2026-10-07-agentic.md），并非独立测试。","source":"ai","title":"基准测试数据由作者自行提供"},{"code":"reads_past_agent_sessions","severity":"info","evidence":"README.md：\"`/ponytail band import` adds your past Claude Code sessions to it, with their cost, agent time and tokens\"；数据保存位置为 \"By default ... on your machine, in `~/.config/ponytail/band/`\"。","source":"ai","title":"读取过往的代理会话"},{"code":"warning_about_fake_copies","severity":"info","evidence":"README.md：\"Only install ponytail from `DietrichGebert/ponytail` on GitHub or `@dietrichgebert/ponytail` on npm. It never ships `.exe` or `.dll` files; a copy that does is not mine.\"","source":"ai","title":"关于假冒副本的警告"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["AI coding assistant account (for example Claude Code, Codex or Gemini CLI)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"coderabbit","name":"CodeRabbit","confirmed":true}],"declaredAlternatives":[],"stars":188329,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":4610,"growthDays":7,"history":[{"date":"2026-10-26","stars":177149},{"date":"2026-10-27","stars":178192},{"date":"2026-10-28","stars":179204},{"date":"2026-10-29","stars":180194},{"date":"2026-10-30","stars":180967},{"date":"2026-10-31","stars":181525},{"date":"2026-11-01","stars":182171},{"date":"2026-11-02","stars":182899},{"date":"2026-11-03","stars":183719},{"date":"2026-11-04","stars":184456},{"date":"2026-11-05","stars":185170},{"date":"2026-11-06","stars":185792},{"date":"2026-11-07","stars":186332},{"date":"2026-11-08","stars":186858},{"date":"2026-11-09","stars":187603},{"date":"2026-11-10","stars":188329}],"firstPublishedAt":"2026-10-27T01:10:12.621Z","commitSha":"96608e078b2d5ea607c5c3b8b5a54d1c1fba6846","verifiedAt":"2026-10-30T01:41:16.328Z","lastCommitAt":"2026-10-29T11:59:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An add-on for AI coding assistants that pushes them to write less, simpler code and reuse what already exists.","summary":"Ponytail is one set of instructions (a prompt) plus small plugins for AI coding assistants such as Claude Code, Codex, Copilot CLI, Cursor, Gemini CLI and OpenCode. It tells the assistant to reuse existing code and built-in features before writing new code, while keeping validation, security, tests and accessibility. It adds review, audit and debt commands and, in Claude Code, a bar above the prompt with the cost, tokens and time of each task. It is for programmers who already use one of these assistants; in Claude Code you install it with two commands.","easeRationale":"Same on Windows, Mac and Linux: you need an AI coding assistant already installed plus Node.js on your PATH for the two lifecycle hooks, then you type the install commands (in Claude Code, /plugin marketplace add DietrichGebert/ponytail and /plugin install ponytail@ponytail as two separate prompts).","costNote":"Ponytail is free (MIT). It only works inside an AI coding assistant, which usually needs a paid plan or paid API usage; the package does not state prices.","notVerified":["Whether the claimed savings in code, time and cost hold on your own projects.","Exactly what the two Node.js hooks do on each prompt and whether any data leaves your machine.","Whether the install works with the version of your AI coding assistant, especially on Windows.","How the extra instructions added to every turn change your AI usage bill.","What the local board page stores about your past sessions."]},"zh-Hans":{"summaryShort":"一款面向 AI 编程助手的插件,推动它们写更少、更简单的代码,并复用已有内容。","summary":"Ponytail 是一套指令(prompt),外加适配 Claude Code、Codex、Copilot CLI、Cursor、Gemini CLI 和 OpenCode 等 AI 编程助手的小插件。它要求助手在写新代码之前先复用现有代码和内置功能,同时不牺牲校验、防护、测试和无障碍这几方面。它新增了审查、审计和技术债相关命令,在 Claude Code 中还会在输入框上方显示一条状态栏,展示每个任务的花费、令牌数和耗时。它面向已经在使用这些助手之一的程序员;在 Claude Code 中用两条命令即可安装。","easeRationale":"在 Windows、Mac 和 Linux 上流程相同:你需要先安装好一个 AI 编程助手,并让 Node.js 可在 PATH 中被找到,供两个生命周期钩子使用,然后输入安装命令(在 Claude Code 中,依次作为两条独立指令运行 /plugin marketplace add DietrichGebert/ponytail 和 /plugin install ponytail@ponytail)。","costNote":"Ponytail 免费(MIT 许可证)。它只能在 AI 编程助手内使用,通常需要付费计划或付费 API 用量;安装包中没有说明具体价格。","notVerified":["所宣称的代码量、时间和成本节省,在你自己的项目中是否同样成立。","两个 Node.js 钩子在每次提示时具体做了什么,以及是否有数据离开你的电脑。","安装是否能配合你所用的 AI 编程助手版本正常工作,尤其是在 Windows 上。","每个回合新增的额外指令会如何影响你的 AI 用量账单。","本地看板页面会保存关于你过去会话的哪些信息。"]}}},{"id":"R_kgDOVbKw8g","slug":"djihads80__eglantine","name":"Eglantine","owner":"Eglantine-Community","fullName":"Eglantine-Community/Eglantine","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"10 天内获得 5425 个星标","source":"check","title":"成立时间短但星标很多的仓库"},{"code":"young_owner_account","severity":"attention","evidence":"所有者账号是 1 天前创建的","source":"check","title":"账号注册时间很短的所有者"},{"code":"runs_commercial_console_games","severity":"high","evidence":"该仓库的描述写的是 \"VERY experimental Nintendo Switch 2 emulator\"，README 里说该模拟器 \"is capable of running Mario Kart World at decent/full speed depending on specs and configurations, and Splatoon Raiders at variable performances\"。这些都是为某款主机销售的任天堂商业游戏，程序本身并不包含它们，所以需要用户自己提供游戏文件。一般来说，主机厂商的条款禁止绕过主机及其游戏内置的保护措施，许多国家也有法律限制绕过版权保护。","source":"ai","title":"运行商业主机游戏"},{"code":"declared_very_experimental","severity":"attention","evidence":"所有者自己的描述开头就是 \"VERY experimental\"，README 里写着 \"A list of supported games will be available in future. Please be patient\"，而这个创建于 2026-10-29 的仓库已经有 79 个发行版（最近三个分别是 2026-11-08 的 v0.1.0-rc3、2026-11-08 的 branch-nx2-experimental-efb9fb18 和 2026-11-07 的 daily-20261107）。要做好遇到崩溃和快速变化版本的准备。","source":"ai","title":"声明为高度实验性"},{"code":"derived_from_other_emulators","severity":"info","evidence":"README 的头部带有 \"SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project\" 和 \"SPDX-FileCopyrightText: 2018 yuzu Emulator Project\"，logo 链接也仍然指向 https://git.eden-emu.dev/eden-emu/eden。大部分代码来自这些更早的项目，而不是来自这个仓库本身。","source":"ai","title":"派生自其他模拟器"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":106,"maxMB":318,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":5761,"downloads":685301,"downloadsWeek":383059,"downloadsDays":6,"growth":2737,"growthDays":7,"history":[{"date":"2026-11-04","stars":3563},{"date":"2026-11-05","stars":4301},{"date":"2026-11-06","stars":4839},{"date":"2026-11-07","stars":5172},{"date":"2026-11-08","stars":5407},{"date":"2026-11-09","stars":5614},{"date":"2026-11-10","stars":5761}],"firstPublishedAt":"2026-11-04T01:49:36.029Z","commitSha":"78743f30c251f66bd77e6eea6c28d8632d642d6f","verifiedAt":"2026-11-08T23:06:05.783Z","lastCommitAt":"2026-11-08T14:57:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An early-stage Nintendo Switch 2 emulator that tries to run Switch 2 games on a Windows, Mac or Linux computer.","summary":"Eglantine is an emulator that tries to run Nintendo Switch 2 games on a normal computer. The owner calls it \"VERY experimental\" and the README is derived from the Eden, yuzu and Sudachi emulator projects. Ready-made files are published for Windows, Mac (Apple chips) and Linux, so there is nothing to compile. The program itself does not come with games: the README names commercial Switch 2 titles, so you have to supply the game files yourself. Written in C++ and licensed under GPLv3. Built for people who like to tinker with emulators.","easeRationale":"On Windows you download a ready .zip (65-110 MB) and open the program, and Mac (.dmg) and Linux (.AppImage) are the same, so all three are equally easy to install. The extra step that makes it a B everywhere is that the emulator comes with no games, so you have to supply the game files yourself; the Mac build is only for Apple-chip Macs.","costNote":"The README says the project takes no donations for now.","notVerified":["Whether any game really runs at playable speed on your own computer, and which games work at all (the README says the compatibility list does not exist yet)","Whether the published Windows, Mac and Linux files were built from the source code in this repository","Whether the Mac and Linux builds work as well as the Windows one, since the project is days old","What exactly you must supply to start a game, because the user handbook was not read here"]},"zh-Hans":{"summaryShort":"一款处于早期阶段的任天堂 Switch 2 模拟器,尝试在 Windows、Mac 或 Linux 电脑上运行 Switch 2 游戏。","summary":"Eglantine 是一款试图在普通电脑上运行任天堂 Switch 2 游戏的模拟器。作者自己称它为 \"VERY experimental\",README 衍生自 Eden、yuzu 和 Sudachi 这几个模拟器项目。Windows、Mac(Apple 芯片)和 Linux 都提供现成文件,不需要自己编译。程序本身不附带游戏:README 点名了一些商业性的 Switch 2 游戏,所以你得自己提供游戏文件。用 C++ 编写,采用 GPLv3 许可。是给喜欢捣鼓模拟器的人做的。","easeRationale":"在 Windows 上,你下载一个现成的 .zip(65-110 MB)并打开程序,Mac(.dmg)和 Linux(.AppImage)的方式一样,所以三者安装起来同样简单。让它在各平台都归为 B 级的额外一步是,这款模拟器不附带游戏,你得自己提供游戏文件;Mac 版本只支持 Apple 芯片的 Mac。","costNote":"README 说项目目前不接受捐赠。","notVerified":["在你自己的电脑上,是否真的有游戏能以可玩的速度运行,哪些游戏能跑起来(README 说兼容性列表目前还不存在)","发布的 Windows、Mac 和 Linux 文件是否是从这个仓库里的源代码编译出来的","Mac 和 Linux 版本是否和 Windows 版本一样好用,毕竟项目才存在了几天","要启动一个游戏具体需要提供什么,因为这里没有去读用户手册"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNTA0NTc1MQ==","slug":"docker__compose","name":"Docker Compose","owner":"docker","fullName":"docker/compose","type":"framework_infra","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"a_compose_file_can_reach_into_your_computer","severity":"attention","evidence":"README.md 中的示例把你当前的文件夹挂载进容器（volumes: - .:/code），发布一个端口（\"5000:5000\"），并拉取别人的镜像（image: redis）。因此，一个从网上复制来的 compose.yaml 可以读写你的文件、在你的机器上打开端口，并运行一个你没有检查过的镜像里的代码。","source":"ai","title":"一个 compose 文件能触及你的电脑"},{"code":"can_report_run_data_to_a_collector","severity":"info","evidence":"go.mod 中包含 OpenTelemetry 的追踪和指标导出器（go.opentelemetry.io/otel/exporters/otlp/otlptrace、.../otlpmetricgrpc）。这段代码会把你命令的详细信息发送到你 Docker 配置中设定的收集端。","source":"ai","title":"可以向收集端报告运行数据"},{"code":"docker_desktop_is_not_free_for_every_company","severity":"info","evidence":"README：\"Docker Compose is included in Docker Desktop for Windows and macOS.\" Compose 本身的许可证是 Apache-2.0，但 Docker Desktop 是一个独立产品，对较大的组织有自己的商业条款。","source":"ai","title":"docker desktop 并非对每家公司都免费"},{"code":"big_version_jump_and_old_version_split_off","severity":"info","evidence":"列出的最新版本是 v5.7.0，go.mod 声明的是 module github.com/docker/compose/v5，而 README 写道：\"The Python version of Compose is available under the `v1` branch\"。为很旧版本编写的 compose 文件和命令，可能表现不一样。","source":"ai","title":"版本跨度大，旧版本已被拆分出去"},{"code":"swarm_users_lose_features","severity":"info","evidence":"README 中的说明：Docker Swarm \"did not adopt the compose specification so is missing some of the recent enhancements in the compose syntax\"，在被 Mirantis 收购之后，\"swarm isn't maintained by Docker Inc, and as such some Docker Compose features aren't accessible to swarm users\"。","source":"ai","title":"swarm 用户会失去一些功能"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["docker","go"],"disk":{"minMB":33,"maxMB":100,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","network_system","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":38682,"downloads":512079721,"downloadsWeek":2362755,"downloadsDays":7,"growth":90,"growthDays":7,"history":[{"date":"2026-10-26","stars":38505},{"date":"2026-10-27","stars":38513},{"date":"2026-10-28","stars":38530},{"date":"2026-10-29","stars":38518},{"date":"2026-10-30","stars":38530},{"date":"2026-10-31","stars":38550},{"date":"2026-11-01","stars":38568},{"date":"2026-11-02","stars":38582},{"date":"2026-11-03","stars":38592},{"date":"2026-11-04","stars":38599},{"date":"2026-11-05","stars":38616},{"date":"2026-11-06","stars":38626},{"date":"2026-11-07","stars":38637},{"date":"2026-11-08","stars":38652},{"date":"2026-11-09","stars":38665},{"date":"2026-11-10","stars":38682}],"firstPublishedAt":"2026-10-27T01:10:12.622Z","commitSha":"5bceecf985347fcf6dd23029ce440b502a0d4301","verifiedAt":"2026-11-03T02:42:57.068Z","lastCommitAt":"2026-11-10T21:40:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Starts a whole set of programs that run in containers with one command, described in a single compose.yaml file.","summary":"The official Docker tool for running several containers together. You write one `compose.yaml` file that lists the parts of your app, for example a web service and a database, and `docker compose up` builds them, connects them and starts them all. It is made for people who already use Docker: on Windows and macOS it comes inside Docker Desktop, and on Linux you download the file for your machine and copy it into Docker's plugin folder. Everything happens in a terminal, and Docker itself must be installed and running.","easeRationale":"The grade is the same on all three systems because Docker itself and a terminal are needed either way. Windows and macOS are the lighter path, since the README says Compose is already included in Docker Desktop; on Linux you download the matching release file, rename it to `docker-compose`, copy it into `$HOME/.docker/cli-plugins` and may have to run `chmod +x` on it. Then you write a YAML file by hand.","costNote":"Compose is free and open (Apache-2.0). On Windows and macOS it arrives with Docker Desktop, which Docker charges larger companies for.","notVerified":["Real disk use, which depends on the container images your compose file pulls, not on the 33 MB program","Whether your installed Docker version works with this release of Compose","What a given `compose.yaml` from the internet will do once started, since that file decides which of your folders and ports the containers may touch","Whether any usage data actually leaves your machine, which depends on your Docker configuration"]},"zh-Hans":{"summaryShort":"用一条命令启动一整套容器化程序，全部写在一个 compose.yaml 文件里。","summary":"这是 Docker 官方提供的、用于同时运行多个容器的工具。你编写一个 compose.yaml 文件，列出应用的各个部分，例如一个网页服务和一个数据库，然后 docker compose up 会构建、连接并启动它们。它面向已经在使用 Docker 的人：在 Windows 和 macOS 上它已经包含在 Docker Desktop 里，在 Linux 上你需要下载适合自己机器的文件，并复制到 Docker 的插件目录中。所有操作都在终端里完成，而且必须先安装并运行 Docker 本身。","easeRationale":"三个系统的评分相同，因为无论如何都需要安装 Docker 本身并使用终端。Windows 和 macOS 是更轻松的路径，因为 README 说 Compose 已经包含在 Docker Desktop 里；在 Linux 上，你需要下载对应版本的文件，把它改名为 docker-compose，复制到 $HOME/.docker/cli-plugins 目录，可能还要对它执行 chmod +x。之后你需要手动编写一个 YAML 文件。","costNote":"Compose 是免费、开放源代码的（Apache-2.0）。在 Windows 和 macOS 上，它随 Docker Desktop 一起提供，而 Docker 会向大型公司收取 Docker Desktop 的费用。","notVerified":["实际占用的磁盘空间，这取决于你的 compose 文件拉取的容器镜像，而不是这个 33 MB 的程序本身","你安装的 Docker 版本是否能配合这个版本的 Compose 使用","从网上获取的某个 compose.yaml 启动后会做什么，因为这个文件决定了容器可以触碰你哪些文件夹和端口","是否真的有使用数据离开你的机器，这取决于你的 Docker 配置"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTU3NzMwNA==","slug":"dolphin-emu__dolphin","name":"Dolphin","owner":"dolphin-emu","fullName":"dolphin-emu/dolphin","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"you_must_bring_your_own_game_files","severity":"attention","evidence":"该仓库不包含任何游戏。Readme 中 DolphinTool 部分处理用户自行提供的文件:\"-i FILE, --input=FILE  Path to disc image FILE.\" 以及 \"-f rvz|iso|gcz|wia, --format=rvz|iso|gcz|wia\",主程序还有 \"-n <16-character ASCII title ID>, --nand_title=<16-character ASCII title ID>  Launch a NAND title\"。关于复制光盘和主机数据的规定因国家而异。","source":"ai","title":"需要自备游戏文件"},{"code":"mixed_licenses_inside_the_repository","severity":"info","evidence":"检查结果将许可证标记为 GPL-2.0-or-later,COPYING 文件写道:\"Most of Dolphin's original source code is licensed under GPLv2+. Part of the source code distributed in this repository is derived from other projects and may have stronger or weaker licensing requirement.\"每个文件的许可标记位于 LICENSES/ 目录中。","source":"ai","title":"仓库内许可证混杂"},{"code":"no_releases_published_on_github","severity":"info","evidence":"项目信息写着 \"GitHub releases: 0\",软件包的 releases 部分写着 \"No GitHub releases.\",检查结果中 lastReleaseAt 为 null,而最近一次提交是在 2026-11-06。","source":"ai","title":"github 上没有发布任何 release"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0-or-later","category":"copyleft","detectedFrom":"reuse"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":15794,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":25,"growthDays":6,"history":[{"date":"2026-11-07","stars":15774},{"date":"2026-11-08","stars":15781},{"date":"2026-11-09","stars":15783},{"date":"2026-11-10","stars":15794}],"firstPublishedAt":"2026-11-06T23:09:33.759Z","commitSha":"4a70679d0adcd17220568f5c18d2d4005eeea0a2","verifiedAt":"2026-11-06T23:01:23.885Z","lastCommitAt":"2026-11-10T21:24:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Play GameCube and Wii games on your computer from disc image files that you provide yourself.","summary":"Dolphin runs GameCube and Wii games on a computer. The Readme says it works on Windows (10 1903 or higher), Linux and macOS (11.0 Big Sur or higher), and asks for a modern dual-core CPU and a graphics card with Direct3D 11.1 or OpenGL 3.3. You open a game file that you supply; the command line also loads a save state (\"-s, --save_state\") or plays a movie file (\"-m, --movie\"). A companion command-line tool, DolphinTool, can convert, verify, extract and read the header of disc images. The repository has no GitHub releases and the Readme points only to build instructions.","easeRationale":"On all three systems the package offers no download link and no install steps: the Readme's \"Building\" section links wiki pages \"Building for Windows\", \"Building for Linux\" and \"Building for macOS\" and says \"Before building, make sure to pull all submodules\", and the project facts say \"GitHub releases: 0\". The uninstall section does mention \"the NSIS installer\" on Windows and \"delete Dolphin.app\" on macOS, so ready builds exist somewhere, but the package does not show where to get them.","notVerified":["Whether installers offered outside GitHub match the code in this repository.","Whether a given game actually runs, and at what speed, on your PC or graphics card.","How much disk space the program and your own game files take.","Whether the Android build mentioned in the Readme installs on a given phone."]},"zh-Hans":{"summaryShort":"在自己的电脑上运行 GameCube 和 Wii 游戏,使用你自己提供的光盘镜像文件。","summary":"Dolphin 可以在电脑上运行 GameCube 和 Wii 游戏。Readme 说它可以在 Windows(10 1903 或更高版本)、Linux 和 macOS(11.0 Big Sur 或更高版本)上运行,需要一块现代双核 CPU 和支持 Direct3D 11.1 或 OpenGL 3.3 的显卡。你打开自己提供的游戏文件;命令行还可以加载存档状态(\"-s, --save_state\")或播放录像文件(\"-m, --movie\")。配套的命令行工具 DolphinTool 可以转换、校验、提取光盘镜像并读取其文件头。该仓库没有发布 GitHub releases,Readme 只给出了构建说明的链接。","easeRationale":"在这三个系统上,该软件包都没有提供下载链接或安装步骤:Readme 的 \"Building\" 部分链接到维基页面 \"Building for Windows\"、\"Building for Linux\" 和 \"Building for macOS\",并写道 \"Before building, make sure to pull all submodules\",项目信息也写着 \"GitHub releases: 0\"。卸载部分确实提到了 Windows 上的 \"the NSIS installer\" 和 macOS 上的 \"delete Dolphin.app\",说明某处确实存在可用的构建版本,但软件包没有说明在哪里获取。","notVerified":["GitHub 之外提供的安装程序是否与此仓库中的代码一致。","某款游戏在你的电脑或显卡上是否真的能运行,以及运行速度如何。","该程序及你自己的游戏文件占用多少磁盘空间。","Readme 中提到的 Android 版本是否能安装到某款手机上。"]}}},{"id":"R_kgDOROahWg","slug":"donchitos__claude-code-game-studios","name":"Claude Code Game Studios","owner":"Donchitos","fullName":"Donchitos/Claude-Code-Game-Studios","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"runs_scripts_automatically","severity":"attention","evidence":"README 在 .claude/hooks 下列出了 15 个事件钩子(“18 scripts (bash)”),会“run automatically on every session”,并在每次调用 Bash、PowerShell、Read、Write 或 Edit 时触发;notify.sh “uses PowerShell for Windows toast notifications”。克隆这个模板意味着,这些来自陌生人仓库的脚本一打开会话就会在你的电脑上运行。","source":"ai","title":"脚本会自动运行"},{"code":"model_usage_can_get_expensive","severity":"attention","evidence":"README:“the three directors ask for Opus, lead-programmer and fifteen specialists name Sonnet”,测量表显示标准严谨程度在写出游戏第一行代码之前,会用 58 分钟写出 30 份文档。每一份文档和每一个角色都是你账户上的付费模型用量。","source":"ai","title":"模型使用费用可能很高"},{"code":"optional_mode_acts_without_asking","severity":"info","evidence":"README:“The guided and autonomous automation settings, and Claude Code's own auto mode, trade some of these prompts for speed”,也就是说默认的“写入前先询问”行为可以被关闭。","source":"ai","title":"可选模式会不经确认自行操作"},{"code":"author_says_extra_process_did_not_help","severity":"info","evidence":"README 自己的测试显示,标准严谨程度的版本被两位盲评者和一次真人试玩评为“last”:“the extra 29 documents bought traceability, not a better game”。更重的设置花费更多,却没有带来更好的结果。","source":"ai","title":"作者自己说额外流程没有帮助"},{"code":"counts_disagree","severity":"info","evidence":"仓库描述写的是“49 AI agents, 72 workflow skills”,而 README 的标题和徽章写的是“49 agents. 74 skills”,结构章节里也写的是 74。","source":"ai","title":"数字互相矛盾"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Anthropic Claude (through Claude Code)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","games","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":38008,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":2673,"growthDays":6,"history":[{"date":"2026-11-05","stars":35730},{"date":"2026-11-06","stars":36065},{"date":"2026-11-07","stars":36512},{"date":"2026-11-08","stars":36937},{"date":"2026-11-09","stars":37490},{"date":"2026-11-10","stars":38008}],"firstPublishedAt":"2026-11-04T22:32:54.892Z","commitSha":"05ad7c8a5b8aa2aa86bbd3439ae7cf7295cac418","verifiedAt":"2026-11-04T22:15:37.425Z","lastCommitAt":"2026-11-09T12:39:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A project template that gives Claude Code a 49-role game studio team and dozens of commands to plan, build and test a game.","summary":"This is a starting folder for a game project, full of instructions for the Claude Code assistant: 49 \"staff\" roles (directors, designers, programmers, QA), 74 slash commands for each step from brainstorming to release, document templates and scripts that check your work. It does not make a game by itself and ships no game: you clone it, open Claude Code inside it and type /start. You need Git, Claude Code, Python 3 and a Bash shell, plus a paid Claude plan or API credits. Godot, Unity and Unreal role sets are included.","easeRationale":"On Linux and Mac it is a C: Bash and Python 3 are usually already there, so you install Claude Code, clone the repository with Git and type /start in the terminal. Windows is a D because the README asks for Git for Windows to get a Bash shell for the hooks on top of Claude Code and Python 3; everything is done by typing commands, there is no installer.","costNote":"The template is free (MIT), but it only works inside Claude Code, which needs a paid Anthropic plan or API credits. The README says three of the roles ask for the Opus model, so heavy use can cost more.","notVerified":["Whether the 49 roles and 74 commands actually lead to a finished, playable game, or mostly to documents.","How much the model usage really costs for one project.","Whether every hook and script works on Windows, Mac and Linux as described.","Whether the README's four-game comparison can be reproduced, since no game is included in the repository.","What each of the 18 scripts does in detail, since only a one-line summary per hook is published."]},"zh-Hans":{"summaryShort":"这是一个项目模板,为 Claude Code 配备了一个拥有 49 个角色的游戏工作室团队,以及数十个命令,用来策划、制作和测试游戏。","summary":"这是一个游戏项目的起始文件夹,里面装满了给 Claude Code 助手的说明:49 个“员工”角色(导演、设计师、程序员、QA),74 个斜线命令覆盖从头脑风暴到发布的每个阶段,还有文档模板和检查你工作成果的脚本。它本身不会做出游戏,也不附带任何游戏:你需要克隆它,在里面打开 Claude Code,然后输入 /start。你需要 Git、Claude Code、Python 3 和一个 Bash 终端,以及付费的 Claude 方案或 API 额度。其中包含 Godot、Unity 和 Unreal 的角色集。","easeRationale":"在 Linux 和 Mac 上评级为 C:Bash 和 Python 3 通常已经自带,你只需安装 Claude Code,用 Git 克隆仓库,然后在终端输入 /start。Windows 评级为 D,因为 README 要求安装 Git for Windows 以获得供钩子使用的 Bash 终端,此外还需要 Claude Code 和 Python 3;整个过程都靠输入命令完成,没有安装程序。","costNote":"这个模板本身免费(MIT 许可),但只能在 Claude Code 里运行,而 Claude Code 需要付费的 Anthropic 方案或 API 额度。README 提到其中三个角色会调用 Opus 模型,大量使用时费用会更高。","notVerified":["这 49 个角色和 74 个命令是否真的能做出一个完整、可玩的游戏,还是大多只产出文档。","一个项目实际使用模型的费用到底有多少。","所有钩子和脚本是否真的像描述那样在 Windows、Mac 和 Linux 上都能正常工作。","README 中的四款游戏对比是否能被复现,因为仓库里并没有附带任何游戏。","18 个脚本各自具体做什么,因为每个钩子只公开了一行摘要。"]}}},{"id":"R_kgDOLIVTjQ","slug":"drewthomasson__ebook2audiobook","name":"ebook2audiobook","owner":"DrewThomasson","fullName":"DrewThomasson/ebook2audiobook","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"attention","signals":[{"code":"owner_warns_legally_acquired_books_only","severity":"info","evidence":"README的第一条提示写道：\"This tool is intended for use with non-DRM, legally acquired eBooks only. The authors are not responsible for any misuse of this software or any resulting legal consequences.\"这个项目不会移除版权保护，它只是读取你本来就能打开的文件。","source":"ai","title":"作者提醒只能用合法获得的书籍"},{"code":"voice_cloning_of_real_people","severity":"attention","evidence":"README的演示列表里包括\"David Attenborough Voice\"和\"Scarlett Voice\"，相关功能是\"Optional voice cloning using your own voice file\"，README还说理想的录音时长\"around 1 to 5 mn\"。一般来说，许多国家都有法律保护个人的声音和形象。","source":"ai","title":"复刻真人的声音"},{"code":"installs_other_software_on_your_machine","severity":"attention","evidence":"README写道：\"Note for MacOS users: homebrew is installed to install missing programs\"以及\"Note for Windows users: scoop is installed to install missing programs without administrator privileges\"。Dockerfile还会从\"https://download.calibre-ebook.com/linux-installer.sh\"获取Calibre，并用\"curl https://sh.rustup.rs -sSf | sh\"安装Rust。所以运行启动脚本会引入这个代码仓库之外的软件。","source":"ai","title":"会在你的设备上安装其他软件"},{"code":"optional_public_link_exposes_the_app","severity":"attention","evidence":"README写道：\"For Public Link: ./ebook2audiobook.command --share\"。这个选项会把Gradio页面发布到一个公开地址，于是任何拿到链接的人都能在它运行期间上传文件，使用你的设备。","source":"ai","title":"可选的公开链接会暴露这个应用"},{"code":"license_text_disagrees_across_files","severity":"info","evidence":"LICENSE文件和GitHub上写的是Apache-2.0，但Dockerfile里有LABEL org.opencontainers.image.licenses=\"MIT\"，pyproject.toml把它归类为\"License :: OSI Approved :: MIT License\"。两者都是宽松许可证，但项目没有统一说明到底是哪一个。","source":"ai","title":"不同文件里的许可证说法不一致"},{"code":"models_downloaded_after_install","severity":"info","evidence":"requirements.txt拉取了\"coqui-tts[languages]\"、\"piper-tts\"和\"huggingface-hub\"，Docker运行时还会挂载一个\"./models:/app/models\"文件夹。语音模型会在你第一次使用某个声音或语言时才下载，这还没算上Radar的检查为安装估算的837 MB到2512 MB。","source":"ai","title":"模型在安装之后才下载"}],"platforms":["windows","mac","linux","web"],"tools":["python","gpu_any"],"buildTools":["python","docker","gpu_nvidia"],"disk":{"minMB":837,"maxMB":2512,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","office","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"speechify","name":"Speechify","confirmed":true},{"slug":"audible","name":"Audible","confirmed":true}],"declaredAlternatives":[],"stars":20526,"downloads":992,"downloadsWeek":0,"downloadsDays":6,"growth":29,"growthDays":6,"history":[{"date":"2026-11-04","stars":20497},{"date":"2026-11-05","stars":20501},{"date":"2026-11-06","stars":20502},{"date":"2026-11-07","stars":20514},{"date":"2026-11-08","stars":20514},{"date":"2026-11-09","stars":20520},{"date":"2026-11-10","stars":20526}],"firstPublishedAt":"2026-11-04T01:49:36.039Z","commitSha":"0a98ac313ec43a470ee1e5feb13dd881ffb3e969","verifiedAt":"2026-11-04T01:44:35.863Z","lastCommitAt":"2026-11-09T16:50:45Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your own e-books and PDFs into audiobooks with computer voices, on your own machine, and can copy a voice you upload.","summary":"ebook2audiobook reads an e-book (EPUB, MOBI, PDF, Word, text and more, with OCR for scanned pages) and reads it out loud into an audiobook file with chapters, in .m4b, .mp3, .wav and other formats. Voices come from free speech models that run on your computer, in many languages, and you can upload a one to five minute recording to copy a voice. You work in a page that opens in your browser. It needs a download of the project and a launcher script that installs what is missing; a graphics card makes it far faster, as the models are very slow on a plain processor.","easeRationale":"The gentlest way in is the web: the author's free demo on Hugging Face Spaces, or the Google Colab notebook where you press the run buttons. On Windows, Mac and Linux you download the project folder and double-click a launcher (ebook2audiobook.cmd, or ebook2audiobook.command on Mac and Linux); it then installs the missing pieces itself (scoop on Windows, Homebrew on Mac) and opens the page at localhost:7860, which still means a download, a terminal window and a long first setup.","costNote":"Free (the LICENSE file is Apache-2.0). No paid service is needed: the voices run on your own machine. The free hosted demo on Hugging Face and the Google Colab notebook are shared machines.","notVerified":["Real disk use and download time, because each speech model and language is fetched the first time you use it.","How long a whole book takes: the README says modern engines are \"very slow on CPU\" without saying how slow.","Whether the automatic installation of scoop or Homebrew goes through cleanly on your machine.","How good the result sounds in your language, out of the 1158 the project claims.","Whether the free hosted demo on Hugging Face is running and how long its queue is.","Whether chapter detection works on your file; the README warns that EPUB has no standard chapter structure and that you should delete unwanted text by hand first."]},"zh-Hans":{"summaryShort":"把你自己的电子书和PDF变成带电脑合成语音的有声书，在你自己的设备上运行，还能复刻你上传的声音。","summary":"ebook2audiobook可以读取电子书（EPUB、MOBI、PDF、Word、文本等格式，扫描页面还支持OCR），并把它朗读成带章节的有声书文件，格式包括.m4b、.mp3、.wav等。朗读声音来自可以在你电脑上运行的免费语音模型，支持多种语言，你还可以上传一段一到五分钟的录音来复刻一个声音。你在浏览器打开的页面中操作。它需要下载项目并运行一个启动脚本来安装缺少的组件；显卡会让速度大幅提升，因为这些模型在普通处理器上运行非常慢。","easeRationale":"最轻松的入门方式是网页版：作者在Hugging Face Spaces上的免费演示，或者Google Colab笔记本，你只需要点运行按钮。在Windows、Mac和Linux上，你需要下载项目文件夹并双击启动脚本（ebook2audiobook.cmd，或在Mac和Linux上是ebook2audiobook.command）；它会自己安装缺少的组件（Windows上用scoop，Mac上用Homebrew），然后打开localhost:7860这个页面，这仍然意味着要下载文件、打开终端窗口，并经历较长的首次设置过程。","costNote":"免费（LICENSE文件是Apache-2.0）。不需要任何付费服务：语音是在你自己的设备上运行的。Hugging Face上的免费托管演示和Google Colab笔记本用的是共享机器。","notVerified":["实际占用的磁盘空间和下载时间，因为每种语音模型和语言都是第一次使用时才下载。","处理一整本书需要多长时间：README说现代引擎\"very slow on CPU\"，但没说明具体有多慢。","scoop或Homebrew的自动安装在你的设备上能否顺利完成。","在项目声称支持的1158种语言里，你用的那一种听起来效果如何。","Hugging Face上的免费托管演示是否在正常运行，排队时间有多长。","章节检测对你的文件是否有效；README提醒EPUB没有统一的章节结构标准，建议你先手动删除不想要的文字。"]}}},{"id":"R_kgDOUplEVg","slug":"droid-deck__droiddeck","name":"DroidDeck","owner":"Droid-Deck","fullName":"Droid-Deck/DroidDeck","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"39 天内获得 17460 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 39 天前","source":"check","title":"所有者账号很新"},{"code":"asks_you_to_turn_off_an_android_process_restriction","severity":"attention","evidence":"README 安装步骤 4：“On Android 14 and newer, turn on Disable child process restrictions in Developer options.”这是一个系统级的开发者选项，并不是只针对这款应用的设置。限制部分还写道：“Firefox sandboxing is reduced under proot.”","source":"ai","title":"要求关闭 android 的进程限制"},{"code":"reads_your_saved_steam_login_token","severity":"attention","evidence":"README，“Runtime, Wine and Proton”：“The arm64 client's sign in was checked by reading Steam's saved token with the ConnectCache format from deadlock-api-ingest.”这款应用会使用保存你 Steam 会话信息的文件。","source":"ai","title":"读取你保存的 steam 登录令牌"},{"code":"runs_store_clients_in_ways_the_stores_do_not_support","severity":"attention","evidence":"README，“Stores”：“Epic's official Linux launcher preview runs through FEX. That setup is outside Epic's listed Linux support.”Battle.net “runs Blizzard's own client in Proton”，并通过“PROTON_SPOOF_STEAMINPUT_VIDPID”把手柄伪装成 Xbox 360 手柄。商店可能把不受支持的配置当作限制账号的理由。","source":"ai","title":"以商店不支持的方式运行商店客户端"},{"code":"only_download_from_the_repository","severity":"info","evidence":"README 直接写明：“DroidDeck has no website. Only download it from this repository's Releases.”在其他任何地方提供的版本都不是这个项目发布的。","source":"ai","title":"只从该仓库下载"},{"code":"works_only_on_some_phone_chips","severity":"info","evidence":"README 安装步骤 1：“Android 9 or newer on Adreno 730 or newer, or any 8xx. Adreno 6xx is experimental. Mali, Xclipse, PowerVR and Adreno 710 are unsupported.”因此大多数手机都不在支持范围内。","source":"ai","title":"只能在部分手机芯片上运行"},{"code":"one_feature_needs_a_paid_program_you_own","severity":"info","evidence":"README，“Audio, video and HUD”：帧生成功能“needs your own copy of Lossless Scaling and ships none of its shaders.”Lossless Scaling 是 Steam 上的一款付费程序。","source":"ai","title":"某个功能需要你自己购买的付费程序"},{"code":"built_from_many_projects_under_different_licences","severity":"info","evidence":"致谢列表中包含数十个借用的组件，分别采用 GPL-2.0、GPL-3.0、LGPL-2.1、MIT、BSD、MPL-2.0、Apache-2.0 和 BSL-1.0 许可证，其中包括 itch.io 编译的 7-Zip，带有“unRAR restriction on RAR code”的限制。如果你再分发或修改这款应用，上述每一项条款都会在 GPL-3.0 之上额外适用。","source":"ai","title":"由许多采用不同许可证的项目拼装而成"}],"platforms":[],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["games","hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[{"target":"world-of-warcraft","stance":"claims","evidence":"- World of Warcraft addon imports replace an addon's folder from a temporary copy, as [WowUp](https://github.com/WowUp/WowUp) does, and read flavour folders the way [WoWDeck](https://github.com/sigboe"}],"stars":19275,"downloads":1512237,"downloadsWeek":207074,"downloadsDays":7,"growth":1604,"growthDays":7,"history":[{"date":"2026-10-26","stars":14902},{"date":"2026-10-27","stars":15324},{"date":"2026-10-28","stars":15786},{"date":"2026-10-29","stars":16150},{"date":"2026-10-30","stars":16484},{"date":"2026-10-31","stars":16863},{"date":"2026-11-01","stars":17199},{"date":"2026-11-02","stars":17442},{"date":"2026-11-03","stars":17671},{"date":"2026-11-04","stars":17865},{"date":"2026-11-05","stars":18066},{"date":"2026-11-06","stars":18260},{"date":"2026-11-07","stars":18457},{"date":"2026-11-08","stars":18725},{"date":"2026-11-09","stars":19005},{"date":"2026-11-10","stars":19275}],"firstPublishedAt":"2026-10-27T01:10:12.623Z","commitSha":"4b8f63d84df03aae94f61bb770c781663cdfb115","verifiedAt":"2026-11-03T02:42:04.786Z","lastCommitAt":"2026-11-10T21:33:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android app that turns a phone or handheld into a Steam console, playing Windows PC games and opening a Linux desktop.","summary":"DroidDeck installs on an Android phone or gaming handheld and gives it something close to a Steam Deck: the Steam interface, Windows games running through Proton, a full Linux desktop, and shelves from GOG, Epic, Xbox and other stores. It also handles controllers, smart glasses and VR. You install the APK, let it download a Linux runtime, sign in to Steam and turn on one setting in Android's Developer options. It needs a Qualcomm Adreno 730 or newer graphics chip and about 4 GB free. How well each game runs depends on the device.","easeRationale":"It is an Android-only app, so no desktop system gets a grade. On Android you install an APK, download a Linux runtime of several gigabytes, sign in to Steam and switch on \"Disable child process restrictions\" in Developer options, and it only works on a narrow set of graphics chips: Adreno 730 or newer, with Mali, Xclipse and PowerVR unsupported.","costNote":"The app is free (GPL-3.0) with no ads. The games are not: you still buy them on Steam, GOG, Epic or Xbox, and the frame-generation feature needs your own paid copy of Lossless Scaling.","notVerified":["Whether your own device and your own games actually run, since the project says compatibility and performance vary by device","How much storage it really takes after the runtime, the desktop and the emulators are installed, beyond the 4 GB the README asks for","Whether the APK on the releases page was built from the code in this repository","Whether signing in to Steam, Epic, GOG, Xbox or Battle.net this way causes trouble with those accounts","Battery drain and heat on a phone during a long session"]},"zh-Hans":{"summaryShort":"一款 Android 应用，把手机或掌机变成一台 Steam 游戏机，运行 Windows 游戏并打开 Linux 桌面。","summary":"DroidDeck 安装在 Android 手机或掌上游戏机上，让它变得有点像 Steam 掌机：Steam 界面、通过 Proton 运行的 Windows 游戏、完整的 Linux 桌面，以及来自 GOG、Epic、Xbox 等商店的游戏库。它还支持手柄、智能眼镜和 VR。你安装 APK，让它下载一个 Linux 运行环境，登录 Steam，并在 Android 的开发者选项里打开一个设置。它需要高通 Adreno 730 或更新的图形芯片，以及大约 4 GB 的空闲空间。每款游戏运行得怎样取决于设备。","easeRationale":"这是一款仅支持 Android 的应用，所以没有任何桌面系统获得评分。在 Android 上，你需要安装一个 APK，下载几 GB 大小的 Linux 运行环境，登录 Steam，并在开发者选项中打开“Disable child process restrictions”，而且它只支持很窄的一部分图形芯片：Adreno 730 或更新型号，不支持 Mali、Xclipse 和 PowerVR。","costNote":"这款应用是免费的（GPL-3.0），没有广告。游戏本身不是免费的：你仍然需要在 Steam、GOG、Epic 或 Xbox 上购买，而帧生成功能需要你自己购买的 Lossless Scaling。","notVerified":["你自己的设备和游戏是否真的能运行，因为项目方说兼容性和性能因设备而异","运行环境、桌面和模拟器全部安装完之后，实际占用的存储空间有多少，这超出了 README 所要求的 4 GB","发布页面上的 APK 是否确实是用这个仓库里的代码编译出来的","用这种方式登录 Steam、Epic、GOG、Xbox 或 Battle.net 是否会给这些账号带来麻烦","长时间游玩时手机的耗电和发热情况"]}}},{"id":"R_kgDONhlNYg","slug":"duixcom__duix-avatar","name":"Duix Avatar","owner":"duixcom","fullName":"duixcom/Duix-Avatar","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"not_a_standard_open_source_licence","severity":"attention","evidence":"这个仓库自称\"Truly open-source AI avatar toolkit\"，但LICENSE文件其实是\"DUIX.COM COMMUNITY LICENSE AGREEMENT\"，GitHub无法将其识别为已知的许可证（NOASSERTION）。条款2规定，月活跃用户超过\"1 thousand\"后，\"you must request a commercial license from DUIX.COM, which DUIX.COM may grant to you in its sole discretion\"。另外还有一份单独的\"Duix.Avatar model community Licensing Agreement.pdf\"涉及模型本身。","source":"ai","title":"不是标准的开源许可证"},{"code":"licence_and_readme_give_different_limits","severity":"attention","evidence":"LICENSE把商业授权的门槛设在月活跃用户超过1,000人，而README里的对比表却写着\"enterprises with more than 100,000 users or annual revenue exceeding 10 million USD need to sign a commercial license agreement\"。打算用它生成的视频赚钱的人，光看这个仓库根本分不清到底哪个限制才是真的。","source":"ai","title":"许可证和README给出的限制不一样"},{"code":"company_may_reuse_what_you_build","severity":"attention","evidence":"LICENSE第5(c)条写道：\"By using the DUIX.COM Materials, you grant DUIX.COM a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license to use, reproduce, modify, publicly display, publicly perform, and distribute any case studies, examples, applications, or implementations created using the DUIX.COM Materials for any purpose, including ... marketing\"，而且不付钱给你，也不会说明是你做的。","source":"ai","title":"公司可以重新使用你做出来的东西"},{"code":"makes_videos_of_a_real_person","severity":"attention","evidence":"README写道，该工具\"can precisely clone your appearance and voice\"，数据来源是真实人物的视频。同一个工具也可以用来克隆别人的脸和声音，这就是深度伪造（deepfake）；未经同意使用他人的形象或声音，在很多国家都是违法的。","source":"ai","title":"会制作真实人物的视频"},{"code":"very_large_download_and_disk_use","severity":"attention","evidence":"README里Windows的系统要求写着：\"Must have D Drive ... More than 30GB\"、\"C Drive ... More than 100GB\"，服务端安装还提醒\"download will consume about 70GB of traffic, make sure to use WiFi\"。Radar给出的193 MB到578 MB的估算，只涵盖了那个小小的前端安装包。","source":"ai","title":"下载量和磁盘占用都非常大"},{"code":"service_images_come_from_a_separate_account","severity":"info","evidence":"README的依赖部分从Docker Hub拉取了\"guiji2025/fun-asr\"、\"guiji2025/fish-speech-ziming\"和\"guiji2025/duix.avatar\"。真正干活的那些重量级部分，是在和这个仓库不同的账号名下预先构建好的镜像，所以它们的内容并不在你能在这里看到的代码里。","source":"ai","title":"服务镜像来自另一个账号"},{"code":"quiet_for_months","severity":"info","evidence":"Radar的检查显示，最后一次提交是在2026-04-21（检查前196天），最后一个发行版v1.0.6是在2025-09-28发布的，还有423个未解决的issue。README把社区称为\"very active\"，更新\"frequent\"，但这已经和仓库的实际情况不符了。","source":"ai","title":"已经好几个月没有动静"}],"platforms":["windows","linux"],"tools":["docker","gpu_nvidia"],"buildTools":["node","ffmpeg"],"disk":{"minMB":193,"maxMB":578,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["video","audio","social_marketing"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":16295,"downloads":34387,"downloadsWeek":639,"downloadsDays":6,"growth":71,"growthDays":6,"history":[{"date":"2026-11-04","stars":16224},{"date":"2026-11-05","stars":16249},{"date":"2026-11-06","stars":16255},{"date":"2026-11-07","stars":16262},{"date":"2026-11-08","stars":16272},{"date":"2026-11-09","stars":16280},{"date":"2026-11-10","stars":16295}],"firstPublishedAt":"2026-11-04T01:36:50.362Z","commitSha":"1328feb5871448c8fa3d0e45b3bbc87e7c1d458a","verifiedAt":"2026-11-04T01:25:09.265Z","lastCommitAt":"2026-04-21T07:06:08Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a person's face and voice from a video, then makes videos of that avatar reading any text, all on your own computer.","summary":"Duix.Avatar makes talking-head videos. You feed it a video of a person, it builds an avatar of their face and a copy of their voice, and from then on you type text (English, Chinese, Japanese, Korean, French, German, Arabic or Spanish) and it produces a video of that avatar speaking, with the lips matched to the sound. Everything runs on your own machine with no internet. The price is heavy hardware: Windows 10 or Ubuntu 22.04, an NVIDIA card like an RTX 4070, 32 GB of memory, Docker, and about 70 GB of downloads plus over 100 GB of free disk.","easeRationale":"On Windows there is a ready installer for the front end, but it does nothing on its own: first you install WSL and Docker Desktop, run docker-compose in the project's /deploy folder and wait about half an hour for roughly 70 GB of service images, with an NVIDIA card and drivers in place. Ubuntu is the same work plus installing Docker, the graphics driver and the NVIDIA Container Toolkit by hand in the terminal.","costNote":"No charge to download or run. The licence is the company's own \"DUIX.COM Community License\", not a standard open-source one, and it requires a paid commercial licence above a user limit. The company also sells a hosted avatar API as the easier route.","notVerified":["How good the result looks: the project's own table calls the open-source lip sync a \"usable effect\" against the \"stunning and higher definition effect\" of its paid API.","Whether it runs at all on a card other than the tested RTX 4070 and 5090, and how long one video takes.","What is inside the three prebuilt Docker images, since they are downloaded ready-made.","Whether it really never contacts the internet while working, as the README claims.","The real disk use after training a model and keeping several avatars and videos."]},"zh-Hans":{"summaryShort":"从视频中复制一个人的脸和声音，然后让这个虚拟形象读出任何文字，全部在你自己的电脑上完成。","summary":"Duix.Avatar可以制作人物说话的视频。你给它一段某人的视频，它会生成这个人的脸部虚拟形象和声音的复制，之后你输入文字（支持英语、中文、日语、韩语、法语、德语、阿拉伯语或西班牙语），它就会生成这个虚拟形象说话的视频，嘴唇动作和声音对得上。整个过程都在你自己的电脑上运行，不需要联网。代价是对硬件要求很高：需要Windows 10或Ubuntu 22.04系统，一块像RTX 4070这样的NVIDIA显卡，32 GB内存，Docker，大约70 GB的下载量，以及超过100 GB的空闲磁盘空间。","easeRationale":"在Windows上有一个现成的前端安装程序，但它单独运行什么都做不了：你需要先安装WSL和Docker Desktop，在项目的/deploy文件夹里运行docker-compose，等大约半小时下载约70 GB的服务镜像，前提是已经装好NVIDIA显卡和驱动。在Ubuntu上工作量差不多，还要在终端里手动安装Docker、显卡驱动和NVIDIA Container Toolkit。","costNote":"下载和运行都不收费。许可证是公司自己制定的\"DUIX.COM Community License\"，不是标准的开源许可证，超过用户数量限制后需要购买商业许可证。公司还把托管版的虚拟形象API作为更简单的方案出售。","notVerified":["最终效果有多好：项目自己的对比表把开源版本的唇形同步称为\"usable effect\"，而把付费API称为\"stunning and higher definition effect\"。","除了测试过的RTX 4070和5090之外，它能否在其他显卡上运行，以及生成一段视频需要多长时间。","三个预先构建好的Docker镜像里到底是什么内容，因为它们是直接下载现成的。","它是否真的像README所说的那样，在运行时完全不联网。","训练一个模型并保留多个虚拟形象和视频后，实际占用的磁盘空间有多大。"]}}},{"id":"R_kgDOOXuApg","slug":"dyad-sh__dyad","name":"Dyad","owner":"dyad-sh","fullName":"dyad-sh/dyad","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"the_ai_runs_commands_on_your_computer","severity":"attention","evidence":"这款应用自带一个真正的终端,以及可以操控你电脑的工具:`node-pty`和`@xterm/xterm`(终端)、`dugite`(git)、`@vscode/ripgrep`(搜索你的文件)、`kill-port`和`tree-kill`(终止进程)、`ssh2`,还有`@ai-sdk/mcp`和`@modelcontextprotocol/sdk`,让模型可以调用外部工具。以这种方式构建应用意味着AI会在你的电脑上写入文件并执行命令,包括执行它自行选择的软件包的`npm install`。","source":"ai","title":"ai会在你的电脑上执行命令"},{"code":"a_component_that_reads_the_system_keychain","severity":"attention","evidence":"`package.json`里有`\"optionalDependencies\": { \"dyad-keychain-reader\": \"file:native/keychain-reader\" }`,构建步骤在打包前会运行`npm run rebuild:keychain-reader`。这是一段原生代码,专门用来读取Windows和macOS存放密码和密钥的位置。代码库没有说明它会读取什么内容,也没有说明何时读取。","source":"ai","title":"有一个组件会读取系统密钥链"},{"code":"usage_tracking_inside_an_app_sold_as_private","severity":"attention","evidence":"README列出的第一个特点是\"⚡️ Local: Fast, private and no lock-in\",但依赖项里却包含产品分析服务`posthog-js`,还有`stacktrace-js`,以及一个会把截图上传到制作者自有服务的错误报告流程(`VITE_IMAGE_UPLOAD_SERVICE_ORIGIN`,\"dyad-image-uploader\")。PostHog具体记录了什么,代码库中完全没有说明。","source":"ai","title":"一款宣称私密的应用内部却有使用情况追踪"},{"code":"part_of_the_code_is_not_open_source","severity":"info","evidence":"LICENSE文件开头写着\"All content that resides under the 'src/pro/' directory ... is licensed under the license defined in 'src/pro/LICENSE'\",README解释说这是\"Functional Source License 1.1\",这不是一种自由软件许可证。除此之外,`package.json`里声明的是`\"license\": \"MIT\"`,而LICENSE文件实际是Apache-2.0,这就是GitHub把许可证标记为NOASSERTION的原因。","source":"ai","title":"部分代码并非开源"},{"code":"your_requests_can_pass_through_the_makers_service","severity":"info","evidence":"这些脚本指向公司自己的一个中间层——`dev:engine`配合`DYAD_ENGINE_URL=http://localhost:8080/v1`、Google Cloud Run上的一个staging引擎,以及`.env.example`里的`DYAD_GATEWAY_URL`。使用付费计划时,提示词和代码可能会经过Dyad自己的服务,而不是直接发给模型提供方,这正好和\"bring your own keys\"的做法相反。","source":"ai","title":"你的请求可能会经过制作者自己的服务"},{"code":"it_updates_itself_and_moves_fast","severity":"info","evidence":"`update-electron-app`是一个依赖项,所以这款应用会自行获取并安装新版本,自2025年4月以来已经有156个发布版本,检查当天还发布了`v1.23.0-beta.1`,与稳定版`v1.22.0`并列,目前有358个未解决议题。","source":"ai","title":"会自动更新,发展很快"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":170,"maxMB":509,"basis":"installer"},"apiKeys":["An AI service of your choice: OpenAI, Anthropic, Google, xAI, Azure or Amazon Bedrock — or none if you point it at Ollama","GitHub, to push the app it builds (optional)","Supabase, Neon or Vercel, to give the app a database or put it online (optional)"],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["development","ai_agents","local_ai"],"alternativeTo":[{"slug":"v0-by-vercel","name":"v0 by Vercel","confirmed":true},{"slug":"lovable","name":"Lovable","confirmed":true},{"slug":"bolt-new","name":"Bolt.new","confirmed":true},{"slug":"replit","name":"Replit","confirmed":true},{"slug":"cursor","name":"Cursor","confirmed":true}],"declaredAlternatives":[],"stars":22258,"downloads":3997871,"downloadsWeek":37304,"downloadsDays":4,"growth":76,"growthDays":6,"history":[{"date":"2026-11-06","stars":22208},{"date":"2026-11-07","stars":22221},{"date":"2026-11-08","stars":22235},{"date":"2026-11-09","stars":22245},{"date":"2026-11-10","stars":22258}],"firstPublishedAt":"2026-11-05T22:43:48.375Z","commitSha":"85dcdefe5a8980f9bfcbff545d2e7cae60f3fe11","verifiedAt":"2026-11-05T22:34:35.355Z","lastCommitAt":"2026-11-10T20:09:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop program where you describe a web app and an AI builds it on your own computer, with your keys or a local model.","summary":"Dyad is a desktop program for building web apps by describing them. The AI writes the code, runs it and shows the result, all on your own machine, and the project stays in a folder you own. It works with your own keys for OpenAI, Anthropic, Google, xAI, Azure or Bedrock, or with models running locally through Ollama, and it can connect the finished app to GitHub, Supabase, Neon or Vercel. There are ready installers for Windows, Mac and Linux, and the makers also sell a paid plan with their own service in the middle.","easeRationale":"A B on all three: release v1.22.0 has ready files for every system — `dyad-1.22.0.Setup.exe` for Windows (170 MB), `.zip` builds for both kinds of Mac, and `.AppImage`, `.deb` and `.rpm` for Linux — and no sign-up. It is not an A because nothing happens until you paste in a model key or point it at a local Ollama, and Node.js on the machine is what runs the apps it writes.","costNote":"The app downloads without signing up and you can use it with your own paid model keys. There is also a paid Dyad subscription: the build scripts cover `start:out-of-quota` with `DYAD_SIMULATE_FREE_AGENT_QUOTA_EXCEEDED`, plus `payment_past_due`, `subscription_ending`, `subscription_paused` and `healthy` states. No price appears in the repository.","notVerified":["What the keychain-reader component reads, and whether it only looks for your own model keys","What `posthog-js` sends, and whether it can be switched off","What a project costs in model fees, which depends on the key you use and the model you pick","Whether the apps it generates run as shown, and what packages it installs along the way","How much disk a project takes after the AI installs its dependencies, beyond the 170 MB installer"]},"zh-Hans":{"summaryShort":"一款桌面程序,你描述想要的网页应用,AI就在你自己的电脑上把它构建出来,用你自己的密钥或本地模型。","summary":"Dyad是一款桌面程序,你只需描述想要的网页应用即可构建出来。AI会编写代码、运行并展示结果,全部在你自己的电脑上完成,项目保存在你自己拥有的文件夹里。它可以使用你自己的OpenAI、Anthropic、Google、xAI、Azure或Bedrock密钥,也可以通过Ollama使用本地运行的模型,还能把做好的应用连接到GitHub、Supabase、Neon或Vercel。目前已提供Windows、Mac和Linux的现成安装包,制作者还出售一个付费计划,其中加入了他们自己的中间服务。","easeRationale":"三大系统都是B级:v1.22.0版本为每个系统提供了现成文件——Windows版的`dyad-1.22.0.Setup.exe`(170MB)、两种Mac的`.zip`构建包,以及Linux的`.AppImage`、`.deb`和`.rpm`——而且不需要注册。没达到A级是因为在你粘贴模型密钥或指向本地Ollama之前什么都不会发生,而且是电脑上的Node.js在运行它写出来的应用。","costNote":"这款应用无需注册即可下载,你可以用自己付费的模型密钥来使用它。另外还有Dyad的付费订阅:构建脚本里包含`start:out-of-quota`和`DYAD_SIMULATE_FREE_AGENT_QUOTA_EXCEEDED`,以及`payment_past_due`、`subscription_ending`、`subscription_paused`和`healthy`这些状态。代码库里没有出现任何价格。","notVerified":["keychain-reader组件到底读取什么内容,以及它是否只查找你自己的模型密钥","`posthog-js`会发送什么数据,以及能否关闭它","一个项目在模型费用上要花多少钱,这取决于你使用的密钥和选择的模型","它生成的应用是否真的能像展示那样运行,以及在这个过程中会安装哪些软件包","除了170MB的安装包之外,AI安装完依赖之后一个项目实际占用多少磁盘空间"]}}},{"id":"R_kgDOPbFNkw","slug":"earendil-works__pi","name":"Pi","owner":"earendil-works","fullName":"earendil-works/pi","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接接入 shell 执行：curl -fsSL https://pi.dev/install.sh | sh","source":"check","title":"把 curl 直接接到 shell 执行"},{"code":"no_permission_system_by_default","severity":"attention","evidence":"README.md：“Pi does not include a built-in permission system for restricting filesystem, process, network, or credential access. By default, it runs with the permissions of the user and process that launched it.”README 给出的办法是让你自己做容器化或沙盒隔离。","source":"ai","title":"默认没有权限系统"},{"code":"runs_shell_commands_and_edits_files","severity":"attention","evidence":"Pi 是一个在项目文件夹里工作的代理，带有内置工具和 ! 命令（README.md，Permissions & Containerization），因此这个 AI 会在你的机器上修改文件、运行程序。","source":"ai","title":"运行 shell 命令并编辑文件"},{"code":"fetches_model_catalog_at_run_time","severity":"info","evidence":"README.md（Run with Nix）：“At runtime, Pi still overlays newer catalog data from pi.dev as usual”，而且这个工具会用 pi update 自我更新。","source":"ai","title":"运行时会获取模型目录"},{"code":"encourages_publishing_your_sessions","severity":"info","evidence":"README.md 邀请用户公开发布自己的编程会话（“Share your OSS coding agent sessions”，通过 badlogic/pi-share-hf 发布到 Hugging Face）。这是可选的，但一次会话可能包含你的代码和你输入过的任何内容。","source":"ai","title":"鼓励你公开发布自己的会话"},{"code":"new_contributor_reports_auto_closed","severity":"info","evidence":"README.md 顶部的提示：“New issues and PRs from new contributors are auto-closed by default. Maintainers review auto-closed issues daily.”目前有 316 个未解决的 issue。","source":"ai","title":"新贡献者提交的报告会被自动关闭"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":59,"maxMB":177,"basis":"installer"},"apiKeys":["An AI provider account or API key (OpenAI, Anthropic, Google and others)","Hugging Face (optional, only to publish your sessions)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","development"],"alternativeTo":[{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"claude-code","name":"Claude Code","confirmed":true}],"declaredAlternatives":[],"stars":121239,"downloads":1198702,"downloadsWeek":138026,"downloadsDays":7,"growth":1397,"growthDays":7,"history":[{"date":"2026-10-26","stars":118035},{"date":"2026-10-27","stars":118296},{"date":"2026-10-28","stars":118538},{"date":"2026-10-29","stars":118808},{"date":"2026-10-30","stars":119029},{"date":"2026-10-31","stars":119188},{"date":"2026-11-01","stars":119366},{"date":"2026-11-02","stars":119605},{"date":"2026-11-03","stars":119842},{"date":"2026-11-04","stars":120044},{"date":"2026-11-05","stars":120291},{"date":"2026-11-06","stars":120492},{"date":"2026-11-07","stars":120634},{"date":"2026-11-08","stars":120819},{"date":"2026-11-09","stars":121059},{"date":"2026-11-10","stars":121239}],"firstPublishedAt":"2026-10-27T01:10:12.624Z","commitSha":"9c9ce8547e5a88cbe834a53e9a6d83a789d08e47","verifiedAt":"2026-11-03T02:41:05.453Z","lastCommitAt":"2026-11-10T21:56:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free coding agent for the terminal that you point at your own AI account, plus parts to build your own agents.","summary":"Pi is a coding agent you run in a terminal: you open it in a project folder, connect an AI account or API key with /login, and give it tasks. It is deliberately small and meant to be changed, with extensions, skills, prompt templates and themes you can share as packages, and TypeScript libraries for people building their own agents on top of it. It installs with one command on Windows, macOS and Linux and needs Node.js 22.19 or newer. For developers; the AI provider is paid separately.","easeRationale":"On macOS and Linux it is one terminal command (`curl -fsSL https://pi.dev/install.sh | sh`, which can install Node.js for you), then `pi` in a project folder and `/login`. Windows is the same through PowerShell (`irm https://pi.dev/install.ps1 | iex`); everything happens in a terminal, with no window to click.","costNote":"Pi is MIT and free. You connect your own AI provider with `/login` (a subscription or an API key), and that is what costs money.","notVerified":["How well the agent performs, which depends mostly on the AI model you connect","What the install script does step by step before it is run","What your AI provider will charge for a day of use","What the runtime catalog overlay from pi.dev sends and receives","Whether the sandboxing patterns in the docs (micro-VM, Docker, OpenShell) really contain the agent"]},"zh-Hans":{"summaryShort":"一个免费的终端编程代理，接入你自己的 AI 账户，并附带用于搭建自己代理的组件。","summary":"Pi 是一个在终端里运行的编程代理：你在项目文件夹里打开它，用 /login 连接一个 AI 账户或 API 密钥，然后给它分配任务。它刻意做得很小，方便改动，提供扩展、技能（skill）、提示词模板和主题，可以打包分享，还提供 TypeScript 库，方便别人在它基础上搭建自己的代理。它在 Windows、macOS 和 Linux 上都只需一条命令安装，需要 Node.js 22.19 或更新版本。面向开发者；AI 服务商的费用另外单独支付。","easeRationale":"在 macOS 和 Linux 上只需一条终端命令（curl -fsSL https://pi.dev/install.sh | sh，它可以替你安装 Node.js），然后在项目文件夹里运行 pi 并执行 /login。Windows 上通过 PowerShell 同样简单（irm https://pi.dev/install.ps1 | iex）；一切都在终端里完成，没有任何窗口需要点击。","costNote":"Pi 使用 MIT 许可证，免费。你用 /login 连接自己的 AI 服务商（订阅或 API 密钥），这才是真正花钱的地方。","notVerified":["这个代理表现究竟如何，主要取决于你连接的 AI 模型","安装脚本在运行前到底逐步做了什么","你的 AI 服务商会为一天的使用收多少费用","来自 pi.dev 的运行时目录叠加层会发送和接收什么内容","文档里提到的沙盒方案（micro-VM、Docker、OpenShell）是否真的能把代理限制住"]}}},{"id":"R_kgDOUpsZ_w","slug":"ebony-vinyl__dsh-our-free-model","name":"dsh-our-free-model","owner":"Ebony-Vinyl","fullName":"Ebony-Vinyl/dsh-our-free-model","type":"needs_setup","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"C"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"38 天内获得 55891 个星标","source":"check","title":"新仓库，星标很多"},{"code":"sends_official_client_identity_to_the_free_gateway","severity":"high","evidence":"免费通道连接的是一些不要密钥、不收费就发放模型的池子：README 点名了 'OpenCode 的 Zen 网关' 和 'Kilo AI 的公共网关'，并就 Kilo 的那个说 '该池由上游免费提供'（即该池由上游免费提供）。README 自己的文件说明写出了插件为每次调用构造的内容：'src/upstream.js 网关身份：凭据、session/request id 铸造、工具指纹'（即网关身份：凭据、生成会话和请求 ID、工具指纹）。它还写下了 OpenCode 池对调用方的要求：'它要求每请求一枚铸造形状的会话 id、v2 客户端 User-Agent 与一组官方工具声明'（即每次请求都要一个生成形式的会话 ID、v2 客户端的 User-Agent 和一组官方工具声明）。这条通道被迫声明的四个工具名称，在问题表里被称为 '指纹诱饵四件套'（即四个指纹诱饵的组合），那里，从未注册过它们的转发客户端会报告 'Unknown tool bash/read/grep'。这些池子会把模型放给带着服务所期待的客户端身份的调用方。","source":"ai","title":"向免费网关发送官方客户端的身份"},{"code":"uses_other_companies_accounts_outside_their_own_apps","severity":"attention","evidence":"2.8 版新增 '十个账号渠道' —— LobsterAI、Qoder、TRAE、Cline、Loomy、Raccoon、MiniMax Code、ZCode 和 Gemini (Google Code Assist) —— 包含 '登录、账号池、每日积分领取'（即登录、账号池、每日积分领取），它们的设置页面名为 '白嫖模型接入'，字面意思是 \"freeloading model access\"。仓库还附带 `vendor/channel-pack/qoder-auth-wasm.wasm`，这是用于某个厂商登录的已编译代码。通过不属于这些服务的程序去使用它们，并自动领取它们的每日积分，可能违反其条款，并让你丢掉账号。","source":"ai","title":"在其他公司的应用之外使用它们的账号"},{"code":"credentials_hidden_inside_the_plugin","severity":"high","evidence":"README，'EAC channel'：'credentials and endpoints never appear in readable form in the plugin — they exist only as AES-256-GCM ciphertext, and the decryption key is derived at unlock time from three mask shards spread across two files'。文件列表中还附带 src/vault.js、src/vault-anchor.js 和 src/vault-data.js，以及 vendor/channel-pack/qoder-auth-wasm.wasm。","source":"ai","title":"凭据被藏在插件里"},{"code":"stars_required_to_unlock_a_channel","severity":"attention","evidence":"README：EAC 渠道 'only allows conversations after the server side checks your GitHub authorisation (log in and star this repository)'。确定性检查已经记录到 38 天内获得 55891 个星标，每个设置页面都带有 Star 按钮，新增的 '推荐项目' 页面还为同一作者的另外三个仓库显示了 Star 按钮。","source":"ai","title":"需要点星标才能解锁某个渠道"},{"code":"upstream_may_log_your_prompts","severity":"attention","evidence":"README，Kilo 渠道：'this pool is provided free by the upstream, and the upstream states on its model cards that prompts may be recorded to improve the service — please do not send sensitive content'。","source":"ai","title":"上游服务可能记录你输入的内容"},{"code":"local_gateway_keys_readable_by_other_programs","severity":"attention","evidence":"README 'Known limits'：在没有连接服务的组合方式下，'any process on the machine can obtain the forwarding key and the LAN key with one request, and then consume this machine's free quota from another machine'。局域网中继一旦开启，默认绑定到 0.0.0.0，README 还补充说，开启期间 'any user who can reach this host can use that key'。","source":"ai","title":"本地网关的密钥可被其他程序读取"},{"code":"modified_copy_circulating_under_the_same_name","severity":"attention","evidence":"README 的安装警告：'Do not install the bare-name dsh-our-free-model on npm. That package was not published by this repository... index.js and client.js are changed and it also contains src/channel-visibility.js, which this repository never had.'","source":"ai","title":"有被改动的副本以同样的名字流传"},{"code":"self_updating_code_and_pushed_announcements","severity":"attention","evidence":"插件会从一个订阅源下载并替换自己的文件，并热加载它们 ('check for updates -> one-click upgrade'、'src/updater.js')，维护者推送的公告会在一个轮询周期内到达每一个安装，其中 'urgent' 会显示全屏弹窗。README 补充说 'whoever holds the release private key can push any code'。","source":"ai","title":"代码会自行更新，并会推送公告"},{"code":"host_security_scanner_flagged_the_code","severity":"info","evidence":"README 的验收表：DSHEAC AIO 的静态扫描 'records one report line for the env+URL adjacency pattern in src/upstream.js (TROJAN_EXFIL_ENV)'，被描述为只做记录。README 还指出，上游厂商的名字只出现在仓库文档和公告中，从不出现在模型选择器或设置文字里。","source":"ai","title":"宿主程序的威胁扫描器标记了这段代码"},{"code":"mac_download_does_not_match_its_release_tag","severity":"info","evidence":"标记为 v2.8.0-macos-unofficial (2026-11-01) 的发布版里包含文件 `Our.Free.Model_2.7.1_aarch64.dmg`，与较早的 v2.7.4-macos-unofficial 发布版里的文件同名，而插件的 package.json 写的是 version 2.8.0。","source":"ai","title":"Mac 下载文件与其发布标签不符"}],"platforms":["windows","mac"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":25,"maxMB":75,"basis":"installer"},"apiKeys":["Optional: your own key for one of 76 providers (OpenRouter, Groq, Gemini, DeepSeek and others)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"claude-code","name":"Claude Code","confirmed":true}],"declaredAlternatives":[],"stars":68660,"downloads":90851,"downloadsWeek":24524,"downloadsDays":7,"growth":9689,"growthDays":7,"history":[{"date":"2026-10-25","stars":43086},{"date":"2026-10-26","stars":44851},{"date":"2026-10-27","stars":46798},{"date":"2026-10-28","stars":48915},{"date":"2026-10-29","stars":50917},{"date":"2026-10-30","stars":52806},{"date":"2026-10-31","stars":54362},{"date":"2026-11-01","stars":55871},{"date":"2026-11-02","stars":57303},{"date":"2026-11-03","stars":58971},{"date":"2026-11-04","stars":60637},{"date":"2026-11-05","stars":62239},{"date":"2026-11-06","stars":63883},{"date":"2026-11-07","stars":65363},{"date":"2026-11-08","stars":66128},{"date":"2026-11-09","stars":67319},{"date":"2026-11-10","stars":68660}],"firstPublishedAt":"2026-10-25T23:18:50.373Z","commitSha":"999e9f784ea3ea7a11cae383b142697b1ff2c42f","verifiedAt":"2026-11-02T02:55:32.386Z","lastCommitAt":"2026-11-10T12:45:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A plugin for the dsh AI app that offers frontier models with no account, no API key and no usage cap.","summary":"Our Free Model is a plugin for dsh (DeepSeek Harness) that adds frontier models, DeepSeek V4.1 Flash and Kimi K3 among them, with no login, sign-up or API key. It reaches them through other companies' public gateways (OpenCode and Kilo AI). Version 2.8 adds ten channels where you sign in to services such as TRAE, Qoder, Cline or Gemini and it collects their daily credits for you. It also opens an OpenAI- and Anthropic-compatible port so tools such as Claude Code can use the same models, shows a usage dashboard, and accepts your own keys for 76 providers.","easeRationale":"Windows is a B: the project publishes a desktop package whose README says it \"bundles a Node runtime and the standalone service, usable right after installing, with no login and no API key\", though you still pick channels in its settings. macOS is a C, because the .dmg is a community build with no Apple signature and the README's own steps include Terminal commands such as `sudo xattr -rd com.apple.quarantine` to open it. Installing it as a dsh plugin instead needs the dsh host and commands like `dsh plugin --profile web add /path`.","costNote":"No payment and no API key. The README says the EAC channel only answers after the server checks that you logged in with GitHub and starred this repository.","notVerified":["Whether the free gateways keep working, or block this client at any time","What the upstream services keep of the prompts you send","What the encrypted credentials inside the plugin are and who they belong to","Whether a service closes your account for signing in to it through one of the ten channels","Whether the Windows and macOS packages were built from the source in this repository","The real speed and the point at which requests start coming back as 429"]},"zh-Hans":{"summaryShort":"dsh AI 应用的一个插件，提供无需账号、无需 API 密钥、没有使用上限的顶级模型。","summary":"Our Free Model 是 dsh（DeepSeek Harness）的一个插件，它加入了顶级模型，包括 DeepSeek V4.1 Flash 和 Kimi K3，无需登录、注册或 API 密钥。它通过其他公司的公共网关（OpenCode 和 Kilo AI）连接这些模型。2.8 版新增十个渠道，你在 TRAE、Qoder、Cline 或 Gemini 等服务中登录后，它会替你领取这些服务的每日积分。它还会开放一个兼容 OpenAI 和 Anthropic 的端口，让 Claude Code 等工具使用同样的模型，显示用量面板，并接受你自己的 76 个服务商的密钥。","easeRationale":"Windows 是 B：项目发布了桌面安装包，其 README 说它 \"bundles a Node runtime and the standalone service, usable right after installing, with no login and no API key\"，不过你仍需在设置里选择渠道。macOS 是 C，因为 .dmg 是社区构建，没有 Apple 签名，而 README 自己的步骤里包含 `sudo xattr -rd com.apple.quarantine` 这样的终端命令才能打开它。改为作为 dsh 插件安装则需要 dsh 主程序，以及 `dsh plugin --profile web add /path` 这样的命令。","costNote":"无需付费，也无需 API 密钥。README 说，EAC 渠道只有在服务器确认你已用 GitHub 登录并给本仓库点了星标之后才会应答。","notVerified":["这些免费网关是否会一直可用，还是随时会屏蔽这个客户端","上游服务会保留你发送的哪些内容","插件内部那些加密的凭据是什么，属于谁","通过十个渠道之一登录某个服务，会不会导致该服务封掉你的账号","Windows 和 macOS 的安装包是否由本仓库的源代码构建","真实速度，以及请求开始返回 429 的临界点"]}}},{"id":"R_kgDOL5gm3g","slug":"eddieoz__openxrypt","name":"OpenXrypt","owner":"eddieoz","fullName":"eddieoz/openxrypt","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2024-07-01(859 天前)","source":"check","title":"版本更新停滞"},{"code":"features_contradicted_by_own_limitations","severity":"attention","evidence":"主要功能列表宣称支持 \"Group Messages Encryption\" 和 \"Timeline Encryption\",但同一份 README 的 Known limitations 部分却写着 \"Can't encrypt messages for a group yet\"、\"Can't encrypt messages in timeline\" 和 \"Limited use of cryptography through opengpg yet\"。它也不会给图片或表情符号加密。","source":"ai","title":"功能介绍与自述的局限相矛盾"},{"code":"private_key_pasted_into_the_browser","severity":"attention","evidence":"设置步骤要求你把 \"paste your private key\" 粘贴进扩展程序,密钥会 \"locally in your browser's storage\" 保存,常见问题里写着 \"The passphrase is securely stored in session storage\"。README 本身建议创建 \"a new private key for messaging purposes\",而不是使用你已有的密钥。任何能访问该浏览器账户的程序都能接触到这些密钥。","source":"ai","title":"私钥粘贴进浏览器"},{"code":"install_outside_the_chrome_store","severity":"attention","evidence":"安装需要从克隆下来的文件夹里启用 \"Developer mode\" 和 \"Load unpacked\";13 次发布中都没有附带打包好的文件。因此这个扩展程序不会经过 Chrome 网上应用店的审核,也不会自动更新,每次改动你都得手动重新下载。","source":"ai","title":"需在 chrome 应用商店之外安装"},{"code":"no_independent_security_review","severity":"info","evidence":"README 展示了端到端加密和带有 \"SHA-256 hash of your public key's fingerprint\" 的 \"Timeline Obfuscation\",但没有提到任何审计、威胁模型,也没有漏洞联系文件;仓库顶层文件中没有 SECURITY.md。","source":"ai","title":"没有独立的第三方审查"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system","productivity"],"alternativeTo":[],"declaredAlternatives":[],"stars":40,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":40},{"date":"2026-11-09","stars":40},{"date":"2026-11-10","stars":40}],"firstPublishedAt":"2026-11-07T22:56:16.744Z","commitSha":"3558cb48e3fd701effe4c6c7b26037bdd3e6aa74","verifiedAt":"2026-11-07T22:40:12.977Z","lastCommitAt":"2026-05-18T08:23:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Chrome add-on that scrambles your X (Twitter) and WhatsApp Web messages with PGP so only your contact can read them.","summary":"A Chrome extension for people who want their direct messages on X (Twitter) and WhatsApp Web to be unreadable to anyone else. You paste your contacts' PGP public keys and your own private key into the extension, write a message, press Encrypt, and send the scrambled text; messages you receive are turned back into readable text automatically. There is no store listing: you download the folder and load it in Chrome's developer mode, and you need a separate program such as Kleopatra or GPGTools to create your keys first.","easeRationale":"Same work on Windows, macOS and Linux (C): there is no Chrome Web Store listing, so you download or clone the folder, open chrome://extensions/, turn on Developer mode, use \"Load unpacked\" and restart the browser. Before that you have to create PGP keys in another program (the README suggests Kleopatra on Windows and Linux, GPGTools on Mac) and paste your private key into the extension.","costNote":"No price mentioned; the README asks only for optional tips through Buy Me a Coffee or Lightning.","notVerified":["Whether the encryption and key handling are actually implemented correctly, since nobody has reviewed them publicly.","Whether it still works with the current X and WhatsApp Web pages, which change often and which the extension reads directly.","Whether the private key and passphrase really stay only on your machine while you browse.","Whether X or WhatsApp treat posting or sending scrambled text as a rule violation on your account."]},"zh-Hans":{"summaryShort":"一款 Chrome 扩展程序,用 PGP 给你在 X(推特)和 WhatsApp Web 上的消息加密,让只有你的联系人才能读懂。","summary":"一款面向希望在 X(推特)和 WhatsApp Web 上的私信对其他任何人都不可读的用户的 Chrome 扩展程序。你把联系人的 PGP 公钥和自己的私钥粘贴到扩展程序里,写好消息,点击 Encrypt,发送加密后的文本;你收到的消息会自动还原成可读文本。它没有应用商店页面:你需要下载文件夹,在 Chrome 的开发者模式下加载它,而且需要先用 Kleopatra 或 GPGTools 等单独的程序生成密钥。","easeRationale":"在 Windows、macOS 和 Linux 上操作相同,属于 C 级:它没有 Chrome 网上应用店的页面,所以你需要下载或克隆文件夹,打开 chrome://extensions/,开启开发者模式,使用 \"Load unpacked\" 并重启浏览器。在此之前,你要先在另一个程序里生成 PGP 密钥(README 建议 Windows 和 Linux 用 Kleopatra,Mac 用 GPGTools),再把私钥粘贴进扩展程序。","costNote":"未提及价格;README 只是通过 Buy Me a Coffee 或 Lightning 征求可选的打赏。","notVerified":["加密和密钥处理是否真的实现正确,因为还没有人公开审查过。","它是否仍能在当前的 X 和 WhatsApp Web 页面上正常工作,因为这些页面经常变化,而扩展程序是直接读取它们的。","私钥和密码短语在你浏览时是否真的只留在你自己的电脑上。","X 或 WhatsApp 是否会把发布或发送加密文本视为违反账号规则。"]}}},{"id":"R_kgDOMDdSEg","slug":"electricalexis__notagen","name":"NotaGen","owner":"ElectricAlexis","fullName":"ElectricAlexis/NotaGen","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README链接到https://drive.google.com/drive/folders/1iVLkcywzXGcHFodce9nDQyEmK4UDmBtY?usp=sharing","source":"check","title":"外部下载链接"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):transformers:GHSA-29pf-2h5f-8g72、GHSA-37mw-44qp-f5jm、GHSA-4w7r-h757-3r74;gradio:GHSA-39mp-8hj3-5c49、GHSA-5cpq-9538-jm2j、GHSA-6655-8ph2-63j3","source":"check","title":"已知漏洞"},{"code":"the_model_file_comes_from_elsewhere","severity":"attention","evidence":"真正用来生成音乐的内容并不在代码库中:README链接了Hugging Face上的五个`.pth`文件(`weights_notagen_pretrain...`、`weights_notagenx_p_size_16...`),还有另外两个来自不同账号、用于CLaMP 2的文件(`huggingface.co/sander-wood/clamp2`)。PyTorch的`.pth`文件是通过解包保存的Python对象来加载的,这本身就是一种执行代码的方式,所以你下载的内容决定了你电脑上会发生什么。检查还标记出一个指向Google Drive的数据集链接。","source":"ai","title":"模型文件来自别处"},{"code":"where_the_music_it_learned_from_came_from_is_not_said","severity":"attention","evidence":"项目概述中写着\"Pre-training on 1.6M musical pieces\"和\"Fine-tuning on ~9K classical compositions\",却没有指明来源或许可条款;只有那个小型示例数据集被标注了出处(来自OpenScore Lieder的舒伯特歌曲)。项目自己的过滤器也表明模型可能会重复它学到的内容:在`RL/data.py`里,\"Sheets that may plagiarize from ground truth data (ld_sim>0.95) are excluded\"。它写出的乐谱归谁所有,仍是一个悬而未决的问题。","source":"ai","title":"没有说明它学习所用音乐的来源"},{"code":"a_strong_graphics_card_is_the_real_requirement","severity":"info","evidence":"README写道:\"Deploying NotaGen-X inference locally may require 8GB of GPU memory\",而训练方面\"The use of NotaGen-large requires at least 24GB of GPU memory for training and inference\",较小的模型被作为替代方案提供。安装命令锁定了`pytorch==2.3.0 pytorch-cuda=11.8`,所以必须用Nvidia显卡,没有别的选择。","source":"ai","title":"真正的要求是一块强力显卡"},{"code":"training_reports_to_an_online_service","severity":"info","evidence":"`requirements.txt`锁定了`wandb==0.17.2`,也就是Weights & Biases服务,预训练、微调和强化学习脚本都会用到它。如果你用自己的音乐进行训练,除非手动关闭,这次运行的数据就会发送到该服务。","source":"ai","title":"训练过程会上报给一个在线服务"},{"code":"old_pinned_libraries_and_no_commits_for_a_year_and_a_half","severity":"info","evidence":"最后一次提交是在2025-04-21,距本次检查563天,有26个未解决议题,也没有发布版本。`requirements.txt`锁定了`transformers==4.40.0`、`gradio==5.17.1`和`numpy==1.26.4`——正是漏洞检查标记出的版本——所以当前的Python环境可能无法复现它。","source":"ai","title":"依赖版本老旧,一年半没有提交"},{"code":"research_code_configured_by_editing_files","severity":"info","evidence":"没有任何设置界面:使用说明要求你编辑`pretrain/config.py`、`finetune/config.py`、`inference/config.py`、`RL/config.py`、`clamp2/extract_clamp2.py`和`RL/data.py`,并把像`weights_notagen_schubert_p_size_16_p_length_1024_p_layers_20_c_layers_6_h_size_1280_lr_1e-05_batch_1.pth`这样又长的权重文件名分别粘贴进去。最方便使用的版本——一个Colab笔记本和一个ComfyUI节点——都是其他人的贡献,保存在他们自己的代码库里。","source":"ai","title":"需要通过编辑文件来配置的研究代码"}],"platforms":["windows","linux","web"],"tools":[],"buildTools":["python"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":1246,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":1245},{"date":"2026-11-07","stars":1245},{"date":"2026-11-08","stars":1246},{"date":"2026-11-09","stars":1246},{"date":"2026-11-10","stars":1246}],"firstPublishedAt":"2026-11-05T22:43:48.384Z","commitSha":"4553cd743b6b449732baf2a6cc58885fe853b11d","verifiedAt":"2026-11-05T22:35:34.580Z","lastCommitAt":"2025-04-21T04:41:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Writes classical sheet music: you name a period, a composer and the instruments, and it composes a score you can hear and save.","summary":"NotaGen is a research model that composes classical sheet music. You give it a period, a composer's style and the instruments, and it writes the score in ABC notation, which can be turned into MusicXML, PDF or audio. The quickest way to try it is the makers' page on Hugging Face, where you type the prompt and listen. Running it on your own computer means Python with conda, PyTorch with CUDA, downloading the model file by hand and about 8 GB of graphics memory. The code also includes everything used to train and fine-tune it on your own music.","easeRationale":"On the web it is an A: the makers' own page on Hugging Face Space runs NotaGen-X, you type \"Period-Composer-Instrumentation\" and \"preview the audio / pdf scores, and download them\", with nothing installed. On your own Windows or Linux machine it is a D, with no release to download: `conda create --name notagen python=3.10`, PyTorch 2.3.0 with CUDA 11.8, four `pip install` steps, the model file fetched by hand from Hugging Face and roughly 8 GB of graphics memory, following a second set of instructions in `gradio/README.md`. Macs are not covered, since the install line asks for an Nvidia card.","costNote":"MIT code, model files published openly on Hugging Face, and the online demo on Hugging Face Space asks for nothing.","notVerified":["Whether the install still works, with PyTorch 2.3.0, CUDA 11.8 and the pinned libraries","How long one score takes on an 8 GB card, and how much disk the model files take","Whether the music is pleasant or useful, which only listening can tell","Who owns the scores it writes, which depends on training data the repository does not describe","What the online demo on Hugging Face keeps of the prompts you type"]},"zh-Hans":{"summaryShort":"创作古典乐谱:你给出一个时期、一位作曲家和乐器组合,它就能谱出一段可以试听和保存的曲子。","summary":"NotaGen是一个用于作曲的研究模型,能创作古典乐谱。你给它一个时期、一位作曲家的风格和乐器组合,它就会用ABC记谱法写出乐谱,这种格式可以转换成MusicXML、PDF或音频。最快的体验方式是制作者在Hugging Face上的页面,在那里输入提示词就能试听。要在自己的电脑上运行,需要带conda的Python、带CUDA的PyTorch、手动下载模型文件,以及大约8GB的显存。代码里还包含了用自己的音乐进行训练和微调所需的全部内容。","easeRationale":"在网页上是A级:制作者自己在Hugging Face Space上的页面运行着NotaGen-X,你输入\"Period-Composer-Instrumentation\",就能\"preview the audio / pdf scores, and download them\",什么都不用安装。在自己的Windows或Linux电脑上则是D级,没有发布版本可下载:`conda create --name notagen python=3.10`、带CUDA 11.8的PyTorch 2.3.0、四个`pip install`步骤、需要手动从Hugging Face获取的模型文件,以及大约8GB的显存,还要按照`gradio/README.md`里另一套说明操作。Mac不在支持范围内,因为安装步骤要求使用Nvidia显卡。","costNote":"代码采用MIT许可,模型文件在Hugging Face上公开发布,Hugging Face Space上的在线演示不需要任何条件。","notVerified":["按PyTorch 2.3.0、CUDA 11.8和这些锁定版本的库,安装是否仍然有效","在8GB显卡上生成一首曲谱要多久,模型文件又占用多少磁盘空间","这些音乐是否好听或有用,这只有听过才知道","它写出的乐谱归谁所有,这取决于代码库中并未说明的训练数据","Hugging Face上的在线演示会保留你输入的提示词中的哪些内容"]}}},{"id":"R_kgDOVQMnMg","slug":"ely-open-source__elyoffice","name":"Ely Office","owner":"ELY-Open-Source","fullName":"ELY-Open-Source/ElyOffice","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 14 天前","source":"check","title":"所有者账号很新"},{"code":"mit_label_does_not_cover_the_engines_inside","severity":"attention","evidence":"这些应用运行在别人的引擎上：README 的表格里列出 ElyDoc、ElyTable 和 ElyShow 使用“LibreOfficeKit”，ElyPDF 使用“MuPDF, Tesseract”，ElyTree 使用“Git”，构建步骤则是“patched LibreOffice 26.2.6.3”和“patched MuPDF 1.28.5”。许可证部分只写着：“MIT (LICENSE). Third-party components keep their own licences; see docs/licenses.md.”其中一些条款比 MIT 更严格，如果你要再分发或打包这些应用，这一点很重要。","source":"ai","title":"mit 标签并不涉及内部引擎"},{"code":"very_early_version","severity":"info","evidence":"这个仓库创建于 2026-10-19，经过 12 次发布后仍是 0.1.13 版本。应用表格里列出“ElyAgent | Second release”，说明套件里有一部分还不存在。","source":"ai","title":"版本还很早期"},{"code":"only_the_mac_builds_are_signed","severity":"info","evidence":"Release 部分写着：“macOS builds are signed and notarized”，但完全没有提到 Windows 或 Linux 的安装包。可以预期 Windows 会提示发布者未知。","source":"ai","title":"只有 mac 的构建经过签名"},{"code":"it_can_still_use_the_network","severity":"info","evidence":"README 开头写着“Files stay in your folders; no account, no cloud service”，但 Cargo.toml 中包含 HTTP 客户端“ureq = { version = \\\"3\\\", features = [\\\"socks-proxy\\\"] }”，发布流程还会上传“update manifests”和安装包一起。ElyMail 当然也要和你的邮件服务器通信。","source":"ai","title":"依然可以使用网络"},{"code":"mail_sign_in_uses_an_app_password","severity":"info","evidence":"README：“ElyMail signs in to Microsoft with OAuth, and to Gmail and iCloud with an app password.”应用专用密码会让程序完全访问那个邮箱，直到你撤销它为止。","source":"ai","title":"邮件登录使用应用专用密码"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":509,"maxMB":1526,"basis":"installer"},"apiKeys":["Email provider app password (Gmail, iCloud)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","productivity","development"],"alternativeTo":[{"slug":"word","name":"Word","confirmed":true},{"slug":"excel","name":"Excel","confirmed":true},{"slug":"powerpoint","name":"PowerPoint","confirmed":true},{"slug":"acrobat","name":"Acrobat","confirmed":true},{"slug":"microsoft-outlook","name":"Outlook","confirmed":true},{"slug":"gitkraken","name":"GitKraken","confirmed":true}],"declaredAlternatives":[],"stars":1455,"downloads":26929,"downloadsWeek":6926,"downloadsDays":7,"growth":260,"growthDays":7,"history":[{"date":"2026-10-27","stars":1067},{"date":"2026-10-28","stars":1087},{"date":"2026-10-29","stars":1101},{"date":"2026-10-30","stars":1124},{"date":"2026-10-31","stars":1146},{"date":"2026-11-01","stars":1169},{"date":"2026-11-02","stars":1184},{"date":"2026-11-03","stars":1195},{"date":"2026-11-04","stars":1211},{"date":"2026-11-05","stars":1222},{"date":"2026-11-06","stars":1232},{"date":"2026-11-07","stars":1330},{"date":"2026-11-08","stars":1395},{"date":"2026-11-09","stars":1434},{"date":"2026-11-10","stars":1455}],"firstPublishedAt":"2026-10-27T18:40:27.880Z","commitSha":"efa5dc1a8e52d7f40942afc35024aece4fff7aec","verifiedAt":"2026-11-03T02:43:00.252Z","lastCommitAt":"2026-11-10T22:19:25Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"ElyOffice-0.1.15-windows-x86_64-setup.exe","sizeBytes":498764355,"url":"https://github.com/ELY-Open-Source/ElyOffice/releases/download/v0.1.15/ElyOffice-0.1.15-windows-x86_64-setup.exe","releaseUrl":"https://github.com/ELY-Open-Source/ElyOffice/releases/tag/v0.1.15"},{"os":"mac","file":"ElyOffice-0.1.15-arm64.dmg","sizeBytes":867693945,"url":"https://github.com/ELY-Open-Source/ElyOffice/releases/download/v0.1.15/ElyOffice-0.1.15-arm64.dmg","releaseUrl":"https://github.com/ELY-Open-Source/ElyOffice/releases/tag/v0.1.15"},{"os":"linux","file":"ElyDoc-0.1.15-linux-x86_64.AppImage","sizeBytes":181758456,"url":"https://github.com/ELY-Open-Source/ElyOffice/releases/download/v0.1.15/ElyDoc-0.1.15-linux-x86_64.AppImage","releaseUrl":"https://github.com/ELY-Open-Source/ElyOffice/releases/tag/v0.1.15"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A set of desktop apps for documents, spreadsheets, slides, PDFs and email that keeps your files on your own computer.","summary":"Ely Office is a family of separate desktop programs: ElyDoc for text documents (DOCX, ODT), ElyTable for spreadsheets (XLSX, ODS, CSV), ElyShow for presentations (PPTX, ODP), ElyPDF for reading and editing PDFs with text recognition and signatures, ElyMail for email over IMAP, POP3 and SMTP, ElyTree for Git and ElyMark for Markdown. There is no account and no cloud; your files stay in your folders. Installers exist for Windows, Mac and Linux, either as the whole suite or one app at a time. It is still at version 0.1.13.","easeRationale":"Every system gets a ready installer with the engines already inside: a 509 MB setup.exe on Windows, a signed and notarized DMG of about 850 MB on Mac, and AppImage, deb, rpm, Arch and Nix packages on Linux. You can also install a single app, such as the 11 MB ElyMail on Windows, instead of the whole suite.","notVerified":["Whether Word, Excel and PowerPoint files open and save again without losing layout","Whether the installers on the releases page were built from the code in this repository","How the apps behave on a long document or a large spreadsheet","Whether ElyMail connects to your own mail provider and keeps your mail in step","How much disk each app uses once installed, beyond the size of the download"]},"zh-Hans":{"summaryShort":"一套桌面应用，涵盖文档、表格、幻灯片、PDF 和邮件，文件始终保存在你自己的电脑上。","summary":"Ely Office 是一组相互独立的桌面程序：ElyDoc 处理文本文档（DOCX、ODT）、ElyTable 处理电子表格（XLSX、ODS、CSV）、ElyShow 处理演示文稿（PPTX、ODP）、ElyPDF 用于阅读和编辑 PDF，支持文字识别和签名、ElyMail 通过 IMAP、POP3 和 SMTP 处理邮件、ElyTree 处理 Git、ElyMark 处理 Markdown。没有账户，也没有云端；你的文件保留在自己的文件夹里。Windows、Mac 和 Linux 都有安装包，可以整套安装，也可以只装一个应用。目前版本仍是 0.1.13。","easeRationale":"每个系统都有现成的安装包，内置所需引擎：Windows 上是一个 509 MB 的 setup.exe，Mac 上是一个经过签名和公证的约 850 MB 的 DMG，Linux 上有 AppImage、deb、rpm、Arch 和 Nix 包。你也可以只安装单个应用，比如 Windows 上 11 MB 的 ElyMail，而不是整套软件。","notVerified":["Word、Excel 和 PowerPoint 文件是否能打开并再次保存而不丢失排版","发布页面上的安装包是否确实是用这个仓库里的代码编译出来的","这些应用在处理长文档或大型表格时表现如何","ElyMail 是否能连接你自己的邮件服务商，并让邮件保持同步","安装之后每个应用实际占用多少磁盘空间，超出下载包本身的大小"]}}},{"id":"R_kgDOVUt7kg","slug":"enesozturk__dimensional-drawing","name":"Dimensional Drawing","owner":"enesozturk","fullName":"enesozturk/dimensional-drawing","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"low","signals":[{"code":"tuned_for_claude_only","severity":"info","evidence":"README 说明：“Optimized for Claude Motion. The headless renderer is tested with Playwright (Chromium) + ffmpeg. Other animation libraries (Remotion, Motion Canvas, GSAP, Lottie…) are not tested.”动画那一半在其他助手上可能表现不同。","source":"ai","title":"只针对 claude 做过调优"},{"code":"your_component_code_goes_to_the_ai","severity":"info","evidence":"README，“Use it”：“Ask for a drawing and point at the source (a repo file, a Storybook story, a URL, a screenshot)”，示例是“Make a dimensional drawing of the Card component in src/components/Card.tsx”。该文件的内容会被读取并发送给你使用的模型。","source":"ai","title":"你的组件代码会被发给 ai"},{"code":"no_releases_and_quiet_since_the_first_day","severity":"info","evidence":"这个仓库在 GitHub 上有 0 个 release，创建于 2026-10-23，最后一次提交也是同一天，但已经有 1750 个 star。没有版本可以固定使用。","source":"ai","title":"没有发布版本，从第一天起就很安静"},{"code":"local_video_render_downloads_a_browser","severity":"info","evidence":"README，“Make it move”：“cd skills/dimensional-drawing-motion/adapters/headless && npm install”，然后是一个渲染命令。这次安装会带来 Playwright，它会下载自己的 Chromium 副本。","source":"ai","title":"本地视频渲染会下载一个浏览器"}],"platforms":["windows","mac","linux","web"],"tools":["other","node","ffmpeg"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","design_cad","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":2038,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":264,"growthDays":7,"history":[{"date":"2026-10-27","stars":1252},{"date":"2026-10-28","stars":1398},{"date":"2026-10-29","stars":1566},{"date":"2026-10-30","stars":1641},{"date":"2026-10-31","stars":1681},{"date":"2026-11-01","stars":1722},{"date":"2026-11-02","stars":1747},{"date":"2026-11-03","stars":1774},{"date":"2026-11-04","stars":1797},{"date":"2026-11-05","stars":1829},{"date":"2026-11-06","stars":1893},{"date":"2026-11-07","stars":1921},{"date":"2026-11-08","stars":1952},{"date":"2026-11-09","stars":1994},{"date":"2026-11-10","stars":2038}],"firstPublishedAt":"2026-10-27T18:40:27.889Z","commitSha":"5c457437c8c83524308aad3b7443e81adcc13153","verifiedAt":"2026-11-03T02:43:40.734Z","lastCommitAt":"2026-10-23T14:17:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill that has your AI assistant draw an Apple-style technical sheet of a screen component, with every spacing labeled.","summary":"This is a pair of instruction packs you add to an AI assistant. You point it at a button, card or dialog in your code, a Storybook story, a web page or a screenshot, and it produces a one-page technical drawing in the style Apple publishes for its devices: plan view, exploded layers, close-ups of corners, and tables of the spacings, gaps and corner radii as named variables instead of fixed numbers. It leaves out the text on purpose. A second skill animates the finished sheet into a short film, built for Claude's Motion artifact or rendered locally with Node and ffmpeg.","easeRationale":"The easiest route is claude.ai or the Claude desktop app: zip the skill folder and add it under Settings, Capabilities, Skills. On Windows, Mac and Linux you need an AI coding assistant installed first, then clone the repository and copy the folders into its skills directory, or run \"npx skills add\" with Node.js. The optional video render needs Node 18 or later and ffmpeg.","costNote":"The skills are free (MIT). They only work inside an AI assistant, which usually needs a paid plan or paid API usage.","notVerified":["Whether the drawing it produces matches your component's real spacings and corner radii","Whether it works on an assistant other than Claude, which is the only one the author tuned","How much your AI assistant charges in tokens for one sheet or one film","Whether the local video render works on your machine after Playwright and ffmpeg are in place"]},"zh-Hans":{"summaryShort":"一个让你的 AI 助手画出苹果风格技术图纸的技能，标注屏幕组件的每一处间距。","summary":"这其实是一对加入 AI 助手的指令包。你把它指向代码里的一个按钮、卡片或对话框，一个 Storybook story、一个网页或一张截图，它就会生成一张苹果为其设备发布的那种风格的单页技术图纸：平面视图、分解的图层、角部特写，以及把间距、留白和圆角用命名变量而不是固定数字表示的表格。它有意不显示文字内容。第二个技能会把做好的图纸变成一段短片，专为 Claude 的 Motion artifact 制作，也可以用 Node 和 ffmpeg 在本地渲染。","easeRationale":"最简单的方式是在 claude.ai 或 Claude 桌面应用里：把技能文件夹打包成 zip，添加到设置、能力、技能里。在 Windows、Mac 和 Linux 上，你需要先安装一个 AI 编程助手，然后克隆这个仓库，把文件夹复制到它的技能目录，或者用 Node.js 运行“npx skills add”。可选的视频渲染需要 Node 18 或更新版本以及 ffmpeg。","costNote":"这些技能是免费的（MIT）。它们只能在 AI 助手里运行，而助手通常需要付费计划或付费的 API 使用量。","notVerified":["它生成的图纸是否符合你组件真实的间距和圆角","它是否能在 Claude 以外的助手上运行，因为作者只针对 Claude 做过调优","你的 AI 助手为一张图纸或一段视频收多少 token 费用","装好 Playwright 和 ffmpeg 之后，本地视频渲染是否能在你的机器上正常运行"]}}},{"id":"R_kgDORy4B9g","slug":"ericosiu__ai-marketing-skills","name":"AI Marketing Skills","owner":"ericosiu","fullName":"ericosiu/ai-marketing-skills","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"automated_cold_outreach_and_visitor_tracking","severity":"attention","evidence":"Outbound Engine 被描述为 \"ICP definition to emails in inbox — fully automated\",Sales Pipeline 承诺 \"Turn anonymous website visitors into qualified pipeline\",并配有一个 \"RB2B Router\" 来完成 \"intent scoring, seniority-based company dedup, and agency classification before routing to outbound sequences\",\"Deal Resurrector\" 还包括 \"tracking departed contacts to their new companies\"。在很多国家,大量发送陌生开发邮件和识别浏览过你网站的人都受到反垃圾邮件和数据保护法律的限制,包括欧洲的 GDPR 和巴西的 LGPD。","source":"ai","title":"自动化陌生开发和访客追踪"},{"code":"skills_can_publish_and_send_on_your_behalf","severity":"attention","evidence":"一些技能把真实账号和渠道的钥匙交给了 AI 代理:\"Shortform Production ... deliver approved clips through the API | Metricool Delivery\"、\"Deck Generator ... Google Slides Builder\",增长引擎运行的实验会 \"run, measure, and optimize themselves\"。一旦出错,内容会直接发布或发送给真实的人,而不是存在草稿箱里。","source":"ai","title":"技能可以代表你发布和发送内容"},{"code":"repository_promotes_the_vendor_services","severity":"info","evidence":"README 被包裹在该代理机构的营销漏斗里:指向 singlebrain.com 和 singlegrain.com 的链接都带有 \"utm_source=github&utm_medium=skill_repo&utm_campaign=ai_marketing_skills\",还有 \"Want these built and managed for you?\" 以及一个订阅通讯的入口。像 \"battle-tested on real pipelines generating millions in revenue\" 这样的说法来自供应商自己。","source":"ai","title":"仓库在为供应商的服务做推广"},{"code":"telemetry_described_as_opt_in","severity":"info","evidence":"README 的 Telemetry 部分写道:\"Local logging always — see your own usage stats in ~/.ai-marketing-skills/analytics/\"、\"Remote reporting optional — only if you explicitly opt in on first run\",收集的内容包括 \"skill name, duration, success/fail, version, OS\"。它还会运行版本检查(telemetry/version_check.py)。","source":"ai","title":"遥测被描述为可选"},{"code":"no_single_install_each_skill_differs","severity":"info","evidence":"仓库顶层约有 30 个 skill 文件夹,没有 releases,也没有根目录的安装方式:\"Each skill category has its own README with setup instructions\",每个文件夹都有自己的 requirements.txt 和 .env.example,所以需要配置的内容因 skill 而异。","source":"ai","title":"没有统一安装方式,每个技能各不相同"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Google Gemini (video understanding skills)","Google Search Console (SEO skills)","Gong (revenue intelligence skills)","RB2B and Instantly (outbound skills)","Metricool (short-form delivery skill)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","social_marketing","video"],"alternativeTo":[{"slug":"semrush","name":"Semrush","confirmed":true},{"slug":"phantombuster","name":"PhantomBuster","confirmed":true},{"slug":"jasper","name":"Jasper","confirmed":true},{"slug":"surfer-seo","name":"SEO","confirmed":true}],"declaredAlternatives":[],"stars":3803,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":23,"growthDays":6,"history":[{"date":"2026-11-07","stars":3790},{"date":"2026-11-08","stars":3795},{"date":"2026-11-09","stars":3797},{"date":"2026-11-10","stars":3803}],"firstPublishedAt":"2026-11-06T22:57:16.599Z","commitSha":"d2237ec431608c9ab0551d03a69479b000603f11","verifiedAt":"2026-11-06T22:43:12.153Z","lastCommitAt":"2026-11-10T16:41:45Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A large set of marketing and sales workflows for Claude Code: experiments, cold outbound, SEO, short video and finance reports.","summary":"This repository collects around 30 ready-made \"skills\" for Claude Code and similar AI assistants, written by a marketing agency. Each folder holds a SKILL.md plus Python scripts for one job: growth experiments, turning website visitors into sales leads, scoring content, cold outbound email, SEO briefs, YouTube thumbnails, short-form video editing with FFmpeg, finance reports. You copy the folder you want into your assistant's skills directory, install its Python packages and fill in a .env file with keys for the outside services it uses.","easeRationale":"The same work on all three systems, and it is a C: you clone the repository, copy a whole skill folder into .claude/skills, then per the Quick Start \"pip install -r requirements.txt\", copy .env.example to .env and add keys before running commands such as \"python experiment-engine.py create --hypothesis ...\". Some video skills also need FFmpeg installed.","costNote":"The skills are MIT licensed and free. They only run inside an AI assistant such as Claude Code (normally a paid plan), and several of them call paid services (for example Gong, RB2B, Instantly, Metricool, Gemini).","notVerified":["Whether each script still works against the outside services it calls (Gong, RB2B, Instantly, Metricool, Search Console, Gemini), since none of them is pinned or tested here","What a full run costs in AI tokens and in subscriptions to those services","Whether the statistical claims (bootstrap confidence intervals, Mann-Whitney U tests) are implemented correctly","Whether the PII sanitizer really catches your customers' data before it leaves your machine","How much disk space and how much of your own data each skill needs, which the checks could not measure"]},"zh-Hans":{"summaryShort":"面向 Claude Code 的大量营销与销售工作流:增长实验、陌生开发、SEO、短视频和财务报告。","summary":"该仓库收集了约 30 个面向 Claude Code 及类似 AI 助手的现成 \"skills\"(技能),由一家营销代理机构编写。每个文件夹包含一个 SKILL.md 和用于某项工作的 Python 脚本:增长实验、把网站访客转化为销售线索、给内容打分、陌生开发邮件、SEO 简报、YouTube 缩略图、用 FFmpeg 做短视频剪辑、财务报告。你把想要的文件夹复制到你的助手的 skills 目录,安装它的 Python 依赖包,并在 .env 文件里填入它所用外部服务的密钥。","easeRationale":"三个系统上的操作相同,等级为 C:你克隆仓库,把整个 skill 文件夹复制到 .claude/skills,然后按照 Quick Start 运行 \"pip install -r requirements.txt\",把 .env.example 复制为 .env 并填入密钥,再运行类似 \"python experiment-engine.py create --hypothesis ...\" 的命令。部分视频类技能还需要安装 FFmpeg。","costNote":"这些 skills 采用 MIT 许可证,免费提供。它们只能在 Claude Code 这样的 AI 助手里运行(通常是付费套餐),其中几个还会调用付费服务(例如 Gong、RB2B、Instantly、Metricool、Gemini)。","notVerified":["每个脚本是否仍能正常调用它所依赖的外部服务(Gong、RB2B、Instantly、Metricool、Search Console、Gemini),因为这些服务在这里都没有锁定版本,也没有经过测试。","完整运行一次在 AI token 和这些服务订阅费上的实际花费。","统计方法(bootstrap 置信区间、Mann-Whitney U 检验)是否实现正确。","PII 脱敏工具是否真的能在数据离开你的电脑之前拦截客户信息。","每个 skill 需要多少磁盘空间以及多少你自己的数据,这一点检查无法测量。"]}}},{"id":"R_kgDOVPZUMA","slug":"ethanarbuckle__fairplay-decrypt","name":"fairplay-decrypt","owner":"EthanArbuckle","fullName":"EthanArbuckle/fairplay-decrypt","type":"dev_tool","status":"alert","grade":"D","easeByPlatform":{"mac":"D","linux":"D"},"risk":"high","signals":[{"code":"removes_apple_copy_protection","severity":"high","evidence":"这就是这个项目存在的全部目的。项目描述：“Offline FairPlay App Store IPA decryption with independent key derivation”。README：“The decryptor uses each IPA's SINF and SUPF records to derive page keys, decrypt its encrypted arm64 pages, and clear `cryptid`.”FairPlay 是苹果用于 App Store 应用的版权保护机制。苹果的条款禁止绕过它，很多国家的法律也限制绕过版权保护机制，即便是你已经付费购买的应用。","source":"ai","title":"解除苹果的版权保护机制"},{"code":"requires_jailbroken_device","severity":"attention","evidence":"README：“[Credential retrieval](credential-retrieval/README.md) gets it from a supported jailbroken device and writes `credential.bin` to the chosen output directory.”越狱会关闭 iPhone 自身的一部分保护机制，并可能使保修失效。","source":"ai","title":"需要已越狱的设备"},{"code":"handles_apple_account_credential","severity":"attention","evidence":"README：“Decrypt FairPlay-encrypted apps using a 16-byte credential for the account that owns each IPA”，保存为 credential.bin 文件。这个文件与你的苹果账户绑定，任何复制到它的人都能解密用你的账户购买的应用。","source":"ai","title":"处理苹果账户的凭据"},{"code":"very_new_and_unfinished_packaging","severity":"attention","evidence":"创建于 2026-10-19，最后一次推送是在 2026-10-30，GitHub 发行版为 0 个。整个项目只有 .github/、.gitignore、LICENSE、Makefile、README.md、core.py、credential-retrieval/、decrypt.py 和 requirements.txt——都是需要手动运行的脚本，没有安装程序，也没有打包好的版本，却已经有 2286 个星标和 353 次复刻（fork）。","source":"ai","title":"项目非常新，打包也不完整"}],"platforms":["mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["development","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":2411,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":70,"growthDays":7,"history":[{"date":"2026-10-27","stars":2213},{"date":"2026-10-28","stars":2242},{"date":"2026-10-29","stars":2265},{"date":"2026-10-30","stars":2286},{"date":"2026-10-31","stars":2296},{"date":"2026-11-01","stars":2308},{"date":"2026-11-02","stars":2321},{"date":"2026-11-03","stars":2341},{"date":"2026-11-04","stars":2351},{"date":"2026-11-05","stars":2364},{"date":"2026-11-06","stars":2368},{"date":"2026-11-07","stars":2376},{"date":"2026-11-08","stars":2390},{"date":"2026-11-09","stars":2401},{"date":"2026-11-10","stars":2411}],"firstPublishedAt":"2026-10-27T02:46:37.175Z","commitSha":"bb09a61046f5c99eb1795589a307646df04b26d0","verifiedAt":"2026-10-31T01:58:17.359Z","lastCommitAt":"2026-11-08T15:54:02Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Command-line tool that removes Apple's FairPlay copy protection from iPhone app files (IPA).","summary":"fairplay-decrypt is a command-line tool that takes an App Store app file (IPA) and writes a copy without Apple's FairPlay encryption, so the program inside can be opened and inspected. It is for people who study how iOS apps are built. You need macOS or Linux with Python 3.9+, and a 16-byte credential from the Apple account that owns the IPA; the included code pulls that credential from a jailbroken device, and building it on macOS needs Xcode and ldid. Removing copy protection is restricted or illegal in several countries.","easeRationale":"macOS and Linux are both D: the README asks you to install Python 3.9+, create a virtual environment, install the cryptography package and then run decrypt.py in the terminal, and there are no releases to download (GitHub releases: 0). Before any of that you need the 16-byte credential, which the credential-retrieval part takes from a supported jailbroken device, and on macOS building that part needs Xcode and ldid.","costNote":"Free and open source: the repository has a LICENSE file with the GNU GPL v3 text and GitHub reports the licence as GPL-3.0. You still need an Apple account that owns the apps and a jailbroken device to obtain the credential.","notVerified":["Whether the tool really decrypts current IPA files, and whether the credential retrieval works on today's iOS versions and devices.","What the credential-retrieval code does on the jailbroken device, and whether credential.bin or any app data is sent anywhere — we did not run or fully read the code.","Whether using it is legal where you live, even for apps you bought.","The real disk space needed: the checks give a rough estimate of 100 to 300 MB, and decrypted IPA copies add to that.","Whether the GPL-3.0 licence in the repository covers everything inside credential-retrieval/."]},"zh-Hans":{"summaryShort":"一个命令行工具，用于去掉iPhone应用文件（IPA）上苹果的FairPlay版权保护机制。","summary":"fairplay-decrypt是一个命令行工具，它读取一个App Store的应用文件（IPA），写出一份去掉了苹果FairPlay加密的副本，这样里面的程序就能被打开和检查。它是给研究iOS应用是怎么构建出来的人用的。你需要macOS或Linux，装有Python 3.9及以上版本，还需要拥有该IPA的苹果账号的16字节凭证；仓库自带的代码会从一台越狱设备上获取这个凭证，而且在macOS上编译这部分代码需要Xcode和ldid。移除版权保护机制在不少国家是受限制的，甚至是违法的。","easeRationale":"macOS和Linux的难度评级都是D：README要求你安装Python 3.9及以上版本，创建虚拟环境，安装cryptography包，然后在终端里运行decrypt.py，而且没有可下载的release（GitHub releases: 0）。在这之前，你还需要那个16字节的凭证，仓库里credential-retrieval部分的代码会从一台受支持的越狱设备上获取它，在macOS上编译这部分代码还需要Xcode和ldid。","costNote":"免费且开源：仓库里有一个LICENSE文件，内容是GNU GPL v3许可证文本，GitHub把它标记为GPL-3.0许可证。不过你仍然需要一个拥有这些应用的苹果账号，还需要一台越狱设备才能获取凭证。","notVerified":["这个工具是否真的能解密现在的IPA文件，以及凭证获取这部分代码是否能在今天的iOS版本和设备上用。","credential-retrieval这部分代码在越狱设备上到底做了什么，credential.bin或任何应用数据会不会被发送到别处——我们没有运行也没有完整读过这部分代码。","在你所在的地方使用它是否合法，即使是对你自己购买的应用。","实际需要的硬盘空间：检查结果给出的是大致估计，100到300MB，解密出来的IPA副本还会再占用更多空间。","仓库里的GPL-3.0许可证是否覆盖了credential-retrieval/目录里的所有内容。"]}}},{"id":"R_kgDORanENg","slug":"every-app__open-seo","name":"Open SEO","owner":"every-app","fullName":"every-app/open-seo","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"usage_data_sent_by_default","severity":"attention","evidence":"compose.yaml里有\"Anonymous usage heartbeat controls\"，其中\"OPENSEO_TELEMETRY_DISABLED\"和\"DO_NOT_TRACK\"默认是空的，package.json还依赖\"posthog-js\"和\"posthog-node\"。所以在你自己设置这些变量之前，自托管的这份代码会一直上报使用情况。","source":"ai","title":"默认会发送使用数据"},{"code":"docker_setup_runs_with_no_sign_in","severity":"attention","evidence":"仓库里的compose.yaml设置了\"AUTH_MODE=local_noauth\"，.env.example里把它描述为\"local trusted mode with injected admin user (admin@localhost)\"。它目前没有被暴露出去，只是因为端口以\"127.0.0.1:${PORT}\"的形式发布；一旦有人把这个端口开放到网络上，或者设置了ALLOWED_HOST，整个工具和付费密钥就会在没有密码保护的情况下暴露出去。","source":"ai","title":"Docker安装默认不需要登录"},{"code":"affiliate_link_for_the_paid_data_service","severity":"info","evidence":"README的费用部分链接到DataForSEO的地址是\"https://dataforseo.com/?aff=255379\"——这是一个联盟推广代码，所以当你注册这个工具所需要的服务时，作者会获得一笔分成。","source":"ai","title":"付费数据服务用的是联盟链接"},{"code":"cost_lands_on_your_card_per_request","severity":"attention","evidence":"README写道：\"Bring your own DataForSEO API key and pay only for what you use.\"仓库里甚至还提供了用于查看费用的脚本（\"billing:backlinks\"、\"billing:brand-lookup\"、\"billing:usage\"），这说明单次报告的费用可能不低；如果AI代理在循环中反复调用这些工具，就会在不询问你的情况下花掉你的钱。","source":"ai","title":"费用按请求直接扣在你的卡上"},{"code":"crawler_identity_replay_in_site_audits","severity":"info","evidence":".env.example里写着：\"CRAWLER_ACCESS_SECRET ... lets site audits replay a saved Shopify crawler signature\"，再加上\"web-bot-auth\"和\"http-message-sig\"这两个依赖。这意味着在读取页面时，审计功能可以出示一个带签名的爬虫身份，如果你审计的是不属于你自己的网站，这一点就很值得注意。","source":"ai","title":"网站审计会重放爬虫身份"}],"platforms":["windows","mac","linux","web"],"tools":["docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["DataForSEO (the paid source of all the SEO numbers)","OpenRouter (only for SAM, the in-app AI assistant)","Google OAuth, for the Google Search Console connection","Google Maps (only for the map in Local SEO)","Context.dev (only to let site audits load pages that need JavaScript)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","ai_skills","ai_agents"],"alternativeTo":[{"slug":"semrush","name":"Semrush","confirmed":true},{"slug":"ahrefs","name":"Ahrefs","confirmed":true},{"slug":"moz-pro","name":"Pro","confirmed":true},{"slug":"screaming-frog-seo-spider","name":"SEO Spider","confirmed":true},{"slug":"surfer-seo","name":"SEO","confirmed":true}],"declaredAlternatives":[{"target":"semrush","stance":"claims","evidence":"Open source alternative to Semrush and Ahrefs"},{"target":"ahrefs","stance":"claims","evidence":"Open source alternative to Semrush and Ahrefs"}],"stars":27179,"downloads":268,"downloadsWeek":107,"downloadsDays":6,"growth":807,"growthDays":6,"history":[{"date":"2026-11-04","stars":26372},{"date":"2026-11-05","stars":26511},{"date":"2026-11-06","stars":26622},{"date":"2026-11-07","stars":26749},{"date":"2026-11-08","stars":26883},{"date":"2026-11-09","stars":27041},{"date":"2026-11-10","stars":27179}],"firstPublishedAt":"2026-11-04T01:36:50.369Z","commitSha":"43cbbf96e1bb9b3a5d7a5151de4e0084f3653350","verifiedAt":"2026-11-04T01:26:00.010Z","lastCommitAt":"2026-11-07T06:05:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Keyword research, rank tracking, backlinks and site audits for your website, on your own server and open to your AI agent.","summary":"OpenSEO does the everyday jobs of a paid SEO suite: find keywords, follow where your pages rank, look at competitors and backlinks, audit a site and check how AI assistants mention your brand. It runs on your own server with Docker, or on your own Cloudflare account, and there is a hosted version on the author's site. It also works as an MCP server, so Claude Code, ChatGPT or Codex can run the SEO work for you. The numbers come from DataForSEO, a paid service you sign up for and pay per request.","easeRationale":"The simplest route is the web: open openseo.so, make an account and, to get your own data, add a DataForSEO key. Self-hosting on Windows, Mac or Linux means Docker: copy .env.example to .env, fill in keys and run Compose, or use the Cloudflare route which needs a Cloudflare account and a deploy command in the terminal.","costNote":"The software is free (MIT). The SEO data is not: you need a DataForSEO account and pay them per request. The author's hosted version is free to try, $10/month to support the project, and the README says it charges \"28% extra for every request we make to DataForSEO\".","notVerified":["What a month of real use costs at DataForSEO, which depends entirely on how many reports you run.","Whether the free tier of the hosted version is enough to judge the tool.","What exactly the anonymous usage heartbeat sends.","Whether the Cloudflare free plan is really enough for the no-server route, as the README claims.","How well the MCP tools behave in an AI agent, including how much data and money a single conversation can consume."]},"zh-Hans":{"summaryShort":"为你的网站做关键词研究、排名跟踪、外链和网站审计，运行在你自己的服务器上，并对你的AI代理开放。","summary":"OpenSEO可以完成付费SEO套件的日常工作：查找关键词、跟踪你的页面排名、查看竞争对手和外链、审计网站，还能检查AI助手是怎么提到你的品牌的。它可以用Docker运行在你自己的服务器上，也可以用你自己的Cloudflare账号运行，作者的网站上还提供托管版本。它也可以作为MCP服务器使用，这样Claude Code、ChatGPT或Codex就能替你完成SEO的工作。数据来自DataForSEO，这是一个付费服务，你需要注册并按请求付费。","easeRationale":"最简单的方式是通过网页：打开openseo.so，注册一个账号，如果想用自己的数据，再添加一个DataForSEO密钥。想自己在Windows、Mac或Linux上部署，就要用Docker：把.env.example复制成.env，填好密钥，运行Compose；或者走Cloudflare这条路，需要一个Cloudflare账号，在终端里运行一条部署命令。","costNote":"软件本身是免费的（MIT许可）。SEO数据不是：你需要一个DataForSEO账号，并按请求向他们付费。作者的托管版本可以免费试用，每月10美元用于支持这个项目，README里写着会收取\"28% extra for every request we make to DataForSEO\"。","notVerified":["在DataForSEO上一个月的实际使用成本是多少，这完全取决于你跑了多少次报告。","托管版本的免费额度是否足够用来评估这个工具。","匿名使用心跳包到底发送了什么内容。","像README所说的那样，Cloudflare的免费套餐是否真的足够支撑无服务器的方案。","MCP工具在AI代理里的实际表现如何，包括一次对话会消耗多少数据和金钱。"]}}},{"id":"R_kgDOGVrn2g","slug":"evil0ctal__douyin_tiktok_download_api","name":"Douyin TikTok Download API","owner":"Evil0ctal","fullName":"Evil0ctal/Douyin_TikTok_Download_API","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"bypasses_platform_protections","severity":"high","evidence":"这个项目复刻了各平台用来区分自家应用和其他调用者的请求签名。README 的 “Built with” 部分写道:“Signing: a_bogus, X-Bogus, X-Gnarly, X-Dynosaur and Bilibili's WBI in pure Python”(v4 与 v5 的对比表还补充了 “with a browser fallback”),以及 “Transport: wreq (browser TLS fingerprint emulation)”。关于账号,它写道 “A headless browser mints guest identities, and the pool tops itself up when usable ones run low”,关于项目本身的来源则写道:“This project exists because people wrote down how a signature was built”。这里起作用的规则正是平台自己定的:作者写道,“This fetches data from platforms that have their own terms... Respect those terms and applicable law, respect the people whose content you collect”。","source":"ai","title":"绕过平台的防护机制"},{"code":"downloads_other_peoples_videos","severity":"attention","evidence":"README 提到,它下载视频和图集时“without a watermark (it picks the clean stream the platform already publishes rather than stripping anything)”,并把一切都归档起来,“so a post deleted upstream is still here”。作者自己的说明写道“do not redistribute work that is not yours”:你拿到的文件属于创作者本人。","source":"ai","title":"下载他人的视频"},{"code":"collects_other_peoples_personal_data","severity":"attention","evidence":"功能表列出了作者资料、评论和评论回复,以及粉丝和关注列表,并且“Everything parsed is archived”到你的 PostgreSQL 里。保存他人的资料和评论可能会受到你所在地隐私法律的约束;README 把这个责任交给了你:“Respect those terms and applicable law, respect the people whose content you collect”。","source":"ai","title":"收集他人的个人数据"},{"code":"account_ban_risk_when_importing_your_login","severity":"attention","evidence":"README 写道:“An imported login is used at your own risk: platforms rarely ban an account for reading data with it, but a ban cannot be ruled out”。有几个功能(抖音搜索、某作者点赞过的帖子、哔哩哔哩完整评论)只有用你自己登录的 cookie 才能用。","source":"ai","title":"导入你自己的登录信息存在账号被封的风险"},{"code":"install_script_downloaded_and_run","severity":"info","evidence":"快速开始部分会从网上下载一个 shell 脚本并直接运行它:“curl -fsSL https://raw.githubusercontent.com/.../install/install.sh -o install.sh”,然后“bash install.sh”。README 自己也建议先读一读这个脚本(“less install.sh”)。","source":"ai","title":"下载并运行了安装脚本"}],"platforms":["windows","mac","linux"],"tools":["docker","python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["Internal tokens for its own download and transcription helpers (you generate them yourself; no outside service is called)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","social_marketing","ai_skills"],"alternativeTo":[{"slug":"4k-video-downloader","name":"4K Video Downloader","confirmed":true},{"slug":"tikhub-io","name":"TikHub.io","confirmed":true},{"slug":"bright-data","name":"Bright Data","confirmed":true}],"declaredAlternatives":[],"stars":21335,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":176,"growthDays":6,"history":[{"date":"2026-11-05","stars":21186},{"date":"2026-11-06","stars":21223},{"date":"2026-11-07","stars":21257},{"date":"2026-11-08","stars":21285},{"date":"2026-11-09","stars":21313},{"date":"2026-11-10","stars":21335}],"firstPublishedAt":"2026-11-04T22:47:04.154Z","commitSha":"5aa03528d2e974a0de23faf86597b5af3420df55","verifiedAt":"2026-11-04T22:40:08.769Z","lastCommitAt":"2026-11-09T13:34:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Your own server that collects posts, authors, comments and watermark-free videos from TikTok, Douyin and Bilibili.","summary":"A self-hosted service that reads TikTok, Douyin and Bilibili: one post, an author's posts, comments, sounds, hashtags, live rooms and search results. It downloads videos and photo albums without a watermark, keeps everything in your own database, and can turn a downloaded video into text offline (Mandarin). It has a web console, a REST API, a command line and an MCP server so AI assistants can use it. You bring a machine with Docker and type commands; a public read-only demo lets you look first.","easeRationale":"Linux is the intended home and is still a D: there is no installer you double-click, you need Docker and Docker Compose, a cloned repository, an .env file with passwords you generate, and extra container profiles for the browser, downloader and transcription parts. Mac (including Apple Silicon) and Windows run the same Docker images, so they are no easier. A guided shell script can do the steps for you, but it is still typed in a terminal.","costNote":"Free and open source (Apache-2.0); the README says sponsors pay for it. You pay only for the machine or server you run it on.","notVerified":["Whether it still fetches correctly today: the platforms change their signatures and endpoints, and the project has to follow.","How much disk the archive and downloaded videos use over time (the checks' 100-300 MB estimate covers the software, not your collection).","Whether the published Docker images match the code in the repository.","How good the offline transcription is outside Mandarin; the README says other languages \"come out poorly\".","Whether the public demo instance stays available, since the README says it can be switched off without notice."]},"zh-Hans":{"summaryShort":"搭建属于你自己的服务器,从 TikTok、抖音和哔哩哔哩收集帖子、作者信息、评论,以及无水印视频。","summary":"这是一个自己搭建的服务,可以读取 TikTok、抖音和哔哩哔哩的内容:单条帖子、某作者的全部帖子、评论、音乐、话题标签、直播间和搜索结果。它能下载无水印的视频和图集,把所有内容保存在你自己的数据库里,还能离线把下载的视频转成文字(支持普通话)。它提供网页控制台、REST API、命令行,以及供 AI 助手使用的 MCP 服务器。你只需准备一台装了 Docker 的机器并输入命令;还有一个公开的只读演示可以先看看。","easeRationale":"Linux 是这个项目主要面向的环境,评级依然是 D:没有可以双击运行的安装程序,你需要 Docker 和 Docker Compose、一个克隆下来的仓库、一个由你自己生成密码的 .env 文件,以及浏览器、下载器和转录部分各自额外的容器配置。Mac(包括 Apple Silicon)和 Windows 运行的是同样的 Docker 镜像,所以也不会更简单。有一个引导式的 shell 脚本可以帮你完成这些步骤,但依然需要在终端里输入命令。","costNote":"免费且开源(Apache-2.0);README 说有赞助者在资助这个项目。你只需为运行它的机器或服务器付费。","notVerified":["它现在是否还能正常抓取数据:各平台会不断更改签名和接口,项目必须随之跟进。","随着时间推移,归档和下载的视频会占用多少磁盘空间(检测里 100-300 MB 的估算只是软件本身,不包括你收集的内容)。","发布的 Docker 镜像是否与仓库中的代码一致。","普通话以外的离线转录效果如何;README 提到其他语言 \"come out poorly\"(效果不好)。","公开的演示实例能否一直保持可用,因为 README 说它可能随时被关闭而不事先通知。"]}}},{"id":"R_kgDOJtToIg","slug":"evolution-foundation__evolution-api","name":"Evolution API","owner":"evolution-foundation","fullName":"evolution-foundation/evolution-api","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖中存在已知漏洞(OSV):baileys: GHSA-qvv5-jq5g-4cgg","source":"check","title":"存在已知漏洞"},{"code":"unofficial_whatsapp_access_risks_the_number","severity":"attention","evidence":"README 写道:“WhatsApp API - Baileys: A free API based on WhatsApp Web, leveraging the Baileys library... this method relies on the web version of WhatsApp and may have limitations compared to official APIs”。这条路径没有得到 Meta 的授权:用这种方式让一个普通账号自动化运作违反了 WhatsApp 的条款,号码可能被封。批量发送还可能违反当地的营销和消费者保护规定。","source":"ai","title":"非官方的 WhatsApp 访问方式会让号码面临风险"},{"code":"license_adds_conditions_beyond_apache","severity":"attention","evidence":"GitHub 把这个许可证标记为 NOASSERTION(无法判定),LICENSE 文件解释了原因:“Apache License 2.0, with the following additional conditions... you may not remove or modify the LOGO or copyright information”,以及“the user is required to display a clear notification within the system that Evolution API is being utilized. Failure to comply with this requirement may result in the necessity for a commercial license”。","source":"ai","title":"许可证在 Apache 之外附加了额外条件"},{"code":"stores_other_peoples_messages_and_contacts","severity":"attention","evidence":".env.example 默认开启了全部保存:DATABASE_SAVE_DATA_NEW_MESSAGE=true、DATABASE_SAVE_DATA_CONTACTS=true、DATABASE_SAVE_DATA_CHATS=true、DATABASE_SAVE_DATA_HISTORIC=true。你的数据库最终会保存客户的对话内容和电话号码,这会让你承担隐私方面的责任(如 LGPD、GDPR)。","source":"ai","title":"存储他人的消息和联系人"},{"code":"telemetry_on_by_default","severity":"attention","evidence":".env.example 默认写的是 TELEMETRY_ENABLED=true,README 说它会“collects anonymous telemetry data (routes used, most accessed routes, API version)”。除非你自己把它设为 false,否则它会把数据回传给项目方。","source":"ai","title":"默认开启了遥测功能"},{"code":"open_by_default_settings","severity":"info","evidence":".env.example 把 CORS_ORIGIN 设为 *(任何网站都可以从浏览器调用这个 API),docker-compose.dev.yaml 会把 8080 端口暴露在所有网络接口上,而 docker-compose.yaml 则只绑定到 127.0.0.1。除非你自己加上,否则它前面没有 HTTPS。","source":"ai","title":"默认设置是开放的"},{"code":"development_slowed_and_newest_builds_are_previews","severity":"info","evidence":"检测记录显示,最后一次提交是在 2026-05-06(距本次分析 182 天前),最新发布是在 2026-05-17。最新的两个版本 2.4.0-rc2 和 2.4.0-rc1 都标记为候选发布版;上一个稳定版本是 2025-12-05 发布的 2.3.7。","source":"ai","title":"开发放缓,最新版本只是预览版"}],"platforms":["windows","mac","linux"],"tools":["node","docker","ffmpeg"],"buildTools":["node","docker","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta WhatsApp Cloud API (only for the official route)","OpenAI (optional, for voice-to-text and AI replies)","Pusher (optional, for events)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["social_marketing","automation","development"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"twilio-whatsapp-api","name":"WhatsApp API","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":10083,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":49,"growthDays":6,"history":[{"date":"2026-11-05","stars":10045},{"date":"2026-11-06","stars":10053},{"date":"2026-11-07","stars":10061},{"date":"2026-11-08","stars":10066},{"date":"2026-11-09","stars":10073},{"date":"2026-11-10","stars":10083}],"firstPublishedAt":"2026-11-04T22:32:54.907Z","commitSha":"fa09d37892cdbb1d65a250155d293d92230c5b30","verifiedAt":"2026-11-04T22:17:39.728Z","lastCommitAt":"2026-05-06T17:38:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A server you host that plugs WhatsApp into bots, CRMs and automations through a REST API.","summary":"Evolution API is a server that sits between WhatsApp and your other tools. You connect a WhatsApp number (by QR code through the unofficial Baileys route, or with Meta's official Cloud API) and then send and receive messages over a REST API, with ready connections to Typebot, Chatwoot, n8n, Dify, OpenAI, RabbitMQ, Kafka and S3. It is made for developers and agencies: you need a server with Node.js 20+, PostgreSQL or MySQL and ideally Redis, or the Docker Compose files in the repository.","easeRationale":"Linux on a server is the usual home and is still a D: no installer, you clone the repository, fill in a long .env file, provide PostgreSQL or MySQL and Redis, run database migrations and start it with Docker Compose or Node.js. Mac and Windows run the same stack from the terminal (package.json even has separate Windows migration scripts), so they are no easier.","costNote":"The code is free, but the LICENSE adds conditions: keeping the logo and showing a notice that Evolution API is in use, or buying a commercial licence. Meta's official Cloud API route charges per message, and the OpenAI and Pusher connections are paid services.","notVerified":["Whether the unofficial WhatsApp Web route still works today, since Meta changes it and the library has to follow.","How likely a number is to be blocked in real use.","Whether the published Docker image matches the code in this repository.","How much the Meta Cloud API and the OpenAI connections cost for your message volume.","Whether the flagged vulnerable version of the baileys library affects your setup in practice."]},"zh-Hans":{"summaryShort":"一个你自己搭建的服务器,通过 REST API 把 WhatsApp 接入机器人、CRM 和自动化流程。","summary":"Evolution API 是一个介于 WhatsApp 和你其他工具之间的服务器。你可以连接一个 WhatsApp 号码(通过非官方的 Baileys 方式扫码,或使用 Meta 官方的 Cloud API),然后通过 REST API 收发消息,并已经预置了与 Typebot、Chatwoot、n8n、Dify、OpenAI、RabbitMQ、Kafka 和 S3 的连接。它是为开发者和代理商设计的:你需要一台安装了 Node.js 20+、PostgreSQL 或 MySQL,最好还有 Redis 的服务器,或者使用仓库里提供的 Docker Compose 文件。","easeRationale":"Linux 服务器是常见的运行环境,评级依然是 D:没有安装程序,你需要克隆仓库、填写一份很长的 .env 文件、准备 PostgreSQL 或 MySQL 以及 Redis、运行数据库迁移,然后用 Docker Compose 或 Node.js 启动它。Mac 和 Windows 也是在终端里运行同样的技术栈(package.json 里甚至专门有 Windows 的迁移脚本),所以也不会更简单。","costNote":"代码本身是免费的,但 LICENSE 文件附加了条件:必须保留标志,并显示一条说明表示正在使用 Evolution API,否则就需要购买商业许可。Meta 官方的 Cloud API 按消息计费,OpenAI 和 Pusher 的连接也是付费服务。","notVerified":["非官方的 WhatsApp Web 方式现在是否还能用,因为 Meta 会不断调整,库也必须跟着更新。","在实际使用中,号码被封的可能性有多大。","发布的 Docker 镜像是否与这个仓库里的代码一致。","按照你的消息量,Meta 的 Cloud API 和 OpenAI 连接大概要花多少钱。","被标记为存在漏洞的 baileys 库版本,实际上是否会影响你的部署。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMzEyODM0NTI=","slug":"excalidraw__excalidraw","name":"Excalidraw","owner":"excalidraw","fullName":"excalidraw/excalidraw","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"web":"A"},"risk":"low","signals":[{"code":"shared_boards_leave_your_computer","severity":"info","evidence":"README：托管的应用提供 \"Real-time collaboration\"、\"Shareable links\" 和 \"End-to-end encryption\"，并且仓库里有一个 firebase-project/ 文件夹，因此协作时你的绘图会经过该项目的服务器，而你只留给自己的绘图则保存在浏览器中。","source":"ai","title":"共享的白板会离开你的电脑"},{"code":"tagged_releases_lag_behind_the_code","severity":"info","evidence":"检查显示最近一次发行版发布于 2026-04-21（v0.18.1），而最近一次提交是在 2026-10-27，所以 npm 软件包和线上网站并不处于同一个版本点。","source":"ai","title":"带标签的发行版落后于代码"},{"code":"many_open_issues","severity":"info","evidence":"GitHub 数据：3428 个未解决的 issue，对应 134669 个星标。","source":"ai","title":"未解决的 issue 很多"}],"platforms":["web"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["design_cad","productivity","development"],"alternativeTo":[{"slug":"miro","name":"Miro","confirmed":true},{"slug":"lucidchart","name":"Lucidchart","confirmed":true},{"slug":"microsoft-visio","name":"Visio","confirmed":true},{"slug":"figma","name":"Figma","confirmed":true}],"declaredAlternatives":[],"stars":135454,"downloads":32223,"downloadsWeek":0,"downloadsDays":7,"growth":339,"growthDays":7,"history":[{"date":"2026-10-26","stars":134616},{"date":"2026-10-27","stars":134669},{"date":"2026-10-28","stars":134750},{"date":"2026-10-29","stars":134810},{"date":"2026-10-30","stars":134881},{"date":"2026-10-31","stars":134938},{"date":"2026-11-01","stars":134989},{"date":"2026-11-02","stars":135054},{"date":"2026-11-03","stars":135115},{"date":"2026-11-04","stars":135168},{"date":"2026-11-05","stars":135215},{"date":"2026-11-06","stars":135242},{"date":"2026-11-07","stars":135295},{"date":"2026-11-08","stars":135336},{"date":"2026-11-09","stars":135400},{"date":"2026-11-10","stars":135454}],"firstPublishedAt":"2026-10-27T01:10:12.625Z","commitSha":"14bb42edee3d6fee6be264e3ee78670b3b9cee79","verifiedAt":"2026-10-28T01:36:53.315Z","lastCommitAt":"2026-11-05T17:10:20Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Draw diagrams, sketches and screen layouts in your browser in a hand-drawn style, and share the same board with other people.","summary":"An infinite whiteboard that runs in the browser. You draw boxes, arrows, circles and free-hand lines that come out looking hand-drawn, add images and reusable shape libraries, and export to PNG, SVG or an open .excalidraw file. The public site excalidraw.com needs no account or install, saves your drawing in the browser and works offline; it also offers real-time collaboration and share links that the project says are end-to-end encrypted. Developers can embed the same editor in their own app as an npm package.","easeRationale":"On the web it is an A: excalidraw.com opens in any browser with nothing to install and no account, and it keeps working offline. None of the 15 releases carries a desktop installer, and running your own copy instead needs Docker or Node, which is a job for developers.","costNote":"The code and the public site are free and MIT-licensed. The same team sells Excalidraw+, a paid hosted version with extra features.","notVerified":["Whether the encryption of shared boards works as the project describes.","How long a shared link and its data stay on the project's servers.","Whether running your own copy with the included Dockerfile and docker-compose.yml works on your machine.","How much a drawing with many images can grow before the browser storage runs out."]},"zh-Hans":{"summaryShort":"在浏览器里以手绘风格绘制图表、草图和界面布局，并把同一块画板分享给其他人。","summary":"一块在浏览器里运行的无限画板。你可以画出方框、箭头、圆形和自由线条，它们看起来都是手绘风格，还能添加图片和可复用的图形库，并导出为 PNG、SVG 或开放的 .excalidraw 文件。公开网站 excalidraw.com 不需要账户也不需要安装，会把你的画作保存在浏览器里，并且可以离线使用；它还提供实时协作功能和分享链接，项目方表示这些链接是端到端加密的。开发者可以把同一个编辑器作为 npm 包嵌入自己的应用。","easeRationale":"在网页端评分为 A：excalidraw.com 在任何浏览器中都能打开，无需安装也无需账户，而且可以离线使用。15 个发布版本中没有一个带有桌面安装程序，想运行自己的副本则需要 Docker 或 Node，这是开发者才会做的事。","costNote":"代码和公开网站都是免费的，采用 MIT 许可证。同一个团队出售付费的托管版本 Excalidraw+，提供额外功能。","notVerified":["共享画板的加密是否真的像项目所说的那样工作。","分享链接及其数据会在项目的服务器上保留多长时间。","使用自带的 Dockerfile 和 docker-compose.yml 运行自己的副本是否能在你的电脑上正常工作。","包含大量图片的画作在浏览器存储用尽之前能增长到多大。"]}}},{"id":"R_kgDOJsyiaQ","slug":"facebookresearch__audiocraft","name":"AudioCraft","owner":"facebookresearch","fullName":"facebookresearch/audiocraft","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","linux":"E"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-5pcm-hx3q-hm94","source":"check","title":"依赖项存在已知漏洞"},{"code":"models_are_non_commercial_only","severity":"attention","evidence":"README的许可证部分写道：\"The models weights in this repository are released under the CC-BY-NC 4.0 license as found in the LICENSE_weights file.\"MIT许可证只覆盖代码本身，所以用发布的模型生成的音乐或声音不能用于商业用途。","source":"ai","title":"模型只能用于非商业用途"},{"code":"no_changes_for_over_a_year","severity":"attention","evidence":"Radar的检查记录显示，最后一次提交是在2025-03-13，距检查当天已经600天，GitHub上从来没有发布过正式版本，还有401个未解决的issue。没有人在修复安装问题，也没有人在支持更新的显卡。","source":"ai","title":"一年多没有更新了"},{"code":"pinned_to_old_versions","severity":"attention","evidence":"README写道：\"AudioCraft requires Python 3.9, PyTorch 2.1.0\"；requirements.txt里固定了\"torch==2.1.0\"、\"torchvision==0.16.0\"、\"torchtext==0.16.0\"、\"numpy<2.0.0\"、\"xformers<0.0.23\"、\"spacy==3.7.6\"。这些版本都是2023年的，这也是Radar检查在固定版本的torch里发现已知漏洞的原因，而且较新的NVIDIA显卡需要比这更新的PyTorch才能支持。","source":"ai","title":"版本固定在很旧的版本上"},{"code":"models_downloaded_at_first_use","severity":"info","evidence":"README的FAQ里\"Where are the models stored?\"这一条说明，模型权重来自Hugging Face，会存到一个缓存文件夹（\"AUDIOCRAFT_CACHE_DIR\"），musicgen-melody还会通过Torch Hub额外下载Demucs。所以真正的下载量和磁盘占用是在你第一次生成内容时才发生的，会超出Radar估算的1050 MB到4750 MB。","source":"ai","title":"模型在第一次使用时才下载"},{"code":"research_code_not_an_application","severity":"info","evidence":"README写道：\"AudioCraft is a PyTorch library for deep learning research on audio generation\"，仓库里存放的是训练配方（egs/、config/）和演示笔记本（jasco_demo.ipynb、demos/）。不会编程的人，用那些基于MusicGen做好的现成应用会更合适。","source":"ai","title":"这是研究代码，不是应用程序"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":1050,"maxMB":4750,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai","development"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true},{"slug":"epidemic-sound","name":"Epidemic Sound","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true}],"declaredAlternatives":[],"stars":23839,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":19,"growthDays":6,"history":[{"date":"2026-11-04","stars":23820},{"date":"2026-11-05","stars":23826},{"date":"2026-11-06","stars":23830},{"date":"2026-11-07","stars":23833},{"date":"2026-11-08","stars":23834},{"date":"2026-11-09","stars":23835},{"date":"2026-11-10","stars":23839}],"firstPublishedAt":"2026-11-04T01:36:50.381Z","commitSha":"896ec7c47f5e5d1e5aa1e4b260c4405328bf009d","verifiedAt":"2026-11-04T01:26:52.417Z","lastCommitAt":"2025-03-13T16:01:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Meta's research code for making music and sound effects from a text description, with AI models that run on your own machine.","summary":"AudioCraft is the research code behind Meta's audio models: MusicGen (music from a text description, and from a melody you hum), AudioGen (sound effects from text), EnCodec (audio compression), MAGNeT, JASCO and AudioSeal watermarking. It is a Python library with example notebooks, not a program with a window: you install it with pip and write or run code. It expects Python 3.9, PyTorch 2.1.0, ffmpeg and an NVIDIA graphics card. The code is MIT, but the trained models are licensed for non-commercial use only.","easeRationale":"There is nothing to open on either Windows or Linux: the README only shows pip commands for a Python library, plus installing ffmpeg, and the examples are notebooks and training recipes. You also have to get a specific old combination to work — Python 3.9 with PyTorch 2.1.0 on a CUDA graphics card — which is the hard part on a recent machine.","costNote":"No charge. The code is MIT, but the trained models are CC-BY-NC 4.0: non-commercial use only, so what you generate cannot be sold or used in paid work.","notVerified":["Whether the 2023 version pins still install on a current machine and a current graphics card.","How much graphics memory each model needs, and how long one minute of music takes.","Real disk use after the models are downloaded.","What the models were trained on, which matters for anyone worrying about the rights to the music that comes out.","Whether the examples run on Windows, where the README gives no specific instructions."]},"zh-Hans":{"summaryShort":"这是Meta的研究代码，可以根据文字描述生成音乐和音效，AI模型在你自己的电脑上运行。","summary":"AudioCraft是Meta音频模型背后的研究代码，包括：MusicGen（根据文字描述生成音乐，也可以用你哼出的旋律生成）、AudioGen（根据文字生成音效）、EnCodec（音频压缩）、MAGNeT、JASCO，以及AudioSeal水印系统。它是一个带示例笔记本的Python库，不是带界面的程序：你需要用pip安装它，然后编写或运行代码。它要求Python 3.9、PyTorch 2.1.0、ffmpeg和一块NVIDIA显卡。代码本身是MIT许可，但训练好的模型只允许非商业用途。","easeRationale":"无论是Windows还是Linux，都没有什么界面可以打开：README里只有Python库的pip安装命令，加上安装ffmpeg，示例也只是笔记本和训练配方。而且你还得让一套特定的老版本组合跑起来——Python 3.9配合PyTorch 2.1.0，再加上支持CUDA的显卡——这在一台新电脑上才是真正麻烦的地方。","costNote":"不收费。代码是MIT许可，但训练好的模型是CC-BY-NC 4.0许可：只能用于非商业用途，所以你用它生成的内容不能拿去卖，也不能用在有偿的工作里。","notVerified":["2023年固定下来的这些版本，是否还能在现在的电脑和现在的显卡上装起来。","每个模型需要多少显存，生成一分钟音乐要花多长时间。","模型下载完之后实际占用多少磁盘空间。","这些模型是用什么数据训练的，这对担心生成音乐版权问题的人来说很重要。","这些示例在Windows上能不能跑，因为README并没有给出针对Windows的具体说明。"]}}},{"id":"R_kgDOPrlnxA","slug":"facebookresearch__mhr","name":"Momentum Human Rig","owner":"facebookresearch","fullName":"facebookresearch/MHR","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"mac":"D","linux":"D","web":"C"},"risk":"low","signals":[{"code":"model_files_downloaded_separately","severity":"info","evidence":"仅有代码还不够:你必须运行 \"mhr-download-assets\" 或 \"pixi run download-assets\" 来获取项目结构中列出的 rig 文件(compact_v6_1.model、corrective_blendshapes_lod?.npz、lod?.fbx、mhr_model.pt),这些文件来自 GitHub 的 release,并不在仓库本身里。","source":"ai","title":"模型文件需单独下载"},{"code":"pip_install_marked_experimental","severity":"info","evidence":"README 在谈到 pip 方式时写道:\"pip installation is experimental. Some dependencies (e.g., pymomentum) may not resolve correctly on all platforms\",而项目自己的 Pixi 配置也只声明了两个平台:platforms = [\"osx-arm64\", \"linux-64\"]。文档中完全没有提到 Windows。","source":"ai","title":"pip 安装被标记为实验性"},{"code":"research_code_little_packaging","severity":"info","evidence":"目前只有一个 release(v1.0.1,2025-11-19),没有任何平台的二进制文件,README 的内容也是围绕 demo.py、一个笔记本和一篇 arXiv 论文(arXiv:2511.15586)展开的。它是作为研究代码发布的,而不是作为产品。","source":"ai","title":"研究性代码,打包程度不高"}],"platforms":["mac","linux","web"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["design_cad","development"],"alternativeTo":[{"slug":"reallusion-character-creator","name":"Character Creator","confirmed":true}],"declaredAlternatives":[],"stars":902,"downloads":14473,"downloadsWeek":554,"downloadsDays":2,"growth":11,"growthDays":6,"history":[{"date":"2026-11-08","stars":898},{"date":"2026-11-09","stars":902},{"date":"2026-11-10","stars":902}],"firstPublishedAt":"2026-11-07T22:56:16.752Z","commitSha":"dd4340315901ba2e367133b86971e0caf8e91dd5","verifiedAt":"2026-11-07T22:41:03.489Z","lastCommitAt":"2026-10-26T17:36:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A ready-made 3D human body with face expressions that you shape and pose from Python code.","summary":"A Python package from Meta that gives you a full-body 3D human figure you can change with numbers: 45 values for the body shape, 204 for the pose of the skeleton and 72 for facial expressions, in seven levels of detail. It is made for researchers and 3D or computer-vision developers who need bodies and faces generated inside their own code, and it works with PyTorch on processor or graphics card. There is no app or window: you install it with Pixi or pip on macOS or Linux, download the model files and write Python.","easeRationale":"The easiest way in is the browser (C): the README links a visualization notebook that opens in Google Colab, where you install nothing but still run code cells. On macOS and Linux it is a D: you clone the repository, install the Pixi package manager, run \"pixi install\" and \"pixi run download-assets\", then write Python against PyTorch, and the README marks the simpler pip route as experimental because \"pymomentum\" may not resolve on every platform.","costNote":"Apache License 2.0 for the code, as stated in the README and LICENSE.","notVerified":["Whether the install really completes, since it depends on PyTorch and pymomentum resolving on your machine.","Real disk space after the model files are downloaded and unzipped.","Whether it works on Windows at all, as the project only declares macOS (Apple Silicon) and Linux.","How fast it runs on a processor without a graphics card.","Under which terms the downloaded rig files themselves may be used, since the README states the licence for the repository."]},"zh-Hans":{"summaryShort":"一个现成的 3D 人体模型,带面部表情,可以通过 Python 代码来调整形状和姿势。","summary":"这是 Meta 推出的一个 Python 包,提供一个可以用数字调整的全身 3D 人体模型:45 个数值控制身材形状,204 个控制骨骼姿势,72 个控制面部表情,共有七种细节级别。它面向需要在自己代码里生成人体和面部的研究人员以及 3D 或计算机视觉开发者,并支持在处理器或显卡上运行 PyTorch。它没有应用程序或窗口界面:你在 macOS 或 Linux 上用 Pixi 或 pip 安装,下载模型文件,然后编写 Python 代码。","easeRationale":"最简单的入门方式是浏览器(C 级):README 链接了一个可视化笔记本,可以在 Google Colab 中打开,你不用安装任何东西就能运行代码单元。在 macOS 和 Linux 上属于 D 级:你需要克隆仓库,安装 Pixi 包管理器,运行 \"pixi install\" 和 \"pixi run download-assets\",然后用 PyTorch 编写 Python 代码,README 把更简单的 pip 方式标记为实验性,因为 \"pymomentum\" 在某些平台上可能无法正确解析。","costNote":"代码采用 Apache License 2.0 许可证,README 和 LICENSE 文件中都有说明。","notVerified":["安装是否真的能顺利完成,因为这取决于 PyTorch 和 pymomentum 能否在你的机器上正确解析。","下载并解压模型文件之后实际占用的磁盘空间。","它是否能在 Windows 上运行,因为项目只声明支持 macOS(Apple Silicon)和 Linux。","在没有显卡、只用处理器的情况下运行速度有多快。","下载下来的 rig 文件本身可以在什么条款下使用,因为 README 说明的只是仓库代码的许可证。"]}}},{"id":"R_kgDOPYG9FA","slug":"farion1231__cc-switch","name":"CC Switch","owner":"farion1231","fullName":"farion1231/cc-switch","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"readme_is_mostly_paid_advertising_with_referral_links","severity":"attention","evidence":"“Sponsor”部分列出了大约三十五家 API 代理商，每家都带有追踪链接和代码，例如“register using this link and enter the 'cc-switch' promo code during first recharge to get 10% off”，以及带联系方式的“Want to appear here?”。这个项目靠这些链接带来的注册获得收入。","source":"ai","title":"readme 大部分内容是带推广链接的付费广告"},{"code":"advertised_services_resell_models_far_below_official_prices","severity":"attention","evidence":"README 中的赞助文字包括“Claude Code at 19% of original price, GPT at just 1%”、“Claude Code / Codex / Gemini official channels at 38% / 2% / 9% of original price”、“your drop-in replacement for a Claude.ai subscription ... at roughly 30% of direct API cost, with no VPN required”，以及“prices as low as 7% of official rates”。这些渠道从何而来无法在此核实，通过代理商购买某家厂商的模型，也可能违反你与该厂商之间的约定。","source":"ai","title":"推广的服务以远低于官方价格转卖模型"},{"code":"it_holds_your_keys_and_edits_other_programs_settings","severity":"attention","evidence":"README，Why CC Switch：“pick a preset, enter your key, and switch in one click. Switching replaces only the connection details - endpoint, key, and model”。这款应用保存你添加的每个服务商的密钥，并写入多达十二个其他程序的配置文件。","source":"ai","title":"它保存你的密钥并修改其他程序的设置"},{"code":"routing_mode_passes_every_request_through_the_app","severity":"attention","evidence":"README：“Routing forwards requests through your machine, converting API formats and failing over automatically, so you can use GPT in Claude Code and Claude in Codex”，使用页面还会记录请求日志和花费。","source":"ai","title":"路由模式会让每个请求都经过这款应用"},{"code":"project_warns_about_copycat_websites","severity":"info","evidence":"仓库描述结尾写着“Only official website: ccswitch.io”，README 也重复了“The Only Official Website: ccswitch.io”，这说明可能存在使用同一名称的其他网站。","source":"ai","title":"项目提醒要注意仿冒网站"},{"code":"some_linux_systems_not_supported","severity":"info","evidence":"README，System Requirements：Linux 需要“glibc 2.35+ and WebKitGTK 4.1 - e.g. Ubuntu 22.04+, Debian 12+, and recent Fedora releases; RHEL / Rocky / Alma 8-9 are not supported yet”。Flatpak“not included in official releases”。","source":"ai","title":"部分 linux 系统不受支持"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":18,"maxMB":55,"basis":"installer"},"apiKeys":["A key or plan from at least one AI provider"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","productivity"],"alternativeTo":[{"slug":"openrouter","name":"OpenRouter","confirmed":true}],"declaredAlternatives":[{"target":"invideo-ai","stance":"claims","evidence":"CCSub is a stable, affordable AI API relay platform — your drop-in replacement for a Claude.ai subscription"},{"target":"claude-code","stance":"claims","evidence":"Gemini CLI, or Grok Build, CC Switch replaces only the key fields in the config file: the endpoint, key, model name, and API protocol (plus the reasoning effort for Codex and the auth method for Gemin"}],"stars":153329,"downloads":22876302,"downloadsWeek":467943,"downloadsDays":7,"growth":1809,"growthDays":7,"history":[{"date":"2026-10-26","stars":148969},{"date":"2026-10-27","stars":149388},{"date":"2026-10-28","stars":149737},{"date":"2026-10-29","stars":150071},{"date":"2026-10-30","stars":150405},{"date":"2026-10-31","stars":150644},{"date":"2026-11-01","stars":150912},{"date":"2026-11-02","stars":151223},{"date":"2026-11-03","stars":151520},{"date":"2026-11-04","stars":151818},{"date":"2026-11-05","stars":152097},{"date":"2026-11-06","stars":152357},{"date":"2026-11-07","stars":152586},{"date":"2026-11-08","stars":152738},{"date":"2026-11-09","stars":153050},{"date":"2026-11-10","stars":153329}],"firstPublishedAt":"2026-10-27T01:10:12.626Z","commitSha":"7afceec14b977ecc2f4373e8ffb40e7223b435ac","verifiedAt":"2026-11-03T02:43:53.315Z","lastCommitAt":"2026-11-10T15:08:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop app that switches which AI service your coding tools use, without editing their settings files by hand.","summary":"CC Switch is a small desktop app for people who use AI coding tools (Claude Code, Codex, Gemini CLI, OpenCode and others). You keep your providers and keys in one window and switch which one each tool uses with a click, instead of editing JSON, TOML or YAML files yourself. It can also forward requests through your own machine so one tool can use another service's models, and it keeps MCP servers, skills and prompts in one place. It is free; the services it connects to are yours to pay for.","easeRationale":"macOS is the smoothest: a .dmg the README says is \"code-signed and notarized by Apple\", or one Homebrew command. Windows has an .msi installer and a portable zip. All three are B because after installing you still have to add a provider and paste a key; on Linux you also have to check that your system has glibc 2.35+ and WebKitGTK 4.1.","costNote":"The app is free and MIT-licensed. The services it points your tools at are paid, and the README is largely a list of paid providers with referral links and discount codes.","notVerified":["Whether the app really works with each of the twelve tools it lists, since all of them are other people's programs.","How the keys you paste are stored on disk and whether anything is sent anywhere.","Whether the advertised resellers deliver what their texts promise, or how long they stay in business.","Whether the released installers were built from the code published here.","Whether switching a tool's provider leaves the rest of that tool's settings untouched, as the README states."]},"zh-Hans":{"summaryShort":"一款桌面应用，能切换编程工具使用的 AI 服务，不用手动编辑它们的设置文件。","summary":"CC Switch 是一款小型桌面应用，面向使用 AI 编程工具（Claude Code、Codex、Gemini CLI、OpenCode 等）的人。你把各个服务商和密钥都保存在一个窗口里，点一下就能切换每个工具使用哪一个，而不用自己去改 JSON、TOML 或 YAML 文件。它还能把请求通过你自己的机器转发，让一个工具使用另一个服务的模型，并把 MCP 服务器、技能和提示词都集中在一处。它本身是免费的；它连接的那些服务费用需要你自己支付。","easeRationale":"macOS 上最顺畅：一个 README 称为“code-signed and notarized by Apple”的 .dmg，或者一条 Homebrew 命令。Windows 有 .msi 安装包和便携版 zip。三个系统都是 B 级，因为安装之后你还要自己添加一个服务商并粘贴密钥；在 Linux 上你还要检查系统是否满足 glibc 2.35+ 和 WebKitGTK 4.1。","costNote":"这款应用免费，采用 MIT 许可证。它指向的那些服务是收费的，README 很大一部分其实是一份带推广链接和折扣码的付费服务商列表。","notVerified":["它列出的十二款工具是否都真的能配合使用，因为这些都是别人开发的程序。","你粘贴的密钥在磁盘上是怎么保存的，以及是否有内容被发送到别处。","这些被推广的代理商是否真的兑现了宣传文字里的承诺，以及它们能经营多久。","发布的安装包是否确实是用这里公开的代码编译出来的。","按 README 所说，切换一个工具的服务商是否会让该工具其他设置保持不变。"]}}},{"id":"R_kgDOMf5_Zg","slug":"feizc__fluxmusic","name":"FluxMusic","owner":"feizc","fullName":"feizc/FluxMusic","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"license_forbids_commercial_use","severity":"attention","evidence":"LICENSE.txt是\"Attribution-NonCommercial-ShareAlike 4.0 International\"许可证,其中规定\"NonCommercial means not primarily intended for or directed towards commercial advantage or monetary compensation\"。GitHub把许可证标记为NOASSERTION,所以读取许可证字段的工具什么也看不到。","source":"ai","title":"许可证禁止商业使用"},{"code":"no_setup_instructions","severity":"attention","evidence":"README.md里唯一的环境说明是\"You can refer to the [link](https://github.com/black-forest-labs/flux) to build the running environment\",而顶层文件(LICENSE.txt、README.md、model.py、sample.py、train.py、scripts/、config/、modules/、audioldm2/)里没有依赖文件、安装文件或Docker文件。","source":"ai","title":"没有安装说明"},{"code":"development_stopped","severity":"attention","evidence":"本次检查所用快照中的最后一次提交是2024-12-10,距检查时间695天,也没有任何GitHub发布版本,而README仍然承诺\"The full version of main results will be released right way\"。","source":"ai","title":"开发已经停止"},{"code":"model_files_hosted_outside_github","severity":"info","evidence":"README.md的下载表格把所有模型都指向了Hugging Face:VAE和声码器在`cvssp/audioldm2`,T5-XXL在`stabilityai/stable-diffusion-3-medium-diffusers`里,CLAP-L在`laion/larger_clap_music`,五个FluxMusic检查点在`feizhengcong/FluxMusic`。这些文件各自都有自己的许可证和条款。","source":"ai","title":"模型文件托管在github之外"},{"code":"research_quality_not_product","severity":"info","evidence":"README.md写道:\"This is a research project, and it is recommended to try advanced products\",并附上一个商业音乐网站的链接,发布的检查点是200K步的模型,\"trained on a sub-training set and used for ploted the scaling experiments\"。","source":"ai","title":"研究水准,并非成品"},{"code":"training_data_not_included","severity":"info","evidence":"README.md:\"Considering copyright issues, the data used in the paper needs to be downloaded by oneself.\"只有\"a clean subset\"作为Hugging Face数据集发布。","source":"ai","title":"未提供训练数据"}],"platforms":["linux"],"tools":["python","gpu_nvidia"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":1714,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":1,"growthDays":6,"history":[{"date":"2026-11-06","stars":1713},{"date":"2026-11-07","stars":1713},{"date":"2026-11-08","stars":1713},{"date":"2026-11-09","stars":1714},{"date":"2026-11-10","stars":1714}],"firstPublishedAt":"2026-11-05T22:43:48.394Z","commitSha":"126bdcedf6df5e0816a8291c8dad2ea808d1a311","verifiedAt":"2026-11-05T22:24:24.593Z","lastCommitAt":"2024-12-10T06:35:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code and model files that turn a written description into a short music clip on your own graphics card.","summary":"The code from a research paper on text-to-music generation: you write a description like \"calm piano with rain\" and the model produces a music clip. It is made for researchers, not for listeners: there is no app or installer, you build the Python environment yourself following another project's instructions, download four extra model files from Hugging Face plus one FluxMusic checkpoint, and run training or sampling commands. You need Python and a strong graphics card. The authors call it a research project and point visitors to a commercial site instead.","easeRationale":"Linux is the only documented path and it is an E: the README has no install instructions of its own and sends you to the Flux repository to \"build the running environment\", then you download the VAE, vocoder, T5-XXL and CLAP-L files plus a checkpoint by hand and run `torchrun` or `python sample.py` from the command line. There is no release, no installer and no requirements file in the top-level listing.","costNote":"No price. The repository licence is Creative Commons Attribution-NonCommercial-ShareAlike 4.0, which does not allow commercial use.","notVerified":["How the generated music actually sounds and how long each clip is","Which graphics card and how much video memory are really needed, since the README never says","Whether the environment built from the Flux repository instructions still works with current library versions","Real disk space taken by the T5-XXL, CLAP-L, VAE, vocoder and checkpoint downloads","Whether it runs on Windows or Mac, which the project never mentions","The licence terms of each model file downloaded from Hugging Face"]},"zh-Hans":{"summaryShort":"研究代码和模型文件,能把一段文字描述变成一小段音乐,用你自己的显卡生成。","summary":"这是一篇文本生成音乐研究论文的代码:你写一句描述,比如\"calm piano with rain\",模型就会生成一段音乐片段。它是为研究者而不是听众做的:没有应用程序也没有安装包,你需要按照另一个项目的说明自己搭建Python环境,再从Hugging Face下载四个额外的模型文件,加上一个FluxMusic检查点,然后运行训练或采样命令。你需要Python和一块强力显卡。作者把它称为一个研究项目,并把访客引向了一个商业网站。","easeRationale":"Linux是唯一有文档说明的途径,而且是E级:README没有自己的安装说明,而是让你去Flux的代码库\"build the running environment\",然后你需要手动下载VAE、声码器、T5-XXL和CLAP-L文件,再加一个检查点,并在命令行里运行`torchrun`或`python sample.py`。顶层目录里没有发布版本、没有安装包,也没有依赖文件。","costNote":"没有价格。该代码库的许可证是Creative Commons Attribution-NonCommercial-ShareAlike 4.0,不允许商业使用。","notVerified":["生成的音乐实际听起来如何,每段片段有多长","实际需要什么显卡、多少显存,因为README里从未说明","按Flux代码库说明搭建的环境,是否仍能配合目前版本的库正常运行","T5-XXL、CLAP-L、VAE、声码器和检查点这些下载内容实际占用的磁盘空间","能否在Windows或Mac上运行,项目完全没有提及","从Hugging Face下载的每个模型文件各自的许可条款"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTgzMjIyODY=","slug":"felixse__fluentterminal","name":"Fluent Terminal","owner":"felixse","fullName":"felixse/FluentTerminal","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2022-03-13(1700 天前)","source":"check","title":"版本更新停滞"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 1.8/10","source":"check","title":"scorecard 评分偏低"},{"code":"declared_unmaintained_by_author","severity":"attention","evidence":"README 开头写着 \"⚠️ Project Status: Unmaintained — This project is no longer being maintained\",并推荐了另一个程序:\"I highly recommend using Windows Terminal as an alternative.\"。目前有 240 个未关闭的 issue,最近一次发布是 2022 年 3 月,漏洞和缺陷不太可能再得到修复。","source":"ai","title":"作者声明已不再维护"},{"code":"install_scripts_need_admin","severity":"info","evidence":"README 中的包管理器安装方式都要求提升权限的 shell(\"From an elevated/admin shell, execute choco install fluent-terminal\",以及 Scoop 的 \"nonportable\" bucket),资源管理器右键菜单功能则是通过 \"Explorer Context Menu Integration\" 文件夹里一个单独的安装脚本添加的。","source":"ai","title":"安装脚本需要管理员权限"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"mobaxterm","name":"MobaXterm","confirmed":true},{"slug":"warp","name":"Warp","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":9603,"downloads":187894,"downloadsWeek":32,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":9603},{"date":"2026-11-09","stars":9602},{"date":"2026-11-10","stars":9603}],"firstPublishedAt":"2026-11-07T22:56:16.759Z","commitSha":"67504455c74b17853dd6fa8cc684860cd0afb21b","verifiedAt":"2026-11-07T22:41:36.640Z","lastCommitAt":"2026-10-23T18:23:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Windows terminal window for PowerShell, CMD, WSL and SSH, with tabs and themes. The author says it is no longer maintained.","summary":"A terminal program for Windows: a window where you type commands, with tabs, several windows, themes you can import from iTerm, editable shortcuts, search, and shell profiles for PowerShell, CMD, WSL or your own. It connects to other machines over SSH and Mosh and can be added to the Explorer right-click menu. It installs from the Microsoft Store and the interface is translated into more than 30 languages. The README now opens with a notice that the project is unmaintained and points to Windows Terminal instead.","easeRationale":"On Windows it is an A: you search for Fluent Terminal in the Microsoft Store and install it, or double-click the .appxbundle file from a release. The only requirement listed is Windows 10 Fall Creators Update or later. It exists for Windows only, as it is built on UWP.","costNote":"No price mentioned; the GPL-3.0 licence means anyone redistributing changes must publish them.","notVerified":["Whether the Microsoft Store listing is still available and still installs on current Windows versions.","Whether the app still works with today's WSL, PowerShell and SSH after years without a release.","Whether the .appxbundle in the 2022 release installs by double-clicking on a current Windows machine.","Disk space and memory it uses while running."]},"zh-Hans":{"summaryShort":"一个适用于 PowerShell、CMD、WSL 和 SSH 的 Windows 终端窗口,带标签页和主题。作者说该项目已不再维护。","summary":"一个面向 Windows 的终端程序:一个你可以在其中输入命令的窗口,支持标签页、多个窗口、可从 iTerm 导入的主题、可编辑的快捷键、搜索,以及针对 PowerShell、CMD、WSL 或自定义的 shell 配置文件。它可以通过 SSH 和 Mosh 连接到其他机器,还能添加到资源管理器的右键菜单里。它通过 Microsoft Store 安装,界面被翻译成 30 多种语言。README 现在一开头就提示该项目已不再维护,并建议改用 Windows Terminal。","easeRationale":"在 Windows 上属于 A 级:你在 Microsoft Store 里搜索 Fluent Terminal 并安装,或者双击某个 release 里的 .appxbundle 文件。列出的唯一要求是 Windows 10 Fall Creators Update 或更高版本。它只支持 Windows,因为是基于 UWP 构建的。","costNote":"未提及价格;GPL-3.0 许可证意味着任何重新分发改动的人都必须公开这些改动。","notVerified":["Microsoft Store 页面是否还在,以及是否仍能在当前版本的 Windows 上安装。","这款应用在多年没有发布新版本之后,是否还能配合现在的 WSL、PowerShell 和 SSH 正常工作。","2022 年那次发布中的 .appxbundle 是否还能在当前的 Windows 机器上通过双击安装。","它运行时占用的磁盘空间和内存。"]}}},{"id":"R_kgDORHiFvg","slug":"finrandojin__alexandria-audiobook","name":"Alexandria","owner":"Finrandojin","fullName":"Finrandojin/alexandria-audiobook","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"voice_cloning_from_a_short_sample","severity":"attention","evidence":"功能列表写着 \"Voice Cloning - Clone any voice from a 5-15 second reference audio sample\",Training 标签页会用你自己的录音微调出一个可长期使用的声音副本。未经本人同意复制真人的声音,在一些国家可能是违法的,README 对此没有提到需要获得同意。","source":"ai","title":"用一小段样本克隆声音"},{"code":"web_interface_and_api_without_a_password","severity":"attention","evidence":"网页界面和整个 REST API(上传、生成、训练、下载)都在 4200 端口上响应,README 里完全没有提到任何登录方式,Dockerfile 设置了 \"ENV ALEXANDRIA_HOST=0.0.0.0\",docker-compose 发布了 \"4200:4200\"。在共享电脑或服务器上,任何能访问这个端口的人都能使用你的模型和文件,而且 Colab 那条路径还会开一个 ngrok 隧道通向它。","source":"ai","title":"网页界面和 api 都没有登录验证"},{"code":"your_book_text_goes_to_the_language_model_you_choose","severity":"info","evidence":"第二步写着 \"Click Generate Annotated Script — this sends the book to your LLM\",Setup 标签页需要填写 \"LLM Base URL\" 和 \"LLM API Key\"。如果你选择像 OpenAI API 这样的云服务,整本书的文字就会离开你的电脑;用 LM Studio 或 Ollama 的话则留在本地。","source":"ai","title":"你书里的文字会发给你选择的语言模型"},{"code":"rights_over_the_book_and_the_result","severity":"info","evidence":"Script 标签页接受 .epub 文件,整个流程就是为了朗读 \"any book or novel\" 而设计的。把一本不是你写的书做成有声书,尤其是再分享或出售它,需要版权所有者的许可。","source":"ai","title":"对书籍和生成结果的权利"},{"code":"large_downloads_on_first_use","severity":"info","evidence":"README 写道 \"The TTS models are not included in the install. They download automatically from Hugging Face the first time you generate audio\",\"Each model variant is ~3.5 GB\",磁盘需求是 \"~20 GB (8 GB venv/PyTorch, ~7 GB for model weights, working space for audio)\"。下载期间界面可能看起来像卡住了。","source":"ai","title":"第一次使用时要下载大文件"}],"platforms":["windows","linux","mac"],"tools":["python","gpu_nvidia","ffmpeg","docker"],"buildTools":["docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenAI or another OpenAI-compatible language model service (not needed if you run LM Studio or Ollama on your own machine)","Hugging Face (HF_TOKEN) for downloading the speech models","ngrok, only for the Google Colab route"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai","office"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"speechify","name":"Speechify","confirmed":true},{"slug":"audible","name":"Audible","confirmed":true}],"declaredAlternatives":[],"stars":1097,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":9,"growthDays":6,"history":[{"date":"2026-11-08","stars":1093},{"date":"2026-11-09","stars":1094},{"date":"2026-11-10","stars":1097}],"firstPublishedAt":"2026-11-07T22:56:16.766Z","commitSha":"be6343d95d94db5f88973aa25575a390f3ce46aa","verifiedAt":"2026-11-07T22:42:53.478Z","lastCommitAt":"2026-08-02T11:05:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a book file into a multi-voice audiobook on your own computer, with AI voices, chapters and MP3 or M4B export.","summary":"You give it a book as .txt, .md or .epub and it produces a narrated audiobook with a different voice for each character. An AI language model you supply splits the text into lines with speakers and tone directions, then a local speech model reads them; you can fix any line in a browser editor and export MP3, a chaptered M4B or separate tracks for Audacity. It needs an NVIDIA card with 8 GB of video memory for decent speed, about 20 GB of disk, a separate language-model server, and it installs through Pinokio or Docker. A Google Colab notebook is also offered.","easeRationale":"Windows and Linux are the best case and still a D: the install goes through Pinokio or Docker with the NVIDIA Container Toolkit, you need an NVIDIA card with at least 8 GB of video memory and driver 550+, about 20 GB of disk, and you must separately install and start a language-model server such as LM Studio or Ollama before the first button works. On a Mac it is also a D and the README says Apple Silicon runs on processor only, \"functional but slow\"; AMD cards need Linux with ROCm.","costNote":"The program is free (MIT). It can run entirely on your machine with a free local language model, but if you point it at a paid service such as the OpenAI API you pay that service for every book you process.","notVerified":["Whether the Pinokio install finishes on your machine, since it builds a Python environment and PyTorch at install time.","The speed figures in the README (3-6x real time), measured by the author on an AMD RX 7900 XTX with 24 GB.","How good the narration and the character voices actually sound for your book and language.","Real disk and video-memory use after the models, the LoRA adapters and the generated audio pile up.","Whether the Google Colab notebook still works on the free tier and how long a book takes there.","How much a cloud language model would cost to annotate a full-length book."]},"zh-Hans":{"summaryShort":"在你自己的电脑上把一本书变成多人配音的有声书,带 AI 配音、章节划分,可导出 MP3 或 M4B。","summary":"你给它一本 .txt、.md 或 .epub 格式的书,它就会生成一本每个角色都有不同配音的有声书。你提供的 AI 语言模型会把文本拆分成带说话人和语气提示的台词,再由本地语音模型朗读出来;你可以在浏览器编辑器里修改任意一行,并导出 MP3、带章节的 M4B,或供 Audacity 使用的独立音轨。它需要一张显存 8 GB 的 NVIDIA 显卡才能有不错的速度,大约 20 GB 的磁盘空间,一个单独的语言模型服务器,通过 Pinokio 或 Docker 安装。它也提供一个 Google Colab 笔记本。","easeRationale":"Windows 和 Linux 是最好的情况,仍然属于 D 级:安装要通过 Pinokio 或带 NVIDIA Container Toolkit 的 Docker 完成,你需要一张显存至少 8 GB、驱动版本 550 以上的 NVIDIA 显卡,大约 20 GB 磁盘空间,而且必须先单独安装并启动一个语言模型服务器(比如 LM Studio 或 Ollama),第一个按钮才能用。在 Mac 上也是 D 级,README 说 Apple Silicon 只能用处理器运行,\"functional but slow\";AMD 显卡则需要搭配 ROCm 的 Linux。","costNote":"该程序是免费的(MIT 许可证)。配合免费的本地语言模型时可以完全在你自己的电脑上运行,但如果你接入像 OpenAI API 这样的付费服务,每处理一本书都要向那项服务付费。","notVerified":["Pinokio 的安装是否真的能在你的电脑上顺利完成,因为它在安装时要搭建 Python 环境和 PyTorch。","README 里的速度数据(3-6 倍实时速度),是作者在一台有 24 GB 显存的 AMD RX 7900 XTX 上测出来的。","针对你的书和语言,旁白和角色配音实际听起来效果如何。","模型、LoRA 适配器和生成的音频累积起来之后,实际占用的磁盘和显存。","Google Colab 笔记本在免费额度下是否还能用,在那里处理一本书要多长时间。","用云端语言模型给一整本书做标注大概要花多少钱。"]}}},{"id":"R_kgDOTsxOiw","slug":"firecrawl__anydoc","name":"AnyDoc","owner":"firecrawl","fullName":"firecrawl/anydoc","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"scanned_pdfs_are_sent_whole_to_the_company","severity":"attention","evidence":"README OCR 部分：“Only documents that need OCR leave the machine, and the whole document goes, since Parse has no page selection.”这是可选项，通过 --ocr hosted 开启，而且“No signup needed”，所以不需要账号就能上传一整份机密文档。","source":"ai","title":"扫描版 pdf 会整份发送给公司"},{"code":"the_comparison_table_is_the_authors_own","severity":"info","evidence":"README，Benchmark：评分来自项目自己运行的“an LLM judge (Claude Sonnet 5)”，而且“the corpus is not redistributable and is not in the repo”。项目外部的人无法重复这个让 anydoc 在每一行都排第一的测试。","source":"ai","title":"对比表格是作者自己做的"},{"code":"released_versions_are_two_months_behind_the_code","severity":"info","evidence":"最新发布版本是 2026-08-27 的 v0.2.4，而最后一次提交是 2026-10-23。你安装的包比你在仓库里读到的代码更旧，而且有 130 个未解决的 issue。","source":"ai","title":"发布的版本比代码落后两个月"},{"code":"free_tool_next_to_the_companys_paid_service","severity":"info","evidence":"README：“It powers Firecrawl Parse, so if you'd rather not run it yourself, the hosted API gives you the same conversion plus our OCR models for the scanned pages anydoc can't read on its own.”免费部分是通向付费产品的一道门。","source":"ai","title":"免费工具紧挨着公司的付费服务"}],"platforms":["windows","mac","linux","web"],"tools":["node"],"buildTools":["rust"],"disk":{"minMB":8,"maxMB":24,"basis":"installer"},"apiKeys":["Firecrawl"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","development","ai_skills"],"alternativeTo":[],"declaredAlternatives":[],"stars":23477,"downloads":8215,"downloadsWeek":822,"downloadsDays":7,"growth":143,"growthDays":7,"history":[{"date":"2026-10-26","stars":23131},{"date":"2026-10-27","stars":23161},{"date":"2026-10-28","stars":23197},{"date":"2026-10-29","stars":23236},{"date":"2026-10-30","stars":23261},{"date":"2026-10-31","stars":23281},{"date":"2026-11-01","stars":23296},{"date":"2026-11-02","stars":23317},{"date":"2026-11-03","stars":23334},{"date":"2026-11-04","stars":23348},{"date":"2026-11-05","stars":23369},{"date":"2026-11-06","stars":23396},{"date":"2026-11-07","stars":23416},{"date":"2026-11-08","stars":23436},{"date":"2026-11-09","stars":23462},{"date":"2026-11-10","stars":23477}],"firstPublishedAt":"2026-10-27T01:10:12.627Z","commitSha":"e7178e099b6bf9171e2c52b832cbdb4134b69112","verifiedAt":"2026-11-03T02:44:23.870Z","lastCommitAt":"2026-10-23T22:55:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A fast converter that turns Word, PowerPoint, Excel, PDF, EPUB and other documents into clean Markdown text.","summary":"anydoc reads office documents and gives back Markdown: Word, PowerPoint, Excel, OpenDocument, RTF, EPUB, CSV and text-based PDFs, including tables, footnotes, speaker notes and equations. It is mostly meant for programmers, as a Rust, Node.js or Python package, but there is also a command you can run with one line and a demo page in the browser where files are converted on your own machine. It can be added to an AI assistant as a skill so it can read documents. Scanned PDFs are not read on your computer: those need the company's paid online service.","easeRationale":"The demo page at firecrawl.github.io/anydoc needs nothing at all: it runs the converter in the browser and the README says files \"are converted locally and never leave your machine\". On Windows, Mac and Linux you need Node.js (or Python, or Rust) and you type a command such as \"npx @firecrawl/anydoc report.docx\"; there is no window to open.","costNote":"The library and the command are free (MIT). Text recognition for scanned PDFs runs on Firecrawl's paid service, which the README says works without signing up, with \"FIRECRAWL_API_KEY for higher limits\".","notVerified":["Whether the Markdown comes out faithful for your own documents, since the scores were judged by an AI on a corpus nobody else has","What Firecrawl keeps or logs of a document sent for text recognition, and for how long","Whether the prebuilt binaries on npm, PyPI and the releases page were built from this code","How large a free text-recognition allowance you get before you need a key","How it behaves on very large or unusual files beyond the fixed internal limits"]},"zh-Hans":{"summaryShort":"一个快速转换器，把 Word、PowerPoint、Excel、PDF、EPUB 等文档转换成干净的 Markdown 文本。","summary":"anydoc 读取办公文档并输出 Markdown：Word、PowerPoint、Excel、OpenDocument、RTF、EPUB、CSV 以及基于文本的 PDF，包括表格、脚注、演讲者备注和公式。它主要面向程序员，以 Rust、Node.js 或 Python 包的形式提供，但也有一条命令可以直接运行，还有一个浏览器演示页面，文件会在你自己的机器上转换。它可以作为技能添加到 AI 助手里，让助手能读取文档。扫描版 PDF 不会在你本机读取：那部分需要公司的付费在线服务。","easeRationale":"firecrawl.github.io/anydoc 上的演示页面不需要任何安装：它在浏览器里运行转换器，README 说文件“are converted locally and never leave your machine”。在 Windows、Mac 和 Linux 上，你需要安装 Node.js（或 Python、Rust），然后输入一条命令，例如“npx @firecrawl/anydoc report.docx”；没有任何窗口需要打开。","costNote":"这个库和命令是免费的（MIT）。扫描版 PDF 的文字识别运行在 Firecrawl 的付费服务上，README 说不需要注册就能使用，“FIRECRAWL_API_KEY for higher limits”可以提高额度。","notVerified":["对你自己的文档来说，生成的 Markdown 是否忠实，因为评分是由一个 AI 在一份别人都没有的语料库上评出来的","送去做文字识别的文档，Firecrawl 会保留或记录哪些内容，保留多久","npm、PyPI 和发布页面上的预编译二进制文件是否确实是用这份代码编译出来的","在需要密钥之前，能获得多少免费的文字识别额度","在固定的内部限制之外，它处理超大或不常见文件时表现如何"]}}},{"id":"R_kgDOLunVBg","slug":"firecrawl__firecrawl","name":"Firecrawl","owner":"firecrawl","fullName":"firecrawl/firecrawl","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"self_host_no_auth_by_default","severity":"attention","evidence":"docker-compose.yaml：'USE_DB_AUTHENTICATION: ${USE_DB_AUTHENTICATION:-false}'、'HOST: \"0.0.0.0\"' 以及端口 '\"${PORT:-3002}:${INTERNAL_PORT:-3002}\"'：除非另行配置，你自己运行的 API 会在没有身份验证的情况下发布到网络上；Postgres 的默认密码是 'postgres'。","source":"ai","title":"自托管默认不要求身份验证"},{"code":"cloud_has_extra_features","severity":"info","evidence":"README：'The cloud version at firecrawl.dev includes additional features'。","source":"ai","title":"云端版本有额外功能"},{"code":"scraping_terms_responsibility","severity":"info","evidence":"README：'It is the sole responsibility of end users to respect websites' policies when scraping.'","source":"ai","title":"抓取时遵守网站条款是使用者的责任"},{"code":"latest_code_fetched_at_setup","severity":"info","evidence":"README 中的智能体设置：'npx -y firecrawl-cli@latest init --all --browser' 和 'curl -s https://firecrawl.dev/agent-onboarding/SKILL.md' 会在安装时拉取最新的命令行工具和说明，而不是固定版本。","source":"ai","title":"设置时获取最新代码"}],"platforms":["web","windows","mac","linux"],"tools":["docker"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Firecrawl cloud API key (hosted service; not needed when you run your own copy)","OpenAI or compatible LLM endpoint (AI extraction features when you run your own copy)"],"cost":{"model":"freemium"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["development","automation","ai_skills"],"alternativeTo":[{"slug":"bright-data","name":"Bright Data","confirmed":true}],"declaredAlternatives":[],"stars":200471,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":2141,"growthDays":7,"history":[{"date":"2026-10-26","stars":195704},{"date":"2026-10-27","stars":196084},{"date":"2026-10-28","stars":196429},{"date":"2026-10-29","stars":196769},{"date":"2026-10-30","stars":197117},{"date":"2026-10-31","stars":197415},{"date":"2026-11-01","stars":197696},{"date":"2026-11-02","stars":198014},{"date":"2026-11-03","stars":198330},{"date":"2026-11-04","stars":198615},{"date":"2026-11-05","stars":198941},{"date":"2026-11-06","stars":199220},{"date":"2026-11-07","stars":199493},{"date":"2026-11-08","stars":199774},{"date":"2026-11-09","stars":200121},{"date":"2026-11-10","stars":200471}],"firstPublishedAt":"2026-10-27T01:10:12.628Z","commitSha":"33f295dc1f09133c02fc1e096d6685abda3464a7","verifiedAt":"2026-10-29T01:39:17.108Z","lastCommitAt":"2026-11-10T18:42:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A web scraping service that turns websites into clean text or data that AI apps and agents can read.","summary":"Firecrawl searches the web, reads pages (including ones built with JavaScript) and returns clean Markdown, JSON or screenshots for AI agents. It can crawl whole sites, click through pages and parse PDFs. It is for developers building AI tools, used through code libraries (Python, Node.js and others), a CLI or an MCP server. The easiest path is the hosted service with an API key; running your own copy needs Docker and several services. The cloud version has extra features not in the open-source code. Users are responsible for respecting websites' rules.","easeRationale":"The easiest route is the web: sign up at firecrawl.dev and try it in the browser playground with an API key, no install. Running it on your own Windows, Mac or Linux machine is much harder: there are no ready downloads, and the docker-compose stack starts Redis, RabbitMQ, Postgres and a Playwright browser service.","costNote":"Running your own copy of the AGPL code is free; the hosted service at firecrawl.dev uses an API key and credits (prices not stated in the README) and has extra features.","notVerified":["Whether the Docker stack you run yourself builds and runs on Windows with Docker Desktop.","Real memory use when running it yourself; the compose file allows up to 8 GB for the API and 4 GB for the browser service.","Which features are missing from the open-source version compared with the cloud service.","Real cost of the hosted service for a given number of pages.","Whether the free tier of the hosted playground is enough for real work."]},"zh-Hans":{"summaryShort":"一个网页抓取服务,把网站内容转换成干净的文本或数据,供 AI 应用和代理读取。","summary":"Firecrawl 可以搜索网页、读取页面(包括用 JavaScript 构建的页面),并为 AI 代理返回干净的 Markdown、JSON 或截图。它能爬取整个网站、在页面间点击跳转,并解析 PDF。它适合构建 AI 工具的开发者,可通过代码库(Python、Node.js 等)、CLI 或 MCP 服务器使用。最简单的方式是使用托管服务并配一个 API 密钥;自己运行则需要 Docker 和多个服务。云端版本有一些开源代码里没有的额外功能。用户需要自行负责遵守各网站的规则。","easeRationale":"最简单的方式是通过网页:在 firecrawl.dev 注册,用 API 密钥在浏览器的演练场里试用,无需安装任何东西。要在自己的 Windows、Mac 或 Linux 电脑上运行就难得多:没有现成的下载包,docker-compose 组合需要同时启动 Redis、RabbitMQ、Postgres 和一个 Playwright 浏览器服务。","costNote":"自己运行 AGPL 代码是免费的;firecrawl.dev 上的托管服务需要 API 密钥和额度(README 未说明价格),并提供额外功能。","notVerified":["你自己运行的 Docker 组合能否在装有 Docker Desktop 的 Windows 上构建和运行。","自己运行时的实际内存占用;compose 文件允许 API 最多使用 8 GB,浏览器服务最多使用 4 GB。","与云端服务相比,开源版本缺少哪些功能。","处理一定数量页面时,托管服务的实际费用。","托管演练场的免费额度是否足够用于实际工作。"]}}},{"id":"R_kgDOT3Zahg","slug":"fireredteam__fireredtts3","name":"FireRedTTS3","owner":"FireRedTeam","fullName":"FireRedTeam/FireRedTTS3","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖存在已知漏洞(OSV):torch:GHSA-63cw-57p8-fm3p、GHSA-qfhq-4f3w-5fph、GHSA-rrmf-rvhw-rf47;flash_attn:GHSA-7g5w-pq96-8c5w、PYSEC-2026-2477;transformers:GHSA-x9r9-c232-4q39、GHSA-xrqw-3rrv-vx5w、PYSEC-2026-3929","source":"check","title":"已知漏洞"},{"code":"voice_cloning_and_putting_words_in_recordings","severity":"attention","evidence":"该模型可以从一小段参考录音中克隆声音,支持 24 种语言和 21 种中国方言,Instruct 模型还能在词语级别编辑真实音频:\"Semantic Edit ... Content-level editing: insertion / deletion / substitution by instruction\",示例是 \"Replace 'cats' with 'dogs'.\"。这意味着可以让一段真实人物的录音说出他们从未说过的话。README 没有提到生成的音频带有任何水印。","source":"ai","title":"声音克隆并篡改录音内容"},{"code":"authors_restrict_use_to_research","severity":"attention","evidence":"Usage Disclaimer 写道:\"The project incorporates zero-shot voice cloning functionality; Please note that this capability is intended solely for academic research purposes\"、\"DO NOT use this model for ANY illegal activities\" 以及 \"The developers assume no liability for any misuse of this model.\"代码本身采用 Apache-2.0 许可证,所声明的用途意图比许可证本身更窄。","source":"ai","title":"作者将用途限定为研究"},{"code":"model_files_come_from_outside_github","severity":"info","evidence":"权重从 Hugging Face(\"hf download FireRedTeam/FireRedTTS3 --local-dir pretrained_models/\")或 ModelScope 下载,可选的语言检测器则通过 \"curl -L -o fireredtts3/utils/llm_tn/models/lid.176.ftz https://dl.fbaipublicfiles.com/fasttext/...\" 获取。这些都不在仓库里,也没有说明文件大小。","source":"ai","title":"模型文件来自 github 之外"},{"code":"gradio_demo_binds_to_all_interfaces","severity":"info","evidence":"文档给出的命令是 \"python gradio_base.py --host 0.0.0.0 --port 7860\",它会在所有网络接口上响应且没有密码,因此在共享网络或办公网络中,其他设备也能访问这个声音克隆页面。","source":"ai","title":"gradio 演示监听所有网络接口"}],"platforms":["linux","windows"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["Other (LLM_TN_API_KEY)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true}],"declaredAlternatives":[],"stars":1720,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":2,"growthDays":6,"history":[{"date":"2026-11-07","stars":1719},{"date":"2026-11-08","stars":1719},{"date":"2026-11-09","stars":1720},{"date":"2026-11-10","stars":1720}],"firstPublishedAt":"2026-11-06T22:57:16.607Z","commitSha":"7a1f3a7282ff184cc1c7f070556baaf5f08b5216","verifiedAt":"2026-11-06T22:43:58.434Z","lastCommitAt":"2026-09-08T03:53:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that copies a voice from a short clip, reads text in 24 languages, and edits what a recording says.","summary":"FireRedTTS3 is the published code and model for a speech system from a Chinese research team. Give it a few seconds of someone speaking and it reads your text in that voice, across 24 languages and 21 Chinese dialects. A second model invents a voice from a written description (\"gentle young woman, slightly slow, a bit playful\") and edits existing recordings: swap, add or remove words, or change speed, pitch and volume. It is Python code with a small Gradio web page, aimed at people comfortable with machine-learning setups and a graphics card.","easeRationale":"A D on both: you clone the repository, \"pip install -r requirements.txt\" (which pins flash_attn 2.8.3, a package that normally wants an NVIDIA card and a compiler), download the weights with the Hugging Face or ModelScope command line, optionally fetch a language-detection file with curl, and then either write Python or start the Gradio page with \"python gradio_base.py --host 0.0.0.0 --port 7860\".","costNote":"Code and model weights are free (Apache-2.0 code). You pay in hardware or rented GPU time. The optional whole-language text normalisation calls a paid AI endpoint of your choice.","notVerified":["Whether the pinned packages (torch 2.8.0, flash_attn 2.8.3, transformers 5.6.2) install on your machine, and on which cards","How much video memory and disk the two models need, which the README never says (the 950-4450 MB figure is an estimate)","Whether the quality tables, which were produced by the authors and partly scored by Gemini, hold on your own recordings","Whether generated or edited audio carries any mark showing it is synthetic","What the optional LLM text normalisation sends to the endpoint you configure, and what that costs"]},"zh-Hans":{"summaryShort":"研究代码:从一小段录音中复刻声音,用 24 种语言朗读文本,并编辑录音里说的内容。","summary":"FireRedTTS3 是中国一个研究团队发布的语音系统代码和模型。给它几秒钟某人说话的录音,它就能用那个声音朗读你的文本,支持 24 种语言和 21 种中国方言。第二个模型可以根据一段文字描述(\"gentle young woman, slightly slow, a bit playful\")凭空生成一个声音,还能编辑已有的录音:替换、增加或删除词语,或者改变语速、音调和音量。这是一套带有小型 Gradio 网页的 Python 代码,面向熟悉机器学习环境搭建并拥有显卡的人群。","easeRationale":"两个系统上都是 D 级:你克隆仓库,运行 \"pip install -r requirements.txt\"(其中锁定了 flash_attn 2.8.3,这个包通常需要 NVIDIA 显卡和编译器),用 Hugging Face 或 ModelScope 的命令行下载权重,可以选择用 curl 获取一个语言检测文件,然后要么自己写 Python 代码,要么用 \"python gradio_base.py --host 0.0.0.0 --port 7860\" 启动 Gradio 页面。","costNote":"代码和模型权重是免费的(代码采用 Apache-2.0 许可证)。你需要自己承担硬件或租用 GPU 的费用。可选的全语言文本正则化功能会调用你自己选择的付费 AI 接口。","notVerified":["锁定版本的依赖包(torch 2.8.0、flash_attn 2.8.3、transformers 5.6.2)能否在你的机器上安装,以及适用于哪些显卡。","这两个模型各需要多少显存和磁盘空间,README 从未说明(950-4450 MB 这个数字只是估计值)。","由作者制作、部分由 Gemini 打分的质量表现,是否在你自己的录音上也同样成立。","生成或编辑后的音频是否带有任何标记,说明它是合成的。","可选的 LLM 文本正则化功能会向你配置的接口发送什么内容,以及这要花多少钱。"]}}},{"id":"R_kgDOKePR1A","slug":"fishaudio__fish-speech","name":"Fish Speech","owner":"fishaudio","fullName":"fishaudio/fish-speech","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"license_is_not_open_source","severity":"attention","evidence":"GitHub 没有显示任何被认可的许可证，LICENSE 文件是 \"FISH AUDIO COMMUNITY LICENSE AGREEMENT\"：商业使用只被允许 \"only while Your Annual Revenue is less than USD 500,000\"，\"no license is granted under this Agreement to offer a Hosted Service\"，并且该许可证是 \"revocable\"。README 还补充道：\"We will take action against any violation of the license.\"","source":"ai","title":"许可证不是开源许可证"},{"code":"model_weights_under_stricter_license","severity":"attention","evidence":"README 中的许可证说明写道：\"The S2-Pro weights remain under the Fish Audio Research License\"，这是一份和代码所用许可证不同、限制更严的许可证，所以这个更早的旗舰模型可能完全不能用于商业用途。","source":"ai","title":"模型权重受更严格的许可证约束"},{"code":"voice_cloning_can_be_misused","severity":"attention","evidence":"README 写道：\"Rapid voice cloning: 10-30 seconds of reference audio is enough to reproduce a timbre\" 以及 \"Multilingual voice cloning\"。未经允许复制别人的声音可能违反冒充他人、公开形象或版权方面的规定；README 自己的提示也说：\"We do not hold any responsibility for any illegal usage of the codebase. Please refer to your local laws about DMCA and other related laws.\"","source":"ai","title":"声音克隆可能被滥用"},{"code":"heavy_hardware_for_the_new_model","severity":"info","evidence":"模型列表把 Drama 3 列为 \"35B-A3B (3B active per token)\"，compose.yml 的注释写道 \"e.g. cuda:0 to keep Drama 3 on one GPU; unset uses up to four\"，说明这个旗舰模型是按多块显卡来设计的。S2-Pro（4B）才是被描述为小到可以在消费级 GPU 上运行的模型。","source":"ai","title":"新模型需要很重的硬件"},{"code":"newest_version_is_a_beta","severity":"info","evidence":"最新发布版本是 v3.0.0-beta（2026-10-13），再往前是 v2.0.0-beta；最后一个不带 beta 标记的版本是 2025-05-31 发布的 v1.5.1。","source":"ai","title":"最新版本是测试版"}],"platforms":["windows","linux"],"tools":["python","docker","gpu_nvidia"],"buildTools":["python","docker","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[{"target":"word","stance":"claims","evidence":"- Infilling and precise editing: edit audio the way you edit text — replace a single word (\"Wednesday\" → \"Thursday\") while the surrounding timbre and transitions stay natural"}],"stars":33665,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":84,"growthDays":6,"history":[{"date":"2026-11-05","stars":33597},{"date":"2026-11-06","stars":33599},{"date":"2026-11-07","stars":33611},{"date":"2026-11-08","stars":33626},{"date":"2026-11-09","stars":33640},{"date":"2026-11-10","stars":33665}],"firstPublishedAt":"2026-11-04T22:38:15.189Z","commitSha":"0f8a12e8a651062fdee9c33a894bdccd3a8513f1","verifiedAt":"2026-11-04T22:18:30.778Z","lastCommitAt":"2026-10-29T17:05:07Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns text into expressive speech on your own machine and copies a voice from 10-30 seconds of sample audio.","summary":"Fish Speech reads text aloud with the voice you give it: 10 to 30 seconds of sample audio is enough to copy a timbre, inline tags like [whisper] or [angry] steer the emotion, one script can hold several characters, and the newest model can replace a single word in an existing take. You run it yourself through a command line, a web page or an API server, with Python or Docker and an NVIDIA graphics card. The licence is not a normal open-source one: free for personal and research use, limited for business.","easeRationale":"Linux is the best-supported path and is still a D: there is no installer, you set up Python or Docker, download the model weights yourself and start a server or web page from the terminal, with an NVIDIA graphics card (AMD only through the separate ROCm compose file). Windows follows the same documented steps with the same requirements, so it is no easier.","costNote":"Free for research, hobby and evaluation use. Business use is only allowed while your yearly revenue is under USD 500,000; above that, and for any service that lets other people generate voices, you must buy a written licence from Fish Audio.","notVerified":["How much disk and video memory the weights really take: a 35-billion-parameter model is far larger than the checks' 0.95-4.45 GB estimate, which only covers the code and its dependencies.","Whether it runs at a usable speed on a normal home graphics card, since the published speed figures come from an H200 server card.","Whether the Windows instructions work as smoothly as the Linux and Docker ones.","How close the cloned voices and the benchmark scores are to what you get with your own recordings.","Whether your intended use falls inside the licence's free limits; that is a legal question about your own revenue and service."]},"zh-Hans":{"summaryShort":"在你自己的电脑上把文字变成富有感情的语音，只需 10 到 30 秒的样本录音就能复刻一个声音。","summary":"Fish Speech 会用你提供的声音把文字读出来：10到30秒的样本录音就足够复刻一种音色，像 [whisper] 或 [angry] 这样的行内标签可以控制情绪，一段脚本里可以有多个角色，最新的模型还能替换已有录音里的单个词。你可以通过命令行、网页或 API 服务器自己运行它，需要用到 Python 或 Docker，以及一块 NVIDIA 显卡。它的许可证不是常见的开源许可证：个人和研究用途免费，商业用途受到限制。","easeRationale":"Linux 是支持最好的平台，但依然只能打 D：没有安装程序，你需要自己配置 Python 或 Docker，自己下载模型权重，再从终端启动服务器或网页，并需要一块 NVIDIA 显卡（AMD 只能通过单独的 ROCm compose 文件）。Windows 走的是同样的文档步骤，要求也一样，所以并不会更简单。","costNote":"研究、业余爱好和评估用途免费。只要你的年收入低于 50 万美元，商业用途也可以使用；超过这个数额，或者任何让其他人生成语音的服务，都必须向 Fish Audio 购买书面许可证。","notVerified":["权重实际占用的磁盘和显存：这是一个350亿参数的模型，远大于检查结果给出的0.95到4.45 GB的估算，那个估算只涵盖了代码和它的依赖项。","它在普通家用显卡上能否以可用的速度运行，因为公布的速度数据来自 H200 服务器显卡。","Windows 的使用说明是否能像 Linux 和 Docker 那样顺畅。","克隆出的声音和基准测试分数，跟你用自己的录音得到的效果有多接近。","你打算做的用途是否落在许可证的免费范围内；这是一个关于你自己收入和服务方式的法律问题。"]}}},{"id":"R_kgDORSvIXA","slug":"floci-io__floci","name":"floci","owner":"floci-io","fullName":"floci-io/floci","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"needs_the_docker_socket_as_root","severity":"attention","evidence":"为了启动基于 Docker 的服务和网页控制台，README 和 docker-compose.yml 把主机的 Docker 套接字挂载进了容器：\"- /var/run/docker.sock:/var/run/docker.sock\"，文档里给出的运行命令还加上了 \"-u root\"。拥有这个套接字的容器可以控制你电脑上的所有容器和镜像。","source":"ai","title":"需要以 root 身份使用 Docker 套接字"},{"code":"reachable_from_the_network_with_any_credentials","severity":"attention","evidence":"README 写道：\"Credentials can be any non-empty values unless you explicitly enable stricter service-specific auth checks\"，FLOCI_SERVICES_UI_BIND_ADDRESS 的文档说明是 \"Unset publishes on every interface\"。仓库的 docker-compose.yml 发布了 4566 端口，以及 6379-6399、7001-7099 和 9200-9299 这些端口范围，所以任何能连到你电脑的设备都能使用它。","source":"ai","title":"用任意凭证就能从网络访问"},{"code":"downloads_and_runs_many_third_party_images","severity":"info","evidence":"默认镜像列表拉取了其他项目的软件，有些还使用会变动的标签：valkey/valkey:8、postgres:16-alpine、mysql:8.0、mongo:7.0、redpandadata/redpanda:latest、rancher/k3s:latest、registry:2、apache/airflow。控制台在第一次请求 /_floci/ui 时也会 \"pulls the console image\"（拉取控制台镜像）。","source":"ai","title":"下载并运行许多第三方镜像"},{"code":"native_binary_claim_not_in_the_latest_releases","severity":"info","evidence":"README 写道 \"Each release also ships a standalone floci-aws binary for Linux, macOS and Windows\"，但这里记录的最近三个发布版本（2.4.0、2.3.0 和 2.2.0）完全没有列出任何平台的可执行文件。","source":"ai","title":"最新发布版本里没有原生可执行文件"},{"code":"several_services_are_only_pretend","severity":"info","evidence":"服务说明里把一些部分标记为模拟实现：\"Textract: In-process stub\"，\"Transcribe: jobs complete immediately, no real audio processing\"，\"Bedrock Runtime: In-process stub - Dummy Converse and InvokeModel responses\"，\"AWS FIS: experiment execution is a safe control-plane simulation and does not inject faults\"。在这里能跑通的代码，在真实云端上仍然可能失败。","source":"ai","title":"部分服务只是做样子"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"localstack","name":"LocalStack","confirmed":true},{"slug":"amazon-web-services","name":"Web Services","confirmed":true},{"slug":null,"name":"Amazon Web Services (paid cloud accounts used for development and testing)","confirmed":false}],"declaredAlternatives":[{"target":"localstack","stance":"claims","evidence":"Floci is a drop-in replacement for LocalStack Community"}],"stars":31274,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":1763,"growthDays":6,"history":[{"date":"2026-11-05","stars":30323},{"date":"2026-11-06","stars":30840},{"date":"2026-11-07","stars":31000},{"date":"2026-11-08","stars":31088},{"date":"2026-11-09","stars":31197},{"date":"2026-11-10","stars":31274}],"firstPublishedAt":"2026-11-04T22:32:54.916Z","commitSha":"ee038af7721593e891c65764bdfdaca6e9bdde1d","verifiedAt":"2026-11-04T22:19:20.380Z","lastCommitAt":"2026-11-10T22:17:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs Amazon-shaped cloud services like S3, Lambda and DynamoDB on your own machine for development and tests.","summary":"Floci is a local stand-in for Amazon Web Services. You start one container and point the AWS command line, the SDKs, Terraform or your test suite at http://localhost:4566 instead of the real cloud, with no Amazon account and no token. Dozens of services are covered; the ones where fidelity matters (Lambda, RDS, ECS, EC2, Kafka, OpenSearch) start real Docker containers, and a browser console at /_floci/ui shows what you created. It is a tool for programmers: you need Docker and a terminal.","easeRationale":"All three platforms are a D and work the same way: install Docker, write a small compose file or use the separate command-line tool, then set environment variables and type AWS commands in a terminal. There is no installer or graphical setup, and nothing here is meant for someone who does not program.","costNote":"MIT licence, no account, token or paid tier; the README says \"always free\". You pay nothing, but the containers it starts use your own disk, memory and CPU.","notVerified":["How closely each emulated service matches the real Amazon service beyond the project's own test counts.","Whether a standalone binary is really available for your system, since the latest releases list none.","How much disk and memory the whole thing uses once all the extra Docker images are downloaded (the checks could not estimate it).","Whether the published speed and memory comparisons hold on your machine.","How quickly the project follows changes in the real Amazon APIs."]},"zh-Hans":{"summaryShort":"在你自己的电脑上运行类似 Amazon 的云服务，比如 S3、Lambda 和 DynamoDB，用于开发和测试。","summary":"Floci 是 Amazon Web Services 的本地替代品。你启动一个容器，把 AWS 命令行、各种 SDK、Terraform 或你的测试套件指向 http://localhost:4566，而不是真正的云端，不需要 Amazon 账号，也不需要令牌。它覆盖了几十种服务；对真实度要求高的那些（Lambda、RDS、ECS、EC2、Kafka、OpenSearch）会启动真正的 Docker 容器，浏览器里的 /_floci/ui 控制台会显示你创建的内容。这是一个给程序员用的工具：你需要 Docker 和终端。","easeRationale":"三个平台的评级都是 D，用法也一样：安装 Docker，写一个小的 compose 文件，或者用单独的命令行工具，然后设置环境变量，在终端里输入 AWS 命令。没有安装程序，也没有图形界面设置，这一切都不是为不会写代码的人准备的。","costNote":"MIT 许可证，不需要账号、令牌或付费套餐；README 里写着 \"always free\"（始终免费）。你不用付钱，但它启动的容器会占用你自己的磁盘、内存和 CPU。","notVerified":["每个模拟服务跟真实的 Amazon 服务到底有多接近，项目自己的测试数量之外没有别的说明。","你的系统上是否真的能用上独立的可执行文件，因为最新的几个发布版本里都没有列出任何可执行文件。","下载完所有额外的 Docker 镜像之后，整套东西会占用多少磁盘和内存（检查工具无法估算这一点）。","公布的速度和内存对比数据，在你自己的机器上是否也一样。","这个项目跟进 Amazon 真实 API 变化的速度有多快。"]}}},{"id":"R_kgDOO2jIYw","slug":"fluxions-ai__vui","name":"Vui","owner":"fluxions-ai","fullName":"fluxions-ai/vui","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道传给 shell 执行:curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"curl 管道直接执行 shell 脚本"},{"code":"voice_cloning_of_any_speaker","severity":"attention","evidence":"README 写道 \"The model can clone arbitrary voices — upload a sample in the demo UI (or drop a .wav into prompts/) and it will follow that speaker\",命令行方式则是 \"Point --prompt at any .wav. No transcript needed\"。项目自己关于滥用风险的说明写道,该 checkpoint \"is fine-tuned to a curated voice set in part to make these misuses harder, but it is not a substitute for your own judgment\"。未经同意复制别人的声音可能触犯法律。","source":"ai","title":"可克隆任意说话人的声音"},{"code":"install_one_liner_runs_remote_code","severity":"attention","evidence":"文档里给出的安装方式是 \"curl -fsSL https://install.fluxions.ai | bash\",README 说它会 \"Clones into ~/vui, auto-detects Docker vs. native, installs deps (uv, ffmpeg libs, Claude Code CLI), and launches the stack\"。你运行命令那一刻,这个地址提供的任何内容都会在你的电脑上执行,且未经过审查。","source":"ai","title":"一行安装命令会执行远程代码"},{"code":"optional_helper_gets_your_mail_and_claude_login","severity":"attention","evidence":"编排文件默认启动一个 \"claude-task\" 边车容器,并把你的登录信息挂载进去:\"${HOME}/.claude:/root/.claude:ro\",注释写着 \"Claude Code subscription credentials (after claude login on host)\",或者使用 ANTHROPIC_API_KEY。README 说它通过你现有的 Claude Code MCP 处理 \"Gmail / Calendar / Drive / Slack reads, web research\",所以一句说出口的话就可能读取你的邮件。","source":"ai","title":"可选的辅助程序能访问你的邮件和 claude 登录信息"},{"code":"server_has_no_login_and_tunnels_are_documented","severity":"attention","evidence":"这些服务以 \"network_mode: host\" 方式运行在 8080、8443 和 8642 端口上,文档中完全没有提到任何密码,手机那一节还建议使用 \"cloudflared tunnel --url http://localhost:8080 — one command, HTTPS, no account\"。这会把一个由麦克风驱动的助手、你的记忆数据和任务工具暴露给任何知道这个地址的人。","source":"ai","title":"服务器没有登录验证,文档里还给出了隧道方法"},{"code":"what_you_say_is_written_to_disk","severity":"info","evidence":"这个助手会在不同会话之间保存关于你的事实:\"Memories are loaded from ~/.vui/memories.json on startup and rewritten on every add/remove\",编排文件还挂载了 \"./debug_dump:/app/debug_dump\",注释是 \"Optional: write conversation logs to host for inspection\"。","source":"ai","title":"你说的话会被写入磁盘"},{"code":"weights_download_separately_and_one_set_is_unpublished","severity":"info","evidence":"README 写道 \"The TTS weights are not pulled here — they download from Hugging Face on first render\",编排文件里还覆盖了启动命令,注释是 \"load a local checkpoint instead of fetching vui-nano.safetensors from huggingface.co/fluxions/vui (not yet published)\" 和 \"Drop this line once the HF repo has the v2 weights\"。","source":"ai","title":"权重文件需单独下载,其中一套尚未发布"}],"platforms":["linux","mac"],"tools":["python","docker","gpu_nvidia","ffmpeg"],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["Anthropic (or a Claude Code subscription), only for the optional task helper that reads mail and calendar","Serper, Brave or Tavily, only for spoken web searches"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["audio","local_ai","ai_agents"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":false}],"declaredAlternatives":[{"target":"make","stance":"denies","evidence":"The released checkpoint is fine-tuned to a curated voice set in part to make these misuses harder, but it is not a substitute for your own judgment"}],"stars":776,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":777},{"date":"2026-11-09","stars":776},{"date":"2026-11-10","stars":776}],"firstPublishedAt":"2026-11-07T22:56:16.774Z","commitSha":"9dbd98db155e4c2f4116f7cd2898e82314b610c5","verifiedAt":"2026-11-07T22:44:09.194Z","lastCommitAt":"2026-10-10T11:01:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A talking assistant you run on your own machine: speak into the mic in your browser and a local AI answers out loud.","summary":"Two things in one repository: a small speech model that reads text out loud and can copy a voice from a sample, and a complete voice assistant built on it. You open a page on your own machine, talk into the microphone, and it transcribes, asks a language model you run (Ollama or vLLM) and answers out loud, with interruption, memories and timers. It is built for Linux with an NVIDIA card and installs through Docker or a shell script; the Apple Silicon side runs the speech demo but the full assistant there is unfinished. There is also a C version that speaks on a plain processor.","easeRationale":"Linux with an NVIDIA card is the intended setup and still a D: you need Docker with the Compose plugin, the NVIDIA Container Toolkit (installed with sudo), Ollama running on the machine with a model pulled, and the ffmpeg shared libraries, or you run the shell installer and let it do it. On a Mac it is also a D and less complete: the README says the speech demo works on Apple Silicon but \"the rest of the MLX stack is WIP\", including the streaming server. Windows is not mentioned.","costNote":"Apache 2.0 with the weights included and commercial use allowed. The authors also sell a hosted version: the README calls this repository \"the open core\" and says their production API \"ships ongoing model updates and a more advanced turn-taking system\".","notVerified":["Whether the install finishes on your machine, as it builds Python, CUDA and ffmpeg pieces at install time.","The speed and accuracy tables (9x real time on a 4090, word error rates), measured by the authors.","How natural the voice and the turn-taking feel in a real conversation, and how well a cloned voice holds up.","Whether the checkpoints the code asks for are actually published on Hugging Face right now.","Real disk, memory and video-memory use once the speech model, the speech-recognition model and a language model are all loaded.","What the hosted demo at the authors' site does with the audio you send it."]},"zh-Hans":{"summaryShort":"一个会说话的助手,运行在你自己的电脑上:对着浏览器里的麦克风说话,本地 AI 会用语音回答。","summary":"同一个仓库里包含两样东西:一个能朗读文本、还能从样本克隆声音的小型语音模型,以及一个建立在它之上的完整语音助手。你在自己的电脑上打开一个网页,对着麦克风说话,它会转写语音、询问你自己运行的语言模型(Ollama 或 vLLM),再用语音回答,支持打断、记忆和定时器。它主要面向配有 NVIDIA 显卡的 Linux,通过 Docker 或一个 shell 脚本安装;在 Apple Silicon 上只有语音演示能用,完整的助手部分还没完成。另外还有一个纯用处理器朗读的 C 语言版本。","easeRationale":"配有 NVIDIA 显卡的 Linux 是预期的使用环境,仍然属于 D 级:你需要带 Compose 插件的 Docker、NVIDIA Container Toolkit(用 sudo 安装)、在机器上运行并已拉取模型的 Ollama,以及 ffmpeg 的共享库,或者运行 shell 安装脚本让它自动完成这些。在 Mac 上也是 D 级,而且更不完整:README 说语音演示可以在 Apple Silicon 上运行,但 \"the rest of the MLX stack is WIP\",包括流式服务器在内。文档里完全没有提到 Windows。","costNote":"采用 Apache 2.0 许可证,附带权重文件,允许商业使用。作者也出售托管版本:README 把这个仓库称为 \"the open core\",并说他们的生产环境 API \"ships ongoing model updates and a more advanced turn-taking system\"。","notVerified":["安装是否能在你的电脑上顺利完成,因为安装过程中要构建 Python、CUDA 和 ffmpeg 相关组件。","README 里的速度和准确率数据表(在 4090 上达到 9 倍实时速度、词错误率等),是作者自己测出来的。","在真实对话里,声音和对话轮换的自然程度如何,以及克隆出来的声音效果能保持多好。","代码里要求的 checkpoint 文件目前是否真的已经发布在 Hugging Face 上。","当语音模型、语音识别模型和一个语言模型同时加载时,实际占用的磁盘、内存和显存。","作者自家网站上托管的演示会如何处理你发送过去的音频。"]}}},{"id":"R_kgDOTxiuFw","slug":"forgemind-git__forgegrowth-oss","name":"ForgeGrowth OSS","owner":"Forgemind-git","fullName":"Forgemind-git/ForgeGrowth-OSS","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"install_runs_script_from_internet","severity":"attention","evidence":"文档记录的快速开始方式是`bash -c \"$(curl -fsSL https://raw.githubusercontent.com/Forgemind-git/ForgeGrowth-OSS/v3.1.1/scripts/install.sh)\" -- --version v3.1.1 --domain crm.example.com --yes`,这会从网络下载一个脚本并在你的服务器上运行。README还提供了一个现成的提示词,可以把整个安装过程交给通过SSH连接的Claude Code来完成。","source":"ai","title":"安装过程会运行来自网络的脚本"},{"code":"needs_meta_business_approval","severity":"attention","evidence":"这款应用运行在\"Meta's WhatsApp Cloud API, on your own WhatsApp Business number\"之上,消息模板需要经过\"The full Meta lifecycle ... submit, sync, quality score\",而`docs/social-section.md`被描述为\"The Meta app, Instagram login, webhooks and the privacy-policy URL\"。README还警告说,没有HTTPS的话就\"not for WhatsApp — Meta requires HTTPS for webhooks\"。","source":"ai","title":"需要meta商业账号的审批"},{"code":"very_new_project","severity":"attention","evidence":"该代码库创建于2026-08-07,有21个星标,最近的三个发布版本(v3.0.0、v3.1.0和v3.1.1)都在同一天发布,即2026-10-26。没有其他人使用过它的公开记录。","source":"ai","title":"项目非常新"},{"code":"losing_key_loses_stored_credentials","severity":"info","evidence":".env.example里关于FORGECRM_ENCRYPTION_KEY写道:\"LOSING THIS MAKES EVERY STORED CREDENTIAL UNRECOVERABLE. Back it up. CHANGING IT does not re-encrypt anything — existing rows become garbage.\"同样的警告也适用于SOCIAL_ENCRYPTION_KEY。","source":"ai","title":"丢失密钥会导致已存储的凭据全部丢失"},{"code":"stores_customer_personal_data","severity":"info","evidence":"这款应用会保存联系人、Instagram和WhatsApp的对话记录、下载的媒体文件,以及从评论者那里收集的\"name/email/phone into a Google Sheet\",还支持CSV/XLSX导入。这些都是别人的个人数据,存放在你运行、也必须由你来保护的服务器上。","source":"ai","title":"存储客户的个人数据"},{"code":"authors_sell_paid_customisation","severity":"info","evidence":"README.md有一个\"Want it customised?\"板块,附上了一个WhatsApp号码(+91 93422 45724)和contact@forgemind.in,应用内部\"above the sidebar's Collapse control\"的位置也出现了同样的卡片。软件本身采用MIT许可,\"no SaaS tier\"。","source":"ai","title":"作者出售付费定制服务"},{"code":"ai_agents_can_spend_money_and_message_people","severity":"info","evidence":"MCP部分说有\"93 tools in categories you switch on one by one\",按能做的事情分类标注——\"reads, builds, reaches customers, costs money, or deletes\"——默认\"Everything starts off\",Claude构建出来的任何东西都会保存为草稿。","source":"ai","title":"ai代理可以花钱并给人发消息"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Meta WhatsApp Cloud API","Instagram / Meta app","OpenAI","Anthropic","Google (Sheets and Calendar)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation","ai_agents"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[],"stars":22,"downloads":24,"downloadsWeek":1,"downloadsDays":4,"growth":2,"growthDays":6,"history":[{"date":"2026-11-06","stars":21},{"date":"2026-11-07","stars":21},{"date":"2026-11-08","stars":21},{"date":"2026-11-09","stars":22},{"date":"2026-11-10","stars":22}],"firstPublishedAt":"2026-11-05T22:43:48.402Z","commitSha":"ec03a0f43a47f83d391c58e2735623051252d7dd","verifiedAt":"2026-11-05T22:25:26.609Z","lastCommitAt":"2026-10-26T11:27:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs on your own server and answers Instagram and WhatsApp messages for you, with a contact list and follow-ups.","summary":"A business messaging and sales tool you install on your own server. It replies to Instagram comments and DMs with rules or an AI agent, builds WhatsApp conversations on a drag-and-drop canvas using Meta's WhatsApp Cloud API, and keeps contacts, a sales funnel and a follow-up calendar. It can also be connected to Claude as an MCP server. You need Docker, a terminal, about 3 GB of disk, a domain with HTTPS for the messaging parts, and a Meta business account.","easeRationale":"Linux is the smoothest path and still a D: you need Docker with the Compose v2 plugin, a terminal, and one install command, after which the app opens in a browser. Mac is the same after installing Docker Desktop; Windows is the hardest because the installer is a shell script that only runs inside WSL2 — the README says there is \"no PowerShell or `.bat` installer\". For the messaging features you also need a domain pointing at the machine and free ports 80 and 443.","costNote":"The code is free (MIT). WhatsApp sending is billed by Meta per conversation — the README says the app shows \"what every send actually costs\" — and the AI agents need your own OpenAI or Anthropic key. The authors also sell paid customisation.","notVerified":["Whether the one-command installer really works on a fresh server and on WSL2","Whether the Instagram and WhatsApp features behave as the screenshots show, since both depend on approvals from Meta","What Meta will charge you for the WhatsApp messages the app sends","What an AI agent costs per conversation with an OpenAI or Anthropic key","Real disk and memory use beyond the stated ~3 GB and ~1.5 GB of container limits","Whether the security measures described in the README (encryption at rest, rate limits) hold up in the code"]},"zh-Hans":{"summaryShort":"部署在你自己的服务器上,代你回复Instagram和WhatsApp的消息,还有联系人列表和跟进提醒。","summary":"一款商业消息和销售工具,安装在你自己的服务器上。它可以用规则或AI代理回复Instagram的评论和私信,用Meta的WhatsApp Cloud API在拖放式画布上搭建WhatsApp对话流程,并维护联系人、销售漏斗和跟进日程。它还可以作为MCP服务器连接到Claude。你需要Docker、一个终端、大约3GB的磁盘空间、一个带HTTPS的域名(用于消息相关功能),以及一个Meta商业账号。","easeRationale":"Linux是最顺畅的途径,但仍然是D级:你需要带Compose v2插件的Docker、一个终端,以及一条安装命令,之后应用就会在浏览器中打开。安装好Docker Desktop后,Mac的流程相同;Windows最难,因为安装程序是一个只能在WSL2里运行的shell脚本——README说\"no PowerShell or `.bat` installer\"。要使用消息功能,你还需要一个指向这台机器的域名,以及空闲的80和443端口。","costNote":"代码是免费的(MIT)。WhatsApp的发送费用由Meta按对话计费——README说应用会显示\"what every send actually costs\"——而AI代理需要你自己的OpenAI或Anthropic密钥。作者还出售付费定制服务。","notVerified":["这条一键安装命令在全新服务器和WSL2上是否真的管用","Instagram和WhatsApp功能是否真如截图所示那样运作,因为两者都依赖Meta的审批","应用发送的WhatsApp消息,Meta会向你收取多少费用","使用OpenAI或Anthropic密钥时,一个AI代理每次对话要花多少钱","除了声明的约3GB磁盘和约1.5GB容器限制之外,实际的磁盘和内存使用情况","README里描述的保护机制(静态加密、速率限制)是否真的体现在代码里"]}}},{"id":"R_kgDORf475A","slug":"forget-c__jellyfish","name":"Jellyfish","owner":"Forget-C","fullName":"Forget-C/Jellyfish","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"development_appears_stalled","severity":"attention","evidence":"检查记录显示最后一次提交是在 2026-04-20，也就是本次分析之前的 198 天，最后一次发布是在 2026-04-17。最新版本是 v0.3.2，而仓库本身是在 2026-03-06 才创建的，所以这还是一个早期版本，而且已经好几个月没有变化了。","source":"ai","title":"开发看起来已经停滞"},{"code":"running_cost_is_on_you_and_unstated","severity":"attention","evidence":"README 里列出的能力包括 \"multi-provider / multi-model management\"、\"image and video generation tasks\" 和 \"single-shot and batch pre-checks\"：每一集每一个镜头都是一次付费调用外部模型。仓库里没有任何地方说明支持哪些服务商，也没有说明做完一部短剧要花多少钱。","source":"ai","title":"使用成本由你自己承担，且未说明"},{"code":"feature_list_reads_as_work_in_progress","severity":"info","evidence":"README 对自己的范围留有余地：\"Preparation currently supports\"，\"Current task infrastructure supports\"，\"Provide a stable base for future AI workflow expansion\"。部分工作流程被描述为打基础，而不是已经完成的功能。","source":"ai","title":"功能列表读起来像是还在开发中"},{"code":"several_services_published_on_your_machine","severity":"info","evidence":"Docker Compose 部分列出了端口 7788 上的前端、端口 8000 上的后端、端口 3306 上的 MySQL、端口 6379 上的 Redis，以及端口 9000 上的 RustFS 文件存储，这些都是从复制的 .env.example 启动的。README 里没有说明网页有任何登录或访问控制。","source":"ai","title":"多个服务暴露在你的电脑上"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["An AI model provider of your choice for text, image and video generation (set up inside the app)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","image","social_marketing"],"alternativeTo":[{"slug":"ltx-studio","name":"LTX Studio","confirmed":true},{"slug":"invideo-ai","name":"AI","confirmed":true}],"declaredAlternatives":[],"stars":7167,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":46,"growthDays":6,"history":[{"date":"2026-11-05","stars":7124},{"date":"2026-11-06","stars":7127},{"date":"2026-11-07","stars":7134},{"date":"2026-11-08","stars":7144},{"date":"2026-11-09","stars":7153},{"date":"2026-11-10","stars":7167}],"firstPublishedAt":"2026-11-04T22:32:54.923Z","commitSha":"a9678194ddf2d9be3ccbe78d4287d87d5089e123","verifiedAt":"2026-11-04T22:20:10.097Z","lastCommitAt":"2026-04-20T05:41:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A workspace that turns a script into shots, characters and scenes, then orders the images and videos from AI models.","summary":"Jellyfish is a self-hosted studio for short vertical dramas. You paste a script and it splits it into shots and pulls out characters, scenes, props, costumes and dialogue; you confirm each one, keep a shared library so the same character looks the same across shots, then start image and video jobs and follow them in a task centre. The picture and video making is done by AI providers you connect and pay for yourself. You run it with Docker Compose and open a web page on your own machine.","easeRationale":"All three platforms land on D through the same route: install Docker, copy the example .env, run \"docker compose up --build\" and wait for MySQL, Redis, a file store, the backend and the frontend to come up, then open http://localhost:7788 and add your own model providers. There is no installer and no ready binary in any release.","costNote":"The software is free (Apache-2.0), but it does not generate anything on its own: you plug in your own AI providers for text, images and video and pay them per job. The repository gives no prices.","notVerified":["Whether a script really comes out the other end as a finished video, since the repository shows only two screenshots and no sample output.","Which AI providers actually work with it, and how much a short drama costs to generate.","How much disk the database, uploads and generated media use (the checks could not estimate it).","Whether the web page and prompts are fully translated, since the documentation is written in English, Japanese and Chinese in places.","Whether the project is still maintained after months without commits."]},"zh-Hans":{"summaryShort":"一个工作台，把剧本拆成镜头、角色和场景，再向 AI 模型下单生成图片和视频。","summary":"Jellyfish 是一个自行部署的竖屏短剧工作室。你粘贴一段剧本，它会把剧本拆分成镜头，并提取出角色、场景、道具、服装和台词；你逐一确认后，系统会维护一个共享素材库，让同一个角色在不同镜头里保持一致，然后你可以发起图片和视频生成任务，并在任务中心跟踪进度。图片和视频的生成由你自己接入并付费使用的 AI 服务商完成。你用 Docker Compose 运行它，然后在自己的电脑上打开一个网页。","easeRationale":"三个平台都靠同一条路线拿到 D 级评分：安装 Docker，复制示例 .env 文件，运行 \"docker compose up --build\"，等待 MySQL、Redis、文件存储、后端和前端启动，然后打开 http://localhost:7788 并添加你自己的模型服务商。没有安装程序，任何发布版本里也没有现成的可执行文件。","costNote":"软件本身免费（Apache-2.0），但它自己并不生成任何内容：你需要接入自己的文本、图片和视频 AI 服务商，并按任务付费。仓库里没有给出价格。","notVerified":["一段剧本是否真的能在另一端变成一部成片，因为仓库里只有两张截图，没有任何示例成品。","哪些 AI 服务商真的能跟它配合使用，以及生成一部短剧要花多少钱。","数据库、上传文件和生成的媒体文件会占用多少磁盘空间（检查工具无法估算这一点）。","网页和提示语是否完全翻译到位，因为文档部分是用英语、日语和中文写的。","在几个月没有提交代码之后，这个项目是否还在维护。"]}}},{"id":"R_kgDORWQCsw","slug":"fortunatoman__discord-ai-chatbot","name":"Discord AI Chatbot","owner":"fortunatoman","fullName":"fortunatoman/Discord-AI-Chatbot","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单均未检测到任何许可证。","source":"check","title":"没有许可证"},{"code":"uses_bot_detection_evasion","severity":"attention","evidence":"第二个脚本需要 \"undetected-chromedriver\"(\"pip install undetected-chromedriver\"),这个包的作用就是让自动化浏览器看起来像人类访问者。用这种方式操控一个 AI 网站违反了这类网站的服务条款,可能导致账号或地址被封。","source":"ai","title":"使用了反机器人检测规避手段"},{"code":"unnamed_ai_service_receives_your_users_messages","severity":"attention","evidence":"项目描述说它 \"uses Selenium-driven automation to generate intelligent responses\",README 从未说明它操控的是哪个网站,所以你无法知道服务器里每个人的 Discord 消息,包括私信,具体被输入到了哪里。","source":"ai","title":"一个未具名的 ai 服务接收你用户的消息"},{"code":"token_pasted_into_the_source_file","severity":"attention","evidence":"使用方法:\"Replace 'BOT TOKEN HERE' with your bot's token (while keeping '') and execute the script.\"Discord 机器人令牌被直接写在 discordbot.py 里,而不是单独的配置文件中,这样很容易被不小心分享出去或提交到版本库;谁拿到这个令牌,谁就能控制你的机器人。","source":"ai","title":"令牌被直接粘贴进源文件"},{"code":"commit_dates_do_not_match_the_repository","severity":"attention","evidence":"GitHub 显示该仓库创建于 2026-02-22,最后一次推送在 2026-07-19,但最新的提交日期却是 2024-08-20,顶层目录还带有 \"committer.log\"、\"progress.json\" 和 \"notes.txt\",而 README 结尾写着 \"Last updated: 2026-02-22 14:32:33\"。页面显示的活动与代码自身的历史对不上。","source":"ai","title":"提交日期与仓库记录对不上"},{"code":"scripts_duplicated_without_structure","severity":"info","evidence":"该仓库有 discordbot.py、discordbot2.py、一个 \"Version 3/\" 文件夹和一个 \"Repl.it/\" 文件夹,没有依赖清单文件,没有 release,也没有许可证,因此没有唯一可安装的版本,也没有声明允许重用代码。","source":"ai","title":"脚本重复且没有结构"}],"platforms":["windows"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Discord bot token"],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["ai_agents","social_marketing"],"alternativeTo":[{"slug":"chatbase","name":"Chatbase","confirmed":true}],"declaredAlternatives":[],"stars":18,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":18},{"date":"2026-11-08","stars":18},{"date":"2026-11-09","stars":18},{"date":"2026-11-10","stars":18}],"firstPublishedAt":"2026-11-06T22:57:16.614Z","commitSha":"bc0b0bcb79d9387a8b15dc32a9856ae94148b53e","verifiedAt":"2026-11-06T22:44:52.724Z","lastCommitAt":"2024-08-20T16:25:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A small Python script that answers Discord messages with AI by driving a Chrome browser in the background.","summary":"Two short Python scripts that log in as your own Discord bot and reply to messages in servers and in direct messages. Instead of calling an AI service through an official interface, the script opens Google Chrome with Selenium and reads the answers off a web page, so a browser has to stay running. Setup is manual: install Python packages, put a matching chromedriver.exe next to the script, paste your Discord bot token into the source file and run it. The repository is a small personal project with 18 stars, no license file and no release.","easeRationale":"Windows is the only system the instructions cover (they tell you to download \"chromedriver_win32.zip\"), and it is a D: create a bot in the Discord developer portal, install Python with discord.py, selenium and undetected-chromedriver, match a ChromeDriver to your Chrome version, then edit the script to \"Replace 'BOT TOKEN HERE' with your bot's token\" before running it. The README itself says \"The steps below are old\" and points to a video instead.","costNote":"No charge and no license file. The AI answers come from a website driven by the browser, not from a paid interface, so there is no stated cost.","notVerified":["Whether it still works at all: the code was last changed in August 2024 and browser automation breaks whenever the target website changes","Which AI website it drives and what that site's rules say about automated use","What happens to the private messages your users send the bot","Whether the scripts in discordbot.py, discordbot2.py and Version 3/ do the same thing","Disk and memory use, which the checks could not measure, plus the need to keep Chrome open while the bot runs"]},"zh-Hans":{"summaryShort":"一个小型 Python 脚本,在后台操控 Chrome 浏览器,用 AI 回复 Discord 消息。","summary":"两个简短的 Python 脚本以你自己的 Discord 机器人身份登录,在服务器和私信中回复消息。脚本没有通过官方接口调用 AI 服务,而是用 Selenium 打开 Google Chrome,从网页上读取回答,因此必须一直保持浏览器运行。配置过程是手动的:安装 Python 依赖包,把匹配的 chromedriver.exe 放在脚本旁边,把你的 Discord 机器人令牌粘贴进源文件,然后运行它。该仓库是一个小型个人项目,有 18 颗星,没有许可证文件,也没有发布 release。","easeRationale":"说明只覆盖了 Windows 系统(要求下载 \"chromedriver_win32.zip\"),等级为 D:你在 Discord 开发者门户创建一个机器人,安装带有 discord.py、selenium 和 undetected-chromedriver 的 Python,让 ChromeDriver 的版本与你的 Chrome 匹配,然后修改脚本,\"Replace 'BOT TOKEN HERE' with your bot's token\",之后再运行。README 本身写道 \"The steps below are old\",并建议改看一段视频。","costNote":"不收费,也没有许可证文件。AI 回答来自浏览器操控的网站,而不是付费接口,所以没有说明任何费用。","notVerified":["它是否还能正常工作:代码最后一次修改是在 2024 年 8 月,而浏览器自动化只要目标网站一变就会失效。","它操控的是哪个 AI 网站,以及该网站关于自动化使用的规定是什么。","用户发给机器人的私信内容会发生什么。","discordbot.py、discordbot2.py 和 Version 3/ 里的脚本是否做的是同一件事。","磁盘和内存占用,检查无法测量,另外机器人运行时还需要一直保持 Chrome 打开。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1NzM2MDgw","slug":"freecad__freecad","name":"FreeCAD","owner":"FreeCAD","fullName":"FreeCAD/FreeCAD","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"newest_downloads_are_development_builds","severity":"info","evidence":"最近的三个发布版本是“weekly-2026.10.28”、“weekly-2026.10.21”和“26.3rc2”——两个每周开发构建和一个发布候选版。README 说预编译的稳定版包在“latest releases page”，每周的开发版本是单独列出的，所以直接下载列表最上面的文件，得到的会是一个未经测试的构建。","source":"ai","title":"最新的下载版本是开发版构建"},{"code":"large_backlog_of_open_reports","severity":"info","evidence":"这个仓库有 4046 个未解决的 issue，还有 6331 个 fork 和今天的提交。对于这么老、这么大的项目来说这很正常，但这也意味着已知的缺陷会长时间得不到解决。","source":"ai","title":"大量未解决的问题积压"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":415,"maxMB":1245,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"LGPL-2.1","category":"copyleft","detectedFrom":"github"},"categories":["design_cad","development"],"alternativeTo":[{"slug":"autocad","name":"AutoCAD","confirmed":true},{"slug":"solidworks","name":"SolidWorks","confirmed":true},{"slug":"autodesk-fusion","name":"Fusion","confirmed":true}],"declaredAlternatives":[],"stars":35342,"downloads":32247637,"downloadsWeek":259569,"downloadsDays":7,"growth":248,"growthDays":7,"history":[{"date":"2026-10-25","stars":34778},{"date":"2026-10-26","stars":34820},{"date":"2026-10-27","stars":34851},{"date":"2026-10-28","stars":34897},{"date":"2026-10-29","stars":34928},{"date":"2026-10-30","stars":34970},{"date":"2026-10-31","stars":35004},{"date":"2026-11-01","stars":35032},{"date":"2026-11-02","stars":35071},{"date":"2026-11-03","stars":35094},{"date":"2026-11-04","stars":35129},{"date":"2026-11-05","stars":35161},{"date":"2026-11-06","stars":35186},{"date":"2026-11-07","stars":35227},{"date":"2026-11-08","stars":35255},{"date":"2026-11-09","stars":35306},{"date":"2026-11-10","stars":35342}],"firstPublishedAt":"2026-10-25T23:19:01.108Z","commitSha":"f69b1eb064874749739b87bf3d951985ce51bfe3","verifiedAt":"2026-11-03T02:45:01.456Z","lastCommitAt":"2026-11-10T21:02:08Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"FreeCAD_1.1.4-Windows-x86_64-py311-installer.exe","sizeBytes":519314976,"url":"https://github.com/FreeCAD/FreeCAD/releases/download/1.1.4/FreeCAD_1.1.4-Windows-x86_64-py311-installer.exe","releaseUrl":"https://github.com/FreeCAD/FreeCAD/releases/tag/1.1.4"},{"os":"mac","file":"FreeCAD_1.1.4-macOS-arm64-py311.dmg","sizeBytes":649760538,"url":"https://github.com/FreeCAD/FreeCAD/releases/download/1.1.4/FreeCAD_1.1.4-macOS-arm64-py311.dmg","releaseUrl":"https://github.com/FreeCAD/FreeCAD/releases/tag/1.1.4"},{"os":"linux","file":"FreeCAD_1.1.4-Linux-x86_64-py311.AppImage","sizeBytes":820812280,"url":"https://github.com/FreeCAD/FreeCAD/releases/download/1.1.4/FreeCAD_1.1.4-Linux-x86_64-py311.AppImage","releaseUrl":"https://github.com/FreeCAD/FreeCAD/releases/tag/1.1.4"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free program for designing real objects in 3D, from machine parts to buildings, and making measured drawings of them.","summary":"FreeCAD is a 3D design program for real objects of any size. You draw a flat shape with exact measurements and build on it, and because the model keeps its history you can change a number later and the whole part follows. It also turns a 3D model into a dimensioned production drawing. People use it for product design, mechanical engineering, architecture, 3D printing, simulation and machining, and it can be driven by Python. Installers are ready for Windows, Mac and Linux, and on most Linux systems it is in the software centre. The download is 415 MB or more.","easeRationale":"Every system gets a ready package with nothing to install alongside it: a 415 MB installer on Windows, a DMG on Mac for both Intel and Apple chips, and an AppImage on Linux, where it is also in the software centre. Learning to use it is another matter, but getting it running is one download.","notVerified":["Whether drawings and models from other CAD programs import without losing geometry","How it holds up on a large assembly with many parts","Which community add-on workbenches you will need, and whether they are still maintained","Whether the packages on the releases page were built from the code in this repository","How much disk it uses after installation, beyond the size of the download"]},"zh-Hans":{"summaryShort":"一款免费的 3D 设计软件，适用于从机械零件到建筑的各种实物，还能生成带尺寸的图纸。","summary":"FreeCAD 是一款面向任意尺寸实物的 3D 设计软件。你先画出一个带精确尺寸的平面形状，再在它上面继续建模，由于模型保留了整个历史记录，之后修改一个数字，整个零件都会跟着调整。它还能把 3D 模型转换成带尺寸标注的生产图纸。人们用它做产品设计、机械工程、建筑、3D 打印、仿真和加工，还可以用 Python 来驱动它。Windows、Mac 和 Linux 都有现成的安装包，在大多数 Linux 系统上它也能在软件中心找到。下载包大小为 415 MB 或更大。","easeRationale":"每个系统都有现成的安装包，不需要额外安装其他组件：Windows 上是一个 415 MB 的安装程序，Mac 上有同时支持 Intel 和 Apple 芯片的 DMG，Linux 上有 AppImage，而且在软件中心也能找到它。学会使用它是另一回事，但让它跑起来只需要一次下载。","notVerified":["其他 CAD 程序的图纸和模型导入时是否会丢失几何信息","在包含很多零件的大型装配体上它表现如何","你会需要哪些社区附加工作台，以及它们是否仍在维护","发布页面上的安装包是否确实是用这个仓库里的代码编译出来的","安装之后它占用多少磁盘空间，超出下载包本身的大小"]}}},{"id":"R_kgDOJEdyZQ","slug":"friction2d__friction","name":"Friction","owner":"friction2d","fullName":"friction2d/friction","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":27,"maxMB":82,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["video","design_cad"],"alternativeTo":[{"slug":"after-effects","name":"After Effects","confirmed":true},{"slug":"adobe-animate","name":"Animate","confirmed":true}],"declaredAlternatives":[],"stars":1984,"downloads":82813,"downloadsWeek":2051,"downloadsDays":7,"growth":106,"growthDays":7,"history":[{"date":"2026-10-25","stars":1856},{"date":"2026-10-26","stars":1860},{"date":"2026-10-27","stars":1862},{"date":"2026-10-28","stars":1863},{"date":"2026-10-29","stars":1864},{"date":"2026-10-30","stars":1866},{"date":"2026-10-31","stars":1868},{"date":"2026-11-01","stars":1869},{"date":"2026-11-02","stars":1872},{"date":"2026-11-03","stars":1878},{"date":"2026-11-04","stars":1888},{"date":"2026-11-05","stars":1915},{"date":"2026-11-06","stars":1930},{"date":"2026-11-07","stars":1948},{"date":"2026-11-08","stars":1955},{"date":"2026-11-09","stars":1965},{"date":"2026-11-10","stars":1984}],"firstPublishedAt":"2026-10-25T23:19:02.016Z","commitSha":"cfc7481d09f3791457747752c268a779ac55258f","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-09T23:41:05Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"friction-1.0.0-setup-win64.exe","sizeBytes":28407388,"url":"https://github.com/friction2d/friction/releases/download/v1.0.0/friction-1.0.0-setup-win64.exe","releaseUrl":"https://github.com/friction2d/friction/releases/tag/v1.0.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Make 2D animations and motion graphics with shapes, text and images, for video or the web.","summary":"Friction is a desktop program for creating 2D animations, such as animated titles, logos, shapes and drawings. You place vector shapes, text and images on a timeline, animate them with keyframes and export to video or animated SVG for the web. It is for people who make motion graphics and do not want a paid tool. It is based on an older open-source program called enve. Windows, macOS and Linux have ready downloads; the newest versions are still labelled release candidates (1.0.0-rc.3), not a final 1.0.","easeRationale":"The latest release includes a Windows installer (friction-1.0.0-rc.3-setup-win64.exe, about 27 MB) and a portable 7z version, so you download it and open it.","notVerified":["How stable the release-candidate version is for longer projects.","Which video formats export correctly on Windows.","How well it performs with many layers or long animations on an average computer."]},"zh-Hans":{"summaryShort":"用形状、文字和图片制作 2D 动画和动态图形，可用于视频或网页。","summary":"Friction 是一款用于制作 2D 动画的桌面程序，比如动画标题、徽标、形状和图画。你把矢量形状、文字和图片放到时间轴上，用关键帧为它们添加动画，然后导出为视频或适合网页的动画 SVG。它适合制作动态图形、又不想使用付费工具的人。它基于一款更早的开源程序 enve。Windows、macOS 和 Linux 都有现成的下载；最新版本仍标注为候选发布版（1.0.0-rc.3），还不是正式的 1.0。","easeRationale":"最新版本包含 Windows 安装程序（friction-1.0.0-rc.3-setup-win64.exe，约 27 MB）和便携的 7z 版本，下载后打开即可。","notVerified":["候选发布版在较长的项目中稳定性如何。","在 Windows 上哪些视频格式能正确导出。","在普通电脑上处理大量图层或较长动画时表现如何。"]}}},{"id":"R_kgDORIElig","slug":"fspecii__ace-step-ui","name":"ACE-Step UI","owner":"fspecii","fullName":"fspecii/ace-step-ui","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到任何许可协议。","source":"check","title":"没有许可协议"},{"code":"readme_promises_a_licence_the_repository_does_not_have","severity":"attention","evidence":"README 显示了一个 \"License-MIT\" 徽章，并写道 \"This project is open source under the [MIT License](LICENSE)\"，但仓库里并没有 LICENSE 文件，Radar 的检查也完全没有找到许可协议。没有它，无论 README 怎么说，谁都没有正式的权利重用或再分发这份代码。","source":"ai","title":"readme 承诺了一个仓库里并不存在的许可协议"},{"code":"google_ai_library_in_a_local_first_app","severity":"attention","evidence":"README 反复强调 \"100% local\" 和 \"no internet required after setup\"，但 package.json 却依赖 \"@google/genai\": \"^1.38.0\"，用在 \"AI Enhance\" 和 \"Format Assistant\" 功能上。前端里出现谷歌的 AI 库，意味着部分文本功能可能会调用外部服务。","source":"ai","title":"本地优先应用里用了谷歌的 AI 库"},{"code":"big_download_from_outside_github","severity":"attention","evidence":"README 的 Installation 部分写道：\"Download ACE-Step-1.5.7z (~5GB)\"，来自 \"https://files.acemusic.ai/acemusic/win/ACE-Step-1.5.7z\"，这是一个私人网站上的文件，没有给出校验值或签名，而且还自带了一份 Python。","source":"ai","title":"从 GitHub 之外下载的大文件"},{"code":"default_session_secret_plus_network_access","severity":"attention","evidence":".env.example 里自带 \"JWT_SECRET=ace-step-ui-local-secret\" 作为会话密钥，而功能列表里又提供 \"LAN Access - Use from any device on your local network\"，故障排除部分还让你在防火墙里开放 3000 和 3001 端口。这样一来，同一网络里的任何人都能访问到你的音乐库。","source":"ai","title":"默认的会话密钥加上局域网访问"},{"code":"commercial_use_claim_not_backed_by_this_repository","severity":"attention","evidence":"对比表格声称 \"Commercial Use: No restrictions\" 和 \"Ownership: You own everything\"。这些条款应该来自 ACE-Step 1.5 及其模型权重，而那些放在另一个仓库里；这个仓库里没有任何内容说明你生成的音乐属于什么许可协议。","source":"ai","title":"本仓库并不支持的商业使用说法"},{"code":"no_changes_for_four_months","severity":"info","evidence":"Radar 的检查记录显示，最后一次提交是在 2026-06-27，距检查时已过 129 天，期间没有发布任何版本，还有 61 个未关闭的 issue；而 README 却把 ACE-Step 的开发描述为 \"active\"（活跃）和 \"constant\"（持续不断）。","source":"ai","title":"四个月没有更新"}],"platforms":["windows","mac","linux"],"tools":["node","python","gpu_nvidia","ffmpeg"],"buildTools":["node","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Pexels (only for the stock footage in the music video maker)"],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[{"target":"suno","stance":"claims","evidence":"<strong>The Ultimate Open Source Suno Alternative</strong><br>"}],"stars":5198,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":19,"growthDays":6,"history":[{"date":"2026-11-04","stars":5179},{"date":"2026-11-05","stars":5184},{"date":"2026-11-06","stars":5186},{"date":"2026-11-07","stars":5186},{"date":"2026-11-08","stars":5192},{"date":"2026-11-09","stars":5198},{"date":"2026-11-10","stars":5198}],"firstPublishedAt":"2026-11-04T01:36:50.390Z","commitSha":"a1fdf91829ec6f7b98844f80e323529cd155dbf2","verifiedAt":"2026-11-04T01:27:43.189Z","lastCommitAt":"2026-06-27T17:15:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Spotify-like page for making complete songs with AI on your own NVIDIA graphics card, on top of the free ACE-Step model.","summary":"ACE-Step UI is the screen in front of ACE-Step 1.5, a free AI music model. You type a description or full lyrics, choose genre, key, tempo and length, and it writes songs of up to four minutes with vocals on your own machine. It also keeps a library with playlists, trims audio, separates vocals from instruments, and builds a simple music video from stock footage. It is not the model itself: you install ACE-Step 1.5 separately (about 5 GB), start it, then start this interface. An NVIDIA card with at least 4 GB is required.","easeRationale":"Windows is the least painful: a 5 GB portable package of the AI engine, then a start-all.bat that launches the three pieces, or a one-click install through the separate Pinokio app. Linux is the same with uv and shell scripts in a terminal. Everywhere it needs an NVIDIA card, two servers running at once and the right ports open, and on Mac the requirement for an NVIDIA card makes the macOS commands in the README doubtful.","costNote":"No charge and no subscription. You pay in hardware: an NVIDIA card with 4 GB of memory at minimum, 12 GB for the AI helper features.","notVerified":["Whether the music quality really matches the paid services, as the README claims.","What the \"AI Enhance\", \"Format Assistant\" and \"Thinking Mode\" features send out, and whether they need a key of their own.","Real disk use: about 5 GB for the engine and models, plus every song you keep.","How long one song takes on a 4 GB card, and whether it finishes at all.","Whether the macOS instructions work on an Apple machine without an NVIDIA card.","Under which licence the songs you generate stand, since that depends on the separate ACE-Step model."]},"zh-Hans":{"summaryShort":"一个类似 Spotify 的页面，让你用自己的 NVIDIA 显卡，在免费的 ACE-Step 模型基础上制作完整的歌曲。","summary":"ACE-Step UI 是 ACE-Step 1.5（一个免费的 AI 音乐模型）的操作界面。你输入描述或完整歌词，选择曲风、调式、速度和长度，它就能在你自己的机器上生成最长四分钟的带人声歌曲。它还有一个带播放列表的音乐库，可以剪辑音频、把人声和伴奏分离，并用素材画面制作简单的音乐视频。它本身不是模型：你需要单独安装 ACE-Step 1.5（约 5 GB），启动它，然后再启动这个界面。需要一张至少 4 GB 的 NVIDIA 显卡。","easeRationale":"Windows 上最省心：下载一个 5 GB 的便携版 AI 引擎，用 start-all.bat 一次启动三个部分，或者通过另外的 Pinokio 应用一键安装。Linux 上类似，用 uv 和终端里的 shell 脚本。不管哪种系统都需要 NVIDIA 显卡、同时运行两个服务器，并开放正确的端口；而在 Mac 上，需要 NVIDIA 显卡这一点让 README 里的 macOS 命令显得可疑。","costNote":"不收费，也不是订阅制。你花的是硬件成本：至少 4 GB 显存的 NVIDIA 显卡，AI 辅助功能需要 12 GB。","notVerified":["音乐质量是否真的能和付费服务相比，就像 README 里说的那样。","\"AI Enhance\"、\"Format Assistant\" 和 \"Thinking Mode\" 这些功能会发送什么数据出去，以及它们是否需要自己的密钥。","实际占用的磁盘空间：引擎和模型本身约 5 GB，再加上你保存的每一首歌。","在 4 GB 显卡上生成一首歌要多久，以及能不能顺利完成。","macOS 的安装说明在没有 NVIDIA 显卡的苹果电脑上是否真的可行。","你生成的歌曲属于什么许可协议，因为这取决于另外的 ACE-Step 模型。"]}}},{"id":"R_kgDOQ-_Fpw","slug":"fspecii__heartmula-studio","name":"HeartMuLa Studio","owner":"fspecii","fullName":"fspecii/HeartMuLa-Studio","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接通过管道传给 shell 执行:curl -fsSL https://ollama.com/install.sh | sh","source":"check","title":"curl 管道直接执行 shell 脚本"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单均未检测到任何许可证。","source":"check","title":"没有许可证"},{"code":"readme_says_mit_but_no_license_file","severity":"attention","evidence":"README 写道:\"This project is open source under the [MIT License](LICENSE)\",还有一个写着 \"License-MIT\" 的徽章,但顶层文件列表是 \".dockerignore, .github/, Dockerfile, README.md, backend/, docker-compose.yml, frontend/, preview.gif, start.sh\",没有 LICENSE 文件,检查结果也没有检测到任何许可证。没有这个文件,MIT 许可就只是文字上的一句声明。","source":"ai","title":"readme 说是 mit 许可,但没有许可证文件"},{"code":"nvidia_card_with_10gb_required","severity":"attention","evidence":"README 的 Docker 前置条件写道:\"NVIDIA GPU with 10GB+ VRAM (12GB recommended for reliable generation)\",VRAM 表格把 \"<8GB\" 标记为 \"Not supported\"。AMD 显卡只被列为 \"Flash Attention: Disabled (compatibility varies)\"。没有合适 NVIDIA 显卡的电脑无法运行它。","source":"ai","title":"需要显存 10gb 以上的 nvidia 显卡"},{"code":"prebuilt_image_from_personal_docker_hub_account","severity":"info","evidence":"README 在 Docker Hub 上提供了 \"ambsd/heartmula-studio:latest\",与 \"ghcr.io/fspecii/heartmula-studio:latest\" 并列。Docker Hub 上的账号名 \"ambsd\" 和 GitHub 上的仓库所有者 \"fspecii\" 对不上(README 确实链接到了 X 账号 @AmbsdOP),所以那个镜像并非以仓库自己的名义发布。","source":"ai","title":"预构建镜像来自个人 docker hub 账号"},{"code":"models_downloaded_from_the_internet_on_first_run","severity":"info","evidence":"README 写道:\"Models are automatically downloaded from HuggingFace Hub on first run (~5GB)\"。这些 AI 音乐模型并不在仓库里,所以你实际运行的内容是启动时从 HuggingFace 获取的。","source":"ai","title":"模型在首次运行时从网上下载"},{"code":"web_page_opens_on_all_network_interfaces","severity":"info","evidence":"文档给出的启动命令是 \"python -m uvicorn backend.app.main:app --host 0.0.0.0 --port 8000\",Docker 文件发布的端口是 \"8000:8000\"。使用 0.0.0.0 意味着任何能连接到你电脑这个端口的人都能访问该页面,README 也没有提到密码或登录。","source":"ai","title":"网页在所有网络接口上开放"},{"code":"generated_songs_may_copy_a_style_you_upload","severity":"info","evidence":"README 中标记为 \"Experimental\" 的 \"Reference Audio (Style Transfer)\" 写道:\"Use any audio file as a style reference\",还带有一个 \"Style Influence\" 滑块,以及 \"Coming Soon: LoRA Voice Training ... voice consistency that surpasses Suno\"。上传别人的录音或声音来复制其风格,可能涉及该录音或该表演者的权利,而仓库并未给出使用条款。","source":"ai","title":"生成的歌曲可能复制你上传的风格"},{"code":"no_commits_for_about_eight_months","severity":"info","evidence":"检查记录显示最后一次提交是在 2026-02-25,距此次审查已有 254 天,而 README 仍将 LoRA 声音训练标注为 \"Coming Soon\",将参考音频功能标注为 \"Experimental\"。","source":"ai","title":"大约八个月没有提交"}],"platforms":["windows","linux"],"tools":["docker","gpu_nvidia"],"buildTools":["docker","gpu_nvidia"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenRouter"],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":647,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":647},{"date":"2026-11-08","stars":647},{"date":"2026-11-09","stars":647},{"date":"2026-11-10","stars":647}],"firstPublishedAt":"2026-11-06T22:57:16.622Z","commitSha":"bf49e8f2f4c0e73bdc068257d2b740921ecdadd8","verifiedAt":"2026-11-06T22:37:17.429Z","lastCommitAt":"2026-02-25T16:23:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Write a song idea and it generates a full track with vocals on your own PC, using an NVIDIA graphics card.","summary":"HeartMuLa Studio is a web page you run on your own computer to generate complete songs, with vocals and lyrics up to about four minutes, or instrumental tracks. You type a topic and style tags, or upload an audio file as a style reference. It is for people with a gaming or workstation PC: it needs an NVIDIA card with 10 GB or more of video memory, Docker or Python plus Node, and it downloads about 5 GB of AI models on first run. Lyrics can be written by a local Ollama model or by the paid OpenRouter service.","easeRationale":"Linux is the smoothest path and still grade D: you clone the repository and run \"docker compose up -d\", which needs Docker plus the NVIDIA Container Toolkit and an NVIDIA card with 10 GB or more of video memory, then waits for a ~10 GB image and ~5 GB of models. Windows is the same grade through Docker, but the from-source route there asks for Python, Node and, for the faster torch.compile mode, a separate \"triton-windows\" package.","costNote":"The software and the music models cost nothing and run on your own card. Only the optional OpenRouter route for writing lyrics is paid; the local Ollama route is free.","notVerified":["Whether generation really works on each card in the VRAM table, and the quality of the songs it produces","Real disk use: the README mentions a ~10 GB Docker image plus ~5 GB of models, and the checks could not measure it","What is inside the pre-built images on Docker Hub and GitHub Container Registry, since they are built elsewhere","Whether the MIT permission claimed in the README applies, as no license file exists in the repository","The cost of using OpenRouter for lyrics, which depends on the model you pick","Whether any text or audio you type leaves your computer when the Ollama or OpenRouter lyrics feature is on"]},"zh-Hans":{"summaryShort":"写下一个歌曲创意,它就能在你自己的电脑上用 NVIDIA 显卡生成带人声的完整曲目。","summary":"HeartMuLa Studio 是一个在你自己电脑上运行的网页,用来生成带人声和歌词、时长约四分钟以内的完整歌曲,或纯音乐曲目。你输入一个主题和风格标签,或者上传一个音频文件作为风格参考。它适合拥有游戏本或工作站电脑的人:需要一块显存 10GB 以上的 NVIDIA 显卡、Docker 或者 Python 加 Node,首次运行会下载约 5GB 的 AI 模型。歌词可以由本地的 Ollama 模型撰写,也可以用付费的 OpenRouter 服务。","easeRationale":"Linux 是最顺畅的途径,但依然是 D 级:你克隆仓库并运行 \"docker compose up -d\",这需要 Docker、NVIDIA Container Toolkit,以及一块显存 10GB 以上的 NVIDIA 显卡,然后等待约 10GB 的镜像和约 5GB 的模型下载完成。Windows 通过 Docker 的等级相同,但从源码安装那条路径需要 Python、Node,以及为了更快的 torch.compile 模式而单独安装的 \"triton-windows\" 包。","costNote":"软件和音乐模型本身免费,运行在你自己的显卡上。只有写歌词时可选的 OpenRouter 路线是付费的;本地的 Ollama 路线免费。","notVerified":["生成功能在 VRAM 表格里列出的每一块显卡上是否真的能用,以及生成歌曲的质量如何。","实际磁盘占用:README 提到约 10GB 的 Docker 镜像加约 5GB 的模型,检查无法测量这一点。","Docker Hub 和 GitHub Container Registry 上预构建镜像里到底有什么,因为它们是在别处构建的。","README 中声称的 MIT 许可是否真的适用,因为仓库里没有许可证文件。","使用 OpenRouter 写歌词的费用,这取决于你选择的模型。","使用 Ollama 或 OpenRouter 歌词功能时,你输入的文本或音频是否会离开你的电脑。"]}}},{"id":"R_kgDOJEdt9g","slug":"gaizhenbiao__chuanhuchatgpt","name":"ChuanhuChatGPT","owner":"GaiZhenbiao","fullName":"GaiZhenbiao/ChuanhuChatGPT","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞(OSV):gradio: GHSA-26jh-r8g2-6fpr、GHSA-279j-x4gx-hfrh、GHSA-37qc-qgx6-9xjv;langchain: GHSA-3644-q5cj-c5c7、GHSA-3hjh-jh2h-vrg6、GHSA-45pg-36p6-83v9;protobuf: GHSA-7gcm-g887-7qv7、GHSA-8qvm-5x2c-j2w7、PYSEC-2026-1805;python-multipart: GHSA-59g5-xgcq-4qw3、GHSA-5rvq-cxj2-64vf、GHSA-6jv3-5f52-599m","source":"check","title":"已知漏洞"},{"code":"paid_api_key_needed_for_the_main_use","severity":"attention","evidence":"README 快速开始部分:安装后必须把 config_example.json 复制为 config.json,并 \"fill in the API-Key and other settings\"。模型表格列出了 ChatGPT、Azure OpenAI、Google Gemini、Claude、DeepSeek、讯飞星火和 MiniMax,全部是按使用付费的服务。程序本身不含任何模型,也没有免费额度;只有本地模型方式(Ollama、ChatGLM、LLaMA、Qwen)能避开这笔费用。","source":"ai","title":"主要功能需要付费 api 密钥"},{"code":"public_online_demo_is_run_by_another_account","severity":"attention","evidence":"仓库主页指向 \"https://huggingface.co/spaces/JohnSmith9982/ChuanhuChatGPT\",README 把它链接为 \"try online\"。这个页面托管在 HuggingFace 账号 JohnSmith9982 下,而不是仓库所有者 GaiZhenbiao 名下,所以把自己的 API 密钥粘贴到那里,就是把它交给一个你无法控制的服务器。README 自己的建议是先把这个 Space 复制到你自己的账号里。","source":"ai","title":"公开在线演示由另一个账号运营"},{"code":"settings_can_expose_the_chat_to_the_internet","severity":"attention","evidence":"README 部署部分:设置 \"\\\"server_name\\\": \\\"0.0.0.0\\\", \\\"server_port\\\": <your port>\" 以部署到服务器,设置 \"\\\"share\\\": true\" 来 \"get a public link\"。这两项都会让你的聊天窗口,以及保存在其中的密钥,能从你电脑之外被访问;README 没有说明这个链接是否有密码保护。","source":"ai","title":"设置可能把聊天暴露到公网"},{"code":"dependencies_pinned_to_2024_versions","severity":"attention","evidence":"requirements.txt 锁定了 gradio==4.29.0、gradio_client==0.16.1、openai==1.16.2、langchain==0.1.14、anthropic==0.18.1、pydantic==2.5.2、protobuf==3.20.3 和 faiss-cpu==1.7.4。自动检查已经标记出 gradio、langchain、protobuf 和 python-multipart 中的已知漏洞;这些锁定版本也意味着,即使 README 宣传支持 GPT-5,安装的库仍然是 2024 年的。","source":"ai","title":"依赖锁定在 2024 年的版本"},{"code":"local_models_need_a_second_install_and_a_graphics_card","severity":"info","evidence":"本地模型用的库放在单独的 requirements_advanced.txt 里(transformers、torch、llama-cpp-python、auto-gptq、cpm-kernels),Dockerfile 里这一行是被注释掉的:\"# RUN pip install --user --no-cache-dir -r requirements_advanced.txt\"。要在自己的电脑上运行 ChatGLM、LLaMA 或 Qwen,就得安装这第二份依赖列表,并拥有一块显存足够的显卡。","source":"ai","title":"本地模型需要额外安装和一块显卡"},{"code":"midjourney_listed_though_it_has_no_public_api","severity":"info","evidence":"模型表格把 \"Midjourney\" 列在 \"API models\" 里(标注为不支持流式输出)。Midjourney 通过自己的应用出售访问权限,并不发布公开 API,所以这个连接依赖于该服务条款之外的某种途径;仓库没有说明具体是怎么实现的。","source":"ai","title":"midjourney 被列出但并无公开 api"},{"code":"readme_and_install_guide_mainly_in_chinese","severity":"info","evidence":"主 README 是简体中文,readme/ 目录里有英语、日语、韩语、俄语的翻译,而它指向的安装和故障排查页面是中文的 wiki 页面(\"使用教程\"、\"常见问题\")。Telegram 群组是支持渠道。","source":"ai","title":"readme 和安装指南主要是中文"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["OpenAI (ChatGPT, DALL-E 3, fine-tuning)","Google Gemini","Anthropic Claude","DeepSeek"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["local_ai","ai_agents","productivity"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"typingmind","name":"TypingMind","confirmed":true},{"slug":"perplexity","name":"Perplexity","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[],"stars":15244,"downloads":4087,"downloadsWeek":5,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":15243},{"date":"2026-11-08","stars":15244},{"date":"2026-11-09","stars":15244},{"date":"2026-11-10","stars":15244}],"firstPublishedAt":"2026-11-06T22:57:16.629Z","commitSha":"2c7d5304c215520dbc6077f185fc45d8e0c8d919","verifiedAt":"2026-11-06T22:39:14.361Z","lastCommitAt":"2026-09-16T11:43:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A chat page you run on your own computer to talk to ChatGPT, Gemini, Claude or local models, with file Q&A and web search.","summary":"Chuanhu Chat (ChuanhuChatGPT) gives you one chat window for many AI models: paid services such as ChatGPT, Azure OpenAI, Gemini, Claude and DeepSeek through your own key, or models running on your machine through Ollama, ChatGLM, LLaMA and Qwen. It adds saved history with search, answers based on files you upload, web search, an automatic assistant and GPT-3.5 fine-tuning. You install Python, copy the example settings file, paste a key and start it; the page then opens in your browser. The README and the install guide are written mainly in Chinese.","easeRationale":"Every system is grade C and the steps are the same: install Python, clone the project, run \"pip install -r requirements.txt\", rename config_example.json to config.json, paste a key and run \"python ChuanhuChatbot.py\". There are ready starter scripts for the three systems (run_Windows.bat, run_macOS.command, run_Linux.sh), but no installer and no ready download in the 27 releases, and the detailed install guide is the project wiki, in Chinese.","costNote":"The program is free and GPL-3.0. Chatting through ChatGPT, Gemini, Claude or DeepSeek is billed by each of those services on your own key; using local models through Ollama costs nothing but needs a strong computer.","notVerified":["Whether each listed model really answers today, since the pinned libraries are from 2024 while the README announces GPT-5 and DeepSeek R1","Real disk use: the checks estimate 950 to 4450 MB, and the local-model route downloads models on top of that","What the online demo Space does with the API key you type into it","How much each chat costs, which depends on the service and model you choose","Whether the uploaded files used for file-based answers stay on your computer or are sent to the model service","Whether the automatic assistant and web-search features still work, as they depend on outside search services"]},"zh-Hans":{"summaryShort":"一个在自己电脑上运行的聊天网页,可以和 ChatGPT、Gemini、Claude 或本地模型对话,支持基于文件的问答和网页搜索。","summary":"川虎 Chat(ChuanhuChatGPT)为许多 AI 模型提供一个聊天窗口:通过你自己的密钥使用 ChatGPT、Azure OpenAI、Gemini、Claude、DeepSeek 等付费服务,或者通过 Ollama、ChatGLM、LLaMA、Qwen 在自己的电脑上运行模型。它还提供可搜索的保存历史记录、基于你上传文件的回答、网页搜索、自动助手和 GPT-3.5 微调。你需要安装 Python,复制示例配置文件,粘贴密钥并启动程序;之后页面会在浏览器中打开。README 和安装指南主要是中文写的。","easeRationale":"每个系统都是 C 级,步骤相同:安装 Python、克隆项目、运行 \"pip install -r requirements.txt\"、把 config_example.json 改名为 config.json、粘贴密钥并运行 \"python ChuanhuChatbot.py\"。三个系统都有现成的启动脚本(run_Windows.bat、run_macOS.command、run_Linux.sh),但没有安装程序,27 个发行版里也没有现成的下载包,详细的安装指南是项目的 wiki,用中文写的。","costNote":"该程序免费,采用 GPL-3.0 许可。通过 ChatGPT、Gemini、Claude 或 DeepSeek 聊天的费用由这些服务按你自己的密钥收取;通过 Ollama 使用本地模型不花钱,但需要一台性能较强的电脑。","notVerified":["列出的每个模型今天是否真的能用,因为锁定的库是 2024 年的,而 README 宣传的是 GPT-5 和 DeepSeek R1","实际占用的磁盘空间:检查估计为 950 到 4450 MB,本地模型方式还会在此基础上下载模型","在线演示 Space 对你输入的 API 密钥做了什么","每次聊天花费多少,这取决于你选择的服务和模型","用于基于文件回答的上传文件是留在你电脑上,还是被发送到模型服务","自动助手和网页搜索功能是否还能用,因为它们依赖外部搜索服务"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMzcxMjg5NDU=","slug":"garmeeh__next-seo","name":"Next SEO","owner":"garmeeh","fullName":"garmeeh/next-seo","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"low","signals":[{"code":"paid_product_ads_with_referral_links","severity":"info","evidence":"README 开头先是广告,而不是项目本身的介绍:\"Outrank - Get traffic and outrank competitors with Backlinks & SEO-optimized content while you sleep!\",链接指向 https://outrank.so/?via=next-seo(一个推荐链接),还有 \"The only SEO skill your agent needs. Install it now\",以及一个藏在 dub.sh 短链接后面的订阅表单,看不出链接实际指向哪里。","source":"ai","title":"带推荐链接的付费产品广告"},{"code":"version_change_moves_components","severity":"info","evidence":"README 把老用户指向别处:\"Looking for v6 documentation? View Here\" 和 \"Still using <NextSeo /> component in Pages? View docs here\",使用 Pages Router 的用户被告知要从 \"next-seo/pages\" 导入。升级一个旧站点意味着要修改导入语句,并阅读另一套文档。","source":"ai","title":"版本变化导致组件位置改变"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","social_marketing"],"alternativeTo":[],"declaredAlternatives":[],"stars":8542,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":2,"growthDays":6,"history":[{"date":"2026-11-05","stars":8541},{"date":"2026-11-06","stars":8542},{"date":"2026-11-07","stars":8542},{"date":"2026-11-08","stars":8541},{"date":"2026-11-09","stars":8541},{"date":"2026-11-10","stars":8542}],"firstPublishedAt":"2026-11-04T22:32:54.930Z","commitSha":"738ef7e774ee6216f6e90c26a3b5b99dd5d166b4","verifiedAt":"2026-11-04T22:21:08.642Z","lastCommitAt":"2026-07-29T19:23:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A code add-on for Next.js websites that writes the hidden data Google uses to show richer search results.","summary":"Next SEO is a package for websites built with Next.js. It gives you ready-made components that write structured data (JSON-LD) into your pages: articles, recipes, how-to guides, fact checks, organisations, paywalled content and many more, so search engines can understand what a page is about and show richer results. It is not a program you open: you add it to an existing Next.js project with npm and write the components into your code. Free and MIT-licensed, with no account or key.","easeRationale":"Windows, Mac and Linux are all an E for the same reason: there is nothing to install and open. It is a package you add to an existing Next.js project (\"npm install next-seo\") and then use by writing React components in your own source code, so only someone who programs can use it.","costNote":"MIT licence, no paid tier. The author asks for optional donations through Buy Me a Coffee.","notVerified":["Whether Google and other search engines actually show richer results for your pages; that depends on their own rules, not on this package.","Whether every component still matches the current structured-data requirements of the search engines.","Whether it works with Next.js and React versions newer than the ones it declares (Next 13.4+ and React 18.2+).","How much the package adds to your site's page weight in practice."]},"zh-Hans":{"summaryShort":"一个用于 Next.js 网站的代码插件,能写入 Google 用来展示更丰富搜索结果所需的隐藏数据。","summary":"Next SEO 是一个用于 Next.js 网站的软件包。它提供现成的组件,可以把结构化数据(JSON-LD)写入你的页面:文章、食谱、操作指南、事实核查、组织信息、付费内容等等,让搜索引擎能理解页面内容并展示更丰富的搜索结果。它不是一个你打开运行的程序:你需要用 npm 把它添加到已有的 Next.js 项目中,并在代码里写入这些组件。免费,采用 MIT 许可,不需要账号或密钥。","easeRationale":"Windows、Mac 和 Linux 得到相同的 E 评级,原因相同:没有什么需要安装或打开的东西。它是一个添加到现有 Next.js 项目中的软件包(\"npm install next-seo\"),之后通过在自己的源代码里编写 React 组件来使用,所以只有会编程的人才能用它。","costNote":"MIT 许可,没有付费套餐。作者通过 Buy Me a Coffee 接受可选的捐赠。","notVerified":["Google 和其他搜索引擎是否真的会为你的页面展示更丰富的结果;这取决于它们自己的规则,而不是这个软件包。","每个组件是否仍符合搜索引擎当前的结构化数据要求。","它是否适用于比声明版本更新的 Next.js 和 React(Next 13.4+ 和 React 18.2+)。","这个软件包实际上会给你网站的页面增加多少负担。"]}}},{"id":"R_kgDORkolAw","slug":"garrytan__gstack","name":"gstack","owner":"garrytan","fullName":"garrytan/gstack","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"236 天内获得 138691 个 star（约每天 588 个）","source":"check","title":"star 增长速度异常"},{"code":"install_runs_a_setup_script_from_a_pasted_prompt","severity":"attention","evidence":"README.md 让你把一段提示词粘贴进 Claude Code，让代理执行 git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setup。安装是由 AI 而不是你本人执行的，setup 还会在你的机器上注册 hook 并编译二进制文件。","source":"ai","title":"安装过程通过粘贴的提示词运行一个安装脚本"},{"code":"sends_your_code_to_another_ai_by_default","severity":"attention","evidence":"README.md Privacy & Telemetry：“outside reviews (on by default; the first one prints a notice) send code to Codex or Claude Code under your own login and turn off with `gstack-config set codex_reviews disabled`.”多个技能的正常流程中也包含一个“Outside review quality gate”。","source":"ai","title":"默认会把你的代码发送给另一个 ai"},{"code":"updates_itself_from_github_each_session","severity":"attention","evidence":"README.md：“Every Claude Code session starts with a fast auto-update check (throttled to once/hour, network-failure-safe, completely silent)”，在团队模式下更新检查“on by default”。新代码可能在你没有要求的情况下到达你的机器。","source":"ai","title":"每次会话都会自动从 github 更新自己"},{"code":"team_mode_blocks_teammates_work","severity":"info","evidence":"README.md 第二步会向你的仓库提交一个 hook，提交信息为“require gstack for AI-assisted work”；这个 required hook 运行在 Node.js 上，“Swap `required` for `optional` if you'd rather nudge teammates than block them.”","source":"ai","title":"团队模式会阻塞同事的工作"},{"code":"copies_browser_cookies","severity":"attention","evidence":"/setup-browser-cookies 技能会“Copy selected cookies from Chrome, Chromium, Brave, Edge, Windows-only Opera and Opera GX, or macOS-only Comet and Arc into gstack's bundled browser”，这会把你已登录的会话交给一个由 AI 操控的浏览器。","source":"ai","title":"复制浏览器 cookie"},{"code":"advises_turning_off_a_windows_protection","severity":"attention","evidence":"README.md 中的已知问题：gstack 编译出的二进制文件“are built on your machine and unsigned, so Windows 11 with Smart App Control on refuses to start them... Today's workarounds are running gstack inside WSL, or turning Smart App Control off.”","source":"ai","title":"建议关闭 windows 的一项保护机制"},{"code":"opens_a_public_tunnel_for_remote_agents","severity":"info","evidence":"/pair-agent 技能会“Auto-starts ngrok tunnel for remote agents”，让其他代理可以操控 gstack 的浏览器，/ios-qa 还有一个可选的 --tailnet 参数，会把连接的 iPhone 暴露在你的 Tailscale 网络上。","source":"ai","title":"为远程代理打开一个公开隧道"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg","other"],"buildTools":["node","ffmpeg"],"disk":{"minMB":150,"maxMB":650,"basis":"heuristic"},"apiKeys":["Anthropic (Claude Code sign-in; ANTHROPIC_API_KEY for the built-in evals)","OpenAI Codex CLI sign-in (used for the second-opinion reviews, on by default)","Ubicloud (optional, for running heavy test suites on a rented machine)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","automation"],"alternativeTo":[{"slug":"coderabbit","name":"CodeRabbit","confirmed":true},{"slug":"snyk","name":"Snyk","confirmed":false}],"declaredAlternatives":[],"stars":141095,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":2238,"growthDays":7,"history":[{"date":"2026-10-26","stars":137833},{"date":"2026-10-27","stars":137942},{"date":"2026-10-28","stars":138073},{"date":"2026-10-29","stars":138192},{"date":"2026-10-30","stars":138330},{"date":"2026-10-31","stars":138481},{"date":"2026-11-01","stars":138600},{"date":"2026-11-02","stars":138683},{"date":"2026-11-03","stars":138857},{"date":"2026-11-04","stars":139262},{"date":"2026-11-05","stars":139871},{"date":"2026-11-06","stars":140407},{"date":"2026-11-07","stars":140636},{"date":"2026-11-08","stars":140773},{"date":"2026-11-09","stars":140940},{"date":"2026-11-10","stars":141095}],"firstPublishedAt":"2026-10-27T01:10:12.629Z","commitSha":"f47d19dd2109b5a196e854a7dd38e923d12916e8","verifiedAt":"2026-11-03T02:39:42.589Z","lastCommitAt":"2026-11-10T17:37:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A pack of slash commands that turns Claude Code into a product team: planning, code review, browser QA and releases.","summary":"gstack is a free collection of slash commands and scripts that you install inside Claude Code (or another coding agent). Each command plays a role: a founder who challenges your idea, an engineering manager who locks the plan, a designer, a code reviewer, a QA lead that opens a real browser, a security reviewer and a release engineer that opens the pull request. It is written by Garry Tan of Y Combinator and is for people who already code with an AI agent. You install Git, Bun and Claude Code, then clone the repo and run a setup script. Best supported on macOS and Linux.","easeRationale":"macOS and Linux are the best-supported platforms and still need several tools in a terminal: Claude Code, Git and Bun 1.4.2+, then `git clone` into `~/.claude/skills/gstack` and `./setup`, which also builds its own browser. Windows is harder: the README calls it minimal support, needs Git Bash or WSL plus Node.js as well, and the compiled tools are unsigned, so Windows 11 Smart App Control refuses to start them.","costNote":"gstack itself is MIT and free, with no paid tier. The commands only work through a coding agent such as Claude Code, which needs a paid plan or API credit, and some commands (/ubicloud, paid evals) spend money on outside services.","notVerified":["Whether the many skills work as described, and how well they work outside Claude Code (most other agents are marked experimental)","What the productivity claims in the README are worth, since they are the author's own numbers","How much the coding agent, the second-opinion reviews and the paid evals cost per sprint","What exactly leaves your machine during outside reviews and update checks","Whether the setup script's hooks and compiled binaries behave on your own system, especially on Windows"]},"zh-Hans":{"summaryShort":"一套斜线命令，把 Claude Code 变成一个产品团队：规划、代码评审、浏览器 QA 和发布。","summary":"gstack 是一组免费的斜线命令和脚本，安装在 Claude Code（或其他编程代理）内部。每个命令扮演一个角色：挑战你想法的创始人、锁定计划的工程经理、设计师、代码评审员、打开真实浏览器的 QA 负责人、漏洞评审员，以及打开拉取请求的发布工程师。它由 Y Combinator 的 Garry Tan 编写，面向已经在用 AI 代理编程的人。你安装 Git、Bun 和 Claude Code，然后克隆仓库并运行一个安装脚本。macOS 和 Linux 上支持最好。","easeRationale":"macOS 和 Linux 是支持最好的平台，但仍然需要在终端里装好几个工具：Claude Code、Git 和 Bun 1.4.2+，然后在 ~/.claude/skills/gstack 执行 git clone，再运行 ./setup，它还会编译自己的浏览器。Windows 更难：README 称其为最低限度的支持，除了 Node.js 还需要 Git Bash 或 WSL，而且编译出的工具没有签名，所以开启了 Smart App Control 的 Windows 11 会拒绝启动它们。","costNote":"gstack 本身是 MIT 协议、免费的，没有付费档次。这些命令只能通过像 Claude Code 这样的编程代理使用，而后者需要付费计划或 API 额度，一些命令（/ubicloud、付费评测）还会在外部服务上花钱。","notVerified":["众多技能是否真的如描述那样工作，以及它们在 Claude Code 之外表现如何（大多数其他代理都被标记为实验性）","README 里的生产力宣称到底有多少含金量，因为这些数字是作者自己给出的","每个 sprint 在编程代理、二次评审和付费评测上到底要花多少钱","外部评审和更新检查期间，到底有什么内容离开了你的机器","安装脚本的 hook 和编译出的二进制文件在你自己的系统上表现如何，尤其是在 Windows 上"]}}},{"id":"R_kgDOVNLNMw","slug":"gathin23__cleanyourtv","name":"CleanYourTV","owner":"Gathin23","fullName":"Gathin23/cleanyourtv","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"15 天内获得 49267 个星标","source":"check","title":"新仓库，星标很多"},{"code":"tv_debug_port_left_open","severity":"attention","evidence":"README 自己对 \"Isn't ADB access a security hole?\" 的回答：\"Network ADB is plaintext on port 5555, and on some firmware any device on the LAN gets a shell, so close it when you are done.\" 在操作期间，你网络里的任何人都能连到那个 shell；仓库附带 tools/adb-closed.py，用来在事后检查该端口是否已关闭。","source":"ai","title":"电视的调试端口被敞开"},{"code":"can_break_tv_features","severity":"attention","evidence":"关错软件包会带来后果，README 中有记录：在 Google TV 上，设置向导 \"can take Home and crash-loop the system\"，如果 Wi-Fi 是在设置过程中保存的，禁用它会 \"deletes the network\"；在 Chromecast 和 Google TV Streamer 上，则会失去手机投屏。电视连启动都不够时，FAQ 给出的退路是 \"a factory reset from its recovery menu\"。","source":"ai","title":"可能破坏电视的功能"},{"code":"ai_assistant_decides_what_to_disable","severity":"attention","evidence":"推荐的做法是把 PROMPT.md 粘贴给编程助手，由它在自己发现的电视上 \"backs up, debloats, builds and installs the launcher\"：\"the prompt has your coding assistant discover it on your TV and ask before touching anything you might use.\" 你是在信任一个 AI 对你的电视需要哪些系统软件包的判断。仓库自己的主题标签包括 claude-code 和 vibe-coding。","source":"ai","title":"由 AI 助手决定禁用什么"},{"code":"warranty_and_firmware_updates","severity":"info","evidence":"法律部分：\"You run it on a TV you own, at your own risk, and changing a TV's software may affect its warranty or support.\" 固件更新之后 \"the listed packages come back, Home goes to the stock launcher\"，所以这项工作需要重做；有些列表还会禁用电视自带的更新程序。","source":"ai","title":"保修与固件更新"},{"code":"no_prebuilt_app_file","severity":"info","evidence":"FAQ，\"Can I just download the APK?\"：\"Build it with the one-line script... No prebuilt binaries.\" GitHub 上的 11 个发布版本没有附带任何平台的文件。","source":"ai","title":"没有现成的应用文件"},{"code":"lookalike_sites_using_the_name","severity":"info","evidence":"README，Start here：\"Official links: cleanyour.tv and this repo. Other domains using the name, such as cleanyourtv.com, are not run by this project.\" 它还补充说 \"CleanYourTV publishes no app in any app store, and clean-your-tv.net is not ours either.\"","source":"ai","title":"有长得像的网站在使用这个名字"},{"code":"easy_route_is_someone_elses_app","severity":"info","evidence":"对于没有编程助手的人，README 指向另外开发者做的四个手机、Mac 和 Windows 应用 (cleanyourtv-app、Tellyscope、TidyTV、CleanMyTV)，并声明它们 \"are independent projects, not affiliated with or endorsed by CleanYourTV or its maintainer, and each is provided as is under its own license\"。这里没有审查这些程序。","source":"ai","title":"省事的路线是别人做的应用"},{"code":"readme_promotes_a_paid_service_with_a_tracking_link","severity":"info","evidence":"Quick start 里列出了 Claude Code \"on 0G Private Computer\"，链接带有活动标记 (`utm_source=cleanyourtv&utm_medium=github&utm_campaign=readme`)，页头还有一个 Buy Me a Coffee 徽章。","source":"ai","title":"README 用带跟踪参数的链接推广付费服务"}],"platforms":["windows","mac","linux"],"tools":["python","java","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["hardware","network_system"],"alternativeTo":[{"slug":"projectivy-launcher","name":"Launcher","confirmed":true}],"declaredAlternatives":[{"target":"projectivy-launcher","stance":"claims","evidence":"Strip the bloatware from any Android TV or Google TV over Wi-Fi and replace the ad-filled home screen with a clean launcher that looks like an Apple TV"}],"stars":55946,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":4603,"growthDays":7,"history":[{"date":"2026-10-25","stars":35489},{"date":"2026-10-26","stars":39295},{"date":"2026-10-27","stars":41888},{"date":"2026-10-28","stars":43867},{"date":"2026-10-29","stars":45511},{"date":"2026-10-30","stars":46755},{"date":"2026-10-31","stars":47956},{"date":"2026-11-01","stars":49253},{"date":"2026-11-02","stars":50432},{"date":"2026-11-03","stars":51343},{"date":"2026-11-04","stars":51958},{"date":"2026-11-05","stars":52581},{"date":"2026-11-06","stars":53247},{"date":"2026-11-07","stars":53862},{"date":"2026-11-08","stars":54648},{"date":"2026-11-09","stars":55394},{"date":"2026-11-10","stars":55946}],"firstPublishedAt":"2026-10-25T23:19:02.913Z","commitSha":"6e8615eb0be0e843ceeebedae246d69f3e0a88d7","verifiedAt":"2026-11-02T02:56:26.669Z","lastCommitAt":"2026-11-10T14:00:29Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Switches off the ads and tracking apps on an Android or Google TV over Wi-Fi and gives it a clean Apple TV-style home screen.","summary":"A set of scripts and a small home-screen app for TVs running Android TV or Google TV. Over your Wi-Fi, using the TV's own developer options, it notes what is installed, switches off the maker's ad services, tracking and preloaded apps (nothing is uninstalled), then installs a 572 KB launcher with a black home screen, a dock and a grid, in 31 languages. One command puts everything back. It is meant to be driven by a coding assistant, about fifteen minutes; you can also run every command yourself. Lists for 407 TV models are included. Samsung, LG and Roku are not supported.","easeRationale":"Same work on all three systems, and it is terminal work: the computer needs Python 3, a full JDK between 17 and 22, and Android SDK platform 34 with build-tools 34 because the launcher is built on your machine (there is no ready APK), plus developer options and ADB turned on on the TV. Linux and Mac use the shell scripts, Windows the PowerShell twins with `-ExecutionPolicy Bypass`. A Docker route (tools/docker-build.sh) replaces the JDK and SDK with Docker.","notVerified":["Whether it works on your exact TV model, especially one without a tested list","Whether the restore script really puts every TV back the way it was","How long the launcher build takes and how much disk the Android SDK and JDK add","Which of the disabled packages were actually watching or listening, which the project says its own audit cannot prove","Whether a coding assistant follows the prompt's checks on your TV every time"]},"zh-Hans":{"summaryShort":"通过 Wi-Fi 关掉 Android TV 或 Google TV 上的广告和跟踪应用，并换上干净的 Apple TV 风格主屏幕。","summary":"一套脚本加一个小型主屏幕应用，适用于运行 Android TV 或 Google TV 的电视。它通过你的 Wi-Fi，利用电视自带的开发者选项，记录已安装的内容，关闭厂商的广告服务、跟踪和预装应用（不会卸载任何东西），然后安装一个 572 KB 的启动器，带有黑色主屏幕、程序坞和网格，支持 31 种语言。一条命令就能全部恢复。它设计成由编程助手来操作，大约十五分钟；你也可以自己执行每一条命令。附带 407 款电视型号的列表。不支持三星、LG 和 Roku。","easeRationale":"三种系统上的工作相同，而且都是终端操作：电脑需要 Python 3、17 到 22 之间的完整 JDK，以及 Android SDK platform 34 和 build-tools 34，因为启动器要在你的机器上构建（没有现成的 APK），电视上还需要打开开发者选项和 ADB。Linux 和 Mac 使用 shell 脚本，Windows 使用对应的 PowerShell 脚本，并带上 `-ExecutionPolicy Bypass`。有一条 Docker 路线 (tools/docker-build.sh)，用 Docker 代替 JDK 和 SDK。","notVerified":["它能否在你的确切电视型号上工作，尤其是不在已测试列表里的型号","恢复脚本是否真的能把电视恢复原状","构建启动器要花多长时间，Android SDK 和 JDK 会多占多少磁盘","被禁用的软件包中哪些真的在监视或监听，项目自己说它的审计无法证明这一点","编程助手是否每次都会在你的电视上遵守提示词里的检查"]}}},{"id":"R_kgDOLmKbDw","slug":"gcui-art__suno-api","name":"Suno API","owner":"gcui-art","fullName":"gcui-art/suno-api","type":"dev_tool","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):next: GHSA-2464-fhr4-w9xw, GHSA-2xp9-vwfh-vxw4, GHSA-36qx-fr4f-26g5","source":"check","title":"已知漏洞"},{"code":"bypasses_captcha_and_bot_detection","severity":"high","evidence":"绕过 Suno 的防护机制是这个项目的核心。README 的特性列表里提到:\"Solve CAPTCHAs automatically using 2Captcha and Playwright with rebrowser-patches\" 和 \"Automatically keep the account active\";package.json 引入了 @2captcha/captcha-solver、rebrowser-playwright-core、ghost-cursor-playwright 和 user-agents,.env.example 中的 BROWSER_GHOST_CURSOR 用来 \"simulate smooth mouse movements\"。破解一个网站的验证码和机器人检测违反了它的使用条款,在一些国家还违反法律;README 自己也说这个项目 \"unofficial... intended for learning and research purposes only\"。","source":"ai","title":"绕过验证码和机器人检测"},{"code":"your_login_cookie_lives_in_the_server","severity":"attention","evidence":"README 第一步要求你把已登录 suno.com 会话中完整的 Cookie 头复制到 SUNO_COOKIE 中(或者填入 Vercel 的控制面板),API 也接受在请求中传入的 cookie,\"in the Cookie header of your request\"。任何能接触到这台服务器或这个配置的人,都可以在 Suno 上冒充你;Dockerfile 甚至通过一个构建参数(ARG SUNO_COOKIE)把它直接写进去。","source":"ai","title":"你的登录 cookie 保存在服务器里"},{"code":"suggests_several_free_accounts_to_dodge_limits","severity":"attention","evidence":"README 写道:\"This comes in handy when, for example, you want to use multiple free accounts at the same time\",演示环境则 \"bound to a free Suno account, so it has daily usage limits\"。用多个免费账号来绕过服务的配额限制违反了常见的使用条款,可能导致账号被关闭。","source":"ai","title":"建议用多个免费账号来规避限额"},{"code":"no_releases_and_long_pause","severity":"attention","evidence":"检查记录显示没有任何发行版,最后一次提交是在 2026-03-06,距本次分析 243 天前,还有 109 个未关闭的议题。一个依赖未公开的私有接口和验证码破解手段的工具,一旦网站发生变化就会失效,所以无人维护的副本很可能会停止工作。","source":"ai","title":"没有发行版且长期停更"},{"code":"license_described_wrongly_in_the_readme","severity":"info","evidence":"README 的特性列表写着 \"Permissive open-source license, allowing you to freely integrate and modify\",但实际许可证是 LGPL-3.0-or-later,检查工具将其归类为 copyleft(copyleft 许可):对这个库本身的修改必须沿用相同的许可证。","source":"ai","title":"readme 中对许可证的描述有误"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Suno session cookie copied from your browser","2Captcha (paid CAPTCHA-solving service, ruCaptcha in Russia and Belarus)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"LGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["audio","development"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true}],"declaredAlternatives":[{"target":"suno","stance":"claims","evidence":"# replace with your suno-api URL"}],"stars":3306,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":9,"growthDays":6,"history":[{"date":"2026-11-05","stars":3299},{"date":"2026-11-06","stars":3299},{"date":"2026-11-07","stars":3301},{"date":"2026-11-08","stars":3304},{"date":"2026-11-09","stars":3305},{"date":"2026-11-10","stars":3306}],"firstPublishedAt":"2026-11-04T22:32:54.938Z","commitSha":"a2e6a823428903af715d3835d1cb44ffa336021d","verifiedAt":"2026-11-04T22:22:05.131Z","lastCommitAt":"2026-03-06T17:24:19Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An unofficial server that drives the Suno music site with your own account, so programs and AI agents can order songs.","summary":"Suno has no public API, so this project builds one: a small server you host that signs in to Suno with your own session cookie, gets past the site's CAPTCHA with the paid 2Captcha service, and then answers simple requests to make songs, write lyrics, extend a track, split stems or check your remaining credits. It is aimed at programmers and AI agents, not at listeners. You still need your own Suno account and credits, plus money on a 2Captcha balance.","easeRationale":"Mac is the one the README recommends and it is still a D: you open the browser developer tools to copy a login cookie by hand, register and pay at 2Captcha, then run the server with Node.js or Docker Compose and call it over HTTP. Windows and Linux work the same way, but the README warns they \"could get a pretty large number of CAPTCHAs\".","costNote":"The code is free (LGPL-3.0 or later), but it needs a Suno account with credits and a topped-up balance on 2Captcha, which charges per solved CAPTCHA. Neither cost is stated in the repository.","notVerified":["Whether it still works with Suno today, since the site can change its pages, its protections and its CAPTCHA at any time.","How much 2Captcha really costs per song generated.","Whether Suno closes accounts used this way.","Whether the flagged vulnerable version of Next.js (14.1.4) can be reached in your setup.","Who owns the songs produced this way, which depends on Suno's own terms for your plan."]},"zh-Hans":{"summaryShort":"一个非官方服务器,用你自己的账号操作 Suno 音乐网站,让程序和 AI 代理能够下单生成歌曲。","summary":"Suno 没有公开的 API,所以这个项目自己搭建了一个:一台由你自己托管的小型服务器,用你自己的会话 cookie 登录 Suno,借助付费的 2Captcha 服务绕过网站的验证码,然后处理简单的请求来生成歌曲、写歌词、延长曲目、分离音轨或查看剩余额度。它是给程序员和 AI 代理使用的,不是给听众用的。你仍然需要自己的 Suno 账号和额度,以及 2Captcha 账户里的余额。","easeRationale":"README 推荐使用 Mac,但它仍然只是 D 级:你需要打开浏览器的开发者工具手动复制登录 cookie,在 2Captcha 注册并付费,然后用 Node.js 或 Docker Compose 运行服务器,再通过 HTTP 调用它。Windows 和 Linux 的做法相同,但 README 警告说它们 \"could get a pretty large number of CAPTCHAs\"。","costNote":"代码本身是免费的(LGPL-3.0 或更高版本),但需要一个有额度的 Suno 账号,以及 2Captcha 账户里的余额,该服务按解决的验证码数量收费。这两项费用在仓库里都没有说明。","notVerified":["它是否仍然能在今天的 Suno 上正常工作,因为该网站可能随时更改页面、防护机制和验证码。","生成每首歌曲实际需要花多少 2Captcha 费用。","Suno 是否会关闭以这种方式使用的账号。","你的环境中是否能接触到被标记为存在漏洞的 Next.js 版本(14.1.4)。","这样生成的歌曲归谁所有,这取决于 Suno 针对你所用套餐的条款。"]}}},{"id":"R_kgDOMv5pGg","slug":"genmoai__mochi","name":"Mochi","owner":"genmoai","fullName":"genmoai/mochi","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"very_high_hardware_requirement","severity":"attention","evidence":"README.md「Hardware Requirements」:\"it requires approximately 60GB VRAM when running on a single GPU. While ComfyUI can optimize Mochi to run on less than 20GB VRAM, this implementation prioritizes flexibility over memory efficiency. When using this repository, we recommend using at least 1 H100 GPU.\" LoRA微调需要\"one H100 or A100 80GB GPU\"。","source":"ai","title":"硬件要求非常高"},{"code":"gpu_path_not_qualified_in_this_snapshot","severity":"attention","evidence":"README.md安装部分:\"This security refresh validates CPU media helpers and dependency resolution. GPU inference and fine-tuning need qualification before deployment.\" 更新日志止于\"November 26, 2024\",说明近期的工作集中在依赖项上,而不是模型新功能。","source":"ai","title":"此快照中gpu路径未经验证"},{"code":"pip_install_skips_security_override","severity":"attention","evidence":"README.md:\"Use the uv project install above: it applies the Pillow 13 override for MoviePy and Gradio. A plain `pip install -r requirements.txt` does not apply that override.\" pyproject.toml解释了原因:\"MoviePy and Gradio cap Pillow below the security-fixed release.\"","source":"ai","title":"pip install跳过了防护覆盖设置"},{"code":"weights_downloaded_outside_github","severity":"info","evidence":"README.md提供的权重来自Hugging Face(`genmo/mochi-1-preview`)、一个直接的zip链接(`weights.genmo.dev/weights.zip`)和一个BitTorrent磁力链接(`magnet:?xt=urn:btih:441da1af...`)。仓库没有GitHub发行版,所以没有任何内容在那里发布或留有版本记录。","source":"ai","title":"权重在github之外下载"},{"code":"research_preview_quality_limits","severity":"info","evidence":"README.md「Limitations」:\"Mochi 1 is a living and evolving checkpoint... The initial release generates videos at 480p today. In some edge cases with extreme motion, minor warping and distortions can also occur. Mochi 1 is also optimized for photorealistic styles so does not perform well with animated content.\" README中的API示例生成了31帧。","source":"ai","title":"研究预览版的质量存在局限"},{"code":"nsfw_filtering_is_partial","severity":"info","evidence":"README.md「Safety」:模型\"inherently reflect the biases and preconceptions found in their training data. While steps have been taken to limit NSFW content, organizations should implement additional safety protocols and careful consideration before deploying these model weights in any commercial services or products.\"","source":"ai","title":"nsfw内容过滤并不完整"},{"code":"company_backed_with_hosted_service","severity":"info","evidence":"README.md链接到Genmo的博客、其\"[Playground](https://www.genmo.ai/play)\"以及\"[Careers](https://jobs.ashbyhq.com/genmo)\"页面。同一家公司也托管着一个你同样可以自己运行的模型版本。","source":"ai","title":"由公司支持,并提供托管服务"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"kling","name":"Kling","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"ltx-studio","name":"LTX Studio","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":3765,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":4,"growthDays":6,"history":[{"date":"2026-11-06","stars":3762},{"date":"2026-11-07","stars":3764},{"date":"2026-11-08","stars":3764},{"date":"2026-11-09","stars":3765},{"date":"2026-11-10","stars":3765}],"firstPublishedAt":"2026-11-05T22:53:18.550Z","commitSha":"15c9bad662df1fa3e7b53136423695b286484c08","verifiedAt":"2026-11-05T22:45:59.575Z","lastCommitAt":"2026-10-22T23:35:02Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a written prompt into a short 480p video using Genmo's open Mochi 1 model on your own graphics card.","summary":"The code and the open weights of Mochi 1, a text-to-video model from the company Genmo: you type a prompt and it generates a few seconds of 480p video. It comes with a simple web page you start yourself, a command line tool, a Python API and a LoRA trainer for your own videos. It is aimed at people comfortable with a terminal: you install Python 3.11 with uv, install FFmpeg, download the model weights from Hugging Face, and the README asks for roughly 60 GB of video memory on one card.","easeRationale":"Linux is the documented path and it is a D: you clone the repository, install Python 3.11 with uv, install FFmpeg, run `python3 ./scripts/download_weights.py weights/` and then `python3 ./demos/gradio_ui.py` to get a local web page. The commands are Linux shell commands (`source .venv/bin/activate`), and the hardware note recommends \"at least 1 H100 GPU\", with flash-attn and A100/H100 fine-tuning, so it is tied to NVIDIA cards far bigger than a home computer's.","costNote":"Code and weights are Apache-2.0, with no key and no fee. Genmo also runs its own playground with the same model, and renting a suitable graphics card by the hour is a real cost.","notVerified":["Whether GPU generation still works with these dependency versions, which the README itself says needs qualifying","How good the videos look and how long each one takes on a given card","How much disk space the weights really take, since the README lists only dit.safetensors and decoder.safetensors without sizes and the checks could only estimate","Whether it can be made to run on a card with less than 60 GB of video memory without the ComfyUI version","Whether it runs on Windows or Mac, which the project never mentions","What the files behind the direct zip and the magnet link contain"]},"zh-Hans":{"summaryShort":"使用Genmo开放的Mochi 1模型,在自己的显卡上把文字提示变成一段简短的480p视频。","summary":"Mochi 1的代码和开放权重,这是Genmo公司开发的文本生成视频模型:输入一段提示词,就能生成几秒钟的480p视频。项目自带一个你自己启动的简单网页、一个命令行工具、一个Python API,以及一个可用自己视频训练LoRA的工具。它面向熟悉终端的用户:需要用uv安装Python 3.11,安装FFmpeg,从Hugging Face下载模型权重,README要求单卡大约60GB显存。","easeRationale":"Linux是文档中给出的路径,难度评级为D:你需要克隆仓库,用uv安装Python 3.11,安装FFmpeg,运行`python3 ./scripts/download_weights.py weights/`,然后运行`python3 ./demos/gradio_ui.py`来获取本地网页。这些命令都是Linux shell命令(`source .venv/bin/activate`),硬件说明建议\"at least 1 H100 GPU\",微调还需要flash-attn和A100/H100,因此它依赖的NVIDIA显卡远比家用电脑的显卡强大。","costNote":"代码和权重采用Apache-2.0协议,无需密钥,免费使用。Genmo还运营着使用同一模型的自家体验平台,按小时租用一张合适的显卡则是真实的开销。","notVerified":["在这些依赖版本下GPU推理是否仍然可用,README自己也说这一点需要核实","生成的视频效果如何,在特定显卡上每段视频要花多长时间","权重实际占用多少磁盘空间,因为README只列出了dit.safetensors和decoder.safetensors而没有给出大小,检查也只能估算","不使用ComfyUI版本的情况下,能否在显存低于60GB的显卡上运行","是否能在Windows或Mac上运行,项目完全没有提到","直接zip链接和磁力链接背后的文件到底包含什么内容"]}}},{"id":"R_kgDORawBOA","slug":"gentleman-programming__gentle-ai","name":"Gentle-AI","owner":"Gentleman-Programming","fullName":"Gentleman-Programming/gentle-ai","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"install_pipes_script_from_internet_to_shell","severity":"attention","evidence":"README.md 的 Get started 部分：`curl -fsSL https://raw.githubusercontent.com/Gentleman-Programming/gentle-ai/main/scripts/install.sh | bash`。这个脚本会执行当时 main 分支上的任何内容，而且还会修改你 shell 的启动文件（\"install.sh adds one marked line (# gentle-ai installer) to the startup file your shell reads: ~/.zshrc for zsh...\"）；`install.ps1` \"adds the directory to your user Path\"。","source":"ai","title":"安装过程把网上的脚本直接接到 shell 里执行"},{"code":"telemetry_on_by_default","severity":"attention","evidence":"README.md 链接了一个叫 \"Telemetry — What we count, and how to turn it off\" 的文档页面（docs/telemetry.md）。除非你关掉它，否则计数就会发生，而具体统计了什么，README 本身并没有说明。","source":"ai","title":"遥测功能默认开启"},{"code":"rewrites_your_agent_configuration","severity":"attention","evidence":"该仓库的描述写着它 \"configures the AI coding agents you already use: Claude Code, Cursor, OpenCode, Codex, Pi, and more\"，README 里写着 \"Your configs are snapshotted before every write\"。它会编辑决定你的编程助手行为的文件，包括 MCP 服务器和技能；列出的防护措施是一个 \"Security deny-list\"，会 \"Blocks ~/.ssh, .env and credential files\"，另外还有配置备份。","source":"ai","title":"重写你的助手配置"},{"code":"dependency_code_committed_to_the_repository","severity":"info","evidence":"顶层文件列表里包含一个 `node_modules/`，而这是一个 Go 项目，它的 package.json 没有任何依赖，只有一个会失败的测试脚本（\"test\": \"echo \\\"Error: no test specified\\\" && exit 1\"）。因此第三方 JavaScript 文件被直接存放在了仓库里。","source":"ai","title":"依赖代码被提交进了仓库"},{"code":"self_reported_single_run_benchmarks","severity":"info","evidence":"README.md 的 \"Measured\" 部分：前后对比表来自作者自己的运行结果，并写明 \"each cell is a single run (n=1), and cost is API-equivalent\"。这些数字还来自另一个项目 Gentle Shell，而不是只来自这个二进制程序本身。","source":"ai","title":"作者自行报告的单次运行基准测试"},{"code":"review_mode_enabled_by_default","severity":"info","evidence":"README.md：\"Receipt-Driven Development (RDD) is on by default and opt-out: run `gentle-ai review mode disable` to turn it off.\" 额外的审查步骤意味着更多的模型调用，因此也意味着你的助手会花更多钱，除非你把它关掉。","source":"ai","title":"审查模式默认开启"},{"code":"trademark_limits_on_top_of_mit","severity":"info","evidence":"README.md 的商标声明：\"The Gentle AI™ and Engram™ names and logos are trademarks of Alan Buscaglia... The MIT License applies to the code; it does not permit implying endorsement or official affiliation.\" 代码是 MIT 许可的，名称不是。","source":"ai","title":"在 MIT 之外还有商标方面的限制"}],"platforms":["mac","linux","windows"],"tools":[],"buildTools":["node","go"],"disk":{"minMB":18,"maxMB":53,"basis":"installer"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":8741,"downloads":552409,"downloadsWeek":6702,"downloadsDays":1,"growth":173,"growthDays":3,"history":[{"date":"2026-11-09","stars":8668},{"date":"2026-11-10","stars":8741}],"firstPublishedAt":"2026-11-08T23:01:18.722Z","commitSha":"3108ef03fd81d96518eab3b3e469ed69d4e0f52d","verifiedAt":"2026-11-08T22:48:29.191Z","lastCommitAt":"2026-11-10T18:27:46Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A terminal tool that sets up the AI coding assistant you already use with memory, a fixed workflow and skills.","summary":"Gentle-AI writes the configuration for AI coding assistants you already have, such as Claude Code, Cursor, Codex, Copilot or Gemini CLI. It adds a memory store so the assistant keeps project decisions between sessions, a step-by-step working method, a library of skills, MCP servers, personas and a review pass over finished work that is on by default. It does not install any assistant for you and it snapshots each configuration file before writing. For developers: install with Homebrew or a shell script on Mac and Linux, or with Go on Windows, then run `gentle-ai`.","easeRationale":"On Mac the ready binary comes from `brew install gentleman-programming/tap/gentle-ai` (Linux uses a shell script), after which you run `gentle-ai` once to pick your agents and components — so it is an install plus one step, in the terminal. Windows has no binary in the releases: the README installs it with `go install`, which means putting Go 1.25.10 or later on the machine and fixing your PATH by hand.","costNote":"The tool itself is free (MIT), but it only does something on top of an AI coding assistant, and the usual ones are paid subscriptions or paid API usage.","notVerified":["What the telemetry sends and whether turning it off really stops it","Which of the 17 advertised assistants work fully, since the README says features \"can differ\" per integration","How much extra the default review and memory steps add to your agent's bill","Whether the configuration backups let you get your previous setup back cleanly","Whether the Windows path works without Go, given there is no Windows binary in the latest releases"]},"zh-Hans":{"summaryShort":"一个终端工具，为你已经在用的 AI 编程助手配置记忆、固定工作流程和技能。","summary":"Gentle-AI 会为你已经拥有的 AI 编程助手（比如 Claude Code、Cursor、Codex、Copilot 或 Gemini CLI）写入配置。它会添加一个记忆存储，让助手在不同会话之间记住项目的决定，还有一套分步骤的工作方法、一个技能库、MCP 服务器、人设，以及一个默认开启的、针对完成工作的审查环节。它不会替你安装任何助手，而且在写入每个配置文件之前都会先做快照。面向开发者：在 Mac 和 Linux 上用 Homebrew 或一个 shell 脚本安装，在 Windows 上用 Go 安装，然后运行 `gentle-ai`。","easeRationale":"在 Mac 上，现成的二进制文件来自 `brew install gentleman-programming/tap/gentle-ai`（Linux 使用一个 shell 脚本），之后你运行一次 `gentle-ai` 来选择你的代理和组件——所以这是安装加一步，都在终端里完成。Windows 的发行版里没有二进制文件：README 用 `go install` 来安装，这意味着要在机器上装好 Go 1.25.10 或更新版本，并手动修复你的 PATH。","costNote":"这个工具本身是免费的（MIT 许可），但它只是在某个 AI 编程助手之上起作用，而常见的助手都是付费订阅或付费 API 用量。","notVerified":["遥测功能到底发送了什么，关掉它是否真的能停止发送","17 个宣传支持的助手里有多少能完整工作，因为 README 说各功能 \"can differ\"，会因集成对象不同而有差异","默认开启的审查和记忆步骤会给你的助手账单增加多少额外开销","配置备份能否让你干净地恢复之前的设置","既然最新发行版里没有 Windows 二进制文件，Windows 这条路径在没有 Go 的情况下能不能用"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTE1ODM1OTM=","slug":"genymobile__scrcpy","name":"scrcpy","owner":"Genymobile","fullName":"Genymobile/scrcpy","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"requires_usb_debugging_on_the_phone","severity":"attention","evidence":"README「准备工作」部分写道：\"Make sure you enabled USB debugging on your device(s)\"，部分手机还需要开启 \"USB debugging (Security Settings)\"。这项设置会让任何你连接并信任的电脑都能控制手机，所以不再需要时最好把它关掉。","source":"ai","title":"需要在手机上开启 USB 调试"},{"code":"copycat_downloads_elsewhere","severity":"attention","evidence":"README 开头就提出警告：\"This GitHub repo (https://github.com/Genymobile/scrcpy) is the only official source for the project. Do not download releases from random websites, even if their name contains scrcpy.\" 也就是说，这个名字被其他网站盗用，只有本仓库发布页里的文件才是项目官方的。","source":"ai","title":"其他地方的仿冒下载"},{"code":"many_open_issues","severity":"info","evidence":"2942 个未关闭的 issue，对应 153603 个星标。项目很活跃（最新版本 v5.0.1 发布于 2026-10-08），README 把人们先引导到 FAQ，这说明其中大部分可能是关于具体手机型号的支持问题。","source":"ai","title":"大量未关闭的 issue"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":11,"maxMB":32,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["hardware","network_system","video"],"alternativeTo":[{"slug":"vysor","name":"Vysor","confirmed":true}],"declaredAlternatives":[],"stars":154509,"downloads":30989646,"downloadsWeek":248455,"downloadsDays":7,"growth":568,"growthDays":7,"history":[{"date":"2026-10-26","stars":153218},{"date":"2026-10-27","stars":153317},{"date":"2026-10-28","stars":153405},{"date":"2026-10-29","stars":153511},{"date":"2026-10-30","stars":153603},{"date":"2026-10-31","stars":153678},{"date":"2026-11-01","stars":153771},{"date":"2026-11-02","stars":153863},{"date":"2026-11-03","stars":153941},{"date":"2026-11-04","stars":154030},{"date":"2026-11-05","stars":154114},{"date":"2026-11-06","stars":154174},{"date":"2026-11-07","stars":154246},{"date":"2026-11-08","stars":154340},{"date":"2026-11-09","stars":154423},{"date":"2026-11-10","stars":154509}],"firstPublishedAt":"2026-10-27T01:10:12.630Z","commitSha":"a60891aea193d92e7e5c3942700eca63f9d19a5f","verifiedAt":"2026-10-31T01:39:16.365Z","lastCommitAt":"2026-10-08T06:41:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Shows your Android phone screen on the computer and lets you use it with the mouse and keyboard, by cable or Wi-Fi.","summary":"scrcpy puts your Android phone or tablet screen in a window on the computer and lets you tap, type and copy-paste with the computer's mouse and keyboard. It can also forward the phone's sound, record the screen to a video file, use the phone camera as a webcam and work over Wi-Fi. Nothing is installed on the phone and no account or internet connection is needed, but you must switch on USB debugging in the phone's developer options. The downloads are small (10 to 17 MB) and you start it from a terminal when you want options.","easeRationale":"Windows is the easiest: you unzip scrcpy-win64-v5.0.1.zip (11 MB), which already contains everything needed, and the one extra step is turning on USB debugging on the phone as the README requires. On macOS and Linux the project points to per-system instructions that normally mean installing it through a package manager such as Homebrew and starting it from a terminal, and all the extra features (quality, recording, camera) are command-line options like `scrcpy --video-codec=h265 -m1920` on every system.","costNote":"Free and Apache-2.0 licensed. The author only asks for optional donations (GitHub Sponsors, Liberapay, PayPal).","notVerified":["Whether your phone model lets the computer control it: the README notes Xiaomi phones can refuse with an \"INJECT_EVENTS permission\" error until an extra setting is enabled","Whether sound forwarding and camera use work on your phone, since the README limits them to Android 11+ and Android 12+","The real picture quality and delay on your cable, Wi-Fi and phone, as against the claimed 30 to 120 fps and 35 to 70 ms","Whether the macOS and Linux installs are as direct as the Windows one, since the steps live in separate documentation pages","Whether the file you downloaded is the author's, which the project asks you to confirm with the release signatures"]},"zh-Hans":{"summaryShort":"把你的安卓手机屏幕显示在电脑上，让你用鼠标和键盘操作它，支持有线或 Wi-Fi 连接。","summary":"scrcpy 把你的安卓手机或平板屏幕放进电脑上的一个窗口，让你用电脑的鼠标和键盘点击、输入和复制粘贴。它还能转发手机的声音，把屏幕录制成视频文件，把手机摄像头当作网络摄像头使用，并支持通过 Wi-Fi 连接。手机上不会安装任何东西，也不需要账号或网络连接，但你必须在手机的开发者选项中打开 USB 调试。下载包很小（10 到 17 MB），如果需要更多选项，可以从终端启动。","easeRationale":"Windows 上最简单：解压 scrcpy-win64-v5.0.1.zip（11 MB），里面已经包含所需的一切，唯一额外的步骤是按 README 的要求在手机上打开 USB 调试。在 macOS 和 Linux 上，项目指向各系统专门的说明，通常意味着通过像 Homebrew 这样的包管理器安装，再从终端启动，而所有额外功能（画质、录制、摄像头）在各系统上都是命令行选项，例如 `scrcpy --video-codec=h265 -m1920`。","costNote":"免费，采用 Apache-2.0 许可。作者只接受自愿捐赠（GitHub Sponsors、Liberapay、PayPal）。","notVerified":["你的手机型号是否允许电脑控制它：README 指出小米手机可能会出现「INJECT_EVENTS permission」错误，需要先开启一项额外设置","声音转发和摄像头使用能否在你的手机上正常工作，因为 README 把它们限定在安卓 11 及以上和安卓 12 及以上","在你的数据线、Wi-Fi 和手机上，实际的画质和延迟与所说的 30 到 120 fps、35 到 70 ms 相比如何","macOS 和 Linux 的安装是否像 Windows 那样直接，因为相关步骤分散在不同的文档页面","你下载的文件是否来自作者本人，项目要求你用发布版的签名来确认这一点"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMzc3MzMyNjY=","slug":"getpublii__publii","name":"Publii","owner":"GetPublii","fullName":"GetPublii/Publii","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"downloads_come_from_the_vendor_site","severity":"info","evidence":"这里记录的最近三个发行版(v.0.48.2-build-17583、v.0.48.1-build-17582、v.0.48.1-build-17565)都没有附带任何平台的安装文件;README 反而把你指向 getpublii.com/download 上的 \"Download Publii (.exe, .dmg, .deb, .rpm, .AppImage)\",所以你安装的文件并不是来自 GitHub。","source":"ai","title":"下载文件来自厂商自己的网站"},{"code":"still_before_version_one","severity":"info","evidence":"package.json 中写着 \"version\": \"0.48.2\",README 里也是 \"Current version: 0.48.2 - build 17583\",尽管这个仓库是在 2018 年创建的。截至本次分析时,还有 307 个未关闭的议题。","source":"ai","title":"版本号仍未到 1.0"},{"code":"keeps_your_hosting_passwords","severity":"info","evidence":"为了上传网站,这个应用会保存你的服务器凭据:README 列出了 \"standard HTTP/HTTPS servers, Netlify, Amazon S3, GitHub Pages and Google Cloud or SFTP\",Linux 安装包还声明依赖系统密码存储工具 \"gnome-keyring\"。","source":"ai","title":"保存你的主机密码"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Your own hosting account for publishing (SFTP, Netlify, GitHub Pages, Amazon S3 or Google Cloud)"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","development"],"alternativeTo":[{"slug":"wix","name":"Wix","confirmed":true},{"slug":"squarespace","name":"Squarespace","confirmed":true}],"declaredAlternatives":[],"stars":7402,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":12,"growthDays":6,"history":[{"date":"2026-11-05","stars":7392},{"date":"2026-11-06","stars":7395},{"date":"2026-11-07","stars":7396},{"date":"2026-11-08","stars":7396},{"date":"2026-11-09","stars":7398},{"date":"2026-11-10","stars":7402}],"firstPublishedAt":"2026-11-04T22:38:15.198Z","commitSha":"01cfb7ab0aa15f8947ebc5ec35d32793f4f59ca8","verifiedAt":"2026-11-04T22:22:51.276Z","lastCommitAt":"2026-11-07T19:07:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop app to build a blog or small website without coding, then upload it to your own host with one click.","summary":"Publii is a program you install on your computer to build websites and blogs. It works like WordPress to use (write posts, pick a theme, change settings), but the result is a set of plain files, so the published site needs no database and nothing to keep updated. You work offline and then send the site to your host: a normal web server, SFTP, Netlify, GitHub Pages, Amazon S3 or Google Cloud. There are installers for Windows, Mac and Linux, and it is free under the GPL-3.0 licence.","easeRationale":"All three platforms are an A: ready installers are published for Windows (.exe), Mac (.dmg, macOS 12 or newer) and Linux (.deb, .rpm, .AppImage), and the Windows installer is one-click with a desktop shortcut. Nothing else has to be installed, although the Linux packages ask for gnome-keyring to store your upload passwords.","costNote":"The app is free and GPL-3.0 licensed; the project takes donations through Open Collective. You still pay whoever hosts the finished site, if anything.","notVerified":["Whether the installers are signed and notarised on each system (the build settings turn off signing for the Mac disk image and run notarisation as a separate step).","Whether the extra themes and add-ons offered around the app are free or paid.","Whether publishing works smoothly with your particular host.","How much disk the app plus your site files and images actually use.","How the app behaves with a large site (hundreds of posts and images)."]},"zh-Hans":{"summaryShort":"一款桌面应用,无需编程即可创建博客或小型网站,并可以一键上传到你自己的主机。","summary":"Publii 是一个安装在你电脑上、用来制作网站和博客的程序。使用方式和 WordPress 类似(写文章、选主题、改设置),但生成的结果是一组纯静态文件,所以发布后的网站不需要数据库,也没有什么需要持续更新维护的东西。你可以离线编辑,完成后再把网站发布到自己的主机:普通的网页服务器、SFTP、Netlify、GitHub Pages、Amazon S3 或 Google Cloud。它提供 Windows、Mac 和 Linux 的安装程序,并在 GPL-3.0 许可下免费使用。","easeRationale":"三个平台都是 A 级:Windows(.exe)、Mac(.dmg,需要 macOS 12 或更新版本)和 Linux(.deb、.rpm、.AppImage)都提供现成的安装程序,Windows 的安装程序一键完成,还会创建桌面快捷方式。不需要再安装别的东西,不过 Linux 的安装包需要依赖 gnome-keyring 来保存你的上传密码。","costNote":"这个应用是免费的,采用 GPL-3.0 许可;项目通过 Open Collective 接受捐赠。不过你仍然要为最终网站的托管方付费(如果需要付费的话)。","notVerified":["安装程序是否在每个系统上都经过签名和公证(构建设置关闭了 Mac 磁盘镜像的签名,并把公证作为单独的步骤执行)。","围绕这款应用提供的额外主题和插件是免费的还是收费的。","发布功能是否能在你具体使用的主机上顺利运行。","这款应用加上你的网站文件和图片实际会占用多少磁盘空间。","在一个大型网站(数百篇文章和图片)下,这款应用的表现如何。"]}}},{"id":"R_kgDOLx-IRA","slug":"getrepliq__clipify","name":"Clipify","owner":"GetRepliq","fullName":"GetRepliq/Clipify","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"readme_and_license_file_disagree","severity":"attention","evidence":"README 说 \"Clipify is released under the MIT license. This license permits anyone to use, modify, and distribute the software freely\",但仓库里的 LICENSE 文件却是 \"GNU GENERAL PUBLIC LICENSE Version 3\",GitHub 也标注为 GPL-3.0。GPL-3.0 是著佐权(copyleft)协议:它要求任何分发修改版本的人以同样的条款公开源代码,这和 README 承诺的内容不一致。","source":"ai","title":"readme 和许可证文件不一致"},{"code":"no_install_steps_and_empty_requirements_file","severity":"attention","evidence":"仓库顶层的文件是 \".DS_Store, .vscode/, Clipify Build Structure.jpg, LICENSE, README.md, main.py, requirements.txt, tests/\",requirements.txt 没有任何内容。README 列出了 MoviePy、OpenCV、SpeechRecognition、spaCy 和 TensorFlow/PyTorch 作为用到的技术,但仓库里没有任何地方说明要安装哪些版本,也没说怎么启动程序。","source":"ai","title":"没有安装步骤,依赖文件也是空的"},{"code":"no_changes_since_august_2024","severity":"attention","evidence":"检查记录显示最后一次提交是在 2024-08-25,距本次审查 803 天,没有发布版本,未解决的 issue 数为零,而 README 仍然说文档和贡献指南以后才会提供。","source":"ai","title":"自 2024 年 8 月以来没有任何更新"},{"code":"features_described_as_possible_rather_than_built","severity":"info","evidence":"README 自己的措辞对机器学习部分是有条件的:\"TensorFlow/PyTorch: These machine learning frameworks can be utilized to enhance the selection process of video segments\"。仓库里只有一个 main.py,所以说明中描述的语音识别、文本分析、场景检测、尺寸调整和字幕生成这一整套流程,没法对照独立的模块来验证。","source":"ai","title":"功能只是设想,尚未真正实现"},{"code":"readme_images_point_to_a_different_account","severity":"info","evidence":"README 里构建结构的图片是从 \"github.com/Aby-ss/Clipify/assets/...\" 加载的,而这里分析的仓库是 GetRepliq/Clipify。README 的文字是为另一个账号写的,所以其中一些内容可能描述的是项目的另一份副本。","source":"ai","title":"readme 里的图片指向另一个账号"},{"code":"no_mention_of_platform_rules_for_reposting_video","severity":"info","evidence":"项目描述是 \"Instantly transform long videos into engaging, social media-ready clips\",README 还提供了 \"direct sharing to social media platforms\",但仓库里没有任何地方提到你需要拥有所剪视频的权利。剪辑别人的视频并重新发布,要看那个所有者是否允许。","source":"ai","title":"没有提到重新发布视频时的平台规则"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["video","social_marketing"],"alternativeTo":[{"slug":"opusclip","name":"OpusClip","confirmed":true},{"slug":"streamladder","name":"StreamLadder","confirmed":true}],"declaredAlternatives":[],"stars":112,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":8,"growthDays":6,"history":[{"date":"2026-11-07","stars":105},{"date":"2026-11-08","stars":105},{"date":"2026-11-09","stars":111},{"date":"2026-11-10","stars":112}],"firstPublishedAt":"2026-11-06T22:57:16.637Z","commitSha":"1c85e76f18df48f2e9785e41ea79fb1ca0c592af","verifiedAt":"2026-11-06T22:40:06.280Z","lastCommitAt":"2024-08-25T12:04:20Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Python project meant to cut long videos into short social clips with captions, with no setup steps written down.","summary":"Clipify is meant to take a long video, pick the parts worth keeping, crop them to the vertical shape used by Reels, TikTok and Stories, and add captions made from the speech. In practice the repository holds one main.py file, a tests folder and an empty requirements.txt, and the README itself says guidance on how to \"set up the project locally\" will come later. There are no releases and no commits since August 2024, so anyone trying it has to read the code and work out the libraries and commands alone.","easeRationale":"Grade E on all three systems for the same reason: the README has no install or run section at all (\"Detailed guidelines on how to contribute, set up the project locally, and submit pull requests will be provided soon\"), requirements.txt is empty and there are no releases, so you would have to read main.py and install the libraries it needs yourself.","notVerified":["Whether the program runs at all, since no install steps, library versions or releases exist","Which libraries main.py actually imports and whether FFmpeg has to be installed for the video part","Whether the automatic choice of clips and the captions are any good","Real disk use: the checks estimate 100 to 300 MB from the project files alone","Whether it works on Windows, macOS and Linux, as the repository documents no system"]},"zh-Hans":{"summaryShort":"一个 Python 项目,原本打算把长视频剪成带字幕的社交媒体短片,但没有写明具体的安装步骤。","summary":"Clipify 的设想是拿一段长视频,挑出值得保留的片段,把它们裁剪成 Reels、TikTok 和 Stories 使用的竖版比例,再根据语音生成字幕。但实际上,这个仓库只有一个 main.py 文件、一个测试文件夹和一个空的 requirements.txt,README 自己也说关于如何 \"set up the project locally\" 的说明以后才会提供。自 2024 年 8 月以来没有发布版本也没有提交,所以想试用的人只能自己读代码,自己摸清要用哪些库、怎么运行。","easeRationale":"三个系统都是 E 级,原因相同:README 完全没有安装或运行的章节(\"Detailed guidelines on how to contribute, set up the project locally, and submit pull requests will be provided soon\"),requirements.txt 是空的,也没有发布版本,所以你得自己读 main.py,自己装它需要的库。","notVerified":["这个程序能不能跑起来,因为没有安装步骤、库版本号,也没有发布版本","main.py 实际导入了哪些库,视频部分是否需要安装 FFmpeg","自动挑选片段和生成字幕的效果好不好","实际占用的磁盘空间:检查仅凭项目文件估计为 100 到 300 MB","能否在 Windows、macOS 和 Linux 上运行,因为仓库没有记录任何系统"]}}},{"id":"R_kgDOHFhdAg","slug":"ghostty-org__ghostty","name":"Ghostty","owner":"ghostty-org","fullName":"ghostty-org/ghostty","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"mac":"A","linux":"B"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布在 2022-11-17（1447 天前）","source":"check","title":"版本发布已经过时"},{"code":"downloads_come_from_the_website_not_this_repository","severity":"info","evidence":"README 的 Download 部分只指向 \"https://ghostty.org/download\"，GitHub 上唯一的 release 是 2022-11-17 的滚动标签 \"tip\"，里面是 Mac 文件，其中两个还叫 \"debug-fast\" 和 \"debug-slow\"。这就是为什么 Radar 的检查会把这些 release 标记为过时：真正的版本发布在项目自己的网站上，而 Radar 并没有检查那里。","source":"ai","title":"下载来自官网，而不是这个仓库"},{"code":"crash_files_can_hold_sensitive_data","severity":"info","evidence":"README 的 Crash Reports 部分写道：\"Crash reports are not automatically sent anywhere off your machine\"，但警告里又补充说报告 \"does contain the full stack memory of each thread at the time of the crash\"，并且 \"can also contain sensitive data\"。这些文件会留在 ~/.local/state/ghostty/crash 里，直到你删除或者上传它们。","source":"ai","title":"崩溃文件可能包含敏感数据"},{"code":"no_windows_app","severity":"info","evidence":"README 描述了 macOS 应用（SwiftUI、Metal）和 Linux 应用（GTK）；只有 libghostty-vt 这个库被列为 \"compatible for macOS, Linux, Windows, and WebAssembly\"。Windows 用户拿到的是一个给开发者用的库，而不是一个可以安装的终端。","source":"ai","title":"没有 Windows 应用"}],"platforms":["mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"mobaxterm","name":"MobaXterm","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":63966,"downloads":28,"downloadsWeek":14,"downloadsDays":6,"growth":531,"growthDays":7,"history":[{"date":"2026-11-04","stars":63535},{"date":"2026-11-05","stars":63631},{"date":"2026-11-06","stars":63730},{"date":"2026-11-07","stars":63820},{"date":"2026-11-08","stars":63884},{"date":"2026-11-09","stars":63928},{"date":"2026-11-10","stars":63966}],"firstPublishedAt":"2026-11-04T01:36:50.397Z","commitSha":"c7f68a7f48c6a5b357689eb8c2328bc2dab5d8ed","verifiedAt":"2026-11-04T01:28:21.633Z","lastCommitAt":"2026-11-10T21:10:44Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A fast terminal window for Mac and Linux, with tabs, splits and a real settings screen, free and built to feel native.","summary":"Ghostty is a terminal: the window where you type commands. It aims to be quick, complete and to look like a proper Mac or Linux app rather than a plain box, with several windows, tabs, split panes, a settings screen, and drawing done by the graphics card. It also supports modern extras such as showing images in the terminal. On top of the app there is libghostty, a library other developers can embed to put a terminal inside their own program. Free and MIT licensed. It is useful mainly to people who already work in a terminal.","easeRationale":"On Mac you download Ghostty.dmg and drag it in, nothing else. On Linux the project's own releases carry no Linux file: you install it from your distribution's packages, Flatpak or Snap (the repository holds flatpak/, snap/ and PACKAGING.md for the packagers), which is one step more and depends on your distribution.","notVerified":["Whether the installer on ghostty.org is built from exactly the code in this repository.","Which Linux distributions carry an up-to-date package, since the project leaves that to packagers.","How much disk and memory it really uses, which the repository does not state.","Whether the speed claims hold on your machine, as the comparisons in the README come from the project itself."]},"zh-Hans":{"summaryShort":"一个适用于 Mac 和 Linux 的快速终端，有标签页、分屏和真正的设置界面，免费，而且用起来很像原生应用。","summary":"Ghostty 是一个终端：你输入命令的那个窗口。它追求快速、功能齐全，看起来像一个真正的 Mac 或 Linux 应用，而不是一个简陋的方框，支持多个窗口、标签页、分屏、设置界面，并由显卡来完成绘制。它还支持一些现代化的附加功能，比如在终端里显示图片。在这个应用之上还有 libghostty，一个库，其他开发者可以把它嵌入到自己的程序里，从而拥有一个终端。免费，MIT 许可。它主要适合已经在用终端工作的人。","easeRationale":"在 Mac 上，你只需要下载 Ghostty.dmg 再拖进去就行，没别的步骤。在 Linux 上，项目自己发布的版本不包含 Linux 文件：你需要从你所用发行版的软件包、Flatpak 或 Snap 来安装（仓库里有给打包者用的 flatpak/、snap/ 和 PACKAGING.md），这多了一步，而且取决于你用的发行版。","notVerified":["ghostty.org 上的安装包是否确实是用这个仓库里的代码构建的。","哪些 Linux 发行版有更新的软件包，因为这件事项目交给了各个打包者。","它实际占用多少磁盘和内存，仓库里并没有说明。","README 里的速度对比是否在你的机器上也成立，因为那些对比数据来自项目自己。"]}}},{"id":"R_kgDOLJ3URA","slug":"giejay__immich-android-tv","name":"Immich Android TV","owner":"giejay","fullName":"giejay/Immich-Android-TV","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{},"risk":"attention","signals":[{"code":"no_app_store_and_no_checksum_for_the_files","severity":"attention","evidence":"项目信息显示 \"GitHub releases: 69\",最近的三个版本(6.2.0、v6.1.0、v6.0.0)都被标注为 \"no platform binaries\"。仓库没有指向任何应用商店,README 也没有为任何文件发布校验和或签名。","source":"ai","title":"没有应用商店,文件也没有校验和"},{"code":"needs_your_own_immich_server","severity":"info","evidence":"README 写道:\"This app is compatible with Immich server v3.0.0 and newer\",所有功能都是围绕查看 \"those uploaded photos and videos\"。Immich 本身是一个独立的自托管项目(github.com/immich-app/immich);除了被标为 Done 的 \"Demo environment\" 外,这个应用在没有那个服务器时什么也存不了、也显示不了。","source":"ai","title":"需要你自己的 immich 服务器"},{"code":"api_key_with_seventeen_read_permissions","severity":"info","evidence":"README 的 \"Required API Permissions\" 列出了 activity.read、album.download、album.read、archive.read、asset.download、asset.read、asset.view、face.read、folder.read、library.read、memory.read、partner.read、person.read、session.read、tag.asset、tag.read 和 timeline.read。这个存在电视盒子上的密钥,可以读取并下载你的整个照片库,包括已归档的项目和人脸数据。","source":"ai","title":"api 密钥拥有十七项读取权限"},{"code":"firebase_is_part_of_the_build","severity":"info","evidence":"构建步骤 2:\"Create an account at firebase and create a google-services.json file, or cp app/google-services.example app/google-services.json\"。Firebase 是谷歌的一项服务,通常用于崩溃报告或数据分析,仓库没有说明应用向它发送了什么。","source":"ai","title":"firebase 是构建过程的一部分"},{"code":"sign_in_by_phone_goes_through_a_second_project","severity":"info","evidence":"功能表把 \"Sign in by phone (https://github.com/giejay/Immich-Android-TV-Authentication)\" 列为 Done。配对过程因此涉及同一作者的另一个独立程序,这个仓库没有说明用这种方式登录时会经手什么数据。","source":"ai","title":"手机登录要经过第二个项目"},{"code":"demo_server_address_and_keys_live_in_a_build_file","severity":"info","evidence":"构建步骤 3 要求编译应用的人 \"copy app/strings_other.xml.example to app/src/main/res/values/strings_other.xml and modify the address and API keys for your demo server\",功能表把 \"Demo environment\" 标为 Done。这个安装包没有展示任何已编译文件里到底有什么内容。","source":"ai","title":"演示服务器的地址和密钥写在一个构建文件里"},{"code":"several_features_still_marked_todo","severity":"info","evidence":"功能表把 \"Add places/tags view\"、\"Searching in and for albums\" 和 \"Smarter merging of portrait photos\" 标为 Todo,把 \"Casting capabilities\" 标为 \"Out of scope\"。","source":"ai","title":"好几项功能仍标为待办"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Your own Immich server (API key)"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image","hardware","video"],"alternativeTo":[{"slug":"google-photos","name":"Photos","confirmed":true},{"slug":"icloud-photos","name":"iCloud Photos","confirmed":true},{"slug":"plex","name":"Plex","confirmed":true}],"declaredAlternatives":[],"stars":919,"downloads":14522,"downloadsWeek":121,"downloadsDays":3,"growth":6,"growthDays":6,"history":[{"date":"2026-11-07","stars":915},{"date":"2026-11-08","stars":917},{"date":"2026-11-09","stars":918},{"date":"2026-11-10","stars":919}],"firstPublishedAt":"2026-11-06T23:09:33.782Z","commitSha":"037cb7ce0d94f36113afd0bec19b84f23ec705ed","verifiedAt":"2026-11-06T23:02:35.895Z","lastCommitAt":"2026-10-16T07:45:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android TV app that shows the photos and videos from your own Immich server, with slideshows and a screensaver.","summary":"Immich Android TV turns a TV box into a photo frame for a photo server you already run at home. You point the app at your Immich server (version 3.0.0 or newer) with an API key or by signing in from your phone, then browse albums, people, random and seasonal photos, play slideshows with transitions and set the app as the TV screensaver. The repository has 69 releases, and the README's build steps document making the file yourself: \"Build apk with ./gradlew assembleRelease\". Without an Immich server of your own, the README lists only a \"Demo environment\".","easeRationale":"Android TV is the only system it runs on, and there the package supports a D: the README's only install path is building the file yourself (\"Build apk with ./gradlew assembleRelease\", after cloning with submodules and creating or copying a google-services.json), you must already run an Immich server \"v3.0.0 and newer\" and create an API key with the 17 permissions listed, and turning the app into the screensaver needs developer mode, USB debugging and adb commands typed from a computer. Computers get no grade; there is no desktop version.","costNote":"The app is free and GPL-3.0; the README only asks for an optional donation (\"buying me a coffee or a beer\"). Running the Immich server it connects to is your own cost.","notVerified":["Whether it works on your TV box model, since the repository names no tested devices","Whether a build you get from anywhere matches the source in this repository, as the README offers no checksum or signature","What the app sends to Firebase","What passes through the separate sign-in-by-phone project when you use it","How much space and memory the app uses on the TV","How well slideshows and video playback perform on low-powered TV boxes"]},"zh-Hans":{"summaryShort":"一个 Android TV 应用,展示你自己 Immich 服务器上的照片和视频,支持幻灯片播放和屏幕保护程序。","summary":"Immich Android TV 把电视盒子变成一个相框,用来展示你自己在家里搭建的照片服务器上的内容。你用 API 密钥把应用指向你自己的 Immich 服务器(3.0.0 或更新版本),或者在手机上登录,然后浏览相册、人物、随机照片和季节性照片,播放带转场效果的幻灯片,还能把应用设成电视的屏幕保护程序。仓库有 69 个发行版,README 的构建步骤说明需要自己生成安装包:\"Build apk with ./gradlew assembleRelease\"。如果没有自己的 Immich 服务器,README 只提供一个 \"Demo environment\"。","easeRationale":"Android TV 是唯一支持的系统,在那里这个包是 D 级:README 唯一的安装方式是自己编译安装包(\"Build apk with ./gradlew assembleRelease\",要先带子模块克隆仓库,再创建或复制一个 google-services.json),你还得已经有一台运行 \"v3.0.0 and newer\" 的 Immich 服务器,并创建一个拥有列出的 17 项权限的 API 密钥,而把应用设成屏幕保护程序需要开发者模式、USB 调试和从电脑输入 adb 命令。电脑没有评级;没有桌面版本。","costNote":"这个应用免费,采用 GPL-3.0 许可;README 只要求一项可选的捐赠(\"buying me a coffee or a beer\")。运行它所连接的 Immich 服务器则是你自己的开销。","notVerified":["能否在你的电视盒子型号上运行,因为仓库没有列出任何测试过的设备","从别处拿到的安装包是否和这个仓库里的源代码一致,因为 README 没有提供校验和或签名","这个应用向 Firebase 发送了什么","使用手机登录功能时,那个独立项目会经手哪些数据","应用在电视上占用多少空间和内存","幻灯片播放和视频播放在性能较弱的电视盒子上表现如何"]}}},{"id":"R_kgDOODGMVA","slug":"github__github-mcp-server","name":"GitHub MCP Server","owner":"github","fullName":"github/github-mcp-server","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"ai_can_change_things_in_your_account","severity":"attention","evidence":"README：\"Create, update, and manage issues and pull requests\"；工具列表里还包括删除仓库（\"Repository deletion still requires its existing confirmation flow\"）和一个 `governance` 工具组，用于 \"managing rulesets and custom properties at the repository, organization, and enterprise levels\"。README 自己也说要 \"enable the permissions that you feel comfortable granting your AI tools\"，并且有一个 `--read-only` 选项给不想要任何写权限的人用。","source":"ai","title":"AI 可以修改你账户里的东西"},{"code":"assistant_reads_text_written_by_strangers","severity":"attention","evidence":"工具组里包括 `issues`、`discussions`、`notifications` 和 `pull_requests`，README 建议让 \"AI help triage bugs\"。issue 和评论的文本来自互联网上的任何人，而同一个助手又握着你账户的钥匙，所以藏在这些文本里的指令是个真实的风险。","source":"ai","title":"助手会读取陌生人写的文本"},{"code":"some_apps_make_you_paste_the_token_in_a_file","severity":"attention","evidence":"README，\"Handling PATs Securely\"：\"Environment variable support varies by host app and IDE. Some applications (like Windsurf) require hardcoded tokens in config files.\" 给出的建议是运行 \"chmod 600 ~/.your-app/config.json\"，这样 token 就以明文形式留在你的磁盘上。","source":"ai","title":"有些应用要求你把 token 粘贴进文件里"},{"code":"login_credentials_built_into_the_official_image","severity":"info","evidence":"Dockerfile 里的注释：\"OAuth credentials are injected via build secrets so they are not baked into image history; the values are public in practice but kept out of layers\"，README 也说 \"on github.com the official image already includes the app credentials\"。所以这种便捷登录只在 GitHub 自己构建的版本里才有效，自己构建的副本需要用自己的 OAuth 应用。","source":"ai","title":"官方镜像里内置了登录凭据"},{"code":"experimental_tools_behind_a_flag","severity":"info","evidence":"README，Insiders Mode：\"--insiders\" 或 \"GITHUB_INSIDERS=true\"，托管版本对应的网址是 \"https://api.githubcopilot.com/mcp/insiders\"，描述为 \"early access to new features and experimental tools\"。","source":"ai","title":"藏在开关后面的实验性工具"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","go"],"disk":{"minMB":9,"maxMB":27,"basis":"installer"},"apiKeys":["Your GitHub account, either by browser login (OAuth) or a Personal Access Token"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":34366,"downloads":1514172,"downloadsWeek":66979,"downloadsDays":6,"growth":137,"growthDays":6,"history":[{"date":"2026-11-04","stars":34229},{"date":"2026-11-05","stars":34259},{"date":"2026-11-06","stars":34285},{"date":"2026-11-07","stars":34306},{"date":"2026-11-08","stars":34324},{"date":"2026-11-09","stars":34354},{"date":"2026-11-10","stars":34366}],"firstPublishedAt":"2026-11-04T01:36:50.405Z","commitSha":"4bcdcdf8d7beee98c9448e73e257308280924cc4","verifiedAt":"2026-11-04T01:29:11.850Z","lastCommitAt":"2026-11-09T14:28:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant work inside your GitHub account: read code, open and sort issues, review pull requests and watch builds.","summary":"This is GitHub's official MCP server: the bridge that lets AI assistants (GitHub Copilot in VS Code, Claude, Cursor, Codex, Gemini CLI, Zed and others) act on GitHub for you. Once connected, the assistant can read repositories and files, create and update issues and pull requests, follow Actions builds, look at security alerts and read discussions and notifications. GitHub hosts a ready version you connect to with one click, or you run the small program on your own machine. It is for people who already work on GitHub.","easeRationale":"On all three systems the easy path is the same: in VS Code press the install button for the version GitHub hosts and log in through the browser, with nothing to download. Running it yourself is harder: either Docker, or unzip the 9 MB program for your system and paste a small block of settings into your assistant's configuration file.","costNote":"The program is free (MIT) and GitHub's hosted version costs nothing extra. Some of the optional tool groups reach paid GitHub features such as Copilot and the advanced security products.","notVerified":["What the hosted version at api.githubcopilot.com keeps or logs, which is not described in this repository.","Whether the published binaries and the ghcr.io image are built from exactly this code.","How much of your repository content an assistant pulls in during one conversation.","Whether a limited token really blocks every write tool, which can only be seen in use.","Which tools work against GitHub Enterprise Server, where the hosted version is not available."]},"zh-Hans":{"summaryShort":"让 AI 助手在你的 GitHub 账户里工作：读取代码，创建和整理 issue，审查 pull request，还能查看构建情况。","summary":"这是 GitHub 官方的 MCP 服务器：它是让 AI 助手（VS Code 里的 GitHub Copilot、Claude、Cursor、Codex、Gemini CLI、Zed 等）代替你在 GitHub 上操作的桥梁。连接之后，助手可以读取仓库和文件，创建和更新 issue 与 pull request，跟踪 Actions 的构建，查看风险警报，并读取讨论和通知。GitHub 提供一个托管好的版本，一键就能连接，你也可以在自己的机器上运行这个小程序。它适合已经在用 GitHub 的人。","easeRationale":"在三个系统上，最简单的方式都是一样的：在 VS Code 里点安装按钮，连接 GitHub 托管的版本，再通过浏览器登录，什么都不用下载。自己运行就更麻烦一些：要么用 Docker，要么解压适合你系统的 9 MB 程序，再把一小段设置粘贴到你的助手的配置文件里。","costNote":"这个程序是免费的（MIT 许可），GitHub 托管的版本也不额外收费。部分可选的工具组会用到 GitHub 的付费功能，比如 Copilot 和一些进阶的漏洞防护产品。","notVerified":["托管在 api.githubcopilot.com 上的版本会保存或记录什么数据，这个仓库里没有说明。","发布的二进制文件和 ghcr.io 镜像是否确实是用这份代码构建的。","一次对话中，助手会拉取你仓库里多少内容。","受限的 token 是否真的能挡住所有写操作，这一点只能在实际使用中才能看出来。","哪些工具能在 GitHub Enterprise Server 上使用，因为托管版本在那里用不了。"]}}},{"id":"R_kgDOPiFCnQ","slug":"github__spec-kit","name":"Spec Kit","owner":"github","fullName":"github/spec-kit","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"installs_third_party_components","severity":"info","evidence":"README.md 说这个工具可以搜索并安装社区组件：“Spec Kit includes community catalogs... Catalog listings are not audits or endorsements, so review community source code before installing or running it.”像 specify extension add 这样的命令会引入别人写的代码。","source":"ai","title":"安装第三方组件"},{"code":"runs_shell_scripts_in_your_project","severity":"info","evidence":"pyproject.toml 把 scripts/bash、scripts/powershell 和 scripts/python 打包进了安装的软件包里，所以聊天命令会在你项目文件夹内的电脑上运行脚本。","source":"ai","title":"在你的项目里运行 shell 脚本"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":149188,"downloads":946590,"downloadsWeek":0,"downloadsDays":7,"growth":2549,"growthDays":7,"history":[{"date":"2026-10-26","stars":143033},{"date":"2026-10-27","stars":143353},{"date":"2026-10-28","stars":143676},{"date":"2026-10-29","stars":143922},{"date":"2026-10-30","stars":144129},{"date":"2026-10-31","stars":144446},{"date":"2026-11-01","stars":145043},{"date":"2026-11-02","stars":145528},{"date":"2026-11-03","stars":146639},{"date":"2026-11-04","stars":147211},{"date":"2026-11-05","stars":147818},{"date":"2026-11-06","stars":148161},{"date":"2026-11-07","stars":148390},{"date":"2026-11-08","stars":148670},{"date":"2026-11-09","stars":148989},{"date":"2026-11-10","stars":149188}],"firstPublishedAt":"2026-10-27T01:10:12.630Z","commitSha":"ba8363f8ab56b2da456a81299cfdbfee1ef7b871","verifiedAt":"2026-11-03T02:36:00.568Z","lastCommitAt":"2026-11-10T22:06:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Gives your AI coding assistant a step-by-step process: write a spec, plan it, build it, fix bugs, judge ideas.","summary":"Spec Kit is GitHub's free toolkit that adds a structured way of working to AI coding assistants such as GitHub Copilot. You install a small command-line program called Specify, run it inside a project folder, and it drops in templates and chat commands like /speckit-specify, /speckit-plan and /speckit-implement. There are also add-ons for fixing bugs and for deciding whether an idea is worth building. It is for people who already write software with an AI assistant. You need Python 3.11+, the uv installer and a supported AI coding assistant, on Windows, macOS or Linux.","easeRationale":"On Windows, macOS and Linux the steps are the same: install Python 3.11+ and uv, then run two commands (`uv tool install specify-cli` and `specify init`). There is no installer or ready binary, and the commands only do something useful once an AI coding assistant is set up in the same folder.","costNote":"The toolkit itself is free (MIT). You still need an AI coding assistant, and many of those charge a monthly fee.","notVerified":["Whether the chat commands work the same with every supported AI assistant, since each one has its own command syntax","What the AI assistant will actually cost you per project, as the toolkit does not control that","What code is downloaded when you add a community extension, preset or bundle from a catalog","How much disk space a project takes after the templates, workflows and extensions are installed"]},"zh-Hans":{"summaryShort":"给你的 AI 编程助手一套分步流程：写规格、做计划、开发、修复 bug、评估想法。","summary":"Spec Kit 是 GitHub 提供的免费工具包，为 AI 编程助手（例如 GitHub Copilot）带来一套结构化的工作方式。你安装一个叫 Specify 的小型命令行程序，在项目文件夹里运行它，它就会放入模板和聊天命令，例如 /speckit-specify、/speckit-plan 和 /speckit-implement。还有用于修复 bug 和判断一个想法是否值得开发的附加组件。它面向已经使用 AI 助手写软件的人。你需要 Python 3.11+、uv 安装器，以及一个受支持的 AI 编程助手，支持 Windows、macOS 或 Linux。","easeRationale":"在 Windows、macOS 和 Linux 上步骤相同：先安装 Python 3.11+ 和 uv，再运行两条命令（uv tool install specify-cli 和 specify init）。没有安装程序或现成的二进制文件，而且这些命令只有在同一文件夹里配置好 AI 编程助手之后才有用。","costNote":"这个工具包本身是免费的（MIT）。你仍然需要一个 AI 编程助手，而其中很多都要收取月费。","notVerified":["聊天命令在每个受支持的 AI 助手上是否表现一致，因为每个助手都有自己的命令语法","AI 助手每个项目实际会收多少费用，因为这个工具包并不控制这一点","从目录里添加一个社区扩展、预设或合集时，下载的是什么代码","安装了模板、工作流和扩展之后，一个项目会占用多少磁盘空间"]}}},{"id":"R_kgDOJ5QMpw","slug":"gitroomhq__postiz-app","name":"Postiz","owner":"gitroomhq","fullName":"gitroomhq/postiz-app","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖项存在已知漏洞（OSV）：@sentry/nextjs: GHSA-mchm-gjx6-7h55; sweetalert2: GHSA-457r-cqc8-9vj9, GHSA-8jh9-wqpf-q52c, GHSA-pg98-6v7f-2xfv","source":"check","title":"已知漏洞"},{"code":"placeholder_secrets_in_compose_file","severity":"attention","evidence":"docker-compose.yaml 自带 JWT_SECRET: 'random string that is unique to every install - just type random characters here!'，以及固定的数据库密码（POSTGRES_PASSWORD: postiz-password）。如果原样复制到可从互联网访问的服务器上，登录令牌的密钥和数据库密码就会公开。","source":"ai","title":"compose 文件中的示例密钥"},{"code":"open_mcp_client_registration_by_default","severity":"attention","evidence":".env.example 中关于 DCR_VERIFIED_DOMAINS：\"When set (comma separated), POST /oauth/register only accepts redirect_uris whose host matches a listed domain or one of its subdomains; unset or empty means any client can self-register.\" 该变量在示例中被注释掉了。","source":"ai","title":"MCP 客户端默认开放注册"},{"code":"own_platform_apps_and_approval_needed","severity":"info","evidence":"README 对比表，自托管一栏：\"You create your own developer apps on each platform and go through their approval (Meta, YouTube, TikTok can take weeks)\"。数据分析功能同样 \"requires your own app credentials with analytics scopes\"。","source":"ai","title":"需要自建平台应用并等待审批"},{"code":"heavy_server_stack","severity":"info","evidence":"docker-compose.yaml 会启动九个容器：Postiz、postgres:17-alpine、redis:7.2、Sentry Spotlight、elasticsearch:7.17.27、第二个 postgres:16、temporal auto-setup、管理工具和 Temporal UI。普通笔记本电脑通常不会在后台运行这样一整套服务。","source":"ai","title":"服务器组件繁重"},{"code":"network_copyleft_license","severity":"info","evidence":"LICENSE 为 AGPL-3.0：\"public use of a modified version, on a publicly accessible server, gives the public access to the source code of the modified version\"。如果你修改了 Postiz 并让其他人通过网络使用，就必须公开你的修改。","source":"ai","title":"网络型 copyleft 许可证"}],"platforms":["windows","mac","linux","web"],"tools":["docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Your own developer app on each social network (X, LinkedIn, Facebook, Threads, YouTube, TikTok, Reddit, Pinterest, Discord, Slack, Mastodon and others)","OpenAI","Cloudflare R2 storage","Resend (activation e-mails)"],"cost":{"model":"freemium"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","automation","productivity"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"later","name":"Later","confirmed":true}],"declaredAlternatives":[{"target":"manychat","stance":"claims","evidence":"alt=\"ChatbotX\" width=\"500\"/> | The ManyChat alternative that you can self-host, white-label, and resell to your clients"}],"stars":41282,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":469,"growthDays":7,"history":[{"date":"2026-10-28","stars":39967},{"date":"2026-10-29","stars":40114},{"date":"2026-10-30","stars":40238},{"date":"2026-10-31","stars":40377},{"date":"2026-11-01","stars":40603},{"date":"2026-11-02","stars":40727},{"date":"2026-11-03","stars":40813},{"date":"2026-11-04","stars":40899},{"date":"2026-11-05","stars":40966},{"date":"2026-11-06","stars":41027},{"date":"2026-11-07","stars":41081},{"date":"2026-11-08","stars":41139},{"date":"2026-11-09","stars":41202},{"date":"2026-11-10","stars":41282}],"firstPublishedAt":"2026-10-28T01:50:05.578Z","commitSha":"a5affe62c74e39c1e7ac960d50d8418b807d532c","verifiedAt":"2026-10-28T01:38:38.681Z","lastCommitAt":"2026-11-10T08:56:03Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Write, schedule and publish posts to many social networks from one calendar running on your own server.","summary":"A social media management app you host yourself: one calendar to write, schedule and cross-post to X, Instagram, YouTube, TikTok, LinkedIn, Reddit, Facebook, Threads, Pinterest, Discord, Slack, Mastodon and Bluesky, with analytics, team members, a public API and AI writing help. You use it in a browser. Running it means Docker with PostgreSQL, Redis and a Temporal stack, and creating your own developer app on each network, which the README says Meta, YouTube and TikTok can take weeks to approve. The same product is also sold as a paid hosted service.","easeRationale":"The 204 releases carry no installer, so the published way in is the docker-compose file: Linux is the natural home for it and sets the grade, since it starts Postiz plus PostgreSQL, Redis, Temporal and Elasticsearch, and you then fill in dozens of settings before opening http://localhost:4007 in a browser. Windows and macOS need Docker Desktop for the same stack; the browser grade matches because the page only exists once that server is running.","costNote":"Self-hosting is free under AGPL-3.0 (\"Free forever\"; you pay for your own server). Postiz Cloud is a subscription per plan with a 7-day trial.","notVerified":["Whether each network still accepts your own developer app, and how long the approval takes","How much memory and disk the full container stack really needs on your server","What the AI writing, image, video and clipping features cost once you plug in your own provider keys","Whether the published ghcr.io image matches the code in this repository","Whether posting and analytics keep working after each platform changes its API"]},"zh-Hans":{"summaryShort":"在自己服务器上运行的同一个日历里撰写、安排并发布帖子到多个社交网络。","summary":"一个你自己搭建托管的社交媒体管理应用：用一个日历撰写、安排并同时发布到 X、Instagram、YouTube、TikTok、LinkedIn、Reddit、Facebook、Threads、Pinterest、Discord、Slack、Mastodon 和 Bluesky，还提供数据分析、团队成员协作、公开 API 和 AI 写作辅助。你通过浏览器使用它。运行它需要用 Docker 搭配 PostgreSQL、Redis 和 Temporal 技术栈，并在每个平台上创建自己的开发者应用，README 提到 Meta、YouTube 和 TikTok 的审核可能要等上几周。同一款产品也作为付费托管服务出售。","easeRationale":"204 个发布版本都没有安装程序，官方给出的方式是 docker-compose 文件：Linux 是它的天然环境，也决定了评分，因为它要启动 Postiz 以及 PostgreSQL、Redis、Temporal 和 Elasticsearch，之后还要填写几十项设置，才能在浏览器里打开 http://localhost:4007。Windows 和 macOS 需要用 Docker Desktop 运行同一套技术栈；浏览器端的评分与之一致，因为只有服务器运行起来后页面才会存在。","costNote":"自己搭建托管在 AGPL-3.0 协议下是免费的（“永久免费”；你只需支付自己服务器的费用）。Postiz Cloud 则是按套餐订阅收费，提供 7 天试用。","notVerified":["每个平台是否仍然接受你自己的开发者应用，以及审核需要多长时间","完整的容器技术栈在你的服务器上实际需要多少内存和磁盘空间","一旦接入自己的服务商密钥，AI 写作、图片、视频和剪辑功能的实际费用是多少","发布在 ghcr.io 上的镜像是否与这个仓库中的代码一致","在各个平台修改其 API 之后，发布和数据分析功能是否还能正常工作"]}}},{"id":"MDEwOlJlcG9zaXRvcnk4MDI3NjM5MQ==","slug":"gitsquared__edex-ui","name":"eDEX-UI","owner":"GitSquared","fullName":"GitSquared/edex-ui","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"B"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发行在2021-10-18（1842天前）","source":"check","title":"版本更新停滞"},{"code":"archived","severity":"info","evidence":"该仓库已归档（只读，不再维护）","source":"check","title":"已归档"},{"code":"unsigned_release_binaries","severity":"attention","evidence":"README.md：\"Public release binaries are unsigned\"；package.json的构建配置把\"forceCodeSigning\"设为false。Windows和macOS在允许程序运行前都会显示系统警告。","source":"ai","title":"发行版二进制文件未签名"},{"code":"outdated_bundled_runtime","severity":"attention","evidence":"package.json把\"electron\"锁定在\"^12.1.0\"，这是该应用所基于的浏览器引擎2021年的版本。由于仓库已归档，不会再有更新。","source":"ai","title":"内置运行环境已过时"},{"code":"reads_system_and_network_data","severity":"info","evidence":"README.md功能列表：\"Real-time system (CPU, RAM, swap, processes) and network (GeoIP, active connections, transfer rates) monitoring.\" GeoIP查询意味着程序会就你的连接联系一个在线服务。","source":"ai","title":"读取系统和网络数据"},{"code":"windows_feature_limited","severity":"info","evidence":"README.md问答：在Windows上文件浏览器无法跟随终端所在文件夹，会\"reverts back to a 'detached' mode\"。","source":"ai","title":"Windows上功能受限"},{"code":"heavy_on_screen_resources","severity":"info","evidence":"Windows上的安装程序有131 MB，应用会全屏绘制动态图表和一个3D地球（README中致谢了\"ENCOM Globe\"），所以它不是一个轻量终端。","source":"ai","title":"屏幕资源消耗较大"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":131,"maxMB":393,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system"],"alternativeTo":[{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":45039,"downloads":1771929,"downloadsWeek":3479,"downloadsDays":6,"growth":0,"growthDays":7,"history":[{"date":"2026-11-04","stars":45047},{"date":"2026-11-05","stars":45045},{"date":"2026-11-06","stars":45045},{"date":"2026-11-07","stars":45044},{"date":"2026-11-08","stars":45044},{"date":"2026-11-09","stars":45041},{"date":"2026-11-10","stars":45039}],"firstPublishedAt":"2026-11-04T01:36:50.412Z","commitSha":"04a00c4079908788b371c6ecdefff96d0d9950f8","verifiedAt":"2026-11-04T01:20:31.399Z","lastCommitAt":"2021-10-19T22:38:10Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"eDEX-UI-Windows-x64.exe","sizeBytes":70612835,"url":"https://github.com/GitSquared/edex-ui/releases/download/v2.2.8/eDEX-UI-Windows-x64.exe","releaseUrl":"https://github.com/GitSquared/edex-ui/releases/tag/v2.2.8"},{"os":"mac","file":"eDEX-UI-macOS-x64.dmg","sizeBytes":94874695,"url":"https://github.com/GitSquared/edex-ui/releases/download/v2.2.8/eDEX-UI-macOS-x64.dmg","releaseUrl":"https://github.com/GitSquared/edex-ui/releases/tag/v2.2.8"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A full-screen terminal with a sci-fi movie look, plus live charts of your CPU, memory and network.","summary":"eDEX-UI turns a terminal window into a full-screen science-fiction computer interface: the command line sits in the middle, with live CPU, memory, process and network panels around it, a file list that follows the folder you are in, an on-screen keyboard for touchscreens and optional sound effects. It is for people who like the look and already use a terminal, and for show machines and touch displays. There are ready installers for Windows, macOS and Linux, and nothing else to install. The author archived the project in October 2021.","easeRationale":"On Windows the release has a ready .exe installer (131 MB) and on macOS a .dmg, so you download and open it. On Linux the download is an AppImage and the README says you must run `chmod +x` on the file before it will start, which is one extra step.","notVerified":["Whether the 2021 binaries still start on current Windows, macOS and Linux versions","Which online service the GeoIP network panel calls and what it sends","Real disk and memory use once the app is running full screen","Whether the on-screen keyboard and touch support work on a given touchscreen"]},"zh-Hans":{"summaryShort":"一个带科幻电影风格的全屏终端，还能实时显示CPU、内存和网络的图表。","summary":"eDEX-UI 把一个终端窗口变成全屏的科幻电脑界面：命令行位于中间，周围是CPU、内存、进程和网络的实时面板，还有跟随当前所在文件夹变化的文件列表、给触屏用的屏幕键盘，以及可选的音效。它适合喜欢这种风格、已经在用终端的人，也适合展示用的机器和触摸屏。Windows、macOS和Linux都有现成的安装程序，不用再装别的东西。作者在2021年10月把这个项目归档了。","easeRationale":"在Windows上，发行版自带现成的.exe安装程序（131 MB），macOS上是.dmg，下载打开就行。在Linux上，下载的是AppImage，README说启动前必须对文件执行`chmod +x`，这算是多一步。","notVerified":["2021年的二进制文件是否还能在当前版本的Windows、macOS和Linux上启动","GeoIP网络面板调用的是哪个在线服务，它发送了什么内容","应用全屏运行后的实际磁盘和内存占用","屏幕键盘和触控支持是否能在某块具体的触屏上正常工作"]}}},{"id":"R_kgDON4uODQ","slug":"glips__figma-context-mcp","name":"Framelink Figma MCP","owner":"GLips","fullName":"GLips/Figma-Context-MCP","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):@modelcontextprotocol/sdk: GHSA-22jm-h49p-29qw, GHSA-46jr-4wx8-v453, GHSA-6prh-2h8m-c8cw","source":"check","title":"已知漏洞"},{"code":"usage_telemetry","severity":"attention","evidence":"package.json 把分析库 \"posthog-node\": \"^5.29.2\" 列为运行时依赖,README 中没有提到会收集使用数据,也没有说明如何关闭。","source":"ai","title":"使用数据统计"},{"code":"reads_your_design_files","severity":"info","evidence":".env.example 要求提供 FIGMA_API_KEY,这是一个 Figma 个人访问令牌;有了它,服务器就能读取该令牌能访问到的 Figma 文件,并把其中的内容发送给你连接的 AI 助手。","source":"ai","title":"读取你的设计文件"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Figma"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","design_cad"],"alternativeTo":[{"slug":"figma","name":"Figma","confirmed":true}],"declaredAlternatives":[{"target":"figma","stance":"claims","evidence":"When Cursor has access to Figma design data, it's way better at one-shotting designs accurately than alternative approaches like pasting screenshots"},{"target":"cursor","stance":"claims","evidence":"When Cursor has access to Figma design data, it's way better at one-shotting designs accurately than alternative approaches like pasting screenshots"}],"stars":16187,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":30,"growthDays":6,"history":[{"date":"2026-11-05","stars":16169},{"date":"2026-11-06","stars":16174},{"date":"2026-11-07","stars":16174},{"date":"2026-11-08","stars":16177},{"date":"2026-11-09","stars":16180},{"date":"2026-11-10","stars":16187}],"firstPublishedAt":"2026-11-04T22:47:04.165Z","commitSha":"c083d65c7e002923e7cb98f4e3bdafb105e90f6d","verifiedAt":"2026-11-04T22:40:30.348Z","lastCommitAt":"2026-06-24T16:12:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI coding assistant read your Figma screens so it can write the code for that design.","summary":"An MCP server that connects Cursor and other assistants that accept MCP servers to your Figma files. You paste a Figma link in the assistant's chat and it pulls the layout, sizes and colours of that screen and writes code from them. It is for people who already build software with an AI assistant. You need Node.js 20.20.0 or newer and a personal Figma access token; the server itself is free and MIT licensed. It does not replace Figma, it reads designs out of it.","easeRationale":"There is no installer: on Windows, Mac and Linux alike you install Node.js (package.json asks for \">=20.20.0\"), create a Figma access token and paste a small JSON block into your editor's MCP settings file, so all three platforms land at C. The 29 releases ship no binaries, only the npm package.","costNote":"Free and MIT licensed. You need a Figma account to create the access token the server asks for (.env.example: FIGMA_API_KEY).","notVerified":["Whether the PostHog telemetry can be switched off and what it sends","How much of your Figma file content ends up in the AI assistant's context (and with the model provider)","Whether the three known vulnerabilities in the pinned @modelcontextprotocol/sdk 1.29.0 affect normal use","Whether it still works with current editor versions: the last commit was about 4 months before this check"]},"zh-Hans":{"summaryShort":"让 AI 编程助手读取你的 Figma 界面,从而直接写出对应设计的代码。","summary":"一个 MCP 服务器,把 Cursor 和其他支持 MCP 服务器的助手连接到你的 Figma 文件。你在助手的聊天框里粘贴一个 Figma 链接,它就会读取那个界面的布局、尺寸和颜色,并据此写出代码。它是给已经在用 AI 助手开发软件的人使用的。你需要 Node.js 20.20.0 或更新版本,以及一个 Figma 个人访问令牌;服务器本身是免费的,采用 MIT 许可。它不会取代 Figma,只是从里面读取设计内容。","easeRationale":"没有安装程序:在 Windows、Mac 和 Linux 上都是一样的流程,先安装 Node.js(package.json 要求 \">=20.20.0\"),创建一个 Figma 访问令牌,再把一小段 JSON 粘贴到你编辑器的 MCP 设置文件里,所以三个平台都评为 C 级。29 个发行版都没有附带可执行文件,只有 npm 包。","costNote":"免费,采用 MIT 许可。你需要一个 Figma 账号来创建服务器所需的访问令牌(.env.example:FIGMA_API_KEY)。","notVerified":["PostHog 的使用数据统计是否可以关闭,以及它具体发送了哪些内容","你的 Figma 文件内容有多少会进入 AI 助手的上下文(以及传给模型提供方)","固定版本的 @modelcontextprotocol/sdk 1.29.0 中的三个已知漏洞是否会影响正常使用","它是否还能配合当前版本的编辑器正常工作:最后一次提交大约是在本次检查前 4 个月"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNTgwMDM2","slug":"gnachman__iterm2","name":"iTerm2","owner":"gnachman","fullName":"gnachman/iTerm2","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"mac":"A"},"risk":"attention","signals":[{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.8/10","source":"check","title":"scorecard评分偏低"},{"code":"download_not_hosted_on_github","severity":"info","evidence":"该仓库没有GitHub发行版;README把你指向https://iterm2.com/downloads.html和https://iterm2.com/nightly/latest来获取应用程序。你在这里能看到的代码,和你实际安装的文件,来自不同的地方。","source":"ai","title":"下载包没有放在github上"},{"code":"ai_chat_reads_terminal_and_runs_commands","severity":"attention","evidence":"README.md的功能列表写道:\"AI Chat - Built-in LLM chat window that can optionally interact with terminal contents. Link sessions to get context-aware help, run commands on your behalf, or explain output with annotations.\" 你关联的任何内容都会发送给外部AI服务商,而终端窗口里常常存有密钥和密码。","source":"ai","title":"ai聊天功能能读取终端并执行命令"},{"code":"license_statement_inconsistent","severity":"info","evidence":"GitHub显示的是GPL-2.0,LICENSE文件也是GPL第2版的文本,而COPYING却写着\"iTerm2 is licensed under GPL v2 or later... Because of Apache 2.0-licensed dependencies, which are only compatible with GPL v3, the license is effectively GPL v3\",README则写着\"iTerm2 is distributed under the GPLv3 license\"。","source":"ai","title":"许可证说明前后不一致"},{"code":"issues_tracked_outside_github","severity":"info","evidence":"README.md:\"We use GitLab for issues because it provides better support for attachments\",漏洞提交在iterm2.com/bugs,跟踪系统在gitlab.com/gnachman/iterm2/issues。GitHub上72个未关闭的issue并不是真正的待办队列。","source":"ai","title":"问题反馈在github之外处理"},{"code":"build_from_source_installs_a_lot","severity":"info","evidence":"README.md:\"`make setup` will install Homebrew, Xcode, Rust, and all other dependencies\",其中包括通过\"via the official `curl | sh` installer\"安装`rustup`,以及用`sudo`安装SF Symbols包,还有`make dangerous-setup`用来\"skip all confirmation prompts\"。这只影响自己编译项目的人。","source":"ai","title":"从源码编译会安装很多东西"},{"code":"powerful_features_can_act_on_their_own","severity":"info","evidence":"README.md把Triggers描述为\"Regex patterns that fire actions when matched: highlight text, run commands, send notifications, open password manager, set marks, or invoke Python scripts\",还有支持\"click-to-download files via SCP\"的shell集成。这些功能很有用,但也会在匹配到文本后自行采取行动,而不会先询问你。","source":"ai","title":"强大的功能可以自行触发动作"}],"platforms":["mac"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["An AI provider for the built-in chat (your own key)"],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system","development","productivity"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":18271,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":15,"growthDays":6,"history":[{"date":"2026-11-06","stars":18260},{"date":"2026-11-07","stars":18260},{"date":"2026-11-08","stars":18265},{"date":"2026-11-09","stars":18264},{"date":"2026-11-10","stars":18271}],"firstPublishedAt":"2026-11-05T22:43:48.419Z","commitSha":"f365dd26aa136568cb9f0a4211b0a6d9a170a17b","verifiedAt":"2026-11-05T22:28:04.909Z","lastCommitAt":"2026-11-10T03:33:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A replacement for the Mac Terminal app with split panes, saved sessions, images in the window and a built-in AI chat.","summary":"A terminal for Mac, downloaded ready to run from the project's own site. It replaces Apple's Terminal and adds things the original lacks: sessions that survive a crash or an update, a window that drops down on a hotkey, images shown inside the terminal, search across all tabs, a built-in web browser, automation in Python and an optional AI chat that can read what is on screen and run commands for you. Free, with donations, and only for macOS. It is a tool for people who use the command line.","easeRationale":"Mac is an A: the README points to iterm2.com/downloads for a ready application, and a nightly build \"for the bleeding edge without building\". The project is macOS only — the README's badge says \"platform-macOS\" — and building it yourself is a different, much harder path that installs Homebrew, Xcode and Rust.","costNote":"No price; the README asks only for stars and links to a donation page. The optional AI chat needs your own paid key from an AI provider.","notVerified":["Which Mac versions the current download supports","Whether the file on iterm2.com matches the code in this repository","What the AI chat sends to the provider and when, and what it costs","Disk space and memory used in normal daily work","Whether building from source works on a given Xcode version"]},"zh-Hans":{"summaryShort":"替代Mac自带Terminal的应用,支持分屏、会话保存、窗口内显示图片,还内置了AI聊天功能。","summary":"一款Mac终端工具,可以直接从项目自己的网站下载即可使用。它取代了Apple的Terminal,并增加了原版没有的功能:崩溃或更新后仍能恢复的会话、用快捷键呼出的下拉窗口、在终端里直接显示图片、跨所有标签页搜索、内置的网页浏览器、Python自动化,以及一个可选的AI聊天功能,可以读取屏幕内容并代你执行命令。免费使用,接受捐赠,仅支持macOS。它是为使用命令行的人准备的工具。","easeRationale":"Mac的难度评级是A:README指向iterm2.com/downloads获取现成的应用程序,还提供\"for the bleeding edge without building\"的每日构建版。该项目仅支持macOS——README的徽章写着\"platform-macOS\"——自己编译是另一条完全不同、也更难走的路,需要安装Homebrew、Xcode和Rust。","costNote":"没有价格;README只要求点星标,并链接到一个捐赠页面。可选的AI聊天功能需要你自己向AI服务商付费获取密钥。","notVerified":["当前下载版本支持哪些Mac版本","iterm2.com上的文件是否与这个仓库里的代码一致","AI聊天功能把什么内容发给服务商、什么时候发,以及要花多少钱","日常使用中实际占用的磁盘空间和内存","在特定版本的Xcode下从源码编译是否能成功"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0NTM3NTA4","slug":"gnome__gimp","name":"GIMP","owner":"GNOME","fullName":"GNOME/gimp","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"license_in_files_gpl3","severity":"info","evidence":"许可证检查结果为 NOASSERTION，但 COPYING 中包含 GNU 通用公共许可证第 3 版，LICENSE 则说明 libgimp 库使用 GNU 宽通用公共许可证（LGPL）。","source":"ai","title":"文件中标明 GPL3 许可证"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":182,"maxMB":545,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"winget"},"categories":["image","design_cad"],"alternativeTo":[{"slug":"photoshop","name":"Photoshop","confirmed":true},{"slug":"pixelmator-pro","name":"Pixelmator Pro","confirmed":true}],"declaredAlternatives":[],"stars":6577,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":14,"growthDays":7,"history":[{"date":"2026-10-25","stars":6545},{"date":"2026-10-26","stars":6546},{"date":"2026-10-27","stars":6550},{"date":"2026-10-28","stars":6554},{"date":"2026-10-29","stars":6555},{"date":"2026-10-30","stars":6557},{"date":"2026-10-31","stars":6559},{"date":"2026-11-01","stars":6558},{"date":"2026-11-02","stars":6561},{"date":"2026-11-03","stars":6563},{"date":"2026-11-04","stars":6566},{"date":"2026-11-05","stars":6568},{"date":"2026-11-06","stars":6567},{"date":"2026-11-07","stars":6568},{"date":"2026-11-08","stars":6572},{"date":"2026-11-09","stars":6573},{"date":"2026-11-10","stars":6577}],"firstPublishedAt":"2026-10-25T23:19:03.806Z","commitSha":"0f5cab52f31bf467bc0d6e417aeb3d52b83c6c5c","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T22:07:23Z","isMirror":true,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Edit photos, retouch pictures and create digital art with a free, full-featured image editor.","summary":"GIMP (GNU Image Manipulation Program) is a long-running free image editor. You can retouch and crop photos, work with layers and masks, paint with brushes, and open and save many file types, including PDF and SVG. It is for anyone who edits images, from hobbyists to designers. On Windows you download the installer from the official site and run it. This GitHub page is a read-only copy of the code; the main project lives on GNOME's GitLab, and the code here is the unstable 3.3 development branch.","easeRationale":"An official Windows installer exists (gimp-3.2.6-setup.exe, about 182 MB, listed in winget as GIMP.GIMP.3) that you download and run.","notVerified":["Whether the Windows installer (version 3.2.6) was built from this exact code; this mirror holds the 3.3 development branch","How well it performs with very large images on your computer","Whether add-on plug-ins written for the older GIMP 2 work (the INSTALL notes say they do not)"]},"zh-Hans":{"summaryShort":"用一款免费且功能齐全的图像编辑器编辑照片、修饰图片并创作数字艺术。","summary":"GIMP（GNU Image Manipulation Program）是一款历史悠久的免费图像编辑器。你可以修饰和裁剪照片，使用图层和蒙版，用画笔绘画，并打开和保存多种文件类型，包括 PDF 和 SVG。它适合所有需要编辑图像的人，从业余爱好者到设计师。在 Windows 上，你从官方网站下载安装程序并运行即可。这个 GitHub 页面只是代码的只读副本；主项目位于 GNOME 的 GitLab，这里的代码是不稳定的 3.3 开发分支。","easeRationale":"有官方的 Windows 安装程序（gimp-3.2.6-setup.exe，约 182 MB，在 winget 中列为 GIMP.GIMP.3），下载后运行即可。","notVerified":["Windows 安装程序（3.2.6 版）是否由这份代码构建而成；这个镜像保存的是 3.3 开发分支","在你的电脑上处理超大图像时表现如何","为旧版 GIMP 2 编写的附加插件是否可用（INSTALL 说明称不可用）"]}}},{"id":"R_kgDOLHrK7Q","slug":"goenning__google-indexing-script","name":"Google Indexing Script","owner":"goenning","fullName":"goenning/google-indexing-script","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2024-07-08(849 天前)","source":"check","title":"版本更新停滞"},{"code":"works_only_on_some_pages","severity":"attention","evidence":"README:\"This script uses Google Indexing API and it only works on pages with either `JobPosting` or `BroadcastEvent` structured data.\" 大多数网站都没有这两种数据,所以这个工具对它们可能没有任何作用。","source":"ai","title":"只对部分页面有效"},{"code":"powerful_credential_on_your_machine","severity":"info","evidence":"README 要求你把 Google Cloud 服务账号密钥保存为 service_account.json 放在 ~/.gis 中,这个账号拥有\"the `Owner` permission on your sites\";文档列出的一种方式是把私钥作为命令行参数传入,README 自己也把它标为\"(not recommended)\"。","source":"ai","title":"机器上存有高权限凭证"},{"code":"pinned_old_google_library","severity":"info","evidence":"package.json 把\"googleapis\"固定为\"131.0.0\",最后一次提交是在 2024-10-10,说明这个 Google 客户端库已经两年多没有更新了。","source":"ai","title":"固定的 google 库版本老旧"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Google Cloud service account (Indexing API + Search Console API)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":7729,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":6,"growthDays":6,"history":[{"date":"2026-11-05","stars":7723},{"date":"2026-11-06","stars":7726},{"date":"2026-11-07","stars":7723},{"date":"2026-11-08","stars":7724},{"date":"2026-11-09","stars":7726},{"date":"2026-11-10","stars":7729}],"firstPublishedAt":"2026-11-04T22:32:54.953Z","commitSha":"c52c2f8c67ad99594c78021ff4b203decdc74976","verifiedAt":"2026-11-04T22:15:01.386Z","lastCommitAt":"2024-10-10T18:39:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Command-line tool that asks Google to look at the pages of your site, using Google's own indexing API.","summary":"A small command-line tool for site owners. You give it your domain, it reads the sitemaps you already submitted to Google Search Console and asks Google's Indexing API to crawl the pages that are not indexed yet. It is for people comfortable with a terminal: you install Node.js, install the tool with npm, and set up a Google Cloud project with a service account before the first run. Free and MIT licensed. The README warns that the API only works on pages with JobPosting or BroadcastEvent data, and that being indexed is not the same as ranking.","easeRationale":"Same work on Windows, Mac and Linux, and it is more than a few commands: install Node.js 20 or newer, `npm i -g google-indexing-script`, then create a Google Cloud project, enable two APIs, create a service account, give it Owner permission in Search Console and save its JSON key in a ~/.gis folder.","costNote":"The tool is free; Google's Indexing API is free but has daily quotas. The project is sponsored by the paid service SEO Gets.","notVerified":["Whether pages really get indexed, and how fast","Whether the pinned googleapis 131.0.0 still works against Google's current APIs","The real daily quota on your own Google Cloud account","Whether using the Indexing API for pages without JobPosting or BroadcastEvent data breaks Google's own rules for that API"]},"zh-Hans":{"summaryShort":"命令行工具,使用 Google 自己的索引 API,请 Google 查看你网站的页面。","summary":"一个供网站主使用的小型命令行工具。你提供自己的域名,它会读取你已经提交给 Google Search Console 的站点地图,然后请求 Google 索引 API 抓取尚未被收录的页面。它适合习惯使用终端的人:在第一次运行前,你需要安装 Node.js,用 npm 安装这个工具,并配置一个带有服务账号的 Google Cloud 项目。免费,采用 MIT 许可证。README 提醒,该 API 只对带有 JobPosting 或 BroadcastEvent 结构化数据的页面有效,而且被收录并不等于排名靠前。","easeRationale":"在 Windows、Mac 和 Linux 上工作量相同,而且不止是几条命令:安装 Node.js 20 或更新版本,运行 `npm i -g google-indexing-script`,然后创建一个 Google Cloud 项目,启用两个 API,创建一个服务账号,给它 Search Console 的 Owner 权限,并把它的 JSON 密钥保存到 ~/.gis 文件夹中。","costNote":"这个工具是免费的;Google 索引 API 也是免费的,但有每日配额。该项目由付费服务 SEO Gets 赞助。","notVerified":["页面是否真的会被收录,速度有多快","固定版本的 googleapis 131.0.0 是否还能配合 Google 现在的 API 正常工作","你自己 Google Cloud 账号的真实每日配额是多少","对没有 JobPosting 或 BroadcastEvent 数据的页面使用索引 API,是否违反 Google 自己对该 API 的规定"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTE4MDY4Nw==","slug":"gohugoio__hugo","name":"Hugo","owner":"gohugoio","fullName":"gohugoio/hugo","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"theme_needs_extra_tools","severity":"info","evidence":"README.md 列出了针对 Sass、PostCSS、Tailwind CSS 和 JavaScript 的资源处理流程（asset pipelines）；项目的 Dockerfile 会安装 \"git\"、\"nodejs\"、\"npm\" 以及单独的 \"dart-sass\" 程序，所以有些主题在你的电脑上也需要这些额外工具。","source":"ai","title":"主题需要额外工具"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":["docker","go"],"disk":{"minMB":32,"maxMB":95,"basis":"installer"},"apiKeys":["Google Cloud Storage, AWS S3 or Azure Storage (only for the deploy edition's direct upload)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"squarespace","name":"Squarespace","confirmed":true},{"slug":"wix","name":"Wix","confirmed":true}],"declaredAlternatives":[],"stars":90437,"downloads":103982297,"downloadsWeek":820352,"downloadsDays":7,"growth":85,"growthDays":7,"history":[{"date":"2026-10-25","stars":90234},{"date":"2026-10-26","stars":90248},{"date":"2026-10-27","stars":90265},{"date":"2026-10-28","stars":90272},{"date":"2026-10-29","stars":90272},{"date":"2026-10-30","stars":90291},{"date":"2026-10-31","stars":90307},{"date":"2026-11-01","stars":90318},{"date":"2026-11-02","stars":90328},{"date":"2026-11-03","stars":90352},{"date":"2026-11-04","stars":90367},{"date":"2026-11-05","stars":90380},{"date":"2026-11-06","stars":90394},{"date":"2026-11-07","stars":90408},{"date":"2026-11-08","stars":90418},{"date":"2026-11-09","stars":90425},{"date":"2026-11-10","stars":90437}],"firstPublishedAt":"2026-10-25T23:19:04.670Z","commitSha":"a1668da327a05646e98af10b622b75bc9355f287","verifiedAt":"2026-10-30T01:43:26.247Z","lastCommitAt":"2026-11-10T21:26:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turn text files and a theme into a complete website you can host almost anywhere.","summary":"Hugo is a static site generator: you write pages as text files, pick a theme, and Hugo builds a finished website made of plain files you can upload to any web host. People use it for blogs, documentation, portfolios and company sites. It has a built-in preview server that shows your changes as you edit. It runs from the command line, so it suits people who are comfortable typing commands and editing configuration files. Ready downloads exist for Windows, Mac and Linux, in a standard and a deploy edition.","easeRationale":"The same on all three: there are ready downloads (hugo_0.169.0_windows-amd64.zip, a .pkg for macOS, .deb and .tar.gz for Linux), but Hugo has no window of its own — you run it with commands in a terminal, and themes are usually added with Git.","costNote":"Apache-2.0 licence. Hosting the finished website is up to you and may cost money.","notVerified":["Which extra tools your chosen theme needs, such as Git, Dart Sass or Node.","What hosting the finished site will cost you.","How well a given third-party theme is maintained.","Whether the standard edition is enough or you need the deploy edition for your host."]},"zh-Hans":{"summaryShort":"把文本文件和一个主题,变成一个完整的网站,几乎可以托管在任何地方。","summary":"Hugo 是一个静态网站生成器:你把页面写成文本文件,选一个主题,Hugo 就会构建出一个由普通文件组成的成品网站,可以上传到任何网站托管服务。常用于博客、文档、作品集和公司网站。它自带一个预览服务器,边编辑边显示改动效果。它要在命令行中运行,所以适合习惯输入命令和编辑配置文件的人。Windows、Mac 和 Linux 都有现成的下载包,分为标准版和部署版。","easeRationale":"三个系统上都一样:都有现成的下载包(Windows 版 hugo_0.169.0_windows-amd64.zip,macOS 版 .pkg,Linux 版 .deb 和 .tar.gz),但 Hugo 没有自己的窗口界面——要在终端里用命令运行,主题通常用 Git 添加。","costNote":"Apache-2.0 许可证。托管完成后的网站由你自行负责,可能需要付费。","notVerified":["你选的主题需要哪些额外工具,比如 Git、Dart Sass 或 Node。","托管完成后的网站会花费多少钱。","某个第三方主题的维护情况如何。","标准版是否够用,还是你的托管服务需要部署版。"]}}},{"id":"R_kgDOQ7Opvg","slug":"google-labs-code__stitch-skills","name":"Stitch Skills","owner":"google-labs-code","fullName":"google-labs-code/stitch-skills","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"not_an_officially_supported_google_product","severity":"info","evidence":"虽然它位于\"google-labs-code\"组织下,但README结尾写着\"This is not an officially supported Google product. This project is not eligible for the Google Open Source Software Vulnerability Rewards Program.\"没有任何支持承诺,而且在本次检查前82天就是最后一次提交。","source":"ai","title":"不是谷歌官方支持的产品"},{"code":"your_code_and_designs_go_to_a_cloud_service","severity":"attention","evidence":"这些技能的设计目的就是把你的项目发送给Stitch:\"Convert frontend code (React, Vue, etc.) to a Stitch Design via HTML extraction + design system + upload\"、\"Extract a comprehensive DESIGN.md directly from frontend source code\"以及\"Upload local assets (images, mockups, HTML) to a Stitch project\"。你的源代码、界面和设计系统会离开你的电脑,进入谷歌的服务。","source":"ai","title":"你的代码和设计会发送到云服务"},{"code":"skills_run_scripts_through_your_assistant","severity":"info","evidence":"每个技能都在其SKILL.md旁边附带\"scripts/ — Executable enforcers (Validation & Networking)\",安装过程会把它们加入你的助手,助手会以它在你项目中拥有的权限运行它们(例如,\"stitch-loop\"会\"with automated validation\"构建一整个多页网站)。","source":"ai","title":"技能通过你的助手运行脚本"},{"code":"skills_depend_on_each_other","severity":"info","evidence":"README警告说:\"Stitch Design Skills often have inter-dependencies. If you choose to install skills selectively, ensure you include all required dependencies\",所以单独手动安装某个技能可能会让它无法正常使用。","source":"ai","title":"技能之间互相依赖"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Google Stitch (MCP server credentials and environment variables from Stitch's setup guide)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","design_cad","development"],"alternativeTo":[{"slug":"figma","name":"Figma","confirmed":true},{"slug":"v0-by-vercel","name":"v0 by Vercel","confirmed":true},{"slug":"lovable","name":"Lovable","confirmed":true}],"declaredAlternatives":[],"stars":8724,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":46,"growthDays":6,"history":[{"date":"2026-11-08","stars":8700},{"date":"2026-11-09","stars":8712},{"date":"2026-11-10","stars":8724}],"firstPublishedAt":"2026-11-07T22:56:16.783Z","commitSha":"0337446dadde6f8c94210444e2aa9d546126480f","verifiedAt":"2026-11-07T22:44:52.444Z","lastCommitAt":"2026-08-17T20:20:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Add-ons for AI coding assistants that let you design app screens with Google Stitch and turn them into React code.","summary":"A set of skills and plugins you install into an AI coding assistant (Claude Code, Codex, Cursor, Gemini CLI, Antigravity or OpenCode) so you can work with Google Stitch by typing requests: generate app screens from a description, create variants, turn existing frontend code into a Stitch design, pull a design system into a DESIGN.md file, and convert Stitch screens into React, React Native or a Vite dashboard. It is for people who already write software: you need one of those assistants and the Stitch MCP server set up with your own credentials.","easeRationale":"The work is the same on Windows, macOS and Linux (C): you must already have an AI coding assistant, then run a command such as \"npx plugins add google-labs-code/stitch-skills --scope project --target claude-code\" or add the marketplace in Codex, and before anything works you have to follow Stitch's own MCP setup to register the server with credentials. For OpenCode the README says you copy the skill folders by hand and sometimes rename them.","costNote":"The skills are free (Apache 2.0). They do nothing on their own: you need Google Stitch with your own credentials and an AI coding assistant, and those have their own prices and limits.","notVerified":["Whether the Stitch MCP server, its setup flow and the endpoints the skills call still work as described.","Whether each skill loads and behaves the same in every assistant listed, since OpenCode already needs manual changes.","What Google Stitch itself allows and charges for, and what it keeps of the code and designs you upload.","The quality of the React, React Native and Vite code the build skills produce.","What the bundled scripts do when your assistant runs them."]},"zh-Hans":{"summaryShort":"面向AI编程助手的插件,让你用Google Stitch设计应用界面并转换成React代码。","summary":"一套可以安装到AI编程助手(Claude Code、Codex、Cursor、Gemini CLI、Antigravity或OpenCode)中的技能和插件包,让你通过输入请求来使用Google Stitch:根据描述生成应用界面、创建变体、把现有前端代码转换成Stitch设计、把设计系统提取到DESIGN.md文件中,以及把Stitch界面转换成React、React Native或Vite仪表盘。它面向已经会写软件的人:你需要上述助手之一,以及用自己的凭据配置好的Stitch MCP服务器。","easeRationale":"在Windows、macOS和Linux上的操作是一样的(C级):你必须已经有一个AI编程助手,然后运行类似\"npx plugins add google-labs-code/stitch-skills --scope project --target claude-code\"的命令,或者在Codex中添加市场(marketplace),而且在任何功能生效之前,你都得按照Stitch自己的MCP设置流程,用凭据注册服务器。对于OpenCode,README说你要手动复制技能文件夹,有时还要重命名。","costNote":"这些技能是免费的(Apache 2.0)。它们本身什么也做不了:你需要有自己凭据的Google Stitch,以及一个AI编程助手,这些都有各自的价格和限制。","notVerified":["Stitch MCP服务器、其设置流程以及技能调用的接口是否仍按描述正常工作。","每个技能在列出的所有助手中是否都能加载并表现一致,因为OpenCode已经需要手动修改。","Google Stitch本身允许做什么、收取什么费用,以及它对你上传的代码和设计保留了什么。","构建类技能生成的React、React Native和Vite代码的质量如何。","当你的助手运行这些内置脚本时,脚本具体做了什么。"]}}},{"id":"R_kgDOU-uhKw","slug":"google-research__harness2","name":"Harness2","owner":"google-research","fullName":"google-research/harness2","type":"framework_infra","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"21 天内获得 20552 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"agents_run_with_no_sandbox_and_open_network","severity":"attention","evidence":"README，Web access 部分：\"Executor network access is unrestricted, and Codex runs with its sandbox disabled.\" 项目要求你在评测前自行配置网络限制，因此默认情况下，它在你的电脑上运行的 AI 智能体拥有完整的互联网访问权限，且没有沙盒。","source":"ai","title":"智能体在无沙盒且网络开放的环境中运行"},{"code":"not_an_official_google_product","severity":"info","evidence":"README 免责声明（Disclaimer）：\"This is not an officially supported Google product. This project is not eligible for the Google Open Source Software Vulnerability Rewards Program.\" 该仓库位于 google-research 组织下，这可能会给人不同的印象。","source":"ai","title":"并非 Google 官方产品"},{"code":"results_come_from_the_authors_paper","severity":"info","evidence":"README 中引用的提升幅度（\"up to 11.5 points on WorkBuddy-Bench, 4.5 points on JobBench, and 11.8 percentage points on LAB's partial pass rate\"）来自作者自己发表在 arXiv 上的论文 2610.14224，使用 GLM-5 和 Gemini 3.5 Flash 测得。","source":"ai","title":"结果来自作者自己的论文"},{"code":"model_bills_can_add_up","severity":"info","evidence":"README，Run a benchmark 部分：\"Model-backed runs incur API costs; start with a small task selection\"，并且递归预算 --k 会成倍增加每个任务的智能体运行次数。","source":"ai","title":"模型费用可能快速累积"},{"code":"early_code_with_no_releases","severity":"info","evidence":"pyproject.toml 中版本为 0.1.0，GitHub 上没有发布版本（release），创建于 2026-10-06，最后一次提交在 2026-10-14，而检查结果已显示 21 天内获得 20552 个星标。","source":"ai","title":"早期代码，尚无正式发布"}],"platforms":["linux"],"tools":["python","node","docker","other"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["Google Cloud Vertex AI (model access)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[],"declaredAlternatives":[],"stars":23387,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":951,"growthDays":7,"history":[{"date":"2026-10-25","stars":19673},{"date":"2026-10-26","stars":20162},{"date":"2026-10-27","stars":20549},{"date":"2026-10-28","stars":21026},{"date":"2026-10-29","stars":21312},{"date":"2026-10-30","stars":21509},{"date":"2026-10-31","stars":21675},{"date":"2026-11-01","stars":21904},{"date":"2026-11-02","stars":22108},{"date":"2026-11-03","stars":22436},{"date":"2026-11-04","stars":22644},{"date":"2026-11-05","stars":22808},{"date":"2026-11-06","stars":22931},{"date":"2026-11-07","stars":23043},{"date":"2026-11-08","stars":23131},{"date":"2026-11-09","stars":23271},{"date":"2026-11-10","stars":23387}],"firstPublishedAt":"2026-10-25T23:19:05.540Z","commitSha":"1cfe254e8d25b15f37048c3730134686abe60335","verifiedAt":"2026-10-28T01:44:09.748Z","lastCommitAt":"2026-11-02T02:17:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A research framework that rewrites an AI agent's own prompts, tools and memory to make it do better on a task.","summary":"Harness2 is research code from a Google team for improving the system around an AI model: the prompt, guardrails, memory, tool guidance, subagents, skills, scripts and plugins. For each task it proposes edits, tries them out, compares the runs and composes a refined setup, keeping the model itself unchanged, then carries the lessons to the next task. It comes with three professional benchmarks (JobBench, LAB, WorkBuddy-Bench) and runs agents through OpenCode or Codex. It is for researchers: Linux, Python 3.12, uv, Node, Docker and a Vertex AI account are needed, and real runs cost model money.","easeRationale":"Linux is the only system it targets (\"Use Linux, Python 3.12, and uv\", and the package is marked POSIX-only), and it is a D there: you create a virtual environment with uv, and the benchmarks additionally need Node.js 20+, the Google Cloud CLI, Docker for WorkBuddy-Bench and Pandoc and Poppler for LAB, all driven from the terminal. Windows and Mac are not covered.","costNote":"The code is free (Apache-2.0). The README warns that \"Model-backed runs incur API costs\" through your own Vertex AI account; only the built-in demo runs without model credentials.","notVerified":["Whether the reported gains reproduce with other models or on your own tasks.","What a full benchmark sweep really costs in model usage.","How much disk space the three benchmarks and their Docker images take.","Whether the benchmark data and third-party code under third_party/ and dependencies/ can be used for your purpose; they keep their own licences.","What the agents do with unrestricted internet access during a run."]},"zh-Hans":{"summaryShort":"一个研究框架，会重写 AI 智能体自身的提示词、工具和记忆，让它在任务中表现更好。","summary":"Harness2 是 Google 一个团队出品的研究代码，用于改进围绕 AI 模型的整套系统：提示词、防护机制、记忆、工具使用指引、子智能体、技能、脚本和插件。针对每个任务，它会提出修改方案，逐一试验，比较各次运行结果，再组合出一套优化配置，模型本身保持不变，然后把学到的经验带入下一个任务。它自带三个专业基准测试（JobBench、LAB、WorkBuddy-Bench），并通过 OpenCode 或 Codex 运行智能体。它面向研究人员：需要 Linux、Python 3.12、uv、Node、Docker 和一个 Vertex AI 账户，真实运行会产生模型调用费用。","easeRationale":"Linux 是唯一支持的系统（文档写明“使用 Linux、Python 3.12 和 uv”，软件包也标注为仅支持 POSIX），评分为 D：你需要用 uv 创建虚拟环境，几个基准测试还分别需要 Node.js 20+、Google Cloud CLI、运行 WorkBuddy-Bench 用的 Docker，以及运行 LAB 用的 Pandoc 和 Poppler，所有操作都在终端里完成。Windows 和 Mac 都不支持。","costNote":"代码是免费的（Apache-2.0 许可）。README 提醒“涉及模型的运行会产生 API 费用”，费用通过你自己的 Vertex AI 账户支付；只有内置的演示不需要模型凭据。","notVerified":["报告中的效果提升能否在其他模型或你自己的任务上重现。","完整跑一轮基准测试在模型用量上实际要花多少钱。","三个基准测试及其 Docker 镜像占用多少磁盘空间。","third_party/ 和 dependencies/ 目录下的基准数据和第三方代码是否可以用于你的用途；它们各自保留自己的许可证。","智能体在不受限制地访问互联网期间，运行过程中到底会做什么。"]}}},{"id":"R_kgDOUZMgZg","slug":"google__elephant-goldfish-model","name":"Elephant Goldfish Model","owner":"google","fullName":"google/elephant-goldfish-model","type":"content_list","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"46 天内获得 5952 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"not_an_officially_supported_google_product","severity":"info","evidence":"README 的 License 部分：\"[Apache 2.0](LICENSE). This is not an officially supported Google product.\" 它位于 GitHub 的 `google` 组织下，但不提供 Google 的支持。","source":"ai","title":"并非 Google 官方支持的产品"},{"code":"review_stops_on_agents_that_cannot_create_subagents","severity":"info","evidence":"README 的 \"How it works\" 部分：\"Goldfish reviews require fresh subagent contexts. If your agent cannot create them, EGM stops the review.\" 同一部分还补充说：\"A review does not replace tests or human judgment.\"","source":"ai","title":"智能体无法创建子智能体时，审查会中止"},{"code":"adoption_claim_cannot_be_checked","severity":"info","evidence":"README：\"Thousands of Google engineers use EGM, an open-source plugin, to draft specs with deep-context agents\"。仓库中没有任何内容能佐证这个数字。","source":"ai","title":"使用规模的说法无法核实"},{"code":"early_version","severity":"info","evidence":"该仓库创建于 2026-09-13，只有一个发布版本 v0.1.0（2026-10-27），有 53 个未解决的 issue，因此其中的技能和命令仍可能变化。","source":"ai","title":"早期版本"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":17116,"downloads":671,"downloadsWeek":364,"downloadsDays":7,"growth":5830,"growthDays":7,"history":[{"date":"2026-10-29","stars":5941},{"date":"2026-10-30","stars":6623},{"date":"2026-10-31","stars":7777},{"date":"2026-11-01","stars":9011},{"date":"2026-11-02","stars":10634},{"date":"2026-11-03","stars":11286},{"date":"2026-11-04","stars":11687},{"date":"2026-11-05","stars":12369},{"date":"2026-11-06","stars":13030},{"date":"2026-11-07","stars":13743},{"date":"2026-11-08","stars":14878},{"date":"2026-11-09","stars":16333},{"date":"2026-11-10","stars":17116}],"firstPublishedAt":"2026-10-29T01:54:21.319Z","commitSha":"f0ba151b252c9abf1318d9fe857a43265d06de89","verifiedAt":"2026-10-30T01:38:02.853Z","lastCommitAt":"2026-11-10T21:25:07Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Skills that make your coding agent write a design document, then have fresh agents poke holes in it before coding.","summary":"A plugin with seven skills for coding agents (Antigravity, Claude Code, Codex and others). One agent keeps all your project context and drafts the specification for a feature; other agents read only the document and the code it points to, with none of the earlier chat, and list missing assumptions and risks. You decide what to change and approve the design before any code is written. For developers planning a feature, not for small bug fixes. Nothing to install beyond the coding agent you already pay for, plus one install command.","easeRationale":"B wherever your coding agent runs: once you have the agent, installation is one or two commands (`claude plugin marketplace add google/elephant-goldfish-model` then `claude plugin install egm@egm`, or `agy plugin install ...` for Antigravity), then you start a new session and type `/egm-brainstorm`. There is nothing else to install, but you do need a paid agent account.","costNote":"The files are Apache-2.0 and free. Using them needs a coding agent that costs money (Claude Code, Codex, Antigravity and similar), and a review spawns several fresh agents, so one design can use a lot of tokens.","notVerified":["Whether the skills install and run as described in each listed agent, including Claude Cowork and ChatGPT Work.","How many tokens, and how much money, a full brainstorm-draft-review cycle uses on your own plan.","Whether the reviewer findings are actually useful on your codebase, or mostly generic remarks.","What the skills read from your project and send to the model during a review."]},"zh-Hans":{"summaryShort":"一组技能，让你的编程代理先写一份设计文档，再由全新的代理挑出其中的漏洞，然后才开始写代码。","summary":"这是一个插件，为编程代理（Antigravity、Claude Code、Codex 等）提供七项技能。一个代理保留你项目的全部上下文，负责为某个功能起草规格文档；其他代理只读取这份文档以及它指向的代码，不接触之前的对话，列出缺失的假设和风险。你决定要改什么，并在写任何代码之前批准这份设计。适合正在规划功能的开发者，而不是用来修小 bug。除了你已经付费使用的编程代理之外，不需要再安装别的东西，只要执行一条安装命令。","easeRationale":"无论你的编程代理运行在哪里，评级都是 B：一旦有了编程代理，安装只需一两条命令（`claude plugin marketplace add google/elephant-goldfish-model` 然后 `claude plugin install egm@egm`，Antigravity 则用 `agy plugin install ...`），之后开一个新会话并输入 `/egm-brainstorm` 即可。没有其他东西需要安装，但你需要一个付费的代理账户。","costNote":"这些文件采用 Apache-2.0 许可，免费。使用它们需要一个付费的编程代理（Claude Code、Codex、Antigravity 等），而每次评审都会启动多个全新的代理，所以一份设计方案可能会消耗不少 token。","notVerified":["这些技能是否能在列出的每个代理（包括 Claude Cowork 和 ChatGPT Work）中按描述安装和运行","在你自己的套餐下，完整走一遍头脑风暴、起草、评审流程会消耗多少 token、花多少钱","评审者给出的发现是否真的对你的代码库有用，还是大多是泛泛而谈","评审过程中这些技能会从你的项目中读取并发送给模型哪些内容"]}}},{"id":"R_kgDOMGbPlg","slug":"googleapis__mcp-toolbox","name":"MCP Toolbox for Databases","owner":"googleapis","fullName":"googleapis/mcp-toolbox","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"ai_can_run_queries_on_your_database","severity":"attention","evidence":"README:\"Prebuilt generic tools for instant data exploration (e.g., `list_tables`, `execute_sql`) directly from your IDE or CLI.\"使用预置工具集时,AI 助手可以向你连接的数据库发送 SQL 语句,所以你给它的账号权限决定了它能读取或更改什么。","source":"ai","title":"ai 可以在你的数据库上执行查询"},{"code":"credentials_in_plain_config_file","severity":"info","evidence":"README 示例把数据库密码直接写在配置文件里:tools.yaml 中的\"user: toolbox_user\" / \"password: my-password\"。","source":"ai","title":"凭证以明文存在配置文件中"},{"code":"many_open_issues","severity":"info","evidence":"GitHub 上有 431 个未解决的 issue,分散在数十个支持的数据库中,所以每个数据库得到的支持可能并不均衡。","source":"ai","title":"大量未解决的 issue"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["docker","go"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Database credentials (user and password or Google Cloud IAM)","Google Gemini (GEMINI_API_KEY / GOOGLE_API_KEY, only for the example apps in the README)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":16835,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":35,"growthDays":6,"history":[{"date":"2026-11-05","stars":16812},{"date":"2026-11-06","stars":16819},{"date":"2026-11-07","stars":16821},{"date":"2026-11-08","stars":16825},{"date":"2026-11-09","stars":16830},{"date":"2026-11-10","stars":16835}],"firstPublishedAt":"2026-11-04T22:32:54.960Z","commitSha":"44189342ac883a48626433c34be48f85c2ce5990","verifiedAt":"2026-11-04T22:15:56.019Z","lastCommitAt":"2026-11-10T19:27:46Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A server from Google that lets AI assistants read and query your databases in plain language.","summary":"An open-source MCP server, published by Google, that sits between an AI assistant (Gemini CLI, Claude Code, Codex, Cursor and others) and databases such as PostgreSQL, MySQL, SQL Server, Oracle, MongoDB, Redis, BigQuery or Spanner. Ready-made tools let the assistant list tables and run queries, and a tools.yaml file lets developers define their own, narrower tools. It is for developers and data teams: you need a terminal, a database you already have access to and a configuration file. Apache-2.0 licensed and actively developed.","easeRationale":"The same terminal work on Windows, Mac and Linux: the featured quick start runs `npx @toolbox-sdk/server` (so Node.js), or you download the toolbox binary with curl, and in both cases you still write a tools.yaml with database connection details and paste an MCP block into your assistant's settings. Homebrew helps on Mac and Linux, but there is no installer and nothing works until you already have a database to point at.","costNote":"The server is free and open source. The databases or Google Cloud services you point it at may cost money, and the AI assistant you use it with may be paid.","notVerified":["Whether each of the listed databases really works, since each has its own driver","Whether the prebuilt toolsets can be limited to read-only in practice","Disk and memory use of the server once running","What the AI assistant you connect does with the data it reads from your tables"]},"zh-Hans":{"summaryShort":"Google 推出的一个服务器,让 AI 助手能用普通语言读取和查询你的数据库。","summary":"一个由 Google 发布的开源 MCP 服务器,位于 AI 助手(Gemini CLI、Claude Code、Codex、Cursor 等)和数据库(如 PostgreSQL、MySQL、SQL Server、Oracle、MongoDB、Redis、BigQuery 或 Spanner)之间。预置的工具让助手列出表格并执行查询,开发者还可以通过 tools.yaml 文件自定义范围更小的工具。它适合开发者和数据团队:你需要一个终端、一个你已经有权限访问的数据库,以及一个配置文件。采用 Apache-2.0 许可证,积极维护中。","easeRationale":"在 Windows、Mac 和 Linux 上是同样的终端操作:推荐的快速上手方式是运行 `npx @toolbox-sdk/server`(也就是需要 Node.js),或者用 curl 下载 toolbox 的二进制文件,两种方式都仍需要你写一个包含数据库连接信息的 tools.yaml,并把一段 MCP 配置粘贴到你助手的设置里。Homebrew 在 Mac 和 Linux 上能帮点忙,但没有安装程序,而且在你有数据库可以连接之前什么都用不了。","costNote":"这个服务器是免费开源的。你连接的数据库或 Google Cloud 服务可能需要付费,你搭配使用的 AI 助手也可能是付费的。","notVerified":["列出的每个数据库是否真的都能用,因为每个都有自己的驱动","预置的工具集在实际使用中能不能限制为只读","服务器运行后的磁盘和内存占用情况","你连接的 AI 助手如何处理它从你表格中读取的数据"]}}},{"id":"MDEwOlJlcG9zaXRvcnk4NjAwMzUxMA==","slug":"googlechrome__rendertron","name":"Rendertron","owner":"GoogleChrome","fullName":"GoogleChrome/rendertron","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发行是2020-11-04(2192天前)","source":"check","title":"发行版已经过时"},{"code":"archived","severity":"info","evidence":"该仓库已归档(只读,不再维护)","source":"check","title":"已归档"},{"code":"deprecated_by_authors","severity":"attention","evidence":"README.md顶部写道:\"Rendertron is deprecated... Dynamic rendering is not a recommended approach and there are better approaches to rendering on the web. **Rendertron will not be actively maintained at this point.**\" 仓库已归档,最后一个发行版3.1.0发布于2020-11-04。","source":"ai","title":"已被作者弃用"},{"code":"renders_any_url_by_default","severity":"attention","evidence":"接口是`GET /render/<url>`和`GET|POST /screenshot/<url>`,而限制它们的两个选项默认都是关闭的:`renderOnly`(\"restrict the endpoint to only service requests for certain domains\")没有设置,`restrictedUrlPattern`的说明是\"_default `null`_\"。一个暴露在公网上的实例会抓取并截图任何人发给它的任意地址。","source":"ai","title":"默认会渲染任意url"},{"code":"old_unpatched_dependencies","severity":"attention","evidence":"package.json固定了`puppeteer: ^10.0.0`、`koa: ^2.13.1`、`typescript: 4.0.5`以及`engines: { node: >=10 }`,全部来自2020到2021年,而项目已归档,不会再更新。Puppeteer 10还会下载自己那套老旧的Chromium。","source":"ai","title":"依赖项陈旧且未打补丁"},{"code":"public_demo_instance_in_the_docs","severity":"info","evidence":"仓库主页和README中的一个示例使用了一个共享的托管实例,`https://render-tron.appspot.com/render/...`。把自己的页面交给别人的服务器处理,意味着对方能看到你渲染的每一个地址,而一个已归档的项目也不保证这个实例会一直在线。","source":"ai","title":"文档中使用了公开的演示实例"},{"code":"ten_second_render_limit","severity":"info","evidence":"README的FAQ写道:\"There is a hard limit of 10 seconds for rendering\",`timeout`配置默认是10000毫秒。较慢的页面会返回未完成的结果。","source":"ai","title":"渲染限制为十秒"},{"code":"needs_code_changes_on_your_site","severity":"info","evidence":"README.md:\"you'll need to implement the differential serving layer. This checks the user agent to determine whether prerendering is required.\" 这个仓库里只有Express.js的中间件;Firebase、ASP.net和Django的中间件都标注为\"Community maintained\"。","source":"ai","title":"需要在你的网站上修改代码"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","social_marketing"],"alternativeTo":[{"slug":"prerender-io","name":"Prerender.io","confirmed":true}],"declaredAlternatives":[],"stars":5948,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":5950},{"date":"2026-11-07","stars":5950},{"date":"2026-11-08","stars":5949},{"date":"2026-11-09","stars":5949},{"date":"2026-11-10","stars":5948}],"firstPublishedAt":"2026-11-05T22:43:48.428Z","commitSha":"ffc2b331ff751c1f9e84cd6e6d485b802442fdf2","verifiedAt":"2026-11-05T22:28:49.538Z","lastCommitAt":"2022-10-06T07:51:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A small server that opens your web pages in Chrome and hands the finished HTML to search engines. Deprecated by Google.","summary":"A server from the Google Chrome team that loads a web address in an invisible Chrome, waits for the JavaScript to finish and returns the finished HTML, so search engines and other bots see the real content of a JavaScript site. It can also take screenshots. You run it yourself with Node and Chrome installed, and add a piece of middleware to your own site. The authors marked it deprecated and will not maintain it, and the repository is archived and read-only.","easeRationale":"A C on all three, and Linux is where it is normally deployed: you install Node, have Chrome on the machine, then `npm install -g rendertron` and run `rendertron`, or clone and `npm install && npm run build && npm run start`. There is no installer and no graphical window — it is a server plus middleware you add to your own site, and the README now points to Puppeteer's own troubleshooting guide for Docker since \"Rendertron no longer includes a Docker file\".","costNote":"No price (Apache-2.0). If you follow the Google Cloud deployment route, Google charges for the hosting.","notVerified":["Whether it still builds and runs with current versions of Node and Chrome","Whether the hosted render-tron.appspot.com instance still answers","How well search engines treat pages served this way today, given Google's own advice against dynamic rendering","Real memory and disk use with a Chrome running behind it","Whether the community middleware projects still work"]},"zh-Hans":{"summaryShort":"一个小型服务器,在Chrome中打开你的网页并把渲染好的HTML交给搜索引擎。已被Google弃用。","summary":"一个来自Google Chrome团队的服务器,它在一个不可见的Chrome中加载某个网址,等待JavaScript执行完毕,再返回渲染好的HTML,这样搜索引擎和其他爬虫就能看到JavaScript网站的真实内容。它还可以截图。你需要自己安装Node和Chrome并运行它,再给自己的网站加一段中间件。作者已将其标记为弃用,不会再维护,仓库也已归档为只读状态。","easeRationale":"三个系统上的难度评级都是C,通常部署在Linux上:你安装Node、在机器上准备好Chrome,然后执行`npm install -g rendertron`并运行`rendertron`,或者克隆仓库后执行`npm install && npm run build && npm run start`。没有安装程序,也没有图形窗口——它只是一个服务器,加上你要加到自己网站里的中间件,README现在还指向Puppeteer自己的Docker故障排查指南,因为\"Rendertron no longer includes a Docker file\"。","costNote":"没有价格(Apache-2.0)。如果你走Google Cloud的部署路线,Google会就托管收费。","notVerified":["在当前版本的Node和Chrome下是否还能编译和运行","托管实例render-tron.appspot.com是否仍能响应","考虑到Google自己也不建议使用动态渲染,如今搜索引擎对这种方式提供的页面处理得如何","在后台运行一个Chrome实例时实际的内存和磁盘占用","社区维护的中间件项目是否仍然可用"]}}},{"id":"R_kgDOVg4HSg","slug":"govijr__terminal-mac","name":"terminal-mac","owner":"govijr","fullName":"govijr/terminal-mac","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"mac":"D"},"risk":"attention","signals":[{"code":"unsigned_by_apple","severity":"attention","evidence":"README.md:\"Releases are self-signed (no Apple notarization), which the curl installer doesn't need.\"发布的文件是普通的 ZIP(terminal-mac-0.2.0-6-macos-arm64.zip),所以 macOS 没有对它们做过检查。","source":"ai","title":"未经 apple 签名"},{"code":"early_prototype_not_validated_live","severity":"attention","evidence":"README.md 称其为\"an early release candidate\",并说\"Live guest acceptance on a clean Mac is still pending\";测试被描述为\"automated protocol checks, not evidence that a guest was installed, booted, granted permissions, or controlled successfully\"。","source":"ai","title":"早期原型,尚未在真实环境验证"},{"code":"not_a_hardened_sandbox","severity":"attention","evidence":"README.md:\"Default NAT networking permits guest internet access and may reach services on the host or local network. This is **not** a network-isolated or hardened hostile-code environment.\"这台虚拟 Mac 和你的文件是分开的,但它不是用来隔绝不可信代码的防护屏障。","source":"ai","title":"不是加固的隔离环境"},{"code":"install_script_piped_from_internet","severity":"info","evidence":"README.md 中的安装命令是 `curl -fsSL https://github.com/govijr/terminal-mac/releases/latest/download/get.sh | bash`,会直接执行一个下载下来的脚本;README 说安装程序在安装前会校验发布文件的 SHA-256,且不需要管理员权限。","source":"ai","title":"安装脚本直接从网上执行"},{"code":"large_disk_use_and_no_delete_command","severity":"info","evidence":"README.md 提到一个被缓存的 Apple 恢复镜像,以及位于 ~/Library/Application Support/terminal-mac 中的 \"a growing VM disk (64 GiB logical size by default)\",并说 \"There is deliberately no destructive VM-delete command yet\",所以清理需要你自己手动删除文件夹。","source":"ai","title":"占用大量磁盘空间且没有删除命令"},{"code":"very_new_project","severity":"info","evidence":"该仓库创建于 2026-11-03,截至 2026-11-04 检查时已有 1447 个星标;它三个发布版本中最新的 v0.2.0-6 也是同一天发布的。","source":"ai","title":"非常新的项目"}],"platforms":["mac"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","ai_skills","network_system"],"alternativeTo":[{"slug":"parallels-desktop","name":"Parallels Desktop","confirmed":true},{"slug":"browserbase","name":"Browserbase","confirmed":true}],"declaredAlternatives":[],"stars":2548,"downloads":3126,"downloadsWeek":1470,"downloadsDays":5,"growth":2548,"growthDays":7,"history":[{"date":"2026-11-05","stars":1921},{"date":"2026-11-06","stars":2128},{"date":"2026-11-07","stars":2234},{"date":"2026-11-08","stars":2372},{"date":"2026-11-09","stars":2502},{"date":"2026-11-10","stars":2548}],"firstPublishedAt":"2026-11-04T22:32:54.967Z","commitSha":"7009117f9e910c3dd4b4bd863e8bc331bd3c9d96","verifiedAt":"2026-11-04T22:15:00.398Z","lastCommitAt":"2026-11-10T06:06:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs a second, separate Mac inside your Mac, shows it in the terminal, and lets an AI assistant click and type in it.","summary":"This starts a real macOS virtual machine on an Apple silicon Mac and draws its screen inside the terminal (Ghostty, Kitty or iTerm2; other terminals get a normal window). The point is to give an AI assistant a Mac of its own: it installs as an MCP server plus a skill for Claude Code, Cursor, Codex and similar, with screenshot, click, type and scroll commands that never touch your own screen or clipboard. You need an Apple silicon Mac with macOS 14+, several gigabytes of download from Apple and a 64 GiB virtual disk.","easeRationale":"Mac only, and only Apple silicon with macOS 14 or later: you install from the terminal with a one-line script, wait for a multi-gigabyte macOS download, then go through the Mac setup assistant inside the virtual machine, open a helper app in it by hand and grant it Screen Recording and Accessibility. Connecting an AI assistant means editing MCP settings or running another command.","costNote":"MIT licensed. The virtual Mac downloads macOS directly from Apple at no charge.","notVerified":["Whether a virtual Mac really installs, boots and accepts clicks on a clean Apple silicon Mac","How well the terminal picture and typing feel in Ghostty, Kitty or iTerm2","Real disk and memory use once the restore image is cached and the virtual disk grows","Whether the MCP tools work with the AI assistant you use","How Apple's macOS licence terms apply to the number of virtual Macs you run"]},"zh-Hans":{"summaryShort":"在你的 Mac 里运行另一台独立的 Mac,在终端里显示它,还能让 AI 助手在里面点击和输入。","summary":"这会在搭载 Apple silicon 芯片的 Mac 上启动一个真正的 macOS 虚拟机,并把它的屏幕画在终端里(Ghostty、Kitty 或 iTerm2;其他终端会得到一个普通窗口)。它的目的是给 AI 助手一台专属的 Mac:它以 MCP 服务器加上一个适用于 Claude Code、Cursor、Codex 等工具的 skill 的形式安装,提供截图、点击、输入和滚动命令,这些命令不会碰到你自己的屏幕或剪贴板。你需要一台搭载 Apple silicon 芯片、运行 macOS 14 及以上版本的 Mac,从 Apple 下载几个 GB 的内容,并准备一个 64 GiB 的虚拟磁盘。","easeRationale":"仅支持 Mac,而且只支持运行 macOS 14 及以上版本的 Apple silicon 芯片:你通过终端用一行脚本安装,等待几个 GB 的 macOS 下载,然后在虚拟机里完成 Mac 的设置助理流程,手动在里面打开一个辅助应用,并授予它屏幕录制和辅助功能权限。连接 AI 助手则需要修改 MCP 设置或运行另一条命令。","costNote":"采用 MIT 许可证。虚拟 Mac 直接从 Apple 免费下载 macOS。","notVerified":["虚拟 Mac 在一台全新的 Apple silicon 芯片 Mac 上是否真的能安装、启动并接受点击操作","在 Ghostty、Kitty 或 iTerm2 中,终端画面和输入的手感如何","一旦恢复镜像被缓存、虚拟磁盘不断增长后,真实的磁盘和内存占用是多少","MCP 工具是否能配合你使用的 AI 助手正常工作","Apple 的 macOS 许可条款如何适用于你运行的虚拟 Mac 数量"]}}},{"id":"R_kgDOR4-pDw","slug":"graphify-labs__graphify","name":"Graphify","owner":"Graphify-Labs","fullName":"Graphify-Labs/graphify","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"213 天内获得 141233 个 star（约每天 663 个）","source":"check","title":"star 增长速度异常"},{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接接入 shell 执行：curl -LsSf https://astral.sh/uv/install.sh \\| sh","source":"check","title":"把 curl 直接接到 shell 执行"},{"code":"pypi_name_easy_to_confuse","severity":"attention","evidence":"README，Install：“Official package: The PyPI package is graphifyy (double-y). Other graphify* packages on PyPI are not affiliated.”输入看起来正常的名字，装的却是别人的包。","source":"ai","title":"pypi 包名容易混淆"},{"code":"changes_your_assistant_settings_and_intercepts_its_tools","severity":"attention","evidence":"README：“graphify install ... detects your assistant and registers the skill”，会写入像“.claude/skills/graphify/SKILL.md”“AGENTS.md”或“.cursor/rules/”这样的文件，并且“a hook fires automatically before search-style tool calls (and, on Claude Code, before reading source files one by one via the Read/Glob tools)”；在 Pi 上它会安装一个原生扩展。","source":"ai","title":"修改你助手的设置并拦截它的工具调用"},{"code":"documents_and_images_go_to_an_ai_service","severity":"attention","evidence":"README，Privacy：“Docs, PDFs, images: sent to your AI assistant for semantic extraction”，而且“Kimi (MOONSHOT_API_KEY) routes to Moonshot AI servers in China”。只有代码和音频留在本机。","source":"ai","title":"文档和图片会发送给一个 ai 服务"},{"code":"your_questions_are_written_to_a_log_file","severity":"info","evidence":"README，Privacy：“every graphify query, graphify path, graphify explain, and MCP query_graph call is logged to ~/.cache/graphify-queries.log in JSON Lines format”；设置 GRAPHIFY_QUERY_LOG_DISABLE=1 可以关闭这个功能。","source":"ai","title":"你的查询会被写入日志文件"},{"code":"downloads_pages_and_videos_when_asked","severity":"info","evidence":"README，Common commands：“/graphify add https://arxiv.org/abs/1706.03762”和“/graphify add <youtube-url> # transcribe and add a video”；视频扩展会安装 yt-dlp。","source":"ai","title":"按要求下载网页和视频"},{"code":"numbers_measured_by_the_authors","severity":"info","evidence":"“Benchmarks: 0.497 recall@10, 0 LLM credits”表格使用项目自己的测试框架和评判标准，把 graphify 与 mem0、supermemory 做比较，因此这个对比并非来自独立第三方。","source":"ai","title":"数据由作者自己测量"},{"code":"sales_pitch_for_the_paid_cloud_inside_the_readme","severity":"info","evidence":"README 多处重复“Cut agent token spend by 80%”和“Start a free 14-day trial at app.graphify.com”，这家公司被列为 Y Combinator S26 届的创业公司。","source":"ai","title":"readme 里内置了付费云服务的推广"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI","Anthropic","Google Gemini","DeepSeek","Moonshot Kimi"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_skills","productivity"],"alternativeTo":[{"slug":"graphify-cloud","name":"Graphify Cloud","confirmed":true},{"slug":"supermemory","name":"Supermemory","confirmed":true},{"slug":"mem0","name":"Mem0","confirmed":true}],"declaredAlternatives":[],"stars":145583,"downloads":13569,"downloadsWeek":1108,"downloadsDays":7,"growth":3743,"growthDays":7,"history":[{"date":"2026-10-26","stars":136162},{"date":"2026-10-27","stars":136993},{"date":"2026-10-28","stars":137858},{"date":"2026-10-29","stars":138699},{"date":"2026-10-30","stars":139485},{"date":"2026-10-31","stars":140040},{"date":"2026-11-01","stars":140581},{"date":"2026-11-02","stars":141199},{"date":"2026-11-03","stars":141840},{"date":"2026-11-04","stars":142363},{"date":"2026-11-05","stars":142906},{"date":"2026-11-06","stars":143490},{"date":"2026-11-07","stars":143950},{"date":"2026-11-08","stars":144431},{"date":"2026-11-09","stars":144972},{"date":"2026-11-10","stars":145583}],"firstPublishedAt":"2026-10-27T01:10:12.631Z","commitSha":"8e6a6d2129d4aeffc7a0e9ed79b153f7f057f42b","verifiedAt":"2026-11-03T02:40:15.479Z","lastCommitAt":"2026-11-10T21:37:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a folder of code and documents into a map your AI assistant can ask, instead of reading every file.","summary":"Graphify reads a project folder and builds a map of it: files, functions, documents, PDFs and even videos become points connected by links you can follow. Your AI assistant then asks the map (\"what connects login to the database?\") instead of opening file after file. You get an HTML page you can click, a report and a data file. Code is read on your own computer; documents and images need an AI model. You install it in the terminal with Python and run one command to add it to your assistant.","easeRationale":"There is no installer: on all three systems you install Python 3.10+ and the uv or pipx tool, then run two terminal commands (uv tool install graphifyy and graphify install) before typing /graphify in your assistant. Windows and macOS have one-line package-manager commands (winget, Homebrew) but still end in the terminal.","costNote":"The program here is free under Apache-2.0. The README also sells Graphify Cloud, a separate paid service with a 14-day trial, and some optional passes use AI keys you pay for.","notVerified":["Whether the map it builds is correct for your project, since many links are marked INFERRED rather than read from the source.","Whether the assistant really saves work, as the 80% token-saving claim is the company's own.","How much the semantic pass over documents and PDFs costs in API use on a large project.","How much disk and memory a graph of a big repository needs (the estimate here is only a guess, 100-300 MB).","Whether every one of the 20-plus assistants listed is really detected and set up by graphify install."]},"zh-Hans":{"summaryShort":"把一个代码和文档文件夹变成一张地图，让你的 AI 助手可以查询，而不用逐个文件去读。","summary":"Graphify 读取一个项目文件夹并为其建立一张地图：文件、函数、文档、PDF 甚至视频都变成了用链接连接起来、可以追踪的节点。你的 AI 助手随后可以向这张地图提问（“什么把登录和数据库连接起来？”），而不用一个一个打开文件。你会得到一个可点击的 HTML 页面、一份报告和一个数据文件。代码在你自己的电脑上读取；文档和图片需要一个 AI 模型。你在终端用 Python 安装它，运行一条命令就能把它加入你的助手。","easeRationale":"没有安装程序：在三个系统上你都要安装 Python 3.10+ 和 uv 或 pipx 工具，然后运行两条终端命令（uv tool install graphifyy 和 graphify install），再在你的助手里输入 /graphify。Windows 和 macOS 上可以用包管理器（winget、Homebrew）一行命令安装，但最终还是要回到终端里完成。","costNote":"这里的程序是免费的（Apache-2.0）。README 同时也在推销 Graphify Cloud，一个单独收费、带 14 天试用的服务，一些可选的功能还会用到你自己付费的 AI 密钥。","notVerified":["它为你的项目构建的地图是否准确，因为很多链接标记为 INFERRED（推断），而不是从源代码读出来的。","助手是否真的省了工作量，因为 80% 的 token 节省说法是公司自己提出的。","在一个大项目上，对文档和 PDF 做语义分析要花多少 API 费用。","一个大型仓库生成的图谱需要多少磁盘和内存（这里的估算只是一个猜测，100-300 MB）。","列出的二十多个助手是否每一个都真的能被 graphify install 检测并配置好。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyNTg5Njc4NjA=","slug":"graphiteeditor__graphite","name":"Graphite","owner":"GraphiteEditor","fullName":"GraphiteEditor/Graphite","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"web":"A"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2022-02-04（1726 天前）","source":"check","title":"发布版本长期未更新"},{"code":"alpha_stage","severity":"attention","evidence":"README.md：\"Graphite is a free, open source vector and raster graphics engine, [available now](https://editor.graphite.art) in alpha.\" 照片编辑、动态图形和桌面排版等功能被描述为路线图中的计划项目。","source":"ai","title":"处于 alpha 阶段"},{"code":"desktop_app_without_downloads","severity":"info","evidence":"Cargo.toml 列出了 desktop/platform/linux、desktop/platform/mac 和 desktop/platform/win 这些工作区成员，但唯一的发布版本（latest-stable，2022-02-04）没有任何平台的可执行文件。","source":"ai","title":"桌面应用没有可下载的安装包"}],"platforms":["web"],"tools":[],"buildTools":["rust"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["design_cad","image"],"alternativeTo":[{"slug":"illustrator","name":"Illustrator","confirmed":true},{"slug":"photoshop","name":"Photoshop","confirmed":true},{"slug":"affinity-designer","name":"Affinity Designer","confirmed":true},{"slug":"indesign","name":"InDesign","confirmed":false}],"declaredAlternatives":[],"stars":27895,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":60,"growthDays":7,"history":[{"date":"2026-10-25","stars":27728},{"date":"2026-10-26","stars":27736},{"date":"2026-10-27","stars":27743},{"date":"2026-10-28","stars":27769},{"date":"2026-10-29","stars":27776},{"date":"2026-10-30","stars":27785},{"date":"2026-10-31","stars":27802},{"date":"2026-11-01","stars":27814},{"date":"2026-11-02","stars":27822},{"date":"2026-11-03","stars":27835},{"date":"2026-11-04","stars":27843},{"date":"2026-11-05","stars":27842},{"date":"2026-11-06","stars":27852},{"date":"2026-11-07","stars":27856},{"date":"2026-11-08","stars":27870},{"date":"2026-11-09","stars":27881},{"date":"2026-11-10","stars":27895}],"firstPublishedAt":"2026-10-25T23:19:06.404Z","commitSha":"c20da5c3a59d34ad9247ce915059be8118f298bf","verifiedAt":"2026-10-28T01:35:16.445Z","lastCommitAt":"2026-11-10T14:58:36Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Free graphics editor that runs in the browser for vector drawing, design and procedural art.","summary":"Graphite is a free 2D graphics editor you open in the browser at editor.graphite.art. You can draw and edit vector art, apply effects and build procedural designs by connecting nodes, with changes that stay editable. It is aimed at designers, illustrators and digital artists. Nothing to install for the web version. The README says it is in alpha, so parts are unfinished; photo editing, motion graphics and desktop publishing are listed as planned, not finished. Desktop code exists in the repository but there are no ready desktop downloads.","easeRationale":"On the web it is grade A: the README links a hosted editor at editor.graphite.art that opens in a browser with no install. The repository also has desktop folders for Windows, Mac and Linux, but the only GitHub release has no program files, so there is no ready desktop download to grade.","costNote":"Free and community funded; the site asks for donations.","notVerified":["Whether the browser editor handles large or complex artwork without slowing down or crashing","Where your documents are stored when you use the hosted editor, and whether they survive closing the browser","Whether the in-repository desktop app builds and runs on Windows, Mac and Linux","Which file formats can really be opened and exported, and how well they match Illustrator or Photoshop files"]},"zh-Hans":{"summaryShort":"免费的图形编辑器，在浏览器中运行，用于矢量绘图、设计和程序化艺术创作。","summary":"Graphite 是一款免费的 2D 图形编辑器，可以在浏览器中打开 editor.graphite.art 使用。你可以绘制和编辑矢量图形，添加效果，并通过连接节点来构建程序化设计，所做的修改始终保持可编辑。它面向设计师、插画师和数字艺术家。网页版无需安装任何东西。README 说明它目前处于 alpha 阶段，部分功能尚未完成；照片编辑、动态图形和桌面出版都被列为计划中、尚未完成的功能。仓库里有桌面版代码，但没有现成的桌面版下载。","easeRationale":"网页端评分为 A：README 链接到一个托管编辑器 editor.graphite.art，无需安装即可在浏览器中打开。仓库里还有 Windows、Mac 和 Linux 的桌面版文件夹，但唯一的 GitHub 发布版本里没有程序文件，所以没有可供评分的现成桌面版下载。","costNote":"免费，由社区资助；网站接受捐赠。","notVerified":["浏览器编辑器在处理大型或复杂作品时是否会变慢或崩溃","使用托管编辑器时，文档保存在哪里，关闭浏览器后是否还会保留","仓库中的桌面版应用能否在 Windows、Mac 和 Linux 上构建并运行","实际能打开和导出哪些文件格式，以及它们与 Illustrator 或 Photoshop 文件的兼容程度如何"]}}},{"id":"R_kgDOPhqB7A","slug":"growchief__growchief","name":"GrowChief","owner":"growchief","fullName":"growchief/growchief","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"breaks_platform_rules_account_ban","severity":"high","evidence":"README.md，\"Things you should know\"：\"it violates the terms of service of the platforms and can result in a ban. Use it at your own risk and connect only with leads you know.\" 使用它可能让你失去LinkedIn或X账号。","source":"ai","title":"违反平台规则，可能导致账号被封"},{"code":"evades_bot_detection","severity":"attention","evidence":"README.md：\"It uses Playwright together with Patchright for maximum stealthiness\"，\"Human-like automation — GrowChief uses natural mouse movements\"，\"It always runs in headful — our Docker image is already built with xvfb\"。package.json依赖\"patchright\"和\"ghost-cursor\"。这个工具是为了在社交网络的自动防御系统面前表现得像真人而设计的。","source":"ai","title":"规避机器人检测"},{"code":"handles_other_peoples_personal_data","severity":"attention","evidence":".env.example里有APOLLO_API_KEY、DATAGMA_API_KEY和BRIGHTDATA_API_KEY，README描述了一个\"Enrichment waterfall — when you don't provide the account URL but other parameters like email, we use multiple provider to figure out the profile URL\"。你会在收集并发送第三方的联系人数据，这类数据受隐私法规管辖。","source":"ai","title":"处理他人的个人数据"},{"code":"handles_social_account_access","severity":"attention","evidence":"README.md：\"It uses a special technology to authenticate your accounts — you never need to put your username and password directly into the system.\" 服务器仍然为你的社交账号保存着一个已登录的会话，而方法只是笼统地描述了一下。","source":"ai","title":"处理社交账号的登录权限"},{"code":"no_recent_updates","severity":"info","evidence":"可核实的事实：最近一次提交在2025-10-13（检查时的386天前），最新发行版v0.3.0来自2025-09-05。","source":"ai","title":"长期没有更新"},{"code":"install_docs_outside_repository","severity":"info","evidence":"README.md的\"QuickStart / Installation\"只写了\"View https://docs.growchief.com\"，所以步骤不在仓库里，随时可能变化。","source":"ai","title":"安装说明在仓库之外"},{"code":"heavy_resource_use","severity":"attention","evidence":"README.md：\"every time you start automation it opens a Chrome browser (perfect for 1–2 accounts). However, once you scale, you need a smarter system ... as Chrome consumes a lot of memory.\" docker-compose文件还会以256 MB的Java堆内存启动Elasticsearch。","source":"ai","title":"资源占用较高"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Other (AUTH_SECRET)","Google (login with Google)","Resend (sending e-mail)","Beehiiv (newsletter)","Bright Data (proxies)","Apollo (lead data)","Datagma (lead data)","OpenAI"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"phantombuster","name":"PhantomBuster","confirmed":true},{"slug":"expandi","name":"Expandi","confirmed":true},{"slug":"linked-helper","name":"Linked Helper","confirmed":true},{"slug":null,"name":"LinkedIn Helper","confirmed":false}],"declaredAlternatives":[{"target":"expandi","stance":"claims","evidence":"<a href=\"https://growchief.com\">GrowChief</a>: An alternative to: Phantom Buster, Expandi, Zopto, LinkedIn Helper, Meet Alfred, etc.<br /><br />"}],"stars":3509,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":6,"growthDays":6,"history":[{"date":"2026-11-04","stars":3503},{"date":"2026-11-05","stars":3503},{"date":"2026-11-06","stars":3505},{"date":"2026-11-07","stars":3507},{"date":"2026-11-08","stars":3507},{"date":"2026-11-09","stars":3507},{"date":"2026-11-10","stars":3509}],"firstPublishedAt":"2026-11-04T01:36:50.419Z","commitSha":"abb1e37a6f5595d8d105aef5871a2eeb0c22a1dc","verifiedAt":"2026-11-04T01:21:33.850Z","lastCommitAt":"2025-10-13T05:36:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted tool that automatically sends LinkedIn and X connection requests and follow-up messages for you.","summary":"GrowChief is a server you run yourself that logs into your LinkedIn or X account in a hidden Chrome browser and runs step-by-step sales outreach: send a connection request, wait, send a follow-up message, and so on. It has a web screen to build the steps, an API and an n8n node, working-hour limits and support for proxies. It is meant for sales and marketing teams. Installing it means Docker with several services (PostgreSQL, Temporal, Elasticsearch). The README says this kind of automation breaks the social networks' rules and can get your account banned.","easeRationale":"There is no installer: the project ships a docker-compose.yml that starts seven containers (the app, two PostgreSQL databases, Elasticsearch and three Temporal services) and you must fill in environment variables such as DATABASE_URL and AUTH_SECRET. That is the same work on Linux, Windows and macOS, so all three are D and only people comfortable with Docker will get it running.","costNote":"The software is free to self-host. Proxies and the optional lead-finding services (Apollo, Datagma, Bright Data) are paid separately, and the authors also sell a hosted version, GrowChief Cloud.","notVerified":["Whether LinkedIn and X currently detect and block this automation","How the \"special technology\" that keeps your account logged in actually works and where that session is stored","Real memory and disk use with all seven containers running","How much the optional proxy and lead-data services cost in practice","Whether the API and the n8n node still match the current code after a year without releases"]},"zh-Hans":{"summaryShort":"自托管工具，自动在LinkedIn和X上发送加好友请求和跟进消息。","summary":"GrowChief是一个你自己运行的服务器，它在隐藏的Chrome浏览器里登录你的LinkedIn或X账号，按步骤执行销售外联：发送加好友请求、等待、发送跟进消息，依此类推。它有一个网页界面用来搭建这些步骤，还有API和n8n节点、工作时间限制和代理支持。它面向销售和营销团队。安装它需要用Docker运行多个服务（PostgreSQL、Temporal、Elasticsearch）。README说这类自动化行为违反社交网络的规则，可能导致账号被封。","easeRationale":"没有安装程序：项目提供一个docker-compose.yml，启动七个容器（应用本身、两个PostgreSQL数据库、Elasticsearch和三个Temporal服务），你需要填写DATABASE_URL和AUTH_SECRET等环境变量。这在Linux、Windows和macOS上都是同样的工作量，所以三者都是D级，只有熟悉Docker的人才能把它跑起来。","costNote":"软件本身自托管是免费的。代理和可选的线索查找服务（Apollo、Datagma、Bright Data）需要另外付费，作者还出售一个托管版本，GrowChief Cloud。","notVerified":["LinkedIn和X目前是否会检测并拦截这种自动化行为","让账号保持登录状态的\"special technology\"实际是怎么运作的，这个会话又存在哪里","七个容器同时运行时的实际内存和磁盘占用","可选的代理和线索数据服务在实际使用中要花多少钱","一年没有发行新版本后，API和n8n节点是否还与当前代码一致"]}}},{"id":"R_kgDOT1eE0w","slug":"guillaumemeyer__watermarks-remover","name":"Watermarks Remover","owner":"guillaumemeyer","fullName":"guillaumemeyer/watermarks-remover","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"attention","signals":[{"code":"removes_ai_provenance_marks","severity":"attention","evidence":"主要用途是去除标明内容由 AI 生成的标记（C2PA、SynthID 类水印）。README 的 'Ethics and disclaimer' 部分写道：'For privacy and research on your content — not academic fraud or false “human-written” claims.' README 的参考文献引用 'SB 942 / EU AI Act Art. 50' 作为 AI 内容标识的背景。","source":"ai","title":"移除 AI 来源标记"},{"code":"ai_detector_evasion_features","severity":"attention","evidence":"README 更新日志 v0.7.0：'Pangram bulk API as a human-likeness backend'、一种 'humanize' 策略（'names the human-writer rules in the prompt'），以及 'New black-box watermark-stealing module and prompt-corpus downloader (#303)'（顶层文件夹 stealer/）。","source":"ai","title":"规避 AI 检测器的功能"},{"code":"restricted_license_optional_components","severity":"info","evidence":"compose.yaml：'ctrlregen and synthid bake in upstream code that is not publicly redistributable (all-rights-reserved / non-commercial Research License), so they build from source locally and are never pushed to GHCR.'","source":"ai","title":"可选组件的许可证受限"},{"code":"pinned_dependencies_with_advisories","severity":"info","evidence":"README：'Its research-era dependency pins (requirements-ctrlregen.txt — e.g. transformers==4.37.2, diffusers==0.27.2) carry published advisories and are intentionally not current'，这些依赖仅安装在单独的虚拟环境（venv）中，供可选的 CtrlRegen 后端使用。","source":"ai","title":"锁定版本的依赖项带有已公布的漏洞公告"},{"code":"setup_scripts_download_third_party_code","severity":"info","evidence":"README：setup_synthid.sh、setup_ctrlregen.sh 和 setup_markllm.sh 各自会克隆上游仓库（'Clones upstream'）、创建 venv 并安装 torch 等软件包；模型会在首次运行时从 Hugging Face 下载。","source":"ai","title":"安装脚本会下载第三方代码"},{"code":"agent_hook_edits_files","severity":"info","evidence":"README：Claude Code 插件 'registers a PostToolUse hook on Write|Edit|MultiEdit|NotebookEdit'；在 'clean' 模式下，它会对智能体写入的文件 'Strips the marks in place'。","source":"ai","title":"智能体钩子会修改文件"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","ffmpeg"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["Hugging Face (HF_TOKEN, only for optional gated models)","OpenAI-compatible model API for text rewriting (WATERMARKS_REWRITE_API_KEY)","Password you choose for the local service (WATERMARKS_SERVER_API_KEY)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["productivity","image","ai_skills"],"alternativeTo":[{"slug":"undetectable-ai","name":"Undetectable AI","confirmed":true}],"declaredAlternatives":[{"target":"word","stance":"claims","evidence":"Any rewrite replaces the original word choices with the rewriting model's, which flattens tone, voice, and precision"},{"target":"invideo-ai","stance":"claims","evidence":"- Agent skill `remove-ai-marks` (replaces Claude-only `remove-claude-marks`)"}],"stars":27000,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":856,"growthDays":7,"history":[{"date":"2026-10-26","stars":24995},{"date":"2026-10-27","stars":25104},{"date":"2026-10-28","stars":25372},{"date":"2026-10-29","stars":25564},{"date":"2026-10-30","stars":25672},{"date":"2026-10-31","stars":25776},{"date":"2026-11-01","stars":25913},{"date":"2026-11-02","stars":26019},{"date":"2026-11-03","stars":26144},{"date":"2026-11-04","stars":26291},{"date":"2026-11-05","stars":26391},{"date":"2026-11-06","stars":26509},{"date":"2026-11-07","stars":26596},{"date":"2026-11-08","stars":26716},{"date":"2026-11-09","stars":26849},{"date":"2026-11-10","stars":27000}],"firstPublishedAt":"2026-10-27T01:10:12.632Z","commitSha":"5c180932a00c793261177f79aee21db8929f3fe2","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-01T12:46:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Removes hidden AI marks, like invisible characters and file metadata, from text, images and documents you own.","summary":"A Python service plus add-on skills for AI coding assistants (Claude Code, Codex, Cursor) that find and remove AI provenance marks: invisible Unicode characters in text, C2PA and EXIF metadata in images, PDFs and Office files, and, by rewording, statistical text watermarks. It is aimed at technical users. You need Python 3.10+ and a terminal. Text cleaning through the service also needs a rewrite model (local Ollama or an online API). Optional pixel-level image removal downloads about 10 GB of models and works best with an NVIDIA graphics card.","easeRationale":"There is no installer: you install Python 3.10+, run the skill installer script and start the local service from a terminal, and the heavier optional parts need Docker or an NVIDIA graphics card.","costNote":"If you choose an online rewrite model (the README gives DeepSeek as an example), that provider bills you separately.","notVerified":["Whether the Windows commands (py install_skill.py, py service/scripts/server.py) work as documented.","How well text watermark removal holds up against real vendor detectors; the README says no tool can certify a detector will fail.","How much the rewording changes the quality and meaning of cleaned text.","Real cost of an online rewrite model, and disk and graphics memory use of the optional ~10 GB pixel-removal models.","What the stealer/ folder and its prompt-corpus downloader send or fetch over the network."]},"zh-Hans":{"summaryShort":"清除你自己的文本、图片和文档中隐藏的 AI 标记,如不可见字符和文件元数据。","summary":"这是一个 Python 服务,外加适用于 AI 编程助手(Claude Code、Codex、Cursor)的扩展技能,用来查找并清除 AI 来源标记:文本中的不可见 Unicode 字符,图片、PDF 和 Office 文件中的 C2PA 与 EXIF 元数据,以及通过改写来处理的文本统计型水印。它面向技术用户。你需要 Python 3.10+ 和终端。通过该服务清理文本还需要一个改写模型(本地 Ollama 或在线 API)。可选的像素级图片清除会下载约 10 GB 的模型,配合 NVIDIA 显卡效果最好。","easeRationale":"没有安装程序:你需要安装 Python 3.10+,运行技能安装脚本,并在终端中启动本地服务;较重的可选部分需要 Docker 或 NVIDIA 显卡。","costNote":"如果你选择在线改写模型(README 以 DeepSeek 为例),该服务商会另行向你收费。","notVerified":["Windows 命令(py install_skill.py、py service/scripts/server.py)是否如文档所述可用。","文本水印清除面对真实厂商检测器时效果如何;README 称没有任何工具能证明检测器一定失效。","改写会在多大程度上改变清理后文本的质量和含义。","在线改写模型的实际费用,以及可选的约 10 GB 像素级清除模型占用的磁盘和显存。","stealer/ 文件夹及其提示词语料下载器会通过网络发送或获取什么。"]}}},{"id":"R_kgDOJN9KGA","slug":"h2oai__h2ogpt","name":"h2oGPT","owner":"h2oai","fullName":"h2oai/h2ogpt","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次release在2024-06-04(886天前)","source":"check","title":"release已过时"},{"code":"archived","severity":"info","evidence":"该仓库已被归档(只读,不再维护)","source":"check","title":"仓库已归档"},{"code":"known_vulnerabilities","severity":"attention","evidence":"存在已知漏洞的固定依赖项(OSV):gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-39mp-8hj3-5c49;torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p;torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p","source":"check","title":"已知漏洞"},{"code":"some_packages_come_from_the_company_bucket","severity":"attention","evidence":"requirements.txt直接从公司自己的存储而不是常规的Python索引安装了两个软件包:\"InstructorEmbedding @ https://h2o-release.s3.amazonaws.com/h2ogpt/InstructorEmbedding-1.0.1-py3-none-any.whl\"和\"sentence_transformers_old @ https://h2o-release.s3.amazonaws.com/h2ogpt/sentence_transformers_old-2.2.2-py3-none-any.whl\"。安装时这些地址提供的任何内容都会在你的电脑上运行,如果这些地址被下线,安装就会失败。","source":"ai","title":"部分软件包来自公司自己的存储桶"},{"code":"web_interface_published_without_a_password_by_default","severity":"attention","evidence":"compose文件公开了这个界面(\"ports: - '7860:7860'\"),Dockerfile还暴露了7860、8888、5000、5002和5004端口,而登录功能是可选的(\"Authentication in the UI by user/password via Native or Google OAuth\"),这些文件里并没有设置密码。在一台别人能访问到的机器上,你的文档和聊天内容就会对他们公开。","source":"ai","title":"默认情况下网页界面公开且没有密码"},{"code":"optional_extras_pull_gpl_like_packages","severity":"info","evidence":"安装过程把\"reqs_optional/requirements_optional_langchain.gpllike.txt\"列入setup.py拉取的依赖文件之中。项目自身的代码是Apache 2.0许可,但这些可选软件包采用了更严格的许可证,如果你打算在此基础上做商业化产品,这一点很重要。","source":"ai","title":"可选扩展会引入类gpl许可的软件包"},{"code":"old_pins_that_will_not_move","severity":"info","evidence":"依赖文件锁定在2024年前后的版本(\"gradio==4.44.0\"、\"torch==2.2.1\"、\"spacy==3.7.5\"、\"huggingface_hub==0.25.2\"),而仓库是只读的,所以被标记存在已知问题的版本没法在这里升级;需要修复的人得在自己的副本里自行修改。","source":"ai","title":"旧的版本锁定不会再更新"}],"platforms":["windows","mac","linux"],"tools":["python","docker","gpu_nvidia"],"buildTools":["python","docker","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["OpenAI, Azure OpenAI, Anthropic, Google, MistralAI, Groq, Replicate or Together.ai, only if you choose an online model instead of a local one"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["local_ai","office","productivity"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true},{"slug":"typingmind","name":"TypingMind","confirmed":true}],"declaredAlternatives":[],"stars":11924,"downloads":595,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":11926},{"date":"2026-11-09","stars":11925},{"date":"2026-11-10","stars":11924}],"firstPublishedAt":"2026-11-07T22:56:16.791Z","commitSha":"5af700f5bdd155816de33cf95a41ced8393c48dc","verifiedAt":"2026-11-07T22:45:49.525Z","lastCommitAt":"2025-10-09T23:29:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Chat with an AI about your own PDFs, Word files and images on your own machine. The repository is archived.","summary":"A chat program you run yourself: you upload documents (PDF, Word, Excel, images, audio, code, web pages) into a private database and ask questions or request summaries, with the answers coming from a model running on your own machine or from an online service you choose. It also does speech to text, reads answers aloud, generates images and can stand in for the OpenAI API. It runs on Linux, Windows and macOS, best through Docker, and wants an NVIDIA card for the heavier parts. The repository is archived, so nothing more will be fixed here.","easeRationale":"All three are a D. The README says \"Docker is recommended for Linux, Windows, and MAC for full capabilities\", which means Docker plus the NVIDIA runtime and a long build; the plain Linux script has the same capabilities but installs many Python pieces, and the README notes the \"Windows and MAC scripts have less capabilities than using Docker\". On top of that you download model files yourself and the releases contain no ready program.","costNote":"Apache 2.0, free to run. The company sells a separate enterprise product: the README's second heading is \"Try Enterprise Version for Free\" linking to Enterprise h2oGPTe. If you point it at a paid online model instead of a local one, you pay that service.","notVerified":["Whether the install still completes today, since the pinned 2024 versions have to agree with current Python, CUDA and Hugging Face packages.","Whether the Windows and macOS scripts still work, and which features are missing there compared with Docker.","Real disk space and video memory once you download models, embeddings and the image or voice extras.","How good the answers over your own documents are, and how much the paid online models would cost.","Whether the demo and enterprise addresses in the README are still up."]},"zh-Hans":{"summaryShort":"在自己的电脑上,就你自己的PDF、Word文件和图片与AI聊天。该仓库已被归档。","summary":"一个你自己运行的聊天程序:你把文档(PDF、Word、Excel、图片、音频、代码、网页)上传到一个私有数据库,然后提问或要求总结,回答来自运行在你自己电脑上的模型,或者你选择的在线服务。它还能把语音转成文字、朗读回答、生成图片,并且可以替代OpenAI的API。它可以在Linux、Windows和macOS上运行,最好通过Docker,较重的部分需要NVIDIA显卡。该仓库已被归档,所以这里不会再修复任何东西。","easeRationale":"三项评级都是D。README写着\"Docker is recommended for Linux, Windows, and MAC for full capabilities\",这意味着需要Docker加上NVIDIA运行时,并经历一次较长的构建;普通的Linux脚本功能相同,但要安装很多Python组件,README还指出\"Windows and MAC scripts have less capabilities than using Docker\"。除此之外,你还得自己下载模型文件,而且release里没有任何现成的程序。","costNote":"Apache 2.0许可,运行免费。公司另外销售一个企业版产品:README的第二个标题是\"Try Enterprise Version for Free\",链接到Enterprise h2oGPTe。如果你让它调用付费的在线模型而不是本地模型,你就要为那个服务付费。","notVerified":["安装今天是否仍能顺利完成,因为固定在2024年的版本必须和当前的Python、CUDA和Hugging Face软件包兼容。","Windows和macOS脚本是否仍然可用,以及相比Docker缺少哪些功能。","下载模型、嵌入模型以及图片或语音扩展功能之后,实际占用的磁盘空间和显存。","针对你自己文档给出的回答质量如何,以及付费在线模型大概要花多少钱。","README中演示地址和企业版地址是否仍然可以访问。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyODI4NjI3MDI=","slug":"ha6bots__twitch-clips-compilation-generator-tccg-","name":"Twitch Clips Compilation Generator","owner":"HA6Bots","fullName":"HA6Bots/Twitch-Clips-Compilation-Generator-TCCG-","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"high","signals":[{"code":"external_download_links","severity":"attention","evidence":"README链接到https://www.mediafire.com/file/x3zszb11lihvbnd/TCCG_Epic_Music_Assets.zip/file","source":"check","title":"外部下载链接"},{"code":"republishes_other_peoples_clips","severity":"attention","evidence":"整个项目的目的就是收集\"top highlight clips from Twitch for any category\",并输出\"the final mp4 video that you can upload to any platform\"。这些片段属于制作它们的主播:README只提供了\"A credits text file ... which contains links to all the livestreamers whos clips were used\",并表示\"There are no restrictions on downloading the clips\"。上传他人的片段,尤其是带广告的情况,可能会侵犯版权并违反平台条款。","source":"ai","title":"转发他人制作的片段"},{"code":"depends_on_retired_twitch_api","severity":"attention","evidence":"README的\"Automatic Downloader Notes\"部分把请求建立在`https://api.twitch.tv/kraken/clips/top?...`之上,并解释说\"We use the kraken API because it gives access to many more clips then helix\"。Twitch在2022年停用了kraken(v5)API,时间在这里最后一次提交(2021-11-13)之后,所以服务器中查找片段的那部分指向的是一个已经不存在的接口。","source":"ai","title":"依赖已停用的twitch api"},{"code":"development_stopped","severity":"attention","evidence":"在检查时这个快照里的最后一次提交是2021-11-13,距检查时已经1818天,没有任何发行版,还有7个未关闭的issue。该仓库有143个星标和19个fork。","source":"ai","title":"开发已经停止"},{"code":"unencrypted_ftp_with_example_passwords","severity":"attention","evidence":"这三个程序之间通过\"a combination HTTP and FTP servers\"互相通信,文档中的配置自带`ftp_user = VidGen`、`ftp_password = password`、`username = admin`、`password = password`,以及数据库密码为空的`databaseuser = root`。README还宣传了\"Full VPS Support\",这意味着把这些未加密的FTP和HTTP端口直接暴露在公网上。","source":"ai","title":"使用未加密ftp且密码为示例密码"},{"code":"assets_from_file_locker","severity":"attention","evidence":"这个音乐包以MediaFire链接的形式提供——\"[Epic Music (25 copyright free songs)](https://www.mediafire.com/file/x3zszb11lihvbnd/TCCG_Epic_Music_Assets.zip/file)\"——一个放在文件分享网站上、不在GitHub里的zip文件,这25首歌是否\"copyright free\"仅凭作者一人之言。","source":"ai","title":"素材来自文件托管网站"},{"code":"mass_upload_of_similar_videos","severity":"info","evidence":"README宣传的理念是,合集视频\"wrack up a lot of views and can quickly grow a channels follower base\",而账号系统意味着\"multiple people can use the bot at once\",可以开展\"a large scale operation\"。YouTube关于重复、批量生产内容的规定,适用于上传这些视频成果的人。","source":"ai","title":"大量上传相似视频"},{"code":"no_setup_guide","severity":"info","evidence":"README里有一个标题叫\"## Setting up the system\",下面却什么都没有:直接跳到数据库表、文件列表和配置文件,而\"Dependencies\"部分只是一份模块清单(requests、PyQt5、Pymediainfo、Opencv-python、Moviepy、Pyftpdlib、Pysubs2、Pydub、mysql),仓库里也没有requirements文件。","source":"ai","title":"没有安装指南"},{"code":"republishes_clips_made_by_other_streamers","severity":"high","evidence":"项目描述:\"A system of 3 programs that collects clips automatically from Twitch, lets you edit videos and combine clips, and puts them together into a compilation video ready to be uploaded straight to any social media platform\";README:\"Passively downloads and stores top highlight clips from Twitch for any category/categories\"以及\"generating the final mp4 video  that you can upload to any platform\"。这些片段是由主播们录制的:这个项目生成的唯一署名方式是\"A credits text file is also generated, which contains links to all the livestreamers whos clips were used in the video\",而关于下载,README写道\"There are no restrictions on downloading the clips\"。Twitch的条款规定了可以如何使用其服务中的片段,而YouTube和其他平台都要求上传视频的人拥有视频内容的权利。","source":"ai","title":"转发其他主播制作的片段"}],"platforms":["windows","linux"],"tools":["python","ffmpeg","other"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Twitch API (client_id and bearer)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","social_marketing","automation"],"alternativeTo":[{"slug":"streamladder","name":"StreamLadder","confirmed":true},{"slug":"opusclip","name":"OpusClip","confirmed":true}],"declaredAlternatives":[],"stars":144,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":1,"growthDays":6,"history":[{"date":"2026-11-06","stars":143},{"date":"2026-11-07","stars":144},{"date":"2026-11-08","stars":144},{"date":"2026-11-09","stars":144},{"date":"2026-11-10","stars":144}],"firstPublishedAt":"2026-11-05T22:43:48.437Z","commitSha":"a9b99a7e4374f6e1398e8be69413637c26557201","verifiedAt":"2026-11-05T22:29:45.358Z","lastCommitAt":"2021-11-13T15:49:43Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Three Python programs that collect top Twitch clips by game and glue them into a compilation video for YouTube.","summary":"A set of three Python programs with simple windows: one keeps downloading the top Twitch clips for the games you choose into a MySQL database, one lets you swipe through the clips keeping or skipping them, trimming and setting volume, and the third renders the chosen clips into one mp4 with intro, outro, transitions and background music. Made for people who publish gaming compilations. You install Python, MySQL and FFmpeg, get Twitch API credentials and fill in three config files. Last changed in 2021.","easeRationale":"Windows is the platform the author had in mind — \"Python (recommend 3.7+ and 64bit on windows)\" — and it is still a D: there is no installer and no release, you install Python and nine modules by hand, a MySQL server, FFmpeg \"added to system path\", then start three separate programs, let each write its config file, fill in Twitch credentials, FTP users and ports in all three, and restart them. Linux is the same work plus the VPS setup the README suggests for the downloader.","costNote":"No price (MIT). The README recommends renting a VPS to keep the clip downloader running, which costs money.","notVerified":["Whether the clip downloader still works at all, given the Twitch API it calls","Whether the three programs still run with current Python and module versions","How much disk the stored clips take, which depends on how long you leave the downloader running","Whether it runs on a Mac, which the project never mentions","What the MediaFire zip really contains and the true licence of those songs","Whether the videos it makes are accepted by YouTube and the other platforms"]},"zh-Hans":{"summaryShort":"三个Python程序,按游戏收集Twitch上最热门的片段,并把它们拼接成一段给YouTube用的合集视频。","summary":"一套由三个带简单窗口的Python程序组成的工具:一个持续把你选定游戏里Twitch上最热门的片段下载到MySQL数据库中;一个让你逐个浏览这些片段,选择保留或跳过、裁剪并调整音量;第三个把选中的片段拼接成一个mp4文件,带有片头、片尾、转场和背景音乐。这是为发布游戏集锦视频的人做的。你需要安装Python、MySQL和FFmpeg,获取Twitch API凭证,并填写三个配置文件。最后一次更新是在2021年。","easeRationale":"作者设想的平台是Windows——\"Python (recommend 3.7+ and 64bit on windows)\"——难度评级仍然是D:没有安装程序,也没有发行版,你需要手动安装Python和九个模块、一个MySQL服务器、把FFmpeg\"added to system path\",然后分别启动三个独立的程序,让每个程序各自写出配置文件,在三个程序里都填好Twitch凭证、FTP用户和端口,再重启它们。在Linux上同样要做这些事,还要加上README为下载器建议的VPS配置。","costNote":"没有价格(MIT)。README建议租用一台VPS来保持片段下载器持续运行,这需要花钱。","notVerified":["考虑到它调用的Twitch API,这个片段下载器是否还能正常工作","在当前版本的Python和各个模块下,这三个程序是否还能运行","存储的片段占用多少磁盘空间,这取决于你让下载器运行了多久","是否能在Mac上运行,项目完全没有提到","MediaFire上的zip文件实际包含什么内容,那些歌曲的真实授权情况如何","它生成的视频是否能被YouTube和其他平台接受"]}}},{"id":"R_kgDOO4hq-g","slug":"hangwin__mcp-chrome","name":"Chrome MCP Server","owner":"hangwin","fullName":"hangwin/mcp-chrome","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"ai_gets_your_logged_in_browser","severity":"attention","evidence":"README的宣传点是\"Use Your Original Browser: Seamlessly integrate with your existing browser environment (your configurations, login states, etc.)\"以及\"✅ Automatically uses existing login state\"。工具列表包括`chrome_history`、`chrome_bookmark_delete`、`chrome_inject_script`、`chrome_fill_or_select`、`chrome_network_request`,以及带有\"with response bodies\"的`chrome_network_debugger_start`——所以不管助手决定做什么,它都是以你的身份去做的,在你的银行、邮箱和社交账户里面。","source":"ai","title":"ai会获得你已登录状态的浏览器"},{"code":"local_server_without_authentication","severity":"attention","evidence":"推荐的设置让客户端连接到`http://127.0.0.1:12306/mcp`,没有任何令牌,而\"Future Roadmap\"里仍然把\"[ ] Authentication\"列为未完成。同一台电脑上运行的任何程序都能向这个端口发送命令并操控浏览器。","source":"ai","title":"本地服务器没有身份验证"},{"code":"extension_installed_outside_chrome_web_store","severity":"attention","evidence":"安装的第一步是\"Download the latest Chrome extension from GitHub\",第三步是启用\"Developer mode\"并\"Load unpacked\"。这个扩展不经过Chrome网上应用店的审核或更新,所以你需要自己重新下载来更新它。","source":"ai","title":"扩展安装渠道不是chrome网上应用店"},{"code":"early_stage_and_slowed_down","severity":"attention","evidence":"README.md:\"The project is still in its early stages and is under intensive development.\" 在检查时的这个快照里,最后一次提交是2026-01-06,距检查时303天,有237个未关闭的issue,而星标数是12537。","source":"ai","title":"项目尚处早期阶段,进展变慢"},{"code":"asks_you_to_allow_install_scripts","severity":"info","evidence":"pnpm的安装说明要求运行`pnpm config set enable-pre-post-scripts true`,README自己也解释了原因:\"pnpm v7+ disables postinstall scripts by default for security.\" 这个包就是通过那个脚本把自己注册到Chrome的原生消息通信里的。","source":"ai","title":"要求允许安装脚本运行"},{"code":"examples_inspect_other_sites_internals","severity":"info","evidence":"使用示例包括\"Help me modify the current page's style and remove advertisements\"和\"I want to know what the search API for Xiaohongshu is and what the response structure looks like\",这两个例子都是通过脚本注入和网络抓取工具演示的。","source":"ai","title":"示例会查看其他网站的内部细节"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","ai_agents","automation"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":12539,"downloads":93258,"downloadsWeek":277,"downloadsDays":4,"growth":4,"growthDays":6,"history":[{"date":"2026-11-06","stars":12536},{"date":"2026-11-07","stars":12536},{"date":"2026-11-08","stars":12539},{"date":"2026-11-09","stars":12540},{"date":"2026-11-10","stars":12539}],"firstPublishedAt":"2026-11-05T22:43:48.445Z","commitSha":"f48e71751e00bc09725c7e173423cff4f2ccd12a","verifiedAt":"2026-11-05T22:30:36.580Z","lastCommitAt":"2026-01-06T13:05:39Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant drive the Chrome you already use, with your tabs, logins, history and bookmarks.","summary":"A Chrome extension plus a small local program that hands your everyday browser to an AI assistant such as Claude. With more than 20 tools the assistant can open and close tabs, click, fill forms, type, take screenshots, read page text, search your history and bookmarks, inject scripts into pages and record network traffic. Unlike other browser automation it uses your real Chrome, so the AI is already logged in to everything you are. You install Node, a global npm package and the extension by hand.","easeRationale":"A C everywhere, Mac included, and the steps are the same on all three: install Node 20 or newer, run `npm install -g mcp-chrome-bridge`, download the extension from the releases page, turn on Developer mode in `chrome://extensions/`, \"Load unpacked\", then paste a JSON block into your AI client's configuration. No installer, several manual steps, and for the stdio route you also have to look up the package's path by hand.","costNote":"No price (MIT) and no key of its own. You pay whatever your AI assistant costs, and the semantic search runs on your own computer.","notVerified":["What the extension and the local bridge actually send and where, beyond the README's claim of being \"Fully Local\"","Whether anything stops the AI from acting on a page that asks it to, for example a web page with hidden instructions","Whether the extension still works with current Chrome versions after 303 days without commits","Real disk space and memory used by the built-in vector database over time","Whether the 20+ tools work as described on each operating system"]},"zh-Hans":{"summaryShort":"让AI助手操作你正在用的Chrome,包括你的标签页、登录状态、历史记录和收藏夹。","summary":"一个Chrome扩展加上一个小型本地程序,把你日常使用的浏览器交给像Claude这样的AI助手。凭借20多种工具,助手可以打开和关闭标签页、点击、填写表单、输入文字、截图、读取页面文本、搜索你的历史记录和收藏夹、向页面注入脚本并记录网络流量。与其他浏览器自动化方式不同,它使用的是你真实的Chrome,所以AI已经以你的身份登录了一切。你需要手动安装Node、一个全局npm包和这个扩展。","easeRationale":"三个系统的难度评级都是C,Mac也一样,步骤在三者上都相同:安装Node 20或更高版本,运行`npm install -g mcp-chrome-bridge`,从发行版页面下载扩展,在`chrome://extensions/`里开启开发者模式,\"Load unpacked\"加载扩展,然后把一段JSON粘贴到你的AI客户端配置里。没有安装程序,有好几个手动步骤,如果走stdio这条路,你还要自己去查包的路径。","costNote":"没有价格(MIT),也没有自己的密钥。你要支付的是你的AI助手本身的费用,语义搜索则在你自己的电脑上运行。","notVerified":["除了README所说的\"Fully Local\"之外,这个扩展和本地桥接程序实际发送了什么内容、发到哪里","有没有什么机制能阻止AI按照页面的要求去行动,比如一个藏有隐藏指令的网页","在303天没有提交的情况下,这个扩展在当前版本的Chrome上是否还能正常工作","内置的向量数据库随着时间推移实际占用多少磁盘空间和内存","这20多种工具在每种操作系统上是否都如描述那样工作"]}}},{"id":"R_kgDOKDcJiA","slug":"haoheliu__audioldm2","name":"AudioLDM 2","owner":"haoheliu","fullName":"haoheliu/AudioLDM2","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-37q5-v5qm-c9v8","source":"check","title":"已知漏洞"},{"code":"license_conflict_noncommercial_vs_mit","severity":"attention","evidence":"LICENSE 文件是 \"Attribution-NonCommercial-ShareAlike 4.0 International\"，其中写道 \"NonCommercial means not primarily intended for or directed towards commercial advantage or monetary compensation\"，而 setup.py 却设置了 `license=\"MIT\"` 以及分类标签 \"License :: OSI Approved :: MIT License\"。GitHub 显示 NOASSERTION（无法判定），所以没有任何工具能判断你用它制作的内容适用哪一个。","source":"ai","title":"许可证冲突：非商业 vs mit"},{"code":"development_stopped","severity":"attention","evidence":"被检查的快照中最后一次提交是 2024-09-29，距检查日已过去 767 天，没有任何 release，有 75 个未关闭的 issue，TODO 列表里最后一项（\"Add the style-transfer and inpainting code for the audioldm_48k checkpoint\"）仍未勾选，旁边写着 \"PR welcomed\"。","source":"ai","title":"开发已经停止"},{"code":"pinned_old_python_stack","severity":"attention","evidence":"安装说明要求 `python=3.8`，setup.py 锁定了 `transformers==4.30.2`、`numpy<=1.23.5`、`librosa==0.9.2` 和 `gradio<4`。requirements.txt 还设置了 `--extra-index-url https://download.pytorch.org/whl/cu113`，这是一个 CUDA 11.3 的构建版本。这些版本已经有好几年历史，被锁定的 transformers 正是检查中标记出已知漏洞的那个版本。","source":"ai","title":"锁定了陈旧的 python 环境"},{"code":"installs_unpinned_code_from_git","severity":"info","evidence":"requirements.txt 开头就是 `git+https://github.com/huggingface/diffusers.git`，Diffusers 章节还让你运行 `pip install --upgrade git+https://github.com/huggingface/diffusers.git transformers accelerate`。这会安装你安装当天那些分支上的任意内容，而不是固定版本。","source":"ai","title":"从 git 安装未锁定版本的代码"},{"code":"output_quality_varies_with_hardware","severity":"info","evidence":"README 的 \"Random Seed Matters\" 一节写道：\"Sometimes model may not perform well (sounds weird or low quality) when changing into a different hardware. In this case, please adjust the random seed and find the optimal one for your hardware.\" 命令行工具还提供每段文字生成 `-n` 个候选结果，称为 \"Automatic quality control\"。","source":"ai","title":"输出质量因硬件而异"},{"code":"heavy_on_a_mac","severity":"info","evidence":"README 在设备部分写道：\"mps ( Notice that the computation requires about 20GB of RAM. )\" ——也就是说在 Apple Silicon Mac 上，你需要一台内存很大的机器。","source":"ai","title":"在 mac 上很吃资源"},{"code":"faster_path_lives_elsewhere","severity":"info","evidence":"README 写道：\"The Diffusers version of the code runs upwards of **3x faster** than the native AudioLDM 2 implementation, and supports generating audios of arbitrary length\"，并指向 Hugging Face 上的 `cvssp/audioldm2` checkpoint。这个仓库里的代码是速度较慢的原始实现。","source":"ai","title":"更快的路径在别处"}],"platforms":["windows","mac","linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"suno","name":"Suno","confirmed":true},{"slug":"epidemic-sound","name":"Epidemic Sound","confirmed":true}],"declaredAlternatives":[],"stars":2644,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":1,"growthDays":6,"history":[{"date":"2026-11-06","stars":2644},{"date":"2026-11-07","stars":2644},{"date":"2026-11-08","stars":2644},{"date":"2026-11-09","stars":2645},{"date":"2026-11-10","stars":2644}],"firstPublishedAt":"2026-11-05T22:43:48.453Z","commitSha":"b5786c5dc0ae8f766337fdc1b67ab6046586d14d","verifiedAt":"2026-11-05T22:31:47.765Z","lastCommitAt":"2024-09-29T15:05:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Writes a description and gets back a sound effect, a short piece of music or a spoken line, on your own computer.","summary":"A model that turns a written description into audio: sound effects, short music, or speech from a transcription plus a description of the speaker. It also upgrades the quality of an existing recording. You can try it free on the authors' Hugging Face pages, or install it with Python and run either a local web page or a one-line command. Seven different model files are available, including a 48 kHz high-fidelity one. It is research code from a published paper, last changed in 2024.","easeRationale":"The easiest way in is the web, an A: the README links two free Hugging Face Spaces run by the author where you type a description and listen, with nothing installed. Installing it is a C on all three systems: create a Python 3.8 environment with conda, `pip3 install git+https://github.com/haoheliu/AudioLDM2.git`, then either `python3 app.py` for a local web page or `audioldm2 -t \"...\"` in the terminal. It can use an NVIDIA card (cuda), an Apple Silicon Mac (mps) or plain cpu, and text-to-speech also needs espeak installed.","costNote":"No price, but the LICENSE file is Creative Commons Attribution-NonCommercial-ShareAlike 4.0, which does not allow commercial use — while setup.py declares MIT.","notVerified":["Whether the install still works today with Python 3.8 and those pinned versions","Whether the two Hugging Face demo pages are still online","How the generated sound effects, music and speech actually sound","Real disk space taken by the seven model files you may download","How long one clip takes on a plain processor, without a graphics card","Which licence applies to audio you generate, given the conflict between LICENSE and setup.py"]},"zh-Hans":{"summaryShort":"写一段描述，就能得到一个音效、一段短音乐或一句口语配音，全部在你自己的电脑上完成。","summary":"一个把文字描述转换成音频的模型：音效、短音乐，或者根据一段转录文字加上说话人描述生成的语音。它还能提升已有录音的质量。你可以在作者的 Hugging Face 页面上免费试用，也可以用 Python 安装后运行本地网页或一条命令。提供七个不同的模型文件，其中包括一个 48 kHz 高保真版本。这是一篇已发表论文附带的研究代码，最后一次更改是在 2024 年。","easeRationale":"最简单的方式是网页端，评级 A：README 链接了作者运营的两个免费 Hugging Face Spaces，你只需输入描述并试听，无需安装任何东西。安装则是三个系统都是 C 级：用 conda 创建一个 Python 3.8 环境，执行 `pip3 install git+https://github.com/haoheliu/AudioLDM2.git`，然后用 `python3 app.py` 打开本地网页，或在终端里用 `audioldm2 -t \"...\"`。它可以使用 NVIDIA 显卡（cuda）、Apple Silicon Mac（mps）或纯 cpu，文字转语音还需要安装 espeak。","costNote":"没有标价，但 LICENSE 文件是 Creative Commons Attribution-NonCommercial-ShareAlike 4.0，不允许商业使用——而 setup.py 却声明为 MIT。","notVerified":["在 Python 3.8 和这些锁定版本下，今天安装是否还能成功","那两个 Hugging Face 演示页面是否仍在线","生成的音效、音乐和语音实际听起来怎样","你可能下载的七个模型文件实际占用多少磁盘空间","在没有显卡、只用普通处理器时，生成一段音频需要多长时间","由于 LICENSE 和 setup.py 的说法冲突，你生成的音频到底适用哪种许可"]}}},{"id":"R_kgDOG9tUGA","slug":"harlanhong__cvpr2022-dagan","name":"DaGAN","owner":"harlanhong","fullName":"harlanhong/CVPR2022-DaGAN","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"linux":"D","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"存在已知漏洞的固定依赖项(OSV):certifi: GHSA-248v-346w-9cwc, GHSA-43fp-rhv2-5gv8, GHSA-6q48-6c7w-9gw2;fonttools: GHSA-6673-4983-2vx5, GHSA-768j-98cg-p3fv, PYSEC-2026-1388;grpcio: GHSA-496j-2rq6-j6cc, PYSEC-2026-1424;joblib: GHSA-6hrg-qmvc-2xh8, PYSEC-2022-288;markdown: GHSA-5wmx-573v-2qwq, PYSEC-2026-89;numpy: GHSA-fpfv-jqm9-f5jm;pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h;protobuf: GHSA-7gcm-g887-7qv7, GHSA-8gq9-2x98-w8hf, GHSA-8qvm-5x2c-j2w7;scikit-learn: GHSA-jw8x-6495-233v, PYSEC-2024-110;scipy: PYSEC-2023-102, PYSEC-2023-114;tqdm: GHSA-g7vv-2v7x-gj9p, PYSEC-2026-1976;werkzeug: GHSA-29vq-49wr-vm6x, GHSA-2g68-c3qc-8985, GHSA-87hc-h4r5-73f7;zipp: GHSA-jfmj-5v4g-7637, PYSEC-2026-2074","source":"check","title":"已知漏洞"},{"code":"license_file_noncommercial","severity":"attention","evidence":"仓库中的LICENSE.md文件是\"Attribution-NonCommercial 4.0 International\"(知识共享协议),只授予\"for NonCommercial purposes only\"的使用权,而GitHub显示没有明确的许可证(NOASSERTION)。README写道:\"If you have any question or collaboration need (research purpose or commercial purpose), please email fhongac@cse.ust.hk\"。","source":"ai","title":"许可证文件为非商业用途"},{"code":"generates_deepfake_faces","severity":"attention","evidence":"该项目让照片中真实人物的脸说话,并模仿另一个人的头部动作。仓库的主题标签包括\"deepfake\"和\"face-reenactment\",描述写着\"Talking Head Video Generation\"。这样可以制作出某人从未说过那些话的视频,而多个国家对未经本人同意制作的人物影像有所限制。","source":"ai","title":"生成deepfake换脸视频"},{"code":"model_from_outside_file_sharing","severity":"info","evidence":"模型文件不在仓库里:README指向hkustconnect-my.sharepoint.com上的个人OneDrive文件夹,获取\"The pre-trained checkpoint of face depth network and our DaGAN checkpoints\",所以下载内容并没有和代码一起做版本管理。","source":"ai","title":"模型来自外部文件分享服务"},{"code":"no_longer_updated","severity":"info","evidence":"最后一次提交是在2023-12-04,也没有GitHub release。README指向同一作者更新的项目MCNet(ICCV2023),称其为更方便的选择。","source":"ai","title":"已不再更新"}],"platforms":["linux","web"],"tools":["python","gpu_any","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["video","image"],"alternativeTo":[{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"viggle","name":"Viggle","confirmed":true}],"declaredAlternatives":[],"stars":997,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":997},{"date":"2026-11-09","stars":997},{"date":"2026-11-10","stars":997}],"firstPublishedAt":"2026-11-07T22:56:16.798Z","commitSha":"052ad359a019492893c098e8717dc993cdf846ed","verifiedAt":"2026-11-07T22:39:13.652Z","lastCommitAt":"2023-12-04T12:14:44Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that makes a face in a photo talk and move, copying the head motion of a driving video.","summary":"DaGAN is the research code behind a CVPR 2022 paper on talking head video generation. You give it one photo of a face and a short video of someone talking; it produces a video where the face in the photo copies that head motion and expression. It is aimed at researchers and curious technical users: you clone the repository, install Python packages, download a model checkpoint from a OneDrive link, crop the inputs with ffmpeg and run a command. The README also links a free Hugging Face demo page that needs nothing installed.","easeRationale":"On the web the README links a ready Hugging Face demo page, so you only open it and upload a photo and a video. Running it yourself is rated for Linux, the system the README recommends: you install Python packages, build the face-alignment library, fetch a checkpoint from an outside OneDrive link, crop the video with ffmpeg and run a command line, and a GPU is advised.","costNote":"No paid service is involved. The LICENSE.md file is Creative Commons Attribution-NonCommercial 4.0, so commercial use is not granted; the README asks you to email the author about commercial use.","notVerified":["Whether the Hugging Face demo page linked in the README is still online and working","Whether the pinned package versions in requirements.txt still install on a current Python and GPU driver (the file pins torchaudio==0.10.1+rocm4.1)","How well the result looks on your own photo, and how long one video takes on a machine without an NVIDIA GPU","Real disk use after downloading the depth model and the DaGAN checkpoints from the OneDrive links","Whether the OneDrive links still hold the files"]},"zh-Hans":{"summaryShort":"研究代码,让照片里的一张脸说话并做出动作,模仿驱动视频里的头部动作。","summary":"DaGAN是CVPR 2022一篇关于说话人头像视频生成论文背后的研究代码。你给它一张人脸照片和一段某人说话的短视频,它就会生成一段视频,让照片里的脸模仿那段视频的头部动作和表情。它面向研究人员和好奇的技术用户:你需要克隆仓库、安装Python软件包、从OneDrive链接下载模型checkpoint、用ffmpeg裁剪输入内容,然后运行一条命令。README还链接了一个无需安装的免费Hugging Face演示页面。","easeRationale":"在网页上,README链接了一个现成的Hugging Face演示页面,你只需要打开它并上传一张照片和一段视频。自己运行的评级是针对Linux的,这也是README推荐的系统:你需要安装Python软件包、编译face-alignment库、从外部OneDrive链接获取一个checkpoint、用ffmpeg裁剪视频、运行一条命令行,并建议使用GPU。","costNote":"不涉及任何付费服务。LICENSE.md文件采用Creative Commons Attribution-NonCommercial 4.0许可,因此不授予商业用途;README要求你就商业用途给作者发邮件。","notVerified":["README中链接的Hugging Face演示页面是否仍然在线并可以使用。","requirements.txt中固定的软件包版本是否仍能在当前的Python和GPU驱动上安装(文件锁定了torchaudio==0.10.1+rocm4.1)。","用你自己的照片效果如何,以及在没有NVIDIA显卡的机器上生成一段视频要花多长时间。","下载深度模型和DaGAN的checkpoint(来自OneDrive链接)之后,实际占用的磁盘空间。","这些OneDrive链接是否仍然保留着这些文件。"]}}},{"id":"R_kgDOLeediw","slug":"harry0703__moneyprinterturbo","name":"MoneyPrinterTurbo","owner":"harry0703","fullName":"harry0703/MoneyPrinterTurbo","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：python-multipart：GHSA-v9pg-7xvm-68hf、PYSEC-2026-3040","source":"check","title":"存在已知漏洞"},{"code":"bundled_music_unclear_rights","severity":"attention","evidence":"README.md（背景音乐部分）：\"当前项目里面放了一些默认的音乐，来自于 YouTube 视频，如有侵权，请删除。\"（resource/songs 里默认的背景音乐来自 YouTube 视频，如果侵权请自行删除）。用这些歌曲发布的视频可能会收到版权投诉。","source":"ai","title":"附带音乐的版权归属不明确"},{"code":"sponsor_affiliate_links","severity":"info","evidence":"README.md 有一大段付费 AI API 平台的赞助商内容，带有推荐链接，例如 \"https://platform.kimi.com?track_id=track-6eec1e56a4494e52adcaebbcbbefce59&aff=moneyprinterturbo\" 和 \"https://infistar.cc/register?aff=6T4EYXP2\"。功能介绍部分的模型列表也链接到这些赞助商。","source":"ai","title":"赞助商联盟链接"},{"code":"updates_code_at_run_time","severity":"info","evidence":"README.md：\"下载后，建议先双击执行 update.bat 更新到最新代码，然后双击 start.bat 启动\"。你最终运行的代码可能和发布的压缩包不一样。","source":"ai","title":"运行时会更新代码"},{"code":"agent_installs_from_remote_skill","severity":"info","evidence":"README.md 建议把远程文件 \"https://raw.githubusercontent.com/harry0703/MoneyPrinterTurbo/main/docs/skill/SKILL.md\" 交给一个 AI 代理；代理随后会自行在你的终端里安装、配置并运行这个项目。","source":"ai","title":"由代理从远程 skill 自行安装"},{"code":"docker_build_uses_third_party_mirrors","severity":"info","evidence":"Dockerfile 默认设置 ARG DOCKER_BUILD_MIRROR=china，本地构建时会写入阿里云和清华的 Debian 源，并运行 \"uv pip install --system --no-cache --default-index https://mirrors.aliyun.com/pypi/simple/\"。docker-compose.release.yml 使用的预构建镜像则是用官方 PyPI 构建的。","source":"ai","title":"docker 构建使用第三方镜像源"},{"code":"similar_name_package_on_pypi","severity":"info","evidence":"README.md：\"官方 PyPI 包名为 mpt，由 harry0703 维护。PyPI 上名为 MoneyPrinterTurbo 的包由第三方维护。\"","source":"ai","title":"pypi 上有同名近似的包"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":876,"maxMB":2627,"basis":"installer"},"apiKeys":["AI model provider for scripts (OpenAI, Kimi, DeepSeek, Gemini, Claude and others; not needed with local Ollama or a script you write yourself)","Pexels or Pixabay (free stock footage; not needed if you only use your own clips)","Paid voice services (Azure Speech, ElevenLabs, MiniMax, Fish Audio, ModelBest VoxCPM and others)","AI video generation services (MiniMax H3, Seedance, WaveSpeed, OFox, MuAPI)","TikTok, Instagram or YouTube accounts for automatic upload"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","social_marketing","automation"],"alternativeTo":[{"slug":"pictory","name":"Pictory","confirmed":true},{"slug":"invideo-ai","name":"AI","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":false}],"declaredAlternatives":[],"stars":141951,"downloads":174272,"downloadsWeek":10347,"downloadsDays":7,"growth":3629,"growthDays":7,"history":[{"date":"2026-10-26","stars":135481},{"date":"2026-10-27","stars":135856},{"date":"2026-10-28","stars":136132},{"date":"2026-10-29","stars":136310},{"date":"2026-10-30","stars":136616},{"date":"2026-10-31","stars":136917},{"date":"2026-11-01","stars":137189},{"date":"2026-11-02","stars":137525},{"date":"2026-11-03","stars":138322},{"date":"2026-11-04","stars":138853},{"date":"2026-11-05","stars":139731},{"date":"2026-11-06","stars":140225},{"date":"2026-11-07","stars":140712},{"date":"2026-11-08","stars":141015},{"date":"2026-11-09","stars":141564},{"date":"2026-11-10","stars":141951}],"firstPublishedAt":"2026-10-27T01:10:12.634Z","commitSha":"bd29c9ae817f28ce30d9ec819f8b4f054fd0a58e","verifiedAt":"2026-10-31T01:38:14.401Z","lastCommitAt":"2026-11-10T02:52:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Type a topic and it writes a script, finds clips, adds a voice, subtitles and music, and builds a short video.","summary":"MoneyPrinterTurbo turns a topic or keyword into a finished short video for TikTok, YouTube Shorts or Reels. An AI model writes the script, stock footage comes from Pexels or Pixabay (or your own clips), and it adds a voice-over, subtitles and background music. It can also upload the result. On Windows you unzip an 876 MB portable package and run start.bat; on Mac and Linux you install it with Python. You need a key for an AI model service (or a local model through Ollama) and a free stock-footage key. The basic Edge TTS voice needs no key.","easeRationale":"Windows is the easiest: an 876 MB portable package you unzip (in a folder path without spaces or special characters) and start with start.bat, then enter an AI model key and a stock-footage key in the web interface. Mac and Linux have no ready package, so you install Python 3.11, create a virtual environment and run a few commands (pip install mpt, mpt webui).","costNote":"The program is free. Script writing uses an AI model service, usually paid per use (a local model through Ollama is an option). Pexels and Pixabay keys are free, and the Edge TTS voice needs no key. AI-generated video clips from the listed services are paid.","notVerified":["Whether the portable Windows package starts cleanly and what update.bat changes after download.","Real cost per video with each AI model, voice and video service.","Disk use after the optional Whisper subtitle model downloads (README says about 3 GB for large-v3, 1.6 GB for large-v3-turbo).","Whether automatic upload to TikTok, Instagram and YouTube Shorts works and follows those platforms' rules.","The license of the bundled background music and of the stock clips that end up in the final videos.","How good and how accurate the AI-written scripts and the matched footage are."]},"zh-Hans":{"summaryShort":"输入一个主题，它就会写脚本、找素材、配音、加字幕和音乐，做出一个短视频。","summary":"MoneyPrinterTurbo 能把一个主题或关键词变成一个可以直接发到 TikTok、YouTube Shorts 或 Reels 的成品短视频。脚本由 AI 模型撰写，素材来自 Pexels 或 Pixabay（也可以用你自己的素材），并会加上配音、字幕和背景音乐，还能把成品上传出去。在 Windows 上，你解压一个 876 MB 的便携包并运行 start.bat；在 Mac 和 Linux 上则用 Python 安装。你需要一个 AI 模型服务的密钥（或通过 Ollama 使用本地模型），以及一个免费的素材网站密钥。基础的 Edge TTS 配音不需要密钥。","easeRationale":"Windows 上最简单：解压一个 876 MB 的便携包（放在不含空格或特殊字符的文件夹路径下），用 start.bat 启动，再在网页界面里填入 AI 模型密钥和素材网站密钥。Mac 和 Linux 没有现成的安装包，所以要先安装 Python 3.11，创建一个虚拟环境，再运行几条命令（pip install mpt、mpt webui）。","costNote":"这个程序本身免费。写脚本要用 AI 模型服务，通常按使用量付费（也可以通过 Ollama 用本地模型）。Pexels 和 Pixabay 的密钥是免费的，Edge TTS 配音也不需要密钥。所列服务里 AI 生成的视频素材是收费的。","notVerified":["Windows 便携包能否顺利启动，下载后 update.bat 会改动什么。","每种 AI 模型、配音和视频服务的实际使用成本。","下载可选的 Whisper 字幕模型后占用多少磁盘空间（README 说 large-v3 约 3 GB，large-v3-turbo 约 1.6 GB）。","自动上传到 TikTok、Instagram 和 YouTube Shorts 是否有效，以及是否符合这些平台的规则。","随附的背景音乐和最终视频里所用素材的版权情况。","AI 写的脚本质量如何、配的素材是否贴切。"]}}},{"id":"R_kgDOTnjzQQ","slug":"harshithgangone__open-autodm","name":"Open AutoDM","owner":"harshithgangone","fullName":"harshithgangone/open-autodm","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"web":"D"},"risk":"attention","signals":[{"code":"public_triggers_need_meta_review","severity":"attention","evidence":"README：\"in Development mode your app already delivers real events for the accounts on your app (you + testers you add)... When you want the **general public's** comments to trigger your automations, submit Meta's free **App Review** once (a short screencast of your working flow - typically approved in days).\" 在 Meta 批准你的应用之前，陌生人的评论不会触发任何动作。","source":"ai","title":"公开触发需要 meta 审核"},{"code":"account_risk_is_yours","severity":"attention","evidence":"README 免责声明：作者 \"**not liable** for anything that happens through your use of it, including (but not limited to) actions Meta or Instagram takes on your account or app\"，并且 \"**you** are responsible for how you use automation and for your own compliance with Instagram's Terms of Use and Meta's Platform Policies.\" 代码里额外设置了自己的上限（180 DMs/hour，低于 Meta 的 200）以及针对 Meta 368 错误的断路保护。","source":"ai","title":"账号风险由你自己承担"},{"code":"very_new_single_author_project","severity":"attention","evidence":"这个仓库创建于 2026-07-29，只有 20 个星标，没有任何 release，README 是以第一人称由一个人写的（\"Built with ❤️ by @buildwharsha\"）。快照中的最后一次提交是 2026-10-05，距检查日 31 天。","source":"ai","title":"非常新、由一人维护的项目"},{"code":"author_sells_setup_and_asks_for_stars","severity":"info","evidence":"README：\"**⭐ Star this repo now** - it takes 2 seconds, helps other creators find a free alternative to $25/month tools\" 以及 \"Don't want to set it up yourself? I'll deploy the whole system on your own accounts... for a minimal one-time setup charge, for all your Instagram accounts. DM me on Instagram: @buildwharsha\"。","source":"ai","title":"作者出售配置服务并索要星标"},{"code":"security_claims_are_self_described","severity":"info","evidence":"README 声称令牌使用 AES-256-GCM 加密、webhook 通过 HMAC 做 timing-safe 比较来验证、状态 JWT 有效期 10 分钟、所有表都启用了 Row Level Security，并且 \"ownership is checked in SQL on every query\"。这些都只是作者自己的说法，代码从未被人审计过，一旦丢失 TOKEN_ENCRYPTION_KEY，你就 \"must reconnect Instagram and re-run setup\"。","source":"ai","title":"防护说明均为作者自述"},{"code":"stores_data_of_people_who_comment","severity":"info","evidence":"应用保存了 \"contacts, and analytics... scoped per account\"，一个转化漏斗 \"(trigger → delivered → button tap → follow gained)\"，以及 `{username}` 个性化信息。这些都是其他人的 Instagram 账号和互动信息，被存进了你的数据库。","source":"ai","title":"保存了评论者的数据"}],"platforms":["web"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta developer app (Instagram API)","Supabase","Vercel or Cloudflare"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":20,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":20},{"date":"2026-11-07","stars":20},{"date":"2026-11-08","stars":20},{"date":"2026-11-09","stars":20},{"date":"2026-11-10","stars":20}],"firstPublishedAt":"2026-11-05T22:43:48.461Z","commitSha":"78bc5410108e382e62963f8c6641d7fb8301ce74","verifiedAt":"2026-11-05T22:32:38.161Z","lastCommitAt":"2026-10-05T12:55:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Someone comments a keyword on your Instagram post and this sends them your link by DM, on your own free cloud accounts.","summary":"A self-hosted version of the \"comment LINK to get it\" funnel: when someone comments a keyword on your post, replies to your story or sends you a keyword in a DM, the app answers with your link or a card, optionally after asking them to follow you. It also shows a conversion funnel and charts. You deploy one Next.js app to Vercel or Cloudflare, use your own free Supabase database and your own Meta developer app, then follow a setup wizard in the app. It only talks to Instagram through Meta's official API.","easeRationale":"It lives on the web — you deploy it once and then use it in a browser — and getting there is a D: clone the repository, `npm install`, install the Supabase command line tool, `supabase link` and `supabase db push`, fill in six values in a `.env` file including a key you generate with a node command, run `npx vercel`, create your own login row in the Supabase dashboard, and then work through the in-app wizard that creates the Meta app. The README estimates \"~15 minutes\" for the wizard part alone.","costNote":"MIT and free to run on the free tiers of Supabase and Vercel or Cloudflare; those tiers can be outgrown. The author offers to set it up for you \"for a minimal one-time setup charge\" and takes GitHub sponsorships.","notVerified":["Whether a full install really works end to end, since there are no releases and few users","Whether Meta approves app reviews for this kind of use, and how long it takes","Whether the rate limits and delays in the code are enough to keep an account out of trouble","Whether the free tiers of Supabase and Vercel or Cloudflare hold up at any volume","Whether the encryption and permission claims in the README match the code","What the deployment at open-autodm.vercel.app is: a demo, the author's own instance, or both"]},"zh-Hans":{"summaryShort":"有人在你的 Instagram 帖子下评论一个关键词，就会通过私信收到你的链接，全部用你自己的免费云账户完成。","summary":"一个自托管版本的 \"comment LINK to get it\" 引流方式：当有人在你的帖子下评论关键词、回复你的快拍，或者通过私信发给你一个关键词时，应用会回复你的链接或一张卡片，也可以先要求对方关注你。它还会显示转化漏斗和图表。你把一个 Next.js 应用部署到 Vercel 或 Cloudflare，使用你自己的免费 Supabase 数据库和你自己的 Meta 开发者应用，然后在应用里按设置向导操作。它只通过 Meta 的官方 API 和 Instagram 通信。","easeRationale":"它运行在网页上——你只需部署一次，之后在浏览器里使用——但要走到这一步评级是 D：克隆仓库，`npm install`，安装 Supabase 命令行工具，执行 `supabase link` 和 `supabase db push`，在 `.env` 文件里填六个值（包括一个用 node 命令生成的密钥），运行 `npx vercel`，在 Supabase 控制台里创建你自己的登录记录，然后完成应用内创建 Meta 应用的设置向导。README 估计单独完成向导部分需要 \"~15 minutes\"。","costNote":"MIT 许可，在 Supabase 和 Vercel 或 Cloudflare 的免费额度上运行免费；但这些免费额度可能会用超。作者表示可以 \"for a minimal one-time setup charge\" 帮你配置，并接受 GitHub 赞助。","notVerified":["由于没有 release 且用户很少，完整安装是否真的能从头跑到尾","Meta 是否会批准这种用途的应用审核，需要多长时间","代码里的速率限制和延迟是否足以让账号不出问题","Supabase 和 Vercel 或 Cloudflare 的免费额度在稍大的使用量下是否撑得住","README 里关于加密和权限的说法是否与代码一致","open-autodm.vercel.app 这个部署到底是演示、作者自己的实例，还是两者都是"]}}},{"id":"R_kgDORD-p-g","slug":"hbai-ltd__toonflow-app","name":"Toonflow","owner":"HBAI-Ltd","fullName":"HBAI-Ltd/Toonflow-app","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"D"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接管道传给shell执行：curl -fsSL https://bun.com/install | bash","source":"check","title":"通过curl管道执行到shell"},{"code":"self_hosted_server_without_login","severity":"attention","evidence":"README.md第4.4节：\"当前服务使用固定端口 3000，业务页面与 API 没有独立登录鉴权\"，并建议用防火墙屏蔽3000端口，联网访问时使用带身份验证的反向代理。如果你就这样把它跑在服务器上，任何能访问到这个端口的人都能使用它，并读取你的项目。","source":"ai","title":"自托管服务器没有登录验证"},{"code":"advises_disabling_macos_protection","severity":"attention","evidence":"README.md的macOS部分建议，作为最后手段，先执行`sudo xattr -rd com.apple.quarantine /Applications/toonflow.app`，再执行`sudo spctl --master-disable`。第二条命令会关闭macOS对整台机器上所有程序（不只是这个）的已下载应用检查。","source":"ai","title":"建议关闭macOS的保护机制"},{"code":"plugins_downloaded_from_authors_server","severity":"info","evidence":"README.md里链接了一个插件市场 https://api.toonflow.net/console/plugIn，功能亮点表里把\"插件市场\"和MCP列为扩展方式，这意味着额外的代码可以从作者自己的服务器安装进应用里。","source":"ai","title":"插件从作者的服务器下载"},{"code":"license_changed_twice","severity":"info","evidence":"README.md第8节：v1.0.8之前的版本是AGPL-3.0，v1.0.8到v1.1.8是Apache-2.0\"与附加协议\"，现在的许可证是MIT。条款在一年内改了两次。","source":"ai","title":"许可证改了两次"},{"code":"readme_promotes_paid_api_sponsors","severity":"info","evidence":"README.md第1节列出了付费模型API的赞助商（秘塔/MiniMax H3、APIMart、优云智算），附带推荐链接和免费额度，所以推荐的服务商是该项目的商业合作伙伴。","source":"ai","title":"README里推广付费API的赞助商"},{"code":"very_fast_growth_few_issues","severity":"info","evidence":"GitHub数据：创建于2026-01-29，已经有20339个star和3654个fork，但只有3个未解决的issue。这么多fork对应这么多star，再加上几乎没有未解决的issue，这不是一个九个月项目通常会有的模式。","source":"ai","title":"增长非常快但issue很少"},{"code":"docs_hosted_on_external_doc_service","severity":"info","evidence":"README.md里教程和开发者文档的链接指向的是飞书文档（qcn7xdsqgc4z.feishu.cn），而不是仓库里的文件，所以这些说明可能在没有任何commit的情况下被修改或消失。","source":"ai","title":"文档托管在外部文档服务上"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","docker","ffmpeg"],"disk":{"minMB":53,"maxMB":158,"basis":"installer"},"apiKeys":["An image and video model service of your choice (the authors' api.toonflow.net, APIMart, MiniMax and others are suggested in the README)","A language model service of your choice, for script help"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","image","ai_agents"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"ltx-studio","name":"LTX Studio","confirmed":true},{"slug":null,"name":"LibTV","confirmed":false},{"slug":null,"name":"TapNow","confirmed":false}],"declaredAlternatives":[],"stars":21368,"downloads":168970,"downloadsWeek":15513,"downloadsDays":6,"growth":899,"growthDays":6,"history":[{"date":"2026-11-04","stars":20469},{"date":"2026-11-05","stars":20767},{"date":"2026-11-06","stars":20920},{"date":"2026-11-07","stars":21019},{"date":"2026-11-08","stars":21089},{"date":"2026-11-09","stars":21244},{"date":"2026-11-10","stars":21368}],"firstPublishedAt":"2026-11-04T01:41:48.323Z","commitSha":"ecd5325c7f9b38721491382ee6952fa2318b4693","verifiedAt":"2026-11-04T01:22:42.626Z","lastCommitAt":"2026-11-05T07:49:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Make short AI videos and motion comics on one canvas: script, characters, images and video clips, using paid model services.","summary":"Toonflow is a canvas studio for short AI dramas and motion comics: you write the story and the shot list in text boxes, generate character and scene pictures, then turn each shot into a video clip, all on one endless canvas. It does not include the AI models: you connect outside image, video and language model services and pay them per generation, or point it at your own ComfyUI. There are installers for Windows and macOS, an Android APK, and a Docker or Linux server version. The interface has 21 languages; the guides are mostly in Chinese.","easeRationale":"On Windows there is a ready installer (toonflow-2.0.9-Setup.exe, 53 MB) that also installs WebView2, and on macOS a .dmg you drag into Applications; after that you still have to paste in model API keys before anything can be generated, which is the one extra step. Linux is D: there is no Linux installer, so you clone the repository and either run `docker compose up -d --build` or install Bun, FFmpeg and build the server yourself.","costNote":"The program is free (MIT). The models are not: the README's own example says one 2-minute video cost about ¥130 (around US$18) in model calls — ¥120 of it for video.","notVerified":["Whether the desktop installers bundle FFmpeg and everything else needed to export video","Whether the Windows and macOS builds are signed, since the README documents Gatekeeper warnings","Real disk use once models' output, assets and plugins pile up in the data folder","What the app and its plugins send to api.toonflow.net","The real cost per finished video, which depends entirely on the model provider you choose","Whether the Android APK and the \"device link\" feature work on a given phone"]},"zh-Hans":{"summaryShort":"在同一个画布上制作AI短视频和动态漫画：剧本、角色、图片和视频片段，使用付费的模型服务。","summary":"Toonflow是一个用于AI短剧和动态漫画的画布工作室：你在文本框里写故事和分镜列表，生成角色和场景图片，再把每个镜头变成视频片段，这些都在一块无限延展的画布上完成。它本身不包含AI模型：你需要连接外部的图像、视频和语言模型服务，按生成次数付费，或者指向你自己的ComfyUI。它有Windows和macOS的安装程序、一个Android APK，以及Docker或Linux服务器版本。界面支持21种语言，但使用指南大多是中文。","easeRationale":"在Windows上有现成的安装程序（toonflow-2.0.9-Setup.exe，53 MB），它还会安装WebView2；在macOS上是一个拖进“应用程序”文件夹的.dmg；装好之后你还需要粘贴模型的API密钥才能开始生成，这算是多出来的一步。Linux是D级：没有Linux安装程序，你需要克隆仓库，然后运行`docker compose up -d --build`，或者自己安装Bun、FFmpeg并编译服务器。","costNote":"程序本身是免费的（MIT）。模型不是：README自己的例子说，一段2分钟的视频在模型调用上花了大约130元人民币（约18美元），其中120元是视频部分。","notVerified":["桌面安装程序是否自带FFmpeg以及导出视频所需的一切","Windows和macOS的构建是否经过签名，因为README里记录了Gatekeeper的警告","当模型输出、素材和插件在数据文件夹里堆积后的实际磁盘占用","应用和它的插件向api.toonflow.net发送了什么内容","每段成片的实际成本，这完全取决于你选择的模型服务商","Android APK和\"设备联动\"功能在某台具体手机上是否能用"]}}},{"id":"R_kgDOQ1mH3Q","slug":"headroomlabs-ai__headroom","name":"Headroom","owner":"headroomlabs-ai","fullName":"headroomlabs-ai/headroom","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接管道传给shell执行：curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh","source":"check","title":"通过curl管道执行到shell"},{"code":"telemetry_on_by_default","severity":"attention","evidence":"README.md的\"Telemetry\"部分：\"An anonymous beacon is on by default. It reports how compression behaved: ratios, counters, provider and model IDs, OS and architecture.\" 它说提示词和代码永远不会被发送，你可以用HEADROOM_BEACON=off、HEADROOM_TELEMETRY=off、DO_NOT_TRACK=1或HEADROOM_OFFLINE=1关闭它。","source":"ai","title":"默认开启遥测"},{"code":"sees_all_prompts_and_provider_tokens","severity":"attention","evidence":"按设计，每个发往模型的请求都要经过这个本地代理，对于Copilot，它还负责处理凭据：README.md，\"headroom copilot-auth login stores a Headroom-specific Copilot OAuth token\"，以及\"The short-lived upstream Copilot token is held only in the proxy process.\" 任何能读取该进程的东西都能读取你的提示词和令牌。","source":"ai","title":"能看到所有提示词和服务商的令牌"},{"code":"open_data_plane_if_exposed","severity":"attention","evidence":"docker-compose.yml文件开头：\"None of these three services authenticates inbound callers by default: the proxy's /v1/* data plane is open unless HEADROOM_PROXY_TOKEN is set, Qdrant has no API key... Publishing them on 0.0.0.0 therefore hands any peer on your network a relay through the proxy plus direct read/write on the embeddings and graph derived from your prompts.\" 默认提供的文件把所有服务都绑定在127.0.0.1上，以避免这种情况。","source":"ai","title":"暴露后数据层是开放的"},{"code":"edits_other_tools_configuration","severity":"attention","evidence":"README.md：`headroom wrap`会把Serena作为MCP服务器\"installs Serena for semantic code navigation\"安装进`~/.claude.json`，为OpenCode、Cline和Continue\"injects config\"，用`--1m`写入Claude Code的模型选择器，而`headroom learn`会把修正写进CLAUDE.local.md、CLAUDE.md、AGENTS.md、GEMINI.md或GROK.md。有一个`unwrap`命令可以撤销这些改动。","source":"ai","title":"修改其他工具的配置"},{"code":"savings_depend_heavily_on_content","severity":"info","evidence":"仓库简介宣称\"60-95% fewer tokens for JSON\"，而README.md自己的四个基准测试场景显示的是21%、57%、42%和30%，文中还补充说：\"prose and already-dense output compress very little\"。31.7%的输出token数字被工具自己标记为\"[estimated]\"。","source":"ai","title":"节省幅度高度依赖内容类型"},{"code":"free_version_deliberately_weaker","severity":"info","evidence":"README.md的\"Headroom for teams\"部分：\"Headroom OSS is a framework. It was built to show what's possible on one developer's laptop, so it saves less than it could. We are building bigger savings for teams and enterprises in Headroom Enterprise.\"","source":"ai","title":"免费版被刻意限制得更弱"},{"code":"beta_with_very_fast_releases","severity":"info","evidence":"pyproject.toml把它标记为\"Development Status :: 4 - Beta\"，版本号是0.42.0，而GitHub显示这个创建于2026-01-07的仓库已经有175次发行，471个未解决的issue。事情每周都在变化。","source":"ai","title":"测试版本但发行非常频繁"},{"code":"downloads_models_and_binaries_at_install_or_run","severity":"info","evidence":"README.md的\"Corporate networks\"部分列出了它会自行访问的两个下载源：cdn.pyke.io，用于Rust核心所用的ONNX Runtime；huggingface.co，用于kompress-base模型。文中还提到ONNX相关功能在x86上需要支持AVX2的处理器，没有的话会退回到更简单的方法。","source":"ai","title":"安装或运行时下载模型和二进制文件"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","gpu_any","rust"],"disk":{"minMB":15,"maxMB":45,"basis":"installer"},"apiKeys":["Your own model provider (Anthropic, OpenAI, Bedrock and others) — the key you already use passes through the proxy","Other (HEADROOM_PROXY_TOKEN) — a password you make up, needed only if you open the proxy to your network","GitHub (only for the Copilot subscription mode)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_skills"],"alternativeTo":[{"slug":"compresr","name":"Compresr","confirmed":true},{"slug":"the-token-company","name":"The Token Company","confirmed":true}],"declaredAlternatives":[],"stars":79308,"downloads":17193,"downloadsWeek":1439,"downloadsDays":6,"growth":557,"growthDays":7,"history":[{"date":"2026-11-04","stars":78847},{"date":"2026-11-05","stars":78926},{"date":"2026-11-06","stars":79018},{"date":"2026-11-07","stars":79076},{"date":"2026-11-08","stars":79150},{"date":"2026-11-09","stars":79242},{"date":"2026-11-10","stars":79308}],"firstPublishedAt":"2026-11-04T01:36:50.430Z","commitSha":"fe69f381168035ea6527bb7b908d11eeeb01838f","verifiedAt":"2026-11-04T01:23:49.577Z","lastCommitAt":"2026-11-10T18:35:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Shrinks the text an AI coding agent sends to the model, so you pay for fewer tokens and get the same answers.","summary":"Headroom sits between your AI coding agent and the model and squeezes what passes through: tool output, logs, files, search results and chat history. You can use it as a Python or TypeScript library, as a local proxy your agent points at, or as an MCP server inside Claude Code, Codex, Cursor and others. The compressing happens on your own machine. It is a developer tool: you install it with pip or uv and run commands. The repository's own benchmarks show 21% to 57% fewer tokens, depending on how repetitive the text is.","easeRationale":"On all three systems you need Python 3.11 or newer and the terminal: `pip install \"headroom-ai[all]\"` (or `uv tool install`), then commands like `headroom wrap claude` or `headroom proxy --port 8787`. Ready wheels exist for Windows, Linux and Apple Silicon, so nothing has to be compiled; on Intel Macs the README sends you to the Docker install instead, which is harder.","costNote":"Apache-2.0 and free to use. The same company sells a closed \"Headroom Enterprise\" with more savings; you still pay your own model provider for whatever tokens are left.","notVerified":["Whether the answers really stay the same on your own prompts and tools","What the anonymous beacon sends in practice","Real disk use after the models, the local original-text cache (CCR) and the memory databases fill up","How much money it actually saves on your bill, which depends on your workload","Whether every listed agent still works with the current version, given the release pace"]},"zh-Hans":{"summaryShort":"压缩AI编程助手发送给模型的文本，让你为更少的token付费，同时得到一样的回答。","summary":"Headroom位于你的AI编程助手和模型之间，压缩所有经过的内容：工具输出、日志、文件、搜索结果和聊天记录。你可以把它当作Python或TypeScript库使用，当作你的助手指向的本地代理，或者作为Claude Code、Codex、Cursor等工具里的MCP服务器。压缩过程在你自己的机器上完成。这是一个开发者工具：用pip或uv安装，然后运行命令。仓库自己的基准测试显示，根据文本的重复程度不同，能减少21%到57%的token。","easeRationale":"在这三个系统上你都需要Python 3.11或更新版本，以及终端：`pip install \"headroom-ai[all]\"`（或`uv tool install`），然后运行像`headroom wrap claude`或`headroom proxy --port 8787`这样的命令。Windows、Linux和Apple Silicon都有现成的wheel包，不需要编译；在Intel芯片的Mac上，README会让你改用Docker安装，这更难一些。","costNote":"Apache-2.0许可，免费使用。同一家公司出售一个闭源的\"Headroom Enterprise\"，能省得更多；剩下的token你仍然需要向你自己的模型服务商付费。","notVerified":["在你自己的提示词和工具上，回答是否真的保持不变","匿名信标实际上发送了什么内容","模型、本地原文缓存（CCR）和记忆数据库都填满之后的实际磁盘占用","它实际上能在你的账单上省多少钱，这取决于你的工作负载","考虑到发行节奏，列出的每个助手是否都还能配合当前版本使用"]}}},{"id":"R_kgDOQ6ODjQ","slug":"heartmula__heartlib","name":"HeartMuLa","owner":"HeartMuLa","fullName":"HeartMuLa/heartlib","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"needs_a_strong_nvidia_card","severity":"attention","evidence":"README 的 FAQ 标题是\"CUDA out of memory?\",建议把模型拆分到两块显卡上(\"e.g. 2 4090s\"),使用 `--mula_device cuda:0 --codec_device cuda:1`,或者在单卡上用 `--lazy_load true`。脚本默认使用 `cuda`,所以没有为 Mac 或没有 NVIDIA 显卡的电脑提供可用方案。","source":"ai","title":"需要性能强的 nvidia 显卡"},{"code":"slow_generation","severity":"info","evidence":"待办事项列表:\"Release scripts for inference acceleration and streaming inference. The current inference speed is around RTF ≈ 1.0\"——大致是每一分钟音乐需要一分钟的计算时间。","source":"ai","title":"生成速度慢"},{"code":"headline_result_not_released","severity":"info","evidence":"亮点部分写道:\"Our latest internal version of HeartMuLa-7B achieves comparable performance with Suno\",而参数列表里关于 7B 选项却写着\"# `7B` version not released yet.\"仓库的简介也把它称为\"The Most Powerful Open-Source Music Generation Model of 2026\"。","source":"ai","title":"主打成果尚未发布"},{"code":"early_stage_packaging","severity":"info","evidence":"GitHub 上完全没有发布版本,pyproject.toml 里 version = \"0.1.0\",描述为\"A Python Library.\",安装方式是从主分支 `git clone` 加上 `pip install -e .`,所以你拿到的始终是当时最新的代码。","source":"ai","title":"打包方式仍处于早期阶段"}],"platforms":["windows","linux","web"],"tools":[],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":3790,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":7,"growthDays":6,"history":[{"date":"2026-11-05","stars":3785},{"date":"2026-11-06","stars":3786},{"date":"2026-11-07","stars":3787},{"date":"2026-11-08","stars":3788},{"date":"2026-11-09","stars":3789},{"date":"2026-11-10","stars":3790}],"firstPublishedAt":"2026-11-04T22:38:15.205Z","commitSha":"5858ca8f1ffe58d62be7c007ea55f1033c5ca456","verifiedAt":"2026-11-04T22:17:07.507Z","lastCommitAt":"2026-10-08T06:38:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your lyrics plus a few style words into a full song with singing, on your own computer.","summary":"HeartMuLa is a family of open music models from MuLa Labs. The released 3B model writes a song with vocals from a lyrics file and a list of tags such as \"piano,happy,wedding\"; other pieces transcribe lyrics from audio and rebuild 48 kHz stereo sound. You can try the team's hosted demo page in a browser. To run it yourself you need Python, command lines and an NVIDIA graphics card. The README's comparison with Suno is about an internal 7B model that is not released, so what you can run today is the smaller 3B.","easeRationale":"The easy way in is the team's hosted demo on Hugging Face and ModelScope, which needs nothing installed, so web is an A. Running it yourself is command line only on Windows and Linux: git clone, `pip install -e .`, then download several model checkpoints by hand and launch a Python script that expects CUDA devices, which is a D.","costNote":"Apache-2.0 licence, no account or payment. The model files are downloaded free from Hugging Face or ModelScope.","notVerified":["How good the songs sound compared with the demo page","Real disk space after downloading the HeartMuLa and HeartCodec checkpoints","How much video memory a single card actually needs, since the README gives no number","Whether the hosted demo pages stay available and how long the queue is","Who owns the rights to the music you generate, since the repository only states the Apache-2.0 licence of the code and weights"]},"zh-Hans":{"summaryShort":"把你的歌词加上几个风格词,在你自己的电脑上做成一首带演唱的完整歌曲。","summary":"HeartMuLa 是 MuLa Labs 旗下的一系列开放音乐模型。已发布的 3B 模型可以根据歌词文件和一组标签(比如\"piano,happy,wedding\")写出一首带人声的歌;其他部分能从音频中转写歌词,并重建 48 kHz 立体声音频。你可以在浏览器里试用团队提供的在线演示页面。要自己运行它,你需要 Python、命令行操作和一块 NVIDIA 显卡。README 里和 Suno 的对比说的是尚未发布的内部 7B 模型,所以你今天能运行的只是更小的 3B。","easeRationale":"最简单的入门方式是团队在 Hugging Face 和 ModelScope 上提供的在线演示,什么都不用安装,所以网页端评级为 A。自己运行则只能在 Windows 和 Linux 上通过命令行完成:git clone、`pip install -e .`,然后手动下载好几个模型检查点,再运行一个要求有 CUDA 设备的 Python 脚本,这属于 D 级。","costNote":"采用 Apache-2.0 许可证,不需要账号也不需要付费。模型文件可以从 Hugging Face 或 ModelScope 免费下载。","notVerified":["歌曲的效果和演示页面相比怎么样","下载完 HeartMuLa 和 HeartCodec 的检查点后,真实占用的磁盘空间","单块显卡实际需要多少显存,因为 README 没有给出具体数字","在线演示页面能否一直保持可用,排队要等多久","你生成的音乐的权利归谁,因为仓库只说明了代码和权重采用 Apache-2.0 许可证"]}}},{"id":"R_kgDONCHD-g","slug":"henry-23__videochat","name":"VideoChat","owner":"Henry-23","fullName":"Henry-23/VideoChat","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：gradio: GHSA-39mp-8hj3-5c49, GHSA-5cpq-9538-jm2j, GHSA-6655-8ph2-63j3; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; gdown: GHSA-76hw-p97h-883f, PYSEC-2026-2158; requests: GHSA-9hjg-9r4m-mvj7, GHSA-gc5v-m9x4-r6x2, PYSEC-2026-1872; pytorch-lightning: GHSA-75m9-98v2-hjpm, PYSEC-2026-3043, PYSEC-2026-3967; sentencepiece: GHSA-38vq-g6vr-w8wf, PYSEC-2026-1909; opencc: GHSA-7fqq-q52p-2jjg; modelscope: GHSA-fhhq-h4hg-549x, PYSEC-2026-2663; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; deepspeed: GHSA-8cp5-3rf8-8gfh, PYSEC-2024-109; hydra-core: GHSA-2cp2-2r3c-7p7r, GHSA-c3wx-c55w-pxjq, PYSEC-2026-3850; lightning: GHSA-cgwc-qvrx-rf7f, GHSA-mr7h-w2qc-ffc2, GHSA-qqmf-gpg7-g8gw; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8qvm-5x2c-j2w7, PYSEC-2026-1805; requests: GHSA-9hjg-9r4m-mvj7, GHSA-gc5v-m9x4-r6x2, PYSEC-2026-1872","source":"check","title":"已知漏洞"},{"code":"voice_and_face_cloning_can_impersonate","severity":"attention","evidence":"项目介绍承诺 \"可自定义形象与音色，支持音色克隆\"。第 5.1 节把 \"录制好的数字人形象视频\" ——也就是一段录制好的人物视频——当作形象素材，5.2 节则用一段 \"3-10s\" 的参考片段克隆声音。仓库里没有任何地方询问视频或录音中的人是否同意，也没有任何免责声明。","source":"ai","title":"声音和人脸克隆可能被用来冒充他人"},{"code":"development_slowed","severity":"attention","evidence":"被检查快照中的最后一次提交是 2025-12-18，距检查日 322 天，没有任何 release，有 25 个未关闭的 issue。README 自己也提醒 \"requirements中的版本仅供参考，最新依赖请查看对应模块的Repo说明\" ——锁定的版本只是参考，要查看每个组件自己的仓库说明。","source":"ai","title":"开发节奏变慢"},{"code":"configuration_means_editing_source_files","severity":"attention","evidence":"API 密钥要写进 `app.py` 第 14 行（`os.environ[\"DASHSCOPE_API_KEY\"] = \"INPUT YOUR API-KEY HERE\"`），下载 GLM-4-Voice 需要把 `snapshot_download(...)` 调用粘贴进 `app.py`，换一个新形象要修改 `src/thg.py` 里的 `avatar_list`，不想用密钥则要删掉 `src/tts.py` 里 `CosyVoice_API` 的部分。没有任何配置文件。","source":"ai","title":"配置意味着要修改源代码文件"},{"code":"weights_and_code_from_modelscope","severity":"info","evidence":"推荐的安装方式是 `git clone https://www.modelscope.cn/studios/AI-ModelScope/video_chat.git` ——代码和被跟踪的权重文件来自 ModelScope（一个中国的模型平台），而不是这个 GitHub 仓库本身。手动下载指向的是 MuseTalk、GPT-SoVITS 以及 ZhipuAI 的 `glm-4-voice-*` 模型。","source":"ai","title":"权重和代码来自 modelscope"},{"code":"parts_carry_their_own_licences","severity":"info","evidence":"这个仓库本身是 MIT 许可，但它只是把别的项目粘合在一起：FunASR、Qwen、GLM-4-Voice、GPT-SoVITS、CosyVoice、edge-tts 和 MuseTalk。这些模型和它们的权重各有各的条款，决定了你能用生成的形象和音频做什么。","source":"ai","title":"各个组成部分有各自的许可证"},{"code":"free_voice_route_uses_a_browser_service","severity":"info","evidence":"3.2.2 节建议，在没有 API 密钥的情况下，\"使用`Edge_TTS`调用Edge浏览器的免费TTS服务\" ——通过 `edge-tts` 包调用 Edge 浏览器内置的免费文字转语音服务，这是微软为自家浏览器提供的服务，而不是给其他程序用的。","source":"ai","title":"免费语音路线使用的是浏览器自带服务"},{"code":"known_rough_edges","severity":"info","evidence":"README 的 \"已知问题\" 部分列出了资源缺失的报错、因权重下载不完整导致的模型加载失败，以及 \"右侧视频流播放卡顿：需等待Gradio优化Video Streaming效果\"。","source":"ai","title":"已知的粗糙问题"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":1810,"maxMB":6870,"basis":"heuristic"},"apiKeys":["Alibaba Cloud DashScope (Qwen and CosyVoice)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","audio","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":false}],"declaredAlternatives":[],"stars":1331,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":1331},{"date":"2026-11-07","stars":1331},{"date":"2026-11-08","stars":1331},{"date":"2026-11-09","stars":1331},{"date":"2026-11-10","stars":1331}],"firstPublishedAt":"2026-11-05T22:43:48.470Z","commitSha":"303681e69ef1e29c7de5ca1a56435c89ea6f7f67","verifiedAt":"2026-11-05T22:33:40.269Z","lastCommitAt":"2025-12-18T02:50:33Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Hold a spoken conversation with a digital person on screen whose face lip-syncs and answers out loud.","summary":"A demo you run yourself that turns a recorded video of a person into a talking avatar you can speak with: it listens to your microphone, answers with a language model and replies out loud while the face lip-syncs, with the first audio in about 3 seconds. You can add your own avatar video and clone a voice from a 3 to 10 second sample. It is assembled from FunASR, Qwen, GLM-4-Voice, GPT-SoVITS, CosyVoice and MuseTalk, needs Ubuntu, Python 3.10, CUDA and an NVIDIA card with 8 to 20 GB of memory, and the main instructions are in Chinese.","easeRationale":"Linux is the only documented platform and it is a D: the README asks for \"ubuntu 22.04, python 3.10, CUDA 12.2\", a conda environment, `pip install -r requirements.txt` with around a hundred pinned packages, a `git lfs` clone from ModelScope to get the model weights, and an API key pasted into line 14 of `app.py` before `python app.py` opens the local page. The graphics card requirement is roughly 8 GB for the cascade mode and 20 GB for the end-to-end mode, measured on an A100.","costNote":"The code is MIT and free. By default the language model and the voice come from Alibaba Cloud's paid Qwen and CosyVoice APIs; to run everything locally instead you have to edit the source files and provide a much bigger graphics card.","notVerified":["Whether the install still works, since the pinned versions are from 2024 and the README says to check each component itself","How the conversation really feels on a smaller card than the A100 used for the 3 and 7 second figures","What Alibaba Cloud charges for the Qwen and CosyVoice calls in the default setup","Real disk space taken by all the model weights","Whether it can be made to run on Windows or a Mac, which the instructions never cover","How good the lip-sync and the cloned voice are with your own video and audio"]},"zh-Hans":{"summaryShort":"和屏幕上的数字人进行语音对话，对方会对口型并大声回答。","summary":"一个你自己运行的演示程序，把一段录制好的人物视频变成可以对话的说话人形象：它会听你的麦克风说话，用语言模型生成回答，并一边对口型一边大声念出来，首个音频大约 3 秒就能出现。你可以添加自己的形象视频，还能用 3 到 10 秒的样本克隆一个声音。它由 FunASR、Qwen、GLM-4-Voice、GPT-SoVITS、CosyVoice 和 MuseTalk 拼装而成，需要 Ubuntu、Python 3.10、CUDA 以及一块 8 到 20 GB 显存的 NVIDIA 显卡，主要说明是中文的。","easeRationale":"Linux 是唯一有文档说明的平台，评级是 D：README 要求 \"ubuntu 22.04, python 3.10, CUDA 12.2\"，一个 conda 环境，`pip install -r requirements.txt` 安装大约上百个锁定版本的包，用 `git lfs` 从 ModelScope 克隆来获取模型权重，还要把 API 密钥粘贴到 `app.py` 的第 14 行，然后才能运行 `python app.py` 打开本地页面。显卡需求大约是级联模式 8 GB、端到端模式 20 GB，这是在 A100 上测得的。","costNote":"代码是 MIT 许可，免费。默认情况下语言模型和语音都来自阿里云的付费 Qwen 和 CosyVoice API；想完全本地运行就得修改源代码文件，并配备一块大得多的显卡。","notVerified":["由于锁定的版本是 2024 年的，README 也说要自行查看每个组件，安装现在是否还能正常进行尚不确定","在比测出 3 秒和 7 秒数据所用的 A100 更小的显卡上，对话体验到底如何","默认配置下调用 Qwen 和 CosyVoice，阿里云实际收费是多少","所有模型权重实际占用多少磁盘空间","说明文档从未提到的 Windows 或 Mac 上是否能跑起来","用你自己的视频和音频，对口型效果和克隆出的声音到底有多好"]}}},{"id":"R_kgDOVb-mOA","slug":"heydamir__unslop","name":"unslop","owner":"heydamir","fullName":"heydamir/unslop","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"4 天内获得 6330 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"content_license_requires_credit","severity":"info","evidence":"README.md License：“CC BY 4.0. Use it, change it and share it, including commercially, as long as you credit Damir Sokolovsky (@heydamir) and link back to this repository.”这是一种为内容设计的 Creative Commons 许可证，不是常见的软件许可证，带有署名义务。","source":"ai","title":"内容许可证要求署名"},{"code":"edits_your_project_files","severity":"info","evidence":"README.md：“fixes the screen pass by pass like a designer doing a redline review”，以及“It audits the screen and proposes a direction before it edits anything, so you can steer it early.”这个技能会重写你界面的源文件，所以最好先保存好尚未提交的工作。","source":"ai","title":"会修改你项目里的文件"},{"code":"can_connect_paid_image_services","severity":"info","evidence":"README.md Visuals：“it can generate one through an image service like fal, Replicate, OpenAI images or Higgsfield. If none is connected, it offers to connect one as an MCP server, and it shows the prompt and asks before every paid generation.”","source":"ai","title":"可以连接付费的图片生成服务"},{"code":"brand_new_repository","severity":"info","evidence":"GitHub 数据：这个仓库创建于 2026-10-29，最后一次提交是 2026-11-01，已经有 6330 个 star。目前还没有发布版本，也没有修复记录。","source":"ai","title":"全新的仓库"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg","other"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Claude Code or Codex (a paid plan or API use for the model behind the agent)","An image generation service (fal, Replicate, OpenAI images or Higgsfield) connected as an MCP server"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"CC-BY-4.0","category":"other","detectedFrom":"github"},"categories":["ai_skills","design_cad","development"],"alternativeTo":[{"slug":"v0-by-vercel","name":"v0 by Vercel","confirmed":true},{"slug":"claude-design","name":"Claude Design","confirmed":true},{"slug":"figma","name":"Figma","confirmed":false}],"declaredAlternatives":[],"stars":13630,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":6104,"growthDays":7,"history":[{"date":"2026-10-31","stars":3495},{"date":"2026-11-01","stars":4931},{"date":"2026-11-02","stars":6278},{"date":"2026-11-03","stars":7526},{"date":"2026-11-04","stars":8522},{"date":"2026-11-05","stars":9682},{"date":"2026-11-06","stars":10663},{"date":"2026-11-07","stars":11297},{"date":"2026-11-08","stars":11894},{"date":"2026-11-09","stars":12682},{"date":"2026-11-10","stars":13630}],"firstPublishedAt":"2026-10-31T01:54:32.860Z","commitSha":"a6f121eee5f29c2f74976b75456feae8b66afb3a","verifiedAt":"2026-11-03T02:55:14.765Z","lastCommitAt":"2026-11-08T21:16:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill for Claude Code and Codex that redesigns AI-looking pages and dashboards: it fixes spacing, color, icons and copy.","summary":"unslop is one skill folder you copy into Claude Code or Codex. You point it at a screen you built with AI help, or even at a live URL, and it lists what makes it look machine-made (purple gradients, glows, emoji labels, shouty text, uneven spacing), proposes a direction taken from your product, then edits the screen pass by pass: spacing, color, icons, copy, type, layout, details, visuals and motion. It works with Next.js, React, Vue, Svelte, Tailwind or plain HTML and CSS. It is for people who write front-end code and already use Claude Code or Codex.","easeRationale":"The same on all three systems: you need Claude Code or Codex installed, then clone the repository and copy one folder into ~/.claude/skills (or ~/.agents/skills for Codex, which can also fetch it with $skill-installer), and type /unslop or $unslop. The optional story video also needs Playwright and ffmpeg.","costNote":"The skill is free under CC BY 4.0, with credit required. It runs inside Claude Code or Codex, which need a paid plan or API use, and the optional hero images are generated by paid services such as fal, Replicate, OpenAI images or Higgsfield.","notVerified":["Whether the result actually looks better on your own screens, which is a matter of taste.","Whether it works as well on frameworks other than the ones the README names.","How much Claude Code or Codex usage a full multi-pass redesign consumes.","What the story video step installs and how it behaves on Windows.","What a paid image generation costs on each of the services it offers to connect."]},"zh-Hans":{"summaryShort":"一个面向 Claude Code 和 Codex 的技能，重新设计有 AI 痕迹的页面和看板：修正间距、颜色、图标和文案。","summary":"unslop 是一个你复制进 Claude Code 或 Codex 的技能文件夹。你把它指向一个用 AI 帮助做出来的界面，甚至是一个正在运行的网址，它会列出让界面看起来像机器制作的那些痕迹（紫色渐变、光晕、emoji 标签、喊叫式文字、不均匀的间距），根据你的产品提出一个改进方向，然后逐项修改界面：间距、颜色、图标、文案、字体、布局、细节、视觉效果和动效。它支持 Next.js、React、Vue、Svelte、Tailwind 或纯 HTML 和 CSS。它面向会写前端代码、已经在用 Claude Code 或 Codex 的人。","easeRationale":"三个系统的做法相同：你需要先安装好 Claude Code 或 Codex，然后克隆这个仓库，把一个文件夹复制到 ~/.claude/skills（Codex 则是 ~/.agents/skills，也可以用 $skill-installer 获取），再输入 /unslop 或 $unslop。可选的故事视频功能还需要 Playwright 和 ffmpeg。","costNote":"这个技能基于 CC BY 4.0 协议，免费使用，但要求署名。它运行在 Claude Code 或 Codex 里面，这些工具需要付费计划或 API 用量，可选的主视觉图片由 fal、Replicate、OpenAI images 或 Higgsfield 等付费服务生成。","notVerified":["在你自己的界面上，效果是否真的更好，这是一个审美判断的问题。","在 README 没有提到的框架上，它是否同样好用。","一次完整的多轮重新设计会消耗多少 Claude Code 或 Codex 的用量。","故事视频这一步会安装什么，在 Windows 上表现如何。","在它提供连接的每个服务上，一次付费图片生成要花多少钱。"]}}},{"id":"R_kgDORi26-Q","slug":"heygen-com__hyperframes","name":"Hyperframes","owner":"heygen-com","fullName":"heygen-com/hyperframes","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"desktop_app_and_studio_live_outside_this_repository","severity":"attention","evidence":"README 里“Download the desktop app”的链接指向带有 utm 跟踪参数的 hyperframes.studio/studio/download，而这里列出的发布版本“no platform binaries”。那个应用包含什么内容、遵循什么条款，这个仓库里都没有说明。","source":"ai","title":"桌面应用和 studio 存在于这个仓库之外"},{"code":"renders_run_whatever_the_page_contains","severity":"attention","evidence":"示例组合里包含“<script src=\\\"https://cdn.jsdelivr.net/npm/gsap@3/dist/gsap.min.js\\\"></script>”，README 说渲染器“seeks each frame in headless Chrome”。由 AI 助手写出的组合其实就是一个带脚本的网页，渲染它就是在你的机器上运行这些脚本。","source":"ai","title":"渲染会运行页面里的任何内容"},{"code":"writes_skills_and_plugins_into_your_ai_assistant","severity":"info","evidence":"README，Quick Start：“claude plugin marketplace add heygen-com/hyperframes”“claude plugin install hyperframes@hyperframes”并带有“Enable auto-update”，以及“npx hyperframes skills update”，会为 Claude Code、Codex、Cursor、Gemini CLI 等安装技能文件。","source":"ai","title":"会把技能和插件写入你的 ai 助手"},{"code":"one_shipped_skill_needs_the_paid_tier","severity":"info","evidence":"在“On demand”表格中：“/reverse-any-video - Rebuilding any video file as an editable HyperFrames project ... HyperFrames Pro.”","source":"ai","title":"其中一个内置技能需要付费档位"},{"code":"a_company_product_that_suggests_its_own_paid_services","severity":"info","evidence":"这个仓库属于 heygen-com，README 说 HyperFrames“is used in production at HeyGen”；/media-use 技能可以“music, sound effects, images, icons, logos, voiceover, avatar and generated video ... with HeyGen or free local tools”。","source":"ai","title":"这是一款会推荐自家付费服务的公司产品"},{"code":"releases_almost_every_day","severity":"info","evidence":"这个项目自 2026-03-10 以来共发布了 587 个版本，最近三个（v0.9.4、v0.9.5、v0.9.6）日期都是 2026-11-02；版本号仍低于 1.0。","source":"ai","title":"几乎每天都有新版本发布"},{"code":"repository_is_heavy_to_clone","severity":"info","evidence":"README，Development Note：仓库在 Git LFS 中保留了“about 240 MB of .mp4 files”用作回归测试基准，并建议只需要源代码的话设置 GIT_LFS_SKIP_SMUDGE=1。","source":"ai","title":"仓库克隆起来很重"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg"],"buildTools":["node","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Google Gemini (only for automatic image captions)","HeyGen (only for voices, avatars and generated media)","Amazon Web Services (only to render on Lambda)"],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","ai_skills","development"],"alternativeTo":[{"slug":"remotion","name":"Remotion","confirmed":true},{"slug":"after-effects","name":"After Effects","confirmed":true},{"slug":"camtasia","name":"Camtasia","confirmed":false}],"declaredAlternatives":[],"stars":82245,"downloads":13525,"downloadsWeek":2268,"downloadsDays":7,"growth":3890,"growthDays":7,"history":[{"date":"2026-10-29","stars":75479},{"date":"2026-10-30","stars":76054},{"date":"2026-10-31","stars":76588},{"date":"2026-11-01","stars":77144},{"date":"2026-11-02","stars":77742},{"date":"2026-11-03","stars":78355},{"date":"2026-11-04","stars":78869},{"date":"2026-11-05","stars":79480},{"date":"2026-11-06","stars":80048},{"date":"2026-11-07","stars":80525},{"date":"2026-11-08","stars":81022},{"date":"2026-11-09","stars":81615},{"date":"2026-11-10","stars":82245}],"firstPublishedAt":"2026-10-29T01:54:21.324Z","commitSha":"d03bdeae9d992c9bb653bb1e67cf5569d16455f3","verifiedAt":"2026-11-03T02:44:59.171Z","lastCommitAt":"2026-11-10T21:46:26Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a web page you write into a finished video file, so an AI assistant can make videos for you.","summary":"HyperFrames turns an HTML page into an MP4 video. You lay the video out as a page (text, pictures, clips, music) with timing written in the tags, preview it in a browser and then render it: a hidden Chrome goes frame by frame and FFmpeg writes the file. It is built so an AI coding assistant can do the writing for you, with ready-made skills and about 450 reusable pieces (titles, captions, charts, transitions). You run it from the terminal with Node 22 or newer and FFmpeg installed.","easeRationale":"On all three systems the documented way in is the terminal: install Node.js 22 or newer and FFmpeg, then run npx hyperframes init, preview and render. The README also offers a ready desktop app for macOS, Windows and Linux, but it is downloaded from hyperframes.studio and is not part of the files published here.","costNote":"The code here is Apache-2.0 and the README says there are \"no per-render fees or commercial-use thresholds\". One of the eight skills, /reverse-any-video, is marked \"HyperFrames Pro\", and the media skill offers HeyGen's paid services alongside free local tools; an optional Gemini key for image captions is listed at about $0.001 per image.","notVerified":["Whether an AI assistant really produces a watchable video from a plain request.","How long a render takes and how much a minute of video costs in computer time.","Whether the result is identical every time, as the \"deterministic\" claim says.","How much disk the browser, the fonts and the catalog pieces use after install (the estimate here is only a guess).","What the desktop app and the hosted playground send to the company's servers.","Which parts need a HyperFrames Pro or HeyGen account and what those cost."]},"zh-Hans":{"summaryShort":"把你写的网页变成一个成品视频文件，这样 AI 助手就能帮你制作视频。","summary":"HyperFrames 把一个 HTML 页面变成一段 MP4 视频。你像写页面一样排版视频内容（文字、图片、片段、音乐），时间信息写在标签里，在浏览器里预览，然后渲染：一个隐藏的 Chrome 逐帧运行，FFmpeg 写出视频文件。它的设计目的是让 AI 编程助手来替你写这些内容，自带现成的技能和大约 450 个可复用素材（标题、字幕、图表、转场）。你在终端里运行它，需要 Node 22 或更新版本以及 FFmpeg。","easeRationale":"在三个系统上，文档给出的方式都是终端：安装 Node.js 22 或更新版本和 FFmpeg，然后运行 npx hyperframes init、preview 和 render。README 还提供了一个现成的桌面应用，支持 macOS、Windows 和 Linux，但它是从 hyperframes.studio 下载的，并不属于这里发布的文件。","costNote":"这里的代码是 Apache-2.0，README 说没有“per-render fees or commercial-use thresholds”。八个技能中的一个，/reverse-any-video，被标记为“HyperFrames Pro”，媒体技能在免费本地工具的基础上还提供 HeyGen 的付费服务；一个可选的 Gemini 密钥用于图片配字幕，报价约为每张图片 $0.001。","notVerified":["AI 助手是否真的能根据一个简单的请求做出一段能看的视频。","一次渲染要花多长时间，一分钟视频要消耗多少计算时间。","结果是否真的像“deterministic”所说的那样每次都一样。","浏览器、字体和素材库安装之后占用多少磁盘空间（这里的估算只是一个猜测）。","桌面应用和托管的演示环境会向公司服务器发送什么内容。","哪些部分需要 HyperFrames Pro 或 HeyGen 账号，费用是多少。"]}}},{"id":"R_kgDOVHytVw","slug":"heyputer__everything","name":"Everything","owner":"HeyPuter","fullName":"HeyPuter/everything","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"16 天内获得 5249 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"very_young_huge_scope","severity":"info","evidence":"仓库创建于 2026-10-13（约在分析前两周），已列出大约 50 个声称可替代 Photoshop、AutoCAD、Revit、SolidWorks 和 Salesforce 等产品的应用；检查记录显示 16 天内获得 5249 个星标。","source":"ai","title":"项目非常新，范围却极其庞大"},{"code":"demo_links_not_ready","severity":"info","evidence":"README.md：许多 \"Try it\" 链接带有 title=\"Demo URL not yet provided\"，或指向 \"#link-status\"（例如 Notes、Wiki、Expenses、Assets、Property Management、API Client）。","source":"ai","title":"演示链接尚未就绪"}],"platforms":["web","windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","productivity","design_cad"],"alternativeTo":[{"slug":"dropbox","name":"Dropbox","confirmed":true}],"declaredAlternatives":[],"stars":8001,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1180,"growthDays":7,"history":[{"date":"2026-10-26","stars":4163},{"date":"2026-10-27","stars":4418},{"date":"2026-10-28","stars":4961},{"date":"2026-10-29","stars":5217},{"date":"2026-10-30","stars":5948},{"date":"2026-10-31","stars":6112},{"date":"2026-11-01","stars":6309},{"date":"2026-11-02","stars":6539},{"date":"2026-11-03","stars":6821},{"date":"2026-11-04","stars":7100},{"date":"2026-11-05","stars":7263},{"date":"2026-11-06","stars":7449},{"date":"2026-11-07","stars":7569},{"date":"2026-11-08","stars":7653},{"date":"2026-11-09","stars":7805},{"date":"2026-11-10","stars":8001}],"firstPublishedAt":"2026-10-27T01:10:12.635Z","commitSha":"00f4a541a31eed0c86e30934c1842cdd2c3a32bd","verifiedAt":"2026-10-30T01:59:23.801Z","lastCommitAt":"2026-11-10T20:41:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A large set of free web apps, such as a spreadsheet, documents, photo editor and CAD tool, that you can try in a browser.","summary":"The Everything Project, from the company behind Puter, collects about 50 free apps in one repository: office tools (spreadsheet, documents, slides, PDF), team chat, photo, video and audio editors, CAD and engineering tools, business apps like a CRM and invoicing, and simple games. Each app lists the paid product it aims to replace. Many have a \"Try it\" link that opens a hosted version in your browser; some links are still marked as not ready. Running an app yourself needs Node.js 22.12 or newer and a few terminal commands. The project started in October 2026.","easeRationale":"On the web it is grade A: many apps have hosted versions on puter.com that open directly in a browser with nothing to install. On Windows, macOS and Linux you run a copy yourself, which the README says needs Node.js 22.12 or newer plus `npm install` and `npm run dev` inside each app's folder.","costNote":"MIT license. The README does not say whether the hosted versions on puter.com need an account or have limits.","notVerified":["How complete each app is compared with the paid product it names; some may be early prototypes.","Whether the hosted versions need a Puter account, store your files on Puter servers, or have usage limits.","Whether every app runs locally on Windows with the documented npm commands.","How files made in these apps open in the paid programs they replace (for example .docx, .xlsx, .dwg)."]},"zh-Hans":{"summaryShort":"一大批免费的网页应用,包括电子表格、文档、图片编辑器和 CAD 工具,都可以在浏览器里直接试用。","summary":"Everything Project 由 Puter 背后的公司推出,把大约 50 个免费应用集中在一个代码库里:办公工具(电子表格、文档、幻灯片、PDF)、团队聊天、图片、视频和音频编辑器、CAD 和工程工具、CRM 和发票等商业应用,还有一些简单的小游戏。每个应用都标明了它想替代的付费产品。很多应用都有一个“Try it”链接,点击后会在浏览器里打开托管版本;部分链接仍标记为尚未就绪。自己运行一个应用需要 Node.js 22.12 或更新版本,以及几条终端命令。该项目于 2026 年 10 月启动。","easeRationale":"在网页端属于 A 级:很多应用在 puter.com 上有托管版本,直接在浏览器里打开,什么都不用装。在 Windows、macOS 和 Linux 上,你要自己运行一份副本,README 说这需要 Node.js 22.12 或更新版本,还要在每个应用的文件夹里执行 `npm install` 和 `npm run dev`。","costNote":"MIT 许可证。README 没有说明 puter.com 上的托管版本是否需要账号,或是否有使用限制。","notVerified":["每个应用相对于它所说的付费产品完成度如何;有些可能只是早期原型。","托管版本是否需要 Puter 账号、是否会把你的文件存在 Puter 的服务器上,或是否有使用限制。","是否所有应用都能按文档中的 npm 命令在 Windows 上本地运行。","用这些应用制作的文件,在它们要替代的付费软件中(例如 .docx、.xlsx、.dwg)打开效果如何。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMzEwODExMDk=","slug":"hiroppy__fusuma","name":"Fusuma","owner":"hiroppy","fullName":"hiroppy/fusuma","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次 release 在 2021-03-30（2046 天前）","source":"check","title":"release 已经过时"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹或 winget 清单都没有检测到许可证","source":"check","title":"没有许可证"},{"code":"archived","severity":"info","evidence":"该仓库已归档（只读，不再维护）","source":"check","title":"已归档"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.5/10","source":"check","title":"scorecard 评分偏低"},{"code":"license_only_in_package_json","severity":"info","evidence":"package.json 写着 \"license\": \"MIT\"，但顶层文件列表里没有 LICENSE 文件，检查也没有发现任何许可证","source":"ai","title":"许可证只出现在 package json 里"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["office","development"],"alternativeTo":[{"slug":"powerpoint","name":"PowerPoint","confirmed":true}],"declaredAlternatives":[],"stars":5355,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":5355},{"date":"2026-11-07","stars":5355},{"date":"2026-11-08","stars":5355},{"date":"2026-11-09","stars":5355},{"date":"2026-11-10","stars":5355}],"firstPublishedAt":"2026-11-05T22:43:48.479Z","commitSha":"b27ba013331ac572c648981b5d567a2db2cb2268","verifiedAt":"2026-11-05T22:23:58.729Z","lastCommitAt":"2021-12-01T18:05:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns Markdown files into web slides for talks, with themes, speaker notes, a timer and PDF export.","summary":"Fusuma is a command-line tool that builds presentation slides from Markdown (or MDX, which can include React components). You write the text in files, run a command to preview the slides in your browser, and then export them as a web page or a PDF. It also has a presenter view with speaker notes and a timer, and a command that publishes the slides to GitHub Pages. You need Node.js and the terminal to use it. The repository is archived: the last commit was in December 2021 and the last release in March 2021.","easeRationale":"On Windows, macOS and Linux the steps are the same: install Node.js 14.17.6 or newer, then run the install and `npx fusuma init` / `start` commands shown in the README. There is no installer or ready-made program, and no hosted version to try.","notVerified":["Whether the 2021 code still installs and runs on current Node.js versions","Whether PDF export and the GitHub Pages deploy command still work","Whether the presenter mode features (screen capture, voice recording) work in today's Chrome and Firefox","Real disk use after npm downloads all dependencies"]},"zh-Hans":{"summaryShort":"把 Markdown 文件变成用于演讲的网页幻灯片，支持主题、演讲者备注、计时器和导出 PDF。","summary":"Fusuma 是一个命令行工具，能把 Markdown（或可以包含 React 组件的 MDX）变成演示幻灯片。你把文字写在文件里，运行一个命令就能在浏览器里预览幻灯片，然后把它们导出成网页或 PDF。它还有一个带备注和计时器的演讲者视图，以及一个能把幻灯片发布到 GitHub Pages 的命令。使用它需要 Node.js 和终端。这个仓库已经被归档：最后一次提交是在 2021 年 12 月，最后一次 release 是在 2021 年 3 月。","easeRationale":"在 Windows、macOS 和 Linux 上步骤都一样：安装 Node.js 14.17.6 或更新版本，然后运行 README 里给出的安装命令和 `npx fusuma init` / `start`。没有安装程序，也没有现成的程序，也没有可以直接试用的在线版本。","notVerified":["2021 年的代码在现在的 Node.js 版本上是否还能安装和运行","导出 PDF 和发布到 GitHub Pages 的命令是否还能用","演讲者模式的功能（屏幕录制、语音录制）在现在的 Chrome 和 Firefox 里是否还能用","npm 下载完所有依赖后，实际占用的磁盘空间有多大"]}}},{"id":"R_kgDOQwI8Aw","slug":"hkuds__deeptutor","name":"DeepTutor","owner":"HKUDS","fullName":"HKUDS/DeepTutor","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"D"},"risk":"attention","signals":[{"code":"desktop_builds_are_previews_and_unsigned","severity":"attention","evidence":"README 'Desktop — macOS and Windows (Preview)':'Desktop previews are available on dev; the v1.7.0 Release does not include desktop installers',下载来自CI构建产物,会'expire after seven days',而且'Windows preview builds are unsigned and may show a SmartScreen warning.'","source":"ai","title":"桌面构建是未签名的预览版"},{"code":"ai_can_run_commands_on_your_computer","severity":"attention","evidence":"聊天工具列表中包含'exec',Dockerfile 中写道'the restricted-subprocess backend (which the office skills... rely on for exec) is enabled by default via the sandbox_allow_subprocess runtime setting (system.json, default on)'。docker-compose.yml 还提到,在'bare-metal / macOS deployments where the runner is not started'的情况下,应用会'degrades to bwrap (Linux, if installed) or... a restricted subprocess',所以命令被约束的程度取决于你是怎么安装的。","source":"ai","title":"AI可以在你的电脑上执行命令"},{"code":"installs_third_party_code_while_running","severity":"attention","evidence":"Dockerfile 中的说明:'git is required to install CLI apps — most of the CLI-Anything catalog installs with pip install git+…'。发布说明列出了'101 CLI Apps the tutor can run'以及'Install community skills from ClawHub with deeptutor skill install behind a security gate'。compose 文件还会在每次启动时重新应用 DEEPTUTOR_EXTRAS 和 DEEPTUTOR_APT_PACKAGES。","source":"ai","title":"运行期间会安装第三方代码"},{"code":"docker_files_publish_ports_to_the_network","severity":"attention","evidence":"compose.yaml(podman)绑定的是'127.0.0.1:${HOST_PORT_BACKEND:-8001}:8001',但 docker-compose.yml 和 docker-compose.ghcr.yml 发布的是'${DEEPTUTOR_DOCKER_BACKEND_PORT:-8001}:...',没有 127.0.0.1 前缀,所以在这些文件里,应用会在机器的所有网络接口上响应。","source":"ai","title":"Docker文件会把端口暴露到网络"},{"code":"keys_and_accounts_kept_in_the_data_folder","severity":"info","evidence":"docker-compose.yml 中的注释:数据目录包含'accounts/grants/audit (data/system)',并特别指出'never data/system, which holds auth state, provider API keys, and per-owner secrets'是沙箱不应该看到的部分。你配置的一切都保存在你磁盘上的那个文件夹里。","source":"ai","title":"密钥和账户保存在数据文件夹中"}],"platforms":["windows","mac","linux"],"tools":["python","docker"],"buildTools":["python","docker"],"disk":{"minMB":461,"maxMB":1382,"basis":"installer"},"apiKeys":["A language model provider of your choice, set in Settings before use","OpenAI"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["productivity","local_ai","ai_agents"],"alternativeTo":[{"slug":"notebooklm","name":"NotebookLM","confirmed":true},{"slug":"quizlet","name":"Quizlet","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[],"stars":57952,"downloads":17823,"downloadsWeek":13558,"downloadsDays":7,"growth":3234,"growthDays":7,"history":[{"date":"2026-10-27","stars":45915},{"date":"2026-10-28","stars":47991},{"date":"2026-10-29","stars":50068},{"date":"2026-10-30","stars":51779},{"date":"2026-10-31","stars":52820},{"date":"2026-11-01","stars":53726},{"date":"2026-11-02","stars":54317},{"date":"2026-11-03","stars":54718},{"date":"2026-11-04","stars":55028},{"date":"2026-11-05","stars":55409},{"date":"2026-11-06","stars":55886},{"date":"2026-11-07","stars":56332},{"date":"2026-11-08","stars":56921},{"date":"2026-11-09","stars":57574},{"date":"2026-11-10","stars":57952}],"firstPublishedAt":"2026-10-27T18:40:27.908Z","commitSha":"f7aed0744c36c5afef16a4261c31ebee1208cac2","verifiedAt":"2026-11-01T01:45:38.371Z","lastCommitAt":"2026-11-06T08:41:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A personal AI tutor you run yourself: ask about your own documents, get quizzes, study books, reading help and practice.","summary":"DeepTutor is a learning workspace you run on your own computer. You add documents, videos or web sources, and the tutor answers with citations to them, builds quizzes and practice, writes study books and narrated video lessons, keeps a notebook and tracks what you have learned. It brings no AI model of its own: you connect a provider key or a local model such as Ollama or LM Studio. There are preview installers for Windows and macOS that bundle Python and Node.js, plus Docker and pip installs for people comfortable with a terminal.","easeRationale":"Windows and macOS are a B: release v1.7.1 has a Setup .exe, a portable ZIP and a .dmg, and the README says the desktop build 'bundles Python, Node.js, and the full learning workspace, so no terminal or separate runtime installation is needed' — you then pick a provider and model in Settings. Linux is a D, because there is no desktop build and the documented routes are Docker or pip with commands such as 'deeptutor init' and 'deeptutor start'.","costNote":"The software is free (Apache-2.0). You pay whichever model provider you connect, unless you point it at a local model.","notVerified":["Whether the preview desktop builds work and match the source in this repository","What the model calls cost once you use books, quizzes and research for real","Real disk use after indexing your own documents, on top of the 461 MB installer","Whether the restricted sandbox keeps a generated command away from the rest of your files","Whether the Anki, calendar and Word exports come out as described"]},"zh-Hans":{"summaryShort":"一个你自己运行的个人AI导师:就你自己的文档提问,获得测验、学习手册、阅读帮助和练习。","summary":"DeepTutor 是一个在你自己电脑上运行的学习工作区。你添加文档、视频或网页来源,导师会带着引用来回答,生成测验和练习,撰写学习手册和带旁白的视频课程,维护一个笔记本,并追踪你学到的内容。它本身不附带任何AI模型:你需要接入一个服务商的密钥,或者像Ollama、LM Studio这样的本地模型。Windows和macOS有打包了Python和Node.js的预览版安装程序,也为习惯用终端的人提供Docker和pip安装方式。","easeRationale":"Windows 和 macOS 是 B 级:v1.7.1 版本提供了 Setup .exe、便携版 ZIP 和 .dmg,README 中写道桌面版'bundles Python, Node.js, and the full learning workspace, so no terminal or separate runtime installation is needed'——然后你在 Settings 里选择服务商和模型。Linux 是 D 级,因为没有桌面构建,文档记载的途径是 Docker 或 pip,使用像'deeptutor init'和'deeptutor start'这样的命令。","costNote":"软件本身是免费的(Apache-2.0)。你需要向你接入的模型服务商付费,除非指向本地模型。","notVerified":["预览版桌面构建是否能用,是否与这个仓库里的源码一致","真正用于书籍、测验和研究时,模型调用会花多少钱","在索引你自己的文档之后,除了461MB的安装包之外,实际占用多少磁盘空间","受限沙箱是否真的能让生成的命令接触不到你其他的文件","Anki、日历和Word的导出是否真的像描述那样生成"]}}},{"id":"R_kgDOORH-ow","slug":"hkuds__vimax","name":"ViMax","owner":"HKUDS","fullName":"HKUDS/ViMax","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"can_place_real_people_in_generated_video","severity":"attention","evidence":"readme.md的关键功能：\"AutoCameo — place a person or pet from a reference photo into generated stories while maintaining a consistent appearance\"，以及\"Make Your Own Cameo Video ... by uploading your photo\"。同样的功能也能用在没有同意的人的照片上。","source":"ai","title":"可以把真实人物放进生成的视频里"},{"code":"paid_api_cost_grows_fast","severity":"attention","evidence":"readme.md的Usage部分：\"ViMax generates 2 keyframe candidates in parallel and selects one using your vision-capable chat model. This can improve visual quality and consistency, at a higher API cost.\" 没有任何部分在本地运行，所以聊天、图像和视频服务会为每个场景、每个候选项和每个镜头计费。","source":"ai","title":"付费API成本增长很快"},{"code":"api_keys_pasted_into_config_files","severity":"info","evidence":"readme.md让你把`api_key: <YOUR_API_KEY>`写进configs/agent.local.yaml、configs/idea2video.yaml和configs/script2video.yaml。环境变量（VIMAX_LLM_API_KEY等）被作为替代方案提供，但文档记载的做法会把三个密钥以明文留在项目文件夹里。","source":"ai","title":"API密钥被粘贴进配置文件"},{"code":"research_code_edited_by_hand","severity":"info","evidence":"readme.md要求你把想法、要求和风格直接写进main_idea2video.py，把剧本写进main_script2video.py，而不是作为输入传入。这个项目来自一个高校团队（主页是arXiv论文2606.07649），在GitHub上只有一个发行版。","source":"ai","title":"研究代码需要手动编辑"},{"code":"web_ui_is_a_development_server","severity":"info","evidence":"readme.md的Web UI部分：`cd web && npm install && npm run dev`，然后打开http://127.0.0.1:4173。这是开发服务器，没有提到任何登录机制；README建议通过SSH隧道访问远程机器，而不是直接开放端口。","source":"ai","title":"网页界面是一个开发服务器"},{"code":"mac_not_supported","severity":"info","evidence":"readme.md的Quick Start里Environment部分写道：\"OS: Linux, Windows\"。macOS没有被列出，pyproject.toml也只在linux和win32上把torch指向CUDA索引。","source":"ai","title":"不支持Mac"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":960,"maxMB":4420,"basis":"heuristic"},"apiKeys":["A chat/LLM service (OpenAI-compatible, for example OpenRouter or MiniMax)","An image generation service (for example Google Nano Banana or GPT Image 2)","A video generation service (for example Google Veo or Seedance)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","ai_agents","image"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"invideo-ai","name":"AI","confirmed":true},{"slug":"ltx-studio","name":"LTX Studio","confirmed":true}],"declaredAlternatives":[],"stars":12888,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":33,"growthDays":6,"history":[{"date":"2026-11-04","stars":12855},{"date":"2026-11-05","stars":12858},{"date":"2026-11-06","stars":12860},{"date":"2026-11-07","stars":12866},{"date":"2026-11-08","stars":12875},{"date":"2026-11-09","stars":12881},{"date":"2026-11-10","stars":12888}],"firstPublishedAt":"2026-11-04T01:36:50.437Z","commitSha":"fd4b72e7731be6e5f88486206d25ea59849b0285","verifiedAt":"2026-11-04T01:24:40.152Z","lastCommitAt":"2026-09-30T14:35:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Describe an idea and it writes the script, plans the shots and builds a whole video using paid AI image and video services.","summary":"ViMax is a research project that chains AI steps to make a complete video from a sentence, a screenplay or a novel chapter: it writes the story, breaks it into scenes and shots, generates reference pictures, then generates and joins the clips. You drive it from the terminal, a text menu or a local web page. It brings no AI models of its own: you paste in keys for a chat service, an image service and a video service and pay those by the shot. The README lists Linux and Windows only, and you install it with Python and uv.","easeRationale":"There is no installer for either system: you install uv, clone the repository, run `uv sync` (it pulls PyTorch from the NVIDIA CUDA 12.8 index on Linux and Windows), copy configs/agent.example.yaml and hand-edit three blocks of model names, base URLs and API keys before anything runs. The browser interface needs Node.js 18+ and `npm install` on top of that. The README's Environment section lists only Linux and Windows, so there is no grade for macOS.","costNote":"The code is free (MIT). Every shot is generated by a paid service (Google Veo, Nano Banana, Seedance, GPT Image 2 through OpenRouter and similar), and a one-minute video needs many of them.","notVerified":["How much a finished minute of video really costs at today's provider prices","Whether the final videos look like the demo clips in the README","Real disk use after dependencies, reference images and rendered shots pile up","Whether it runs at all without an NVIDIA graphics card","Whether the Web UI and the text interface are stable, since there is only one release"]},"zh-Hans":{"summaryShort":"描述一个想法，它就能写剧本、规划分镜，并使用付费的AI图像和视频服务制作出完整视频。","summary":"ViMax是一个研究项目，它把多个AI步骤串联起来，从一句话、一个剧本或小说的一个章节出发制作出一段完整视频：写故事、拆分成场景和镜头、生成参考图片，然后生成并拼接各个片段。你可以通过终端、文本菜单或本地网页来操作它。它本身不带任何AI模型：你需要粘贴聊天服务、图像服务和视频服务的密钥，并按镜头付费。README只列出了Linux和Windows，安装需要用Python和uv。","easeRationale":"这两个系统都没有安装程序：你需要安装uv，克隆仓库，运行`uv sync`（在Linux和Windows上会从NVIDIA的CUDA 12.8索引拉取PyTorch），复制configs/agent.example.yaml，并手动编辑三处模型名称、基础URL和API密钥的区块，之后才能运行。浏览器界面还需要Node.js 18+和`npm install`。README的Environment部分只列出了Linux和Windows，所以没有macOS的评级。","costNote":"代码是免费的（MIT）。每个镜头都由付费服务生成（Google Veo、Nano Banana、Seedance、通过OpenRouter调用的GPT Image 2等），一分钟的视频需要调用很多次。","notVerified":["按今天的服务商价格，一分钟成片实际要花多少钱","成片是否真的像README里的演示片段那样","依赖项、参考图片和渲染好的镜头堆积起来后的实际磁盘占用","没有NVIDIA显卡的情况下它是否还能运行","由于只有一个发行版，网页界面和文本界面是否稳定"]}}},{"id":"R_kgDOVovjIg","slug":"holyarrs__holyarrs","name":"HolyArrs","owner":"HolyArrs","fullName":"HolyArrs/HolyArrs","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"young_owner_account","severity":"attention","evidence":"作者账号创建于 15 天前","source":"check","title":"作者账号刚注册不久"},{"code":"legal_status_varies_by_country","severity":"attention","evidence":"README 的 \"Ask for a title\" 和 \"It works with what you already run\"：该应用会搜索 \"any Torznab or Newznab indexer, or Prowlarr or Jackett\"，挑选一个发布版，并发送给 qBittorrent、Transmission、Deluge、SABnzbd 或 NZBGet。用这种方式获取剧集、电影和音乐，正是常见的获取受版权保护内容的途径，在很多国家受到限制或被禁止。","source":"ai","title":"各国法律情况不同"},{"code":"depends_on_a_server_run_by_the_author","severity":"attention","evidence":"README：\"Titles, artwork and charts come from the HolyArrs metadata server\" 以及 \"Every install talks to a metadata server I pay for every month\"。也就是说，每一份安装都会连到一个由个人维护的私有服务器，项目还在征集赞助来继续支付费用。","source":"ai","title":"依赖作者自己运营的服务器"},{"code":"first_visitor_becomes_the_owner","severity":"attention","evidence":"README 的 Application setup：\"Open http://your-server:6262 and make the owner account right away. There is no default password. Until you make it, whoever opens HolyArrs first becomes the owner.\" Security 一节也提醒不要在路由器上把 6262 端口映射到外网。","source":"ai","title":"第一个打开的人就成为管理员"},{"code":"very_new_project_fast_releases","severity":"info","evidence":"仓库创建于 2026-11-09，到 2026-11-10 就已经发布了 16 次，版本号到 1.0.15，有 76 个未关闭的问题，作者账号只有 15 天。README 列出了若干尚未实现的功能：没有内置下载器、不能在线搜索字幕、Explore 里的音乐部分还没做完。","source":"ai","title":"项目非常新且发布极为频繁"},{"code":"docker_image_only","severity":"info","evidence":"GitHub 上的 16 个发布版都没有附带任何平台的可执行文件；唯一的运行方式是 Docker Hub 上的 holyarrs/holyarrs 镜像，因此你运行的是一个已经构建好的镜像，而不是能与公开源码逐一对照的文件。","source":"ai","title":"只提供 Docker 镜像"},{"code":"contributor_licence_agreement_and_paid_relicensing","severity":"info","evidence":"README 的 Contributing 和 Licence：\"On your first pull request a bot asks you to accept a short licence agreement (CLA.md)\" 以及 \"Want to build it into a closed product? Email support@HolyArrs.com for a commercial licence\"，也就是说作者保留按其他条款出售这份代码的权利。","source":"ai","title":"贡献者协议与付费重新授权"},{"code":"fetches_shows_movies_and_music_from_indexers","severity":"high","evidence":"README：\"Browse like a streaming app. Ask for anything. It finds it. Straight to your downloader. Subtitles kept. Into your library.\"；\"Sources: any Torznab or Newznab indexer, or Prowlarr or Jackett\"；\"Downloaders: qBittorrent, Transmission, Deluge, SABnzbd, NZBGet\"。售卖或播放电影、剧集和音乐的服务，其条款都不允许你自行保留副本，而许多国家还有法律限制通过 BitTorrent 分享受版权保护的文件。","source":"ai","title":"从索引站点抓取剧集、电影和音乐"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["go"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["automation","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":562,"downloads":0,"downloadsWeek":0,"downloadsDays":0,"growth":3801,"growthDays":0,"history":[],"firstPublishedAt":"2026-11-10T23:52:43.927Z","commitSha":"eac2a68c4638746be936502ec9d969eed9024164","verifiedAt":"2026-11-10T23:35:07.569Z","lastCommitAt":"2026-11-10T21:37:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"One self-hosted web app that finds, downloads and files your shows, movies and music, instead of running six separate apps.","summary":"HolyArrs is a self-hosted web app for a home media server. You browse shows, movies and music like a streaming app, ask for a title, and it searches your indexers, picks a release by your rules, hands it to your download program and files it into your library. It aims to replace Sonarr, Radarr, Lidarr, Bazarr, Prowlarr and Overseerr with one login and one port, and it can import your settings from them. You run it as a Docker container (amd64, arm64 or armv7) and still need a downloader such as qBittorrent or SABnzbd.","easeRationale":"On Linux, the easiest case, there is no installer: you need Docker, a docker-compose.yml you write yourself with your folders and user IDs, a terminal command to start it, and then a separate download program such as qBittorrent or SABnzbd and at least one indexer. Windows and Mac work the same way through Docker Desktop, with the extra rule that the /config folder must be a named volume.","costNote":"Free and open source under AGPL-3.0. Sponsorship is optional, and the author sells a separate commercial licence for anyone who wants to build it into a closed product.","notVerified":["what the metadata server receives from each install, and what happens to your library if that server goes away","whether the import from Sonarr, Radarr and the others really leaves those apps untouched and can be undone","whether one app handles a large library as well as the six programs it replaces, since the 15 MB image and 140 MB of RAM figures are the author's own measurements","how much disk space the database, backups and artwork cache take once your library grows","whether the encryption of stored passwords with the holyarrs.key file protects them as described"]},"zh-Hans":{"summaryShort":"一个自建的网页应用，查找、下载并整理你的剧集、电影和音乐，不用再分别运行六个程序。","summary":"HolyArrs 是一个自己部署的网页应用，用于家庭媒体服务器。你可以像用流媒体应用一样浏览剧集、电影和音乐，点名要某个片子，它就会在你的索引站点里搜索，按你的规则挑选一个发布版，交给你的下载程序，再归档到你的媒体库里。它的目标是用一个登录入口和一个端口取代 Sonarr、Radarr、Lidarr、Bazarr、Prowlarr 和 Overseerr，并且可以导入它们的设置。你以 Docker 容器方式运行它（amd64、arm64 或 armv7），另外仍然需要 qBittorrent 或 SABnzbd 这样的下载程序。","easeRationale":"在最简单的 Linux 上也没有安装程序：你需要 Docker，需要自己写一个带有目录和用户 ID 的 docker-compose.yml，在终端里敲命令启动，然后还要单独准备 qBittorrent 或 SABnzbd 这类下载程序，以及至少一个索引站点。Windows 和 Mac 通过 Docker Desktop 同理，只是多一条规则：/config 目录必须是命名卷。","costNote":"免费开源，采用 AGPL-3.0。赞助是自愿的；如果想把它做进闭源产品，作者另外出售商业许可。","notVerified":["元数据服务器从每个安装实例收到哪些信息，以及如果该服务器停止运行，你的媒体库会怎样","从 Sonarr、Radarr 等程序导入设置是否真的不会改动那些程序，以及能否撤销","一个程序处理大型媒体库是否真能和它取代的六个程序一样好，因为 15 MB 镜像和 140 MB 内存这些数字是作者自己测的","当媒体库变大后，数据库、备份和封面缓存会占用多少磁盘空间","用 holyarrs.key 文件加密保存的密码，是否真的像说明中那样受到保护"]}}},{"id":"R_kgDOH1Ghpw","slug":"homarr-labs__dashboard-icons","name":"Dashboard Icons","owner":"homarr-labs","fullName":"homarr-labs/dashboard-icons","type":"content_list","status":"active","grade":"A","easeByPlatform":{"web":"A"},"risk":"attention","signals":[{"code":"logos_belong_to_other_companies","severity":"attention","evidence":"这个合集由品牌 logo 组成,README 自己的 Legal 部分写道:\"All product names, trademarks, and registered trademarks are the property of their respective owners. Icons are used for identification purposes only and do not imply endorsement.\" 仓库的 Apache-2.0 许可只覆盖这里的文件和代码,并不覆盖图上画的品牌,所以把某个 logo 用在产品或广告里,仍然要看各个商标所有者自己的规定。","source":"ai","title":"logo 属于其他公司"},{"code":"mixes_in_catalogs_with_other_licenses","severity":"info","evidence":"README 写道,MCP 服务器和网站还会 \"search the selfh.st, LobeHub and Simple Icons catalogs ... with their licenses\",外部托管的 Simple Icons 图片是从 \"https://cdn.simpleicons.org/\" 提供的。README 还补充说,它的 \"CC0-1.0 collection license does not waive third-party trademark, patent, or brand-guideline restrictions\"。用这种方式找到的图标不受本仓库许可证的覆盖。","source":"ai","title":"混入了使用其他许可证的目录"},{"code":"assistant_sends_your_service_list_to_their_server","severity":"info","evidence":"MCP 服务器运行在 \"https://dashboardicons.com/api/mcp\",官方宣传的调用是 \"Match every service in a docker-compose.yml or dashboard config to an icon in one call\"。为了给出答案,你的助手会把那个文件里的服务名称发送到项目运营的服务器;README 把工具、限制和登录方式的细节指向了 web/docs/MCP.md。","source":"ai","title":"助手会把你的服务列表发给他们的服务器"},{"code":"assistant_can_submit_icons_in_your_name","severity":"info","evidence":"README 的 MCP 功能部分写道,服务器可以 \"Submit a missing icon or propose an update for you, once you sign in\"。一旦登录后,助手就可以用你的账号向项目提交内容。","source":"ai","title":"助手可以用你的名义提交图标"},{"code":"no_versions_files_come_from_a_third_party_cdn","severity":"info","evidence":"这个项目没有 GitHub 发行版,推荐的图标使用方式是指向 jsDelivr 的实时链接(\"https://cdn.jsdelivr.net/gh/homarr-labs/dashboard-icons\")或 raw.githubusercontent.com。没有固定版本可以锁定,所以图标随时可能被改动或改名,你仪表盘的每个访问者都会从这个外部服务加载图片。","source":"ai","title":"没有版本号,文件来自第三方 cdn"}],"platforms":["web"],"tools":[],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["dashboardicons.com sign-in (only to submit or update icons from an assistant)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["image","design_cad"],"alternativeTo":[{"slug":"icons8","name":"Icons8","confirmed":true},{"slug":"iconfinder","name":"Iconfinder","confirmed":true}],"declaredAlternatives":[],"stars":9061,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":58,"growthDays":6,"history":[{"date":"2026-11-07","stars":9027},{"date":"2026-11-08","stars":9036},{"date":"2026-11-09","stars":9047},{"date":"2026-11-10","stars":9061}],"firstPublishedAt":"2026-11-06T22:57:16.658Z","commitSha":"59f1e9a4108f4f510f41e9588f078631de1e1405","verifiedAt":"2026-11-06T22:42:25.539Z","lastCommitAt":"2026-11-08T19:50:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free collection of 3,400+ app and service logos for home dashboards, in SVG, PNG and WEBP, with a direct link for each.","summary":"Dashboard Icons is a pile of ready-made logos for the apps and services people put on a home dashboard, in SVG, PNG and WEBP, with light and dark versions. You search the collection at dashboardicons.com, download a file or copy a link that you paste into dashboards such as Homarr, Homepage or Dashy. Nothing is installed. There is also a free MCP server that lets Claude Code, Codex, ChatGPT, Cursor and similar assistants find icons for every service in a docker-compose.yml file in one go. Missing icons are requested through a form on the site.","easeRationale":"Grade A on the web, which is the only place you need: you open dashboardicons.com, search, and either download the file or copy a link such as \"https://cdn.jsdelivr.net/gh/homarr-labs/dashboard-icons/svg/plex.svg\". Nothing is installed; the optional MCP server is one command or one line of settings in your assistant, pointing at a page the project already runs.","costNote":"The collection is Apache-2.0 and the README states the MCP server \"is free and needs no API key\". No price appears anywhere in the repository.","notVerified":["Whether each of the 3,400+ logos is current and matches the brand's own guidelines","What dashboardicons.com and its MCP server record about the requests your assistant sends","Which limits apply to the free MCP server, since the README sends that detail to web/docs/MCP.md","Whether a given logo may be used in your own project, which depends on each trademark owner","How big the whole collection is if you clone it, as the checks could only estimate 100 to 500 MB"]},"zh-Hans":{"summaryShort":"一个免费的图标合集,收录 3400 多个应用和服务的 logo,用于家庭仪表盘,提供 SVG、PNG、WEBP 格式,每个都有直接链接。","summary":"Dashboard Icons 收集了大量现成的 logo,供人们放在家庭仪表盘上使用,格式有 SVG、PNG 和 WEBP,分浅色和深色两个版本。你在 dashboardicons.com 上搜索这个合集,下载文件,或者复制一个链接粘贴到 Homarr、Homepage 或 Dashy 之类的仪表盘里。不需要安装任何东西。项目还提供一个免费的 MCP 服务器,让 Claude Code、Codex、ChatGPT、Cursor 等助手能一次性为 docker-compose.yml 文件里的每个服务找到对应的图标。缺少的图标可以通过网站上的表单申请添加。","easeRationale":"在网页端是 A 级,也是你唯一需要的地方:打开 dashboardicons.com,搜索,然后下载文件,或者复制一个像 \"https://cdn.jsdelivr.net/gh/homarr-labs/dashboard-icons/svg/plex.svg\" 这样的链接。不需要安装任何东西;可选的 MCP 服务器只需在你的助手里执行一条命令或加一行设置,指向项目已经在运行的一个页面。","costNote":"这个合集采用 Apache-2.0 许可,README 说 MCP 服务器 \"is free and needs no API key\"。整个仓库里没有出现任何价格。","notVerified":["3400 多个 logo 是否都是最新的,是否符合各品牌自己的规范","dashboardicons.com 和它的 MCP 服务器会记录你的助手发出的请求的哪些信息","免费 MCP 服务器有哪些限制,因为 README 把这个细节放到了 web/docs/MCP.md 里","某个 logo 能不能用在你自己的项目里,这取决于每个商标持有者的规定","整个合集克隆下来有多大,因为检查只能估计出 100 到 500 MB"]}}},{"id":"R_kgDOU6bksw","slug":"horizon-rust__horizon","name":"Horizon","owner":"horizon-rust","fullName":"horizon-rust/horizon","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"30 天内获得 6027 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 6 天前","source":"check","title":"所有者账号很新"},{"code":"unofficial_rewrite_of_a_paid_game","severity":"attention","evidence":"README.md：“Not an official Minecraft product, nor approved by or associated with Mojang or Microsoft”，以及“Minecraft is a trademark of Mojang Synergies AB; its code and assets belong to Mojang and Microsoft under the Minecraft EULA.”一个读取你已安装的游戏并加入在线服务器的重写客户端，要受到 Mojang 条款的约束，而这只有 Mojang 和微软能决定。","source":"ai","title":"对一款付费游戏的非官方重写"},{"code":"build_from_source_involves_decompiled_game_code","severity":"attention","evidence":"README.md 中 CONTRIBUTING 的说明：“Never commit game files, decompiled source or prepared resources”，Cargo.toml 中的注释写道：“Ports mirror vanilla's Java signatures, index loops and nested generic types one to one so they can be checked against the decompiled sources.”拆解 Java 版游戏来做对比，正是 Minecraft EULA 所涉及的内容。","source":"ai","title":"从源代码构建涉及反编译的游戏代码"},{"code":"parity_claim_narrower_than_it_looks","severity":"info","evidence":"README 展示了一个“vanilla parity 100%”的徽章，随后又写道“Player-visible gaps: 833 open”，并解释“100% means 100% of what the harnesses compare, not that nothing else differs.”这项审核同样是项目自己做的。","source":"ai","title":"一致性声明比看起来的范围更窄"},{"code":"ships_telemetry_code","severity":"info","evidence":"工作区中包含 crates/horizon-telemetry 和 crates/horizon-telemetry-server（Cargo.toml），说明游戏内置了遥测功能；README 没有说明收集了什么内容，也没说默认是否开启。","source":"ai","title":"附带遥测代码"},{"code":"launcher_updates_itself_and_the_game","severity":"info","evidence":"README.md：这个启动器“installs and updates the game, prepares its files from your copy of Minecraft, and updates itself”，所以新代码会自己到达你的机器。","source":"ai","title":"启动器会自我更新并更新游戏"},{"code":"ai_agents_can_play_for_you","severity":"info","evidence":"README.md Agent control：游戏自带的 horizon-ctl“reads its state as JSON and acts like a player”，horizon-ctl mcp 还通过 MCP 提供同样的命令，服务器所有者可能把这当作自动化行为来对待。","source":"ai","title":"ai 代理可以代替你玩游戏"},{"code":"checked_against_decompiled_game_code","severity":"high","evidence":"Cargo.toml：“Ports mirror vanilla's Java signatures, index loops and nested generic types one to one so they can be checked against the decompiled sources.”README，Build from source：“horizon-prepare builds the game's resources from your own copy of Java Edition 26.3”，并写明“Never commit game files, decompiled source or prepared resources.”README 自己的许可说明写道：“Minecraft is a trademark of Mojang Synergies AB; its code and assets belong to Mojang and Microsoft under the Minecraft EULA”，并说 Horizon“Not an official Minecraft product, nor approved by or associated with Mojang or Microsoft”——这份 EULA 正是管辖游戏代码和资源的规则，只有 Mojang 和微软能说明它允许什么。","source":"ai","title":"对照反编译的游戏代码进行校验"}],"platforms":["windows","mac","linux"],"tools":["gpu_any"],"buildTools":["rust"],"disk":{"minMB":15,"maxMB":46,"basis":"installer"},"apiKeys":["Microsoft account that owns Minecraft: Java Edition"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":8549,"downloads":1667728,"downloadsWeek":844420,"downloadsDays":7,"growth":2281,"growthDays":7,"history":[{"date":"2026-10-29","stars":4348},{"date":"2026-10-30","stars":4906},{"date":"2026-10-31","stars":5386},{"date":"2026-11-01","stars":5699},{"date":"2026-11-02","stars":6020},{"date":"2026-11-03","stars":6268},{"date":"2026-11-04","stars":6510},{"date":"2026-11-05","stars":6793},{"date":"2026-11-06","stars":7321},{"date":"2026-11-07","stars":7783},{"date":"2026-11-08","stars":8111},{"date":"2026-11-09","stars":8349},{"date":"2026-11-10","stars":8549}],"firstPublishedAt":"2026-10-29T01:54:21.336Z","commitSha":"a9d66a5453b0f8e5272723e16f6723d56149ece7","verifiedAt":"2026-11-03T02:40:28.137Z","lastCommitAt":"2026-11-10T22:27:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Plays Minecraft Java 26.3 through a from-scratch Rust rewrite, using your own paid copy and Microsoft account.","summary":"Horizon is an unofficial rewrite of Minecraft: Java Edition 26.3 in Rust, meant to run the same worlds and rules with better graphics options and no Java. You download the Horizon launcher for Windows, macOS or Linux, sign in with a Microsoft account that owns Java Edition, and it prepares the game from your own copy, keeps itself updated and can also host a server. No Minecraft code or assets are included. It is for Minecraft players with a graphics card that supports Vulkan, OpenGL 3.3 or DirectX 12.","easeRationale":"On Windows you unpack the launcher zip and run it, then sign in with the Microsoft account that owns Java Edition and press Play; the launcher downloads and prepares everything else. macOS has a .dmg and Linux an AppImage with the same one sign-in step; macOS on Intel only arrives \"from the next Nightly\" according to the README.","costNote":"Horizon is free and MIT licensed, but it only works if you already own a paid copy of Minecraft: Java Edition 26.3.","notVerified":["Whether the game really behaves like vanilla outside the tested harnesses, given the 833 player-visible gaps the project itself lists","How it performs on a given graphics card, and whether crashes noted for some Intel and old CPUs still occur","Whether Minecraft servers and their anti-cheat accept a third-party client","What the built-in telemetry sends and whether it can be switched off","How Mojang and Microsoft will treat the project, which is 30 days old and run by a brand-new account"]},"zh-Hans":{"summaryShort":"通过一个从零用 Rust 重写的版本来玩 Minecraft Java 26.3，使用你自己购买的正版游戏和微软账户。","summary":"Horizon 是一个用 Rust 从零重写的非官方 Minecraft: Java Edition 26.3，目标是运行相同的世界和规则，提供更好的图形选项，并且不需要 Java。你下载适用于 Windows、macOS 或 Linux 的 Horizon 启动器，用拥有 Java Edition 的微软账户登录，它会基于你自己的游戏副本准备游戏，自动保持更新，还可以搭建服务器。它不包含任何 Minecraft 的代码或资源。它面向拥有支持 Vulkan、OpenGL 3.3 或 DirectX 12 显卡的 Minecraft 玩家。","easeRationale":"在 Windows 上，你解压启动器的 zip 文件并运行它，然后用拥有 Java Edition 的微软账户登录，点击“Play”；启动器会下载并准备好其余的一切。macOS 有一个 .dmg 文件，Linux 有一个 AppImage，登录步骤同样只有一步；按 README 所说，Intel 版 macOS 只有“from the next Nightly”才能用上。","costNote":"Horizon 是免费的，采用 MIT 许可证，但只有你已经拥有一份付费的 Minecraft: Java Edition 26.3 才能使用。","notVerified":["考虑到项目自己列出的 833 个玩家可见差异，这个游戏在测试框架之外是否真的表现得和原版一样","在特定显卡上的实际表现如何，一些 Intel 和老旧 CPU 上报告的崩溃是否依然存在","Minecraft 服务器和它们的反作弊系统是否接受第三方客户端","内置的遥测功能会发送什么数据，是否可以关闭","Mojang 和微软会如何对待这个只有 30 天历史、由全新账号运营的项目"]}}},{"id":"R_kgDOOXyxGQ","slug":"horsemail__yourtv","name":"YourTV","owner":"horsemail","fullName":"horsemail/yourtv","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{},"risk":"high","signals":[{"code":"plays_tv_channel_lists_from_outside_sources","severity":"high","evidence":"仓库简介把它称为\"安卓电视直播APK\",并写有\"可自定義源(支持webview://格式)\"。README 要求用户自己提供频道列表,构建说明里提到资源文件\"cloudflare.txt/github_private.txt/sources.txt\"是加密保存的,作者的 APK 使用的加密方式和代码里的不同,\"目的保護我的私有資源信息\"。致谢部分指向了外部的频道列表项目:fanmingming/live、my-tv-0 和 my-tv-1。在很多国家,转播电视频道需要频道所有者的授权;仓库没有说明这些列表里的流来自哪里,也没有说明是否获得了授权。","source":"ai","title":"播放来自外部来源的电视频道列表"},{"code":"published_app_does_not_match_the_source_code","severity":"attention","evidence":"README:\"我上傳的APK文件與源碼可能不同步,APK文件比較新,源碼更新一般落後幾天\"以及\"我上傳的APK文件使用的加密解密邏輯與項目內加密解密邏輯...不同,目的保護我的私有資源信息\"。所以你安装的东西没法拿这里公开的内容来核对。","source":"ai","title":"发布的应用和源代码不一致"},{"code":"channel_list_can_change_by_itself","severity":"info","evidence":"FAQ:如果开启了\"應用啟動后更新視頻源\"并设置了源地址,应用会自动更新,\"可能會覆蓋已保存的視頻源文本\"。你安装之后看到的列表,可能会被网上的内容替换。","source":"ai","title":"频道列表可能自行改变"},{"code":"third_party_encryption_website","severity":"info","evidence":"README 指向 https://yourtvcrypto.horsenma.net 来加密和解密源文件,\"兼容Tvbox的接口源加密解密\"。你粘贴在那里的频道列表会离开你的电脑。","source":"ai","title":"第三方加密网站"},{"code":"no_binaries_detected_in_releases","severity":"info","evidence":"Radar 的检查显示 v2.4.3、v2.4.2 和 v2.4.1 都\"no platform binaries\",因为发布文件是安卓 APK,不是桌面安装程序。","source":"ai","title":"发布版本中未检测到二进制文件"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","hardware"],"alternativeTo":[{"slug":"tivimate","name":"TiviMate","confirmed":true}],"declaredAlternatives":[],"stars":1357,"downloads":38740,"downloadsWeek":619,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":1357},{"date":"2026-11-06","stars":1357},{"date":"2026-11-07","stars":1358},{"date":"2026-11-08","stars":1359},{"date":"2026-11-09","stars":1358},{"date":"2026-11-10","stars":1357}],"firstPublishedAt":"2026-11-04T22:47:04.174Z","commitSha":"16ab601d640eff080d501ef00ca1fabc10b1d8f5","verifiedAt":"2026-11-04T22:41:00.469Z","lastCommitAt":"2026-07-03T04:45:45Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Android TV and phone app that plays live TV channel lists and web video streams you point it at.","summary":"YourTV is an Android app (TV boxes, smart TVs and phones, Android 6 or newer) for watching live TV channels. It plays IPTV channel lists and web video pages, using Tencent's X5 browser engine for the `webview://` sources, and supports picture-in-picture and playing with the screen off. You add your own channel lists, or use the ones the author bundles in encrypted files. It installs as an APK, usually with `adb install` or Xiaomi's TV helper. Free, MIT licensed, with documentation in Chinese and an English version (README.en.md).","easeRationale":"It is an Android-only app, so none of the desktop platforms get a grade. On Android the releases hold a ready APK, but it is not in any store: the README says \"建議通過ADB進行安裝\" (install with `adb install YourTV.apk`) or with the Xiaomi TV helper, and you may still have to paste in a channel list.","costNote":"The app is free (MIT). Channel lists come from elsewhere and are not part of the app.","notVerified":["Whether the built-in channel lists still play anything","What the published APK really does, since the author says it differs from this source code","Where the app connects to and what it sends, including the Tencent X5 web engine","Whether it works on your particular TV box or Android version","Whether sideloading an APK breaks the terms of your TV maker"]},"zh-Hans":{"summaryShort":"一款安卓电视和手机应用,播放你指定的直播频道列表和网页视频流。","summary":"YourTV 是一款安卓应用(电视盒子、智能电视和手机,安卓 6 及以上版本),用来看直播电视频道。它播放 IPTV 频道列表和网页视频页面,对 `webview://` 来源使用腾讯的 X5 浏览器引擎,支持画中画和熄屏播放。你可以添加自己的频道列表,也可以使用作者打包在加密文件里的列表。安装方式是 APK,通常用 `adb install` 或小米的电视助手。免费,采用 MIT 许可证,文档有中文版,也有英文版(README.en.md)。","easeRationale":"这是一款只支持安卓的应用,所以没有任何桌面平台会被评级。在安卓上,各个发布版本都带有现成的 APK,但它不在任何应用商店里:README 说\"建議通過ADB進行安裝\"(用 `adb install YourTV.apk` 安装)或用小米的电视助手,而且你可能还是要自己粘贴一份频道列表。","costNote":"这款应用是免费的(MIT)。频道列表来自别处,不是应用本身的一部分。","notVerified":["内置的频道列表是否还能播放出什么内容","发布的 APK 到底做了什么,因为作者说它和这份源代码不一样","这个应用连接到哪里、发送了什么,包括腾讯的 X5 网页引擎","它在你具体的电视盒子或安卓版本上是否能用","绕过应用商店安装 APK 是否违反你电视厂商的条款"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2NDg2ODI2","slug":"hrydgard__ppsspp","name":"PPSSPP","owner":"hrydgard","fullName":"hrydgard/ppsspp","type":"ready_app","status":"alert","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"high","signals":[{"code":"you_must_supply_your_own_game_files","severity":"attention","evidence":"这个项目提供的是模拟器,不是游戏:仓库里不包含任何游戏,README 指向 report.ppsspp.org/games 上的 \"community compatibility feedback\",发行说明描述的是加载用户自己带来的文件(\"Prevent trying to load obviously-corrupt CSO/CHD files\"、\"Basic support for extracting ISO from 7zip files\")。关于复制游戏的规定因国家而异。","source":"ai","title":"你必须自己提供游戏文件"},{"code":"reads_games_protected_by_the_consoles_encryption","severity":"attention","evidence":"发行说明提到 \"Fix crash in UI when viewing a directory with multiple NPDRM ISOs\" 和 \"libkirk concurrency fixes\"。NPDRM 是索尼为 PSP 下载内容设置的版权保护,libkirk 是主机的加密引擎,所以这个程序会读取主机本来锁住的游戏文件。有些国家的法律限制规避版权保护措施。","source":"ai","title":"读取受主机加密保护的游戏"},{"code":"no_psp_system_file_needed","severity":"info","evidence":"README:\"No BIOS file required to play, PPSSPP is an \\\"HLE\\\" emulator. Default settings balance good compatibility and speed.\" 和许多模拟器不同,它不要求你从真实主机上复制软件。","source":"ai","title":"不需要 psp 系统文件"},{"code":"license_file_mixes_two_licenses","severity":"info","evidence":"GitHub 显示没有明确的许可证(NOASSERTION),因为 LICENSE.TXT 开头是 \"与 BSD 兼容的 PSPSDK 版权声明\",针对 \"PSPSDK 的部分内容 ...(定义、常量、头文件)\",之后才是 \"GNU GENERAL PUBLIC LICENSE Version 2\"。检查工具从这个文件里读到的是 GPL-2.0,README 也说项目 \"Originally released under the GPL 2.0 (and later) in November 2012\"。","source":"ai","title":"许可证文件混合了两种许可证"},{"code":"multiplayer_depends_on_fan_run_servers","severity":"info","evidence":"发行说明:\"Added Infrastructure multiplayer support, with automatic DNS configuration\",并注明 \"Only games that people have made \\\"revival servers\\\" for will work\",还有 \"Support for \\\"data.json\\\" live status from adhoc relay servers\"。联机对战会把你的游戏流量发送经过志愿者运营的服务器。","source":"ai","title":"联机对战依赖粉丝运营的服务器"},{"code":"development_builds_offered_outside_releases","severity":"info","evidence":"README:\"To download fresh development builds for Android, Windows and Mac, go to the /devbuilds page\"。这些文件来自项目自己的网站,而不是 GitHub 上的 80 个发行版,而且最新的正式发行版 v1.20.4 是 2026-05-16 发布的,这个包里记录的最后一次提交却是 2026-11-06。","source":"ai","title":"发行版之外还提供开发版构建"},{"code":"many_open_issues","severity":"info","evidence":"仓库有 958 个未解决的 issue,同时有 14843 个星标,发行说明是一长串按游戏逐条列出的修复(\"Fix rendering bug caused by CPU emulation bug in ATV Offroad Fury\"、\"Partially fix rendering problem in Mahjong Artifacts\")。某个游戏是否跑得好是按游戏逐个跟踪的,并不是整体的保证。","source":"ai","title":"未解决的 issue 很多"},{"code":"reads_game_files_the_console_keeps_encrypted","severity":"high","evidence":"包里的发行说明:\"Fix crash in UI when viewing a directory with multiple NPDRM ISOs\" 和 \"libkirk concurrency fixes\"。NPDRM 是索尼为 PSP 下载内容设置的保护系统,kirk 是主机的加密引擎,所以这个程序会打开主机本来加密锁住的游戏文件。许多国家的法律限制规避版权保护措施,README 也补充说 \"No BIOS file required to play, PPSSPP is an \"HLE\" emulator.\"","source":"ai","title":"读取主机加密保护的游戏文件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","ffmpeg"],"disk":{"minMB":19,"maxMB":58,"basis":"installer"},"apiKeys":["RetroAchievements (only for achievements)"],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"file"},"categories":["games"],"alternativeTo":[{"slug":"playstation-plus-premium","name":"PlayStation Plus Premium","confirmed":true}],"declaredAlternatives":[],"stars":14877,"downloads":624844,"downloadsWeek":4067,"downloadsDays":3,"growth":49,"growthDays":6,"history":[{"date":"2026-11-07","stars":14853},{"date":"2026-11-08","stars":14861},{"date":"2026-11-09","stars":14869},{"date":"2026-11-10","stars":14877}],"firstPublishedAt":"2026-11-06T23:09:33.794Z","commitSha":"7c3807b425571801555c359a255020114e846e52","verifiedAt":"2026-11-06T23:03:21.995Z","lastCommitAt":"2026-11-10T20:51:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A program that plays PlayStation Portable games on your computer, at higher resolution than the handheld ever showed.","summary":"PPSSPP runs PlayStation Portable games on a PC, Mac or Linux machine, with sharper graphics, texture upscaling, save states, gamepad support and ad-hoc multiplayer through community relay servers. Each release has a ready file for Windows (a 19 MB zip), macOS (a 36 MB zip) and Linux (an AppImage of about 48 MB), and there are builds for Android and iOS too. No PSP system file is needed, but the games are not included: you supply your own game files. The repository was created in 2012, the last release is v1.20.4 (2026-05-16) and the last commit is from 2026-11-06.","easeRationale":"Windows is the simplest at grade A: the release holds PPSSPP-v1.20.4-Windows-x64.zip (19 MB, also an ARM64 build), you unzip it and open the program. macOS gets PPSSPPSDL-macOS-v1.20.4.zip (36 MB) and Linux gets a single AppImage file (about 48 MB for x86_64, 46 MB for ARM), both of which also just need to be opened, so all three stay at A. Getting your own game files is a separate matter the project does not handle.","costNote":"The builds in this repository cost nothing and the code is GPL-2.0. The release notes mention \"Reflect Gold status in Discord Rich Presence\"; nothing else in this package explains what \"Gold status\" is.","notVerified":["Whether the game you want runs correctly, which the project tracks per game on report.ppsspp.org","Whether the download matches the source code here, as the releases carry no checksum or signature in the README","How well it performs on your machine, since graphics quality and speed depend on the graphics card and the chosen backend","Real disk use beyond the 19 to 58 MB of the installers, once game files and save states are added","What the ad-hoc relay servers and the RetroAchievements service record when you use online features"]},"zh-Hans":{"summaryShort":"一个能在电脑上运行 PlayStation Portable 游戏的程序,分辨率比那台掌机本身显示的还要高。","summary":"PPSSPP 可以在 PC、Mac 或 Linux 电脑上运行 PlayStation Portable 游戏,画面更清晰,支持材质放大、存档点、手柄,还能通过社区中继服务器实现临时联机对战。每个发行版都提供现成的文件:Windows 版(19 MB 的压缩包)、macOS 版(36 MB 的压缩包)和 Linux 版(约 48 MB 的 AppImage),还有 Android 和 iOS 的版本。不需要任何 PSP 系统文件,但游戏不包含在内:需要你自己准备游戏文件。仓库创建于 2012 年,最新发行版是 v1.20.4(2026-05-16),最后一次提交是在 2026-11-06。","easeRationale":"Windows 最简单,是 A 级:发行版里有 PPSSPP-v1.20.4-Windows-x64.zip(19 MB,还有 ARM64 版本),解压后打开程序就行。macOS 有 PPSSPPSDL-macOS-v1.20.4.zip(36 MB),Linux 有单个 AppImage 文件(x86_64 约 48 MB,ARM 约 46 MB),两者同样只需要打开,所以三个系统都是 A 级。自己准备游戏文件是项目不负责的另一件事。","costNote":"这个仓库里的构建都是免费的,代码采用 GPL-2.0 许可。发行说明里提到 \"Reflect Gold status in Discord Rich Presence\";这个包里其他地方都没有解释什么是 \"Gold status\"。","notVerified":["你想玩的游戏能不能正常运行,这个由项目在 report.ppsspp.org 上按游戏逐个跟踪","下载的文件是否和这里的源代码一致,因为发行版在 README 里没有提供校验和或签名","在你的机器上表现如何,因为画质和速度取决于显卡和所选的后端","除了安装包的 19 到 58 MB 之外的实际磁盘占用,还要加上游戏文件和存档","使用联机功能时,临时联机中继服务器和 RetroAchievements 服务会记录什么"]}}},{"id":"R_kgDOVnvzgw","slug":"huamidev__gestaltx","name":"GestaltX","owner":"huamidev","fullName":"huamidev/GestaltX","type":"needs_setup","status":"alert","grade":"E","easeByPlatform":{},"risk":"high","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 0 天前","source":"check","title":"所有者账号很新"},{"code":"bypasses_ios_protection_with_an_exploit","severity":"high","evidence":"这款应用的全部意义就在于写入 iOS 所保护的文件 MobileGestalt。README 中列出了 \"Logs & respring | Actual exploit-stage logs and manual NeoSpring respring\"，并说它 \"Powered by worse_query\"，还说 \"SystemGroup access requires iOS 27\"，也就是说它利用了 iOS 27.x 中的一个漏洞来绕过系统自身的限制。","source":"ai","title":"利用漏洞绕过 iOS 的保护"},{"code":"unlocks_features_apple_restricts_by_region_or_model","severity":"high","evidence":"README 功能表中写着 \"Independent region-code options, Apple Intelligence capability flag, and sales-region options for activation attempts on China-mainland devices\" 以及 \"Apple internal feature flags, internal storage view, Security Research Device flag\"。它会修改苹果用来判断某台设备可以使用哪些功能和地区的标记。","source":"ai","title":"解锁苹果按地区或型号限制的功能"},{"code":"can_leave_the_device_unusable","severity":"attention","evidence":"README 免责声明：\"Back up your data before making changes. Modifying MobileGestalt may cause system instability or require a device restore. You are responsible for the changes you apply and the resulting risks.\"","source":"ai","title":"可能导致设备无法使用"},{"code":"stars_do_not_match_the_repository_age","severity":"attention","evidence":"该仓库和所有者账号都是在 2026-11-09（也就是本次检查当天）创建的，但仓库已经有 1392 个星标、524 个复刻（fork）和 3 个发布版本。检查结果中还标注了 \"Owner account created 0 days ago\"。","source":"ai","title":"星标数量与仓库年龄不符"},{"code":"no_app_file_in_the_releases","severity":"info","evidence":"列出的三个发布版本（v1.0.3、v1.0.2、v1.0.1，均发布于 2026-11-09）都标注为 \"no platform binaries\"，所以如果不自己编译并签名，就无法安装任何东西。","source":"ai","title":"发布版本中没有应用文件"}],"platforms":[],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":4604,"downloads":30630,"downloadsWeek":0,"downloadsDays":0,"growth":4604,"growthDays":1,"history":[{"date":"2026-11-10","stars":4604}],"firstPublishedAt":"2026-11-09T23:02:44.031Z","commitSha":"601be665cf31a94d80480de2bdb2f2ac98829e26","verifiedAt":"2026-11-09T22:46:45.348Z","lastCommitAt":"2026-11-10T07:31:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An iPhone and iPad app that edits the hidden MobileGestalt system file to switch on device features Apple keeps off.","summary":"GestaltX is an app for iPhone and iPad running iOS or iPadOS 27.x. It reads and edits MobileGestalt, the hidden file where iOS records what your device is allowed to do, so you can try to turn on things like Dynamic Island, Always-On Display, Apple Intelligence or a different region code. It keeps backups and lets you undo each change. It is for people who tweak their own device: there is no ready app file to download, so you need a Mac with Xcode plus your own Apple signing certificate and an App Group to build and install it.","easeRationale":"It is an iPhone and iPad app, so no desktop system gets a grade. The three releases ship no app file, and the README says you must sign the app and its share extension yourself with a certificate and provisioning profile sharing a common App Group, which means building it in Xcode on a Mac.","notVerified":["Whether the presets really switch on Dynamic Island, Always-On Display or Apple Intelligence on any given iPhone or iPad.","Whether the iOS 27.x hole it relies on still works after an Apple update.","Whether the backup and revert features restore every original value, or how often a device restore is needed after a bad write.","What the app does with the device identifiers it reads to verify an imported plist, and whether anything leaves the device.","Whether modifying your device this way affects your Apple warranty or breaks local rules on region and feature changes."]},"zh-Hans":{"summaryShort":"一款面向 iPhone 和 iPad 的应用，通过修改隐藏的系统文件 MobileGestalt 来打开苹果关闭的设备功能。","summary":"GestaltX 是一款运行在 iOS 或 iPadOS 27.x 上的 iPhone 和 iPad 应用。它读取并修改 MobileGestalt，这是 iOS 用来记录设备被允许做什么的隐藏文件，让你尝试打开灵动岛、常亮显示、Apple Intelligence 或更改地区代码等功能。它会保留备份，并允许你撤销每一次修改。它面向喜欢折腾自己设备的人：没有现成的应用文件可以下载，你需要一台装有 Xcode 的 Mac，以及你自己的 Apple 签名证书和一个 App Group 才能编译并安装它。","easeRationale":"这是一款 iPhone 和 iPad 应用，所以没有桌面系统获得评分。三个发布版本都不附带应用文件，README 说你必须用共享同一个 App Group 的证书和描述文件，自己为应用及其共享扩展签名，也就是说要在 Mac 上用 Xcode 编译。","notVerified":["这些预设是否真的能在具体某台 iPhone 或 iPad 上打开灵动岛、常亮显示或 Apple Intelligence。","它所依赖的 iOS 27.x 漏洞在苹果更新后是否依然有效。","备份和还原功能是否能恢复所有原始数值，以及写入失败后需要多频繁地做设备还原。","应用读取设备标识符来验证导入的 plist 文件时会如何处理这些信息，以及是否有任何数据离开设备。","以这种方式修改设备是否会影响苹果的保修，或是否违反当地关于地区和功能更改的规定。"]}}},{"id":"R_kgDOQl_Gyg","slug":"hugohe3__ppt-master","name":"PPT Master","owner":"hugohe3","fullName":"hugohe3/ppt-master","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"readme_steers_you_to_api_resellers_it_earns_from","severity":"attention","evidence":"README 的赞助商部分和 .env.example 里的「Sponsor relay shortcuts」推广了带推荐链接和优惠码的付费 API 经销商，例如 \"https://www.packyapi.ai/register?aff=ppt-master\"（\"enter the promo code ppt-master during recharge to get 10% off\"）、\"https://apikey.fan/register?aff=PPT-MASTER\" 和 \"https://runapi.host/register?aff=WMLJ\"。这些经销商转售 Claude、GPT 和 Gemini 的访问权限，\"as low as 7% of official rates\"，并不是厂商自己的渠道。","source":"ai","title":"README 把你引向它能拿到提成的 API 经销商"},{"code":"pdf_support_brings_in_an_agpl_library","severity":"info","evidence":"README 致谢部分：\"The optional PDF converter depends on PyMuPDF, which is licensed under AGPL-3.0 rather than MIT; it is only needed for PDF sources. See conversion.md before redistributing a bundle that includes it.\"","source":"ai","title":"PDF 支持引入了一个 AGPL 许可的库"},{"code":"found_pictures_can_require_a_credit_line","severity":"info","evidence":"README 第 5 步：网页搜索会考虑 \"CC0, Public Domain, Pexels / Pixabay no-attribution licenses, CC BY, and CC BY-SA\"，并且 \"Executor adds a small inline credit whenever the selected image requires attribution\"，可以用 `--strict-no-attribution` 来避免这一点。幻灯片里因此可能带有你需要保留的署名信息。","source":"ai","title":"搜到的图片可能需要署名"},{"code":"the_author_warns_results_depend_on_the_model","severity":"info","evidence":"README「This is a tool, not a wishing well」部分：\"don't expect a finished, perfect deck in one shot\" 以及 \"The cheaper the model, the more there is to do\"；推荐配置是上下文约 100 万令牌的模型，外加付费图片生成。","source":"ai","title":"作者提醒效果取决于所用模型"},{"code":"huge_star_count_with_almost_no_open_issues","severity":"info","evidence":"该仓库创建于 2025-12-10，有 71322 个星标和 5562 个 fork，却只有一个未关闭的 issue，相对于这个规模的项目，公开的问题讨论很少。","source":"ai","title":"星标数极高，但几乎没有未关闭的 issue"},{"code":"update_runs_a_script_that_pulls_code","severity":"info","evidence":"README「Updating Later」部分：对于克隆安装的用户，运行 `python3 skills/ppt-master/scripts/update_repo.py`，它会 \"pulls the latest version and syncs Python dependencies when requirements.txt changes\"。也就是说，你实际运行的代码可能在你没有审查的情况下发生变化。","source":"ai","title":"更新会运行一个拉取代码的脚本"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["An AI model through your assistant (Claude, Kimi, GPT, Gemini)","OpenAI, Google Gemini or another image model, for AI-generated pictures","Pexels or Pixabay, free keys for better photo search"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","ai_skills","productivity"],"alternativeTo":[{"slug":"gamma","name":"Gamma","confirmed":true},{"slug":"microsoft-365-copilot","name":"365 Copilot","confirmed":true}],"declaredAlternatives":[],"stars":75620,"downloads":212453,"downloadsWeek":34794,"downloadsDays":7,"growth":2540,"growthDays":7,"history":[{"date":"2026-10-29","stars":70830},{"date":"2026-10-30","stars":71320},{"date":"2026-10-31","stars":71657},{"date":"2026-11-01","stars":72133},{"date":"2026-11-02","stars":72597},{"date":"2026-11-03","stars":73080},{"date":"2026-11-04","stars":73490},{"date":"2026-11-05","stars":73918},{"date":"2026-11-06","stars":74257},{"date":"2026-11-07","stars":74523},{"date":"2026-11-08","stars":74881},{"date":"2026-11-09","stars":75241},{"date":"2026-11-10","stars":75620}],"firstPublishedAt":"2026-10-29T01:54:21.345Z","commitSha":"5e8856be0da8e6bc85c4aa28acc614478f7b7137","verifiedAt":"2026-10-31T01:42:42.701Z","lastCommitAt":"2026-11-09T13:39:53Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ask an AI assistant to turn a PDF, a document or a topic into a real PowerPoint file you can keep editing.","summary":"PPT Master is a workflow (a \"skill\") you install into an AI assistant such as Claude Code, Cursor or Codex. You put a PDF, a Word file or just a topic in a folder, ask for a deck in the chat, and it writes a .pptx with real PowerPoint shapes, charts, tables, transitions and speaker notes, so you can keep editing it in PowerPoint. It can also reuse your own template or fill an existing deck. Everything runs on your computer; you install Python 3.10 or newer, the assistant, and the project files. The AI model is what costs money.","easeRationale":"C on Windows, Mac and Linux: none of the 38 releases is an installer, so you install Python 3.10 or newer, install an AI assistant, download the project (git clone or a ZIP) and run `pip install -r requirements.txt`; after that you only type in a chat window. Windows takes extra steps for PATH and the execution policy, which the project covers in its own docs/windows-installation.md.","costNote":"MIT and free, with \"no PPT subscription on top\" as the README puts it; what you pay is the AI model that drives it, plus any image generation you turn on. The README recommends a large-context model (Kimi K3 or Claude) and gpt-image-2.5-sunburst or Gemini for pictures, and says cheaper models leave more work for you.","notVerified":["Whether decks come out like the published examples when you use a cheaper model or your own template.","How many tokens, and how much money, one deck costs on your plan.","Whether the Windows guide is enough on a machine with no development tools.","What the scripts send to the image, search and speech services you enable.","Whether the \"fill an existing .pptx\" route really keeps untouched pages byte-for-byte as the README claims."]},"zh-Hans":{"summaryShort":"让 AI 助手把一份 PDF、一份文档或一个主题，变成一个可以继续编辑的真正的 PowerPoint 文件。","summary":"PPT Master 是一个安装到 Claude Code、Cursor 或 Codex 等 AI 助手里的工作流（一个「技能」）。你把一份 PDF、一份 Word 文件，或者只是一个主题放进文件夹，在聊天里要求做一份演示文稿，它就会生成一个带有真正 PowerPoint 形状、图表、表格、过渡效果和演讲者备注的 .pptx 文件，你可以继续在 PowerPoint 里编辑它。它还能重用你自己的模板，或者填充一份已有的演示文稿。一切都在你的电脑上运行；你需要安装 Python 3.10 或更新版本、助手本身，以及项目文件。真正花钱的是 AI 模型。","easeRationale":"在 Windows、Mac 和 Linux 上都是 C 级：38 个发布版本里没有一个是安装程序，所以你要安装 Python 3.10 或更新版本，安装一个 AI 助手，下载项目（git clone 或 ZIP），再运行 `pip install -r requirements.txt`；之后你只需要在聊天窗口里输入文字。Windows 还需要额外处理 PATH 和执行策略，项目在自己的 docs/windows-installation.md 里有说明。","costNote":"采用 MIT 许可，免费，README 的说法是「no PPT subscription on top」（不用额外订阅 PPT 软件）；真正要花钱的是驱动它的 AI 模型，以及你开启的任何图片生成功能。README 推荐使用大上下文模型（Kimi K3 或 Claude），图片用 gpt-image-2.5-sunburst 或 Gemini，并说更便宜的模型会留给你更多要做的事。","notVerified":["用更便宜的模型或你自己的模板时，做出来的演示文稿是否和发布的示例一样好。","在你的套餐里，做一份演示文稿要花多少令牌、多少钱。","在没有开发工具的机器上，Windows 安装指南是否够用。","脚本会向你开启的图片、搜索和语音服务发送什么内容。","「填充已有 .pptx」这条路径是否真的像 README 所说那样，逐字节保留未改动的页面。"]}}},{"id":"R_kgDOVOnfRA","slug":"humblemuzzu__dumbtv","name":"DumbTV","owner":"humblemuzzu","fullName":"humblemuzzu/dumbtv","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"attention","signals":[{"code":"disables_system_apps_on_your_tv","severity":"attention","evidence":"README：\"About 50 [Amazon packages] disabled: ads, sponsored content, ACR, telemetry, crash and usage uploads, experiments, update nags. Disabled, never deleted.\" 运行过程会让电视重启两次；./scripts/restore.sh 可以重新启用它们，恢复出厂设置也会清除一切。","source":"ai","title":"停用你电视上的系统应用"},{"code":"blocks_system_updates","severity":"attention","evidence":"README 的 \"Good to know\" 部分：\"Fire OS updates are blocked while the DNS filter runs, since an update would switch a lot of this back on.\" DNS 过滤器会屏蔽约 400 个亚马逊的遥测、广告和更新主机。","source":"ai","title":"阻止系统更新"},{"code":"downloads_third_party_apps","severity":"info","evidence":"LICENSE 和 README：来自 F-Droid 的 personalDNSfilter（GPL-2.0）、HaGeZi 的拦截列表（GPL-3.0）和 SmartTube（MIT）并未随项目打包，而是由脚本从它们的上游发布页下载；项目还提供用 SmartTube 替换 YouTube 应用（可去除 YouTube 的广告）的选项，默认关闭。","source":"ai","title":"下载第三方应用"},{"code":"coding_agent_runs_the_steps","severity":"info","evidence":"README 快速开始：你可以克隆仓库，在其中打开一个编程智能体并说 \"Clean up my Fire TV\"；智能体会读取 AGENTS.md，并替你在电视上运行这些脚本。","source":"ai","title":"由编程智能体来执行这些步骤"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[{"target":"projectivy-launcher","stance":"claims","evidence":"- Replaces the home screen with [dumbtv home](launcher/README.md): your apps,"}],"stars":2024,"downloads":7965,"downloadsWeek":1529,"downloadsDays":7,"growth":112,"growthDays":7,"history":[{"date":"2026-10-27","stars":1685},{"date":"2026-10-28","stars":1722},{"date":"2026-10-29","stars":1759},{"date":"2026-10-30","stars":1792},{"date":"2026-10-31","stars":1824},{"date":"2026-11-01","stars":1856},{"date":"2026-11-02","stars":1893},{"date":"2026-11-03","stars":1912},{"date":"2026-11-04","stars":1939},{"date":"2026-11-05","stars":1960},{"date":"2026-11-06","stars":1970},{"date":"2026-11-07","stars":1980},{"date":"2026-11-08","stars":2000},{"date":"2026-11-09","stars":2017},{"date":"2026-11-10","stars":2024}],"firstPublishedAt":"2026-10-27T18:40:27.917Z","commitSha":"755991b17a66e753e82f7252368c8850c08fdc71","verifiedAt":"2026-10-27T18:35:41.804Z","lastCommitAt":"2026-11-07T15:44:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Strips the ads, tracking and clutter out of a Fire TV from your computer, and puts it all back with one command.","summary":"dumbtv is a set of scripts you run from your computer against a Fire TV stick, Cube or Fire TV Edition television over ADB, with no root. It disables about 50 Amazon packages for ads, sponsored rows, screen fingerprinting and telemetry, turns off the ad and usage-reporting settings, puts a small DNS filter on the TV for what cannot be disabled, and can replace the home screen with its own launcher showing your apps and HDMI inputs. Your streaming apps stay. One script undoes everything. You need adb and about ten minutes; on Windows, WSL or Git Bash.","easeRationale":"You install Android platform-tools (adb) on your computer, turn on ADB debugging in the TV's hidden developer menu, and run four shell scripts that ask before each step; on Windows that means WSL or Git Bash.","costNote":"MIT licensed scripts and launcher; the catalog of devices is CC BY-SA 4.0.","notVerified":["Whether it works on your exact TV model: the README lists about two dozen devices tested by their owners, some only partly.","Whether a future Fire OS update puts the disabled packages and settings back.","What the TV loses in practice from the disabled packages beyond what the project documents.","Whether Amazon treats the change as a reason to refuse support or warranty on the device."]},"zh-Hans":{"summaryShort":"从你的电脑上清除 Fire TV 里的广告、追踪和杂乱内容，还能用一条命令全部恢复。","summary":"dumbtv 是一套脚本，你在自己的电脑上通过 ADB 对 Fire TV Stick、Cube 或 Fire TV Edition 电视进行操作，不需要 root。它会禁用约 50 个 Amazon 软件包，涉及广告、赞助内容行、屏幕指纹识别和遥测数据，关闭广告和使用情况上报设置，为无法禁用的部分在电视上加一个小型 DNS 过滤器，还可以把主屏幕替换成自带的启动器，显示你的应用和 HDMI 输入源。你的流媒体应用不受影响。一条脚本就能撤销全部改动。你需要 adb 和大约十分钟时间；在 Windows 上要用 WSL 或 Git Bash。","easeRationale":"你需要在电脑上安装 Android platform-tools（adb），在电视隐藏的开发者菜单里打开 ADB 调试，然后运行四个会在每一步之前询问确认的 shell 脚本；在 Windows 上这意味着要用 WSL 或 Git Bash。","costNote":"脚本和启动器采用 MIT 许可证；设备目录采用 CC BY-SA 4.0 许可证。","notVerified":["是否适用于你的具体电视型号：README 列出了大约二十来款由用户自己测试过的设备，其中一些只测试了部分功能。","未来的 Fire OS 更新是否会把已禁用的软件包和设置重新恢复。","除了项目文档里写的内容之外，电视实际还会因为这些被禁用的软件包失去哪些功能。","亚马逊是否会把这种改动当作拒绝提供支持或保修的理由。"]}}},{"id":"R_kgDOVawfcQ","slug":"humpalumps__ultra-water","name":"Ultra Water","owner":"Humpalumps","fullName":"Humpalumps/Ultra-Water","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"brand_new_no_releases","severity":"info","evidence":"项目信息：创建于 2026-10-28，最近一次推送为 2026-10-28，GitHub 上有 0 个发布版本、0 个未关闭的 issue，却已有 318 个星标。目前还没有可用来评判它的版本历史。","source":"ai","title":"全新项目，尚无发布版本"},{"code":"not_packaged_for_reuse","severity":"info","evidence":"package.json 被标记为 \"private\": true，并且还留着另一个项目的标识：\"name\": \"gelatin-game-engine\"，\"description\": \"Gelatin Game Engine\"。它没有作为可安装的包发布，所以要复用就得手动复制 src/ 里的文件。","source":"ai","title":"没有打包成可复用的形式"},{"code":"built_with_ai_agents","severity":"info","evidence":"README：\"The design was argued out with a second agent before building, and again for each round of upgrades.\" 仓库顶层还有一个 .claude/ 文件夹。","source":"ai","title":"由 AI 智能体参与开发"},{"code":"performance_numbers_self_measured","severity":"info","evidence":"README 自己的基准测试部分承认之前公布的数字有误（\"Earlier versions of this README quoted numbers 2-4x lower\"），并称最新的泡沫功能 \"hasn't been re-measured cleanly yet\"。帧率来自作者在页面内自行运行的基准测试，只用了一台电脑和一部手机。","source":"ai","title":"性能数据由作者自己测得"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","games"],"alternativeTo":[],"declaredAlternatives":[],"stars":657,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":38,"growthDays":7,"history":[{"date":"2026-10-30","stars":439},{"date":"2026-10-31","stars":500},{"date":"2026-11-01","stars":536},{"date":"2026-11-02","stars":580},{"date":"2026-11-03","stars":619},{"date":"2026-11-04","stars":627},{"date":"2026-11-05","stars":634},{"date":"2026-11-06","stars":640},{"date":"2026-11-07","stars":643},{"date":"2026-11-08","stars":649},{"date":"2026-11-09","stars":653},{"date":"2026-11-10","stars":657}],"firstPublishedAt":"2026-10-30T01:57:54.307Z","commitSha":"ce9248fa20afbdf63fd4c11709f68d7368aca0a0","verifiedAt":"2026-10-30T01:42:13.924Z","lastCommitAt":"2026-11-02T13:23:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Code that draws a living ocean in a web browser: waves, breaking surf, foam, light patterns underwater and floating objects.","summary":"A water renderer for three.js, the 3D library used in browser games and visualisations. It comes with a demo island scene you can fly around: wind waves, waves that bend and break near the shore, foam, sunlight patterns on the seabed, an underwater view with sun shafts, and crates and boats that bob on the swell. It is source code for people building their own 3D scene, not a finished program. You install Node, run two commands and open the page; phones and tablets get a lighter setting automatically.","easeRationale":"There is no download to install: on any of the three systems you need Node installed, then `npm install` and `npm run dev`, and you open the address the terminal prints (README). The scene itself needs a browser with WebGL2 and a reasonably strong graphics card; the author's numbers come from an RTX 3060.","notVerified":["Whether the quoted 200+ fps holds on a different graphics card, browser or operating system","How it behaves in browsers other than Chrome, and on Macs or iPhones","How much disk the npm install takes (the estimate of 100-500 MB is a guess, not a measurement)","How much work it really is to move the water into an existing three.js project, since there is no documented interface"]},"zh-Hans":{"summaryShort":"在网页浏览器里绘制一片活灵活现的海洋的代码:波浪、拍岸浪花、泡沫、水下光影和漂浮的物体。","summary":"一个面向 three.js 的水面渲染器,three.js 是浏览器游戏和可视化中常用的 3D 库。它自带一个可以飞行环绕的演示海岛场景:风浪、靠近海岸时弯曲破碎的浪花、泡沫、海床上的阳光光斑、带光柱的水下视角,以及随波起伏的木箱和船只。这是给自己搭建 3D 场景的人用的源代码,不是一个成品程序。你需要安装 Node,运行两条命令并打开网页;手机和平板会自动切换到较轻量的画质设置。","easeRationale":"没有可下载安装的程序:三大系统上都需要先安装 Node,然后运行 `npm install` 和 `npm run dev`,再打开终端打印出的地址(见 README)。场景本身需要支持 WebGL2 的浏览器和一块相当强劲的显卡;作者给出的数据来自 RTX 3060。","notVerified":["所说的 200+ fps 在不同显卡、浏览器或操作系统上是否依然成立","在 Chrome 以外的浏览器,以及在 Mac 或 iPhone 上表现如何","npm install 占用多少磁盘空间(100-500 MB 只是估计,并非实际测量)","把这个水面效果移植到现有 three.js 项目中实际要花多少工夫,因为没有现成的文档化接口"]}}},{"id":"R_kgDOTne_vw","slug":"hypit-ai__hypit","name":"Hypit","owner":"hypit-ai","fullName":"hypit-ai/hypit","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/drive/u/2/folders/18J9Fz7mkU3OQNJ-2Res3eIyFQ2cemIK5","source":"check","title":"外部下载链接"},{"code":"custom_restrictive_license","severity":"attention","evidence":"LICENSE 是 \"a modified version of the Apache License 2.0\"，禁止以 \"for commercial gain\" 为目的出售或捆绑 Hypit，也禁止在没有商业许可证的情况下把它作为托管服务运行，并且写明 \"The producer can adjust the open-source agreement to be more strict\"。","source":"ai","title":"自定义的限制性许可证"},{"code":"video_cloning_and_face_swap","severity":"attention","evidence":"描述：\"Clone any viral video with AI agents... swap the face, the words, the B-roll, ship 100 variants in one command\"；README 建议克隆 \"a winning ad from the Meta Ad Library\"。复制他人的视频或面孔可能引发版权和肖像权问题。","source":"ai","title":"视频克隆与换脸"},{"code":"affiliate_referral_links","severity":"info","evidence":"README 中的合作方链接带有推荐码，例如 `https://monid.ai?fpr=alvin-45480b`、`https://www.hiapi.ai/invite/hypit`、`https://beatapi.io/sign-up?aff=hypit`。","source":"ai","title":"带推荐码的联盟推广链接"},{"code":"program_fetched_on_first_use","severity":"info","evidence":"README：\"On first use, your agent checks for the Hypit executable and helps prepare it if needed.\"","source":"ai","title":"首次使用时才获取程序"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg","python","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["AI coding assistant (Claude Code, Codex or similar)","AI image, voice and video generation service (HypiHub or another provider)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["video","social_marketing","ai_skills"],"alternativeTo":[{"slug":"invideo-ai","name":"AI","confirmed":true},{"slug":"opusclip","name":"OpusClip","confirmed":true}],"declaredAlternatives":[{"target":"word","stance":"claims","evidence":"narrator to banana cat, flip the rankings and Ronaldo becomes GOAT, or replace all players with tech founders — same viral structure, different viral video.</p><p>Total cost: <b>$1.15</b>.</p></td>"},{"target":"invideo-ai","stance":"claims","evidence":"Pepe and Doge arguing about doggy arms, replace the creatine with retinol where a pretty boy roasts a tomboy's pores, or ditch the physical product for a CheatGPT app where a himbo jock out-GPAs a PhD"}],"stars":29064,"downloads":3076,"downloadsWeek":827,"downloadsDays":7,"growth":1313,"growthDays":7,"history":[{"date":"2026-10-26","stars":25879},{"date":"2026-10-27","stars":26086},{"date":"2026-10-28","stars":26257},{"date":"2026-10-29","stars":26509},{"date":"2026-10-30","stars":26892},{"date":"2026-10-31","stars":27131},{"date":"2026-11-01","stars":27368},{"date":"2026-11-02","stars":27562},{"date":"2026-11-03","stars":27751},{"date":"2026-11-04","stars":27975},{"date":"2026-11-05","stars":28158},{"date":"2026-11-06","stars":28304},{"date":"2026-11-07","stars":28444},{"date":"2026-11-08","stars":28605},{"date":"2026-11-09","stars":28783},{"date":"2026-11-10","stars":29064}],"firstPublishedAt":"2026-10-27T01:10:12.636Z","commitSha":"e26d9142fd8a6f78bf4f3cdb800d6205f4ecca16","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-09T14:32:26Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Lets an AI coding assistant copy the structure of a short video and make many new versions with new faces, words and clips.","summary":"Hypit is an add-on for AI coding assistants such as Claude Code or Codex. You give the assistant a short video or a description, and it builds an editable recipe with captions, clips and effects, then renders the finished video, including many variants at once. It is aimed at people making social ads, short clips and talking-head videos. You need Node.js, a coding assistant account, a terminal, and usually paid AI image, voice and video services. The README shows example videos costing about $1 each in service fees.","easeRationale":"There is no installer: you run `npx skills add hypit-ai/hypit -g` in a terminal, need Node.js 22.15+, a coding assistant, and the assistant then sets up the Hypit program and media tools.","costNote":"Hypit itself has no fee; the coding assistant and AI generation services are billed separately (README examples: $1.07 to $1.15 per video).","notVerified":["Whether it works on Windows; the README gives no platform list.","Which extra tools (FFmpeg, Python for WhisperX, headless Chrome) the assistant installs during setup.","Real cost per video with your chosen AI services and coding assistant.","What is inside the Google Drive avatar pack linked in the README.","What commands the coding assistant runs on your computer while preparing Hypit."]},"zh-Hans":{"summaryShort":"让 AI 编程助手复制一段短视频的结构，并用新的人脸、文字和片段批量生成多个新版本。","summary":"Hypit 是 Claude Code、Codex 等 AI 编程助手的附加组件。你给助手一段短视频或一段描述，它会生成一份可编辑的“配方”，包含字幕、片段和特效，然后渲染出成品视频，还能一次生成多个变体。它面向制作社交媒体广告、短视频和口播视频的人。你需要 Node.js、一个编程助手账号、终端，通常还需要付费的 AI 图像、语音和视频服务。README 中的示例视频每条的服务费用约为 1 美元。","easeRationale":"没有安装程序：你要在终端中运行 `npx skills add hypit-ai/hypit -g`，需要 Node.js 22.15 及以上版本和一个编程助手，之后由助手来安装 Hypit 程序和媒体工具。","costNote":"Hypit 本身不收费；编程助手和 AI 生成服务另行计费（README 示例：每条视频 1.07 至 1.15 美元）。","notVerified":["是否能在 Windows 上运行；README 没有给出支持的平台列表。","助手在设置过程中会安装哪些额外工具（FFmpeg、用于 WhisperX 的 Python、无头 Chrome）。","使用你所选的 AI 服务和编程助手时，每条视频的真实成本。","README 中链接的 Google Drive 头像包里有什么内容。","编程助手在准备 Hypit 的过程中会在你的电脑上运行哪些命令。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyODQ0NjM1MTM=","slug":"iamvishnusankar__next-sitemap","name":"next-sitemap","owner":"iamvishnusankar","fullName":"iamvishnusankar/next-sitemap","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2023-09-06（1157 天前）","source":"check","title":"发布版本已过时"},{"code":"code_is_newer_than_the_last_published_version","severity":"attention","evidence":"检查记录显示最后一次发布是在 2023-09-06（v4.2.3），而最后一次提交是在 2026-03-10，总共有 236 次发布。使用 \"yarn add next-sitemap\" 安装的人得到的是 2023 年的版本，而不是这个仓库中的代码，README 也没有说明此后发生了哪些变化。","source":"ai","title":"代码比最后发布的版本更新"},{"code":"robots_txt_entries_are_public_and_only_a_request","severity":"info","evidence":"完整示例写入的 robots.txt 包含 \"User-agent: black-listed-bot / Disallow: /sub-path-1\"，并在 sitemap 中排除了 \"'/protected-page', '/awesome/secret-page'\"。任何人都可以读取 robots.txt 文件，遵守规则的爬虫才会选择遵循它；在其中列出私密路径并不能阻止对它的访问，不把它放进 sitemap 同样不能把它隐藏起来。","source":"ai","title":"robots txt 中的条目是公开的，只是一种请求"},{"code":"settings_file_runs_your_own_code_at_build_time","severity":"info","evidence":"配置文件是 JavaScript，包含 \"transform: async (config, path) => ...\" 和 \"additionalPaths: async (config) => ...\" 函数，README 中提到 \"next-sitemap will load environment variables from .env files by default\"。你在其中写的任何内容都会在构建期间运行，并能访问这些变量。","source":"ai","title":"配置文件会在构建时运行你自己的代码"},{"code":"only_works_with_nextjs","severity":"info","evidence":"该工具读取 Next.js 的构建文件夹（\"sourceDir (optional) | next.js build directory. Default .next\"），服务器端辅助函数是 Next.js 的路由处理器（getServerSideSitemap、getServerSideSitemapIndex）。对于用其他技术构建的网站，它什么也做不了。","source":"ai","title":"只能用于 nextjs"},{"code":"sitemap_values_are_guesses_by_default","severity":"info","evidence":"选项表中的默认值是 \"changefreq ... Default daily\"、\"priority ... Default 0.7\" 和 \"autoLastmod ... Default true\"，示例中的 transform 把 lastmod 设为 \"new Date().toISOString()\"。于是，无论页面是否真的发生变化，每个页面都会被标注为每天更新，并且在构建时刻被修改过。","source":"ai","title":"sitemap 的值默认是猜测出来的"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","social_marketing"],"alternativeTo":[{"slug":"screaming-frog-seo-spider","name":"SEO Spider","confirmed":true}],"declaredAlternatives":[],"stars":3751,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":3750},{"date":"2026-11-08","stars":3750},{"date":"2026-11-09","stars":3750},{"date":"2026-11-10","stars":3751}],"firstPublishedAt":"2026-11-06T22:57:16.672Z","commitSha":"144fad921e8ee30f03c73ed72208bfa7355eb75d","verifiedAt":"2026-11-06T22:44:40.665Z","lastCommitAt":"2026-03-10T19:47:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A command for Next.js websites that writes the sitemap.xml and robots.txt files search engines look for.","summary":"next-sitemap is for people who build websites with Next.js. After each build it walks through the pages the site produced and writes sitemap.xml and, if you ask, robots.txt into the public folder, splitting very large sitemaps into numbered files and supporting Google news, image and video entries plus other-language versions of a page. You add it to the project with one command, create a small next-sitemap.config.js with your site address, and run it as a postbuild step. It has no screen of its own and only works inside a Next.js project.","easeRationale":"The same on all three systems, grade C: with Node.js already installed you run \"yarn add next-sitemap\", write a next-sitemap.config.js file with \"siteUrl\" and add \"postbuild\": \"next-sitemap\" to package.json. There is nothing to open or click, and it is only useful inside an existing Next.js project; with pnpm you must also add \"enable-pre-post-scripts=true\" to an .npmrc file.","costNote":"MIT licensed with no paid tier mentioned anywhere in the repository.","notVerified":["Whether it works with the newest Next.js versions, since the published package dates from 2023","Whether search engines actually index the pages it lists, which only they decide","What the code committed after the last release changes compared with the published package","Real disk use, which the checks could only estimate at 100 to 500 MB for the repository and its dependencies"]},"zh-Hans":{"summaryShort":"一个用于 Next.js 网站的命令，可生成搜索引擎需要的 sitemap.xml 和 robots.txt 文件。","summary":"next-sitemap 面向使用 Next.js 搭建网站的人。每次构建后，它会遍历网站生成的页面，并在 public 文件夹中写入 sitemap.xml，如果你需要，还会写入 robots.txt，将非常大的 sitemap 拆分成带编号的多个文件，并支持 Google 新闻、图片和视频条目，以及同一页面的其他语言版本。你只需用一条命令把它添加到项目中，创建一个包含网站地址的小型 next-sitemap.config.js 文件，并把它作为 postbuild 步骤运行。它没有自己的界面，只能在 Next.js 项目内部使用。","easeRationale":"三个系统上的体验相同，等级为 C：在已安装 Node.js 的前提下，运行 \"yarn add next-sitemap\"，编写包含 \"siteUrl\" 的 next-sitemap.config.js 文件，并在 package.json 中加入 \"postbuild\": \"next-sitemap\"。没有界面可打开或点击，而且只有在已有的 Next.js 项目中才有用；使用 pnpm 时还需要在 .npmrc 文件中加入 \"enable-pre-post-scripts=true\"。","costNote":"采用 MIT 许可证，仓库中没有提到任何付费方案。","notVerified":["是否适用于最新版本的 Next.js，因为发布的包可追溯到 2023 年","搜索引擎是否真的会收录它列出的页面，这完全由搜索引擎自行决定","最后一次发布之后提交的代码相对于已发布的包有哪些改动","实际磁盘占用，检查只能估算仓库及其依赖项占用 100 到 500 MB"]}}},{"id":"R_kgDOSN5LZA","slug":"ifixai-ai__ifixai","name":"iFixAi","owner":"ifixai-ai","fullName":"ifixai-ai/iFixAi","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"sends_adversarial_requests_to_your_agent","severity":"attention","evidence":"README：\"Use a staging copy, because the probes are real requests.\" 默认运行包含针对提示词注入（`B12`）以及无权提出请求者发起破坏性操作（`P01`）的探测，因此只应对准你有权测试的智能体。","source":"ai","title":"向你的智能体发送对抗性请求"},{"code":"very_young_repository_with_large_star_count","severity":"info","evidence":"该仓库创建于 2026-04-27，已有 37,084 个星标，却只有 3 个未关闭的 issue；README 开头就是 Product Hunt 和 Trendshift 的推广徽章。","source":"ai","title":"非常新的仓库，星标数却很高"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI","Anthropic, Google Gemini, Azure, Bedrock, OpenRouter or Hugging Face (any one can be the judge instead of OpenAI)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents","ai_skills"],"alternativeTo":[{"slug":"langsmith","name":"LangSmith","confirmed":true},{"slug":"braintrust","name":"Braintrust","confirmed":true}],"declaredAlternatives":[],"stars":44510,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1341,"growthDays":7,"history":[{"date":"2026-10-29","stars":37898},{"date":"2026-10-30","stars":39226},{"date":"2026-10-31","stars":41458},{"date":"2026-11-01","stars":42530},{"date":"2026-11-02","stars":43005},{"date":"2026-11-03","stars":43169},{"date":"2026-11-04","stars":43382},{"date":"2026-11-05","stars":43512},{"date":"2026-11-06","stars":43593},{"date":"2026-11-07","stars":43854},{"date":"2026-11-08","stars":44015},{"date":"2026-11-09","stars":44256},{"date":"2026-11-10","stars":44510}],"firstPublishedAt":"2026-10-29T01:54:21.351Z","commitSha":"28dca412279635361d626574810aae8d91ab6b45","verifiedAt":"2026-10-29T01:37:54.067Z","lastCommitAt":"2026-11-09T23:56:39Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Audits your own AI agent: sends it tricky requests and grades how it answers, from A to F.","summary":"A command-line tool for people who build AI agents. It puts your agent in a simulated version of your business, sends the kind of requests a real day brings, including requests that push the agent past what it is allowed to do, and another AI model grades each reply. You get a pass or fail per check, an A-to-F grade and reports in JSON and Markdown. You need Python 3.10 or newer and a paid model key for the judge. A demo mode with canned answers runs with no key. It also installs as a plugin or skill inside Claude Code, Codex, Cursor and similar.","easeRationale":"There are no ready installers in the 23 releases, so on every platform you install Python 3.10 or newer, run `pip install ifixai` and then a few commands; Windows, Mac and Linux are all C. The README adds Windows-specific steps (`py -m venv`, `python -m ifixai.cli.main`) and warns that a stock Mac ships Python 3.9, which is too old.","costNote":"The code is free (Apache-2.0). A real audit calls a paid model as the judge, plus your own agent's API; the README says `--dry-run` shows how many judge calls a run will make before anything is billed.","notVerified":["Whether the audit grades are accurate, or whether a pass here means your agent behaves well outside the simulated scenarios.","What a real audit costs in judge tokens, especially the `--full-run` with all 61 inspections.","Whether the Claude Code plugin, the Codex plugin and the editor skills install and work as described.","Whether the install works on each platform and which Python versions actually run it.","What the tool sends to the judge provider, and whether any of your agent's answers or business data leave your machine during a run."]},"zh-Hans":{"summaryShort":"审查你自己的 AI 代理：给它发送刁钻的请求，并给它的回答打分，从 A 到 F。","summary":"一个面向 AI 代理开发者的命令行工具。它把你的代理放进一个模拟的业务场景中，发送真实一天里可能出现的各种请求，包括那些试图让代理越权的请求，再由另一个 AI 模型给每个回答打分。你会得到每项检查的通过或不通过、一个 A 到 F 的等级，以及 JSON 和 Markdown 格式的报告。你需要 Python 3.10 或更新版本，以及一个用于评判的付费模型密钥。不带密钥时可以用演示模式，用现成的回答运行。它还可以作为插件或技能安装进 Claude Code、Codex、Cursor 等工具里。","easeRationale":"23 个发行版本里没有现成的安装程序，所以在每个平台上你都要安装 Python 3.10 或更新版本，运行 `pip install ifixai`，再执行几条命令；Windows、Mac 和 Linux 的评级都是 C。README 还补充了 Windows 专属的步骤（`py -m venv`、`python -m ifixai.cli.main`），并提醒出厂自带的 Mac 系统自带的是 Python 3.9，版本太旧。","costNote":"代码本身免费（Apache-2.0 许可）。真正的审查会调用一个付费模型作为评判，再加上你自己代理的 API 调用；README 说 `--dry-run` 能在产生任何费用之前，先显示一次运行会发起多少次评判调用。","notVerified":["审查打分是否准确，或者在这里通过是否意味着你的代理在模拟场景之外也表现良好","一次真正的审查在评判模型的 token 上会花多少钱，尤其是包含全部 61 项检查的 `--full-run`","Claude Code 插件、Codex 插件和编辑器技能是否真的能按描述安装并正常工作","安装在每个平台上是否都能成功，以及哪些 Python 版本真的能运行它","这个工具会向评判模型提供方发送什么内容，以及运行过程中你代理的回答或业务数据是否会离开你的电脑"]}}},{"id":"R_kgDOPTY97g","slug":"iflow-ai__iflow-cli","name":"iFlow CLI","owner":"iflow-ai","fullName":"iflow-ai/iflow-cli","type":"dev_tool","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接管道传给 shell 执行：curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"curl 管道直接执行 shell 脚本"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹或 winget 清单都没有检测到许可证","source":"check","title":"没有许可证"},{"code":"service_announced_shutdown","severity":"high","evidence":"README 顶部的提示：\"iFlow CLI will be shutting down on April 17, 2026 (UTC+8)\" ——这个日期已经过去；这个工具依赖 iFlow 的在线服务和账号登录","source":"ai","title":"服务已宣布关闭"},{"code":"program_source_not_in_repository","severity":"attention","evidence":"顶层文件只有 install.sh、install_beta.sh、README 的各语言翻译、docs_en/、docs_cn/、Formula/ 和 assets/；程序本身来自 npm 包 @iflow-ai/iflow-cli","source":"ai","title":"程序源代码不在这个仓库里"},{"code":"installer_downloads_third_party_binaries","severity":"attention","evidence":"install.sh 会从项目自己的服务器下载一个 nvm 包（PACKAGE_URL=\"https://cloud.iflow.cn/iflow-cli/nvm-${VERSION}.tar.gz\"），解压到 $HOME/.nvm，并在你的 .bashrc/.zshrc/config.fish 里追加内容","source":"ai","title":"安装脚本会下载第三方二进制文件"},{"code":"installer_uses_sudo","severity":"attention","evidence":"当系统缺少 xz 时，install.sh 会运行 \"sudo yum install -y xz\" 或 \"sudo apt-get update && sudo apt-get install -y xz-utils\"","source":"ai","title":"安装脚本会使用 sudo"},{"code":"auto_update_at_run_time","severity":"info","evidence":"README 的功能列表写道：\"Auto-upgrade: iFlow CLI automatically detects if current version is latest\"，所以你机器上的代码可能在你没有要求的情况下发生变化","source":"ai","title":"运行时会自动更新"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["iFlow","OpenAI-compatible API"],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["ai_agents","development"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true},{"slug":"cursor","name":"Cursor","confirmed":true}],"declaredAlternatives":[],"stars":5077,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":1,"growthDays":6,"history":[{"date":"2026-11-06","stars":5076},{"date":"2026-11-07","stars":5076},{"date":"2026-11-08","stars":5077},{"date":"2026-11-09","stars":5077},{"date":"2026-11-10","stars":5077}],"firstPublishedAt":"2026-11-05T22:43:48.488Z","commitSha":"4642808afbc6580ac117d930f6c64ac0d84955c7","verifiedAt":"2026-11-05T22:24:46.332Z","lastCommitAt":"2026-03-20T03:00:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"AI assistant for the terminal that reads your files and does coding and file tasks; the service was announced as closing.","summary":"iFlow CLI is an AI assistant you run in the terminal. You type what you want in plain language and it reads the project files, writes code, organizes files or runs small automations, using AI models from iFlow's online platform. You need Node.js 22 or newer, an iFlow account and an internet connection. The top of the README says iFlow CLI shuts down on 17 April 2026, a date already in the past, and the last commit is from March 2026. The repository holds only install scripts and documentation; the program itself is installed from npm.","easeRationale":"On all three systems you install Node.js 22 or newer and then run one install command (`npm install -g @iflow-ai/iflow-cli`, or the one-line shell script on macOS and Linux), then log in to an iFlow account. There is no installer and no ready-made program.","costNote":"The README says the models on the iFlow platform can be used for free after logging in","notVerified":["Whether login, the free models and the Open Market still work after the announced shutdown date","What the npm package actually does, since its source code is not in this repository","Which data is sent to iFlow's servers when it reads your files","Whether the Windows path (npm only, no install script) gives the same features as macOS and Linux"]},"zh-Hans":{"summaryShort":"面向终端的 AI 助手，能读取你的文件并完成编程和文件整理任务；该服务已被宣布即将关闭。","summary":"iFlow CLI 是一个在终端里运行的 AI 助手。你用日常语言说出想要做的事，它就会读取项目文件、写代码、整理文件或运行一些小型自动化任务，使用的是 iFlow 在线平台提供的 AI 模型。你需要 Node.js 22 或更新版本、一个 iFlow 账号和网络连接。README 顶部写着 iFlow CLI 将于 2026 年 4 月 17 日关闭，这个日期已经过去，最后一次提交是在 2026 年 3 月。这个仓库里只有安装脚本和文档；程序本身是通过 npm 安装的。","easeRationale":"三个系统的步骤都一样：安装 Node.js 22 或更新版本，然后运行一条安装命令（`npm install -g @iflow-ai/iflow-cli`，或者在 macOS 和 Linux 上用那条一行的 shell 脚本），再登录一个 iFlow 账号。没有安装程序，也没有现成的程序。","costNote":"README 说登录后可以免费使用 iFlow 平台上的模型","notVerified":["登录、免费模型和 Open Market 在宣布的关闭日期之后是否还能用","这个 npm 包到底做了什么，因为它的源代码不在这个仓库里","它读取你的文件时，会把哪些数据发送到 iFlow 的服务器","Windows 这条路径（只能用 npm，没有安装脚本）能不能提供和 macOS、Linux 一样的功能"]}}},{"id":"R_kgDOSQHxtA","slug":"im-anishraj__instagram-comment-to-dm","name":"Instagram Comment to DM","owner":"im-anishraj","fullName":"im-anishraj/instagram-comment-to-dm","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"meta_approval_needed_before_real_use","severity":"attention","evidence":"需求列表中提到需要一个\"Meta Developer App\"和一个\"Instagram Business or Creator account\",仓库里还附带了 META_APP_REVIEW.md,以及\"Public Privacy, Terms, Data Deletion, and Meta App Review support pages\"。在 Meta 审核并批准你自己的应用之前,私信回复只能在 Meta 的测试账号上使用。","source":"ai","title":"正式使用前需要 meta 批准"},{"code":"automated_dms_can_cost_you_the_account","severity":"info","evidence":"该项目通过 Meta Graph API 自动发送私信(\"The worker sends a private reply using the comment ID\")。README 把这些回复称为\"Meta-compliant\",但责任在你身上:消息发送规则由 Meta 制定,违反规则可能导致 Instagram 账号受限。","source":"ai","title":"自动发送私信可能导致账号受限"},{"code":"very_early_project","severity":"info","evidence":"只有一个发布版本(v0.1.0,2026-04-29),package.json 的版本号是\"0.1.0\",最后一次提交是在 2026-05-24,README 里说截图和演示 GIF 仍然\"planned in issue #17\",所以还没有产品的实际画面。","source":"ai","title":"项目非常早期"},{"code":"beta_login_library","severity":"info","evidence":"package.json 依赖\"next-auth\": \"^5.0.0-beta.31\"——一个测试版的库在负责一个存有 Instagram 访问令牌的应用的登录功能。","source":"ai","title":"登录库仍是测试版"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram developer app (INSTAGRAM_APP_ID, INSTAGRAM_APP_SECRET, FACEBOOK_APP_SECRET)","Resend (magic-link login email)","Stripe (subscriptions and webhooks)","Secrets you generate yourself (NEXTAUTH_SECRET, CRON_SECRET, ENCRYPTION_KEY, WEBHOOK_VERIFY_TOKEN)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":41,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":1,"growthDays":6,"history":[{"date":"2026-11-05","stars":40},{"date":"2026-11-06","stars":40},{"date":"2026-11-07","stars":40},{"date":"2026-11-08","stars":41},{"date":"2026-11-09","stars":41},{"date":"2026-11-10","stars":41}],"firstPublishedAt":"2026-11-04T22:32:54.984Z","commitSha":"a29a6b144828fa3f887bcc7ce843f48d82ab3be4","verifiedAt":"2026-11-04T22:19:18.899Z","lastCommitAt":"2026-05-24T06:35:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted tool that sends an Instagram direct message when someone comments a keyword like LINK.","summary":"CampaignCue is a web app you host yourself. When someone comments a keyword such as LINK or PRICE on your Instagram post or reel, Meta sends it a notification and the app replies with a private message. It also has campaign pages, tracked links with click counts, client reports and workspaces for agencies. It is for developers or agencies with a technical person: you need Node.js, Docker for the database and queue, an Instagram Business account, a Meta developer app, and Resend and Stripe accounts. The code is MIT; a paid hosted version is planned.","easeRationale":"Identical work on Windows, Mac and Linux, and it is not an install but a deployment: git clone, npm install, `docker-compose up -d` for PostgreSQL and Redis, 17 environment variables to fill in, two Prisma database commands, then a web process and a worker process, plus accounts at Meta, Resend and Stripe before anything can send a message.","costNote":"The self-hosted core is MIT and free. You still need outside accounts (Resend for login emails, Stripe for billing), and the author plans a paid hosted version with the agency features.","notVerified":["Whether Meta approves an app like this, and how long review takes","Whether the comment-to-DM flow actually works end to end on a live account","Whether it still works with the current Meta Graph API version, since the code set v25.0 in May 2026","How well the Instagram tokens are protected in your own database","The real monthly cost of the Resend, Stripe and hosting accounts"]},"zh-Hans":{"summaryShort":"一款自行搭建部署的工具,当有人在评论里打出 LINK 这类关键词时,会发 Instagram 私信。","summary":"CampaignCue 是一个你自己搭建部署的网页应用。当有人在你的 Instagram 帖子或 reel 下评论 LINK 或 PRICE 这样的关键词时,Meta 会发来一个通知,应用就会用私信回复。它还有活动页面、带点击次数统计的追踪链接、客户报告和面向代理机构的工作区功能。它适合有技术人员的开发者或代理机构:你需要 Node.js、用于数据库和队列的 Docker、一个 Instagram 商业账号、一个 Meta 开发者应用,以及 Resend 和 Stripe 账号。代码采用 MIT 许可证;计划推出一个付费的托管版本。","easeRationale":"在 Windows、Mac 和 Linux 上工作量相同,而且这不是一次安装,而是一次部署:git clone、npm install、用 `docker-compose up -d` 启动 PostgreSQL 和 Redis、填写 17 个环境变量、运行两条 Prisma 数据库命令,然后启动一个网页进程和一个 worker 进程,还要在 Meta、Resend 和 Stripe 注册账号,之后才能发出第一条消息。","costNote":"自行搭建的核心部分采用 MIT 许可证,免费。你仍然需要一些外部账号(Resend 用于登录邮件,Stripe 用于计费),作者计划推出带有代理机构功能的付费托管版本。","notVerified":["Meta 是否会批准这样的应用,审核需要多久","评论转私信的流程在真实账号上是否真的能端到端跑通","它是否还能配合 Meta Graph API 目前的版本工作,因为代码里设定的是 2026 年 5 月的 v25.0","你自己数据库里的 Instagram 令牌受到了怎样的保护","Resend、Stripe 和主机账号的真实月费是多少"]}}},{"id":"R_kgDORZOZYw","slug":"imbad0202__academic-research-skills","name":"Academic Research Skills","owner":"Imbad0202","fullName":"Imbad0202/academic-research-skills","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"B"},"risk":"attention","signals":[{"code":"noncommercial_license_only","severity":"attention","evidence":"LICENSE是Creative Commons Attribution-NonCommercial 4.0，package.json写着\"license\": \"CC-BY-NC-4.0\"：\"NonCommercial — You may not use the material for commercial purposes.\"咨询工作、公司研究，或任何你因此获得报酬的用途都不在许可范围内，这也不是一个被认可的开源许可证。","source":"ai","title":"许可证仅限非商业用途"},{"code":"ai_citations_still_wrong_after_checks","severity":"attention","evidence":"README.md展示表中，阶段2.5的完整性报告\"flagged 15 problem references (8 with bibliographic errors, 6–8 likely fabricated)\"，而Post-Publication Audit发现\"21 of the 68 final references still had problems after 3 rounds of integrity checks\"。每一条引用仍然需要人工检查。","source":"ai","title":"经过检查后AI引用仍有错误"},{"code":"academic_integrity_responsibility","severity":"attention","evidence":"README.md：\"It can draft text, including a whole paper in full mode\"以及\"You remain the author, and you answer for every claim you submit.\" 这套工具包含一个用于AI使用声明的\"disclosure\"模式。许多期刊和大学会限制或要求披露这类帮助。","source":"ai","title":"学术责任仍然在你身上"},{"code":"integrity_checks_do_not_verify_the_research","severity":"info","evidence":"README.md，\"Integrity and verification boundary\"：\"ARS does not establish that procedures were actually performed, raw data are authentic, or results reproduce; a consistently reported fabrication can pass these checks.\" 部分检查被描述为抽样检查或由LLM中介完成。","source":"ai","title":"完整性检查并不验证研究本身"},{"code":"review_panel_not_calibrated","severity":"info","evidence":"README.md，Academic Paper Reviewer：\"Current live reviews and Schema 6 packages remain NOT_CALIBRATED\"，\"No numerical total is mapped to Accept, Minor Revision, Major Revision, or Reject\"，而claim审计的校准测试\"drives the runner with a stub judge that returns the gold labels, so it checks the tooling, not a live judge\"。","source":"ai","title":"评审小组尚未校准"},{"code":"sends_your_text_to_outside_services","severity":"info","evidence":"README.md指向docs/DATA_FLOWS.md，说明\"what leaves your machine (bibliographic resolvers, optional consent-gated cross-model calls, the plugin update check)\"。除了Anthropic，你的部分工作内容还可能发往像Semantic Scholar这样的引文数据库，也可以选择发给第二个AI模型做交叉核查。","source":"ai","title":"把你的文本发送给外部服务"},{"code":"declared_alternative_false_match","severity":"info","evidence":"自动检查从README的一段话里读出了\"invideo-ai\"，但那段话实际讲的是AI撰写的同行评审（\"the AI review won 62% of decided pairs without web retrieval\"）。这个项目和视频毫无关系，所以这里不会把那个付费产品列为替代品。","source":"ai","title":"声明的替代产品并不匹配"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","node"],"disk":{"minMB":200,"maxMB":800,"basis":"heuristic"},"apiKeys":["Anthropic (an API key, or a paid Claude plan)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["ai_skills","office","productivity"],"alternativeTo":[{"slug":"paperpal","name":"Paperpal","confirmed":true},{"slug":"elicit","name":"Elicit","confirmed":true},{"slug":"grammarly","name":"Grammarly","confirmed":true}],"declaredAlternatives":[{"target":"invideo-ai","stance":"claims","evidence":"In the additional-sources area (alternative data, theory, or literature), the AI review won 62% of decided pairs without web retrieval (n=85) and 100% with it (n=99), and ARS's domain reviewer recomme"}],"stars":56350,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":1132,"growthDays":7,"history":[{"date":"2026-11-04","stars":55366},{"date":"2026-11-05","stars":55493},{"date":"2026-11-06","stars":55635},{"date":"2026-11-07","stars":55735},{"date":"2026-11-08","stars":55912},{"date":"2026-11-09","stars":56150},{"date":"2026-11-10","stars":56350}],"firstPublishedAt":"2026-11-04T01:36:50.446Z","commitSha":"2e02041a3b821a160d98c9398f173c484ac759a8","verifiedAt":"2026-11-04T01:26:17.248Z","lastCommitAt":"2026-11-10T04:33:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A set of Claude Code skills that walks a researcher from literature search to writing, peer review and a finished paper.","summary":"Academic Research Skills (ARS) is a pack of skills, slash commands and prompts you install inside Claude Code. It guides the whole path of an academic paper: searching the literature, screening studies, planning the argument, drafting, checking every citation, running a simulated review panel and handling reviewer comments. It stops for your confirmation at each stage and the author stays you. It is text and prompts, not a program: you need Claude Code (or a paid claude.ai plan) and you pay Anthropic for the tokens, roughly US$3-7 for a 15,000-word paper by the project's own estimate.","easeRationale":"On claude.ai (paid plans) you only add the repository under Customize → Plugins and install it, so the web route is B — but the README warns that hooks and plugin agents do not run in plain chat there. On Windows, macOS and Linux it is C: you first install the Claude Code command-line tool, set an ANTHROPIC_API_KEY and then run `/plugin marketplace add` and `/plugin install`; Pandoc, tectonic and Python are extra installs only if you want DOCX or PDF output.","costNote":"The skills themselves are free but non-commercial only (CC BY-NC 4.0). docs/PERFORMANCE.md estimates about US$3-7 of Anthropic tokens for a 15,000-word paper at 2026-09 list prices, before cache discounts.","notVerified":["Whether the drafts and reviews are good enough for a real journal submission","How much the full pipeline really costs in tokens for your own paper","Which parts work in each install channel (the project itself ships docs/CONTROL_AVAILABILITY.md because availability varies)","Whether the optional Python scripts, Pandoc and tectonic steps work on Windows","What exactly leaves your machine during a run"]},"zh-Hans":{"summaryShort":"一套Claude Code技能，带领研究者从文献检索走到写作、同行评审和完成一篇论文。","summary":"Academic Research Skills（ARS）是一套安装在Claude Code里的技能、斜线命令和提示词。它引导学术论文的整个流程：检索文献、筛选研究、规划论证、写初稿、核查每一条引用、运行一个模拟评审小组，并处理评审意见。每个阶段它都会停下来等你确认，作者仍然是你。它是文本和提示词，不是一个程序：你需要Claude Code（或一个付费的claude.ai套餐），并按token向Anthropic付费，按项目自己的估算，一篇15000字的论文大约需要3到7美元。","easeRationale":"在claude.ai（付费套餐）上，你只需要在Customize → Plugins里添加仓库并安装，所以网页路径是B级——但README提醒，在那里的普通聊天中，hooks和插件代理不会运行。在Windows、macOS和Linux上是C级：你要先安装Claude Code命令行工具，设置ANTHROPIC_API_KEY，然后运行`/plugin marketplace add`和`/plugin install`；只有想要DOCX或PDF输出时才需要额外安装Pandoc、tectonic和Python。","costNote":"这些技能本身是免费的，但只能用于非商业用途（CC BY-NC 4.0）。docs/PERFORMANCE.md估算，按2026年9月的标价、未计缓存折扣，一篇15000字的论文大约需要3到7美元的Anthropic token费用。","notVerified":["草稿和评审是否足够好，可以真正投给期刊","整条流程对你自己的论文实际要花多少token费用","每种安装渠道下哪些部分能正常工作（项目自己附带docs/CONTROL_AVAILABILITY.md，因为可用性会有差异）","可选的Python脚本、Pandoc和tectonic步骤在Windows上是否能用","运行期间到底有什么内容离开你的机器"]}}},{"id":"R_kgDOGyI-8A","slug":"immich-app__immich","name":"Immich","owner":"immich-app","fullName":"immich-app/immich","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"copyleft_license_agpl","severity":"info","evidence":"LICENSE 是 GNU Affero General Public License v3:如果你通过网络向其他人提供修改后的版本,许可要求你向这些用户公开你修改后的源代码。","source":"ai","title":"copyleft的AGPL许可"},{"code":"install_script_downloads_at_run_time","severity":"info","evidence":"install.sh 会创建 ./immich-app,然后在运行`docker compose up`之前从`https://github.com/immich-app/immich/releases/latest/download`下载 docker-compose.yml 和 example.env,所以实际运行的内容取决于那一刻获取到的文件。","source":"ai","title":"安装脚本在运行时下载文件"},{"code":"you_are_responsible_for_backups","severity":"attention","evidence":"README.md 开头就是一句提醒:\"Always follow 3-2-1 backup plan for your precious photos and videos!\"——除非你自己安排备份,否则项目只保存你唯一的一份副本。","source":"ai","title":"备份由你自己负责"},{"code":"pre_release_version_published","severity":"info","evidence":"最新发布版本标记为 v3.4.0-rc.1(2026-10-30),属于候选发布版;最新的稳定版本标签是 v3.3.5。","source":"ai","title":"发布了预发行版本"}],"platforms":["linux","mac","windows"],"tools":["docker"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image","video","network_system"],"alternativeTo":[{"slug":"google-photos","name":"Photos","confirmed":true},{"slug":"icloud-photos","name":"iCloud Photos","confirmed":true},{"slug":null,"name":"Apple iCloud Photos","confirmed":false}],"declaredAlternatives":[],"stars":118721,"downloads":6422555,"downloadsWeek":73432,"downloadsDays":7,"growth":596,"growthDays":7,"history":[{"date":"2026-10-26","stars":117324},{"date":"2026-10-27","stars":117426},{"date":"2026-10-28","stars":117529},{"date":"2026-10-29","stars":117630},{"date":"2026-10-30","stars":117746},{"date":"2026-10-31","stars":117828},{"date":"2026-11-01","stars":117911},{"date":"2026-11-02","stars":118019},{"date":"2026-11-03","stars":118125},{"date":"2026-11-04","stars":118248},{"date":"2026-11-05","stars":118331},{"date":"2026-11-06","stars":118403},{"date":"2026-11-07","stars":118463},{"date":"2026-11-08","stars":118564},{"date":"2026-11-09","stars":118633},{"date":"2026-11-10","stars":118721}],"firstPublishedAt":"2026-10-27T01:10:12.637Z","commitSha":"3184b1532fd74a0b777207f8a86e9efe7dafce65","verifiedAt":"2026-11-01T01:39:27.616Z","lastCommitAt":"2026-11-10T19:51:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Keeps your photos and videos on your own computer or server, with a phone app that backs them up automatically.","summary":"Immich is a photo and video library you run yourself instead of leaving everything with a cloud service. You install it on a computer or home server with Docker, open it in a browser, and the phone app uploads new pictures by itself. It groups photos by date and place, recognises faces, searches by what appears in the image, and supports several users, shared albums and raw files. You need Docker and a machine that stays on; the official docs list the hardware requirements.","easeRationale":"D on all three: there is no installer or binary in the releases, and the documented way to run it is Docker Compose plus an .env file, which also means keeping a database and a background server running. Linux is the path the project documents best; on Windows and macOS you need Docker Desktop and a machine left switched on.","costNote":"AGPL-3.0 software with no license fee; you pay for the computer or server and the disk space.","notVerified":["Whether it installs and runs on your machine with Docker Desktop on Windows or macOS","How much disk space and memory your own library will actually need","Whether face recognition and image search work well on your photos","Whether the phone app's automatic backup catches every new picture"]},"zh-Hans":{"summaryShort":"把你的照片和视频保存在你自己的电脑或服务器上,配有手机App自动备份。","summary":"Immich 是一个你自己运行的照片和视频库,不用把一切都交给云服务。你用Docker在电脑或家庭服务器上安装它,在浏览器里打开,手机App会自动上传新照片。它按日期和地点给照片分组,识别人脸,根据图片内容搜索,并支持多用户、共享相册和raw文件。你需要Docker和一台保持开机的机器;官方文档列出了硬件要求。","easeRationale":"三个平台都是 D 级:发布版本里没有安装程序或二进制文件,文档中记载的运行方式是 Docker Compose 加一个 .env 文件,这也意味着要维护一个数据库和一个后台运行的服务器。Linux 是项目文档记载得最好的路径;在 Windows 和 macOS 上你需要 Docker Desktop,并让机器保持开机。","costNote":"AGPL-3.0 软件,没有许可费;你需要支付电脑或服务器以及磁盘空间的费用。","notVerified":["在 Windows 或 macOS 上用 Docker Desktop 是否能安装并正常运行","你自己的照片库实际会用掉多少磁盘空间和内存","人脸识别和图像搜索在你的照片上效果好不好","手机App的自动备份是否能捕捉到每一张新照片"]}}},{"id":"R_kgDOVQa3Cw","slug":"impacifist__windows-gpu-unlock","name":"Windows GPU UNLOCK","owner":"windows-GPU-UNLOCK","fullName":"windows-GPU-UNLOCK/Windows-GPU-UNLOCK","type":"needs_setup","status":"alert","grade":"E","easeByPlatform":{"windows":"E"},"risk":"high","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 0 天前","source":"check","title":"所有者账号注册时间很短"},{"code":"bypasses_hardware_restriction","severity":"high","evidence":"这个项目的目的就是破解消费级 NVIDIA 显卡上厂商设置的锁。仓库描述：'Windows GPU UNLOCK: RTX3090 UEFI port...'；README：'Experimental UEFI GPU unlock and direct Windows handoff'，相对于一个 'controlled locked baseline' 测得的提升为 'FP16 1.88-1.92x, BF16 1.92-1.95x'。它基于 rough2515-hub/rtx3080-unlock-public 和 Zagitta/gpu-unlock。","source":"ai","title":"绕过硬件限制"},{"code":"requires_disabling_security","severity":"attention","evidence":"README「Internal headless quick start」：'Use the exact supported hardware, Secure Boot disabled, a fully decrypted Windows C: with BitLocker protection off, and a connected keyboard.'","source":"ai","title":"需要关闭系统防护功能"},{"code":"changes_boot_configuration","severity":"attention","evidence":"README：`Headless.ps1 -EnableAutomatic -ConfirmedSuccess` 'places headless first and Windows Boot Manager second and disables Windows Fast Startup'。README 也描述了失败的情况：'On failure or if Windows does not return, fully power off and choose F8 -> Windows Boot Manager.'","source":"ai","title":"改动系统启动配置"},{"code":"results_not_established","severity":"info","evidence":"README：'There is no contemporary controlled locked baseline for this headless boot, so these values do not establish an unlock percentage. They do not establish game or ComfyUI gains, long-term stability, sleep/hibernate resume or repeated automatic cold-boot reliability.' 最新的 v0.3.1 标签是 'a source-only pre-release'。","source":"ai","title":"结果尚未得到证实"},{"code":"very_narrow_hardware_support","severity":"info","evidence":"README「GPU support」：'Success reports apply to specific cards and systems, not every card of the same GPU model.' 兼容性表格列出的是具体的主板加显卡组合，例如 'ASUS TUF B650M + RTX 3090 FE'。","source":"ai","title":"支持的硬件范围非常有限"}],"platforms":["windows"],"tools":["python","gpu_nvidia","other"],"buildTools":[],"disk":{"minMB":2,"maxMB":7,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":2064,"downloads":1148,"downloadsWeek":354,"downloadsDays":7,"growth":397,"growthDays":7,"history":[{"date":"2026-10-27","stars":1043},{"date":"2026-10-28","stars":1173},{"date":"2026-10-29","stars":1267},{"date":"2026-10-30","stars":1333},{"date":"2026-10-31","stars":1413},{"date":"2026-11-01","stars":1482},{"date":"2026-11-02","stars":1534},{"date":"2026-11-03","stars":1667},{"date":"2026-11-04","stars":1802},{"date":"2026-11-05","stars":1881},{"date":"2026-11-06","stars":1937},{"date":"2026-11-07","stars":1980},{"date":"2026-11-08","stars":2004},{"date":"2026-11-09","stars":2036},{"date":"2026-11-10","stars":2064}],"firstPublishedAt":"2026-10-27T18:40:27.934Z","commitSha":"1f8776ee9c6ee32f5920adaa4baabc9181379bd4","verifiedAt":"2026-10-31T01:35:33.562Z","lastCommitAt":"2026-11-10T03:26:44Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Experimental boot code that releases an NVIDIA RTX 3000 card from its firmware lock before Windows starts.","summary":"A set of experimental UEFI boot files and PowerShell scripts that take an NVIDIA RTX 3060, 3070 Ti, 3080, 3090 or 3090 Ti out of the state the card's firmware leaves it in, then hand control straight to Windows without a restart. The authors report roughly double the measured speed in some tests. It only works on a short list of exact motherboard and card combinations. You must compile it yourself with Python and a pinned clang version, supply five firmware files from your own machine, turn off Secure Boot and BitLocker, and change your boot order.","easeRationale":"There is no binary: the README asks for Python 3.9+, clang/lld-link exactly 20.1.8, five locally obtained firmware assets, your own partition GUIDs, a 1 GiB extra EFI partition, Secure Boot off and BitLocker off, then several administrator PowerShell commands.","costNote":"GPL-2.0-only. Releases are source-only, with no ready program to download.","notVerified":["Whether the build works and the machine comes back to Windows on hardware that is not one of the listed profiles","Whether the reported speed gains hold in games or everyday work rather than the CUDA benchmark","What happens to sleep, hibernate and repeated cold boots, which the README says were not tested","Whether the GPU is left in a state that affects warranty or long-term reliability"]},"zh-Hans":{"summaryShort":"一套实验性启动代码，能在 Windows 启动之前解除 NVIDIA RTX 3000 显卡固件的锁定状态。","summary":"一套实验性的 UEFI 启动文件和 PowerShell 脚本，可以让 NVIDIA RTX 3060、3070 Ti、3080、3090 或 3090 Ti 显卡脱离其固件默认留下的状态，然后不重启就直接把控制权交给 Windows。作者报告在部分测试中测得的速度大约提升了一倍。它只适用于一份很短的主板和显卡组合清单。你必须自己用 Python 和一个固定版本的 clang 编译，提供从自己机器上获取的五个固件文件，关闭 Secure Boot 和 BitLocker，并修改启动顺序。","easeRationale":"没有现成的二进制文件：README 要求准备 Python 3.9+、精确到 20.1.8 版本的 clang/lld-link、本地获取的五个固件文件、你自己分区的 GUID、一个额外的 1 GiB EFI 分区、关闭 Secure Boot 和 BitLocker，然后再运行好几条管理员权限的 PowerShell 命令。","costNote":"仅采用 GPL-2.0 协议。发布的内容只有源代码，没有可以直接下载的现成程序。","notVerified":["在不属于清单所列配置的硬件上，编译是否能成功、系统是否能正常回到 Windows","报告中的速度提升，在游戏或日常使用中是否依然成立，而不仅仅体现在 CUDA 基准测试里","睡眠、休眠以及反复冷启动会发生什么，README 说这些情况没有经过测试","显卡是否会因此进入某种影响保修或长期可靠性的状态"]}}},{"id":"R_kgDORnHMQA","slug":"inbharatai__socialflow","name":"SocialFlow","owner":"inbharatai","fullName":"inbharatai/SocialFlow","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES文件夹和winget清单都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"publishes_with_your_social_logins","severity":"attention","evidence":"发布是通过操控一个已登录你账号的浏览器完成的,而不是通过官方接口:README的表格列出\"LinkedIn | Playwright browser | Session persist\",X / Twitter、Facebook、Instagram、Medium、Substack、beehiiv、MailerLite和Brevo也是如此,Publisher代理会\"Post via Playwright browser automation or API\"。LinkedIn、Instagram、Facebook和X的条款都禁止用你的账号进行自动化访问,这样使用的账号可能会被限制或封禁。","source":"ai","title":"用你自己的社交账号登录状态发布"},{"code":"passwords_in_config_file","severity":"attention","evidence":".env.example要求把某服务的密码写进一个纯文本文件:\"HEYGEN_EMAIL=your-heygen-email@example.com\"和\"HEYGEN_PASSWORD=your-heygen-password\"。README说凭据存储后会\"encrypted with Fernet (AES-128-CBC)\",浏览器会话则保存在本地的\"sessions/\"目录中。","source":"ai","title":"配置文件中存有密码"},{"code":"license_in_readme_only","severity":"info","evidence":"README写着\"MIT License. Built by InBharat AI\",但顶层文件(.env.example、.gitignore、README.md、SOCIALFLOW-README.md、backend/、configs/、frontend/、start.sh)中没有任何许可证文件,GitHub也显示没有许可证。","source":"ai","title":"许可证只出现在readme里"},{"code":"posts_on_its_own_schedule","severity":"info","evidence":"README里的时间表显示程序自己在撰写和发布内容:\"8:07 AM | Full pipeline: news fetch → content generation → review\"以及\"11:00 AM | Publish approved LinkedIn + X posts\"。AI写的文字会以你的名义发出去,审核把关也是由这个项目自己来执行的。","source":"ai","title":"按自己的计划发帖"},{"code":"very_young_project","severity":"info","evidence":"该仓库创建于2026-03-14,有43个星标、12个fork,没有release;分析快照中最后一次提交是在2026-04-10。","source":"ai","title":"项目非常年轻"}],"platforms":["linux","mac"],"tools":["python","ffmpeg","other"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenAI","Anthropic","Google Gemini","Reddit (OAuth)"],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","ai_agents","automation"],"alternativeTo":[{"slug":"socialrails","name":"SocialRails","confirmed":true},{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true}],"declaredAlternatives":[],"stars":44,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":3,"growthDays":6,"history":[{"date":"2026-11-08","stars":43},{"date":"2026-11-09","stars":44},{"date":"2026-11-10","stars":44}],"firstPublishedAt":"2026-11-07T22:56:16.807Z","commitSha":"d450ae0584cccd81923ac7b99f6fd98e2d17d69d","verifiedAt":"2026-11-07T22:40:04.474Z","lastCommitAt":"2026-04-10T11:33:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted assistant that finds news, writes posts and images, and publishes them to 12 social accounts for you.","summary":"SocialFlow runs on your own computer and does a social media routine by itself: it collects news, plans posts, writes them in your brand voice, makes branded images and publishes them on a schedule to 12 places, from LinkedIn and Instagram to Reddit and Substack. It is for someone comfortable with a terminal: you create a Python environment, install the packages and the Playwright browser, run the server and open a dashboard at localhost:8000. Text is written by a local Ollama model by default, or by OpenAI, Anthropic or Gemini with your key.","easeRationale":"On Linux, the system the quick start is written for (bash commands and a start.sh file), you install Python, create a virtual environment, install the packages, install the Playwright Chromium browser, install Ollama and pull an 8B model, then log into each social account in the browser window. Mac works the same way with the same steps; there is no installer and no ready binary for either.","costNote":"The program is free. Writing posts is free with the local Ollama model; the README says image generation uses OpenAI's paid image model at about $0.015-0.020 per image, and the avatar videos go through a HeyGen account, which is a paid service.","notVerified":["Whether the browser automation still works on each of the 12 platforms, which change their pages often","Whether the posts it writes alone are good enough to publish without reading them first","How much the OpenAI image generation really costs in a month of daily posting","Whether the local Ollama model runs at a usable speed on a computer without a strong graphics card","What the program sends out, since the README says nothing leaves your machine"]},"zh-Hans":{"summaryShort":"自托管助手,帮你查找新闻、撰写帖子和图片,并发布到12个社交账号。","summary":"SocialFlow运行在你自己的电脑上,独自完成一整套社交媒体工作:收集新闻、规划帖子、用你的品牌语气撰写内容、制作带品牌风格的图片,并按计划发布到12个平台,从LinkedIn、Instagram到Reddit和Substack。它适合习惯使用终端的人:你需要创建一个Python环境、安装软件包和Playwright浏览器、运行服务器,然后在localhost:8000打开一个控制面板。文本默认由本地的Ollama模型编写,也可以用你自己的密钥换成OpenAI、Anthropic或Gemini。","easeRationale":"在Linux上(快速入门指南就是为它写的,用bash命令和一个start.sh文件),你需要安装Python、创建虚拟环境、安装软件包、安装Playwright的Chromium浏览器、安装Ollama并拉取一个8B模型,然后在浏览器窗口里登录每个社交账号。Mac的步骤完全一样;两者都没有安装程序,也没有现成的可执行文件。","costNote":"这个程序是免费的。用本地Ollama模型写帖子是免费的;README说生成图片用的是OpenAI的付费图片模型,大约每张图片$0.015-0.020,而头像视频要通过HeyGen账号,这是一项付费服务。","notVerified":["浏览器自动化在12个平台各自的页面上是否仍然有效,因为这些平台经常改版。","它独自撰写的帖子是否足够好,可以不经过事先阅读就直接发布。","在每天都发帖的一个月里,OpenAI的图片生成实际要花多少钱。","在没有强力显卡的电脑上,本地Ollama模型的运行速度是否够用。","既然README说没有任何东西离开你的电脑,这个程序实际对外发送了什么。"]}}},{"id":"R_kgDON1G5Sg","slug":"index-tts__index-tts","name":"IndexTTS","owner":"index-tts","fullName":"index-tts/index-tts","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"restrictive_custom_license","severity":"attention","evidence":"LICENSE 是 \"bilibili Model Use License Agreement\",不是标准的开源许可证:月活用户超过 1 亿或收入超过人民币 10 亿时需要另行申请许可,禁止将其用于 \"to improve any AI model\"(本项目以外的其他 AI 模型),且所有者 \"reserve[s] the right to revoke the license\"","source":"ai","title":"自定义许可证限制较多"},{"code":"license_forbids_high_risk_use","severity":"info","evidence":"LICENSE 第 4.2 节 \"Prohibited High-Risk Use\" 列出了医疗诊断、自动驾驶、军事应用、关键基础设施控制、大规模生物识别监控以及自动化的信贷或招聘决策","source":"ai","title":"许可证禁止高风险用途"},{"code":"voice_cloning_impersonation_risk","severity":"attention","evidence":"README 描述了从单个参考片段进行零样本声音克隆(\"clones a voice from a single reference audio clip\"),这使得在未经同意的情况下复制真人声音成为可能;仓库中还附带了单独的 DISCLAIMER 文件","source":"ai","title":"声音克隆存在冒充风险"},{"code":"downloads_models_at_run_time","severity":"info","evidence":"README 写道:\"Some small models are downloaded automatically on first run\",示例音频文件会在 WebUI 首次启动时 \"downloaded on demand from HuggingFace/ModelScope the first time the WebUI starts\"","source":"ai","title":"运行时下载模型"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":1010,"maxMB":4570,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":25032,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":89,"growthDays":6,"history":[{"date":"2026-11-06","stars":24964},{"date":"2026-11-07","stars":24974},{"date":"2026-11-08","stars":24992},{"date":"2026-11-09","stars":25010},{"date":"2026-11-10","stars":25032}],"firstPublishedAt":"2026-11-05T22:43:48.496Z","commitSha":"d9e41aac89fd00b3d71497fddb287b7f24613712","verifiedAt":"2026-11-05T22:25:26.336Z","lastCommitAt":"2026-09-29T16:06:52Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Clones a voice from one short recording and reads your text aloud in Chinese, English, Japanese, Spanish or Arabic.","summary":"IndexTTS, from bilibili, turns written text into speech using a voice copied from a single short reference recording. You can set the emotion (from another clip, from sliders or from a description), the speaking speed and the pronunciation of specific words. It runs on your own computer through a page in the browser. You need the terminal, Python with the uv tool, an NVIDIA graphics card with CUDA 12.8 for good speed, and a few gigabytes of model files downloaded from HuggingFace or ModelScope.","easeRationale":"On Windows and Linux, the best case, you clone the repository, install the uv tool, run `uv sync --all-extras`, install a model-download tool, fetch several gigabytes of model files and then start `uv run webui.py`; CUDA Toolkit 12.8 or newer is needed and the README warns that DeepSpeed may be hard to install on Windows. On Mac there is no CUDA support in the project's own setup, so speed depends on the processor.","costNote":"No price; the license asks very large companies to request a separate licence and says to email indexspeech@bilibili.com for commercial use","notVerified":["Real disk use and video memory needed after downloading the 2.5 and 2 model files","Whether it installs and runs on a Mac without an NVIDIA card, and how slow it is there","Whether the quality in each of the five languages matches the tables in the README","Which files the automatic first-run downloads bring in"]},"zh-Hans":{"summaryShort":"从一段简短的录音中克隆声音,并用中文、英语、日语、西班牙语或阿拉伯语朗读你的文本。","summary":"IndexTTS 来自哔哩哔哩(bilibili),使用从一段简短参考录音中复制的声音,把书面文字转换成语音。你可以设置情绪(来自另一段片段、滑动条或一段描述)、说话速度和特定词语的发音。它通过浏览器中的一个页面,在你自己的电脑上运行。你需要终端、带有 uv 工具的 Python、一块支持 CUDA 12.8 的 NVIDIA 显卡以获得良好速度,以及从 HuggingFace 或 ModelScope 下载的几个 GB 的模型文件。","easeRationale":"在 Windows 和 Linux 上,最理想的情况是:克隆仓库,安装 uv 工具,运行 `uv sync --all-extras`,安装一个模型下载工具,获取几个 GB 的模型文件,然后启动 `uv run webui.py`;需要 CUDA Toolkit 12.8 或更新版本,README 提醒 DeepSpeed 在 Windows 上可能很难安装。在 Mac 上,项目自身的配置不支持 CUDA,所以速度取决于处理器。","costNote":"没有标价;许可证要求规模很大的公司另行申请许可,并说明商业用途请发邮件至 indexspeech@bilibili.com","notVerified":["下载 2.5 和 2 两个模型文件后,实际占用的磁盘空间和显存","在没有 NVIDIA 显卡的 Mac 上能否安装运行,以及速度有多慢","五种语言各自的质量是否与 README 中的表格相符","首次运行时的自动下载会带来哪些文件"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMzE2MzA5Mw==","slug":"inkscape__inkscape","name":"Inkscape","owner":"inkscape","fullName":"inkscape/inkscape","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"github_repo_is_placeholder","severity":"info","evidence":"该仓库只有 README.md 和一张图片；README.md 写着 \"Inkscape is being developed on GitLab\"（gitlab.com/inkscape/inkscape）。这里最后一次推送是 2022-03-03，也没有许可证文件，所以 GitHub 上的活跃度和许可证数据并不能反映这个项目的真实情况。","source":"ai","title":"GitHub 仓库只是个占位页"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":162,"maxMB":487,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"winget"},"categories":["design_cad","image"],"alternativeTo":[{"slug":"illustrator","name":"Illustrator","confirmed":true},{"slug":"coreldraw","name":"CorelDRAW","confirmed":true},{"slug":"affinity-designer","name":"Affinity Designer","confirmed":true}],"declaredAlternatives":[],"stars":4109,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":26,"growthDays":7,"history":[{"date":"2026-10-25","stars":4058},{"date":"2026-10-26","stars":4058},{"date":"2026-10-27","stars":4062},{"date":"2026-10-28","stars":4066},{"date":"2026-10-29","stars":4071},{"date":"2026-10-30","stars":4073},{"date":"2026-10-31","stars":4075},{"date":"2026-11-01","stars":4075},{"date":"2026-11-02","stars":4079},{"date":"2026-11-03","stars":4083},{"date":"2026-11-04","stars":4087},{"date":"2026-11-05","stars":4096},{"date":"2026-11-06","stars":4099},{"date":"2026-11-07","stars":4101},{"date":"2026-11-08","stars":4102},{"date":"2026-11-09","stars":4106},{"date":"2026-11-10","stars":4109}],"firstPublishedAt":"2026-10-25T23:19:07.264Z","commitSha":"1232eac68aff12978b208ff3ebfbbcc3ecbd7022","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2022-03-03T20:52:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Draw logos, icons, illustrations, diagrams and maps as vector graphics that stay sharp at any size.","summary":"Inkscape is a vector drawing program used by designers and hobbyists to make illustrations, icons, logos, diagrams, maps and web graphics. Its own file format is SVG, an open standard. It runs on Windows, macOS and Linux and is installed with a normal installer. This GitHub page is only a pointer: the code is developed on GitLab and the program is downloaded from inkscape.org.","easeRationale":"A ready Windows installer is available (the checks found one of about 162–487 MB), so you download it and open it.","costNote":"The README describes it as free and open-source software.","notVerified":["The real license and recent activity, which live in the GitLab repository, not on GitHub","Whether the installer the checks found is the current version from inkscape.org","How well it opens files made in Illustrator or CorelDRAW"]},"zh-Hans":{"summaryShort":"绘制标志、图标、插图、图表和地图，作为矢量图形，在任何尺寸下都保持清晰。","summary":"Inkscape 是一款矢量绘图程序，设计师和爱好者用它来制作插图、图标、标志、图表、地图和网页图形。它自己的文件格式是 SVG，这是一种开放标准。它可在 Windows、macOS 和 Linux 上运行，用常规安装程序安装。这个 GitHub 页面只是一个指引：代码在 GitLab 上开发，程序则从 inkscape.org 下载。","easeRationale":"有现成的 Windows 安装程序（检查中发现了一个约 162–487 MB 的），下载后打开即可。","costNote":"README 将其描述为免费的开源软件。","notVerified":["真实的许可证和近期活跃情况，这些信息在 GitLab 仓库而不是 GitHub 上","检查中发现的安装程序是否就是 inkscape.org 上的当前版本","它打开 Illustrator 或 CorelDRAW 制作的文件时效果如何"]}}},{"id":"R_kgDOVW5g1w","slug":"interplains__zero-breach-toolkit","name":"Zero Breach Toolkit","owner":"interplains","fullName":"interplains/Zero-Breach-Toolkit","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定的依赖项存在已知漏洞 (OSV)：cryptography: GHSA-g6cj-pr64-35w5, GHSA-jwv3-5hgf-82ww, GHSA-m2h6-j472-rp4c","source":"check","title":"存在已知漏洞"},{"code":"anonymous_publisher_and_unsigned_apps","severity":"attention","evidence":"README：\"The project is anonymous, so do not take its word for anything\" 以及 \"There is no signed Windows program, and none is planned while the project stays anonymous: code-signing certificates require a named legal identity\"。它还警告 \"Your antivirus may flag these programs\"，并说对于自己发布页上的文件，这属于误报。每个发布版本都会公布 SHA-256、GPG 签名和 GitHub 构建记录，你可以用 `gh attestation verify` 核对。","source":"ai","title":"发布者匿名，应用未签名"},{"code":"reports_hold_private_personal_data","severity":"attention","evidence":"README：报告 \"contain private details from your computer; keep them out of shared and synced folders\"，并告诉你不要把报告贴到 issue 跟踪器里。iPhone 的检查会读取信息（短信、iMessage、WhatsApp）、通话、日历和浏览记录。","source":"ai","title":"报告中包含私人个人数据"},{"code":"phone_check_needs_usb_debugging","severity":"attention","evidence":"安卓的步骤要求你打开开发者选项和 USB 调试，并为这台电脑接受 \"Always allow\"。README 说完成后要再把 USB 调试关掉，但一直保持这种状态的手机会更容易受到攻击。","source":"ai","title":"检查手机需要开启 USB 调试"},{"code":"very_new_repository_with_many_stars","severity":"info","evidence":"该仓库创建于 2026-10-25，已有 4,162 个星标、437 个 fork 和 21 个发布版本，其中三个仅在 2026-11-01 一天发布。没有长期的历史可供判断。","source":"ai","title":"很新的仓库，星标很多"},{"code":"optional_command_downloads_and_runs_another_program","severity":"info","evidence":"README，\"Guided acquisition (experimental)\"：可选的 --collect-android 参数 \"downloads official AndroidQF v1.8.3, checks its pinned SHA-256, asks for the device owner's explicit consent and runs the collector\"。README 称之为 \"the only point at which the tool runs a program it did not ship\"，并说采集会改变手机的状态。","source":"ai","title":"一个可选命令会下载并运行另一个程序"},{"code":"one_optional_test_opens_your_browser_online","severity":"info","evidence":"README，\"Start menu and follow-up\"：浏览器一项提供了广告拦截测试，它 \"only works inside a browser, so the tool asks first and then opens the ArchiveGenocide test page in your default browser. The browser goes online for it; the tool does not.\" README 说扫描本身不发出网络请求，Python 的网络调用也被屏蔽了，同时指出 \"This is not an operating-system network sandbox\"。","source":"ai","title":"一项可选测试会让你的浏览器联网"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python"],"disk":{"minMB":18,"maxMB":54,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["network_system"],"alternativeTo":[{"slug":"iverify","name":"iVerify","confirmed":true}],"declaredAlternatives":[],"stars":4643,"downloads":11020,"downloadsWeek":2364,"downloadsDays":7,"growth":277,"growthDays":7,"history":[{"date":"2026-10-28","stars":2785},{"date":"2026-10-29","stars":3614},{"date":"2026-10-30","stars":3869},{"date":"2026-10-31","stars":4006},{"date":"2026-11-01","stars":4160},{"date":"2026-11-02","stars":4284},{"date":"2026-11-03","stars":4366},{"date":"2026-11-04","stars":4416},{"date":"2026-11-05","stars":4472},{"date":"2026-11-06","stars":4511},{"date":"2026-11-07","stars":4548},{"date":"2026-11-08","stars":4584},{"date":"2026-11-09","stars":4621},{"date":"2026-11-10","stars":4643}],"firstPublishedAt":"2026-10-28T01:50:05.614Z","commitSha":"c711c01a75d23e8812b471fb21295b7c29c97941","verifiedAt":"2026-11-02T02:53:07.721Z","lastCommitAt":"2026-11-05T14:11:47Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"zero-breach-toolkit-windows-x64.exe","sizeBytes":19003412,"url":"https://github.com/interplains/Zero-Breach-Toolkit/releases/download/v2.0.4/zero-breach-toolkit-windows-x64.exe","releaseUrl":"https://github.com/interplains/Zero-Breach-Toolkit/releases/tag/v2.0.4"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Checks a Windows, Mac or Linux computer, an Android phone or an iPhone backup for traces of known spyware.","summary":"A free scanner that looks for traces of spyware already described in public research, such as Pegasus, Predator, Graphite and Candiru. It reads browser records, startup settings, logs and files, and can also read an Android collection or a local iPhone backup made on your computer. Version 2 is a double-click app for Windows, Mac and Linux in 27 languages; the same check also runs from the Python source. It is written for journalists, activists and anyone worried about being targeted. A standard check takes 15 to 20 minutes.","easeRationale":"Windows is the simplest: one `zero-breach-toolkit-windows-x64.exe` you double-click, then \"More info → Run anyway\" because it is not code-signed. On Mac you drag the .dmg app to Applications and macOS refuses it the first time, so you have to click \"Open Anyway\" in System Settings → Privacy & Security, and give the app Full Disk Access for Safari's records to be read. On Linux you first extract a .tar.gz and run the program inside it. Checking a phone is harder on every platform: an Android phone needs USB debugging, and an iPhone has to be backed up to the computer first.","costNote":"Apache-2.0 code; the indicator lists it downloads are published free by Amnesty International's Security Lab, the Citizen Lab, the MVT project and Echap.","notVerified":["Whether the checks really run to the end on each version of Windows, macOS, Linux, Android and iOS.","Whether a result is correct: the project itself says a clean result is not proof the device is clean, and a match means \"look closer\", not that you are infected.","Whether the app and the single Python file attached to each release really rebuild to the published SHA-256.","How much disk and time a full check takes on a machine with many files, since it fingerprints files up to 512 MiB each.","Whether the pinned dependency versions with known vulnerabilities (cryptography) affect the parts of the tool that are used, beyond what the requirements file claims.","Whether the 27 translations are accurate."]},"zh-Hans":{"summaryShort":"检查 Windows、Mac 或 Linux 电脑、安卓手机或 iPhone 备份中是否有已知间谍软件的痕迹。","summary":"一款免费的扫描工具，查找公开研究中已有描述的间谍软件痕迹，例如 Pegasus、Predator、Graphite 和 Candiru。它读取浏览器记录、启动设置、日志和文件，也可以读取安卓采集数据，或在你电脑上制作的本地 iPhone 备份。第 2 版是适用于 Windows、Mac 和 Linux 的双击即用应用，支持 27 种语言；同样的检查也可以通过 Python 源代码运行。它写给记者、活动人士以及任何担心自己成为监视目标的人。一次标准检查需要 15 到 20 分钟。","easeRationale":"Windows 最简单：一个 `zero-breach-toolkit-windows-x64.exe`，双击后再点 \"More info → Run anyway\"，因为它没有代码签名。在 Mac 上，你把 .dmg 里的应用拖到“应用程序”，macOS 第一次会拒绝运行，所以你要在 System Settings → Privacy & Security 里点 \"Open Anyway\"，并给应用 Full Disk Access 权限，才能读取 Safari 的记录。在 Linux 上，你先解压一个 .tar.gz，再运行里面的程序。检查手机在每个平台上都更难：安卓手机需要开启 USB 调试，iPhone 则必须先备份到电脑上。","costNote":"代码采用 Apache-2.0；它下载的指标列表由 Amnesty International 的 Security Lab、Citizen Lab、MVT 项目和 Echap 免费发布。","notVerified":["检查是否真的能在每个版本的 Windows、macOS、Linux、Android 和 iOS 上运行到结束。","结果是否正确：项目自己说，结果干净并不能证明设备是干净的，匹配到的结果意思是 \"look closer\"，而不是说你已被感染。","应用和每个发布版本附带的那个 Python 单文件，是否真的能重新构建出已公布的 SHA-256。","在文件很多的机器上，一次完整检查要花多少磁盘和时间，因为它会为每个最大 512 MiB 的文件计算指纹。","固定下来的、带有已知漏洞的依赖版本 (cryptography) 是否会影响工具中实际用到的部分，而不只是需求文件所说的那样。","27 种语言的翻译是否准确。"]}}},{"id":"R_kgDOH1Ppow","slug":"invoke-ai__invokeai","name":"InvokeAI","owner":"invoke-ai","fullName":"invoke-ai/InvokeAI","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"the_models_have_their_own_licences","severity":"attention","evidence":"程序本身使用 Apache-2.0 许可，但仓库中还包含 LICENSE-SD1+SD2.txt、LICENSE-SDXL.txt、LICENSE-HiDiffusion.txt、LICENSE-PiD.txt 和 Stable_Diffusion_v1_Model_Card.md，模型列表中也包括 \"Flux.1 Dev\" 和 \"Flux.2 Dev\" 等系列。你下载的模型受这些独立条款约束，可能限制商业使用，而应用自身的许可并不涵盖它们。","source":"ai","title":"模型有各自独立的许可"},{"code":"the_installer_lives_in_another_repository","severity":"info","evidence":"这个仓库最新的三个发布版本（v7.0.0-beta.1、v6.14.4、v7.0.0-alpha.4）都标注为 \"no platform binaries\"；README 把你指向另一个项目去安装：\"To get started with Invoke, [Download the Launcher](https://github.com/invoke-ai/launcher/releases/latest)\"。那部分代码不在这个仓库里。","source":"ai","title":"安装程序放在另一个仓库里"},{"code":"the_newest_release_is_a_test_version","severity":"info","evidence":"最新发布版本是 v7.0.0-beta.1（2026-11-09），上一个是 v7.0.0-alpha.4；最新的正式版本是 v6.14.4（2026-11-03）。该项目共有 235 个发布版本和 318 个未解决的 issue。","source":"ai","title":"最新版本是测试版"},{"code":"models_are_downloaded_and_take_a_lot_of_disk","severity":"info","evidence":"应用本身不自带模型：它有一个 \"Model Manager & Support\"，模型列表列出了 25 个以上需要获取的系列，从 SD 1.5 到 Flux.2 和 Qwen Image。检查报告的大小（23 到 69 MB）只是程序本身；每个下载的模型通常有好几 GB。","source":"ai","title":"模型需要下载，占用大量磁盘空间"},{"code":"it_opens_as_a_web_page_served_by_your_own_computer","severity":"info","evidence":"README：\"Invoke runs a locally hosted web server & React UI\"。你是在浏览器里访问运行在自己电脑上的服务器，所以任何能连到那个端口的设备都能访问这个应用。","source":"ai","title":"它以网页形式运行，由你自己的电脑提供服务"}],"platforms":["windows","mac","linux"],"tools":["gpu_nvidia"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":23,"maxMB":69,"basis":"installer"},"apiKeys":["Online image and video model services for the API-only models (Nano Banana, GPT Image, Wan)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["image","local_ai"],"alternativeTo":[{"slug":"midjourney","name":"Midjourney","confirmed":true},{"slug":"adobe-firefly","name":"Firefly","confirmed":true},{"slug":"photoshop","name":"Photoshop","confirmed":true}],"declaredAlternatives":[],"stars":29054,"downloads":355026,"downloadsWeek":0,"downloadsDays":0,"growth":0,"growthDays":0,"history":[{"date":"2026-11-10","stars":29054}],"firstPublishedAt":"2026-11-09T23:02:44.108Z","commitSha":"83507c67a9a17ff0b84bd7a5b12b74673ce4aa6f","verifiedAt":"2026-11-09T22:52:34.293Z","lastCommitAt":"2026-11-10T08:42:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Generates and edits pictures with AI models running on your own computer, with a paint canvas and node workflows.","summary":"Invoke is a free program for creating and editing images with AI models that run on your own computer. You install it with its launcher, download the models you want inside the app, and work in a browser window on your machine: a prompt box, a paint canvas for filling in or extending parts of a picture, galleries and boards, and node workflows for repeatable jobs. It supports many model families, from Stable Diffusion to FLUX and Qwen Image, and a few online-only models. It is aimed at artists and studios and needs a capable graphics card.","easeRationale":"On Windows, Mac and Linux the README points at one download, \"Download the Launcher\", which sets the program up for you; after that the extra step is picking and downloading models inside the app before you can generate anything. There is no terminal work on the normal route, but you do need a suitable graphics card, and the pyproject file shows the Mac build is held back to older graphics libraries because newer ones run out of memory.","costNote":"The program is free under the Apache-2.0 licence. The company behind it asks for sponsorship and says Invoke \"serves as the foundation for multiple commercial products\", and a few models in the list work only through a paid online service (\"Nano Banana (API Only)\", \"GPT Image (API Only)\", \"Wan (API Only)\").","notVerified":["Which graphics cards are enough for each model family, and how much video memory the newer FLUX and Qwen models need.","Real disk use once you download the models you want, on top of the program itself.","How well it performs on a Mac, where the project pins older graphics libraries, and on AMD cards on Linux.","Whether the separate launcher download installs what this repository contains.","What the API-only models cost per image or video, since that is billed by the outside service.","Whether the local web server is reachable from other machines on your network in the default setup."]},"zh-Hans":{"summaryShort":"在你自己的电脑上运行 AI 模型来生成和编辑图片，带有绘画画布和节点工作流。","summary":"Invoke 是一款免费软件，用于在你自己的电脑上运行 AI 模型来创建和编辑图片。你通过它的启动器安装程序，在应用内下载想要的模型，并在本机的浏览器窗口中工作：提示词输入框、用于填补或扩展图片局部的绘画画布、图库和面板，以及用于重复性工作的节点工作流。它支持多种模型系列，从 Stable Diffusion 到 FLUX 和 Qwen Image，还有少数仅在线使用的模型。它面向艺术家和工作室，需要一块性能不错的显卡。","easeRationale":"在 Windows、Mac 和 Linux 上，README 都指向同一个下载项 \"Download the Launcher\"，它会替你配置好程序；之后多出来的一步是在应用内挑选并下载模型，然后才能开始生成。正常流程不需要操作终端，但你确实需要一块合适的显卡，而且 pyproject 文件显示 Mac 版本被锁定在较旧的图形库上，因为较新的库会耗尽内存。","costNote":"该程序基于 Apache-2.0 许可免费提供。背后的公司会寻求赞助，并表示 Invoke \"serves as the foundation for multiple commercial products\"，列表中的一些模型只能通过付费的在线服务使用（\"Nano Banana (API Only)\"、\"GPT Image (API Only)\"、\"Wan (API Only)\"）。","notVerified":["每种模型系列需要什么级别的显卡才够用，以及较新的 FLUX 和 Qwen 模型需要多少显存。","下载完想要的模型之后，除程序本身外的实际磁盘占用。","它在 Mac（该项目锁定了较旧的图形库）上以及在 Linux 的 AMD 显卡上表现如何。","单独下载的启动器是否安装的就是这个仓库里的内容。","仅限 API 的模型每张图片或每段视频的实际费用，因为这由外部服务计费。","在默认设置下，本机的网络服务器是否能被同一网络中的其他设备访问。"]}}},{"id":"R_kgDOPZ417g","slug":"iofficeai__aionui","name":"AionUi","owner":"iOfficeAI","fullName":"iOfficeAI/AionUi","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):builder-util-runtime: GHSA-p2f4-r6v6-j797","source":"check","title":"已知漏洞"},{"code":"crash_reporting_despite_local_only_claim","severity":"attention","evidence":"README 的问答部分写道\"All data is stored locally in a SQLite database. Nothing is uploaded to any server.\",但 package.json 依赖\"@sentry/electron\": \"^7.10.0\",这是一个会把崩溃和错误报告发送到远程服务器的服务。","source":"ai","title":"声称只在本地却仍上报崩溃信息"},{"code":"agent_can_change_your_files_unattended","severity":"attention","evidence":"README 自己的表格写着\"Yes — built-in agent with full file access\",并列出\"YOLO / Full-Auto Mode — supported agents can expose an unattended mode\"以及用于计划任务的\"Cron — 24/7 unattended\"。你没在旁边盯着时出的一个差错,可能会重命名、移动或覆盖你的文件。","source":"ai","title":"代理可以在无人看管时更改你的文件"},{"code":"remote_access_opens_a_door","severity":"attention","evidence":"WebUI 模式\"Supports LAN, cross-network, and server deployment. QR code or password login\",Dockerfile 里设置了\"ENV ALLOW_REMOTE=true\"并暴露了 3000 端口。任何能访问到那个地址并通过密码验证的人,都可以操控一个能访问你文件的代理。","source":"ai","title":"远程访问打开了一道门"},{"code":"newest_installers_not_on_github","severity":"info","evidence":"GitHub 上最后一个带安装文件的发布版本是 v2.1.47-final(2026-08-05);之后的版本,一直到 v2.2.2(2026-09-09),都没有附带任何文件,尽管代码仍在每天更新。新的下载只能来自项目自己的网站 aionui.com。","source":"ai","title":"最新的安装程序不在 github 上"},{"code":"component_downloaded_after_install","severity":"info","evidence":"Dockerfile 里的注释:\"officecli (the Office preview component, auto-installed at runtime by the backend) is a .NET binary\"。部分 Office 功能是在你安装应用之后才从网上下载的。","source":"ai","title":"组件在安装后才下载"},{"code":"paid_referral_links_in_readme","severity":"info","evidence":"README 宣传了与 Kimi 的合作关系,带有联盟链接(\"kimi.com/code?aff=aionui\"、\"platform.kimi.ai?track_id=...&aff=aionui\"),还为\"for our contributors\"提供免费的付费套餐,所以 README 里的一些推荐会给项目带来收入。","source":"ai","title":"readme 中含有付费推荐链接"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":452,"maxMB":1355,"basis":"installer"},"apiKeys":["An AI model provider of your choice (Gemini, OpenAI, Anthropic, DeepSeek, Kimi, OpenRouter and 30+ more)","Telegram, Lark, DingTalk or WeChat bot token (only for chatting with it from those apps)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","office","automation"],"alternativeTo":[{"slug":"claude-cowork","name":"Claude Cowork","confirmed":true},{"slug":"manus","name":"Manus","confirmed":true}],"declaredAlternatives":[],"stars":34194,"downloads":933354,"downloadsWeek":0,"downloadsDays":5,"growth":160,"growthDays":6,"history":[{"date":"2026-11-05","stars":34060},{"date":"2026-11-06","stars":34075},{"date":"2026-11-07","stars":34091},{"date":"2026-11-08","stars":34126},{"date":"2026-11-09","stars":34165},{"date":"2026-11-10","stars":34194}],"firstPublishedAt":"2026-11-04T22:32:54.992Z","commitSha":"119dd2b4768f8885cb35a512a8d9b6b81f576dfd","verifiedAt":"2026-11-04T22:20:42.912Z","lastCommitAt":"2026-11-05T05:31:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop app where an AI assistant works on your own files: organizes folders, builds slides, Word and Excel.","summary":"AionUi is a desktop program with a built-in AI agent that works on the files on your computer: it tidies folders, renames in batches, reads and writes Word, Excel and PowerPoint files, analyses spreadsheets, generates images and previews the results. You install it and paste an API key from a provider you choose (Gemini, OpenAI, Anthropic, DeepSeek, Kimi and others, or a local model through Ollama). It can also drive AI command-line tools you already have, run tasks on a schedule, and be reached from a phone through a browser or a Telegram bot.","easeRationale":"Each platform has a ready installer (.exe for Windows, .dmg for Mac, .deb for Linux, plus `brew install aionui`), and the README's three steps are install, paste an API key, start working — an installer plus one simple step, so B everywhere. Minimum stated: Windows 10, macOS 10.15, Ubuntu 18.04, 4 GB of memory.","costNote":"The app is free (Apache-2.0). You pay the AI provider for what you use; free tiers or a local model through Ollama or LM Studio can avoid that.","notVerified":["Whether the 21 built-in assistants produce documents as good as the animated demos","What the Sentry crash reporting sends and whether it can be turned off","How much you will actually spend on API calls for a day of agent work","Whether every one of the 20+ listed command-line agents really is detected","What is inside the installers on the project's website, since they are not on GitHub","Whether the WebUI password login is strong enough to expose to the internet"]},"zh-Hans":{"summaryShort":"一款桌面应用,AI 助手在里面处理你自己的文件:整理文件夹、制作幻灯片、Word 和 Excel。","summary":"AionUi 是一款内置 AI 代理的桌面程序,能处理你电脑上的文件:整理文件夹、批量改名、读写 Word、Excel 和 PowerPoint 文件、分析表格、生成图像并预览结果。安装之后,粘贴一个你自选服务商的 API 密钥即可(Gemini、OpenAI、Anthropic、DeepSeek、Kimi 等,或通过 Ollama 使用本地模型)。它还能驱动你已有的 AI 命令行工具,按计划运行任务,并通过浏览器或 Telegram 机器人从手机上访问。","easeRationale":"每个平台都有现成的安装程序(Windows 用 .exe、Mac 用 .dmg、Linux 用 .deb,还可以 `brew install aionui`),README 里的三个步骤是安装、粘贴 API 密钥、开始使用——一个安装程序加一个简单步骤,所以各平台都是 B 级。文档给出的最低要求:Windows 10、macOS 10.15、Ubuntu 18.04,4 GB 内存。","costNote":"这款应用是免费的(Apache-2.0)。你按使用量向 AI 服务商付费;免费额度或通过 Ollama、LM Studio 使用本地模型可以避免这笔费用。","notVerified":["21 个内置助手生成的文档是否真的和动画演示一样好","Sentry 崩溃报告发送了什么内容,是否可以关闭","一天的代理工作实际会在 API 调用上花多少钱","列出的 20 多个命令行代理是不是每一个都真的能被识别","项目官网上的安装程序里到底有什么,因为它们不在 GitHub 上","WebUI 的密码登录是否足够牢固,能不能放心暴露在互联网上"]}}},{"id":"MDEwOlJlcG9zaXRvcnk5MDMyMDQ5NA==","slug":"ish-app__ish","name":"iSH","owner":"ish-app","fullName":"ish-app/ish","type":"ready_app","status":"active","grade":"A","easeByPlatform":{},"risk":"low","signals":[{"code":"no_files_in_releases","severity":"info","evidence":"GitHub 上的 701 个 release(builds/813、builds/812、builds/811)都不包含任何平台的二进制文件;该应用通过 README 中的 App Store 和 TestFlight 链接分发","source":"ai","title":"发行版中没有文件"},{"code":"license_conflict_with_app_store","severity":"info","evidence":"LICENSE.IOS 中写道,开发者 \"are aware that the terms of service that apply to apps distributed via Apple's App Store services may conflict with rights granted under the iSH license, the GNU General Public License, version 2 or 3\",并承诺不会仅因这一冲突而追究违规责任","source":"ai","title":"许可证与 app store 存在冲突"},{"code":"author_warns_about_the_code","severity":"info","evidence":"README 中 \"A note on the interpreter\" 一节提到,模拟器里几乎所有的 gadget 都是用汇编写的,作者称之为 \"a horrible decision with regards to readability, maintainability\",而且 \"almost no comments\"","source":"ai","title":"作者对代码提出警示"},{"code":"many_open_issues","severity":"info","evidence":"GitHub 上有 646 个未解决的 issue,README 建议 \"check the issues tab, and the commit logs\" 来了解项目目前的状态","source":"ai","title":"大量未解决的 issue"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"file"},"categories":["network_system","development"],"alternativeTo":[{"slug":"blink-shell","name":"Blink Shell","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":20698,"downloads":107598,"downloadsWeek":1612,"downloadsDays":4,"growth":32,"growthDays":6,"history":[{"date":"2026-11-06","stars":20677},{"date":"2026-11-07","stars":20683},{"date":"2026-11-08","stars":20688},{"date":"2026-11-09","stars":20696},{"date":"2026-11-10","stars":20698}],"firstPublishedAt":"2026-11-05T22:43:48.503Z","commitSha":"de5ab2c4b1039551203c1106e75f715cd917544b","verifiedAt":"2026-11-05T22:26:42.438Z","lastCommitAt":"2026-11-07T00:46:48Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs a small Linux system with a command line inside an iPhone or iPad, by emulating a PC processor.","summary":"iSH puts a Linux command line on an iPhone or iPad. It emulates a 32-bit x86 processor and translates Linux system calls, so it can run an Alpine Linux environment and its command-line programs on the phone itself, without a server. It is for people comfortable with a terminal: once open, you type shell commands. The app is listed on the App Store and has a TestFlight beta; this repository holds the source code, and the build instructions (Xcode, Meson, Ninja, Clang) are meant for developers. The code is under the GPL, with an extra note about App Store terms.","easeRationale":"It is an iPhone and iPad app, so none of the computer systems get a grade: the README's first link is its App Store page, where it installs like any other app. On a Mac or Linux computer only the developer build is described (Xcode, or Python with Meson, Ninja, Clang, LLD, sqlite3 and libarchive), plus making an Alpine file system by hand.","notVerified":["Whether the app on the App Store is built from this source code and how old that version is","How fast emulated Linux programs actually run on a given iPhone or iPad, and the battery cost","Which Alpine Linux packages install and work inside it","How much storage the Linux files take after installing packages","Whether the app keeps running in the background or is stopped by iOS"]},"zh-Hans":{"summaryShort":"通过模拟 PC 处理器,在 iPhone 或 iPad 内部运行一个带命令行的小型 Linux 系统。","summary":"iSH 在 iPhone 或 iPad 上提供一个 Linux 命令行。它模拟一个 32 位 x86 处理器并转换 Linux 系统调用,因此可以直接在手机上运行一个 Alpine Linux 环境及其命令行程序,不需要服务器。它适合习惯使用终端的人:打开之后,你输入 shell 命令即可。这个应用已上架 App Store,并有 TestFlight 测试版;这个仓库保存着源代码,其中的构建说明(Xcode、Meson、Ninja、Clang)是给开发者看的。代码采用 GPL 许可,并附有一条关于 App Store 条款的附加说明。","easeRationale":"这是一个 iPhone 和 iPad 应用,因此没有任何电脑系统获得评分:README 的第一个链接就是它在 App Store 上的页面,在那里它像其他应用一样安装。在 Mac 或 Linux 电脑上,只描述了面向开发者的构建方式(Xcode,或带有 Meson、Ninja、Clang、LLD、sqlite3 和 libarchive 的 Python),外加手动创建一个 Alpine 文件系统。","notVerified":["App Store 上的应用是否基于这份源代码构建,以及那个版本有多旧","在特定的 iPhone 或 iPad 上,模拟运行的 Linux 程序实际速度如何,耗电情况怎样","哪些 Alpine Linux 软件包能够安装并在其中正常运行","安装软件包之后,Linux 文件占用多少存储空间","应用在后台是否能持续运行,还是会被 iOS 终止"]}}},{"id":"R_kgDOVaRZQw","slug":"jacke-pp__open-yoki","name":"Open Yoki","owner":"jacke-pp","fullName":"jacke-pp/open-yoki","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 77 天前","source":"check","title":"所有者账号很新"},{"code":"crypto_promotion_main_use","severity":"attention","evidence":"这位主播是围绕加密货币直播而打造的：README 说她在 Binance Square 上直播，借助公开价格和 CoinAnk 的合约数据来 \"懂行情\"，并且在 TwinOasis 广场上 \"还能喊单，喊的币会在公开的战绩榜上跟踪涨跌\"。一个向观众吹捧币种的 AI 人设，可能触碰金融推广方面的规定，尽管 README 写着 \"没有钱包、交易、下注、红包抽奖、发币相关功能\"。","source":"ai","title":"主要用途是推广加密货币"},{"code":"ai_posts_on_your_accounts_unattended","severity":"attention","evidence":"可选的 \"heart\" 是一个长期运行的 AI（Claude Code 或 DeepSeek），它有自己的笔记文件夹，并 \"决定发什么\"：它写的内容会通过 ops/square_post.py 发到 Binance Square，并通过 X API 发到 X，还能在每日美元预算内搜索网页（HEART_WEB_USD，默认 20）。所描述的流程里，没有任何一步要求有人在发布前先批准。","source":"ai","title":"AI 无人看管地在你的账号上发帖"},{"code":"voice_cloning_and_digital_face","severity":"info","evidence":"\"做你自己的主播\" 的第 4 步：\"复刻声音 (app/tools/volc-clone.mjs)\"，并把形象和动作上传到 Vidu。克隆不属于你的声音或面孔，或把结果冒充真人，在一些国家可能违反平台规则和法律；README 引用的 TwinOasis 规则中包括 \"不假装是真人\"。","source":"ai","title":"声音克隆与数字人脸"},{"code":"asks_for_your_douyin_cookie","severity":"attention","evidence":".env.example：\"DOUYIN_COOKIE=\" - \"抖音 cookie 文件的路径（文件里放浏览器复制出来的 cookie）\"。把你自己浏览器里的 cookie 复制到服务器上，等于把你已登录的抖音会话交给了该服务器，而用它进行自动读取可能导致账号被限制。","source":"ai","title":"要求提供你的抖音 cookie"},{"code":"snapshot_of_a_closed_product","severity":"info","evidence":"README：\"这份代码截自 yoki 10 月 24 日跑的版本（v150），之后的新功能分批再放\" - 这份代码是线上产品在 10 月 24 日的截取版本，之后的新功能会分批发布。GitHub 上没有发布版本，唯一给出的联系方式是在 X 上私信 @Jacke_pp。","source":"ai","title":"闭源产品的快照"},{"code":"chat_ingest_key_lets_anyone_speak_to_her","severity":"info","evidence":"README 的 Security 部分：\"弹幕接入的钥匙谁拿到都能往她的弹幕里塞话\" - 谁拿到接入密钥，就能向主播的弹幕里推送任何消息，所以这把密钥必须留在读取你直播间的那台机器上。","source":"ai","title":"聊天接入密钥让任何人都能对她说话"}],"platforms":["linux"],"tools":["node","ffmpeg","other"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenRouter (the model that talks)","Volcengine Doubao voice, plus a voice ID","Vidu (the moving digital face, needed for a real stream)","Volcengine RTC (sends the picture to the stream)","Anthropic or DeepSeek (the optional long-term memory, the \"heart\")","X (posting and replying)","Binance Square OpenAPI (publishing posts)","CoinAnk (crypto market data)","TikHub, plus a Douyin cookie (trending topics)","fal.ai (backup voice)","Telegram bot (fan and team groups)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","video","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true}],"declaredAlternatives":[],"stars":1809,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":855,"growthDays":7,"history":[{"date":"2026-10-30","stars":477},{"date":"2026-10-31","stars":592},{"date":"2026-11-01","stars":731},{"date":"2026-11-02","stars":838},{"date":"2026-11-03","stars":954},{"date":"2026-11-04","stars":1073},{"date":"2026-11-05","stars":1204},{"date":"2026-11-06","stars":1333},{"date":"2026-11-07","stars":1457},{"date":"2026-11-08","stars":1578},{"date":"2026-11-09","stars":1698},{"date":"2026-11-10","stars":1809}],"firstPublishedAt":"2026-10-30T01:57:54.315Z","commitSha":"2447d748dac4839a8ca238c8bf697ebfd12d398c","verifiedAt":"2026-10-30T01:45:01.738Z","lastCommitAt":"2026-10-28T09:20:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs your own AI live streamer on a server: a cloned voice, a moving digital face, chat replies, a diary and its own posts.","summary":"The code behind \"yoki\", an AI host that streams nightly on Binance Square. It builds a 16:9 live picture with a real-time digital face, a cloned voice and subtitles, pushes it out by RTMP, reads the chat and answers in seconds, remembers regulars, keeps a diary and posts on its own to Binance Square and X. The face, voice, personality and accounts of the original host are not included: you supply your own. It is made for an Ubuntu server (4 cores, 8 GB) and needs several paid services. The README is in Chinese, with an English version.","easeRationale":"This is server software: the README asks for an x86_64 Ubuntu machine with 4 cores and 8 GB, Node 22 or newer, ffmpeg and a headless Chrome, then deploy/install.sh, systemd services and Caddy in front, a key file under /etc/livehost/, keys from several paid services, a cloned voice and a digital face built on Vidu. The author says the step-by-step guide in SETUP.md is \"written for the AI that sets it up for you\". You can start it on your own computer to check that it loads, but without keys there is no voice and no face.","costNote":"The code is free (MIT), the running costs are not. The README's own table: the Vidu real-time digital human is billed per second of connection and is the largest item, Volcengine RTC a few yuan per stream, Doubao voice about 3 yuan per 10,000 characters plus a paid voice-cloning slot, OpenRouter a few cents per stream, and the optional Claude \"heart\" a few dollars a day. A server is extra.","notVerified":["What a night of streaming really costs, since the Vidu face is billed by the second of connection","Whether it works outside Ubuntu, and whether the servers suggested (Hong Kong, Singapore, Japan) are needed for the services to answer","Whether Binance Square, X or Douyin allow this kind of automated account and posting","What the \"heart\" writes and publishes over days, since it keeps its own notes and decides on its own","How much disk the recordings use (the README warns SQUARE_TAP can take several GB a night)","Whether the English README matches the Chinese one, which is the one kept up to date"]},"zh-Hans":{"summaryShort":"在服务器上运行你自己的 AI 直播主播:克隆的声音、会动的数字人脸、聊天回复、日记,以及自己发的帖子。","summary":"“yoki”背后的代码,一个每晚在 Binance Square 直播的 AI 主播。它会生成一幅 16:9 的实时画面,带实时数字人脸、克隆的声音和字幕,通过 RTMP 推流出去,读取聊天并在几秒内回复,记住常客,写日记,还会自主在 Binance Square 和 X 上发帖。原主播的人脸、声音、个性和账号都不包含在内:需要你自己提供。它是为 Ubuntu 服务器(4 核、8 GB)设计的,还需要几项付费服务。README 是中文的,附有英文版本。","easeRationale":"这是服务器软件:README 要求一台 x86_64 的 Ubuntu 机器,4 核、8 GB 内存,Node 22 或更新版本,ffmpeg 和无头 Chrome,然后运行 deploy/install.sh,配置 systemd 服务和前置的 Caddy,在 /etc/livehost/ 下放一个密钥文件,以及几项付费服务的密钥、一份克隆的声音和基于 Vidu 构建的数字人脸。作者说 SETUP.md 里的分步指南是“写给帮你搭建的 AI 看的”。你可以在自己电脑上启动,检查它能否加载,但没有密钥就没有声音也没有人脸。","costNote":"代码是免费的(MIT),运行成本不是。README 自带的表格显示:Vidu 实时数字人按连接秒数计费,是最大的一项开支;Volcengine RTC 每场直播几元人民币;豆包语音大约每万字 3 元,外加一个付费的声音克隆名额;OpenRouter 每场直播几分钱;可选的 Claude“心脏”每天几美元。服务器费用另算。","notVerified":["一晚直播到底要花多少钱,因为 Vidu 人脸是按连接秒数计费的","是否能在 Ubuntu 之外的系统上运行,以及是否必须使用建议的服务器地区(香港、新加坡、日本)才能让这些服务正常响应","Binance Square、X 或抖音是否允许这种自动化账号和发帖行为","这个“心脏”在日积月累中会写些什么、发些什么,因为它自己记笔记并自行决定","录制内容会占用多少磁盘空间(README 警告 SQUARE_TAP 一晚可能占用好几个 GB)","英文版 README 是否与中文版一致,因为中文版才是持续更新的那一份"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1NDgyMDQ2NQ==","slug":"jackyandroid__androidtvlauncher","name":"Android TV Launcher","owner":"JackyAndroid","fullName":"JackyAndroid/AndroidTVLauncher","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2020-03-16(2424 天前)","source":"check","title":"版本更新停滞"},{"code":"signing_key_published_in_the_repository","severity":"attention","evidence":"顶层文件列表中包含\"AndroidTVLauncherKey.jks\"——安卓的签名密钥库被提交到了仓库里。任何拿到它的人都能编译出一个你的电视会当成这个应用更新来接受的 APK。","source":"ai","title":"仓库中发布了签名密钥"},{"code":"no_downloadable_file_in_any_release","severity":"attention","evidence":"README 提供了\"[Download Apk](https://github.com/JackyAndroid/AndroidTVLauncher/releases)\"链接,但这个包里列出的三个发布版本——2.1(2020-03-16)、V2.0(2016-09-14)和 v1.0(2016-04-09)——全都显示\"no platform binaries\",所以承诺的现成下载根本不存在可供检查。","source":"ai","title":"任何发布版本都没有可下载的文件"},{"code":"abandoned_and_built_for_very_old_android","severity":"attention","evidence":"简介写道:\"This is a leanback style tv launcher(minSdkVersion 17)\"(即 2012 年的 Android 4.2)。最近一次发布是 2020-03-16 的 2.1 版,最近一次提交是 2022-11-14,而\"next step to improve\"里写的仍然是\"Follow up the official Leanback library new features\"。","source":"ai","title":"项目已被放弃,且是为很老的安卓版本构建的"},{"code":"replacing_the_home_screen_is_hard_to_undo","severity":"info","evidence":"这个项目是一个电视启动器,也就是接管主屏幕按钮的应用。如果它崩溃或者不适配你的电视,用遥控器切换回原来的主屏幕可能会很麻烦。","source":"ai","title":"更换主屏幕之后很难恢复"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[{"slug":"projectivy-launcher","name":"Launcher","confirmed":true}],"declaredAlternatives":[],"stars":3146,"downloads":17220,"downloadsWeek":131,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":3148},{"date":"2026-11-06","stars":3147},{"date":"2026-11-07","stars":3147},{"date":"2026-11-08","stars":3147},{"date":"2026-11-09","stars":3146},{"date":"2026-11-10","stars":3146}],"firstPublishedAt":"2026-11-04T22:47:04.183Z","commitSha":"aa20c0a53da7faedd6e20b6bb10f5933a6fdcdd6","verifiedAt":"2026-11-04T22:41:27.819Z","lastCommitAt":"2022-11-14T01:55:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Replaces the home screen of an Android TV box with a Leanback-style one that lists your apps, photos and videos.","summary":"AndroidTVLauncher (also called CatLauncher) is an Android TV home screen in Google's old \"Leanback\" style: rows of cards with your installed apps, a photo viewer, media details and a built-in video player based on ijkplayer. It targets Android 4.2 and newer (minSdkVersion 17). The README links \"Download Apk\" to the releases page, but no release file is listed there, so in practice you build the app from the Gradle project yourself, sideload it on the TV box and set it as the home screen. GPL-3.0; last release 2020-03-16, last code change 2022-11-14.","easeRationale":"It is an Android TV app, so none of the desktop platforms get a grade. The README only links \"Download Apk\" to the GitHub releases page and no file is listed for any of the three releases, the newest being 2.1 from 2020-03-16, so the only route left is building this Gradle project yourself, sideloading the result on the TV box and choosing it as the home screen.","costNote":"Free, GPL-3.0. No account and no purchase.","notVerified":["Whether any installable APK is in fact offered behind the README's \"Download Apk\" link","Whether the project still builds with current Android tooling and runs on a current Android TV or Google TV device","Whether the built-in ijkplayer-based video player still plays current formats","What the launcher connects to while showing media details"]},"zh-Hans":{"summaryShort":"把安卓电视盒子的主屏幕换成 Leanback 风格的界面,列出你的应用、照片和视频。","summary":"AndroidTVLauncher(也叫 CatLauncher)是一个采用 Google 旧版\"Leanback\"风格的安卓电视主屏幕:一行行卡片展示你安装的应用、一个照片查看器、媒体详情,以及一个基于 ijkplayer 的内置视频播放器。它面向 Android 4.2 及以上版本(minSdkVersion 17)。README 里的\"Download Apk\"链接指向发布页面,但那里没有列出任何发布文件,所以实际上你需要自己从 Gradle 项目编译这个应用,在电视盒子上绕过应用商店安装,并把它设为主屏幕。GPL-3.0 许可证;最近一次发布是 2020-03-16,最近一次代码改动是 2022-11-14。","easeRationale":"这是一款安卓电视应用,所以没有任何桌面平台会被评级。README 只是把\"Download Apk\"链接指向 GitHub 的发布页面,而三个发布版本——最新的是 2020-03-16 的 2.1 版——都没有列出任何文件,所以唯一的办法是自己编译这个 Gradle 项目,在电视盒子上绕过应用商店安装结果,并选择它作为主屏幕。","costNote":"免费,GPL-3.0 许可证。不需要账号,也不需要购买。","notVerified":["README 的\"Download Apk\"链接背后是否真的提供了可安装的 APK","这个项目是否还能用现在的安卓工具链编译,能否在当前的安卓电视或 Google TV 设备上运行","内置的基于 ijkplayer 的视频播放器是否还能播放现在的格式","这个启动器在显示媒体详情时连接到了什么地方"]}}},{"id":"R_kgDOVRA2_g","slug":"jakeschincariol__day-trader-skill","name":"Day Trader Skill","owner":"Jakeschincariol","fullName":"Jakeschincariol/day-trader-skill","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"15 天内获得 5247 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"very_new_repository_high_star_count","severity":"attention","evidence":"该仓库创建于 2026-10-20,最后一次 push 也是同一天,2026-10-20——截至 2026-11-04 本次检查时,已经 15 天没有提交——但已经有 5247 个星标和 1277 个 fork。","source":"ai","title":"非常新的仓库却有很高的星标数"},{"code":"can_send_orders_to_your_broker","severity":"attention","evidence":"README:\"/trader-broker | Order Executor | Sends approved orders to your broker, paper trading by default, never without your yes.\"同一份 README 还写道\"Most people who day trade lose money. You can lose money.\"如果你关闭模拟交易或批准了一个订单,就可能亏掉真实的钱。","source":"ai","title":"可以向你的经纪商发送订单"},{"code":"unofficial_data_source","severity":"info","evidence":"README:\"Yahoo Finance's chart endpoint is not an official API: it can lag, change or rate-limit you. The scanner and the entry timer read 1-minute or 5-minute bars, not every tick.\"","source":"ai","title":"非官方数据来源"},{"code":"sends_your_name_and_email_to_sec","severity":"info","evidence":"README:\"The SEC tools (news filings, the analyst) ask you to identify yourself: put your name and email in `trader/config.json` (`\"sec_user_agent\": \"Your Name you@example.com\"`) or in the `SEC_USER_AGENT` environment variable.\"这个姓名和邮箱会随每一次向 SEC EDGAR 发出的请求一起发出去。","source":"ai","title":"把你的姓名和邮箱发送给 sec"},{"code":"simulation_presented_as_odds","severity":"info","evidence":"Time Machine 给出的百分比,例如\"target first 29.4 %, stop first 25.5 %\",只是根据这只股票最近 60 天的 5 分钟 K 线构建出来的。README 自己也说:\"The future can do things those 60 days never did... Treat its odds as a reality check, never as a promise.\"","source":"ai","title":"把模拟结果当成概率呈现"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Alpaca (brokerage, only to send real or paper orders)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","productivity","automation"],"alternativeTo":[{"slug":"tradingview","name":"TradingView","confirmed":true},{"slug":"benzinga-pro","name":"Pro","confirmed":true}],"declaredAlternatives":[],"stars":6637,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1720,"growthDays":7,"history":[{"date":"2026-10-27","stars":2328},{"date":"2026-10-28","stars":2653},{"date":"2026-10-29","stars":3247},{"date":"2026-10-30","stars":3619},{"date":"2026-10-31","stars":3979},{"date":"2026-11-01","stars":4328},{"date":"2026-11-02","stars":4591},{"date":"2026-11-03","stars":4917},{"date":"2026-11-04","stars":5242},{"date":"2026-11-05","stars":5547},{"date":"2026-11-06","stars":5755},{"date":"2026-11-07","stars":5949},{"date":"2026-11-08","stars":6167},{"date":"2026-11-09","stars":6387},{"date":"2026-11-10","stars":6637}],"firstPublishedAt":"2026-10-27T18:40:27.943Z","commitSha":"3d0f4ef6986f4984b7c6bd46f7ff4f87744c3abd","verifiedAt":"2026-11-04T22:41:57.520Z","lastCommitAt":"2026-10-20T14:17:50Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Eleven add-ons for Claude that scan stock charts, follow news and plan day trades, with your approval before any order.","summary":"Day Trader skill is a pack of 11 instruction files for Claude and Claude Code, plus small Python scripts that read free public data: Yahoo Finance charts, Nasdaq symbol lists, SEC EDGAR and US House filings. They scan charts for setups, follow news, time the entry, set a stop and a target, size the position to your budget and replay the session 10,000 times before you risk money. It is for people who already use Claude and trade stocks. You need Python 3.8 or newer and there is nothing to pip install; real orders go through your own broker and only after you approve each one.","easeRationale":"Windows, macOS and Linux are all C: you must already have Claude or Claude Code, then run one install step (paste the repository link with \"install skill\", or /plugin marketplace add Jakeschincariol/day-trader-skill and /plugin install day-trader-skill@day-trader-skill), and have Python 3.8 or newer for the scripts, with nothing to pip install. Sending real orders needs a brokerage connector on top of that.","costNote":"The skills are free (MIT) and the README says there is no signup and no API key. They only work inside Claude, which usually needs a paid plan or API usage, and real orders need your own brokerage account.","notVerified":["Whether the scripts still work against Yahoo Finance, Nasdaq Trader, SEC EDGAR and the House Clerk today, since none of those is an official supported interface.","Whether the numbers in the Tesla example can be reproduced outside the offline test fixtures.","Whether the risk and budget checks really block an order that breaks your limits.","Whether anything other than an order you approved can reach your brokerage account.","How much Claude usage a full scan plus 10,000 replays consumes on a paid plan.","How much disk space it uses: the checks could not estimate it."]},"zh-Hans":{"summaryShort":"11 个 Claude 插件，扫描股票图表、追踪新闻、规划日内交易，下单前都需要你批准。","summary":"Day Trader skill 是一套包含 11 个指令文件的插件，供 Claude 和 Claude Code 使用，外加一些读取免费公开数据的小型 Python 脚本：Yahoo Finance 图表、纳斯达克代码列表、SEC EDGAR 和美国众议院的申报文件。它扫描图表寻找交易机会，追踪新闻，把握入场时机，设置止损和目标，按你的预算调整仓位大小，并在你真正投入资金前把这次交易重放 10000 次。它适合已经在用 Claude 并从事股票交易的人。你需要 Python 3.8 或更新版本，不需要用 pip 安装任何东西；真实下单要经过你自己的券商，而且只有在你批准每一单之后才会执行。","easeRationale":"Windows、macOS 和 Linux 都是 C 级：你必须已经有 Claude 或 Claude Code，然后只需一步安装（用 \"install skill\" 粘贴仓库链接，或者执行 /plugin marketplace add Jakeschincariol/day-trader-skill 和 /plugin install day-trader-skill@day-trader-skill），脚本需要 Python 3.8 或更新版本，不需要用 pip 安装任何东西。要真实下单，还需要额外接入券商连接器。","costNote":"这些技能是免费的（MIT 许可），README 说不需要注册也不需要 API 密钥。它们只能在 Claude 里运行，通常需要付费套餐或按 API 用量付费，真实下单还需要你自己的券商账户。","notVerified":["脚本如今是否仍能正常访问 Yahoo Finance、Nasdaq Trader、SEC EDGAR 和美国众议院文书处，因为这些都不是官方支持的接口。","特斯拉示例里的数字，能否在离线测试样例之外重现。","风险和预算检查是否真的能挡住超出你限额的订单。","除了你批准的订单之外，是否还有其他东西能触达你的券商账户。","一次完整扫描加上 10000 次重放，在付费套餐上会消耗多少 Claude 用量。","它占用多少磁盘空间：检查无法估算这一点。"]}}},{"id":"R_kgDOU6HY4w","slug":"jakeschincariol__replica-skill","name":"Replica","owner":"Jakeschincariol","fullName":"Jakeschincariol/replica-skill","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"23 天内获得 5794 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"app_cloning_legal_risk","severity":"attention","evidence":"描述：\"clone any app: reverse-engineer it, rebuild it... then fix what its users hate\"；README：\"so you have an app you can sell\" 以及 \"talk to a lawyer before launch if there is money on the line. This is not legal advice.\"","source":"ai","title":"克隆应用的法律风险"},{"code":"single_day_repository","severity":"info","evidence":"创建于 2026-10-03，最近一次推送也是 2026-10-03，此后没有任何提交，却已有 5794 个星标和 1017 个 fork。","source":"ai","title":"只有一天动静的仓库"},{"code":"deploys_to_production","severity":"info","evidence":"README.md：`/replica-deploy` \"sets up production, gives you the DNS records for your domain, and ships it when you say go\"。","source":"ai","title":"部署到生产环境"}],"platforms":["windows","mac","linux","web"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Claude account (Claude Code or Claude chat)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":8784,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1367,"growthDays":7,"history":[{"date":"2026-10-26","stars":5797},{"date":"2026-10-27","stars":6068},{"date":"2026-10-28","stars":6281},{"date":"2026-10-29","stars":6487},{"date":"2026-10-30","stars":6673},{"date":"2026-10-31","stars":6820},{"date":"2026-11-01","stars":6999},{"date":"2026-11-02","stars":7175},{"date":"2026-11-03","stars":7417},{"date":"2026-11-04","stars":7624},{"date":"2026-11-05","stars":7765},{"date":"2026-11-06","stars":7920},{"date":"2026-11-07","stars":8181},{"date":"2026-11-08","stars":8428},{"date":"2026-11-09","stars":8641},{"date":"2026-11-10","stars":8784}],"firstPublishedAt":"2026-10-27T01:10:12.639Z","commitSha":"77c9436fb3d18c3d58169efb8caf4fe906b0dc51","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-10-03T12:07:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Eleven instruction sets for Claude that guide it to study an existing app and build, rebrand and launch your own version.","summary":"Replica is a pack of eleven \"skills\", written instructions that teach Claude a step-by-step method. They walk Claude through studying an existing app from its public pages, planning and building a similar app, testing it, reading user reviews for complaints to fix, giving it a new name and brand, and putting it online. It is aimed at indie founders who want to build a competing product. You need Claude (ideally Claude Code) and Python 3.8+ for the small helper scripts. The generated app needs its own development and hosting setup. The README says it copies features, not code, logos or content.","easeRationale":"Installing is one command in Claude Code (or pasting the link into Claude), but the helper scripts need Python and the app it builds needs a full development and hosting setup.","costNote":"The skills are free (MIT) and need no API key, but they run inside Claude, and the app you build needs its own hosting and services.","notVerified":["Whether the skills really stop Claude from copying another app's code, text or images, as the README promises.","How much Claude usage it takes to build a full app with all eleven steps.","Whether selling a clone built this way is allowed under the law and trademarks in your country.","How well the generated app works compared with the original."]},"zh-Hans":{"summaryShort":"十一套给 Claude 的指令，引导它研究现有应用，并构建、更换品牌和发布你自己的版本。","summary":"Replica 是一个包含十一项“skills”的合集，即教 Claude 按步骤做事的书面说明。它们带着 Claude 依次完成：从公开页面研究现有应用，规划并构建类似的应用，测试，阅读用户评价以找出可修复的抱怨，给应用起新名字和新品牌，并上线发布。它面向想打造竞品的独立创始人。你需要 Claude（最好是 Claude Code）和 Python 3.8+ 来运行小型辅助脚本。生成的应用需要自己的开发和托管环境。README 称它复制的是功能，而不是代码、标志或内容。","easeRationale":"安装只需在 Claude Code 中运行一条命令（或把链接粘贴给 Claude），但辅助脚本需要 Python，它所构建的应用则需要完整的开发和托管环境。","costNote":"这些 skills 免费（MIT）且不需要 API 密钥，但它们在 Claude 内运行，你构建的应用需要自己的托管和服务。","notVerified":["这些 skills 是否真的像 README 承诺的那样，阻止 Claude 复制其他应用的代码、文字或图片。","用全部十一个步骤构建一个完整应用需要消耗多少 Claude 用量。","在你所在的国家，出售以这种方式做出的仿制应用是否符合法律和商标规定。","生成的应用与原应用相比运行得如何。"]}}},{"id":"R_kgDOUfXmpg","slug":"jaredpalmer__kev","name":"Kev","owner":"jaredpalmer","fullName":"jaredpalmer/kev","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"40 天内获得 9621 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"package_name_collision","severity":"attention","evidence":"README.md 警告：\"PyPI name collision: `pip install kev` installs a **different package** — K.E.V. ORM ... This project is **not** published to PyPI. Install from source as below.\"","source":"ai","title":"包名与另一个包重名"},{"code":"server_without_key_by_default","severity":"info","evidence":"README.md：\"The server binds to 127.0.0.1 (`--host 0.0.0.0` to accept other machines) and is open by default; set `KEV_API_KEY` to require `Authorization: Bearer <key>` on `/v1/*`.\"","source":"ai","title":"服务器默认不设密钥"},{"code":"downloads_code_at_setup","severity":"info","evidence":"部署说明会从互联网获取并运行一个脚本：\"curl -LO https://raw.githubusercontent.com/jaredpalmer/kev/main/skills/kev-deploy/scripts/kev_serve.py\"，随后是 \"modal deploy kev_serve.py\"；微调路径则使用 \"npx skills add jaredpalmer/kev@kev-finetune\"。","source":"ai","title":"安装配置时会下载代码"},{"code":"accuracy_claims_self_measured","severity":"info","evidence":"README 中与 Jev 服务的对比表是项目自己做的，并且它也这样说明：\"We don't know what Jev was trained on, so this isn't a controlled comparison of the two architectures.\"","source":"ai","title":"准确率说法由项目自己测得"}],"platforms":["linux","mac","web"],"tools":[],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["Modal (only to deploy a hosted endpoint)","Vercel AI Gateway (only to compare against the Jev service)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","local_ai"],"alternativeTo":[{"slug":"jev","name":"Jev","confirmed":true}],"declaredAlternatives":[],"stars":9931,"downloads":1254,"downloadsWeek":135,"downloadsDays":7,"growth":131,"growthDays":7,"history":[{"date":"2026-10-26","stars":9571},{"date":"2026-10-27","stars":9621},{"date":"2026-10-28","stars":9662},{"date":"2026-10-29","stars":9694},{"date":"2026-10-30","stars":9721},{"date":"2026-10-31","stars":9737},{"date":"2026-11-01","stars":9753},{"date":"2026-11-02","stars":9771},{"date":"2026-11-03","stars":9800},{"date":"2026-11-04","stars":9822},{"date":"2026-11-05","stars":9841},{"date":"2026-11-06","stars":9859},{"date":"2026-11-07","stars":9877},{"date":"2026-11-08","stars":9895},{"date":"2026-11-09","stars":9915},{"date":"2026-11-10","stars":9931}],"firstPublishedAt":"2026-10-27T01:10:12.640Z","commitSha":"2074cc44256200bb37540103fa4572901faa10f3","verifiedAt":"2026-10-28T01:36:25.859Z","lastCommitAt":"2026-10-22T19:27:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Small AI models that answer yes/no, multiple-choice and rating questions about a text, run on your own machine.","summary":"Kev is a family of small AI models that read a piece of text and answer set questions about it: yes or no, pick one option, or give a rating, each with a probability. It is for developers who classify or route content, such as support tickets, and want it running on their own machine or server instead of a paid API. The server copies TypeSafe's System One API, so code written for Jev can point at Kev. You need Python 3.12 or 3.13, the uv tool and the terminal; a GPU for speed. A browser demo exists for the two smallest models.","easeRationale":"The best grade is on the web: the README links a Hugging Face Space that runs the two smaller models with nothing to install. Running it yourself is grade C on Mac and Linux, because you install Python 3.12 or 3.13 plus the uv tool, clone the repository and run a few terminal commands; the bigger models need a large GPU (Kev-27B asks for an 80 GB card or a 96-128 GB Mac), and Windows is not mentioned in the instructions.","costNote":"The code and the model weights are free. Running it costs hardware: the README quotes rented GPUs from $0.80 to $6.25 per hour and about $1 for one fine-tuning run.","notVerified":["Whether the published accuracy and speed numbers reproduce on your own data and hardware","Real disk use after the model files download (the 0.8B adapter is small, the 27B model is 51 GB of weights)","Whether it runs on Windows, which the instructions do not cover","The real monthly cost of a hosted endpoint on Modal, including the warm container option that keeps a paid GPU running while idle","Whether the Hugging Face browser demo keeps or logs the text you paste into it"]},"zh-Hans":{"summaryShort":"在自己电脑上运行的小型 AI 模型，可以针对一段文本回答是/否、多选和评分类问题。","summary":"Kev 是一系列小型 AI 模型，能读取一段文本并回答固定类型的问题：是或否、从选项中挑一个，或给出一个评分，每个答案都带有概率。它面向需要对内容进行分类或分流的开发者，比如处理客服工单，并希望让它运行在自己的电脑或服务器上，而不是使用付费 API。这个服务器复刻了 TypeSafe 的 System One API，所以为 Jev 写的代码也可以直接指向 Kev。你需要 Python 3.12 或 3.13、uv 工具和终端；用 GPU 可以提升速度。两个最小的模型还提供了浏览器演示版本。","easeRationale":"最好的评分在网页端：README 链接到一个 Hugging Face Space，可以运行两个较小的模型，无需安装任何东西。自己运行的评分在 Mac 和 Linux 上是 C，因为需要安装 Python 3.12 或 3.13 加 uv 工具，克隆仓库，再运行几条终端命令；更大的模型需要大显存的 GPU（Kev-27B 要求 80 GB 显存的显卡或 96-128 GB 内存的 Mac），说明文档里没有提到 Windows。","costNote":"代码和模型权重都是免费的。运行需要花钱买硬件：README 里提到租用 GPU 的价格在每小时 0.80 到 6.25 美元之间，一次微调大约花费 1 美元。","notVerified":["公开的准确率和速度数据在你自己的数据和硬件上能否重现","下载模型文件后实际占用的磁盘空间（0.8B 的适配器很小，27B 模型的权重有 51 GB）","能否在 Windows 上运行，说明文档里没有涉及这一点","在 Modal 上托管一个端点的实际月费是多少，包括让付费 GPU 在空闲时也保持运行的“热容器”选项","Hugging Face 的浏览器演示是否会保留或记录你粘贴进去的文本"]}}},{"id":"R_kgDOK64U5Q","slug":"jarodmica__ai-voice-cloning","name":"AI Voice Cloning","owner":"JarodMica","fullName":"JarodMica/ai-voice-cloning","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在2024-04-23（929天前）","source":"check","title":"发布已过时"},{"code":"archived","severity":"info","evidence":"该仓库已被归档（只读，不再维护）","source":"check","title":"已归档"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖中存在已知漏洞（OSV）：gradio：GHSA-26jh-r8g2-6fpr、GHSA-279j-x4gx-hfrh、GHSA-37qc-qgx6-9xjv；transformers：GHSA-282v-666c-3fvg、GHSA-29pf-2h5f-8g72、GHSA-37mw-44qp-f5jm；starlette：GHSA-2c2j-9gv5-cj73、GHSA-82w8-qh3p-5jfq、GHSA-86qp-5c8j-p5mr","source":"check","title":"已知漏洞"},{"code":"clones_the_voice_of_real_people","severity":"attention","evidence":"项目的全部目的就是从样本中复制一个声音，README中RVC那一节让你带上\"a **trained** RVC .pth file that you get from RVC or online\"，放进`models/rvc_models/`。让一个真实的人看起来说了他从未说过的话，可能侵犯声音和形象权，也可能触犯多个国家中关于冒充他人的法律。","source":"ai","title":"克隆真实人物的声音"},{"code":"installs_code_from_many_outside_sources","severity":"attention","evidence":"Dockerfile在安装过程中直接从第三方地址拉取代码和模型包：`curl -L -o /tmp/rvc.zip https://huggingface.co/Jmica/rvc/resolve/main/rvc_lightweight.zip`、`pip install git+https://github.com/liyaodev/fairseq`、`git+https://github.com/JarodMica/rvc-tts-pipeline.git@lightweight`、`git+https://github.com/m-bain/whisperx.git`，还会克隆一份`pyfastmp3decoder`。这些都没有锁定固定版本，所以你安装当天得到的内容会不一样。","source":"ai","title":"从许多外部来源安装代码"},{"code":"compiled_binary_committed_to_the_repository","severity":"attention","evidence":"顶层文件列表中有`deepspeed-0.8.3+6eca037c-cp39-cp39-win_amd64.whl`，这是一个预先编译好的Windows Python包，存放在仓库里并由安装脚本安装。它的内容在源码中无法查看。","source":"ai","title":"仓库中提交了编译好的二进制文件"},{"code":"ready_package_hosted_outside_github","severity":"info","evidence":"README的Setup部分写道：\"Head over to the releases tab and download the latest package on Hugging Face\" —— Windows安装包并不是GitHub发布版的附件，这也是为什么Radar的检查在v3.0、v2.0和v1.0发布版中都发现\"no platform binaries\"（没有平台二进制文件）。","source":"ai","title":"现成安装包托管在github之外"},{"code":"web_interface_without_a_password","severity":"info","evidence":"README的Building and Running in Docker部分写道：\"Visit `http://localhost:7860` or remotely with `http://<ip>:7860`\"。这个Gradio页面没有登录环节，所以任何能连到那个端口的人都能使用机器上的声音和文件。","source":"ai","title":"网页界面没有密码"},{"code":"author_points_to_newer_projects","severity":"info","evidence":"README写道：\"I don't have any further plans to maintain this repo (Tortoise TTS) so I will be archiving it... There are better alternatives out there that work just as well or better than Tortoise: XTTS, GPT-SoVITS, F5TTS, StyleTTS2, Zonos, etc.\" 代码只是保持在\"in a working state\"（可运行的状态）。","source":"ai","title":"作者建议改用更新的项目"}],"platforms":["windows","linux"],"tools":["python","docker","gpu_any","ffmpeg"],"buildTools":["python","docker","gpu_any","ffmpeg"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true}],"declaredAlternatives":[],"stars":774,"downloads":0,"downloadsWeek":0,"downloadsDays":1,"growth":0,"growthDays":6,"history":[{"date":"2026-11-09","stars":774},{"date":"2026-11-10","stars":774}],"firstPublishedAt":"2026-11-08T23:01:18.734Z","commitSha":"684794b62babe5ddbddcf0ba15d31187fd0ba8c7","verifiedAt":"2026-11-08T22:49:36.084Z","lastCommitAt":"2025-06-09T01:21:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from a few audio samples and makes it read any text on your own PC, with an NVIDIA graphics card.","summary":"AI Voice Cloning is a web page that runs on your own computer, built around the Tortoise TTS model, where you upload samples of a voice and have it read text you type. It can also train on your own recordings, transcribe them with Whisper or WhisperX, and pass the result through RVC to make it sound closer to the target voice. It needs an NVIDIA graphics card: on Windows you unpack a ready package and run start.bat, on Linux you build a Docker image. The author archived the project in 2025 and points to newer models instead.","easeRationale":"Windows is the easier side and still a D: you need an NVIDIA card, then you unpack a 7-Zip package hosted on Hugging Face and run start.bat, which downloads most of the models on the first start, with more downloaded later during generation and training. Linux means Docker plus the NVIDIA Container Toolkit and building the image yourself with ./setup-docker.sh.","costNote":"GPL-3.0; everything runs on your own machine, so the cost is the hardware and the electricity.","notVerified":["How much disk the models really take after the first start and after training, since they are downloaded as you go","Whether the setup still completes today, with unpinned installs from GitHub and Hugging Face and two years without maintenance","Whether an AMD card works through the setup-rocm or DirectML scripts, which the README does not cover","How close the cloned voice gets, and how long generation and training take on a given graphics card","Whether the known vulnerabilities in the pinned gradio, transformers and starlette versions can be reached in normal use"]},"zh-Hans":{"summaryShort":"只用几段语音样本就能复刻一个声音，让它在你自己的电脑上朗读任何文本，需要一张NVIDIA显卡。","summary":"AI Voice Cloning是一个在你自己电脑上运行的网页，基于Tortoise TTS模型，你上传一段声音样本，就能让它朗读你输入的文本。它也可以用你自己的录音来训练，用Whisper或WhisperX转写文字，再把结果经过RVC处理，让声音更接近目标说话人。它需要一张NVIDIA显卡：在Windows上你解压一个现成的安装包并运行start.bat，在Linux上你要自己构建Docker镜像。作者在2025年把项目归档，并建议改用更新的模型。","easeRationale":"Windows是比较简单的一侧，但评级仍是D：你需要一张NVIDIA显卡，然后解压一个托管在Hugging Face上的7-Zip安装包并运行start.bat，它会在第一次启动时下载大部分模型，之后在生成和训练过程中还会下载更多。Linux则需要Docker加上NVIDIA Container Toolkit，并自己用./setup-docker.sh构建镜像。","costNote":"GPL-3.0许可；一切都在你自己的机器上运行，所以成本只是硬件和电费。","notVerified":["模型在第一次启动后以及训练之后到底占用多少磁盘空间，因为它们是逐步下载的","在GitHub和Hugging Face上都没有锁定版本的安装、又两年没有维护的情况下，安装流程今天是否还能顺利完成","AMD显卡能否通过setup-rocm或DirectML脚本运行，README并没有说明这一点","克隆出来的声音到底有多像，以及在某款显卡上生成和训练各需要多长时间","gradio、transformers和starlette这些锁定版本中已知的漏洞，在正常使用中是否真的会被触发"]}}},{"id":"R_kgDOIvux3g","slug":"jdx__mise","name":"mise","owner":"jdx","fullName":"jdx/mise","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接管道到shell执行:curl -fsSL https://mise.run | sh","source":"check","title":"curl管道直接执行到shell"},{"code":"downloads_and_runs_third_party_programs","severity":"attention","evidence":"这正是这个工具的用途:README 中说它会\"from a registry of 1,000+ tools or directly from backends like GitHub releases, Aqua, Cargo, npm, and PyPI\"安装工具。你要的每个工具都是从别人的服务器下载,然后在你的机器上运行。`mise lock`会记录确切的版本和校验和,但这是可选的。","source":"ai","title":"下载并运行第三方程序"},{"code":"project_file_can_set_variables_and_run_commands","severity":"info","evidence":"文件夹里的 mise.toml 可以声明环境变量、额外的 PATH 条目,以及带`run = \"...\"`命令的任务。README 中说变量和 .env 文件会\"automatically as you enter a directory\"自动加载,所以打开别人写的文件夹会改变你终端的环境。","source":"ai","title":"项目文件可以设置变量并运行命令"},{"code":"can_change_machine_wide_settings","severity":"info","evidence":"README,\"Set up a machine with mise bootstrap\":`mise bootstrap`和`mise dot apply`会\"apply system packages, dotfiles, Git repositories, macOS defaults, and background services\",`mise bootstrap --adopt you/dotfiles`会从一个 Git 仓库获取这些配置。可以先用`--dry-run`预览。","source":"ai","title":"可以更改整台机器的设置"},{"code":"large_download_for_a_command_line_tool","severity":"info","evidence":"发布文件对于单个程序来说很大:mise-v2026.10.25-windows-x64.exe 有240MB,Linux x64 二进制文件有168MB,这还不包括它为你安装的任何工具。","source":"ai","title":"命令行工具的下载体积很大"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","docker","rust"],"disk":{"minMB":240,"maxMB":719,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":36012,"downloads":97415393,"downloadsWeek":8272910,"downloadsDays":7,"growth":286,"growthDays":7,"history":[{"date":"2026-10-26","stars":35423},{"date":"2026-10-27","stars":35461},{"date":"2026-10-28","stars":35506},{"date":"2026-10-29","stars":35547},{"date":"2026-10-30","stars":35573},{"date":"2026-10-31","stars":35598},{"date":"2026-11-01","stars":35632},{"date":"2026-11-02","stars":35694},{"date":"2026-11-03","stars":35726},{"date":"2026-11-04","stars":35771},{"date":"2026-11-05","stars":35817},{"date":"2026-11-06","stars":35868},{"date":"2026-11-07","stars":35905},{"date":"2026-11-08","stars":35945},{"date":"2026-11-09","stars":35981},{"date":"2026-11-10","stars":36012}],"firstPublishedAt":"2026-10-27T01:10:12.641Z","commitSha":"99734bca80530cc446acbe435c7b189cee44febe","verifiedAt":"2026-11-01T01:39:33.118Z","lastCommitAt":"2026-11-10T22:25:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"One command-line tool that installs the right version of Node, Python or Go per project, loads its variables and runs its tasks.","summary":"mise keeps each project's programming tools, environment variables and tasks in a single mise.toml file. You write node = \"24\" or python = \"3.14\" and it downloads that exact version for that folder, so two projects can use different versions without clashing. It can also load .env files and secrets when you enter the folder, and run named tasks such as build or test. It is made for programmers who work in a terminal. One command installs it on Windows, macOS and Linux, and it needs no other tool first.","easeRationale":"Every platform is one command: on Windows `winget install jdx.mise`, and on macOS and Linux the install script from mise.run, which puts the program in ~/.local/bin. Each release also has a ready binary for Windows, macOS and Linux. After installing you add one line to your shell start-up file so it works in every new terminal, and from then on everything is typed commands.","costNote":"MIT licence. The README links a sponsors page, but nothing in the tool is paid.","notVerified":["How much disk space it ends up using, since every tool version it installs is kept in its own folder.","Whether a particular tool in the 1,000+ registry really installs and works on your platform, especially on Windows.","Whether the install script at mise.run is the same code as the repository at any given moment.","Whether shell activation conflicts with another version manager you already use, such as nvm or asdf."]},"zh-Hans":{"summaryShort":"一个命令行工具,为每个项目安装正确版本的Node、Python或Go,加载其变量并运行其任务。","summary":"mise 把每个项目的编程工具、环境变量和任务都保存在一个 mise.toml 文件里。你写 node = \"24\" 或 python = \"3.14\",它就会为那个文件夹下载那个确切的版本,这样两个项目可以使用不同版本而不冲突。它还能在你进入文件夹时加载 .env 文件和密钥,并运行像 build 或 test 这样的命名任务。它是为在终端里工作的程序员设计的。一条命令就能在 Windows、macOS 和 Linux 上安装它,而且事先不需要任何其他工具。","easeRationale":"每个平台都只需要一条命令:在 Windows 上是`winget install jdx.mise`,在 macOS 和 Linux 上是来自 mise.run 的安装脚本,会把程序放到 ~/.local/bin。每个发布版本也都有适用于 Windows、macOS 和 Linux 的现成二进制文件。安装后你需要在你的 shell 启动文件里加一行,这样它才能在每个新终端里生效,从那之后一切都是输入命令。","costNote":"MIT 许可。README 链接了一个赞助页面,但工具本身没有任何付费内容。","notVerified":["最终会占用多少磁盘空间,因为它安装的每个工具版本都保存在自己的文件夹里。","1000多个注册项里的某个具体工具是否真的能在你的平台上安装并正常工作,尤其是在 Windows 上。","mise.run 上的安装脚本在某个时刻是否与仓库里的代码完全一致。","shell 激活是否会和你已经在用的另一个版本管理器(比如 nvm 或 asdf)冲突。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNjE2NTEyNTY=","slug":"jellyfin__jellyfin-androidtv","name":"Jellyfin for Android TV","owner":"jellyfin","fullName":"jellyfin/jellyfin-androidtv","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"low","signals":[{"code":"needs_your_own_media_server","severity":"info","evidence":"README.md：\"Jellyfin for Android TV is a Jellyfin client for Android TV, Nvidia Shield, and Amazon Fire TV devices.\" 客户端只能显示服务器上已有的内容；如果没有在别处安装自己的Jellyfin服务器，就没有东西可以看。","source":"ai","title":"需要你自己的媒体服务器"},{"code":"latest_releases_are_betas","severity":"info","evidence":"最新的三个发行版是v0.20.0-beta.4（2026-10-15）、v0.20.0-beta.3和v0.20.0-beta.2，而这个创建于2018年的仓库版本号还停留在0.x。应用商店里提供的是稳定版本。","source":"ai","title":"最新版本都是测试版"},{"code":"no_app_file_in_github_releases","severity":"info","evidence":"GitHub发行版里没有列出任何平台的二进制文件。README.md让你去Google Play、Amazon Appstore、F-Droid，或者repo.jellyfin.org/releases/client/androidtv/的下载存档里获取可安装的应用。","source":"ai","title":"GitHub发行版里没有应用文件"},{"code":"build_from_source_is_developer_work","severity":"info","evidence":"README.md的Building部分：\"The app uses Gradle and requires the Android SDK\"，而`assembleDebug`生成的APK\"uses a different app-id from our stable builds\"。另外：\"The `master` branch ... is unstable and may contain breaking changes or unresolved bugs.\"","source":"ai","title":"从源码构建是开发者的工作"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["video","hardware"],"alternativeTo":[{"slug":"plex","name":"Plex","confirmed":true},{"slug":"emby","name":"Emby","confirmed":true}],"declaredAlternatives":[],"stars":4726,"downloads":526567,"downloadsWeek":4470,"downloadsDays":6,"growth":20,"growthDays":6,"history":[{"date":"2026-11-04","stars":4706},{"date":"2026-11-05","stars":4707},{"date":"2026-11-06","stars":4713},{"date":"2026-11-07","stars":4717},{"date":"2026-11-08","stars":4718},{"date":"2026-11-09","stars":4721},{"date":"2026-11-10","stars":4726}],"firstPublishedAt":"2026-11-04T01:36:50.456Z","commitSha":"f299a2c4e2aa8c81869b4c201be9fc77ce208037","verifiedAt":"2026-11-04T01:28:04.416Z","lastCommitAt":"2026-11-10T16:55:33Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Watch the films, series and music from your own Jellyfin server on an Android TV, Fire TV or Nvidia Shield.","summary":"This is the television app of the Jellyfin project: it shows the films, series, music and photos kept on your own Jellyfin server on the big screen, with a remote-control interface for Android TV, Amazon Fire TV and Nvidia Shield. It brings no content and no server of its own — someone has to install the Jellyfin server on a computer or NAS first, and then you type that address into the TV app. It is free, has no ads and no accounts beyond your own server, and it is translated by volunteers. There is no version for Windows, macOS or Linux in this repository.","easeRationale":"This is a TV app, not a computer program, so there is no grade for Windows, macOS or Linux. On a TV it is a one-tap install from Google Play, the Amazon Appstore or F-Droid, and then one simple step: typing the address of your Jellyfin server — which is why it is B and not A. Setting up that server is separate work, not part of this app.","costNote":"Free and with no paid tier. The Jellyfin project is funded by donations (Open Collective).","notVerified":["Whether a given TV box or Fire TV stick plays your files smoothly without the server converting them first","Which audio and subtitle formats work on your device","How much space the app and its cache take on the device","Whether the beta builds in GitHub behave like the stable app in the stores"]},"zh-Hans":{"summaryShort":"在Android TV、Fire TV或Nvidia Shield上观看你自己Jellyfin服务器里的电影、剧集和音乐。","summary":"这是Jellyfin项目的电视应用：它把你自己Jellyfin服务器里保存的电影、剧集、音乐和照片显示在大屏幕上，为Android TV、Amazon Fire TV和Nvidia Shield提供遥控器界面。它本身不带任何内容或服务器——得先有人在电脑或NAS上安装Jellyfin服务器，然后你再把那个地址输入到电视应用里。它是免费的，没有广告，除了你自己的服务器之外也没有账号，并由志愿者翻译。这个仓库里没有Windows、macOS或Linux版本。","easeRationale":"这是一个电视应用，不是电脑程序，所以没有Windows、macOS或Linux的评级。在电视上，只需从Google Play、Amazon Appstore或F-Droid一键安装，然后再做一件简单的事：输入你的Jellyfin服务器地址——这就是为什么它是B级而不是A级。配置那台服务器是另外的工作，不属于这个应用的一部分。","costNote":"免费，没有付费档。Jellyfin项目靠捐赠维持（Open Collective）。","notVerified":["某台具体的电视盒子或Fire TV Stick是否能顺畅播放你的文件，而不需要服务器先转码","你的设备支持哪些音频和字幕格式","应用和它的缓存在设备上占用多少空间","GitHub上的测试版构建表现是否和商店里的正式版一样"]}}},{"id":"R_kgDOTxPUmA","slug":"jenesyx__openreply-fa","name":"OpenReply","owner":"Jenesyx","fullName":"Jenesyx/openreply-fa","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"rebuild_of_another_project","severity":"info","evidence":"README 的致谢写道:\"This bilingual Persian edition was rebuilt, localized, and is maintained by Arta\";最初的 OpenReply 由 Diwen Huang 构建,基于 Anish Raj 的 instagram-comment-to-dm。GitHub 并未将这个仓库标记为 fork","source":"ai","title":"是另一个项目的重建版本"},{"code":"very_young_and_little_used","severity":"attention","evidence":"26 个 star,没有 GitHub release,仓库创建和最后一次 push 是在同一天,2026-08-07,距这次检查 90 天","source":"ai","title":"项目很新,使用者很少"},{"code":"needs_meta_app_review_and_business_account","severity":"attention","evidence":"仓库中附带了 META_APP_REVIEW.md,README 要求 \"a Meta developer app\",并说明 \"The Instagram account you connect has to be a Business or Creator account\";项目描述还补充道 \"the link below will not allow you to oauth; you must self-host with your own api keys\"","source":"ai","title":"需要 meta 应用审核和商业账号"},{"code":"server_holds_instagram_access_tokens","severity":"attention","evidence":".env.example 中包含 ENCRYPTION_KEY、INSTAGRAM_APP_SECRET、FACEBOOK_APP_SECRET 和 WEBHOOK_VERIFY_TOKEN;该应用会读取你 Instagram 的私信对话(\"Inbox\")并发送消息,因此运营这台服务器的人掌握着可以操控这个账号的凭据","source":"ai","title":"服务器保存着 instagram 访问令牌"},{"code":"docs_suggest_handing_keys_to_an_ai_assistant","severity":"attention","evidence":"README 中 \"Set it up with your AI assistant\" 一节写道:\"Paste it into your assistant inside a clone of this repo, hand over your keys as it asks, and it will walk you through connecting Instagram and going live\"","source":"ai","title":"文档建议把密钥交给 ai 助手"},{"code":"example_database_passwords","severity":"info","evidence":"docker-compose.yml 使用 POSTGRES_PASSWORD: postgres,并发布了 5432 和 6379 端口;.env.example 中附带的是占位密钥,正式使用前必须替换","source":"ai","title":"示例数据库密码"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram app (app ID and secret)","Resend (login emails)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"novinhub","name":"Novinhub","confirmed":true}],"declaredAlternatives":[],"stars":26,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":26},{"date":"2026-11-07","stars":26},{"date":"2026-11-08","stars":26},{"date":"2026-11-09","stars":26},{"date":"2026-11-10","stars":26}],"firstPublishedAt":"2026-11-05T22:43:48.511Z","commitSha":"468e73c62276f5f36d2feae547274f9264c8333b","verifiedAt":"2026-11-05T22:27:40.375Z","lastCommitAt":"2026-08-07T13:48:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Sends an Instagram DM with your link when someone comments a keyword on your post, running on your own server.","summary":"OpenReply watches the comments on your Instagram posts and, when a comment matches a keyword you chose, sends that person a private message with your link through Meta's official API; it can also post a public reply. It has tracked links with click counts, an optional \"follow first\" step, several accounts and a shared workspace. You host it yourself: a Next.js web app plus a separate worker, with PostgreSQL and Redis. You also need a Meta developer app for Instagram, a Business or Creator account and a Resend account for login emails.","easeRationale":"The steps are the same on Windows, macOS and Linux: install Node, clone the repository, fill in a .env file with six or more secrets, start Postgres and Redis with docker-compose, run the database migration, then keep two processes running (`npm run dev` and `npm run worker`). The README itself says \"the Meta app setup is the part that takes real time\".","costNote":"The code is free (MIT). It needs third-party accounts — Meta, Resend and hosting — which the README says can be on free plans, plus an always-on service for the worker","notVerified":["Whether Meta approves the permissions this needs for a given app, and how long that takes","Whether automatic DMs of this kind stay within Instagram's rules for a given account over time","Whether the code works as described, since there are no releases and almost no users yet","The real monthly cost of hosting the always-on worker, database and Redis","What data the tracked-link redirect records about the people who click"]},"zh-Hans":{"summaryShort":"当有人在你的帖子下评论某个关键词时,在你自己的服务器上发送一条带有你的链接的 Instagram 私信。","summary":"OpenReply 会监控你 Instagram 帖子下的评论,当某条评论匹配你选定的关键词时,就通过 Meta 的官方 API 给那个人发送一条带有你的链接的私信;它也可以发布一条公开回复。它带有可统计点击次数的跟踪链接、一个可选的 \"follow first\" 步骤、多个账号以及一个共享工作区。你需要自己托管:一个 Next.js 网页应用加上一个独立的 worker,配合 PostgreSQL 和 Redis。你还需要一个用于 Instagram 的 Meta 开发者应用、一个 Business 或 Creator 账号,以及一个用于登录邮件的 Resend 账号。","easeRationale":"在 Windows、macOS 和 Linux 上步骤相同:安装 Node,克隆仓库,在 .env 文件中填写六个或更多密钥,用 docker-compose 启动 Postgres 和 Redis,运行数据库迁移,然后让两个进程保持运行(`npm run dev` 和 `npm run worker`)。README 本身也说 \"the Meta app setup is the part that takes real time\"。","costNote":"代码是免费的(MIT)。它需要第三方账号——Meta、Resend 和托管服务——README 说这些可以使用免费套餐,此外 worker 还需要一个一直在线的服务","notVerified":["对于某个具体应用,Meta 是否会批准所需的权限,以及需要多长时间","这种自动私信长期来看是否符合 Instagram 对某个账号的规则","代码是否真的像描述的那样运作,因为目前还没有 release,几乎也没有用户","托管这个一直在线的 worker、数据库和 Redis 的实际月费是多少","跟踪链接的跳转会记录点击者的哪些数据"]}}},{"id":"R_kgDOVk_FEw","slug":"jenkinwoo__sheeridverify","name":"sheeridVerify","owner":"JenkinWoo","fullName":"JenkinWoo/sheeridVerify","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"fills_and_submits_the_sheerid_student_check","severity":"high","evidence":"整个项目围绕 SheerID 运作，这是许多公司在给予教育优惠前用来核实学生和教师身份的服务。README.md 中写道：配置文件保存\"获授权的国家、学校、data.organization.emailDomain\"，页面不再从用户处获取邮箱地址，而是\"按 First name + 点 + Last name @xxx.edu 格式自动拼接生成\"，多个任务可以同时运行（\"web.concurrency\"），当页面显示\"无法确认您是真人\"并出现\"重试\"按钮时，脚本会点击它，每次运行最多点击 3 次，超过后返回 \"HUMAN_VERIFICATION_RETRY_EXHAUSTED\"。同一份 README 说明该项目用于教学和研究，并写明\"不得伪造身份、学籍或材料\"和\"仅应使用自己有权使用的真实地址与资料\"。验证服务会自行制定关于自动提交的规则。","source":"ai","title":"填写并提交 sheerid 的学生身份验证"},{"code":"browser_with_fingerprint_settings_and_proxy","severity":"high","evidence":"package.json 依赖 \"camoufox\"、\"playwright-core\" 和 \"proxy-chain\"。README.md 说启动器会关闭浏览器自带的广告拦截器——\"浏览器启动时排除 Camoufox 自带的 uBlock Origin，减少验证资源被扩展拦截的可能性\"——并且\"随机输入、保持窗口和指纹配置\"适用于浏览器这条分支。一个 HTTP 或 SOCKS5 代理（带用户名和密码）会被浏览器页面请求、状态检查、证明文件下载和上传共用；留空则\"留空或只有空格时强制直连\"。","source":"ai","title":"带指纹配置和代理的浏览器"},{"code":"default_web_password_and_public_exposure_steps","severity":"attention","evidence":"README.md 写道\"本地默认访问密码为 JenkinWoo\"，部署一节还说明了如何通过 ONE_WEB_HOST=0.0.0.0 监听整个网络。README 确实要求事先修改——\"对外开放前必须通过 ONE_WEB_PASSWORD 更换密码，并配置 HTTPS 入口\"——但只要默认密码仍然有效，任何能访问到该页面的人都可以创建任务并读取任务列表。","source":"ai","title":"默认网页密码与对外暴露步骤"},{"code":"handles_identity_documents_and_personal_data","severity":"attention","evidence":"README.md 中，任务需要填写\"名、姓、出生日期\"，并上传 PNG、JPEG 或 PDF 格式的证明文件，该文件也可以从某个网址获取。配置了代理后，页面流量、文件下载和上传都会经过该代理运营方，对方能看到这条连接。","source":"ai","title":"处理身份证件和个人数据"},{"code":"brand_new_repository_many_forks_same_day","severity":"info","evidence":"项目信息：创建于 2026-11-06，最后一次推送也在 2026-11-06，没有任何发布版本，却已经有 538 个星标和 356 个复刻。代码背后没有任何历史记录，同一天出现这么多复刻意味着已经有可能被修改过的副本在流传。","source":"ai","title":"全新仓库但当天就有大量复刻"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"ISC","category":"permissive","detectedFrom":"github"},"categories":["automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":2005,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":2005,"growthDays":4,"history":[{"date":"2026-11-07","stars":1627},{"date":"2026-11-08","stars":1856},{"date":"2026-11-09","stars":1925},{"date":"2026-11-10","stars":2005}],"firstPublishedAt":"2026-11-06T23:09:33.806Z","commitSha":"bc7bf16d7a65e26c6e80b69d76526ee95e06794a","verifiedAt":"2026-11-06T23:04:10.752Z","lastCommitAt":"2026-11-06T09:56:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Fills in and submits the SheerID student check from your own computer and uploads your proof file.","summary":"sheeridVerify drives the student and teacher status check run by SheerID, the service many companies use before granting an education discount. A local web page and a command line take a verification link, a name and a birth date, open the form in the Camoufox browser, fill it in, submit it and upload a PNG, JPEG or PDF as proof; an HTTP or SOCKS5 proxy is optional. You need Node.js 22.15 or newer, the source code and a hand-edited JSON config file. The README is in Chinese and calls the project teaching and research material.","easeRationale":"On Windows, Mac and Linux alike you install Node.js 22.15 or newer, download the source, run npm ci and npm run browser:install, copy and hand-edit config/sheerid.local.json, then start a local web page at 127.0.0.1:8787. On Linux you also need a graphical desktop for the browser step (the README uses DISPLAY=:1), and the server deployment it documents adds a systemd service.","costNote":"The software is free. The README treats a proxy as optional (\"留空或只有空格时强制直连\"); if you use a proxy service, that is your own cost.","notVerified":["Whether a check actually succeeds, since the README promises no result (\"也不保证验证成功\")","What the configured proxy operator can see or keep from the verification traffic and the uploaded document","Whether the code does only what the README describes, including what the run records in .runs/ contain","Whether the local web page stays limited to your own computer in your setup","Whether your own use of it fits the verification service's rules and the law where you live"]},"zh-Hans":{"summaryShort":"在你自己的电脑上填写并提交 SheerID 的学生身份验证，并上传你的证明文件。","summary":"sheeridVerify 自动完成由 SheerID 执行的学生和教师身份核实，许多公司在提供教育优惠前都会使用这项服务。一个本地网页和一条命令行接收验证链接、姓名和出生日期，在 Camoufox 浏览器中打开表单，填写并提交，然后上传 PNG、JPEG 或 PDF 作为证明；HTTP 或 SOCKS5 代理是可选的。你需要 Node.js 22.15 或更高版本、源代码以及手动编辑的 JSON 配置文件。README 用中文写成，并将该项目称为教学与研究资料。","easeRationale":"在 Windows、Mac 和 Linux 上流程相同：安装 Node.js 22.15 或更高版本，下载源代码，运行 npm ci 和 npm run browser:install，复制并手动编辑 config/sheerid.local.json，然后在 127.0.0.1:8787 启动本地网页。在 Linux 上，浏览器这一步还需要图形桌面环境（README 使用 DISPLAY=:1），文档中记录的服务器部署还会添加一个 systemd 服务。","costNote":"该软件是免费的。README 将代理视为可选项（\"留空或只有空格时强制直连\"）；如果你使用代理服务，费用由你自己承担。","notVerified":["验证是否真的会成功，因为 README 并不承诺任何结果（\"也不保证验证成功\"）","配置的代理运营方能从验证流量和上传的文件中看到或保留什么","代码是否只做 README 所描述的事情，包括 .runs/ 中的运行记录包含什么内容","在你的环境中，本地网页是否真的只能在你自己的电脑上访问","你自己的使用方式是否符合验证服务的规则以及你所在地的法律"]}}},{"id":"R_kgDOUjqjyA","slug":"jev-chat__jev-chat-jarvis","name":"Jev Chat","owner":"jev-chat","fullName":"jev-chat/jev-chat-jarvis","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"36 天内获得 8125 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 36 天前","source":"check","title":"所有者账号很新"},{"code":"reads_other_apps_chats","severity":"attention","evidence":"描述：“before you reply, Jev reads the chat, judges intent and risk, and drafts replies”；主题包括“accessibility-service”、“ocr”、“llm”、“whatsapp”和“qq”。该应用会读取其他应用里的私人对话（包括他人发来的消息），并通过 AI 模型处理；这些文字会被发送到哪里，仓库中没有任何地方说明。","source":"ai","title":"读取其他应用的聊天内容"},{"code":"no_readme","severity":"attention","evidence":"所分析的提交在顶层只有 .github/、.gitignore、LICENSE、NOTICE、cn/ 和 global/，没有 README，因此安装步骤、数据处理方式和费用都没有文档说明。","source":"ai","title":"没有 README"},{"code":"claimed_platforms_without_builds","severity":"info","evidence":"描述中声称支持“Android · Windows · macOS · iOS”，但两个发布版本（2026-10-06 的 global-v0.1.0 和 2026-10-05 的 v1.7）都没有附带任何平台的程序文件。","source":"ai","title":"宣称支持的平台没有可用的构建版本"},{"code":"young_project_and_owner","severity":"info","evidence":"仓库创建于 2026-09-21，已有 8125 个星标和 1306 个 fork，检查记录显示所有者账号创建于 36 天前；该项目只有五周。","source":"ai","title":"项目和所有者都很新"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","productivity"],"alternativeTo":[],"declaredAlternatives":[],"stars":8321,"downloads":3723,"downloadsWeek":530,"downloadsDays":7,"growth":90,"growthDays":7,"history":[{"date":"2026-10-26","stars":8115},{"date":"2026-10-27","stars":8125},{"date":"2026-10-28","stars":8147},{"date":"2026-10-29","stars":8166},{"date":"2026-10-30","stars":8190},{"date":"2026-10-31","stars":8200},{"date":"2026-11-01","stars":8207},{"date":"2026-11-02","stars":8216},{"date":"2026-11-03","stars":8231},{"date":"2026-11-04","stars":8258},{"date":"2026-11-05","stars":8267},{"date":"2026-11-06","stars":8283},{"date":"2026-11-07","stars":8286},{"date":"2026-11-08","stars":8300},{"date":"2026-11-09","stars":8311},{"date":"2026-11-10","stars":8321}],"firstPublishedAt":"2026-10-27T01:10:12.642Z","commitSha":"bd3fe42eebd04b58b2b1d4a7f5310fddfb0bb91b","verifiedAt":"2026-10-28T01:37:06.788Z","lastCommitAt":"2026-10-16T08:48:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A phone helper that reads your chat screen, guesses what the other person means and drafts replies for you.","summary":"Jev is a chat assistant app. According to its description it reads the conversation on your screen in apps such as WhatsApp and QQ, using Android accessibility access and text recognition, judges the other person's intent and possible risk, and drafts replies you fill in with one tap; you decide whether to send. The repository has no README at the analyzed commit, so there are no install steps, privacy details or pricing to go on. The two GitHub releases carry no files for Windows, Mac, Linux or the web, and there is a separate Chinese edition (cn/) and global edition (global/).","easeRationale":"Nothing here can be installed on Windows, Mac, Linux or the browser: the two releases list no platform binaries and the repository has no README with install steps, so it is grade E. The description points at phones, which Radar does not grade.","costNote":"The code is MIT-licensed. The package does not say whether the AI features need an account, a paid plan or the website chatjevs.com.","notVerified":["Where the chat text and screen captures are sent for AI analysis, and whether they are kept","Whether any downloadable app file matches the source code in the cn/ and global/ folders","Whether the Windows, macOS and iOS versions named in the description exist anywhere","Whether the AI features need an account or a paid plan on chatjevs.com","Whether reading another person's messages this way is allowed by the chat apps' own rules in your country"]},"zh-Hans":{"summaryShort":"一个手机助手，读取你的聊天界面，猜测对方的意思并帮你起草回复。","summary":"Jev 是一款聊天助手应用。根据其描述，它会借助 Android 的无障碍权限和文字识别，读取你屏幕上诸如 WhatsApp 和 QQ 等应用里的对话，判断对方的意图和可能存在的风险，并起草回复让你一键填入；是否发送由你决定。在分析所用的提交版本中，仓库没有 README，因此没有安装步骤、隐私细节或价格可供参考。GitHub 上的两个发布版本都没有附带 Windows、Mac、Linux 或网页版的文件，仓库里还分别有一个中国版（cn/）和一个全球版（global/）。","easeRationale":"这里的东西都没法安装在 Windows、Mac、Linux 或浏览器上：两个发布版本都没有列出任何平台的二进制文件，仓库也没有包含安装步骤的 README，所以评分为 E。描述指向的是手机应用，而 Radar 不对手机应用评分。","costNote":"代码采用 MIT 许可证。软件包没有说明 AI 功能是否需要账户、付费套餐，或者是否依赖官网 chatjevs.com。","notVerified":["聊天文本和屏幕截图会被发送到哪里进行 AI 分析，以及它们是否会被保留","是否有任何可下载的应用文件与 cn/ 和 global/ 文件夹中的源代码相符","描述中提到的 Windows、macOS 和 iOS 版本是否真的存在","AI 功能是否需要在 chatjevs.com 上注册账户或购买付费套餐","以这种方式读取他人的消息，是否符合你所在国家对这些聊天应用自身规则的要求"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1NzE3NzA=","slug":"jgm__pandoc","name":"Pandoc","owner":"jgm","fullName":"jgm/pandoc","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"B","mac":"B","linux":"B","web":"A"},"risk":"low","signals":[{"code":"conversions_can_lose_formatting","severity":"info","evidence":"README.md：\"one should not expect perfect conversions between every format and every other\"；pandoc 保留结构，\"but not formatting details such as margin size\"，而且 \"some document elements, such as complex tables, may not fit into pandoc's simple document model\"。","source":"ai","title":"转换可能丢失格式"},{"code":"ready_binaries_cannot_use_some_add_ons","severity":"info","evidence":"INSTALL.md：\"the statically linked Pandoc binaries provided by us (or those available on Conda Forge) have a limitation. They are unable to utilise Lua filters that rely on Lua modules written in C.\"","source":"ai","title":"现成的二进制文件无法使用某些扩展"},{"code":"pdf_output_needs_a_separate_large_install","severity":"info","evidence":"INSTALL.md：\"By default, Pandoc creates PDFs using LaTeX\"，并提醒 \"a full MacTeX installation uses four gigabytes of disk space\"，建议改用 BasicTeX 或 TinyTeX。","source":"ai","title":"PDF 输出需要单独安装一个较大的程序"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":[],"disk":{"minMB":41,"maxMB":122,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[],"declaredAlternatives":[],"stars":46884,"downloads":51436661,"downloadsWeek":621741,"downloadsDays":7,"growth":92,"growthDays":7,"history":[{"date":"2026-10-26","stars":46703},{"date":"2026-10-27","stars":46719},{"date":"2026-10-28","stars":46736},{"date":"2026-10-29","stars":46750},{"date":"2026-10-30","stars":46761},{"date":"2026-10-31","stars":46769},{"date":"2026-11-01","stars":46778},{"date":"2026-11-02","stars":46787},{"date":"2026-11-03","stars":46792},{"date":"2026-11-04","stars":46808},{"date":"2026-11-05","stars":46826},{"date":"2026-11-06","stars":46840},{"date":"2026-11-07","stars":46848},{"date":"2026-11-08","stars":46858},{"date":"2026-11-09","stars":46868},{"date":"2026-11-10","stars":46884}],"firstPublishedAt":"2026-10-27T01:10:12.643Z","commitSha":"d24f39568be2666e035d72c4f4345f1f7926ab25","verifiedAt":"2026-10-31T01:44:47.735Z","lastCommitAt":"2026-11-10T21:24:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Converts documents between dozens of formats, such as Word, Markdown, HTML, EPUB and PDF, by typing a command.","summary":"Pandoc is a long-running, widely used converter for text documents. It turns files from one format into another, for example Word to Markdown, Markdown to Word, HTML or EPUB e-books, and slides. It is for writers, students, researchers and anyone who moves text between tools. It has no windows or buttons: you type commands in a terminal, or try the project's online demo in a browser. Installers exist for Windows, Mac and Linux. Making PDFs needs an extra program such as MiKTeX (LaTeX). Conversions keep the structure of a document but not every formatting detail.","easeRationale":"The project's online demos at pandoc.org/try and pandoc.org/app open in a browser with nothing to install, which is an A, but the README calls them limited. On Windows, Mac and Linux it is a B: a ready installer (`pandoc-3.13-windows-x86_64.msi`, `.pkg` for Mac, `.deb` for Linux) sets everything up and adds pandoc to the path, and the extra step is that you use it by typing commands, with LaTeX installed separately if you want PDF files.","costNote":"Open source under GPL version 2 or later.","notVerified":["How well complex Word layouts, tables and styles survive a conversion.","Disk space used by a LaTeX install such as MiKTeX if you need PDF output.","How complete the online demos at pandoc.org/try and pandoc.org/app are compared with the installed program, and whether files you drop in stay on your computer."]},"zh-Hans":{"summaryShort":"输入一条命令，就能在 Word、Markdown、HTML、EPUB、PDF 等几十种格式之间转换文档。","summary":"Pandoc 是一款历史悠久、使用广泛的文本文档转换工具。它能把文件从一种格式转成另一种，比如 Word 转 Markdown、Markdown 转 Word、转成 HTML 或 EPUB 电子书，还能转成幻灯片。它适合写作者、学生、研究人员，以及任何需要在不同工具之间搬运文本的人。它没有窗口也没有按钮：你在终端里输入命令，或者在浏览器里试用项目的在线演示。Windows、Mac 和 Linux 都有安装程序。生成 PDF 需要额外安装像 MiKTeX 这样的程序（LaTeX）。转换会保留文档的结构，但不是所有格式细节都能保留。","easeRationale":"项目在 pandoc.org/try 和 pandoc.org/app 的在线演示不用安装任何东西就能在浏览器打开，属于 A 级，但 README 说它们功能有限。在 Windows、Mac 和 Linux 上属于 B 级：一个现成的安装程序（`pandoc-3.13-windows-x86_64.msi`、Mac 上的 `.pkg`、Linux 上的 `.deb`）会配置好一切并把 pandoc 加入 path，额外的步骤是你要通过输入命令来使用它，如果想要 PDF 文件，还要单独安装 LaTeX。","costNote":"开源，采用 GPL 第 2 版或更新版本的许可。","notVerified":["复杂的 Word 排版、表格和样式在转换后能保留多少。","如果需要 PDF 输出，安装 MiKTeX 这类 LaTeX 软件会占用多少磁盘空间。","pandoc.org/try 和 pandoc.org/app 上的在线演示和安装后的程序相比功能完整到什么程度，以及放进去的文件是否留在你的电脑上。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk5MjQ0Mzk4MA==","slug":"jgraph__drawio-desktop","name":"draw.io Desktop","owner":"jgraph","fullName":"jgraph/drawio-desktop","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"background_auto_update","severity":"info","evidence":"README 的 Updates 部分：Windows 的 .exe、macOS 和 AppImage 版本会“new versions download in the background”，Security 部分则说它“checks github.com once a day for a newer version and downloads it from an AWS S3 bucket”。可以用 `DRAWIO_DISABLE_UPDATE=true` 或 `--disable-update` 关闭。","source":"ai","title":"后台自动更新"},{"code":"external_media_in_diagrams","severity":"info","evidence":"README 的 Security 部分：“a diagram can reference external media ... these are fetched when the diagram is opened ... Opening a diagram from an untrusted source may therefore trigger a request to the referenced URL, which can reveal metadata such as your IP address to that server; no diagram content is transmitted.”","source":"ai","title":"图表中的外部媒体"},{"code":"closed_to_contributions","severity":"info","evidence":"README：“draw.io is closed to contributions (unless a maintainer permits it, which is extremely rare).” 代码是公开的，但不接受外部人员提交的修复。","source":"ai","title":"不接受外部贡献"},{"code":"one_windows_installer_needs_admin_rights","severity":"info","evidence":"README 的 Windows 安装部分：NSIS 安装程序“Installs **per-machine** into `Program Files` and **requires administrator privileges**”，它的更新也会要求管理员权限。.msi 和 zip 则不需要。","source":"ai","title":"其中一个 Windows 安装程序需要管理员权限"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":142,"maxMB":427,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["office","productivity","design_cad"],"alternativeTo":[{"slug":"microsoft-visio","name":"Visio","confirmed":true},{"slug":"lucidchart","name":"Lucidchart","confirmed":true},{"slug":"miro","name":"Miro","confirmed":true}],"declaredAlternatives":[],"stars":63925,"downloads":150113696,"downloadsWeek":1238739,"downloadsDays":7,"growth":122,"growthDays":7,"history":[{"date":"2026-10-26","stars":63664},{"date":"2026-10-27","stars":63682},{"date":"2026-10-28","stars":63697},{"date":"2026-10-29","stars":63718},{"date":"2026-10-30","stars":63727},{"date":"2026-10-31","stars":63731},{"date":"2026-11-01","stars":63751},{"date":"2026-11-02","stars":63776},{"date":"2026-11-03","stars":63803},{"date":"2026-11-04","stars":63817},{"date":"2026-11-05","stars":63832},{"date":"2026-11-06","stars":63850},{"date":"2026-11-07","stars":63876},{"date":"2026-11-08","stars":63893},{"date":"2026-11-09","stars":63912},{"date":"2026-11-10","stars":63925}],"firstPublishedAt":"2026-10-27T01:10:12.644Z","commitSha":"e56f6f3f8689503b328456081186de0ab5594a11","verifiedAt":"2026-10-29T01:44:38.825Z","lastCommitAt":"2026-11-05T07:55:31Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"draw.io-32.6.10-windows-installer.exe","sizeBytes":144077080,"url":"https://github.com/jgraph/drawio-desktop/releases/download/v32.6.10/draw.io-32.6.10-windows-installer.exe","releaseUrl":"https://github.com/jgraph/drawio-desktop/releases/tag/v32.6.10"},{"os":"mac","file":"draw.io-universal-32.6.10.dmg","sizeBytes":259134411,"url":"https://github.com/jgraph/drawio-desktop/releases/download/v32.6.10/draw.io-universal-32.6.10.dmg","releaseUrl":"https://github.com/jgraph/drawio-desktop/releases/tag/v32.6.10"},{"os":"linux","file":"drawio-x86_64-32.6.10.AppImage","sizeBytes":152706651,"url":"https://github.com/jgraph/drawio-desktop/releases/download/v32.6.10/drawio-x86_64-32.6.10.AppImage","releaseUrl":"https://github.com/jgraph/drawio-desktop/releases/tag/v32.6.10"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free desktop app for drawing flowcharts, org charts and network diagrams, working offline.","summary":"draw.io Desktop is the official offline version of the draw.io (diagrams.net) diagram editor. You can draw flowcharts, UML, network and org charts, mind maps and floor plans, and export them to PDF, PNG or SVG. It is for anyone who needs diagrams at work or school. Windows, macOS and Linux installers are ready to download, including a per-user MSI and a portable zip for Windows that need no administrator rights. The README says diagrams stay on your computer and the app only goes online to check for updates.","easeRationale":"Every release ships ready files for all three systems: a Windows .exe installer plus a per-user .msi and a portable .zip that need no administrator rights, .dmg files for Intel and Apple Silicon Macs, and .deb, .rpm and .AppImage for Linux. You download one and open it.","costNote":"Free under GPL v3. The README says support is only through the repository and \"There is no private ticketing support for non-paying users\".","notVerified":["Whether the app really makes no network calls besides update checks, as the README states","How well it opens and saves Visio (.vsdx) files compared with Visio itself","Real disk use after installation","Whether the Linux AppImage starts on your system, since the README describes a clash with older AppImageLauncher versions"]},"zh-Hans":{"summaryShort":"一个免费的桌面应用，用于绘制流程图、组织结构图和网络图，可以离线使用。","summary":"draw.io Desktop 是 draw.io（diagrams.net）图表编辑器的官方离线版本。你可以绘制流程图、UML 图、网络图和组织结构图、思维导图以及平面图，并导出为 PDF、PNG 或 SVG。适合任何在工作或学习中需要画图的人。它为 Windows、macOS 和 Linux 都提供了现成的安装包，其中包括 Windows 下无需管理员权限的按用户安装 MSI 和便携版 zip 包。README 说图表只留在你自己的电脑上，这个应用联网只是为了检查更新。","easeRationale":"每个发行版本都为三大系统提供了现成文件：Windows 的 .exe 安装程序，加上一个按用户安装的 .msi 和一个不需要管理员权限的便携版 .zip，Intel 和 Apple Silicon Mac 的 .dmg 文件，以及 Linux 的 .deb、.rpm 和 .AppImage。下载一个打开就行。","costNote":"基于 GPL v3 许可，免费。README 说明支持只通过代码仓库提供，并且“不为未付费用户提供私人工单支持”。","notVerified":["这个应用是否真的像 README 所说那样，除了检查更新外不进行任何网络通信","打开和保存 Visio（.vsdx）文件的效果与 Visio 本身相比如何","安装后的真实磁盘占用情况","Linux 版 AppImage 在你的系统上能否正常启动，因为 README 提到它与较旧版本的 AppImageLauncher 存在冲突"]}}},{"id":"R_kgDOVFv6Gg","slug":"jhaizhou-ops__d2r-mobile","name":"D2R Mobile","owner":"jhaizhou-ops","fullName":"jhaizhou-ops/d2r-mobile","type":"needs_setup","status":"alert","grade":"E","easeByPlatform":{"mac":"E"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"21 天内获得 5026 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"needs_console_game_files_and_keys","severity":"high","evidence":"README.md：“If what you have is a whole game package (`.xci` / `.nsp`), `./setup.sh <game package> <prod.keys>` first extracts the game files into place and then carries on (install hactool first)”。获取 prod.keys 并解密 Switch 游戏包，意味着绕过主机自身的保护机制，这在不少国家属于违法行为，即使你拥有正版游戏也是如此。","source":"ai","title":"需要主机游戏文件和密钥"},{"code":"derived_from_commercial_game","severity":"attention","evidence":"README.md：这个程序重写器“Rewrites only the few instructions that cannot run unchanged under Apple's kernel”，并且“shaders are translated offline from the game's originals”；仓库中带有“shader interface descriptions, the graphics API function table, addresses inside the program and the like”（docs/DERIVED.md）。没有展示来自游戏发行商的许可，因此这个项目随时可能被下架或封禁。","source":"ai","title":"脱胎于一款商业游戏"},{"code":"very_young_repo_many_stars","severity":"info","evidence":"仓库创建于 2026-10-12，有 4765 个星标、1266 个 fork，截至 2026-10-28 已有 23 个发布版本，所以这一切都只有大约两周的历史。","source":"ai","title":"非常年轻的仓库却有大量星标"},{"code":"claims_measured_by_authors","severity":"info","evidence":"60 fps、3.8 W 以及启动时间的数据，还有 Android 和 HarmonyOS 的数字，都是这个移植版的作者在自己的设备上测出来的（“Measured by the port's authors on their phones”）。Mac 版的键盘和游戏手柄被描述为“not been operated by a person yet”，Apple TV 版则只“Reached the character-selection screen”。","source":"ai","title":"数据由作者自己测量"},{"code":"self_signed_install_expires","severity":"info","evidence":"README.md 在 iPhone 签名部分写道：“a free Apple ID works too (no iCloud save sync, and the app has to be re-signed and reinstalled every 7 days)”。","source":"ai","title":"自签名安装会过期"}],"platforms":["mac"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":5253,"downloads":5077326,"downloadsWeek":474965,"downloadsDays":7,"growth":176,"growthDays":7,"history":[{"date":"2026-10-27","stars":4763},{"date":"2026-10-28","stars":4827},{"date":"2026-10-29","stars":4873},{"date":"2026-10-30","stars":4910},{"date":"2026-10-31","stars":4931},{"date":"2026-11-01","stars":4996},{"date":"2026-11-02","stars":5023},{"date":"2026-11-03","stars":5077},{"date":"2026-11-04","stars":5127},{"date":"2026-11-05","stars":5150},{"date":"2026-11-06","stars":5180},{"date":"2026-11-07","stars":5199},{"date":"2026-11-08","stars":5222},{"date":"2026-11-09","stars":5234},{"date":"2026-11-10","stars":5253}],"firstPublishedAt":"2026-10-27T02:52:00.652Z","commitSha":"fae8c680fe98f79179cf3fab85f7a6c14ee28600","verifiedAt":"2026-11-03T02:33:08.929Z","lastCommitAt":"2026-11-04T10:19:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Tools that make a Diablo II: Resurrected Switch copy you own run on an Apple silicon Mac, iPhone or Android phone.","summary":"D2R Mobile is a compatibility layer that runs the Nintendo Switch version of Diablo II: Resurrected (with the Reign of the Warlock expansion) on iPhone, iPad, Apple silicon Mac, Android and HarmonyOS, with touch controls and extra endgame modes. It ships no game content: you supply your own game files, and the tools rewrite the game program and translate its shaders on your computer. The Mac route means installing Xcode and Python tools and building from source. You must extract the files from your own Switch copy, which needs console keys and extra tools.","easeRationale":"On an Apple silicon Mac it is grade E: you need Xcode, Homebrew and llvm, the game files from your own Switch copy, then run ./setup.sh and dm/tools/mac_app.sh to build the app yourself; there is no ready Mac download. Windows and Linux can only be used to prepare an install package for a phone, not to play, and phones are not graded by Radar.","costNote":"The project is free (MIT). You need the Switch copy of the game with the Reign of the Warlock expansion, which you buy separately; iCloud save sync also needs a paid Apple developer account.","notVerified":["Whether it is lawful where you live to extract and decrypt the files from your own Switch copy","Whether the frame rate, battery and launch figures hold on your own device","Real disk use: the game resources are about 5.6 GB plus the build workspace on your computer","Whether the tools work with any game version other than the Switch 1.0.34.0 release they target","Whether the files offered on the releases page match the source code in this repository"]},"zh-Hans":{"summaryShort":"一套工具,让你自己拥有的 Switch 版《暗黑破坏神II:狱火重生》在搭载苹果芯片的 Mac、iPhone 或安卓手机上运行。","summary":"D2R Mobile 是一个兼容层,可以把任天堂 Switch 版《暗黑破坏神II:狱火重生》(含 Reign of the Warlock 扩展包)运行在 iPhone、iPad、搭载苹果芯片的 Mac、安卓和 HarmonyOS 设备上,带有触控操作和额外的终局玩法模式。它不包含任何游戏内容:你需要自己提供游戏文件,工具会在你的电脑上重写游戏程序并转译其着色器。Mac 路线需要安装 Xcode 和 Python 工具,并从源代码构建。你必须从自己拥有的 Switch 游戏卡中提取文件,这需要主机密钥和额外的工具。","easeRationale":"在搭载苹果芯片的 Mac 上评级为 E:你需要安装 Xcode、Homebrew 和 llvm,准备好来自你自己 Switch 游戏卡的文件,然后运行 ./setup.sh 和 dm/tools/mac_app.sh 自己构建应用;没有现成的 Mac 下载包。Windows 和 Linux 只能用来为手机准备安装包,不能用来玩游戏,而且 Radar 不对手机评级。","costNote":"该项目是免费的(MIT 许可)。你需要单独购买带 Reign of the Warlock 扩展包的 Switch 版游戏;通过 iCloud 同步存档还需要付费的 Apple 开发者账号。","notVerified":["在你所在地,提取并解密自己 Switch 游戏卡里的文件是否合法","帧率、续航和启动时间等数据,在你自己的设备上是否成立","实际占用的存储空间:游戏资源约 5.6 GB,再加上电脑上的构建工作区","这些工具是否支持 Switch 1.0.34.0 以外的其他游戏版本","发布页面提供的文件是否与本仓库中的源代码一致"]}}},{"id":"R_kgDOVm1wdg","slug":"jhaizhou-ops__s2arm","name":"S2Arm","owner":"jhaizhou-ops","fullName":"jhaizhou-ops/S2Arm","type":"needs_setup","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B"},"risk":"high","signals":[{"code":"needs_console_keys_and_dumped_games","severity":"high","evidence":"要使用这个应用，你必须自己提供主机密钥和游戏文件。仓库描述写着 \"Bring your own games and keys\"，入门指南第 2 步写着 \"Import your own keys and games\"，法律声明部分写着 \"The project ships no games and no keys; obtain your own game files legally.\" 所以密钥和游戏文件都必须来自项目之外，而许多国家的法律限制绕过设备或游戏的复制保护。","source":"ai","title":"需要主机密钥和自己转储的游戏"},{"code":"built_on_switch_emulator_code","severity":"attention","evidence":"README 致谢部分：\"Ryujinx (Ryujinx Team and Contributors, MIT): upstream for system services and shader translation; S2Arm's compatibility layer is built on it.\" Ryujinx 曾是一款任天堂 Switch 模拟器，在任天堂联系其作者后停止了开发，README 还声明 S2Arm \"is not affiliated with or endorsed by Nintendo\"。","source":"ai","title":"构建在 Switch 模拟器代码之上"},{"code":"ai_assistant_told_to_run_commands_on_your_computer","severity":"attention","evidence":"README 中的 \"Have an AI do it for you\" 部分让你把仓库里某个文件的链接发给你的 AI 助手，并说 \"it needs to be able to run commands on your computer\"，接着列出了诸如 \"/Applications/S2Arm.app/Contents/Resources/s2armctl start\" 之类的命令。助手要遵循的指令存放在这个仓库里，随时可能更改。","source":"ai","title":"要求 AI 助手在你的电脑上执行命令"},{"code":"very_new_project_with_many_open_issues","severity":"attention","evidence":"该仓库创建于 2026-11-08，也就是本次检查的前一天，但已经有 8 个发布版本（最新为 v0.1.7，发布于 2026-11-09）和 154 个未解决的 issue，共 1716 个星标和 341 个复刻。所有版本号都低于 1.0。","source":"ai","title":"项目非常新但未解决的 issue 很多"},{"code":"saves_do_not_carry_over_from_earlier_builds","severity":"info","evidence":"README 入门指南：\"Back up your saves before updating the app. The release uses the new app identifier `io.s2arm.*`, which the system treats as a new app: saves from the beta builds do not carry over automatically.\"","source":"ai","title":"存档不会从旧版本自动带过来"}],"platforms":["mac","windows"],"tools":[],"buildTools":[],"disk":{"minMB":388,"maxMB":1163,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":2358,"downloads":20713,"downloadsWeek":0,"downloadsDays":0,"growth":2358,"growthDays":2,"history":[{"date":"2026-11-10","stars":2358}],"firstPublishedAt":"2026-11-09T23:08:44.771Z","commitSha":"ce99720410fa6a5f8af27ae33f724a62becf8ce1","verifiedAt":"2026-11-09T23:03:54.950Z","lastCommitAt":"2026-11-10T22:22:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your own Nintendo Switch games into an app that runs natively on iPhone, iPad or an Apple silicon Mac.","summary":"S2Arm has a desktop launcher, for Mac or Windows, that takes Switch game files you supply and rewrites the game code to ARM64 so it runs on your iPhone, iPad or Apple silicon Mac without an emulator in between. All your games sit in one app. The launcher is graphical and claims no command line, but you must bring your own game files and your own console keys, sign in with your Apple ID to install on the phone, and on a Mac install the free Xcode for iPhone builds. Games need 12 GB or more each. It ships no games and no keys.","easeRationale":"Both launchers come as ready downloads (a 241 MB Mac disk image, a 388 MB Windows zip) and the README says the welcome page is three steps with no command line, but you still have to add your own console keys and game files and sign in with your Apple ID; on a Mac, building for iPhone also means installing the free Xcode from the App Store. Windows only builds and installs for iPhone and iPad, it cannot play games.","costNote":"The app and the launcher are free (Apache-2.0). Installing on an iPhone or iPad uses your own Apple ID; the README says the whole library takes one app slot on your developer account.","notVerified":["Whether the performance figures in the README hold up: \"211 games surveyed with a 95% first-try build success rate\", 60 fps on an iPhone 17 Pro Max, and the 6 to 34 minute build times.","Which games actually work, beyond the statement that 32-bit games and games needing a 36-bit address space are out.","How much disk space you need in total, on the computer and on the phone, once a 12 to 23 GB game is imported and built.","Whether the Mac disk image and the Windows zip on the releases page match the source in this repository.","Whether the app installed with a free Apple ID keeps working, and for how long, before it has to be signed again.","What the rules are in the user's country for extracting keys from a console and converting a game you own."]},"zh-Hans":{"summaryShort":"把你自己的任天堂 Switch 游戏转换成一个应用，原生运行在 iPhone、iPad 或搭载 Apple 芯片的 Mac 上。","summary":"S2Arm 提供一个桌面启动器，支持 Mac 或 Windows，它会接收你自己提供的 Switch 游戏文件，把游戏代码重写为 ARM64，这样就能不借助模拟器直接运行在你的 iPhone、iPad 或搭载 Apple 芯片的 Mac 上。所有游戏都集中在一个应用里。这个启动器是图形界面的，号称无需命令行，但你必须自带游戏文件和自己的主机密钥，用 Apple ID 登录才能安装到手机上，在 Mac 上编译 iPhone 版本还需要安装免费的 Xcode。每个游戏需要 12 GB 或更多空间。它不附带任何游戏或密钥。","easeRationale":"两个启动器都是可以直接下载使用的（一个 241 MB 的 Mac 磁盘镜像，一个 388 MB 的 Windows 压缩包），README 说欢迎页面只有三步、无需命令行，但你仍然需要添加自己的主机密钥和游戏文件，并用 Apple ID 登录；在 Mac 上，为 iPhone 编译还意味着要从 App Store 安装免费的 Xcode。Windows 版只能为 iPhone 和 iPad 编译并安装，无法在电脑上玩游戏。","costNote":"该应用和启动器都是免费的（Apache-2.0）。安装到 iPhone 或 iPad 需要使用你自己的 Apple ID；README 说整个游戏库在你的开发者账号里只占用一个应用名额。","notVerified":["README 中的性能数据是否站得住脚：\"211 games surveyed with a 95% first-try build success rate\"、在 iPhone 17 Pro Max 上达到 60 fps，以及 6 到 34 分钟的编译时间。","除了声明 32 位游戏和需要 36 位地址空间的游戏不支持之外，究竟哪些游戏真的能用。","导入并编译一个 12 到 23 GB 的游戏后，在电脑和手机上总共需要多少磁盘空间。","发布页面上的 Mac 磁盘镜像和 Windows 压缩包是否与这个仓库里的源代码一致。","用免费 Apple ID 安装的应用能持续使用多久，之后是否需要重新签名。","在用户所在国家，提取主机密钥并转换自己拥有的游戏，适用什么样的法律规定。"]}}},{"id":"R_kgDOUBHRMQ","slug":"jiafeimiao-k-cat__discimagestudio","name":"DiscImageStudio","owner":"JiaFeiMiao-K-Cat","fullName":"JiaFeiMiao-K-Cat/DiscImageStudio","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"burning_can_ruin_the_disc","severity":"attention","evidence":"README.md：直接烧录功能是实验性的，\"请只使用可报废的测试介质\"，以及\"刻录中断、断电或生成速度不足仍可能使盘片报废\"——中断、断电或电脑太慢都可能毁掉光盘。对于BD-R，\"中断后同一盘片通常不能继续写入\"：中断的烧录无法恢复。","source":"ai","title":"烧录可能毁掉光盘"},{"code":"interface_windows_only","severity":"info","evidence":"Windows版本是带图形界面的.exe，但Mac和Linux的下载包叫DiscImageStudio-CLI-...tar.gz，README把它们称为\"实验性 CLI\"；另外说明CLI只会在本地生成镜像，真正的烧录要交给其他软件。","source":"ai","title":"界面仅支持windows"},{"code":"result_depends_on_hardware","severity":"info","evidence":"README.md：\"实际可见效果取决于盘片、刻录机、固件与写入策略\"——图案是否真的能看出来，取决于光盘品牌、刻录机、它的固件和写入策略，应用只为少数品牌内置了预设。Windows ARM64版本从未在真实光盘上验证过（\"尚未验证实盘刻录\"）。","source":"ai","title":"效果取决于硬件"},{"code":"copyleft_license_with_ported_code","severity":"info","evidence":"GPL-3.0-or-later许可，README说明CD交织代码是从CDImage项目移植并修改而来，因此任何重新分发编译版本或修改版本的人，也必须一并提供源代码。","source":"ai","title":"copyleft许可证，包含移植的代码"}],"platforms":["windows","linux","mac"],"tools":[],"buildTools":[],"disk":{"minMB":168,"maxMB":505,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image","hardware"],"alternativeTo":[{"slug":"nero-burning-rom","name":"Burning ROM","confirmed":true}],"declaredAlternatives":[],"stars":9214,"downloads":25389,"downloadsWeek":1923,"downloadsDays":1,"growth":3871,"growthDays":7,"history":[{"date":"2026-11-09","stars":8890},{"date":"2026-11-10","stars":9214}],"firstPublishedAt":"2026-11-08T23:01:18.743Z","commitSha":"df188fa7977703a028f153597216abca6c58988e","verifiedAt":"2026-11-08T22:44:32.954Z","lastCommitAt":"2026-11-10T07:15:41Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"DiscImageStudio-v1.5.0-win-x64.exe","sizeBytes":176663162,"url":"https://github.com/JiaFeiMiao-K-Cat/DiscImageStudio/releases/download/v1.5.0/DiscImageStudio-v1.5.0-win-x64.exe","releaseUrl":"https://github.com/JiaFeiMiao-K-Cat/DiscImageStudio/releases/tag/v1.5.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Etches a picture you can see with your own eyes onto a blank CD, DVD or Blu-ray, using an ordinary disc burner.","summary":"Disc Image Studio turns a picture into disc data arranged so that, once burned, the pattern is visible on the surface of the disc. On Windows it is a ready program with a design page (layers, dithering, gamma, brightness), a preview of how the disc will look, and export to ISO, RAW or WAV+CUE for burning software; it can also burn straight to a blank disc, which the project calls experimental. You can add normal files to the disc too. It needs a blank CD, DVD or single-layer BD-R and a burner. Mac and Linux only get an experimental command-line version.","easeRationale":"On Windows you download DiscImageStudio-v1.5.0-win-x64.exe and run it: the .NET runtime is already inside, and the graphical design and export pages are there. On Mac and Linux there is no interface at all, only an experimental command-line package you unpack and drive with commands like `./disc-image-studio cd-generate --input picture.png`, and it cannot burn the disc, only make the image file.","notVerified":["Whether the picture really comes out visible on your own disc brand and burner","Whether the claim of no network use, no uploads and no telemetry holds in practice","How often the experimental direct burning fails and wastes a disc","Whether the calibration workflow reaches the accuracy the README's photos suggest","Whether the Mac and Linux command-line builds produce images other burning software accepts"]},"zh-Hans":{"summaryShort":"用普通刻录机，把一张用肉眼就能看到的图案刻在空白CD、DVD或蓝光光盘上。","summary":"Disc Image Studio把一张图片转换成光盘数据，经过排列后，烧录完成后图案就能在光盘表面看到。在Windows上它是一个现成的程序，有设计页面（图层、抖动、伽马、亮度）、光盘效果预览，并能导出为ISO、RAW或WAV+CUE供烧录软件使用；它也可以直接烧录到空白光盘，项目把这个功能称为实验性。你还可以往光盘里添加普通文件。它需要一张空白的CD、DVD或单层BD-R，以及一台刻录机。Mac和Linux只有实验性的命令行版本。","easeRationale":"在Windows上你下载DiscImageStudio-v1.5.0-win-x64.exe并运行：.NET运行时已经内置，图形化的设计和导出页面都在。Mac和Linux上完全没有界面，只有一个实验性的命令行包，你解压后用类似`./disc-image-studio cd-generate --input picture.png`这样的命令来操作，而且它不能烧录光盘，只能生成镜像文件。","notVerified":["图案在你自己的光盘品牌和刻录机上是否真的能看出来","宣称的不联网、不上传、无遥测，在实际使用中是否真的做到了","实验性的直接烧录功能多久会失败一次并浪费一张光盘","校准流程能否达到README照片所暗示的精度","Mac和Linux的命令行版本生成的镜像，其他烧录软件是否能接受"]}}},{"id":"R_kgDOKbB_hA","slug":"jianchang512__pyvideotrans","name":"pyVideoTrans","owner":"jianchang512","fullName":"jianchang512/pyvideotrans","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接通过管道传给 shell 执行:curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"curl 管道直接执行到 shell"},{"code":"voice_cloning_impersonation_risk","severity":"attention","evidence":"README 的功能列表中写道:\"Voice Cloning: Integrates models like F5-TTS, CosyVoice, GPT-SoVITS for zero-shot voice cloning\",这可以从一段样本中复制真人的声音","source":"ai","title":"声音克隆存在冒充风险"},{"code":"dependency_conflicts_overridden","severity":"attention","evidence":"pyproject.toml 固定了大约 250 个包的版本,然后添加了一段带有注释 \"# 忽略版本冲突错误\" 的 override-dependencies 配置,还加上了 `index-strategy = \"unsafe-best-match\"`,这让安装器可以从任意配置的索引中选择一个包","source":"ai","title":"依赖冲突被强制忽略"},{"code":"docker_build_downloads_third_party_binary","severity":"info","evidence":"Dockerfile 在构建过程中会从一个个人网站下载一份 FFmpeg 编译版本:wget -q https://johnvansickle.com/ffmpeg/releases/ffmpeg-release-amd64-static.tar.xz","source":"ai","title":"docker 构建时下载第三方二进制文件"},{"code":"author_disclaimer_about_copyrighted_video","severity":"info","evidence":"README 的 Disclaimer 写道:\"Users are solely responsible for any legal consequences arising from the use of this software (including but not limited to calling third-party APIs or processing copyrighted video content)\"","source":"ai","title":"作者就版权视频发出的免责声明"},{"code":"readme_promotes_paid_sponsors","severity":"info","evidence":"README 开头有一个 \"Sponsors\" 表格,推广 Recall.ai 和 OfoxAI 这两个付费 API 服务,并带有 utm 跟踪链接","source":"ai","title":"readme 推广付费赞助商"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":70,"maxMB":210,"basis":"installer"},"apiKeys":["OpenAI","Azure Speech","ElevenLabs","DeepL","Google Gemini","DeepSeek"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["video","audio","local_ai"],"alternativeTo":[{"slug":"rask-ai","name":"Rask AI","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true}],"declaredAlternatives":[],"stars":19697,"downloads":220330,"downloadsWeek":0,"downloadsDays":4,"growth":62,"growthDays":6,"history":[{"date":"2026-11-06","stars":19653},{"date":"2026-11-07","stars":19668},{"date":"2026-11-08","stars":19672},{"date":"2026-11-09","stars":19688},{"date":"2026-11-10","stars":19697}],"firstPublishedAt":"2026-11-05T22:43:48.520Z","commitSha":"7921c819f9f09ac64a611d57cd676df28fd0b56a","verifiedAt":"2026-11-05T22:28:40.045Z","lastCommitAt":"2026-11-10T12:27:33Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Translates a video into another language: writes the subtitles from the speech, translates them and dubs it with an AI voice.","summary":"pyVideoTrans takes a video or audio file, writes down what is said, translates it, reads the translation with a synthetic voice and puts the new audio and subtitles back on the video. It can tell speakers apart, give each one a different voice, clone a voice from a sample and let you correct the text at each step. It can work offline with local models or connect to online services such as OpenAI, Azure, DeepL, Gemini and ElevenLabs. On Windows 10 or 11 there is a ready package to unzip and run; on macOS and Linux the README describes installing from source.","easeRationale":"Windows 10 and 11 get the best grade: the README's quick start is to download the pre-packaged version from the releases page, unzip it to a folder without spaces and double-click sp.exe, then pick a recognition model or enter a key for an online service inside the app. For macOS and Linux the README only describes the source route — Python 3.10, FFmpeg installed by hand with brew or apt, the uv tool, `uv sync` and `uv run sp.py` — or a Docker image you build yourself.","costNote":"The program is free (GPL-3.0) and can run with free pieces (Edge-TTS, local Whisper); the optional online services it connects to, such as OpenAI, Azure or ElevenLabs, are paid","notVerified":["Which files the Windows package contains and whether local models come with it or download on first use","Whether the Mac files on the releases page are a ready app, since the README documents only the source route for macOS","Real disk use and video memory after downloading recognition, translation and voice models","The quality of the automatic dubbing and lip timing for a given language pair","How much the optional online services cost for a long video"]},"zh-Hans":{"summaryShort":"把视频翻译成另一种语言:从语音生成字幕,翻译字幕,再用 AI 配音。","summary":"pyVideoTrans 接收一个视频或音频文件,把其中说的话写下来,翻译它,用合成语音朗读译文,再把新的音频和字幕放回视频里。它可以区分不同的说话人,给每个人配上不同的声音,从一个样本克隆声音,并且在每一步都让你修改文本。它可以离线使用本地模型,也可以连接 OpenAI、Azure、DeepL、Gemini、ElevenLabs 等在线服务。在 Windows 10 或 11 上有现成的压缩包,解压即可运行;在 macOS 和 Linux 上,README 介绍的是从源代码安装。","easeRationale":"Windows 10 和 11 得到最高评级:README 的快速上手方式是从 release 页面下载打包好的版本,解压到一个路径中不含空格的文件夹,双击 sp.exe,然后在应用内选择一个识别模型,或输入某个在线服务的密钥。对于 macOS 和 Linux,README 只描述了从源代码安装的方式——Python 3.10、用 brew 或 apt 手动安装的 FFmpeg、uv 工具、`uv sync` 和 `uv run sp.py`——或者自己构建的 Docker 镜像。","costNote":"这个程序是免费的(GPL-3.0),可以配合免费组件使用(Edge-TTS、本地 Whisper);它可连接的可选在线服务,比如 OpenAI、Azure 或 ElevenLabs,是收费的","notVerified":["Windows 安装包里包含哪些文件,本地模型是随包提供还是首次使用时下载","release 页面上的 Mac 文件是否是现成可用的应用,因为 README 对 macOS 只记录了源代码安装方式","下载识别、翻译和语音模型之后,实际占用的磁盘空间和显存","对于特定的语言对,自动配音和口型同步的质量如何","处理一段长视频时,可选在线服务的费用是多少"]}}},{"id":"R_kgDORKrMKQ","slug":"jihe520__social-push","name":"Social Push","owner":"jihe520","fullName":"jihe520/social-push","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单均未检测到许可证","source":"check","title":"没有许可证"},{"code":"turns_on_remote_control_of_your_browser","severity":"attention","evidence":"安装步骤 3 要求你打开 `chrome://inspect/#remote-debugging` 并启用\"Allow remote debugging for this browser instance\",默认设置是 `--auto-connect`,也就是\"自动链接用户使用的浏览器\"。只要这个功能开着,你电脑上的任何程序都能操控这个你已经登录了各种账号的浏览器。","source":"ai","title":"开启了对你浏览器的远程控制"},{"code":"saves_your_logged_in_session_to_a_file","severity":"attention","evidence":"指南的最后一步是 `agent-browser state save ~/my-state.json`,还提供了 `--profile ~/my-profile`,所以你的社交媒体会话会以明文文件保存在磁盘上;任何复制了这些文件的人都能以你的身份登录。","source":"ai","title":"把你已登录的会话保存到文件里"},{"code":"automated_posting_can_get_accounts_blocked","severity":"attention","evidence":"功能列表承诺可以在小红书、X、知乎、微博、微信公众号、掘金和 Linux.do 上\"一句话发布内容\",做法是操控这些网页。这些平台的条款都禁止自动化使用,账号可能因此被限制或封禁。","source":"ai","title":"自动发帖可能导致账号被限制"},{"code":"skill_rewrites_its_own_instructions","severity":"info","evidence":"功能介绍:\"Self-Evolution(自我进化)- 网页改版后可自动检测并修复 workflow,无需手动维护代码\"——当某个网站改版时,AI 会去修改 skill 自己的工作流文件,所以它下一次的行为可能和你审查过的不一样。","source":"ai","title":"skill 会改写自己的指令"},{"code":"confusing_credit_for_the_browser_tool","severity":"info","evidence":"文档第一行把功劳归给 agent-browser,并链接到\"github.com/anthropics/agent-browser\",但安装命令却是 `npx skills add https://github.com/vercel-labs/agent-browser --skill agent-browser`。它实际安装的工具,来源和链接暗示的并不是同一个所有者。","source":"ai","title":"浏览器工具的来源标注令人困惑"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Claude Code (paid Claude plan or Anthropic API credits)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["ai_skills","social_marketing"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true}],"declaredAlternatives":[],"stars":550,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":4,"growthDays":6,"history":[{"date":"2026-11-05","stars":547},{"date":"2026-11-06","stars":547},{"date":"2026-11-07","stars":547},{"date":"2026-11-08","stars":547},{"date":"2026-11-09","stars":549},{"date":"2026-11-10","stars":550}],"firstPublishedAt":"2026-11-04T22:32:55.014Z","commitSha":"f7cd0d0e5c634a9d9a7cc28ce5e35c8cc30440d1","verifiedAt":"2026-11-04T22:22:22.475Z","lastCommitAt":"2026-06-28T09:08:50Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Claude Code skill that drives your browser to write posts for you on Xiaohongshu, X, Zhihu, Weibo and more.","summary":"A skill for Claude Code. You type \"/social-push post this article to Xiaohongshu\" and the assistant opens your browser and fills the post in for you, using the agent-browser tool instead of fixed scripts. It covers Xiaohongshu, X, Zhihu, Weibo, WeChat official accounts, Juejin and Linux.do, and each workflow is a Markdown file you can copy. You need Claude Code, Node.js and agent-browser with Chromium, and you log in to each site by hand once. Documentation in Chinese and English; the repository has no licence.","easeRationale":"Same terminal work everywhere and several pieces to line up: Claude Code, `npm install -g agent-browser`, `npx skills add jihe520/social-push`, `agent-browser install` to download Chromium, turning on \"Allow remote debugging for this browser instance\" in Chrome, and a manual first login on each platform. The hand-login instructions are written for Mac (`open \"path\" --args --remote-debugging-port=9222`), so Windows and Linux users have to adapt them.","costNote":"The skill files are free, but it only works inside Claude Code, which needs a paid Claude plan or API credits.","notVerified":["Whether it really only saves drafts and waits for your confirmation, as the README claims","Whether each platform's workflow still matches the current pages, since the last change is from June 2026","How many Claude tokens, and how much money, a single post costs","Whether you may use the files at all, since the repository has no licence","What the \"self-evolution\" step writes into the skill files"]},"zh-Hans":{"summaryShort":"一个 Claude Code 的 skill,能操作你的浏览器,帮你在小红书、X、知乎、微博等平台发帖。","summary":"这是一个 Claude Code 的 skill。你输入\"/social-push post this article to Xiaohongshu\",助手就会打开你的浏览器并帮你填好帖子内容,用的是 agent-browser 工具,而不是固定脚本。它覆盖小红书、X、知乎、微博、微信公众号、掘金和 Linux.do,每个平台的工作流都是一个你可以复制的 Markdown 文件。你需要 Claude Code、Node.js 和带有 Chromium 的 agent-browser,并且要在每个网站手动登录一次。文档有中文和英文版本;仓库没有许可证。","easeRationale":"各处都是同样的终端操作,而且要凑齐好几样东西:Claude Code、`npm install -g agent-browser`、`npx skills add jihe520/social-push`、用 `agent-browser install` 下载 Chromium、在 Chrome 里打开\"Allow remote debugging for this browser instance\",以及在每个平台先手动登录一次。手动登录的说明是按 Mac 写的(`open \"path\" --args --remote-debugging-port=9222`),所以 Windows 和 Linux 用户得自己改。","costNote":"这个 skill 的文件是免费的,但它只能在 Claude Code 里运行,而 Claude Code 需要付费套餐或 API 额度。","notVerified":["它是否真的只保存草稿并等你确认,就像 README 宣称的那样","各平台的工作流是否还和现在的页面匹配,因为最近一次改动是在 2026 年 6 月","发一条帖子要花多少 Claude token,折算成钱是多少","你到底能不能使用这些文件,因为仓库没有许可证","\"自我进化\"这一步会往 skill 文件里写入什么内容"]}}},{"id":"R_kgDOTaUzTw","slug":"jingyi-wu-richael__rachel-digital-human-production","name":"Rachel Digital Human Production","owner":"Jingyi-Wu-Richael","fullName":"Jingyi-Wu-Richael/rachel-digital-human-production","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"cloning_a_voice_and_face_needs_that_persons_consent","severity":"attention","evidence":"该工作流程使用 \"inputs/portrait.jpg\" 和 \"inputs/voice-source.mp3\"，生成一段由那个人说出你所写脚本的视频。README 自己的风险提示章节写道 \"Use this only for authorized voice, portrait, script, and publishing workflows\"，并说明安装它 \"does not grant access to ... legal clearance\"。未经本人同意使用其面孔或声音，会在许多国家涉及肖像权和声音权问题，也会违反这两项服务各自的规则。","source":"ai","title":"克隆某人的声音和面孔需要本人同意"},{"code":"two_paid_accounts_required_before_anything_happens","severity":"attention","evidence":"README 的 Environment Variables 部分写着 \"export MINIMAX_API_KEY\" 和 \"export HEYGEN_API_KEY\"，并说明用户需要自备 \"their own accounts, billing, permissions, and API keys\"。没有任何环节在本地运行：每一次语音克隆和每一段视频都是对 MiniMax 和 HeyGen 的付费调用，该技能也明确要求先生成一个 15 秒的预览，以限制这笔花费。","source":"ai","title":"开始之前需要两个付费账号"},{"code":"the_assistant_holds_your_keys_while_it_works","severity":"info","evidence":"这些密钥是在 Codex 执行步骤所在的 shell 中导出的，该技能声明的目标是 \"Keep API keys, Authorization headers, signed URLs, private portraits, voice samples, and generated client videos out of the skill package\"。但助手在执行操作说明的过程中仍然会读取这些值以及你的输入文件。","source":"ai","title":"助手在工作期间持有你的密钥"},{"code":"most_of_the_work_is_instructions_not_code","severity":"info","evidence":"该仓库只包含 .gitignore、LICENSE、README.md、SKILL.md 以及 agents/、references/ 和 scripts/ 这几个文件夹，其中唯一列出的脚本是 init_job_state.py（创建状态文件）和 preflight_assets.py（检查素材）。对 MiniMax 和 HeyGen 的实际调用都交给助手去完成，由它遵照书面操作说明执行，因此实际运行的内容取决于它如何理解这些指令。","source":"ai","title":"大部分工作是说明文字而不是代码"},{"code":"one_day_of_work_and_no_updates_since","severity":"info","evidence":"该仓库创建于 2026-07-16，并在当天就完成了推送，此后 113 天里没有任何提交，没有发布版本，没有描述，也没有主题标签，却已经积累了 1083 个星标和 159 个复刻。付费 API 会不断变化；而这里的内容自第一天起就再未调整过。","source":"ai","title":"只工作了一天，此后再无更新"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["MiniMax (voice cloning and narration)","HeyGen (image-to-video)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","video","audio"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":1085,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":5,"growthDays":6,"history":[{"date":"2026-11-07","stars":1083},{"date":"2026-11-08","stars":1084},{"date":"2026-11-09","stars":1085},{"date":"2026-11-10","stars":1085}],"firstPublishedAt":"2026-11-06T22:57:16.686Z","commitSha":"a20258b8eb38eb370cf3402fb4414754b4c44c2a","verifiedAt":"2026-11-06T22:45:37.740Z","lastCommitAt":"2026-07-16T10:57:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Codex skill that walks an AI assistant through making a talking-head video from one photo and a voice sample.","summary":"This is a skill for the Codex assistant, not a program with a screen. You copy the folder into Codex's skills directory, set two API keys, and then ask Codex to produce a talking-head video: it checks that your script, portrait photo and voice sample are usable, clones the voice with MiniMax, builds the video from the photo with HeyGen, makes a 15-second preview and waits for your approval before the full 1080p version, keeping track of each job in a file. Both services are paid and the accounts, credits and permissions are yours to arrange.","easeRationale":"Grade C on all three systems: you must already use Codex, copy the folder into its skills directory (\"cp -R rachel-digital-human-production ~/.codex/skills/\"), export MINIMAX_API_KEY and HEYGEN_API_KEY, and have Python for the two helper scripts (init_job_state.py, preflight_assets.py). The copy command as written is a macOS or Linux shell line; on Windows you copy the folder yourself.","costNote":"The skill itself is MIT and free. README: \"The skill expects the user to provide their own accounts, billing, permissions, and API keys\" for MiniMax and HeyGen, and voice cloning plus video generation are billed by those services. The repository states no prices.","notVerified":["Whether the described steps still match the current MiniMax and HeyGen APIs, as nothing has changed since July 2026","What each video costs, since the repository names no prices and both services bill per use","What MiniMax and HeyGen keep of the portrait, the voice sample and the finished video","Whether the assistant really stops for your approval before the full 1080p render, which depends on it following the playbook","The quality of the resulting video and cloned voice"]},"zh-Hans":{"summaryShort":"一个 Codex 技能，引导 AI 助手根据一张照片和一段语音样本制作出人物说话的视频。","summary":"这是一个供 Codex 助手使用的技能，不是一个有界面的程序。你把这个文件夹复制到 Codex 的技能目录中，设置两个 API 密钥，然后让 Codex 制作一段人物说话的视频：它会检查你的脚本、人像照片和语音样本是否可用，用 MiniMax 克隆语音，用 HeyGen 根据照片生成视频，先制作一个 15 秒的预览并等待你批准，之后才生成完整的 1080p 版本，并在一个文件中记录每个任务的进度。这两项服务都需要付费，账号、额度和权限都需要你自己准备。","easeRationale":"三个系统上均为 C 级：你必须已经在使用 Codex，把文件夹复制到它的技能目录中（\"cp -R rachel-digital-human-production ~/.codex/skills/\"），导出 MINIMAX_API_KEY 和 HEYGEN_API_KEY，并安装 Python 以运行两个辅助脚本（init_job_state.py、preflight_assets.py）。文中给出的复制命令是 macOS 或 Linux 的 shell 命令；在 Windows 上需要自己手动复制该文件夹。","costNote":"这个技能本身采用 MIT 许可，免费使用。README 写道：\"The skill expects the user to provide their own accounts, billing, permissions, and API keys\"，这里指的是 MiniMax 和 HeyGen，语音克隆和视频生成都由这两项服务收费。仓库中没有给出具体价格。","notVerified":["所描述的步骤是否仍然符合 MiniMax 和 HeyGen 目前的 API，因为自 2026 年 7 月以来没有任何更新","每段视频的费用是多少，因为仓库没有给出价格，而这两项服务都是按使用量收费","MiniMax 和 HeyGen 会保留人像照片、语音样本和成片中的哪些内容","助手是否真的会在完整的 1080p 渲染之前停下来等待你批准，这取决于它是否遵循了操作说明","最终视频和克隆语音的质量如何"]}}},{"id":"R_kgDOVdrchw","slug":"joeblau__subvox","name":"Subvox","owner":"joeblau","fullName":"joeblau/subvox","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"mac":"E"},"risk":"attention","signals":[{"code":"unproven_prototype","severity":"attention","evidence":"README 开头一句:这是'A macOS prototype that tests whether AirPods Pro head-motion data... carries enough information'。'Known limits'部分还提到耳机动作数据'roughly 25 Hz on AirPods',而且'Templates are user- and session-specific'。","source":"ai","title":"尚未验证的原型"},{"code":"records_microphone_and_motion_to_disk","severity":"info","evidence":"README 'Experiment protocol':录音保存在'~/Documents/Subvox/Sessions/<timestamp>-<condition>/',包含'audio.caf: microphone audio, if the mic is running'以及每一份动作采样;训练好的词保存在'~/Library/Application Support/Subvox/dataset.json'。","source":"ai","title":"将麦克风和动作数据写入磁盘"},{"code":"reads_words_not_spoken_aloud","severity":"info","evidence":"项目声明的目标是通过耳机动作把'silently mouthed words into text'。README 中说每个人必须先录制自己的模板('Templates are user- and session-specific'),所以它不是针对别人的。","source":"ai","title":"读取没有说出声的词"},{"code":"specific_hardware_required","severity":"info","evidence":"README 'Build & run':'Requires macOS 14+, AirPods Pro or another headphone that supports head tracking, connected as the Mac's audio output.'","source":"ai","title":"需要特定硬件"},{"code":"very_new_repository","severity":"info","evidence":"GitHub 数据:创建于2026-10-31,唯一的一次commit就在当天,0次发布,却已经有337颗star。","source":"ai","title":"非常新的仓库"}],"platforms":["mac"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","audio"],"alternativeTo":[],"declaredAlternatives":[],"stars":816,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":105,"growthDays":7,"history":[{"date":"2026-11-01","stars":577},{"date":"2026-11-02","stars":671},{"date":"2026-11-03","stars":711},{"date":"2026-11-04","stars":727},{"date":"2026-11-05","stars":743},{"date":"2026-11-06","stars":772},{"date":"2026-11-07","stars":784},{"date":"2026-11-08","stars":799},{"date":"2026-11-09","stars":810},{"date":"2026-11-10","stars":816}],"firstPublishedAt":"2026-11-01T01:55:44.564Z","commitSha":"591d02838521c3d33107bc9fde1a6fd620f96d58","verifiedAt":"2026-11-01T01:40:32.288Z","lastCommitAt":"2026-10-31T01:49:43Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Mac experiment that tries to turn words you only mouth into text, using the head-motion sensors in AirPods Pro.","summary":"Subvox is a research prototype for macOS that tests whether the head-motion data from AirPods Pro carries enough information to turn silently mouthed words into text. You train it by holding Space and mouthing a word several times, and it then matches new movements against those recordings and shows whether the result beats chance. It is for curious technical people: there is no download, so you install xcodegen with Homebrew and build the app in Xcode. It needs macOS 14 or newer and headphones with head tracking.","easeRationale":"Mac is an E: there are no releases to download, and the README's only instructions are 'brew install xcodegen', 'xcodegen generate' and building the project in Xcode. You also need AirPods Pro or other head-tracking headphones set as the Mac's audio output, and you grant Motion access on first launch.","notVerified":["Whether it really recognises mouthed words for a new person, and how many words it can tell apart","Whether the project still builds with the current Xcode and macOS","How much disk the recorded sessions use over time","Whether anything is sent off the computer, which can only be seen by running it","How much the result changes when you take the headphones off and put them back on"]},"zh-Hans":{"summaryShort":"一个Mac实验项目,利用AirPods Pro的头部动作传感器,尝试把你只是用嘴型默念的词转成文字。","summary":"Subvox 是一个 macOS 研究原型,测试 AirPods Pro 的头部动作数据是否携带了足够的信息,把只用嘴型默念的词转成文字。你通过按住空格键并反复做出某个词的口型来训练它,之后它会把新的动作与那些录制数据进行匹配,并显示结果是否比随机猜测更好。它适合好奇的技术爱好者:没有下载包,你需要用 Homebrew 安装 xcodegen,然后在 Xcode 中编译这个应用。它需要 macOS 14 或更高版本,以及支持头部追踪的耳机。","easeRationale":"Mac 是 E 级:没有可下载的发布版本,README 里唯一的说明是'brew install xcodegen'、'xcodegen generate'以及在 Xcode 中编译这个项目。你还需要把 AirPods Pro 或其他支持头部追踪的耳机设为 Mac 的音频输出,并在首次启动时授予 Motion(运动数据)访问权限。","notVerified":["它是否真的能为一个新用户识别默念的词,能区分多少个词","这个项目在当前的 Xcode 和 macOS 下是否还能编译","随着时间推移,录制的会话会占用多少磁盘空间","是否有任何数据被发送到电脑之外,这只能通过运行它才能看到","摘下耳机再戴上之后,结果会有多大变化"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1MTAxMTQx","slug":"jqlang__jq","name":"jq","owner":"jqlang","fullName":"jqlang/jq","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A","web":"A"},"risk":"low","signals":[{"code":"license_file_mixes_several_licenses","severity":"info","evidence":"GitHub 没有为该仓库识别出标准许可证，因为 COPYING 里除了 MIT 文本之外，还分别列出了文档（CC BY 3.0）、decNumber（ICU 许可证）、dtoa.c 和 g_fmt.c（Lucent）、来自 Heimdal 的代码以及 NetBSD 的 strptime 的单独条款。确定性检查把该文件读作 MIT。要在产品中复用这些代码，就得把所有这些声明都读一遍。","source":"ai","title":"许可证文件混合了多种许可证"},{"code":"release_files_are_bare_executables","severity":"info","evidence":"README：'The platform-specific assets are standalone executables, not archives.' 没有安装程序，也没有自动更新，所以要自己放置文件，之后也要自己替换。仓库附带 KEYS 和 sig/ 文件夹，因此可以用项目的签名来核对已发布的文件。","source":"ai","title":"发布文件是单独的可执行文件"},{"code":"no_release_for_four_months","severity":"info","evidence":"最新版本是 2026-06-20 的 jq-1.8.2，而代码最近一次修改是在 2026-10-25，所以 6 月以来的修复只存在于源代码中，没有进入可下载的二进制文件。","source":"ai","title":"四个月没有发布新版本"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["docker"],"disk":{"minMB":1,"maxMB":3,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":36097,"downloads":312510454,"downloadsWeek":2479824,"downloadsDays":7,"growth":68,"growthDays":7,"history":[{"date":"2026-10-26","stars":35966},{"date":"2026-10-27","stars":35979},{"date":"2026-10-28","stars":35992},{"date":"2026-10-29","stars":35983},{"date":"2026-10-30","stars":35991},{"date":"2026-10-31","stars":36001},{"date":"2026-11-01","stars":36012},{"date":"2026-11-02","stars":36024},{"date":"2026-11-03","stars":36029},{"date":"2026-11-04","stars":36036},{"date":"2026-11-05","stars":36051},{"date":"2026-11-06","stars":36049},{"date":"2026-11-07","stars":36060},{"date":"2026-11-08","stars":36076},{"date":"2026-11-09","stars":36089},{"date":"2026-11-10","stars":36097}],"firstPublishedAt":"2026-10-27T01:10:12.645Z","commitSha":"ac35c7dcbb991fa0f3c3196dd00ca479211e9847","verifiedAt":"2026-10-28T01:42:51.265Z","lastCommitAt":"2026-11-02T07:08:10Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"jq-windows-amd64.exe","sizeBytes":1035264,"url":"https://github.com/jqlang/jq/releases/download/jq-1.8.2/jq-windows-amd64.exe","releaseUrl":"https://github.com/jqlang/jq/releases/tag/jq-1.8.2"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Reads JSON data in a terminal and pulls out, filters or reshapes exactly the parts you want.","summary":"jq is a small command-line program for JSON, the format most websites and APIs answer in. You give it a filter such as '.version' and it pulls out, sorts, counts or rebuilds the pieces you asked for, which makes it the usual way to read an API answer or a config file from a terminal or a script. It is for programmers and system administrators. The download is a single file of 1-2 MB for Windows, Mac or Linux with nothing else to install, there is a Docker image, and jqlang.org hosts a playground where you can try filters in a browser.","easeRationale":"Each release has a ready executable for every system (jq-windows-amd64.exe, jq-macos-arm64, jq-linux-amd64, all 1-2 MB) with no runtime to install; on Mac and Linux the README only asks you to rename the file to jq and run `chmod +x jq`. The browser playground at play.jqlang.org needs nothing at all. Using it still means typing commands and learning jq's filter language.","costNote":"Free and open source (MIT; the documentation is CC BY 3.0 and part of the code carries the ICU licence).","notVerified":["Whether Windows or macOS warns about the downloaded file and what it takes to allow it","How it behaves on very large JSON files or on broken JSON","Whether the version your package manager (apt, brew, winget) installs matches 1.8.2","What the online playground does with data you paste into it"]},"zh-Hans":{"summaryShort":"在终端中读取 JSON 数据,精准提取、过滤或重新组织你想要的部分。","summary":"jq 是一个用于处理 JSON 的小型命令行程序,JSON 是大多数网站和 API 返回数据的格式。你给它一个筛选表达式,比如 '.version',它就能提取、排序、计数或重组你想要的部分,因此它成了在终端或脚本中读取 API 响应或配置文件的常用方式。它适合程序员和系统管理员使用。下载包是面向 Windows、Mac 或 Linux 的单个 1-2 MB 文件,不需要安装其他任何东西,还有 Docker 镜像,jqlang.org 网站上有一个在浏览器里就能试用筛选表达式的演练场。","easeRationale":"每个发布版本都为每种系统提供现成的可执行文件(jq-windows-amd64.exe、jq-macos-arm64、jq-linux-amd64,均为 1-2 MB),不需要安装任何运行环境;在 Mac 和 Linux 上,README 只要求你把文件改名为 jq 并运行 `chmod +x jq`。play.jqlang.org 上的浏览器演练场则什么都不需要。不过使用它仍然需要输入命令,并学习 jq 的筛选语言。","costNote":"免费且开源(MIT 许可;文档采用 CC BY 3.0 许可,部分代码带有 ICU 许可)。","notVerified":["Windows 或 macOS 是否会对下载的文件发出警告,以及需要怎样操作才能允许运行","处理特别大的 JSON 文件或格式错误的 JSON 时表现如何","你的包管理器(apt、brew、winget)安装的版本是否与 1.8.2 一致","在线演练场会如何处理你粘贴进去的数据"]}}},{"id":"R_kgDORh-C1Q","slug":"juancadile__instabot","name":"Instabot","owner":"juancadile","fullName":"juancadile/instabot","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"very_young_and_little_used","severity":"attention","evidence":"28 个 star,没有 GitHub release,仓库创建和最后一次 push 是在同一天,2026-03-09,距这次检查 241 天","source":"ai","title":"项目很新,使用者很少"},{"code":"author_warns_about_account_suspension","severity":"attention","evidence":"README 的 Disclaimer 写道:\"The authors and contributors of this project accept no liability for any account suspension, ban, legal claim, or other consequence arising from the use of this software. Use at your own risk.\" 它还提到,把它用在第三方账号上 \"may require additional Meta approvals\"","source":"ai","title":"作者就账号被封提出警示"},{"code":"server_holds_instagram_access_token","severity":"attention","evidence":".env.example 要求提供 INSTAGRAM_PAGE_ACCESS_TOKEN、META_APP_SECRET 和 ADMIN_API_KEY;运营这台服务器的人掌握着能够读取和发送这个账号消息的凭据","source":"ai","title":"服务器保存着 instagram 访问令牌"},{"code":"collects_personal_data_of_other_people","severity":"info","evidence":"README 的功能描述:\"Conversational email flow — CTA button → ask for email\" 和 \"PostgreSQL — Leads, DM logs, and email status tracked in a database\",因此你要为其他人的邮箱地址和消息承担责任","source":"ai","title":"收集他人的个人数据"},{"code":"readme_promotes_authors_paid_services","severity":"info","evidence":"\"Need Help?\" 一节提供付费的定制搭建、功能开发、培训和咨询服务,并链接到作者为期 10 周的付费训练营;默认的邮件模板 bienvenido.html \"Golem Lab branded\"","source":"ai","title":"readme 推广作者的付费服务"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram (app secret and page access token)","Resend (emails)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true}],"declaredAlternatives":[{"target":"manychat","stance":"claims","evidence":"Open-source ManyChat alternative"}],"stars":28,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":28},{"date":"2026-11-07","stars":28},{"date":"2026-11-08","stars":28},{"date":"2026-11-09","stars":28},{"date":"2026-11-10","stars":28}],"firstPublishedAt":"2026-11-05T22:43:48.528Z","commitSha":"438f5efac698554dbf5e539086df879b8cac11a2","verifiedAt":"2026-11-05T22:29:16.890Z","lastCommitAt":"2026-03-09T16:24:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Sends an automatic Instagram DM when someone comments a keyword, then collects their email and sends them a link.","summary":"InstaBot is a small server you run yourself for Instagram direct messages. When someone comments one of your keywords on a post, it sends them a DM with a button; the conversation then asks for their email, waits, reminds them and delivers a link or file, by DM and by email. It also replies to story mentions and keeps leads and message logs in a PostgreSQL database. You need Node.js, a Meta developer app with an Instagram access token, a database and somewhere to host it; the README uses Railway and says it costs about 5 dollars a month.","easeRationale":"Windows, macOS and Linux all take the same route and get the same grade: clone the repository, run npm install, fill in a .env file with five or more values, edit keywords.json by hand, provide a PostgreSQL database, and then register the webhook address in Meta's developer console (with ngrok if you run it on your own machine). There is no interface to click through and nothing to double-click.","costNote":"The code is free (MIT). The README estimates about 5 dollars a month for hosting on Railway, and the email features need a Resend account","notVerified":["Whether it still works with Meta's current API, since there are no releases and no commits for eight months","Whether the Instagram permissions it needs really work without Meta's app review, as the disclaimer claims","The real monthly cost of hosting plus the database","Whether the database tables are created automatically on first start","How the 10-minute reminder behaves if the server restarts"]},"zh-Hans":{"summaryShort":"当有人评论某个关键词时自动发送 Instagram 私信,然后收集对方的邮箱并发送一个链接。","summary":"InstaBot 是一个由你自己运行的小型服务器,用于 Instagram 私信。当有人在帖子下评论你设定的某个关键词时,它会给这个人发送一条带按钮的私信;接下来的对话会询问对方的邮箱,等待,发提醒,然后通过私信和邮件发送一个链接或文件。它还会回复故事中的提及,并把潜在客户和消息记录保存在一个 PostgreSQL 数据库中。你需要 Node.js、一个带有 Instagram 访问令牌的 Meta 开发者应用、一个数据库和一个托管的地方;README 使用 Railway,并说大约每月花费 5 美元。","easeRationale":"Windows、macOS 和 Linux 走的是同一条路线,得到的评级也相同:克隆仓库,运行 npm install,在 .env 文件中填入五个或更多的值,手动编辑 keywords.json,准备一个 PostgreSQL 数据库,然后在 Meta 开发者控制台中注册 webhook 地址(如果在自己的机器上运行,还需要用 ngrok)。没有可以点来点去的界面,也没有什么东西可以双击。","costNote":"代码是免费的(MIT)。README 估算在 Railway 上托管大约每月 5 美元,邮件相关功能需要一个 Resend 账号","notVerified":["它是否仍能配合 Meta 当前的 API 工作,因为已经八个月没有 release 也没有 commit","正如免责声明所说,它所需的 Instagram 权限是否真的可以在没有 Meta 应用审核的情况下使用","托管加数据库的实际月费是多少","首次启动时数据库表是否会自动创建","服务器重启时,10 分钟的提醒会如何表现"]}}},{"id":"R_kgDOR5h10Q","slug":"juliusbrussee__caveman","name":"Caveman","owner":"JuliusBrussee","fullName":"JuliusBrussee/caveman","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"210天内116532颗star(约每天555颗)","source":"check","title":"异常的star增长速度"},{"code":"telemetry_on_by_default","severity":"attention","evidence":"README 的 Privacy 一节:\"The CLI sends usage stats by default: commands run, token counts, caveman-blocks commands and usage counts, a random install ID, OS, and IP\",并说可以用`caveman telemetry off`或`DO_NOT_TRACK=1`关闭。","source":"ai","title":"遥测默认开启"},{"code":"prompts_sent_to_vendor_cloud_when_routing_on","severity":"attention","evidence":"README 的 Privacy 一节关于 Auto 的部分:\"each ask sends Caveman Cloud your latest ask (with whatever your agent attaches to it, such as CLAUDE.md, memory or @-mentioned files) ... On the Free plan Caveman may keep it to improve routing\",被路由到 Caveman Cloud 模型的请求会\"passes through Caveman Cloud, whole (system prompt, tool results, code)\"。","source":"ai","title":"开启路由后,提示词会发送到厂商云端"},{"code":"installer_rewrites_other_tools_settings","severity":"attention","evidence":"INSTALL.md:安装程序会\"Auto-detects every supported agent installed on your machine\",\"runs that agent's native install path (plugin / extension / rule file / `npx skills add`)\",并且\"Wires Claude Code hooks and statusline badge on top, plus a Codex start-of-session hook\"——它会修改大约30个其他程序的文件和hooks。","source":"ai","title":"安装程序会重写其他工具的设置"},{"code":"install_script_piped_from_web","severity":"info","evidence":"INSTALL.md 提供了`curl -fsSL https://raw.githubusercontent.com/JuliusBrussee/caveman/v4.0.1/install.sh | bash`和`irm ...install.ps1 | iex`,并用自己的话警告:\"Piping a script straight into a shell runs it sight-unseen.\"这些脚本接着会执行`npx -y github:JuliusBrussee/caveman#v4.0.1`,固定在一个所有者可以随时移动的标签上。","source":"ai","title":"安装脚本通过管道从网络执行"},{"code":"self_reported_and_partial_benchmarks","severity":"info","evidence":"README 自己的说明:33.2%的会话数字\"is from August 2026 on an older engine\",HTML结果\"9.9% worse\",Auto的对比说\"the run is half done\",分数差距\"is too small to tell apart at this size\",而Blocks部分是\"Pre-release (v0.1.0-rc.5). Python only, no sandbox yet.\"","source":"ai","title":"自行报告且不完整的基准测试"},{"code":"release_binaries_unsigned_by_github","severity":"info","evidence":"只有 bin-v4.0.2 这个发布版本带有二进制文件(caveman-proxy、caveman-engine、caveman-mcp、cavemem 等,覆盖三个系统);INSTALL.md 说运行时二进制文件会\"checked against a checksum list signed with a key built into the CLI\",这一点无法从外部验证。","source":"ai","title":"发布的二进制文件未经GitHub签名"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","docker","go"],"disk":{"minMB":46,"maxMB":137,"basis":"installer"},"apiKeys":["Caveman Cloud account (for the Auto router)","Your own model provider key (Anthropic, OpenAI and others)"],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","ai_agents"],"alternativeTo":[{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"portkey","name":"Portkey","confirmed":true}],"declaredAlternatives":[],"stars":118866,"downloads":900080,"downloadsWeek":126967,"downloadsDays":7,"growth":1655,"growthDays":7,"history":[{"date":"2026-10-26","stars":115311},{"date":"2026-10-27","stars":115544},{"date":"2026-10-28","stars":115817},{"date":"2026-10-29","stars":116105},{"date":"2026-10-30","stars":116334},{"date":"2026-10-31","stars":116527},{"date":"2026-11-01","stars":116718},{"date":"2026-11-02","stars":116944},{"date":"2026-11-03","stars":117211},{"date":"2026-11-04","stars":117491},{"date":"2026-11-05","stars":117738},{"date":"2026-11-06","stars":117991},{"date":"2026-11-07","stars":118155},{"date":"2026-11-08","stars":118341},{"date":"2026-11-09","stars":118615},{"date":"2026-11-10","stars":118866}],"firstPublishedAt":"2026-10-27T01:10:12.647Z","commitSha":"9b68f7ed12b05e7efaff18e65c5a3fc2849fd0a4","verifiedAt":"2026-11-01T01:42:35.445Z","lastCommitAt":"2026-11-09T01:48:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes AI coding assistants reply in a short \"caveman\" style and shrinks the logs and files they read, to spend fewer tokens.","summary":"A set of add-ons for AI coding assistants such as Claude Code, Codex, Cursor and Gemini CLI. It installs instructions that make the assistant answer in a clipped style, plus a local proxy that compresses logs, JSON, CSV and web pages before the model reads them, and an optional router that picks a model per question. You install it in a terminal with one npx command and Node 22.13 or newer; it then edits the settings of each assistant it finds on your machine.","easeRationale":"C on all three: you need Node.js 22.13 or newer and a terminal, then `npx @caveman-ai/cli@latest`, which detects your assistants and changes their configuration. Windows needs PowerShell 5.1 or newer for the one-line form; the released files are command-line binaries, not an app you open.","costNote":"Apache-2.0 code, free to run. The \"Auto\" router needs a free Caveman account, and Caveman Cloud is a separate paid service (app.caveman.so).","notVerified":["Whether the token savings hold on your own projects and models","Whether the outside citations and blog results quoted in the README say what the README says they say","What exactly the installer changed in each assistant's configuration, and whether uninstalling restores it","What the telemetry and the Auto router send once running, since that needs watching the traffic"]},"zh-Hans":{"summaryShort":"让AI编程助手用简短的\"caveman\"风格回复,并压缩它们读取的日志和文件,从而减少token消耗。","summary":"一套适用于 Claude Code、Codex、Cursor 和 Gemini CLI 等 AI 编程助手的附加组件。它会安装指令,让助手用简短的风格回答,还有一个本地代理,会在模型读取之前压缩日志、JSON、CSV 和网页,以及一个可选的路由器,按问题选择模型。你在终端里用一条 npx 命令和 Node 22.13 或更高版本安装它;然后它会修改它在你机器上找到的每个助手的设置。","easeRationale":"三个平台都是 C 级:你需要 Node.js 22.13 或更高版本以及一个终端,然后执行`npx @caveman-ai/cli@latest`,它会检测你的助手并更改它们的配置。Windows 上单行命令需要 PowerShell 5.1 或更高版本;发布的文件是命令行二进制文件,不是可以打开的应用。","costNote":"代码采用 Apache-2.0 许可,运行免费。\"Auto\"路由器需要一个免费的 Caveman 账户,而 Caveman Cloud 是一个独立的付费服务(app.caveman.so)。","notVerified":["token节省效果在你自己的项目和模型上是否成立","README中引用的外部引述和博客结果是否真的如README所说","安装程序到底改动了每个助手配置中的哪些内容,卸载后是否能恢复","遥测和Auto路由器在运行时会发送什么,这需要观察流量才能知道"]}}},{"id":"R_kgDOVcA9Ag","slug":"juneyaooo__rea-video","name":"Rea Video","owner":"JuneYaooo","fullName":"JuneYaooo/rea-video","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"11 天内获得 5441 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"an_ai_assistant_is_told_to_install_and_run_it_for_you","severity":"attention","evidence":"README 的 Quick start 部分让你把这段内容粘贴进 Claude Code 或 Codex：\"Install https://github.com/JuneYaooo/rea-video, then break down this video: /path/to/video.mp4\"，并说 \"The agent installs the CLI, learns the workflow and runs the whole thing.\" 你的助手会遵循写在这个仓库里的安装步骤，而这些步骤随时可能更改。","source":"ai","title":"要求 AI 助手替你安装并运行"},{"code":"the_installer_changes_packages_in_your_python_setup","severity":"attention","evidence":"pyproject.toml 中关于可选 OCR 附加项的注释：\"rapidocr 会顺带装 opencv-python（带 GUI 的那套）... scripts/install_skill.sh 装完会删掉它、重装 headless\" —— 安装脚本会移除 opencv-python 包并重新安装无界面（headless）版本，这可能影响你电脑上其他的 Python 程序。","source":"ai","title":"安装脚本会改动你 Python 环境里的软件包"},{"code":"it_writes_into_your_ai_assistant_folders","severity":"info","evidence":"README 的手动安装部分：\"bash scripts/install_skill.sh  # installs the CLI (pip, or pipx / a local .venv if pip is blocked) and links the skill into ~/.claude, ~/.codex, ~/.agents\"。仓库里还有一个 .claude-plugin/ 文件夹和一个 skills/ 文件夹。","source":"ai","title":"会写入你的 AI 助手文件夹"},{"code":"downloads_a_model_the_first_time_it_runs","severity":"info","evidence":"pyproject.toml 依赖注释：\"onnxruntime>=1.16,   # 人声/配乐分离（UVR MDX-Net，首次运行自动下载约 67 MB 模型）\" —— 首次运行时会自动获取约 67 MB 的模型文件，用于分离人声和配乐。","source":"ai","title":"首次运行时会下载一个模型"},{"code":"other_peoples_videos_are_used_as_examples","severity":"info","evidence":"README：docs/cases/videos/ 中的样例是 TikTok 和抖音的片段，重制部分写道 \"The originals and OiiOii versions come from the articles and are shown for comparison only.\" FAQ 对 \"Can I just repost someone else's video?\" 的回答是 \"No ... Don't reuse other people's footage, music, characters or likeness.\"","source":"ai","title":"用别人的视频作为示例"}],"platforms":["mac","linux"],"tools":["python","ffmpeg"],"buildTools":["python","ffmpeg"],"disk":{"minMB":160,"maxMB":420,"basis":"heuristic"},"apiKeys":["Gemini or Doubao video-capable model"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","ai_skills","social_marketing"],"alternativeTo":[],"declaredAlternatives":[],"stars":6282,"downloads":0,"downloadsWeek":0,"downloadsDays":0,"growth":4292,"growthDays":5,"history":[{"date":"2026-11-10","stars":6282}],"firstPublishedAt":"2026-11-09T23:02:44.131Z","commitSha":"356815c0d61291811551a2b8813b2b571d9b3238","verifiedAt":"2026-11-09T22:48:48.668Z","lastCommitAt":"2026-11-10T17:19:02Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Breaks a video down shot by shot, says what was filmed, what was AI-made or coded, and how to remake it.","summary":"rea-video is a command-line tool and a skill for AI assistants such as Claude Code or Codex. You point it at a video file and it cuts it into shots, then for each shot it judges whether it was filmed, generated by AI, built in code or screen-recorded, and writes an HTML report with frames, evidence, camera and lighting notes, captions and the audio breakdown, plus a plan to remake it. It is aimed at video creators and marketers who are comfortable with a terminal. You need Python 3.10 or newer and ffmpeg; a paid vision model is optional.","easeRationale":"There is no ready app: on a Mac or Linux you install ffmpeg and Python 3.10 or newer, clone the repository and run the shell script scripts/install_skill.sh, or let an AI assistant run those commands for you. The install instructions only cover Mac (brew) and Linux (apt), so Windows gets no grade.","costNote":"The offline analysis is free. The README says a video-capable model (Gemini or Doubao) is optional and that \"you pay whatever that service charges\"; automatic remaking with the separate capsule-cinema project calls paid models after you confirm.","notVerified":["Whether the accuracy table (98.3% to 100% on the how-it-was-made call) holds for your own videos; the numbers come from the authors' own test sets.","What the install script does in detail and what it leaves behind in ~/.claude, ~/.codex and ~/.agents.","Real disk use and running time after the audio models are downloaded and a long video is analysed.","What a run costs when you connect a paid vision model, since every uncertain shot is sent to it.","Whether it works on Windows, for example through WSL, since the documented install only covers Mac and Linux.","What the website reavideo.ai offers and whether anything there is paid; the repository only links to it."]},"zh-Hans":{"summaryShort":"把视频按镜头逐一拆解，判断哪些是实拍、哪些是 AI 生成或代码制作，并说明如何重新制作。","summary":"rea-video 是一个命令行工具，也是面向 Claude Code 或 Codex 等 AI 助手的技能（skill）。你指定一个视频文件，它会把视频切分成镜头，然后对每个镜头判断是实拍、AI 生成、代码制作还是屏幕录制，并生成一份 HTML 报告，包含画面、证据、运镜和灯光笔记、字幕以及音频拆解，还附带一份重新制作的方案。它面向习惯使用终端的视频创作者和营销人员。你需要 Python 3.10 或更高版本以及 ffmpeg；付费的视觉模型是可选的。","easeRationale":"没有现成的应用：在 Mac 或 Linux 上，你需要安装 ffmpeg 和 Python 3.10 或更高版本，克隆仓库并运行脚本 scripts/install_skill.sh，或者让 AI 助手替你执行这些命令。安装说明只覆盖了 Mac（brew）和 Linux（apt），所以 Windows 没有评分。","costNote":"离线分析是免费的。README 说具备视频能力的模型（Gemini 或 Doubao）是可选的，并说 \"you pay whatever that service charges\"；用独立的 capsule-cinema 项目自动重新制作时，在你确认之后会调用付费模型。","notVerified":["准确率表格（判断制作方式的准确率为 98.3% 到 100%）是否适用于你自己的视频；这些数字来自作者自己的测试集。","安装脚本具体做了什么，以及它在 ~/.claude、~/.codex 和 ~/.agents 中留下了什么。","下载完音频模型并分析一段长视频之后，实际的磁盘占用和运行时间。","接入付费视觉模型后一次运行的实际花费，因为每个不确定的镜头都会被发送过去。","它能否在 Windows 上运行，比如通过 WSL，因为文档中的安装说明只覆盖了 Mac 和 Linux。","reavideo.ai 这个网站提供什么内容，是否有付费项目；仓库只是链接到它。"]}}},{"id":"R_kgDOVVOmXw","slug":"justaresearcher__gpu-unlock-ada","name":"GPU Unlock Ada","owner":"JustAResearcher","fullName":"JustAResearcher/gpu-unlock-ada","type":"dev_tool","status":"alert","grade":"E","easeByPlatform":{"linux":"E"},"risk":"high","signals":[{"code":"bypasses_vendor_firmware_limit","severity":"high","evidence":"这正是该项目的全部目的。README：“This Linux kernel module enables the SM speed-select override before the stock NVIDIA driver starts”；RTX 4090 的路径“writes SS0 0x88888888 and SS1 0x8, then performs FLR before the unmodified NVIDIA driver binds”，并且“The payload uses instruction addresses derived from published decrypted debug-booter analysis”，针对的是 NVIDIA 的签名固件。去掉厂商写进显卡固件里的限制，违反 NVIDIA 的条款，也可能被视为规避技术保护措施，这在一些国家受到限制。","source":"ai","title":"绕过厂商固件设定的限制"},{"code":"can_leave_the_machine_or_card_in_a_bad_state","severity":"attention","evidence":"README 的 Check and disable 部分：“If the module reports that firmware did not stop and its DMA buffers are being kept, power off before trying again... Do not force-unload it in this state.” 它还提到要“Keep an unmodified stock entry available”，在 nvidia-drm 接管之前屏幕会一直黑屏，并且解锁后的 RTX 4090“peaked at 346 W against 313 W stock”。","source":"ai","title":"可能让电脑或显卡处于异常状态"},{"code":"anonymous_author_new_repository_kernel_code","severity":"attention","evidence":"所有者账号为“JustAResearcher”；仓库创建于 2026-10-23，已有 4276 个星标和 556 个 fork。代码在 Linux 内核中运行，并在 GPU 驱动加载之前直接访问硬件（BAR0、DMA），其载荷（payload）来自第三方在 Hugging Face 上传的文件（TOTORONG/AD103_win_unlocker）。","source":"ai","title":"作者匿名、仓库很新且涉及内核代码"},{"code":"results_from_single_cards","severity":"info","evidence":"README 的 Experimental Ada targets 部分：“Results come from one or a few cards per ID; other boards, memory configurations and laptop models remain unverified”，以及“Laptops differ in VBIOS ROM exposure, embedded-controller power rails and FLR handling, so a result on one model does not carry over to another.”","source":"ai","title":"结果只来自个别显卡"},{"code":"firmware_downloaded_at_build_time","severity":"info","evidence":"README 的 Build and install 部分：“Builds fetch the pinned NVIDIA 610.43.03 source and binary inputs over HTTPS and verify their SHA-256 hashes”，以及“Prebuilt firmware packages are not distributed by this repository”，因此在构建过程中会把 NVIDIA 的材料下载到你的电脑上。","source":"ai","title":"构建时下载固件"}],"platforms":["linux"],"tools":["python","other"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":4862,"downloads":5493,"downloadsWeek":602,"downloadsDays":7,"growth":123,"growthDays":7,"history":[{"date":"2026-10-26","stars":3836},{"date":"2026-10-27","stars":4275},{"date":"2026-10-28","stars":4418},{"date":"2026-10-29","stars":4518},{"date":"2026-10-30","stars":4588},{"date":"2026-10-31","stars":4626},{"date":"2026-11-01","stars":4669},{"date":"2026-11-02","stars":4705},{"date":"2026-11-03","stars":4739},{"date":"2026-11-04","stars":4766},{"date":"2026-11-05","stars":4790},{"date":"2026-11-06","stars":4811},{"date":"2026-11-07","stars":4831},{"date":"2026-11-08","stars":4846},{"date":"2026-11-09","stars":4854},{"date":"2026-11-10","stars":4862}],"firstPublishedAt":"2026-10-27T01:14:24.002Z","commitSha":"d28c485418c3d9d034cd4bac19200bc33dd3453b","verifiedAt":"2026-10-28T01:40:40.027Z","lastCommitAt":"2026-10-26T04:21:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Linux kernel module that lifts NVIDIA's tensor speed limit on GeForce RTX 40 cards, about doubling one measured rate.","summary":"A kernel module for Linux, for people who are comfortable compiling software. It loads before the NVIDIA driver, writes to the card's speed-select registers and lifts the limit NVIDIA puts on FP16, BF16 and FP8 maths with FP32 accumulate: the RTX 4090 figures in the README go from about 165 to about 331 TF/s on instruction rates, while real application gains are reported as much smaller or absent. You build packages from source, add options to the kernel command line with your card's PCI address, and cold-boot. Only the RTX 4090 is treated as supported.","easeRationale":"Linux is the only system it runs on and it is a job for developers: clone the repository, run make install for your distribution (kernel headers and a working DKMS toolchain are required), let the build download and generate the pinned NVIDIA firmware inputs, then edit the kernel command line with the exact PCI address from lspci -D and power the machine fully off and on. The Windows folder is a report of one manual boot and ships no binaries or installer.","notVerified":["Whether your exact card, board maker and laptop model unlock at all","Whether the extra speed shows up in the programs you actually use, beyond the register-level numbers","Long-term stability, heat and power draw after the change","Whether the card still works normally after a driver or kernel update","What the change does to your warranty and to vendor support"]},"zh-Hans":{"summaryShort":"一个 Linux 内核模块,解除 NVIDIA 在 GeForce RTX 40 系列显卡上设置的张量运算速度限制,让某项实测速率提升了约一倍。","summary":"这是一个面向习惯自行编译软件的用户的 Linux 内核模块。它在 NVIDIA 驱动加载之前运行,写入显卡的速度选择寄存器,解除 NVIDIA 对 FP16、BF16 和采用 FP32 累加的 FP8 运算设下的限制:README 中 RTX 4090 的指令速率数据从约 165 TF/s 提升到约 331 TF/s,而实际应用程序中报告的收益要小得多,甚至没有提升。你需要从源代码构建软件包,用自己显卡的 PCI 地址给内核命令行添加选项,然后彻底冷启动。只有 RTX 4090 被视为受支持的型号。","easeRationale":"它只能在 Linux 上运行,而且是开发者才能完成的工作:克隆仓库,针对自己的发行版运行 make install(需要内核头文件和可用的 DKMS 工具链),让构建过程下载并生成固定版本的 NVIDIA 固件输入文件,再用 lspci -D 得到的准确 PCI 地址编辑内核命令行,然后彻底关机再开机。Windows 文件夹里只是一次手动启动的记录,不提供任何二进制文件或安装程序。","notVerified":["你手上具体型号的显卡、板卡厂商和笔记本机型是否真的能解锁","除了寄存器层面的数字外,这种额外速度是否真的体现在你实际使用的程序中","改动之后长期的稳定性、发热和功耗情况","驱动或内核更新之后,显卡是否仍能正常工作","这种改动对保修和厂商支持会产生什么影响"]}}},{"id":"R_kgDOTKgHjQ","slug":"justvugg__colibri","name":"Colibri","owner":"JustVugg","fullName":"JustVugg/colibri","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"very_large_downloads","severity":"attention","evidence":"程序文件本身很小，但模型不是：README要求\"22 GB free on the disk for the smallest model\"，它的表格里列出的下载量有23 GB（Qwen3.6-35B）、429 GB（GLM-5.2）和1.56 TB（Kimi K3）。要提前考虑磁盘空间和下载时间。","source":"ai","title":"下载文件非常大"},{"code":"model_licences_restrict_commercial_use","severity":"info","evidence":"引擎是Apache-2.0，但模型有自己的条款：README标注了\"Qwen-Image-2.1: text to picture, non-commercial licence\"和\"Muse Spark Open (816B, 42B active): text and tools, under Meta's licence agreement\"。在工作中使用前要检查每个模型的许可证。","source":"ai","title":"模型许可证限制商业用途"},{"code":"speed_can_be_very_slow","severity":"info","evidence":"项目自己的测量数据显示不同模型之间差距有多大：GLM-5.2是\"0.05-0.1 tok/s cold on a 25 GB laptop\"，Kimi K3是\"about 9.4 s per token\"，相比之下较小的Qwen3.6-35B每秒能生成6到30个token。在一台普通电脑上，最大的模型一次只能给出一个词。","source":"ai","title":"速度可能非常慢"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","gpu_any"],"disk":{"minMB":25,"maxMB":74,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["local_ai","development"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true}],"declaredAlternatives":[],"stars":49729,"downloads":113907,"downloadsWeek":17627,"downloadsDays":6,"growth":2490,"growthDays":7,"history":[{"date":"2026-11-04","stars":47691},{"date":"2026-11-05","stars":48141},{"date":"2026-11-06","stars":48428},{"date":"2026-11-07","stars":48696},{"date":"2026-11-08","stars":49046},{"date":"2026-11-09","stars":49345},{"date":"2026-11-10","stars":49729}],"firstPublishedAt":"2026-11-04T01:36:50.466Z","commitSha":"0a5504a1dc64f86fcd0fe7db298b0448c730e7e6","verifiedAt":"2026-11-04T01:23:50.393Z","lastCommitAt":"2026-10-31T21:29:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs very large open AI models on an ordinary computer, keeping part of the model in memory and reading the rest from disk.","summary":"colibri chats with big open AI models on the machine you already own. Only the part a model needs for each word stays in memory; the rest is read from the disk as it is asked for, so a 744-billion-parameter model can start on a computer with 24 GB of RAM, slowly. It comes with a chat page in your browser, a terminal chat and a local server that other apps and coding tools can talk to. You need at least 8 GB of RAM, 22 GB of free disk for the smallest model and a download; a graphics card is optional.","easeRationale":"Windows is the easiest: you download the ZIP, double-click START-HERE.bat and, in the README's words, \"If Python is missing, it offers to install it for you\", then answer one question and it downloads the model and opens the dashboard in your browser. On Mac and Linux the same thing takes terminal commands and installing tools first (\"sudo apt install git python3 build-essential\", or Xcode tools plus Homebrew packages), so those are a C.","costNote":"Free and open source (Apache-2.0); the README has a \"Buy me a coffee\" donation button.","notVerified":["Whether the speeds in the README's tables hold on your own computer and disk","Whether the Windows script really installs Python and sets up the engine without a hitch","How much disk space and memory you end up using for the model you pick, plus its expert cache","Whether the model downloads work without a Hugging Face account","Whether the answers match the quality of the same model run on a cloud service"]},"zh-Hans":{"summaryShort":"在普通电脑上运行非常大的开放AI模型，把模型的一部分留在内存里，其余部分从磁盘读取。","summary":"colibri让你在自己已有的电脑上和大型开放AI模型对话。每次生成一个词时，只有模型所需的那部分留在内存里，其余部分按需从磁盘读取，所以一个7440亿参数的模型也能在一台只有24 GB内存的电脑上启动，只是速度会慢。它自带浏览器聊天页面、终端聊天工具，以及其他应用和编程工具可以连接的本地服务器。你至少需要8 GB内存，最小模型需要22 GB的空闲磁盘空间，还要下载模型；显卡是可选的。","easeRationale":"Windows最简单：下载ZIP，双击START-HERE.bat，用README的话说，\"If Python is missing, it offers to install it for you\"，然后回答一个问题，它就会下载模型并在浏览器里打开控制台。在Mac和Linux上，同样的事情需要先用终端命令安装工具（\"sudo apt install git python3 build-essential\"，或者Xcode工具加Homebrew包），所以这两个是C级。","costNote":"免费且开源（Apache-2.0）；README里有一个\"Buy me a coffee\"的捐赠按钮。","notVerified":["README表格里的速度数据在你自己的电脑和磁盘上是否依然成立","Windows脚本是否真的能顺利安装Python并配置好引擎","你最终选的模型加上它的专家缓存会占用多少磁盘空间和内存","模型下载是否可以在没有Hugging Face账号的情况下完成","回答质量是否能和同一个模型在云服务上运行时相当"]}}},{"id":"R_kgDOQFcxuA","slug":"k-dense-ai__scientific-agent-skills","name":"Scientific Agent Skills","owner":"K-Dense-AI","fullName":"K-Dense-AI/scientific-agent-skills","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接管道传给shell执行：curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"通过curl管道执行到shell"},{"code":"skills_can_run_code_on_your_machine","severity":"attention","evidence":"README.md的Security Disclaimer：\"Agent Skills are powerful — they can instruct your AI agent to run arbitrary code, install packages, make network requests, and modify files on your system. A malicious or poorly written skill has the potential to steer your coding agent into harmful behavior.\" 作者还补充说他们\"cannot guarantee that every skill has been exhaustively reviewed\"，并建议\"Do not install everything at once.\"","source":"ai","title":"技能可以在你的电脑上运行代码"},{"code":"community_contributions_less_reviewed","severity":"attention","evidence":"README.md：\"Skills authored by K-Dense (K-Dense-AI) have been through our internal review process. Community-contributed skills have been reviewed to the best of our ability, but with limited resources\"，以及\"the repository now includes many community contributions that we may not have reviewed as thoroughly\"。每周的自动扫描结果发布在docs/security-report.md里。","source":"ai","title":"社区贡献的审核较少"},{"code":"medical_and_clinical_limits","severity":"attention","evidence":"有几组技能涉及和患者相关的工作，README明确限定了它们的范围：\"Healthcare AI & Biosignal Research - EHR and model research, physiological signal analysis, and retrospective validation—not patient-specific diagnosis, treatment, alarms, or deployment decisions\"，临床报告会保持\"visibly marked drafts\"的状态，等待有资质的人审核。这里的任何东西都不是医疗器械或诊断。","source":"ai","title":"医疗和临床方面的限制"},{"code":"dependencies_are_not_installed_for_you","severity":"info","evidence":"README.md的Prerequisites：\"Installing the skill files does not install their dependencies\"，以及\"Use a separate environment for each scientific workflow: some skills require different Python versions or incompatible package pins.\" 让一个工作流程跑起来可能意味着一次很长的软件包安装过程。","source":"ai","title":"依赖项不会自动帮你安装"},{"code":"unverified_popularity_claim","severity":"info","evidence":"仓库简介写道\"The #1 Agent Skills library for science, used by 250,000+ scientists worldwide\"和\"177 ready-to-use validated skills\"。这两个数字都没有给出来源，而同一份README里2.71.0版本的说明又提到\"all 181 skills\"。","source":"ai","title":"未经验证的受欢迎程度说法"},{"code":"some_skills_removed_over_licensing","severity":"info","evidence":"README.md的\"What's new in 2.72.0\"：docx、pdf、pptx和xlsx这几个技能\"which were vendored from anthropics/skills under Anthropic's own license, are no longer bundled\"。仓库本身是MIT许可证，但个别技能曾带有其他许可证。","source":"ai","title":"部分技能因许可证问题被移除"},{"code":"promotes_the_company_products","severity":"info","evidence":"README.md开头有好几处在推广同一家公司的另一个产品（K-Dense BYOK）、一个网络研讨会、社交账号，以及\"Enterprise Ready - Commercial support available for advanced needs\"，还请读者给仓库加星标。","source":"ai","title":"宣传公司自己的其他产品"},{"code":"declared_alternative_false_match","severity":"info","evidence":"自动检查从\"Open Notebook (self-hosted NotebookLM alternative — PDFs, videos, audio, web pages...)\"这一行里读出了\"invideo-ai\"。这一行说的是一个集成技能和NotebookLM，和付费视频产品无关，所以这里不会列出invideo-ai。","source":"ai","title":"声明的替代产品并不匹配"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["The model provider behind your AI assistant (Anthropic, OpenAI, Google and others)","Scientific and search services used by individual skills (Exa, Parallel, Benchling, NCBI, Materials Project, OpenRouter and more)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","productivity"],"alternativeTo":[{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[{"target":"invideo-ai","stance":"claims","evidence":"- Research notebooks: Open Notebook (self-hosted NotebookLM alternative — PDFs, videos, audio, web pages; 16+ AI providers; multi-speaker podcast generation)"},{"target":"notebooklm","stance":"claims","evidence":"- Research notebooks: Open Notebook (self-hosted NotebookLM alternative — PDFs, videos, audio, web pages; 16+ AI providers; multi-speaker podcast generation)"}],"stars":56728,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":755,"growthDays":7,"history":[{"date":"2026-11-04","stars":56101},{"date":"2026-11-05","stars":56212},{"date":"2026-11-06","stars":56321},{"date":"2026-11-07","stars":56397},{"date":"2026-11-08","stars":56487},{"date":"2026-11-09","stars":56601},{"date":"2026-11-10","stars":56728}],"firstPublishedAt":"2026-11-04T01:36:50.473Z","commitSha":"a25cf9ec00e2745d03f8704acf6f6919694ccd9a","verifiedAt":"2026-11-04T01:29:03.723Z","lastCommitAt":"2026-11-09T09:17:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"177 ready-made skills that teach an AI assistant to use real scientific tools and databases in biology, chemistry and medicine.","summary":"This is a library of 177 written skills you install into an AI assistant such as Claude Code, Cursor, Codex or Gemini CLI. Each one teaches the assistant how to use a specific scientific tool, Python package or public database correctly — gene data, molecules, mass spectrometry, clinical trial records, satellite images, statistics and figures — with the conventions and checks that field expects. The skills are only text and helper scripts: you still need Python 3.13, the uv package manager, a paid AI assistant and the scientific packages each skill uses.","easeRationale":"On macOS and Linux it is C: install Python 3.13 and uv, then one command such as `npx skills add K-Dense-AI/scientific-agent-skills` or `gh skill install`, and after that each skill still needs its own packages installed. Windows is D because the Prerequisites list it only \"with WSL2\", so you first set up a whole Linux environment inside Windows.","costNote":"The skills are free (MIT). The AI assistant that runs them is not: you need a paid plan or your own API keys, and some skills call services that charge or require registration.","notVerified":["Whether a given skill still matches the current version of the package or database it documents","Whether the results of a workflow are scientifically correct — the skills guide the assistant, they do not check the science","How much disk space and how many packages one scientific workflow really needs","What each skill sends to outside services and which ones charge for it","How thoroughly any individual community-contributed skill was reviewed"]},"zh-Hans":{"summaryShort":"177个现成的技能，教AI助手在生物学、化学和医学领域使用真实的科学工具和数据库。","summary":"这是一个包含177个文本技能的库，你把它安装进像Claude Code、Cursor、Codex或Gemini CLI这样的AI助手里。每个技能教助手正确使用某个特定的科学工具、Python包或公共数据库——基因数据、分子、质谱、临床试验记录、卫星图像、统计和图表——并遵循该领域所期望的规范和检查方式。这些技能只是文本和辅助脚本：你仍然需要Python 3.13、uv包管理器、一个付费的AI助手，以及每个技能所用到的科学软件包。","easeRationale":"在macOS和Linux上是C级：先安装Python 3.13和uv，再运行一条命令，比如`npx skills add K-Dense-AI/scientific-agent-skills`或`gh skill install`，之后每个技能还需要单独安装自己的依赖包。Windows是D级，因为Prerequisites只写了\"with WSL2\"，意味着你得先在Windows里搭建一整套Linux环境。","costNote":"这些技能是免费的（MIT）。运行它们的AI助手不是：你需要一个付费套餐或自己的API密钥，有些技能会调用收费或需要注册的服务。","notVerified":["某个具体技能是否仍与它所记录的软件包或数据库的当前版本一致","工作流程的结果是否在科学上正确——这些技能只是引导助手，并不会检查科学内容本身","一个科学工作流程实际需要多少磁盘空间和多少个软件包","每个技能向外部服务发送了什么内容，哪些服务会收费","任何社区贡献的技能到底经过了多严格的审核"]}}},{"id":"R_kgDOR1vNbQ","slug":"k2-fsa__omnivoice","name":"OmniVoice","owner":"k2-fsa","fullName":"k2-fsa/OmniVoice","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"voice_cloning_can_impersonate_people","severity":"attention","evidence":"README.md的Disclaimer：\"Users are strictly prohibited from using this model for unauthorized voice cloning, voice impersonation, fraud, scams, or any other illegal or unethical activities.\" 这个工具能从一段3到10秒的录音克隆出声音，所以只要你有某人的录音，就能复刻出他的声音，代码里没有任何地方会检查是否获得同意。","source":"ai","title":"语音克隆可以冒充他人"},{"code":"local_web_demo_opens_to_the_network","severity":"attention","evidence":"文档里给出的命令是`omnivoice-demo --ip 0.0.0.0 --port 8001`（README.md的Quick Start和Command-Line Tools部分）。绑定到0.0.0.0意味着你网络里的每台设备都能打开这个声音克隆页面，而且没有提到任何密码。","source":"ai","title":"本地网页演示会对网络开放"},{"code":"downloads_models_on_first_run","severity":"info","evidence":"`OmniVoice.from_pretrained(\"k2-fsa/OmniVoice\")`会在第一次运行时从Hugging Face获取权重，如果你省略了`ref_text`，它还会下载一个Whisper模型来转录你的参考音频。检查结果估算磁盘占用在950 MB到4450 MB之间。","source":"ai","title":"首次运行时会下载模型"},{"code":"third_party_download_mirrors_suggested","severity":"info","evidence":"README.md建议用`export HF_ENDPOINT=\"https://hf-mirror.com\"`来下载模型，用`uv sync --default-index \"https://mirrors.aliyun.com/pypi/simple\"`来下载软件包。两者都会把下载请求导向一个不是官方来源的镜像站。","source":"ai","title":"建议使用第三方下载镜像"},{"code":"quality_claims_are_the_authors_own","severity":"info","evidence":"\"600+ languages\"、\"state-of-the-art\"以及速度数字都来自作者自己的README和论文（arXiv:2604.00688）。README自己也提醒，声音设计\"is trained on Chinese and English data only\"，\"may produce unstable results for some low-resource languages\"，而跨语言克隆会带上参考声音的口音。","source":"ai","title":"质量方面的说法来自作者自己"},{"code":"early_version_research_code","severity":"info","evidence":"版本0.2.1，共有六个发行版，最新的发布于2026-07-16，确定性检查记录的最近一次提交是在2026-08-24，也就是检查时的71天前。pyproject.toml把目标用户列为\"Science/Research\"和\"Developers\"。","source":"ai","title":"早期版本的研究代码"}],"platforms":["windows","mac","linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true}],"declaredAlternatives":[],"stars":16480,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":192,"growthDays":6,"history":[{"date":"2026-11-04","stars":16288},{"date":"2026-11-05","stars":16318},{"date":"2026-11-06","stars":16351},{"date":"2026-11-07","stars":16372},{"date":"2026-11-08","stars":16403},{"date":"2026-11-09","stars":16432},{"date":"2026-11-10","stars":16480}],"firstPublishedAt":"2026-11-04T01:36:50.483Z","commitSha":"08be0b4ccbac3e13e374e86fbfead4b4cac343e2","verifiedAt":"2026-11-04T01:30:05.340Z","lastCommitAt":"2026-08-24T03:03:01Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from a few seconds of audio and makes it read any text, in more than 600 languages, on your own computer.","summary":"OmniVoice is a text-to-speech model you run yourself. Give it 3 to 10 seconds of someone speaking and it reads your text in that voice, or describe a voice instead (female, low pitch, British accent) and it invents one. It can add laughter or sighs and fix pronunciations. It runs on an NVIDIA, Apple Silicon or Intel Arc graphics chip, with a simple web page started from the terminal; the authors also host a free online demo you can try without installing anything. Free and Apache-2.0, with a 600+ language claim from the authors' paper.","easeRationale":"The easiest way in is the web: the authors keep a free Hugging Face Space (and a Google Colab notebook), so you open a page and type text, which is why web is A. Installing it is D on all three systems: you create a Python environment, install the right PyTorch build for your exact graphics chip (CUDA 12.8 wheels for NVIDIA, a separate Intel Arc index, Apple Silicon for Macs), then `pip install omnivoice` and run `omnivoice-demo` from the terminal.","costNote":"Free software and free model weights. The cost is your own hardware: a graphics card and roughly 1 to 4.5 GB of disk for the downloaded model.","notVerified":["How good the cloned voice sounds in your language, especially outside Chinese and English","Whether it runs usefully on a computer without a graphics card, and how slow that is","Real disk and memory use after the model and the Whisper transcriber are downloaded","Whether the hosted online demo stays available and what it does with the audio you upload","Whether the speed figures hold on anything other than the H100 card used in the benchmark"]},"zh-Hans":{"summaryShort":"只需几秒钟的音频就能复刻一个声音，让它用600多种语言朗读任意文本，并在你自己的电脑上运行。","summary":"OmniVoice是一个你自己运行的文本转语音模型。给它3到10秒某人说话的音频，它就能用那个声音朗读你的文本；或者你可以描述一个声音（女性、低音调、英式口音），它会凭空造出一个。它能加入笑声或叹息，还能修正发音。它运行在NVIDIA、Apple Silicon或Intel Arc显卡上，有一个从终端启动的简单网页；作者还提供一个免费的在线演示，不用安装任何东西就能试用。免费且采用Apache-2.0许可，600多种语言的说法来自作者自己的论文。","easeRationale":"最简单的使用方式是网页：作者维护着一个免费的Hugging Face Space（还有一个Google Colab笔记本），你打开页面输入文字就行，这就是为什么网页是A级。在这三个系统上安装都是D级：你需要创建Python环境，为你具体的显卡安装对应的PyTorch版本（NVIDIA用CUDA 12.8的wheel包，Intel Arc用单独的索引，Mac用Apple Silicon版本），然后执行`pip install omnivoice`，再从终端运行`omnivoice-demo`。","costNote":"软件免费，模型权重也免费。成本在于你自己的硬件：一块显卡，以及下载模型所需的大约1到4.5 GB磁盘空间。","notVerified":["克隆出来的声音在你的语言里听起来怎么样，尤其是中文和英文之外的语言","在没有显卡的电脑上它是否能正常使用，速度会有多慢","下载模型和Whisper转录器之后的实际磁盘和内存占用","托管的在线演示是否会一直可用，它对你上传的音频做了什么处理","速度数据在基准测试所用的H100显卡之外是否依然成立"]}}},{"id":"R_kgDOH6w6Yw","slug":"k2-fsa__sherpa-onnx","name":"sherpa-onnx","owner":"k2-fsa","fullName":"k2-fsa/sherpa-onnx","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"B","mac":"C","linux":"C","web":"A"},"risk":"low","signals":[{"code":"voice_cloning_included","severity":"info","evidence":"README列出了演示页面，比如\"Voice cloning with ZipVoice (Chinese+English)\"和\"Voice cloning with Pocket TTS (English)\"。未经许可复制他人的声音在一些国家可能违法。","source":"ai","title":"包含语音克隆功能"},{"code":"models_downloaded_separately","severity":"info","evidence":"README有一个\"Links for pre-trained models\"部分，指向外部下载页面，提供语音识别、文本转语音、VAD、关键词检测和说话人识别模型。程序文件本身是22-66 MB，但你额外添加的模型并不计入其中。","source":"ai","title":"模型需要单独下载"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":22,"maxMB":66,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai","development"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"otter-ai","name":"Otter.ai","confirmed":true},{"slug":"google-cloud-speech-to-text","name":"Cloud Speech-to-Text","confirmed":true}],"declaredAlternatives":[],"stars":15866,"downloads":38716831,"downloadsWeek":535808,"downloadsDays":6,"growth":124,"growthDays":6,"history":[{"date":"2026-11-04","stars":15742},{"date":"2026-11-05","stars":15763},{"date":"2026-11-06","stars":15793},{"date":"2026-11-07","stars":15813},{"date":"2026-11-08","stars":15826},{"date":"2026-11-09","stars":15844},{"date":"2026-11-10","stars":15866}],"firstPublishedAt":"2026-11-04T01:36:50.492Z","commitSha":"111617b114bc12c895a202cdc5ee2db3403d53f2","verifiedAt":"2026-11-04T01:22:44.421Z","lastCommitAt":"2026-11-10T07:45:42Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns speech into text and text into speech on your own computer or device, with no internet connection needed.","summary":"sherpa-onnx is a speech toolkit that runs on your own machine instead of a cloud service. It can transcribe speech, read text out loud, clone a voice, tell speakers apart, detect when someone is talking, add punctuation, clean up noisy audio and separate voice from music. It is mostly for people who build programs: there are APIs for 12 languages, from C++ and Python to Java and Rust, plus Android, iOS and web builds. For a quick test there are ready Windows programs and browser demos, and you download the speech models separately.","easeRationale":"The easiest route is the browser: the README points to hosted demo pages where, in its words, \"All you need is a browser\". On Windows the releases include ready programs such as sherpa-onnx-non-streaming-tts-x64-v2.0.1.exe and sherpa-onnx-streaming-asr-x64-v2.0.1.exe (19-22 MB), where the only extra step is downloading a speech model, so B. On Mac and Linux the releases are code libraries instead of finished programs, so you install Python or another language and run commands, which is a C.","costNote":"Apache-2.0 licence, no paid tier and no account.","notVerified":["How good the transcription and the synthetic voices sound in your own language, since quality depends on which model you download","How much disk space you end up using after downloading models, which is not part of the 22-66 MB program files","Whether the ready Windows programs work without extra Windows components","Whether every listed board, phone system and NPU really works as the tables claim","Whether the hosted browser demos keep running, since they are on outside services"]},"zh-Hans":{"summaryShort":"在你自己的电脑或设备上把语音转成文字、把文字转成语音，不需要联网。","summary":"sherpa-onnx是一个语音工具包，运行在你自己的机器上，而不是依赖云服务。它可以转录语音、朗读文本、克隆声音、区分说话人、检测是否有人在说话、添加标点、清理带噪音的音频，以及把人声和音乐分开。它主要面向开发程序的人：提供12种语言的API，从C++和Python到Java和Rust，还有Android、iOS和网页版本。想快速体验的话，有现成的Windows程序和浏览器演示，语音模型要单独下载。","easeRationale":"最简单的方式是浏览器：README指向托管的演示页面，用它自己的话说，\"All you need is a browser\"。在Windows上，发行版包含现成的程序，比如sherpa-onnx-non-streaming-tts-x64-v2.0.1.exe和sherpa-onnx-streaming-asr-x64-v2.0.1.exe（19-22 MB），唯一的额外步骤是下载一个语音模型，所以是B级。在Mac和Linux上，发行版是代码库而不是成品程序，所以你需要安装Python或其他语言并运行命令，这是C级。","costNote":"Apache-2.0许可，没有付费档，不需要账号。","notVerified":["转录效果和合成声音在你自己的语言里听起来怎么样，这取决于你下载的模型","下载模型之后实际占用多少磁盘空间，这部分不包含在22-66 MB的程序文件里","现成的Windows程序是否需要额外的Windows组件才能运行","列出的每块开发板、电话系统和NPU是否真的像表格里说的那样可用","托管在浏览器里的演示是否能一直运行，因为它们依赖外部服务"]}}},{"id":"R_kgDORkbrOA","slug":"katzca__autosocial","name":"AutoSocial","owner":"Katzca","fullName":"Katzca/AutoSocial","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"pipeline_for_reposting_other_peoples_videos","severity":"high","evidence":"README.md的功能列表：\"Download recent TikTok videos with yt-dlp and fan them out into queues\"，\"Scan/download TikTok profiles into autodownload/profile_downloads\"，以及\"Run FFmpeg-based video uniquification from the dashboard or CLI, with an optional user-provided logo overlay\"。.env.example用WATCH_CHANNEL、WATCH_MAX_VIDEOS、WATCH_MIN_VIEWS和AUTO_POST_PLATFORMS=tiktok,instagram,youtube把这些串联起来。这条链条会拿走别人的频道内容，修改每份副本（加上片头、logo、结尾停顿），然后自动发布到三个平台。README.md自己的Responsible Use部分说这个项目\"does not grant permission to post content you do not own or have permission to use\"，并要求用户\"avoid spam, deceptive behavior, unauthorized scraping, and posting without consent\"。","source":"ai","title":"转发他人视频的流水线"},{"code":"automates_sites_against_their_rules","severity":"attention","evidence":"它用Playwright直接操控真实的上传页面——.env.example设置了TIKTOK_UPLOAD_URL=https://www.tiktok.com/tiktokstudio/upload、INSTAGRAM_UPLOAD_URL和YOUTUBE_UPLOAD_URL，还有HEADLESS和像POST_DELAY_MS=15000这样的延迟设置——而不是使用平台自己的发布API。README.md：\"Prefer official platform APIs where they are available for your workflow\"以及\"review platform limits before enabling scheduled automation\"。账号可能因此被限制或封禁。","source":"ai","title":"违反规则自动操作网站"},{"code":"keeps_logged_in_sessions_in_folders","severity":"attention","evidence":"README.md：\"Persist Playwright login sessions under `.profiles/<account>/<platform>`\"以及\"Sessions are stored on disk and reused between runs.\" 任何复制了这些文件夹的人——或者README警告过的、不小心公开发布的fork——都能在没有密码的情况下进入你的社交账号。","source":"ai","title":"把已登录的会话保存在文件夹里"},{"code":"dashboard_without_any_login","severity":"attention","evidence":"README.md的Dashboard Security部分：\"The dashboard binds to 127.0.0.1 by default and has no authentication layer. Keep it local.\" 设置DASHBOARD_ALLOW_REMOTE=true会把它开放出去，README写道\"anyone who can reach the dashboard can operate the local automation controls\"。","source":"ai","title":"控制面板没有任何登录验证"},{"code":"very_early_and_paused","severity":"info","evidence":"package.json标注的版本是0.1.0，并写着\"private\": true，GitHub上完全没有发行版，确定性检查记录的最近一次提交是在2026-06-10，也就是检查时的146天前。README.md还提到测试覆盖的是\"local logic that does not require live social platform access\"，所以发布流程本身并没有在CI里被测试过。","source":"ai","title":"处于非常早期且已暂停"},{"code":"you_supply_an_external_downloader","severity":"info","evidence":"README.md的Requirements部分：\"Optional: `yt-dlp.exe` in `autodownload/` for downloader features\"，以及\"Windows is the primary target for the bundled yt-dlp.exe workflow\"。你需要自己找到并放置那个程序，下载功能才能正常工作。","source":"ai","title":"你需要自己提供外部下载工具"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg"],"buildTools":["node","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation","video"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"later","name":"Later","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true}],"declaredAlternatives":[],"stars":1483,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":22,"growthDays":6,"history":[{"date":"2026-11-04","stars":1461},{"date":"2026-11-05","stars":1464},{"date":"2026-11-06","stars":1475},{"date":"2026-11-07","stars":1479},{"date":"2026-11-08","stars":1481},{"date":"2026-11-09","stars":1485},{"date":"2026-11-10","stars":1483}],"firstPublishedAt":"2026-11-04T01:36:50.499Z","commitSha":"6deb560ee58ea1e6a040e1ee8e6ca269bd1576b5","verifiedAt":"2026-11-04T01:31:00.705Z","lastCommitAt":"2026-06-10T13:28:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A local dashboard that posts short videos to TikTok, Instagram and YouTube on a schedule by driving your browser.","summary":"AutoSocial Studio is a control panel you run on your own computer to publish short videos to TikTok, Instagram and YouTube. You drop video files into folders, write captions in text files, and a hidden browser logs into each site with your saved session and uploads them on a schedule. It also downloads TikTok videos with yt-dlp and can alter copies of a video with FFmpeg. It keeps your logins on your machine instead of a paid service. You need Node.js, FFmpeg and the terminal; there is no login on the dashboard and no finished release yet.","easeRationale":"Windows is the platform the author targets, but it is still D there and on macOS and Linux: there is no release to download, so you clone the repository and run `npm ci`, `npx playwright install chromium` and `npm run doctor`, put FFmpeg and ffprobe on your PATH, copy .env.example to .env, start the dashboard from the terminal and then log in to each social account by hand. The downloader feature additionally expects you to place yt-dlp.exe in the project folder.","costNote":"Free, MIT licensed, with no paid tier and no service fees.","notVerified":["Whether the upload flows still work, since the platforms' pages change and the last commit is months old","Whether TikTok, Instagram or YouTube detect this kind of automation and act on the account","How well the scheduler behaves when an upload fails halfway","Real disk use once video queues, downloads and browser profiles pile up","Whether it runs smoothly on macOS and Linux, which the author does not target"]},"zh-Hans":{"summaryShort":"一个本地控制面板，通过操控你的浏览器，按计划把短视频发布到TikTok、Instagram和YouTube。","summary":"AutoSocial Studio是一个在你自己电脑上运行的控制面板，用来把短视频发布到TikTok、Instagram和YouTube。你把视频文件放进文件夹，在文本文件里写说明文字，一个隐藏的浏览器会用你保存的会话登录每个网站，并按计划上传。它还能用yt-dlp下载TikTok视频，并用FFmpeg修改视频副本。它把你的登录信息留在自己的机器上，而不是依赖付费服务。你需要Node.js、FFmpeg和终端；控制面板本身没有登录机制，目前也还没有正式发行版。","easeRationale":"Windows是作者主要针对的平台，但在Windows、macOS和Linux上都是D级：没有可下载的发行版，你需要克隆仓库，运行`npm ci`、`npx playwright install chromium`和`npm run doctor`，把FFmpeg和ffprobe加入PATH，把.env.example复制为.env，从终端启动面板，再手动登录每个社交账号。下载功能还要求你把yt-dlp.exe放进项目文件夹。","costNote":"免费，MIT许可，没有付费档，也没有服务费。","notVerified":["由于平台页面会变化，而最近一次提交已经是几个月前，上传流程是否仍然有效","TikTok、Instagram或YouTube是否会检测到这类自动化行为并对账号采取措施","当上传中途失败时，调度器的表现如何","视频队列、下载内容和浏览器配置堆积起来后的实际磁盘占用","它在作者没有针对的macOS和Linux上是否能顺畅运行"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0MjcyMzI2OA==","slug":"kde__kdenlive","name":"Kdenlive","owner":"KDE","fullName":"KDE/kdenlive","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":129,"maxMB":388,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"premiere-pro","name":"Premiere Pro","confirmed":true},{"slug":"final-cut-pro","name":"Final Cut Pro","confirmed":true}],"declaredAlternatives":[],"stars":6084,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":42,"growthDays":7,"history":[{"date":"2026-10-25","stars":5934},{"date":"2026-10-26","stars":5938},{"date":"2026-10-27","stars":5967},{"date":"2026-10-28","stars":5987},{"date":"2026-10-29","stars":5999},{"date":"2026-10-30","stars":6008},{"date":"2026-10-31","stars":6015},{"date":"2026-11-01","stars":6026},{"date":"2026-11-02","stars":6038},{"date":"2026-11-03","stars":6042},{"date":"2026-11-04","stars":6046},{"date":"2026-11-05","stars":6052},{"date":"2026-11-06","stars":6056},{"date":"2026-11-07","stars":6066},{"date":"2026-11-08","stars":6073},{"date":"2026-11-09","stars":6078},{"date":"2026-11-10","stars":6084}],"firstPublishedAt":"2026-10-25T23:19:08.130Z","commitSha":"583aca73aafeec9043fb23ee4cd3fd3b91821c36","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T16:07:31Z","isMirror":true,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Edit videos on a timeline: cut clips, add titles, effects, transitions and music, then export the finished film.","summary":"Kdenlive is a free video editor from the KDE community. You place clips on a timeline, trim and arrange them, add titles, transitions, video and audio effects, and export the result in common video formats. It suits anyone from family-video makers to people doing longer projects. On Windows you download the installer from the official site and run it. This GitHub page is a read-only copy; the main project lives on KDE's own GitLab.","easeRationale":"An official Windows installer exists (kdenlive-26.08.2.exe, about 129 MB, listed in winget as KDE.Kdenlive) that you download and run.","notVerified":["Whether the Windows installer (version 26.08.2) was built from this exact code","How smoothly it edits and exports high-resolution video on your computer","Real disk use once you add projects, proxy clips and render files"]},"zh-Hans":{"summaryShort":"在时间线上剪辑视频：裁剪片段，添加标题、特效、转场和音乐，然后导出成片。","summary":"Kdenlive 是 KDE 社区推出的免费视频编辑器。你把片段放到时间线上，进行裁剪和排列，添加标题、转场、视频和音频特效，并以常见视频格式导出成品。无论是制作家庭视频的人，还是做较长项目的人，都可以使用。在 Windows 上，你从官方网站下载安装程序并运行即可。这个 GitHub 页面是只读副本；主项目托管在 KDE 自己的 GitLab 上。","easeRationale":"官方提供 Windows 安装程序（kdenlive-26.08.2.exe，约 129 MB，在 winget 中名为 KDE.Kdenlive），下载并运行即可。","notVerified":["Windows 安装程序（版本 26.08.2）是否由这份代码构建而成","在你的电脑上编辑和导出高分辨率视频是否流畅","添加项目、代理片段和渲染文件后实际占用的磁盘空间"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0Mzk3NTY3Mw==","slug":"kde__krita","name":"Krita","owner":"KDE","fullName":"KDE/krita","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":158,"maxMB":473,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"photoshop","name":"Photoshop","confirmed":true},{"slug":"clip-studio-paint","name":"Clip Studio Paint","confirmed":true},{"slug":"corel-painter","name":"Painter","confirmed":true}],"declaredAlternatives":[],"stars":10725,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":39,"growthDays":7,"history":[{"date":"2026-10-25","stars":10634},{"date":"2026-10-26","stars":10638},{"date":"2026-10-27","stars":10652},{"date":"2026-10-28","stars":10655},{"date":"2026-10-29","stars":10664},{"date":"2026-10-30","stars":10669},{"date":"2026-10-31","stars":10671},{"date":"2026-11-01","stars":10676},{"date":"2026-11-02","stars":10682},{"date":"2026-11-03","stars":10686},{"date":"2026-11-04","stars":10694},{"date":"2026-11-05","stars":10699},{"date":"2026-11-06","stars":10703},{"date":"2026-11-07","stars":10710},{"date":"2026-11-08","stars":10717},{"date":"2026-11-09","stars":10722},{"date":"2026-11-10","stars":10725}],"firstPublishedAt":"2026-10-25T23:19:09.035Z","commitSha":"51beeb564a010b3d0d03c3db0c68d3b3b1c2cbd1","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T15:06:43Z","isMirror":true,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Draw and paint digital art, comics and illustrations with brushes, layers and drawing-tablet support.","summary":"Krita is a free digital painting program from the KDE community. It is made for artists: comic artists, illustrators, concept artists and texture painters use it to create artwork from a blank canvas to the finished file. This GitHub page is only a copy of the code; the real project lives on KDE's own servers and the program is downloaded from krita.org. Radar's checks found a Windows installer.","easeRationale":"The checks found a Windows installer (about 158–473 MB on disk), so you download it and open it like any other program.","costNote":"GNU GPL version 3.","notVerified":["How well your specific drawing tablet and pen pressure work with it","Performance with very large canvases on your computer","The exact installer on krita.org, since this GitHub mirror has no release files of its own"]},"zh-Hans":{"summaryShort":"用画笔、图层并支持数位板，绘制数字艺术、漫画和插画。","summary":"Krita 是 KDE 社区出品的免费数字绘画程序。它为艺术家而做：漫画家、插画师、概念艺术家和贴图绘制师用它从空白画布一直创作到成品文件。这个 GitHub 页面只是代码的副本；真正的项目位于 KDE 自己的服务器上，程序从 krita.org 下载。Radar 的检查发现了 Windows 安装程序。","easeRationale":"检查发现了 Windows 安装程序（磁盘上约 158–473 MB），下载后像打开其他程序一样打开即可。","costNote":"GNU GPL 第 3 版。","notVerified":["你的数位板和压感笔在其中的表现如何","在你的电脑上处理超大画布时的性能","krita.org 上的具体安装程序，因为这个 GitHub 镜像没有自己的发布文件"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0MjczMTM2Ng==","slug":"kde__okular","name":"Okular","owner":"KDE","fullName":"KDE/okular","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"license_files_in_licenses_folder","severity":"info","evidence":"顶层文件列表中有一个 LICENSES/ 文件夹，所以许可证文本看起来是按文件分别存放的，尽管 GitHub 没有检测到单一的许可证文件。","source":"ai","title":"许可证文件放在 LICENSES 文件夹中"}],"platforms":["windows","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0-only","category":"copyleft","detectedFrom":"reuse"},"categories":["office","productivity"],"alternativeTo":[{"slug":"acrobat","name":"Acrobat","confirmed":true}],"declaredAlternatives":[],"stars":1571,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":7,"growthDays":7,"history":[{"date":"2026-10-25","stars":1550},{"date":"2026-10-26","stars":1551},{"date":"2026-10-27","stars":1552},{"date":"2026-10-28","stars":1553},{"date":"2026-10-29","stars":1555},{"date":"2026-10-30","stars":1555},{"date":"2026-10-31","stars":1555},{"date":"2026-11-01","stars":1560},{"date":"2026-11-02","stars":1563},{"date":"2026-11-03","stars":1564},{"date":"2026-11-04","stars":1565},{"date":"2026-11-05","stars":1565},{"date":"2026-11-06","stars":1565},{"date":"2026-11-07","stars":1566},{"date":"2026-11-08","stars":1566},{"date":"2026-11-09","stars":1570},{"date":"2026-11-10","stars":1571}],"firstPublishedAt":"2026-10-25T23:32:49.542Z","commitSha":"896b2fe4d8a113f58a92a1e26b3e42e2ce33ef12","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T19:23:15Z","isMirror":true,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Open and read PDFs, comic books, PostScript files and images, and add notes and highlights to PDFs.","summary":"Okular is the KDE document viewer. It opens PDF, PostScript, comic book files and several image formats, and lets you add notes and highlights that are saved as standard PDF annotations. It is for anyone who reads or marks up documents on a computer. This GitHub repository is a read-only copy of KDE's own code server and has no downloads. KDE offers an official Windows installer (Windows Package Manager ID KDE.Okular, version 26.08.1), and the README sends users to okular.kde.org for downloads.","easeRationale":"KDE publishes an official Windows installer (an .exe listed in winget as KDE.Okular 26.08.1) that you download and open; the CMake build steps in the README are only for developers.","costNote":"Free KDE software; no paid tier is mentioned.","notVerified":["Download size and disk space of the Windows installer, since the winget lookup did not report a size.","Whether every document format listed in the README opens in the Windows version.","Whether a macOS version is available.","Which license applies to which part, since GitHub did not detect a single license file.","How well annotations made in Okular show up in other PDF readers."]},"zh-Hans":{"summaryShort":"打开并阅读 PDF、漫画、PostScript 文件和图片，还能在 PDF 上添加注释和高亮。","summary":"Okular 是 KDE 的文档查看器。它可以打开 PDF、PostScript、漫画文件和多种图片格式，并能添加注释和高亮，保存为标准的 PDF 批注。适合所有在电脑上阅读或标注文档的人。这个 GitHub 仓库是 KDE 自有代码服务器的只读副本，没有下载文件。KDE 提供官方 Windows 安装程序（Windows 程序包管理器 ID 为 KDE.Okular，版本 26.08.1），README 也引导用户到 okular.kde.org 下载。","easeRationale":"KDE 发布了官方 Windows 安装程序（在 winget 中列为 KDE.Okular 26.08.1 的 .exe），下载后打开即可；README 里的 CMake 编译步骤只面向开发者。","costNote":"KDE 的免费软件；未提到任何付费版本。","notVerified":["Windows 安装程序的下载大小和磁盘占用，因为 winget 查询没有提供大小。","README 列出的所有文档格式能否在 Windows 版中打开。","是否有 macOS 版本。","各部分分别适用哪种许可证，因为 GitHub 没有检测到单一的许可证文件。","在 Okular 中做的批注在其他 PDF 阅读器里显示得如何。"]}}},{"id":"R_kgDOKhTjZg","slug":"kedreamix__linly-talker","name":"Linly-Talker","owner":"Kedreamix","fullName":"Kedreamix/Linly-Talker","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2024-04-23(925 天前)","source":"check","title":"版本更新停滞"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):joblib: GHSA-6hrg-qmvc-2xh8, PYSEC-2022-288; gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-37qc-qgx6-9xjv; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-37q5-v5qm-c9v8; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8qvm-5x2c-j2w7, PYSEC-2026-1805; gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2; scikit_learn: GHSA-jw8x-6495-233v, PYSEC-2024-110; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8qvm-5x2c-j2w7, PYSEC-2026-1805; basicsr: GHSA-86w8-vhw6-q9qq, PYSEC-2026-1215; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; modelscope: GHSA-fhhq-h4hg-549x, PYSEC-2026-2663; lightning: GHSA-cgwc-qvrx-rf7f, GHSA-mr7h-w2qc-ffc2, GHSA-qqmf-gpg7-g8gw","source":"check","title":"已知漏洞"},{"code":"makes_a_real_person_say_things","severity":"attention","evidence":"该项目根据\"a one-minute voice sample\"克隆一个声音,并让任何上传的照片动起来;README 里对应章节的标题就是\"Voice Clone - Stealing Your Voice Quietly During Conversations\",ER-NeRF 部分还写道\"Taking inspiration from the likeness of Obama\"。这样使用别人的面孔或声音,可能违反法律和大多数平台的规则。","source":"ai","title":"让真实的人说出某些话"},{"code":"unofficial_access_to_paid_ai_services","severity":"attention","evidence":"两份依赖文件都安装了\"g4f\"(GPT4Free),README 的聊天模型列表里也有一项\"GPT4Free\"。这个库能在没有账号的情况下访问付费服务商的聊天服务,这违反了那些服务商的条款。","source":"ai","title":"非官方途径访问付费 ai 服务"},{"code":"models_from_file_sharing_links","severity":"attention","evidence":"模型文件和\"Windows All-in-One Package\"通过百度网盘(密码`linl`)、Quark 网盘(pan.quark.cn)和一个阿里云 Docker 镜像提供,此外还有 Hugging Face 和 ModelScope。这些网盘上的文件没有校验值,所以你无法确认自己运行的到底是什么。","source":"ai","title":"模型来自文件分享链接"},{"code":"development_moved_on","severity":"info","evidence":"2026.02 的更新指向了另一个仓库 Linly-Talker-Stream,称之为\"A Brand New Real-time Streaming Interactive Architecture\";这里最近一次提交是在 2026-02-10,唯一的发布版本是 2024 年 4 月的 v1.0。","source":"ai","title":"开发重心已经转移"},{"code":"one_voice_feature_already_broken","severity":"info","evidence":"README 里关于 Edge TTS 的提醒:\"it seems that Microsoft has restricted certain IPs... The solution is still unstable for now\",并建议改用 CosyVoice。","source":"ai","title":"一个语音功能已经失效"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":1010,"maxMB":4570,"basis":"heuristic"},"apiKeys":["Google Gemini (only if you pick GeminiPro as the chat model)","OpenAI (only if you pick ChatGPT as the chat model)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","audio","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true}],"declaredAlternatives":[],"stars":3481,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":2,"growthDays":6,"history":[{"date":"2026-11-04","stars":3479},{"date":"2026-11-05","stars":3480},{"date":"2026-11-06","stars":3479},{"date":"2026-11-07","stars":3480},{"date":"2026-11-08","stars":3481},{"date":"2026-11-09","stars":3481},{"date":"2026-11-10","stars":3481}],"firstPublishedAt":"2026-11-04T22:32:55.023Z","commitSha":"dc831b3eea1202728c45042dd92de73ade8c3abd","verifiedAt":"2026-11-04T22:23:17.608Z","lastCommitAt":"2026-02-10T05:17:02Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a photo into a talking avatar that listens, answers with an AI and speaks in a cloned voice.","summary":"Linly-Talker is a web page you run on your own computer that puts several open models together: you upload a photo, speak or type, an AI model answers, text-to-speech reads the answer and a lip-sync model (SadTalker, Wav2Lip, MuseTalk or ER-NeRF) makes the picture talk. It can also copy a voice from about a minute of recorded speech. It is for people comfortable with the terminal: Python, an NVIDIA card, ffmpeg and several gigabytes of model files. Free, MIT licensed, and the author has moved on to a newer repository.","easeRationale":"Both platforms are terminal work with an NVIDIA card: a conda environment, PyTorch matched to your CUDA version, ffmpeg, several `mim install` steps, pytorch3d from source for the NeRF part, then gigabytes of model files to download and move into the right folders. The README itself warns \"The installation process is very slow\". Windows is slightly helped by an \"all-in-one package\", but it is still a sequence of conda commands, so both stay at D.","costNote":"MIT licence, no payment. The chat part can use a free local model (Qwen, Linly) or an online one such as Gemini or ChatGPT, which you pay for.","notVerified":["Whether the long install still works, given pinned packages from 2024 and dependencies pulled from many sources","How much video memory each avatar model really needs and how long one video takes","Real disk use after downloading the chat, voice and avatar models","Whether the Colab notebook in the repository still runs","What the Windows all-in-one package on the Quark drive contains"]},"zh-Hans":{"summaryShort":"把一张照片变成会说话的虚拟形象:能听、能用 AI 回答,还能用克隆的声音说出来。","summary":"Linly-Talker 是一个你在自己电脑上运行的网页,把好几个开源模型组合在一起:上传一张照片,说话或打字,一个 AI 模型给出回答,文字转语音读出这个回答,再由一个唇形同步模型(SadTalker、Wav2Lip、MuseTalk 或 ER-NeRF)让照片\"开口说话\"。它还能根据大约一分钟的录音复刻一个声音。它适合习惯使用终端的人:需要 Python、一块 NVIDIA 显卡、ffmpeg,以及好几个 GB 的模型文件。免费,采用 MIT 许可证,作者已经转向了一个更新的仓库。","easeRationale":"两个平台都需要带 NVIDIA 显卡的终端操作:配置 conda 环境、安装与你的 CUDA 版本匹配的 PyTorch、ffmpeg、好几步 `mim install`、为 NeRF 部分从源码编译 pytorch3d,然后下载好几个 GB 的模型文件并放到正确的文件夹里。README 自己也提醒\"The installation process is very slow\"。Windows 有一个\"一体化安装包\"能稍微帮点忙,但仍然要走一串 conda 命令,所以两个平台都是 D 级。","costNote":"MIT 许可证,不需要付费。聊天部分可以用免费的本地模型(Qwen、Linly),也可以用 Gemini 或 ChatGPT 这样的在线模型,但那些要付费。","notVerified":["鉴于依赖包的版本固定在 2024 年,且来自很多不同的来源,这个冗长的安装过程是否还能跑通","每个虚拟形象模型实际需要多少显存,生成一段视频要花多长时间","下载完聊天、语音和虚拟形象模型之后,真实占用的磁盘空间","仓库里的 Colab notebook 是否还能运行","Quark 网盘上的 Windows 一体化安装包里到底有什么内容"]}}},{"id":"R_kgDOTGon8Q","slug":"kenkaiii__oh-daddy","name":"Oh Daddy","owner":"KenKaiii","fullName":"KenKaiii/oh-daddy","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"web":"E"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"存在已知漏洞的固定依赖项(OSV):next: GHSA-2464-fhr4-w9xw, GHSA-2xp9-vwfh-vxw4, GHSA-39w2-rjm5-chcv","source":"check","title":"已知漏洞"},{"code":"author_referral_link_in_setup","severity":"info","evidence":"部署说明里的托管链接带有作者自己的推荐码:\"a Railway account ([create one here](https://railway.com?referralCode=Q2XTYD))\",所以作者可能会从你的注册中获利。","source":"ai","title":"部署说明中有作者的推荐链接"},{"code":"setup_needs_ai_coding_tool","severity":"info","evidence":"唯一有文档记录的部署方式是运行一个AI代理命令:\"1. Fork this repo. 2. Open it in Claude Code / GG Coder. 3. Run /setup-railway.\"仓库里有.gg/、AGENTS.md和CLAUDE.md文件;README里没有一步步的手动说明。","source":"ai","title":"设置过程需要AI编程工具"},{"code":"very_young_project","severity":"info","evidence":"创建于2026-06-27,最后一次提交在2026-07-04,8个星标,没有release,package.json写着\"version\": \"0.1.0\"和\"private\": true。","source":"ai","title":"项目非常年轻"},{"code":"acts_on_your_social_account","severity":"info","evidence":"它会以你的身份回复并发送私信:\"they get a public reply right away plus a DM straight to their inbox. No human involved\",使用Meta OAuth和\"Meta's Private Replies API\"。Meta限制自动化消息发送,README设定了\"one hard cap of 195 comments/hour, just under Meta's Graph API ceiling\"。","source":"ai","title":"代表你的社交账号采取行动"}],"platforms":["web"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta (Instagram and Facebook Graph API)"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":8,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":8},{"date":"2026-11-09","stars":8},{"date":"2026-11-10","stars":8}],"firstPublishedAt":"2026-11-07T22:56:16.817Z","commitSha":"5b5f1eb967451f152d6e916b141a67dd4b0e452e","verifiedAt":"2026-11-07T22:40:51.151Z","lastCommitAt":"2026-11-09T22:39:29Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted tool that answers a keyword comment on Instagram or Facebook with a public reply and a DM.","summary":"Oh Daddy does one job: when someone comments a keyword under your Instagram or Facebook post, it posts a public reply and sends that person a DM through Meta's official API, with a 24-hour limit per person so nobody gets the same answer twice. It has four pages (dashboard, automations, accounts, settings) and no chatbot builder. You run it yourself: fork the repository, deploy it to Railway with the included command, and connect your Meta account. The author compares it to ManyChat at about $5 a month of hosting.","easeRationale":"It is a web app you host yourself, and that is the only way to use it: the README asks you to fork the repository, open it in an AI coding assistant (Claude Code or GG Coder), run a /setup-railway command with the Railway CLI logged in, and then add Meta credentials in a wizard, which means creating an app in Meta's developer portal. There is no installer and no hosted version to sign up for.","costNote":"The code is free (AGPL-3.0). You pay for your own server: the README estimates \"~$5/month total, your own hosting\" on Railway, including the Postgres database.","notVerified":["Whether the Railway setup command really works end to end and what the server costs in a busy month","How hard it is to get the Meta app permissions approved for Instagram comments and private replies","Whether the per-person cooldown and the hourly cap behave as described under real traffic","Whether Meta keeps allowing this kind of automatic reply on your account","Whether the stored Meta tokens are encrypted as the README describes"]},"zh-Hans":{"summaryShort":"自托管工具,对Instagram或Facebook上带关键词的评论自动公开回复并发送私信。","summary":"Oh Daddy只做一件事:当有人在你的Instagram或Facebook帖子下评论一个关键词时,它会发布一条公开回复,并通过Meta官方API给这个人发一条私信,每人每24小时限制一次,这样没人会收到两次相同的回复。它有四个页面(仪表盘、自动化、账号、设置),没有聊天机器人搭建器。你需要自己运行它:fork这个仓库,用自带的命令部署到Railway,然后连接你的Meta账号。作者把它和ManyChat相比,托管费用大约每月$5。","easeRationale":"这是一个你自己托管的网页应用,而且这是唯一的使用方式:README要求你fork这个仓库,在AI编程助手(Claude Code或GG Coder)里打开它,在已登录的Railway CLI中运行/setup-railway命令,然后在一个向导里添加Meta凭据,这意味着要在Meta开发者门户里创建一个应用。没有安装程序,也没有可以直接注册使用的托管版本。","costNote":"代码是免费的(AGPL-3.0)。你要为自己的服务器付费:README估计在Railway上\"~$5/month total, your own hosting\",包括Postgres数据库。","notVerified":["Railway的设置命令是否真的能从头到尾顺利跑通,以及在繁忙的一个月里服务器要花多少钱。","要让Meta批准Instagram评论和私信回复所需的应用权限有多难。","每人冷却时间和每小时上限在真实流量下是否如描述那样运作。","Meta是否会一直允许你的账号使用这种自动回复。","保存的Meta令牌是否像README描述的那样被加密。"]}}},{"id":"R_kgDOTXWCMg","slug":"kevinbadi__social-agents","name":"Social Agents","owner":"kevinbadi","fullName":"kevinbadi/social-agents","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"C","linux":"D"},"risk":"attention","signals":[{"code":"useless_without_a_paid_service","severity":"attention","evidence":"README:\"CreatorOS is the service underneath: it holds your connected socials and does the actual posting, replying, and analytics\";客户端只通过 cos_live_ 密钥与 CreatorOS API 通信,管理密钥或套餐的请求会被拒绝,提示为\"Manage your plan and API keys in the CreatorOS app\"","source":"ai","title":"没有付费服务就毫无用处"},{"code":"unattended_agent_can_run_up_api_bills","severity":"attention","evidence":"README,Railway 部署路径:\"set a spend limit in the Anthropic Console (console.anthropic.com → Billing → Limits) before deploying. The service runs an agent unattended; an uncapped key is an uncapped bill\"","source":"ai","title":"无人值守的代理可能让 api 账单失控"},{"code":"agent_acts_on_your_accounts_on_its_own","severity":"attention","evidence":"初始的定时任务会发帖、\"triage comments & DMs, reply on-brand\",并按计划启动评论转私信漏斗;文档所述的最终状态是\"fully autonomous, with you only reviewing what Social Agents surface\"","source":"ai","title":"代理会自行操作你的账号"},{"code":"agent_follows_instructions_from_external_videos","severity":"attention","evidence":"README:在构建新的自动化模式之前,代理会\"checks the index, fetches the tutorial, and follows the taught pattern\",内容来自 social-agents/knowledge/TUTORIALS.md,这是作者的 YouTube 教程列表,因此仓库之外的内容也能左右代理的行为","source":"ai","title":"代理会听从外部视频中的指示"},{"code":"installs_dependencies_on_first_start","severity":"info","evidence":"快速上手部分写道\"No npm install needed — first start installs for you\",package.json 会在 start、dashboard、worker 和 test 之前运行 scripts/ensure-deps.mjs","source":"ai","title":"首次启动时自动安装依赖"},{"code":"early_version_no_releases","severity":"info","evidence":"package.json 版本号为 0.1.0,没有 GitHub releases,98 颗星,仓库创建于 2026-07-13","source":"ai","title":"早期版本,没有发行版"}],"platforms":["windows","mac","linux"],"tools":["node","ffmpeg"],"buildTools":["node","ffmpeg"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["CreatorOS","Anthropic (or an Anthropic-compatible model)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","ai_agents","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"manychat","name":"ManyChat","confirmed":true}],"declaredAlternatives":[],"stars":100,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":2,"growthDays":6,"history":[{"date":"2026-11-04","stars":98},{"date":"2026-11-05","stars":98},{"date":"2026-11-06","stars":98},{"date":"2026-11-07","stars":100},{"date":"2026-11-08","stars":100},{"date":"2026-11-09","stars":100},{"date":"2026-11-10","stars":100}],"firstPublishedAt":"2026-11-05T22:43:48.537Z","commitSha":"7f9e1ce87f184274521d7d1fb216f663133907b6","verifiedAt":"2026-11-05T22:30:13.316Z","lastCommitAt":"2026-09-25T20:50:45Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"AI agents that post to your social accounts, reply to comments and DMs and report results, driven from your computer.","summary":"Social Agents is a set of AI agents you run on your own computer to look after your social accounts: they interview you about your brand, write and schedule posts (short video, long video, carousels, threads), reply to comments and messages, run comment-to-DM funnels and send you a weekly report. There is also a local dashboard showing what they did. It does not talk to Instagram or TikTok itself: everything goes through CreatorOS, a separate online service where you connect the accounts and get a paid API key. You also need Node.js 20+ and either a Claude plan or an Anthropic key.","easeRationale":"macOS gets the best grade: with Node.js 20 or newer you clone the repository and run `npm start creatoros social-agents`, which installs the dependencies and asks for your CreatorOS key, and scheduled jobs can run locally as launchd services. On Windows and Linux that local option does not exist — the README's other path is deploying a Railway worker with your CreatorOS and Anthropic keys — and the drop-a-video skill also needs ffmpeg and whisper installed.","costNote":"The code is MIT, but it only works with a CreatorOS account and its cos_live_ API key, plus AI usage on a Claude plan or an Anthropic key; neither price is stated in the repository","notVerified":["What CreatorOS costs and whether its account connections work for a given network","How much the AI usage costs per month when the agents run on a schedule","Whether the code-level endpoint allowlist really blocks everything outside the documented table","What the agents actually publish when running unattended, and how often mistakes need correcting","Whether the whole flow works on Windows and Linux, since the local schedule option is macOS-only"]},"zh-Hans":{"summaryShort":"AI 代理会在你的社交账号上发帖、回复评论和私信,并汇报结果,一切由你自己的电脑驱动。","summary":"Social Agents 是一套在你自己的电脑上运行的 AI 代理,用来打理你的社交账号:它们会就你的品牌对你进行访谈,撰写并安排发帖(短视频、长视频、轮播图、话题串),回复评论和私信,运行\"评论转私信\"转化漏斗,并给你发送每周报告。还有一个本地仪表盘显示它们做了什么。它本身不会直接连接 Instagram 或 TikTok:一切都通过 CreatorOS 进行,这是一个独立的在线服务,你在其中连接账号并获取付费 API 密钥。你还需要 Node.js 20 及以上版本,以及 Claude 套餐或 Anthropic 密钥。","easeRationale":"macOS 获得最高评分:使用 Node.js 20 或更高版本,克隆仓库并运行 `npm start creatoros social-agents`,它会安装依赖并要求输入你的 CreatorOS 密钥,定时任务可以作为 launchd 服务在本地运行。在 Windows 和 Linux 上没有这种本地方案——README 给出的另一条路径是部署一个带有你的 CreatorOS 和 Anthropic 密钥的 Railway worker——而\"丢视频\"功能还需要安装 ffmpeg 和 whisper。","costNote":"代码采用 MIT 许可,但只能配合 CreatorOS 账号及其 cos_live_ API 密钥使用,另外还需要 Claude 套餐或 Anthropic 密钥产生的 AI 使用费用;这两项价格仓库中都没有说明","notVerified":["CreatorOS 的收费是多少,以及它的账号连接对某个具体平台是否真的管用","当代理按计划定时运行时,每月的 AI 使用费用是多少","代码层面的端点白名单是否真的能拦住文档表格之外的一切","代理在无人值守运行时实际会发布什么内容,以及需要多频繁地纠正错误","整个流程在 Windows 和 Linux 上是否都能用,因为本地定时方案只支持 macOS"]}}},{"id":"R_kgDOVcv29w","slug":"kevinraymond__florilegium","name":"Florilegium","owner":"kevinraymond","fullName":"kevinraymond/florilegium","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"C"},"risk":"low","signals":[{"code":"brand_new_project","severity":"info","evidence":"GitHub 信息：创建于 2026-10-30，两个版本（v1.0 和 v1.1）都在同一天打了标签。目前还没有它能在不同 TeX 发行版上正常工作的记录。","source":"ai","title":"全新的项目"},{"code":"authored_layout_tuned_for_one_page_size","severity":"info","evidence":"README.md 的表格里写道，手工排版的版本 \"adapts to\" \"nothing (tuned for A4, 68 mm margin)\"，只有自动生成的版本才会适配页面布局。在其他纸张尺寸上，手工摆放的版本可能放不下。","source":"ai","title":"手工排版只适配了一种页面尺寸"}],"platforms":["windows","mac","linux","web"],"tools":["other","ffmpeg"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","design_cad"],"alternativeTo":[],"declaredAlternatives":[],"stars":1619,"downloads":900,"downloadsWeek":486,"downloadsDays":7,"growth":211,"growthDays":7,"history":[{"date":"2026-10-31","stars":1158},{"date":"2026-11-01","stars":1299},{"date":"2026-11-02","stars":1369},{"date":"2026-11-03","stars":1408},{"date":"2026-11-04","stars":1425},{"date":"2026-11-05","stars":1442},{"date":"2026-11-06","stars":1548},{"date":"2026-11-07","stars":1569},{"date":"2026-11-08","stars":1588},{"date":"2026-11-09","stars":1606},{"date":"2026-11-10","stars":1619}],"firstPublishedAt":"2026-10-31T01:54:32.991Z","commitSha":"138f9f2603736eba8ea9014b33b381bdd5a503f5","verifiedAt":"2026-10-31T01:44:13.626Z","lastCommitAt":"2026-11-10T20:18:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A LaTeX add-on that paints a different blue floral border in the margin of every page of your document.","summary":"florilegium is a package for LuaLaTeX documents. It draws Delft-blue flower margins where every petal, leaf and stem is computed while the page is being typeset, so each page gets its own garden and the same page always comes out the same. The margin can react to the text: a decorated first letter, a flower beside a heading or wherever you ask, and a garden that ages from spring to autumn across the document. One colour setting changes the whole palette. There is also a pure PostScript version for Ghostscript or a printer.","easeRationale":"The easiest route is the browser: the README says you can paste florilegium.sty, the six .lua files and one .tex file into an Overleaf project and set the compiler to LuaLaTeX. On your own computer you need a LaTeX distribution with LuaLaTeX plus make and a terminal (or build the provided Docker image), and the animations and gallery images additionally need ffmpeg, Python with Pillow, pdftoppm or Ghostscript.","costNote":"Free and open source under MIT.","notVerified":["Whether it compiles without changes in your own TeX distribution and version.","The real time and memory per page on your machine, beyond the author's \"about a second a page\".","How it behaves with page sizes, margins and document classes other than the ones it was tuned for.","Whether a given printer handles the PostScript version, which the README says falls back to opaque colours."]},"zh-Hans":{"summaryShort":"一个 LaTeX 扩展包，在你文档每一页的页边上画出一幅不同的蓝色花卉图案。","summary":"florilegium 是一个用于 LuaLaTeX 文档的扩展包。它在页边绘制代尔夫特蓝色的花卉，每一片花瓣、叶子和花茎都是在排版页面时实时计算出来的，所以每一页都有自己独特的花园，而同一页永远会得到相同的结果。页边还能对正文内容做出反应：一个装饰性的首字母、标题旁边的一朵花，或者你指定的其他位置，整份文档中花园还会从春天一直老去到秋天。一个颜色设置就能改变整套配色。另外还有一个纯 PostScript 版本，适用于 Ghostscript 或打印机。","easeRationale":"最简单的方式是在浏览器里用：README 说你可以把 florilegium.sty、六个 .lua 文件和一个 .tex 文件粘贴进一个 Overleaf 项目，把编译器设为 LuaLaTeX。在自己的电脑上，你需要一个带 LuaLaTeX 的 LaTeX 发行版，外加 make 和一个终端（或者编译项目提供的 Docker 镜像），动画和图库的图片还需要 ffmpeg、带 Pillow 的 Python、pdftoppm 或 Ghostscript。","costNote":"免费，采用 MIT 许可的开源项目。","notVerified":["在你自己的 TeX 发行版和版本上，是否能不做任何修改就编译通过。","在你的机器上，每页实际花费的时间和内存，而不只是作者所说的\"about a second a page\"。","在不同于它所适配的页面尺寸、页边距和文档类型上，它的表现如何。","特定打印机是否能处理好这个 PostScript 版本，README 说它会退回到不透明的颜色。"]}}},{"id":"R_kgDOJuIwkQ","slug":"kevinwang676__bark-voice-cloning","name":"Bark Voice Cloning","owner":"KevinWang676","fullName":"KevinWang676/Bark-Voice-Cloning","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2","source":"check","title":"存在已知漏洞"},{"code":"voice_cloning_impersonation_risk","severity":"attention","evidence":"该界面可以从上传的录音中克隆声音,并替换现有音频中的声音;README 还宣传了一个\"AI song cover generator\"。仓库自带的 Disclaimer 中写道要\"obtain proper consent before cloning or converting any voice\"","source":"ai","title":"声音克隆存在冒充风险"},{"code":"docker_image_builds_a_different_project","severity":"attention","evidence":"Dockerfile 并没有使用本仓库的代码:它运行\"git clone https://github.com/C0untFloyd/bark-gui\",安装那个项目的依赖,并启动它的 webui.py","source":"ai","title":"docker 镜像构建的是另一个项目"},{"code":"installs_binary_wheel_from_personal_repository","severity":"attention","evidence":"Windows 上的 requirements.txt 会从一个个人 release 页面拉取预编译包:fairseq@https://github.com/Sharrnah/fairseq/releases/download/v0.12.4/fairseq-0.12.4-cp310-cp310-win_amd64.whl","source":"ai","title":"从个人仓库安装二进制 wheel 包"},{"code":"pinned_to_old_versions","severity":"attention","evidence":"requirements.txt 固定了 gradio==3.33.0 和 gradio_client==0.2.7,这两个版本是 2023 年 10 月为绕开一个错误而选定的,而检查发现那个 gradio 版本存在已知漏洞","source":"ai","title":"固定在旧版本上"},{"code":"hardcoded_colab_paths_in_local_code","severity":"info","evidence":"README,\"Important note for local runs\":\"The default path in app.py is set for Colab (/content/...). If you run locally, you may need to update that destination path\"","source":"ai","title":"本地代码中硬编码了 colab 的路径"},{"code":"downloads_models_on_first_run","severity":"info","evidence":"README:\"On first run, Bark checkpoints are downloaded into ./models/\" 以及 \"HuBERT + tokenizer for voice cloning are downloaded into ./models/hubert/\"","source":"ai","title":"首次运行时下载模型"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","docker","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["OpenAI (only for the OpenAI TTS notebooks)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true},{"slug":"kits-ai","name":"Kits AI","confirmed":true}],"declaredAlternatives":[],"stars":2944,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":2944},{"date":"2026-11-07","stars":2944},{"date":"2026-11-08","stars":2944},{"date":"2026-11-09","stars":2944},{"date":"2026-11-10","stars":2944}],"firstPublishedAt":"2026-11-05T22:43:48.545Z","commitSha":"54cdcf224e7cf557628ad4bdd05113c891635603","verifiedAt":"2026-11-05T22:31:03.908Z","lastCommitAt":"2026-05-31T03:45:07Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from a recording and uses it to read text out loud or to change the voice in another recording.","summary":"This repository collects ready-to-run voice cloning tools. The main part is a Bark web interface with three tabs: create a voice from an uploaded recording, read text with that voice, and swap the voice in an existing recording. A second interface, Sambert, covers Chinese voice cloning with labelling, training and use. The rest is a list of one-click Colab notebooks for other open models (GPT-SoVITS, XTTS, VALL-E X, F5-TTS, CosyVoice, OpenVoice and more). Running it on your own computer needs Python and, in practice, a graphics card; the notebooks run in the browser on Google's machines.","easeRationale":"The browser route earns the best grade: you open one of the Colab notebooks and run the setup cells, with nothing installed on your computer, though you need a Google account and the session's graphics card. On Windows, macOS and Linux you install Python 3.10 or newer and the requirements by hand, a graphics card is recommended \"CPU works but is slow\", and the README warns that the \"Create Voice\" path in app.py still points at the Colab folder /content/... and may need editing.","costNote":"The code is free (MIT); the Colab notebooks use Google's free tier, which limits how long a session with a graphics card lasts","notVerified":["Whether the 2023-era dependencies still install today, on Colab or locally","Whether the HuggingFace demos linked in the README are still online","Real disk use and video memory after the model files download","How close the cloned voice sounds to the original, in English and in Chinese","Whether the many third-party notebooks still work, since each depends on another project"]},"zh-Hans":{"summaryShort":"从一段录音中复制一个声音,用它朗读文字,或替换另一段录音中的声音。","summary":"这个仓库收集了一批可以直接使用的声音克隆工具。主要部分是一个带三个标签页的 Bark 网页界面:从上传的录音创建一个声音、用这个声音朗读文字、在现有录音中替换声音。第二个界面 Sambert 用于中文声音克隆,包含标注、训练和使用。其余部分是一份面向其他开放模型(GPT-SoVITS、XTTS、VALL-E X、F5-TTS、CosyVoice、OpenVoice 等)的一键 Colab 笔记本列表。在自己的电脑上运行需要 Python,实际上还需要显卡;这些笔记本在浏览器中运行,使用的是 Google 的机器。","easeRationale":"浏览器方案得到最高评分:你打开其中一个 Colab 笔记本并运行设置单元格,电脑上什么都不用安装,不过需要一个 Google 账号和该会话分配到的显卡。在 Windows、macOS 和 Linux 上,你需要手动安装 Python 3.10 及以上版本和各项依赖,官方建议使用显卡,\"CPU works but is slow\",而且 README 提醒 app.py 里的\"Create Voice\"路径仍然指向 Colab 的文件夹 /content/...,可能需要自己修改。","costNote":"代码是免费的(MIT);Colab 笔记本使用 Google 的免费额度,这会限制带显卡的会话能持续多久","notVerified":["2023 年那批依赖项现在是否还能在 Colab 或本地安装成功","README 中链接的 HuggingFace 演示是否还在线","模型文件下载完成后,实际占用的磁盘空间和显存","克隆出来的声音和原声有多像,包括英语和中文","众多第三方笔记本是否还能正常运行,因为每一个都依赖另一个项目"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzOTY1Njk1Mzg=","slug":"khoj-ai__khoj","name":"Khoj","owner":"khoj-ai","fullName":"khoj-ai/khoj","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"telemetry_on_by_default","severity":"attention","evidence":"docker-compose.yml 把关闭选项留成了注释状态:\"# Uncomment the line below to disable telemetry. Telemetry helps us prioritize feature development and understand how people are using Khoj\" / \"# - KHOJ_TELEMETRY_DISABLE=True\"","source":"ai","title":"遥测功能默认开启"},{"code":"example_settings_are_wide_open","severity":"attention","evidence":"docker-compose.yml 自带 KHOJ_DJANGO_SECRET_KEY=secret、KHOJ_ADMIN_PASSWORD=password、POSTGRES_PASSWORD=postgres,并以 --anonymous-mode 运行服务器;它自己的注释也提醒\"Uncomment the necessary lines below to make your instance publicly accessible. Proceed with caution, especially if you are using anonymous mode\"","source":"ai","title":"示例配置完全处于开放状态"},{"code":"optional_setup_gives_access_to_host_docker","severity":"attention","evidence":"docker-compose.yml 中,为启用\"computer\"功能:\"# uncomment line below to mount docker socket to allow khoj to use its computer\" / \"# - /var/run/docker.sock:/var/run/docker.sock\",这会让容器控制主机上的 Docker","source":"ai","title":"可选配置会让其访问主机的 docker"},{"code":"ai_runs_code_and_browses_for_you","severity":"info","evidence":"compose 文件中包含一个沙箱服务(ghcr.io/khoj-ai/terrarium),\"used by Khoj to run code\",还有一个可选的远程 E2B 沙箱、一个 SearxNG 搜索服务,以及一个在 5900 端口暴露 VNC 的 khoj-computer 容器","source":"ai","title":"AI 会替你运行代码和浏览网页"},{"code":"latest_releases_are_betas_without_files","severity":"info","evidence":"最近的三个发行版(2.0.0-beta.28、beta.27、beta.26,均为 2026 年 3 月)都没有附带任何平台的二进制文件,而最后一次提交是在 2026 年 8 月","source":"ai","title":"最新的发行版是没有文件的测试版"},{"code":"open_core_with_paid_cloud","severity":"info","evidence":"pyproject.toml 中有一个包含 stripe 和 twilio 的\"prod\"附加项,README 还提供\"Khoj Enterprise\"作为云端、本地部署或混合形式的服务","source":"ai","title":"开放核心加付费云服务"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","docker","gpu_any"],"disk":{"minMB":74,"maxMB":223,"basis":"installer"},"apiKeys":["OpenAI","Anthropic","Google Gemini","Serper, Firecrawl, Exa or Olostep (web search and page reading)","E2B (remote code sandbox)"],"cost":{"model":"freemium"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["productivity","ai_agents","local_ai"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"perplexity","name":"Perplexity","confirmed":true},{"slug":"notion-ai","name":"Notion AI","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[],"stars":38100,"downloads":186149,"downloadsWeek":382,"downloadsDays":5,"growth":130,"growthDays":7,"history":[{"date":"2026-11-05","stars":38008},{"date":"2026-11-06","stars":38026},{"date":"2026-11-07","stars":38051},{"date":"2026-11-08","stars":38069},{"date":"2026-11-09","stars":38091},{"date":"2026-11-10","stars":38100}],"firstPublishedAt":"2026-11-05T22:43:48.554Z","commitSha":"ae229ca894c0b80ad84664afcfdde523b5e87057","verifiedAt":"2026-11-05T22:32:04.352Z","lastCommitAt":"2026-08-02T01:55:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A personal AI that answers from your own documents and from the web, with custom agents and scheduled research.","summary":"Khoj is a personal AI assistant you can run on your own machine. You point it at your files (PDF, Word, Markdown, org-mode, Notion, images) and it searches them and answers questions, alone or together with results from the web. You can build agents with their own instructions and tools, schedule research that arrives as a newsletter, generate images and talk to it. It works with online models (GPT, Claude, Gemini) or local ones through Ollama. Self-hosting means running several containers; there is also a hosted version at app.khoj.dev.","easeRationale":"The hosted version earns the best grade: the README says \"You can use Khoj right away at https://app.khoj.dev — no setup required\". Running it yourself on Windows, macOS or Linux is a different job: the docker-compose.yml in this package starts five services (the server, a Postgres database with pgvector, a SearxNG search engine, a Python sandbox and an optional \"computer\" container), and you edit environment variables by hand to add model keys and change the default passwords.","costNote":"Self-hosting the AGPL-3.0 code is free; the hosted app at app.khoj.dev has free and paid plans, and the repository ships Stripe and Twilio as production extras","notVerified":["What the hosted plans cost and what the free one allows","What the telemetry sends when it is left on","Real disk use and memory after the search and transcription models download","Which files the release pages for Windows, macOS and Linux contain and whether they still work","How well answers from your own documents hold up on a large collection"]},"zh-Hans":{"summaryShort":"一个个人 AI,能根据你自己的文档和网络内容来回答问题,支持自定义代理和定时研究任务。","summary":"Khoj 是一个可以在你自己的机器上运行的个人 AI 助手。你把它指向你的文件(PDF、Word、Markdown、org-mode、Notion、图片),它会搜索这些文件并回答问题,既可以单独回答,也可以结合网络搜索结果。你可以创建带有自己指令和工具的代理,安排以邮件简报形式送达的定时研究任务,还能生成图片并与它对话。它可以配合在线模型(GPT、Claude、Gemini)使用,也可以通过 Ollama 使用本地模型。自己托管意味着要运行多个容器;另外在 app.khoj.dev 也有一个托管版本。","easeRationale":"托管版本获得最高评分:README 写道\"You can use Khoj right away at https://app.khoj.dev — no setup required\"。自己在 Windows、macOS 或 Linux 上运行则是另一回事:这个包里的 docker-compose.yml 会启动五个服务(服务器本身、带 pgvector 的 Postgres 数据库、SearxNG 搜索引擎、一个 Python 沙箱,以及一个可选的\"computer\"容器),你还要手动编辑环境变量来添加模型密钥并修改默认密码。","costNote":"自己托管这份 AGPL-3.0 代码是免费的;app.khoj.dev 上的托管应用有免费和付费套餐,仓库还把 Stripe 和 Twilio 作为生产环境的附加项","notVerified":["托管套餐的价格,以及免费套餐能做到什么程度","遥测功能开启时会发送哪些数据","下载搜索和转录模型后,实际占用的磁盘空间和内存","Windows、macOS 和 Linux 的发行版页面里有哪些文件,它们是否还能用","在很大的文档集合上,基于你自己文档给出的答案效果如何"]}}},{"id":"R_kgDOVo2lNQ","slug":"kianuni__erosion-cn","name":"Erosion-CN","owner":"Kianuni","fullName":"Kianuni/Erosion-CN","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{},"risk":"high","signals":[{"code":"young_owner_account","severity":"attention","evidence":"作者账号创建于 0 天前","source":"check","title":"作者账号刚注册不久"},{"code":"bypasses_device_restrictions","severity":"high","evidence":"README 第 8 节：该应用会写入 MobileGestalt 键值，解锁苹果只在更新硬件或其他地区提供的功能，例如在 iPhone 15 Pro / 16 / 17 上 \"解锁 iOS 27 进阶 AI 能力\"，用于 \"个人实时翻译 (Personal Translator)\" 和 \"iOS 27 音频智能 (Audio Intelligence)\"，以及 \"国行设备仅需开启「解除地区限制」\"。第 10 节则把硬件设备树槽位直接写进二进制 CacheData，以绕过会忽略它们的系统校验。","source":"ai","title":"绕过设备限制"},{"code":"relies_on_sandbox_escape","severity":"high","evidence":"README 开头：该应用 \"基于 worse_query\"，并注明来自 forcequitOS 与 0xjohnnydev 的 \"沙盒研究\"，第 10 节标题为 \"全链路沙盒逃逸与文件编辑安全加固\"。它需要利用 iOS 漏洞，才能把文件写到自己的应用容器之外。","source":"ai","title":"依赖沙盒逃逸"},{"code":"can_make_device_unbootable","severity":"attention","evidence":"README 的兼容性说明：\"操作前请务必通过电脑完整备份设备重要数据！修改系统底层配置存在潜在风险\"。项目名本身就带 \"防白苹果\"（开机循环），第 7 节删除了上游有 \"软砖\" 风险的功能，第 6 节还提醒备份不匹配可能引发 \"基带/eSIM 异常\"。","source":"ai","title":"可能导致设备无法开机"},{"code":"very_new_repository_many_stars","severity":"attention","evidence":"GitHub 数据：仓库创建于 2026-11-10，当天就有 1007 个星标和 233 个分支，并在同一天发布了 9 次；检查还发现作者账号只有 0 天。","source":"ai","title":"新仓库却有很多星标"},{"code":"interface_and_docs_chinese_only","severity":"info","evidence":"仓库描述以及整个 README、USAGE_GUIDE.md 和 RESTORE_CHA.md 都是简体中文，该分支宣称的目标就是 \"100% 简体中文界面本地化\"。","source":"ai","title":"界面和文档只有中文"},{"code":"release_files_need_resigning","severity":"info","evidence":"README 的安装部分：发布包提供 \"Erosion_v0.7.2_zh-Hans.ipa\"，这是一个未签名的应用，只能通过带 App Group 权限的第三方侧载工具安装；检查也未在发布中找到任何平台的可执行文件。","source":"ai","title":"发布文件需要自行重新签名"}],"platforms":[],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":1007,"downloads":0,"downloadsWeek":0,"downloadsDays":0,"growth":6818,"growthDays":0,"history":[],"firstPublishedAt":"2026-11-10T23:52:43.943Z","commitSha":"2764867bdbba0f69170b2a0d835179b20116fdc6","verifiedAt":"2026-11-10T23:32:32.917Z","lastCommitAt":"2026-11-10T14:19:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Chinese-language iPhone and iPad app that edits hidden iOS settings to turn on features Apple limits by model or region.","summary":"Erosion-CN is a Simplified Chinese fork of the Erosion project. It reads and rewrites MobileGestalt, the hidden file where iOS records what a device model is allowed to do, so you can switch on Apple Intelligence, camera and audio options and region-limited settings that Apple leaves off on your device. It also backs up app and game save data and clears app caches. Writing needs iOS or iPadOS 27.0-27.3 beta 1; on 26.x only the read-only parts work. The .ipa has to be sideloaded with a tool that keeps an App Group entitlement, such as AltStore, SideStore, Feather or Sideloadly.","easeRationale":"No desktop or web grade applies: this is an iPhone and iPad app, so none of the four platforms get a grade. On iOS you download the .ipa from the releases page and install it with a sideloading tool, re-signing it with your own Apple ID so it keeps an App Group entitlement, and the main features only write on iOS 27.","notVerified":["whether the unlocked Apple Intelligence, camera and audio options actually work on each iPhone or iPad model","whether the SHA-256 backup check and the automatic rollback really restore the original state after a failed write","whether a given sideloading tool keeps the App Group entitlement the app depends on","what the app does on iOS versions outside the supported 26.x-27.3 beta 1 range, or after a future iOS update closes the sandbox hole"]},"zh-Hans":{"summaryShort":"面向 iPhone 和 iPad 的中文应用，通过修改 iOS 隐藏配置，开启苹果按机型或地区限制的功能。","summary":"Erosion-CN 是 Erosion 项目的简体中文分支。它会读取并改写 MobileGestalt——iOS 用来记录某个机型允许做什么的隐藏文件，从而让你打开苹果在你设备上默认关闭的 Apple Intelligence、相机与音频选项以及受地区限制的设置。它还能备份应用和游戏存档，并清理应用缓存。写入功能需要 iOS 或 iPadOS 27.0-27.3 beta 1；在 26.x 上只有只读部分可用。.ipa 必须用能保留 App Group 权限的工具侧载安装，例如 AltStore、SideStore、Feather 或 Sideloadly。","easeRationale":"本项目不适用桌面或网页评级：这是一款 iPhone 和 iPad 应用，因此四个平台都不给等级。在 iOS 上，你从发布页下载 .ipa，用侧载工具安装，并用自己的 Apple ID 重新签名以保留 App Group 权限；而主要功能只有在 iOS 27 上才能写入。","notVerified":["被解锁的 Apple Intelligence、相机和音频选项，在每一种 iPhone 或 iPad 机型上是否真的可用","SHA-256 备份校验和自动回滚，在写入失败后是否真能恢复原状","某个具体的侧载工具是否会保留该应用所依赖的 App Group 权限","在支持范围 26.x-27.3 beta 1 之外的 iOS 版本上，或者未来 iOS 更新堵上沙盒漏洞之后，这个应用会怎样"]}}},{"id":"R_kgDOPYbcgQ","slug":"kittenml__kittentts","name":"KittenTTS","owner":"KittenML","fullName":"KittenML/KittenTTS","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"the_models_are_not_under_the_code_license","severity":"attention","evidence":"README 的 License 部分写道：\"That covers the code in this repository. The models are licensed separately and their terms may differ. Each model repository carries its own licensing ... do not assume the code's license extends to the weights.\" KittenTTS 2 \"released under the Stellon Labs Community License\"，这是该公司自行制定的许可证，因此检查发现的 Apache-2.0 许可仅适用于代码本身。","source":"ai","title":"模型不受代码许可证约束"},{"code":"cloning_a_voice_needs_that_persons_consent","severity":"attention","evidence":"README 的 Features 部分写道：\"Voice cloning -- Clone any speaker from 5-30 seconds of audio, no fine-tuning\"，而针对 \"cloning an unseen recording\" 测得的相似度为 0.81。未经本人同意复制某人的声音，会在许多国家涉及声音权和肖像权问题，而该仓库并没有规定可以使用谁的录音。","source":"ai","title":"克隆声音需要本人同意"},{"code":"license_fields_inside_the_project_disagree","severity":"info","evidence":"setup.py 中声明 \"License :: OSI Approved :: MIT License\"，而 LICENSE 文件是 Apache License 2.0，pyproject.toml 则写着 license = {text = \"Apache 2.0\"}。这两个打包文件对同一份代码给出了不同的许可证说明。","source":"ai","title":"项目内部的许可证字段相互矛盾"},{"code":"watermarking_library_among_the_dependencies","severity":"info","evidence":"pyproject.toml、requirements.txt 和 setup.py 中的依赖列表都包含 \"resemble-perth\"，也就是 Perth 音频水印库，它被归在注释 \"s3gen vocoder (vendored in kittenml/kittentts2/s3gen.py)\" 之下。README 从未提到水印，因此你生成的音频是否带有水印并没有说明。","source":"ai","title":"依赖项中包含一个水印库"},{"code":"parts_are_labelled_alpha_or_beta","severity":"info","evidence":"setup.py 中标注为 \"Development Status :: 3 - Alpha\"；表达控制功能被标记为 \"Beta. Emotion control steers delivery rather than guaranteeing it\"；而更快的 C++ 路径写着 \"Kitten-TTS 2 C++ is in beta\"。README 还提到该模型 \"truncates or drifts into repetition when asked for a whole script in one pass\"。","source":"ai","title":"部分内容被标注为 alpha 或 beta"},{"code":"models_are_downloaded_at_first_use","severity":"info","evidence":"模型权重保存在单独的 HuggingFace 仓库中（\"KittenML/kitten-tts-2\"，492 MB，还有一个无损的 954 MB 选项），前提条件要求 \"About 2 GB of disk space for the model and audio decoder, plus Python dependencies\"。程序会在你首次运行时从网上获取这些文件，检查估算总占用为 1250 到 5250 MB。","source":"ai","title":"模型在首次使用时下载"},{"code":"github_releases_lag_behind_the_code","severity":"info","evidence":"GitHub 上最新的发布版本是 2026-02-24 的 0.8.1，而 pyproject.toml 和 setup.py 中的版本号是 0.9.7，最后一次提交是在 2026-10-27。\"pip install kittenml\" 实际会给你什么版本，这里没有任何发布说明描述。","source":"ai","title":"github 发布版本落后于代码"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":1250,"maxMB":5250,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai","development"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"speechify","name":"Speechify","confirmed":true}],"declaredAlternatives":[],"stars":16831,"downloads":3511934,"downloadsWeek":186516,"downloadsDays":3,"growth":37,"growthDays":6,"history":[{"date":"2026-11-07","stars":16808},{"date":"2026-11-08","stars":16821},{"date":"2026-11-09","stars":16828},{"date":"2026-11-10","stars":16831}],"firstPublishedAt":"2026-11-06T22:59:23.792Z","commitSha":"b9517e3050adced306219f8bea5f117d87ecfaa7","verifiedAt":"2026-11-06T22:46:52.076Z","lastCommitAt":"2026-10-27T01:43:42Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns written text into speech on your own computer: 115 voices, 20 languages and voice cloning from a short recording.","summary":"KittenTTS reads text aloud with models that run on your own machine, no account and no internet call needed once the files are downloaded. The main model has 1.7 billion parameters, 115 voices and 20 languages, and can copy a speaker from 5 to 30 seconds of audio; there are also tiny models of 15 to 80 million parameters for weak computers. It is a Python library: you install it with pip and write a few lines of code, or start the included Gradio page for a simple screen. A graphics card is optional, but about 6 GB of free memory is not.","easeRationale":"Grade C on all three systems: with Python 3.10 or later you run \"pip install kittenml\", then either write a few lines of Python or run \"pip install kittenml gradio\" and \"python gradio_demo.py\" for a local page. Windows may also need Microsoft's Visual C++ runtime, the faster Mac path means installing the extra MLX backend, and the NVIDIA route on Linux asks for driver 580 or newer, a C compiler and Python development headers.","costNote":"The code is Apache-2.0 and the models are downloaded at no charge. The same company sells \"integration assistance, custom voice development, and enterprise licensing\" and runs a hosted API at platform.kittenml.com, described in the README as free.","notVerified":["Whether the voices and the cloning really sound as reported, since the numbers come from the project's own measurements","What the Stellon Labs Community License allows, as its text lives in the model repository and not here","Whether generated audio carries a watermark, which the dependency on resemble-perth raises but the README does not address","Real disk and memory use on your machine, beyond the \"about 2 GB\" and \"about 6 GB of RAM\" the README states","How fast it is on your processor, as the claim of running in real time on a CPU is the project's own","What the hosted API at platform.kittenml.com records about the text you send it"]},"zh-Hans":{"summaryShort":"在你自己的电脑上把书面文字转换成语音：115 种声音、20 种语言，还能从一段简短录音克隆声音。","summary":"KittenTTS 使用在你自己的设备上运行的模型朗读文字，文件下载完成后既不需要账号，也不需要联网调用。主模型拥有 17 亿个参数、115 种声音和 20 种语言，并能根据 5 到 30 秒的音频复制说话人的声音；此外还有面向性能较弱电脑的 1500 万到 8000 万参数的小型模型。它是一个 Python 库：你可以用 pip 安装它并编写几行代码，也可以启动自带的 Gradio 页面来获得一个简单的界面。显卡是可选的，但大约 6 GB 的空闲内存并非可选。","easeRationale":"三个系统上均为 C 级：在已安装 Python 3.10 或更高版本的前提下，运行 \"pip install kittenml\"，然后要么编写几行 Python 代码，要么运行 \"pip install kittenml gradio\" 和 \"python gradio_demo.py\" 来获得一个本地页面。在 Windows 上可能还需要微软的 Visual C++ 运行库，Mac 上更快的路径意味着要安装额外的 MLX 后端，而 Linux 上使用 NVIDIA 的路径则需要 580 或更新的驱动程序、一个 C 编译器以及 Python 开发头文件。","costNote":"代码采用 Apache-2.0 许可，模型下载不收费。同一家公司出售 \"integration assistance, custom voice development, and enterprise licensing\"，并在 platform.kittenml.com 运营一个托管 API，README 中称其为免费。","notVerified":["这些声音和克隆效果是否真的如宣称的那样，因为相关数字来自项目自身的测量","Stellon Labs Community License 具体允许什么，因为其条款在模型仓库中，不在这里","生成的音频是否带有水印，对 resemble-perth 的依赖引出了这个问题，但 README 没有说明","在你的设备上实际占用多少磁盘和内存，而不仅是 README 所写的 \"about 2 GB\" 和 \"about 6 GB of RAM\"","在你的处理器上运行速度有多快，因为可以在 CPU 上实时运行的说法来自项目自身","platform.kittenml.com 上托管的 API 会记录你发送的文本的哪些内容"]}}},{"id":"R_kgDOVhF2rg","slug":"kkkkhazix__agent-autoloop","name":"agent-autoloop","owner":"KKKKhazix","fullName":"KKKKhazix/agent-autoloop","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"install_script_piped_from_internet","severity":"attention","evidence":"INSTALL.md 和 README.md 给出的安装命令是 `curl -fsSL https://raw.githubusercontent.com/KKKKhazix/agent-autoloop/main/install.sh | sh`,会立刻执行一个下载下来的脚本;install.sh 随后会把仓库克隆到 ~/.agent-autoloop/app,并用 `\"$PROGRAM\" service install` 安装一个后台服务。","source":"ai","title":"安装脚本直接从网上执行"},{"code":"ai_runs_commands_unattended","severity":"attention","evidence":"这整个项目的重点就是无人看管的工作:README.md 说中枢会根据你的规则决定\"交给哪个 Agent、去哪个目录干\",而交付流程\"审代码 → 等 CI → 合并 PR → 等部署\"会在打上 `deliver` 标签后自行合并拉取请求。项目说在删除数据、重启服务或花钱之前会先问你,但这个承诺只有实际运行才能验证。","source":"ai","title":"ai 在无人看管的情况下运行命令"},{"code":"very_new_project","severity":"info","evidence":"该仓库创建于 2026-11-03,截至 2026-11-04 检查时已经有 2404 个星标和一个发布版本(v1.0.0,2026-11-03),所以还没有经过时间验证的记录。","source":"ai","title":"项目非常新"},{"code":"features_not_tested_by_author","severity":"info","evidence":"README.md 说四个 webhook 模板只针对本地模拟服务器做过检查,GitLab webhook 从未连接过真正的 GitLab,列出的其他助手(DeepSeek Harness、pi、OpenClaw、Hermes、Cursor、CodeBuddy)的命令行是从它们各自的文档里抄来的,\"我们还没有逐个跑过\"。","source":"ai","title":"部分功能作者自己也没测试过"},{"code":"docs_mostly_in_chinese","severity":"info","evidence":"主 README.md、INSTALL.md,以及 install.sh 里的终端提示信息都是中文;有一个英文的 README.en.md,但你需要照着做的安装说明和示例都是中文。","source":"ai","title":"文档大多是中文"}],"platforms":["mac","linux"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","automation","development"],"alternativeTo":[{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"n8n-cloud","name":"Cloud","confirmed":true},{"slug":null,"name":"n8n","confirmed":false}],"declaredAlternatives":[],"stars":4420,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":4420,"growthDays":7,"history":[{"date":"2026-11-05","stars":3211},{"date":"2026-11-06","stars":3646},{"date":"2026-11-07","stars":3941},{"date":"2026-11-08","stars":4092},{"date":"2026-11-09","stars":4266},{"date":"2026-11-10","stars":4420}],"firstPublishedAt":"2026-11-04T22:32:55.030Z","commitSha":"8f8e2eef3cfaa8f7a418818adc442450fe6bd884","verifiedAt":"2026-11-04T22:14:09.798Z","lastCommitAt":"2026-11-10T20:06:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Keeps a terminal AI assistant working in the background on a schedule or on events, and pings you only when it needs a decision.","summary":"A background service for people who already use terminal AI assistants such as Claude Code or Codex. You write rules in a JSON file (for example: every morning at 9, or when a new GitHub issue arrives) and it starts the assistant, follows the job through, and sends you a card on desktop, Feishu, Slack or Discord when it needs your choice. While it waits, nothing is running. It works on macOS and Linux and you need Node 24, Git, curl and an assistant you have already logged into. Windows is not supported yet.","easeRationale":"On macOS and Linux it is all terminal: install Node 24, Git and curl, run an install script, log into a command-line AI assistant, then write rules by hand in a JSON file. There is no grade for Windows because the README says it does not run there yet.","costNote":"The software is MIT and free. The AI work runs on your own assistant account, so you pay the model bill; the README says that for heavy automation the model providers prefer an API key over a personal subscription.","notVerified":["Whether the install script and the background service really do only what the documents describe on your machine","Whether it works end to end on macOS and Linux with your own assistant login","How much the model usage costs once rules are running every few minutes","Whether the promise of asking before deleting data, restarting services or sending messages holds in practice","Whether the announced Windows version arrives","Real disk use as the append-only log and database grow over time"]},"zh-Hans":{"summaryShort":"让终端里的 AI 助手按计划或事件在后台持续工作,只在需要你做决定时才通知你。","summary":"这是一个为已经在用终端 AI 助手(比如 Claude Code 或 Codex)的人准备的后台服务。你在一个 JSON 文件里写规则(比如:每天早上 9 点,或者有新的 GitHub issue 出现时),它就会启动助手、把任务跟进到底,并在需要你做选择时,通过桌面通知、飞书、Slack 或 Discord 给你发一张卡片。等待的时候,什么都不会运行。它支持 macOS 和 Linux,你需要 Node 24、Git、curl,以及一个你已经登录过的助手。Windows 暂时还不支持。","easeRationale":"在 macOS 和 Linux 上全部是终端操作:安装 Node 24、Git 和 curl,运行一个安装脚本,登录一个命令行 AI 助手,再手动在 JSON 文件里写规则。Windows 没有评级,因为 README 说它目前还跑不起来。","costNote":"这个软件采用 MIT 许可证,免费。AI 的工作运行在你自己的助手账号上,所以模型费用由你承担;README 提到,对于高强度的自动化,模型提供商更倾向于用 API 密钥而不是个人订阅。","notVerified":["安装脚本和后台服务在你的机器上是否真的只做文档描述的事情","用你自己的助手账号登录后,在 macOS 和 Linux 上是否能端到端跑通","规则每隔几分钟运行一次之后,模型使用的费用会是多少","在删除数据、重启服务或发送消息前会先询问你的承诺,在实际使用中是否站得住","宣布的 Windows 版本是否真的会推出","随着只增量写入的日志和数据库不断增长,真实的磁盘占用情况"]}}},{"id":"R_kgDOUxY7BQ","slug":"kkkkhazix__aihot","name":"AIHOT","owner":"KKKKhazix","fullName":"KKKKhazix/AIHOT","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"28 天内获得 8675 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"author_not_professional_developer","severity":"info","evidence":"README.md：\"我不是专业的开发者。…这套代码是我和 AI 一起重写的\"（作者表示自己不是专业开发者，代码是和 AI 一起重写的）。","source":"ai","title":"作者不是专业开发者"},{"code":"port_open_on_all_interfaces","severity":"info","evidence":".env.example：\"网站在本机开的端口（docker compose 用，默认 3000，所有网卡都能访问）\" — 默认情况下，网站端口可以从所有网络接口访问，而不仅限于这台电脑。","source":"ai","title":"端口对所有网络接口开放"}],"platforms":["windows","mac","linux"],"tools":["docker","node"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["OpenAI-compatible AI model (DeepSeek by default; Qwen or Zhipu also work)","Embeddings API (OpenAI or Alibaba DashScope)","SocialData (X posts)","Dajiala (WeChat articles)","Jina Reader","Feishu (notifications and login)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["local_ai","productivity","automation"],"alternativeTo":[{"slug":"feedly","name":"Feedly","confirmed":true},{"slug":"inoreader","name":"Inoreader","confirmed":true}],"declaredAlternatives":[],"stars":10275,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":891,"growthDays":7,"history":[{"date":"2026-10-26","stars":8675},{"date":"2026-10-27","stars":8733},{"date":"2026-10-28","stars":8818},{"date":"2026-10-29","stars":8886},{"date":"2026-10-30","stars":8947},{"date":"2026-10-31","stars":8983},{"date":"2026-11-01","stars":9064},{"date":"2026-11-02","stars":9243},{"date":"2026-11-03","stars":9384},{"date":"2026-11-04","stars":9539},{"date":"2026-11-05","stars":9630},{"date":"2026-11-06","stars":9695},{"date":"2026-11-07","stars":9728},{"date":"2026-11-08","stars":9757},{"date":"2026-11-09","stars":10055},{"date":"2026-11-10","stars":10275}],"firstPublishedAt":"2026-10-27T01:10:12.649Z","commitSha":"aa8e5b1bc4895267f98abe59c87dfed8a22bf35f","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T20:44:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Run your own news website that gathers articles from sources you pick and uses AI to choose the best ones and write a daily digest.","summary":"AIHOT is the engine behind the Chinese AI news site aihot.news. You host it yourself: it collects posts from RSS feeds, web pages, X accounts and WeChat accounts, asks an AI model to score and summarise them, groups stories about the same event, ranks what is trending and publishes daily, weekly and monthly digests. Titles and summaries are written in Chinese by default. It suits people who want a news site for their own field. You need Docker, Node.js 24 and a key for an OpenAI-compatible AI service such as DeepSeek.","easeRationale":"You need Docker and Node.js 24, then you clone the repository, generate the settings with a script and start several services with docker compose from a terminal; running without Docker on Windows needs WSL2.","costNote":"The software is free (MIT), but every step calls a paid AI model API; optional collectors (SocialData for X, Dajiala for WeChat, Jina Reader) charge per request.","notVerified":["How much the AI model calls cost per day with a real list of sources.","Whether the Docker setup runs smoothly on Windows with Docker Desktop.","How well the scoring and summaries work for fields other than AI, or for languages other than Chinese.","Real disk use once the database and image cache grow."]},"zh-Hans":{"summaryShort":"运行你自己的新闻网站：从你选定的来源收集文章，用 AI 挑选最好的内容并撰写每日摘要。","summary":"AIHOT 是中文 AI 新闻网站 aihot.news 背后的引擎。你需要自己托管：它从 RSS 订阅、网页、X 账号和微信账号收集内容，请 AI 模型打分并总结，把同一事件的报道归为一组，对热门内容排序，并发布每日、每周和每月摘要。标题和摘要默认用中文撰写。它适合想为自己所在领域做一个新闻网站的人。你需要 Docker、Node.js 24，以及 DeepSeek 等兼容 OpenAI 的 AI 服务的密钥。","easeRationale":"你需要 Docker 和 Node.js 24，然后克隆代码仓库，用脚本生成设置，并在终端用 docker compose 启动多个服务；在 Windows 上不用 Docker 运行则需要 WSL2。","costNote":"软件免费（MIT），但每一步都要调用付费的 AI 模型 API；可选的采集器（用于 X 的 SocialData、用于微信的 Dajiala、Jina Reader）按请求收费。","notVerified":["用一份真实的来源列表时，AI 模型调用每天要花多少钱。","Docker 方案在装有 Docker Desktop 的 Windows 上是否运行顺畅。","打分和摘要在 AI 以外的领域或中文以外的语言上效果如何。","数据库和图片缓存增长后实际占用多少磁盘空间。"]}}},{"id":"R_kgDOR643pQ","slug":"kkkkhazix__khazix-skills","name":"Khazix Skills","owner":"KKKKhazix","fullName":"KKKKhazix/khazix-skills","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"the_cleaning_skill_can_delete_your_files","severity":"attention","evidence":"storage-analyzer 会打开一份 HTML 报告，你可以在其中\"直接点按钮移到废纸篓 / 删除\"。被标记为绿色的项目可能被助手一次性清除，相关规则还补充道\"用户已明确授权具体清理对象时，Agent 也可代为执行，无需重复确认\"。模型一旦判断失误，就是一个被删除的文件。","source":"ai","title":"清理技能可以删除你的文件"},{"code":"a_full_disk_scan_is_read_by_the_assistant","severity":"attention","evidence":"该技能会\"扫一遍整机磁盘\"，报告中会列出具体的文件夹，给出的例子是一个被识别为 B 站离线视频缓存、大小为 97 GB 的容器。它找到的文件名和路径都会传给助手，而对大多数人来说这是一个在线模型，于是你磁盘上的结构就离开了你的电脑。","source":"ai","title":"完整的磁盘扫描结果会被助手读取"},{"code":"scan_runs_a_local_web_page_with_a_token","severity":"info","evidence":"规则说明本地服务运行在\"127.0.0.1 + 随机端口 + token\"上，页面中的删除操作会逐一确认，扫描本身是只读的，并且该服务在 30 分钟没有请求后会自动退出。","source":"ai","title":"扫描功能通过带令牌的本地网页运行"},{"code":"news_skill_depends_on_the_authors_own_service","severity":"info","evidence":"aihot 技能说明\"查询和整理都在 aihot.news 服务端完成\"，它本身只负责提问和转发答案。你得到的结果，以及该服务器从你的问题中获取的信息，都由同一位作者在这个仓库之外决定，而他把自己介绍为一家媒体公司的创始人。","source":"ai","title":"新闻技能依赖作者自己的服务"},{"code":"a_skill_is_instructions_your_assistant_will_follow","severity":"info","evidence":"安装的方式就是对助手说\"帮我安装这个 skill\"并附上仓库链接，之后\"Agent 会自己 clone 到对应目录\"，包含它的 references/、scripts/、assets/ 和 agents/ 文件夹。从那以后，你的助手就会遵循别人写的文字并运行别人写的脚本；neat-freak 自己也指出，在文件内部发现的命令不应被当作你的授权。","source":"ai","title":"技能就是你的助手将会遵循的指令"},{"code":"written_for_a_chinese_speaking_audience","severity":"info","evidence":"主 README 是中文的，附带一个较短的 README.en.md；其中的说明链接到微信文章，而 khazix-writer 的存在就是为了以作者的语气撰写微信长文，并列出了它拒绝使用的中文表达。这个写作技能在那种风格之外几乎没什么用处。","source":"ai","title":"面向中文读者编写"},{"code":"install_only_from_this_repository_or_its_release","severity":"info","evidence":"README 写道：\"第三方市场可能滞后或漏带资源，完整内容以此仓库和 Release 为准\"。它还提醒不要同时保留同一个技能的插件安装版本和手动复制的版本，并且过去教程里使用的旧技能名称已经不再分发。","source":"ai","title":"只从本仓库或其发布版安装"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","network_system","productivity"],"alternativeTo":[{"slug":"cleanmymac","name":"CleanMyMac","confirmed":true},{"slug":"perplexity","name":"Perplexity","confirmed":true}],"declaredAlternatives":[],"stars":22158,"downloads":198,"downloadsWeek":87,"downloadsDays":3,"growth":202,"growthDays":6,"history":[{"date":"2026-11-07","stars":22056},{"date":"2026-11-08","stars":22068},{"date":"2026-11-09","stars":22113},{"date":"2026-11-10","stars":22158}],"firstPublishedAt":"2026-11-06T22:57:16.695Z","commitSha":"35bd05b466ceb83a727b28afb54b45b69a6c3b0a","verifiedAt":"2026-11-06T22:48:03.062Z","lastCommitAt":"2026-11-10T20:13:19Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Six ready-made skills for AI assistants: write goals, clean up disk space, tidy project docs, research and write.","summary":"A pack of six skills the author uses daily, for Claude Code, Codex and other assistants that follow the Agent Skills standard. leader turns a vague idea into a written goal an assistant can work on for hours; storage-analyzer scans the whole disk and opens a colour-coded report in your browser with buttons to send files to the trash; aihot fetches the day's AI news from aihot.news; neat-freak brings project docs, CLAUDE.md and the assistant's memory back in line; hv-analysis writes a long research report as PDF; khazix-writer copies the author's writing style.","easeRationale":"Grade C on all three systems, because what you install into is an assistant you must already have: in Claude Code you type \"/plugin marketplace add KKKKhazix/khazix-skills\" then \"/plugin install khazix-skills@khazix-skills\", or you ask any supporting assistant to install one folder from the repository. There is nothing to double-click, and the disk-cleaning skill only claims macOS and Windows, so on Linux five of the six apply.","costNote":"MIT licensed, and the news skill states it needs no API key and no MCP server. Any cost is what your assistant already charges for the tokens it uses.","notVerified":["Whether each skill behaves as described, since what runs is an assistant reading instructions","Whether the colour grading always puts your files in the right group before anything is deleted","What aihot.news records about the questions your assistant sends it","Which tools the included scripts need on your machine, as the repository is mostly Python and states no requirement","Whether the disk-cleaning skill works on Linux, which the README does not claim","How many tokens, and therefore how much money, a long goal or a 10,000-word report costs in your assistant"]},"zh-Hans":{"summaryShort":"面向 AI 助手的六个现成技能：撰写目标、清理磁盘空间、整理项目文档、做调研和写作。","summary":"这是作者每天都在使用的六个技能的合集，适用于 Claude Code、Codex 以及其他遵循 Agent Skills 标准的助手。leader 把一个模糊的想法变成一个书面目标，供助手花几个小时去完成；storage-analyzer 扫描整个磁盘，并在浏览器中打开一份带颜色标记的报告，附有把文件送进回收站的按钮；aihot 从 aihot.news 获取当天的 AI 新闻；neat-freak 让项目文档、CLAUDE.md 和助手的记忆重新保持一致；hv-analysis 生成一份较长的研究报告并导出为 PDF；khazix-writer 模仿作者本人的写作风格。","easeRationale":"三个系统上均为 C 级，因为你安装的内容实际运行在一个你必须已经拥有的助手里：在 Claude Code 中输入 \"/plugin marketplace add KKKKhazix/khazix-skills\"，再输入 \"/plugin install khazix-skills@khazix-skills\"，或者让任何支持的助手从仓库中安装某一个文件夹。没有任何东西可以双击打开，而磁盘清理技能只声明支持 macOS 和 Windows，因此在 Linux 上六个技能中有五个适用。","costNote":"采用 MIT 许可证，新闻技能声明不需要 API 密钥，也不需要 MCP 服务器。唯一的费用就是你的助手本身已经在收取的 token 费用。","notVerified":["每个技能是否真的像描述的那样运作，因为实际运行的是一个读取指令的助手","颜色分级是否总能在删除任何文件之前把它们归入正确的组","aihot.news 会记录你的助手发送给它的问题中的哪些内容","所附带的脚本在你的电脑上需要哪些工具，因为该仓库主要是 Python 代码，也没有说明任何依赖要求","磁盘清理技能是否真的能在 Linux 上运行，README 并未声明这一点","在你的助手里，一个较长的目标或一份一万字的报告会消耗多少 token，也就是花多少钱"]}}},{"id":"R_kgDOUz1Obg","slug":"knifelemon__filee","name":"Filee","owner":"KnifeLemon","fullName":"KnifeLemon/Filee","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"B","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"29 天内获得 16380 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"self_updating_app","severity":"info","evidence":"README.md：“Filee shows new releases in its menu and updates itself”，并且会在“shortly after start and then every few hours”访问 api.github.com（可以在 Settings 中关闭）。","source":"ai","title":"应用会自行更新"},{"code":"downloads_engines_at_runtime","severity":"info","evidence":"README.md：可选引擎（Ghostscript、LibreOffice、Pandoc、calibre、FFmpeg）“downloaded only when needed”，下载地址固定在 `engines.json` 中。","source":"ai","title":"运行时下载引擎"},{"code":"readme_suggests_disabling_protection","severity":"info","evidence":"README.md：如果 Filee 在过旧的 Windows 10 上无法启动，请以管理员身份运行 `Set-ProcessMitigation -Name Filee.exe -Disable UserShadowStack`；推荐的解决办法是更新 Windows。","source":"ai","title":"README 建议关闭一项保护"},{"code":"older_releases_unsigned","severity":"info","evidence":"README.md：“Releases before it, up to 2.0.0, are not signed”以及“macOS packages are ad-hoc signed and not notarized”。","source":"ai","title":"旧版本没有签名"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":127,"maxMB":381,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","productivity","image"],"alternativeTo":[{"slug":"acrobat","name":"Acrobat","confirmed":true}],"declaredAlternatives":[],"stars":25383,"downloads":827520,"downloadsWeek":191536,"downloadsDays":7,"growth":4060,"growthDays":7,"history":[{"date":"2026-10-26","stars":13246},{"date":"2026-10-27","stars":14778},{"date":"2026-10-28","stars":16354},{"date":"2026-10-29","stars":17428},{"date":"2026-10-30","stars":18485},{"date":"2026-10-31","stars":19288},{"date":"2026-11-01","stars":20080},{"date":"2026-11-02","stars":20733},{"date":"2026-11-03","stars":21323},{"date":"2026-11-04","stars":21954},{"date":"2026-11-05","stars":22555},{"date":"2026-11-06","stars":23024},{"date":"2026-11-07","stars":23618},{"date":"2026-11-08","stars":24174},{"date":"2026-11-09","stars":24652},{"date":"2026-11-10","stars":25383}],"firstPublishedAt":"2026-10-27T01:10:12.650Z","commitSha":"58bbe56ffd746076e557e4dc48b1679c10211fd6","verifiedAt":"2026-10-29T01:38:04.826Z","lastCommitAt":"2026-11-10T22:32:18Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Filee-2.4.0-win-Setup.exe","sizeBytes":138550336,"url":"https://github.com/KnifeLemon/Filee/releases/download/v2.4.0/Filee-2.4.0-win-Setup.exe","releaseUrl":"https://github.com/KnifeLemon/Filee/releases/tag/v2.4.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free app that converts images, PDFs, Office files, video, music and e-books by dragging them onto a menu at your cursor.","summary":"Filee is a free file converter for Windows, Mac and Linux. You hold a key, drag files, and drop them on the format you want in a round menu that opens at your cursor. It handles images, PDF, Word, Excel, PowerPoint, Korean HWP, e-books, archives, fonts, CAD and 3D files, and can merge, split or compress PDFs. Files stay on your computer. Video, audio and some older formats need extra free engines that the app downloads when needed. The Windows installer needs nothing else installed; the Mac version is marked as a preview.","easeRationale":"Windows is the easiest: a signed Setup.exe (also on the Microsoft Store) and a portable .zip that need nothing else installed. On Mac you open the .dmg, drag the app to Applications and then allow the first start in Privacy & Security because the build is not notarized. On Linux you unpack the package and run ./install.sh, which needs Python 3, and drag gestures only work on X11.","costNote":"MIT license; donations accepted.","notVerified":["How faithfully Word, Excel, PowerPoint and HWP files keep their layout after conversion.","Real disk space after the app downloads optional engines such as LibreOffice or FFmpeg.","Whether the drag-and-drop menu works with every Windows setup and keyboard.","What the self-updater downloads and how updates are verified.","Whether the Mac preview and the Linux build work as well as the Windows one."]},"zh-Hans":{"summaryShort":"一个免费的应用，把图片、PDF、Office 文件、视频、音乐和电子书拖到鼠标指针处的菜单上就能转换格式。","summary":"Filee 是一个适用于 Windows、Mac 和 Linux 的免费文件转换器。你按住一个键，拖动文件，放到鼠标指针处弹出的圆形菜单里你想要的格式上即可。它支持图片、PDF、Word、Excel、PowerPoint、韩文 HWP、电子书、压缩包、字体、CAD 和 3D 文件，还能合并、拆分或压缩 PDF。文件始终留在你自己的电脑上。视频、音频以及一些较旧的格式需要额外的免费引擎，应用会在需要时自动下载。Windows 安装包不需要额外安装其他东西；Mac 版本标注为预览版。","easeRationale":"Windows 上最简单：一个已签名的 Setup.exe（也上架了 Microsoft Store）和一个不需要额外安装任何东西的便携版 .zip。在 Mac 上，你打开 .dmg，把应用拖到应用程序文件夹，然后在系统的隐私设置里允许第一次启动，因为这个构建没有经过公证。在 Linux 上，你解包后运行 ./install.sh，这需要 Python 3，而且拖拽手势只在 X11 下可用。","costNote":"采用 MIT 许可；接受捐赠。","notVerified":["Word、Excel、PowerPoint 和 HWP 文件转换后版式保留得有多准确","应用下载 LibreOffice 或 FFmpeg 等可选引擎之后，真实的磁盘占用空间","拖放菜单是否能在各种 Windows 配置和键盘下正常工作","自动更新程序会下载什么内容，以及更新是如何被验证的","Mac 预览版和 Linux 版本是否能像 Windows 版本一样好用"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2NzMxNDMy","slug":"koalaman__shellcheck","name":"ShellCheck","owner":"koalaman","fullName":"koalaman/shellcheck","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"the_web_version_sends_your_script_to_someone_elses_server","severity":"attention","evidence":"README：\"Paste a shell script on https://www.shellcheck.net for instant feedback\"，该网站 \"is always synchronized to the latest git commit\"。脚本里的任何内容——服务器名称、路径、密钥——都会离开你的电脑。本地安装不需要联网。","source":"ai","title":"网页版会把你的脚本发送到别人的服务器"},{"code":"new_versions_can_break_an_automated_build","severity":"info","evidence":"README：\"It's a good idea to manually install a specific ShellCheck version regardless. This avoids any surprise build breaks when a new version with new warnings is published.\" 如果你让构建在出现警告时失败，就应该固定版本。","source":"ai","title":"新版本可能会破坏自动化构建"},{"code":"last_release_is_from_2025_while_the_code_keeps_moving","severity":"info","evidence":"最新发布版 v0.11.0 来自 2025-08-04，但最后一次提交是在 2026-10-26，有 1089 个未关闭的 issue，对应 40262 个星标。14 年里只发布了 30 个版本：项目还活着，但发布得很少，所以包管理器给你的版本可能大幅落后于代码。","source":"ai","title":"最新发布版来自 2025 年，而代码仍在持续更新"},{"code":"extra_fiddling_on_windows","severity":"info","evidence":"README：在原生 Windows 上 \"the system may use a legacy codepage\"，所以你得选一个 TrueType 字体并运行 `chcp 65001`，还要在 PowerShell ISE 里加一行额外的编码设置，否则输出会乱码。","source":"ai","title":"在 windows 上需要额外折腾"}],"platforms":["web","windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":40318,"downloads":67696674,"downloadsWeek":1555295,"downloadsDays":7,"growth":41,"growthDays":7,"history":[{"date":"2026-10-26","stars":40244},{"date":"2026-10-27","stars":40256},{"date":"2026-10-28","stars":40259},{"date":"2026-10-29","stars":40259},{"date":"2026-10-30","stars":40262},{"date":"2026-10-31","stars":40266},{"date":"2026-11-01","stars":40270},{"date":"2026-11-02","stars":40276},{"date":"2026-11-03","stars":40277},{"date":"2026-11-04","stars":40282},{"date":"2026-11-05","stars":40287},{"date":"2026-11-06","stars":40296},{"date":"2026-11-07","stars":40298},{"date":"2026-11-08","stars":40303},{"date":"2026-11-09","stars":40309},{"date":"2026-11-10","stars":40318}],"firstPublishedAt":"2026-10-27T01:10:12.651Z","commitSha":"919d37c4d77a7ce16cfc76ebdd781dd6e8081fa5","verifiedAt":"2026-10-31T01:45:07.765Z","lastCommitAt":"2026-11-09T01:29:52Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Reads shell scripts and points out mistakes with an explanation, in the terminal, your editor or on a web page.","summary":"ShellCheck reads a bash or sh script without running it and lists the problems it finds: forgotten quotes, comparisons that are always true, misused commands, things that work in bash but not in sh. Each warning has a number with a long explanation on the project wiki. You can paste a script at shellcheck.net, run `shellcheck yourscript` in a terminal, see the warnings inside editors such as VS Code, Vim or Emacs, or add it to an automated build. It is free, GPL-3.0, and has existed since 2012.","easeRationale":"Nothing to install on the web: \"Paste a shell script on https://www.shellcheck.net for instant feedback\", which the README calls the easiest way to try it. On a computer it stays at C because it is a terminal program with no window of its own: one command installs it (`winget install --id koalaman.shellcheck` on Windows, `brew install shellcheck` on macOS, `sudo apt install shellcheck` on Linux) and then you type `shellcheck yourscript` or wire it into your editor.","costNote":"Free and GPL-3.0; contributions must be GPLv3 as well. The website and all packages cost nothing.","notVerified":["Whether it finds the particular bug in your script: it only reads the text and never runs it","How many of its warnings are beside the point in your script, which only you can judge","Whether shellcheck.net keeps or logs the scripts people paste","Which version your package manager gives you, which can be older than v0.11.0","Disk use, which the automatic check could not determine, although the downloads are 2 to 14 MB"]},"zh-Hans":{"summaryShort":"读取 shell 脚本并指出错误，附带说明，可以在终端、编辑器或网页上使用。","summary":"ShellCheck 在不运行 bash 或 sh 脚本的情况下读取它，并列出发现的问题：忘记加引号、永远为真的比较、用错的命令、在 bash 里能用但在 sh 里不行的写法。每条警告都有一个编号，在项目的 wiki 上有详细说明。你可以把脚本粘贴到 shellcheck.net，在终端里运行 `shellcheck yourscript`，在 VS Code、Vim 或 Emacs 等编辑器里直接看到警告，或者把它加入自动化构建流程。它免费，采用 GPL-3.0 许可，从 2012 年就存在了。","easeRationale":"网页版不用安装任何东西：\"Paste a shell script on https://www.shellcheck.net for instant feedback\"，README 称这是最简单的试用方式。在电脑上则停留在 C 级，因为它是没有自己窗口的终端程序：一条命令就能安装（Windows 上 `winget install --id koalaman.shellcheck`，macOS 上 `brew install shellcheck`，Linux 上 `sudo apt install shellcheck`），然后你输入 `shellcheck yourscript` 或把它接入编辑器。","costNote":"免费，采用 GPL-3.0 许可；贡献的代码也必须是 GPLv3。网站和所有软件包都不收费。","notVerified":["它是否能找到你脚本里具体的那个错误：它只读文本，从不运行脚本","它的哪些警告在你的脚本里其实无关紧要，这只有你自己能判断","shellcheck.net 是否会保存或记录人们粘贴上去的脚本","你的包管理器给你的是哪个版本，可能比 v0.11.0 旧","磁盘占用情况，自动检查无法确定，不过下载包大小在 2 到 14 MB 之间"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1Njg5OTI4NA==","slug":"kong__insomnia","name":"Insomnia","owner":"Kong","fullName":"Kong/insomnia","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"account_required_for_most_features","severity":"info","evidence":"README.md：“Insomnia does not require an account if you decide to use the local Scratch Pad, but to access most capabilities of the product we require an account.”","source":"ai","title":"大部分功能需要账号"},{"code":"analytics_component","severity":"info","evidence":"package.json 的 workspaces 包含“packages/insomnia-analytics”，allowScripts 包含“@scarf/scarf@1.4.0”: true（Scarf 是一个统计软件包使用情况的库）。","source":"ai","title":"含使用情况统计组件"},{"code":"bundles_local_ai_model_runner","severity":"info","evidence":"package.json 的 allowScripts 列出了“node-llama-cpp@3.18.1”: true，这是一个可以在你自己的电脑上运行 AI 语言模型的库。这也是一款请求工具附带如此大文件的原因之一。","source":"ai","title":"内置本地 AI 模型运行组件"},{"code":"large_download","severity":"info","evidence":"对一款请求工具来说，发布文件很大：Insomnia.Core-13.4.0.exe 为 620 MB，Linux 的 AppImage 为 622 MB；确定性检查估算的磁盘占用在 620 MB 到 1859 MB 之间。","source":"ai","title":"下载体积很大"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":620,"maxMB":1859,"basis":"installer"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"postman","name":"Postman","confirmed":true}],"declaredAlternatives":[],"stars":40138,"downloads":142944146,"downloadsWeek":443433,"downloadsDays":7,"growth":26,"growthDays":7,"history":[{"date":"2026-10-26","stars":40079},{"date":"2026-10-27","stars":40089},{"date":"2026-10-28","stars":40096},{"date":"2026-10-29","stars":40101},{"date":"2026-10-30","stars":40106},{"date":"2026-10-31","stars":40105},{"date":"2026-11-01","stars":40107},{"date":"2026-11-02","stars":40110},{"date":"2026-11-03","stars":40112},{"date":"2026-11-04","stars":40122},{"date":"2026-11-05","stars":40132},{"date":"2026-11-06","stars":40135},{"date":"2026-11-07","stars":40133},{"date":"2026-11-08","stars":40131},{"date":"2026-11-09","stars":40133},{"date":"2026-11-10","stars":40138}],"firstPublishedAt":"2026-10-27T01:10:12.652Z","commitSha":"f6f38495946a1342c94a691dccaddb7c09f31169","verifiedAt":"2026-10-30T01:42:36.700Z","lastCommitAt":"2026-11-10T10:41:21Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Insomnia.Core-13.4.0.exe","sizeBytes":649769152,"url":"https://github.com/Kong/insomnia/releases/download/core%4013.4.0/Insomnia.Core-13.4.0.exe","releaseUrl":"https://github.com/Kong/insomnia/releases/tag/core%4013.4.0"},{"os":"mac","file":"Insomnia.Core-13.4.0.dmg","sizeBytes":401444007,"url":"https://github.com/Kong/insomnia/releases/download/core%4013.4.0/Insomnia.Core-13.4.0.dmg","releaseUrl":"https://github.com/Kong/insomnia/releases/tag/core%4013.4.0"},{"os":"linux","file":"Insomnia.Core-13.4.0.AppImage","sizeBytes":652072586,"url":"https://github.com/Kong/insomnia/releases/download/core%4013.4.0/Insomnia.Core-13.4.0.AppImage","releaseUrl":"https://github.com/Kong/insomnia/releases/tag/core%4013.4.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop app for developers to send test requests to web APIs, read the answers and keep them organized.","summary":"Insomnia is a desktop app for people who build or use web APIs, the connections programs use to talk to each other. You can send requests (REST, GraphQL, gRPC, WebSockets and more), read the responses, write tests, design API specs and share collections with a team. Projects can stay fully on your computer, in a Git repository, or in the maker's cloud. There are installers for Windows, Mac and Linux. Without an account you only get a local Scratch Pad; most features need a free account, and some need a paid plan.","easeRationale":"All three platforms have ready downloads you just open: Windows installers (Insomnia.Core-13.4.0.exe and an NSIS .exe), a Mac .dmg, and .deb, .rpm, .snap and AppImage files for Linux. You can start with the local Scratch Pad, though most features ask you to create a free account.","costNote":"Free plan; paid subscription plans add things like unlimited collaboration, Git Sync, organizations and SAML/OIDC logins.","notVerified":["What usage data the app sends and whether you can turn it off.","Exactly which features stay free and which need a paid plan today.","Whether the app works fully offline with Local Vault and no account.","Real disk use after install; the Windows installers are about 530 to 620 MB.","Whether the bundled local AI model runner is used by default and what it downloads."]},"zh-Hans":{"summaryShort":"一款桌面应用,供开发者向网络 API 发送测试请求、查看返回结果并保持条理。","summary":"Insomnia 是一款面向构建或使用网络 API(程序之间互相通信的连接方式)的人群的桌面应用。你可以发送请求(REST、GraphQL、gRPC、WebSockets 等),查看返回结果,编写测试,设计 API 规范,并与团队共享集合。项目可以完全留在你自己的电脑上、放在 Git 仓库里,或放在厂商的云端。Windows、Mac 和 Linux 都有安装包。不登录账号只能使用本地的 Scratch Pad;大多数功能需要一个免费账号,部分功能需要付费计划。","easeRationale":"三个平台都有现成的下载包,打开即可:Windows 安装程序(Insomnia.Core-13.4.0.exe 和一个 NSIS .exe)、Mac 的 .dmg,以及 Linux 的 .deb、.rpm、.snap 和 AppImage 文件。你可以先用本地的 Scratch Pad,不过大多数功能会要求你创建一个免费账号。","costNote":"有免费计划;付费订阅计划增加了无限协作、Git Sync、组织账户和 SAML/OIDC 登录等功能。","notVerified":["应用会发送哪些使用数据,是否可以关闭。","目前到底哪些功能保持免费,哪些需要付费计划。","应用是否能靠 Local Vault 完全离线使用,且不需要账号。","安装后实际占用的磁盘空间;Windows 安装包大约 530 到 620 MB。","内置的本地 AI 模型运行器是否默认启用,以及它会下载什么内容。"]}}},{"id":"R_kgDONeluaA","slug":"krillinai__opencreator","name":"OpenCreator","owner":"krillinai","fullName":"krillinai/OpenCreator","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"D"},"risk":"attention","signals":[{"code":"downloads_videos_from_platforms","severity":"attention","evidence":"其中一个工具会下载属于他人的视频:\"Analyze individual public videos from YouTube, Bilibili, X, TikTok, Instagram, Douyin, Facebook, Xiaohongshu, Pinterest, and Reddit; compare available formats and download video or audio. Some sources may require platform cookies.\"架构图中列出使用的工具是yt-dlp。YouTube的条款禁止未经许可下载视频,重新使用或重新配音他人的视频需要获得该视频的权利。","source":"ai","title":"从各平台下载视频"},{"code":"ai_agent_acts_on_your_computer","severity":"info","evidence":"该应用运行一个可访问你文件的AI代理:\"Enter a task and choose the permission level, Profile, model, and reasoning effort\",时间线会显示\"tool calls, file changes, approvals\"。README还提醒它会使用你现有的Codex配置:\"OpenCreator uses the active $CODEX_HOME by default, so confirm the impact before changing global Skills or MCP configuration.\"","source":"ai","title":"AI代理在你的电脑上执行操作"},{"code":"paid_subscription_inside_the_app","severity":"info","evidence":"这个开源应用在销售开发者自己服务的套餐:\"Subscription plans are selected inside OpenCreator. Purchasing opens Stripe Checkout, and managing a subscription opens Stripe Billing Portal\",登录时\"opens https://www.open-creator.ai/device in your browser\"。","source":"ai","title":"应用内有付费订阅"},{"code":"updates_bundled_components_from_internet","severity":"info","evidence":"该应用会获取内置工具的新版本:\"Open Settings → Third-party Components to inspect the yt-dlp nightly version currently in use... OpenCreator checks for updates every seven days but never installs them automatically.\"","source":"ai","title":"从网上更新内置组件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","ffmpeg"],"disk":{"minMB":514,"maxMB":1543,"basis":"installer"},"apiKeys":["OpenAI / ChatGPT sign-in for Codex","Volcengine (Doubao speech)","Google Gemini","Kling"],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","audio","ai_agents"],"alternativeTo":[{"slug":"rask-ai","name":"Rask AI","confirmed":true},{"slug":"4k-video-downloader","name":"4K Video Downloader","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"runway","name":"Runway","confirmed":true}],"declaredAlternatives":[],"stars":14292,"downloads":81044,"downloadsWeek":1460,"downloadsDays":2,"growth":137,"growthDays":6,"history":[{"date":"2026-11-08","stars":14237},{"date":"2026-11-09","stars":14267},{"date":"2026-11-10","stars":14292}],"firstPublishedAt":"2026-11-07T22:56:16.825Z","commitSha":"b5b9235cedbb363d1cc19bdfa3b5ab20b92f651f","verifiedAt":"2026-11-07T22:41:51.439Z","lastCommitAt":"2026-11-08T22:26:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop workspace that translates and dubs videos, downloads clips, and makes images, voiceovers and posts with AI.","summary":"OpenCreator, formerly KrillinAI, is a desktop workspace for people who publish videos. It transcribes, translates and dubs a video with subtitles in over 120 languages, downloads public videos from ten platforms, generates images, thumbnails and short AI videos, writes articles and scripts, and reads text aloud. You can click through the tools or ask an AI assistant to do the work, since it runs on top of Codex CLI. There are ready installers for Windows and macOS; everything runs on your machine, but the AI models need a sign-in or API keys.","easeRationale":"On Windows there is a 514 MB installer (OpenCreator-3.3.0-win-x64.exe) with Codex CLI included, and after installing you still have to sign in with a ChatGPT account or paste an API key before any AI task runs; macOS has the same path with .dmg files for Apple Silicon and Intel. Linux gets no desktop app, only KrillinAI CLI and Server archives, or running the web version from source with Node.js 22, pnpm and the Codex CLI.","costNote":"The program is free and open source (Apache-2.0). Subscriptions are sold inside the app: \"Purchasing opens Stripe Checkout\", with credits for the official model services. You can instead use your own ChatGPT sign-in or provider API keys, which you pay per use.","notVerified":["Whether the installer really includes every media tool (ffmpeg, yt-dlp, Whisper) or downloads parts on first use","Real disk use after the first launch, on top of the 514 MB installer","How much the model providers cost for a full video translation with dubbing","Whether the translation and dubbing quality holds up in the language pair you need","Whether the video downloader still works on each of the ten platforms"]},"zh-Hans":{"summaryShort":"桌面工作台,可翻译和配音视频、下载视频片段,并用AI制作图片、配音和帖子。","summary":"OpenCreator(原名KrillinAI)是面向发布视频用户的桌面工作台。它可以为视频转录、翻译并配音,字幕支持120多种语言,从十个平台下载公开视频,生成图片、缩略图和AI短视频,撰写文章和脚本,并朗读文字。你可以点击工具操作,也可以让AI助手完成工作,因为它运行在Codex CLI之上。Windows和macOS都有现成的安装包;所有内容都在你自己的电脑上运行,但AI模型需要登录或API密钥。","easeRationale":"Windows上有一个514MB的安装包(OpenCreator-3.3.0-win-x64.exe),内置Codex CLI,安装后仍需用ChatGPT账号登录或粘贴API密钥才能运行任何AI任务;macOS的流程相同,提供适用于Apple Silicon和Intel的.dmg文件。Linux没有桌面应用,只有KrillinAI的CLI和Server压缩包,或者用Node.js 22、pnpm和Codex CLI从源代码运行网页版。","costNote":"该程序免费且开源(Apache-2.0协议)。应用内会出售订阅服务:\"Purchasing opens Stripe Checkout\"(购买会打开Stripe结账页面),提供官方模型服务的额度。你也可以改用自己的ChatGPT登录或其他服务商的API密钥,按使用量付费。","notVerified":["安装包是否真的包含所有媒体工具(ffmpeg、yt-dlp、Whisper),还是首次使用时再下载","首次启动后的实际磁盘占用,在514MB安装包之外","完整翻译并配音一段视频,模型服务商要收多少费用","你需要的语言对,翻译和配音质量是否过关","视频下载器在十个平台上是否都还能正常使用"]}}},{"id":"R_kgDOVSLamg","slug":"l0ng-ai__openidea","name":"OpenIDEA","owner":"l0ng-ai","fullName":"l0ng-ai/openidea","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"12 天内获得 5053 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"very_early_version_of_a_big_program","severity":"info","evidence":"该仓库创建于 2026-10-21，在不到两周内发布了 30 个版本后已经到了 0.1.34（Cargo.toml: version = \"0.1.34\"），还有 222 个未解决的 issue。一个完整的编程工作区在 0.1 版本时很可能仍有缺漏。","source":"ai","title":"大型程序的极早期版本"},{"code":"speed_comparison_measured_by_the_project","severity":"info","evidence":"README 表格中 1.1 秒的启动时间和 169 MB 的内存占用，对比 IntelliJ IDEA 的 13.2 秒和 2011 MB，是项目自己在一台电脑上测量的（\"Windows 11, 32-core AMD, 62 GB RAM\"），并注明 \"IDEA analyses Python more deeply, so 'ready' doesn't mean the same work on both sides\"。","source":"ai","title":"由项目自己测量的速度对比"},{"code":"copies_the_look_of_a_paid_product","severity":"info","evidence":"该项目自称是 \"The IDE you know, rebuilt in Rust\"，带有 \"IntelliJ IDEA-style keymap\"，甚至有一个导入 IntelliJ 主题的脚本（`theme:import-intellij`），同时又声明 \"Not affiliated with JetBrains. IntelliJ IDEA is a trademark of JetBrains s.r.o.\"。","source":"ai","title":"模仿付费产品的外观"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","rust"],"disk":{"minMB":53,"maxMB":160,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"jetbrains-ides","name":"IDEs","confirmed":true},{"slug":"datagrip","name":"DataGrip","confirmed":true},{"slug":"postman","name":"Postman","confirmed":true}],"declaredAlternatives":[],"stars":8383,"downloads":178452,"downloadsWeek":105728,"downloadsDays":7,"growth":2773,"growthDays":7,"history":[{"date":"2026-11-02","stars":5001},{"date":"2026-11-03","stars":5610},{"date":"2026-11-04","stars":6110},{"date":"2026-11-05","stars":6722},{"date":"2026-11-06","stars":7414},{"date":"2026-11-07","stars":7647},{"date":"2026-11-08","stars":7823},{"date":"2026-11-09","stars":8144},{"date":"2026-11-10","stars":8383}],"firstPublishedAt":"2026-11-02T03:04:39.362Z","commitSha":"4f64ca3149c0c91a0c8c92060c20ef50dad5139f","verifiedAt":"2026-11-03T02:55:37.467Z","lastCommitAt":"2026-11-10T21:57:11Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"OpenIDEA_0.1.40_x64-setup.exe","sizeBytes":52023410,"url":"https://github.com/l0ng-ai/openidea/releases/download/v0.1.40/OpenIDEA_0.1.40_x64-setup.exe","releaseUrl":"https://github.com/l0ng-ai/openidea/releases/tag/v0.1.40"},{"os":"mac","file":"OpenIDEA_0.1.40_aarch64.dmg","sizeBytes":62422471,"url":"https://github.com/l0ng-ai/openidea/releases/download/v0.1.40/OpenIDEA_0.1.40_aarch64.dmg","releaseUrl":"https://github.com/l0ng-ai/openidea/releases/tag/v0.1.40"},{"os":"linux","file":"OpenIDEA_0.1.40_amd64.AppImage","sizeBytes":142580216,"url":"https://github.com/l0ng-ai/openidea/releases/download/v0.1.40/OpenIDEA_0.1.40_amd64.AppImage","releaseUrl":"https://github.com/l0ng-ai/openidea/releases/tag/v0.1.40"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free programming workspace that copies IntelliJ IDEA's layout and shortcuts, written in Rust to start faster.","summary":"A desktop program for writing code that follows the habits of IntelliJ IDEA: the same keyboard shortcuts, the same tool windows and the same search-everywhere. It includes an editor with code completion, a debugger, Git, a terminal and built-in database and HTTP clients, and says it covers more than 60 languages. Ready installers exist for Windows, macOS and Linux and nothing else needs installing. It is for programmers, is free under Apache 2.0, and is still at version 0.1, about two weeks old.","easeRationale":"Every system has a ready download: an installer (.exe or .msi) on Windows, a .dmg on macOS, and .deb, .rpm or AppImage files on Linux, each around 40 to 120 MB, with nothing else to install first.","costNote":"Free and open source under Apache 2.0.","notVerified":["Whether the Windows and macOS installers are signed by the publisher, since the project says nothing about it.","Whether the speed and memory numbers hold on an ordinary computer and on your own projects.","How complete the code understanding really is for each of the 60+ languages at this early version.","Whether the debugger, database and HTTP tools work for your setup.","How much disk space it uses after it indexes a large project, beyond the installer size."]},"zh-Hans":{"summaryShort":"一个免费的编程工作区，照搬 IntelliJ IDEA 的布局和快捷键，用 Rust 编写，启动更快。","summary":"一款用来写代码的桌面程序，沿用 IntelliJ IDEA 的使用习惯：相同的键盘快捷键、相同的工具窗口和相同的全局搜索。它包含带代码补全的编辑器、调试器、Git、终端，以及内置的数据库和 HTTP 客户端，并称支持 60 多种语言。Windows、macOS 和 Linux 都有现成的安装包，无需再安装其他东西。它面向程序员，采用 Apache 2.0 协议免费提供，目前仍是 0.1 版本，上线约两周。","easeRationale":"每个系统都有现成的下载：Windows 上是安装程序（.exe 或 .msi），macOS 上是 .dmg，Linux 上是 .deb、.rpm 或 AppImage 文件，大小各在 40 到 120 MB 左右，事先不需要安装任何其他东西。","costNote":"免费开源，采用 Apache 2.0 协议。","notVerified":["Windows 和 macOS 的安装包是否由发布者签名，因为项目对此只字未提。","速度和内存的数据在普通电脑和你自己的项目上是否也成立。","在这个很早期的版本里，对 60 多种语言中每一种的代码理解到底有多完整。","调试器、数据库和 HTTP 工具在你的环境里是否能用。","为大型项目建立索引后，除安装包大小之外还会占用多少磁盘空间。"]}}},{"id":"R_kgDOIkq04Q","slug":"langbot-app__langbot","name":"LangBot","owner":"langbot-app","fullName":"langbot-app/LangBot","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"runs_code_and_tools_the_model_chooses","severity":"attention","evidence":"Dockerfile 编译了 Google 的 nsjail,以便\"the Box runtime isolate code without a host Docker socket\",并安装了 Node,\"so the sandbox (nsjail/Docker box) can run npx-based stdio MCP servers\"。因此,机器人可以在对话过程中执行代码并启动外部工具;Docker CLI 也被打包进来,用于可选的 box 服务,该服务\"can drive the mounted host Docker socket\",如果你开启这个选项,就相当于把容器交给了你机器上的 Docker 控制权。","source":"ai","title":"运行模型自行选择的代码和工具"},{"code":"plugins_from_a_marketplace","severity":"attention","evidence":"README 宣传\"Hundreds of plugins\",并在 space.langbot.app 提供了一个 Plugin Market,plugins/README.md 是\"Official plugin source index\",并注明\"Plugins are installed separately and are not bundled in Core releases\"。你安装的插件代码来自第三方,并在你的机器人内部运行。","source":"ai","title":"来自市场的插件"},{"code":"connects_personal_wechat_and_qq_accounts","severity":"attention","evidence":"支持平台表格列出了\"WeChat | Personal & Official Account\"和\"QQ | Personal & Official API (Channel, DM, Group)\",依赖项中包括\"gewechat-client\"和\"nakuru-project-idk\"。因此,机器人可以绑定到个人账号,而不仅仅是官方机器人账号;每个聊天服务对此都有自己的规则。","source":"ai","title":"连接个人微信和 QQ 账号"},{"code":"management_api_also_open_to_assistants","severity":"info","evidence":"README 说 LangBot\"exposes a built-in Model Context Protocol endpoint at /mcp, mirroring the HTTP API so an agent can manage bots, pipelines, plugins, and models programmatically\",并且你可以\"Authenticate with the same API key ... no login flow required\"。任何拿到这个密钥的人都可以管理整个安装。","source":"ai","title":"管理 api 同样向助手开放"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["A model provider such as OpenAI, Anthropic or DeepSeek (a local model through Ollama or LM Studio replaces it)","A bot account on each chat platform you connect (Discord, Telegram, Slack, WeCom, Lark, DingTalk and so on)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatbase","name":"Chatbase","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":18883,"downloads":31879,"downloadsWeek":787,"downloadsDays":6,"growth":154,"growthDays":6,"history":[{"date":"2026-11-04","stars":18729},{"date":"2026-11-05","stars":18744},{"date":"2026-11-06","stars":18765},{"date":"2026-11-07","stars":18791},{"date":"2026-11-08","stars":18820},{"date":"2026-11-09","stars":18854},{"date":"2026-11-10","stars":18883}],"firstPublishedAt":"2026-11-06T23:09:33.818Z","commitSha":"5ef31d965d695843eae095b513a536267e1eefe9","verifiedAt":"2026-11-06T23:04:44.793Z","lastCommitAt":"2026-11-10T19:32:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs your own AI chat bot on Discord, Telegram, Slack, WeChat and more, set up from a panel in the browser.","summary":"LangBot is a server you run yourself that turns an AI model into a chat bot on messaging apps: Discord, Telegram, Slack, LINE, QQ, WeChat, WeCom, Lark, DingTalk, KOOK and Matrix. You choose a model (OpenAI, Anthropic, DeepSeek, Gemini, or a local one through Ollama), can feed it your documents as a knowledge base, and manage it all in a browser panel instead of editing files. It starts with one command through uv, or with Docker Compose. You still need a bot account on each chat app and, for most models, a paid key.","easeRationale":"Windows, Mac and Linux take the same route and get the same grade: install the uv tool, run \"uvx langbot\", then open http://localhost:5300 and finish in the browser panel, where you paste the model key and connect each chat app. Docker Compose is the other documented route and is heavier. There is no installer to double-click, and no release carries a binary.","costNote":"The program is free under Apache-2.0. The models usually cost per use (OpenAI, Anthropic, DeepSeek and the listed gateways); Ollama or LM Studio on your own machine avoids that. The README also points to a version hosted by the maintainers: \"LangBot Cloud (Recommended) ... Zero deployment, ready to use\" at space.langbot.app/cloud.","notVerified":["Whether the connector for each chat app still works, since those platforms change their rules and APIs often","What the model calls would actually cost for your message volume","What LangBot Cloud, the hosted version the README recommends, includes or charges","Whether a personal WeChat or QQ account stays usable while a bot drives it","Real disk and memory use once a knowledge base with its vector database is filled","What any plugin or MCP server you install from the market does inside your bot"]},"zh-Hans":{"summaryShort":"在 Discord、Telegram、Slack、微信等平台运行你自己的 AI 聊天机器人,通过浏览器面板完成设置。","summary":"LangBot 是一个你自己运行的服务器,它把一个 AI 模型变成即时通讯应用上的聊天机器人:Discord、Telegram、Slack、LINE、QQ、微信、企业微信、飞书、钉钉、KOOK 和 Matrix。你可以选择一个模型(OpenAI、Anthropic、DeepSeek、Gemini,或通过 Ollama 运行的本地模型),可以把你的文档作为知识库提供给它,并在浏览器面板里管理这一切,而不用编辑文件。它通过 uv 用一条命令启动,或者用 Docker Compose。你仍然需要在每个聊天应用上有一个机器人账号,并且大多数模型需要付费密钥。","easeRationale":"Windows、Mac 和 Linux 走相同的路线,得到相同的评级:安装 uv 工具,运行\"uvx langbot\",然后打开 http://localhost:5300,在浏览器面板里完成设置,在那里粘贴模型密钥并连接每个聊天应用。Docker Compose 是另一条有文档说明的路线,但更繁琐。没有可以双击安装的安装程序,任何发行版本里都不带二进制文件。","costNote":"该程序基于 Apache-2.0 协议免费提供。模型通常按使用量收费(OpenAI、Anthropic、DeepSeek 和列出的网关);在自己的机器上使用 Ollama 或 LM Studio 可以避免这一点。README 还提到了一个由维护者托管的版本:\"LangBot Cloud (Recommended) ... Zero deployment, ready to use\",地址是 space.langbot.app/cloud。","notVerified":["每个聊天应用的连接器是否仍然有效,因为这些平台经常更改规则和 API","按你的消息量计算,调用模型实际会花多少钱","README 推荐的托管版本 LangBot Cloud 包含什么、收费多少","个人微信或 QQ 账号在被机器人操作时是否仍然可用","知识库连同其向量数据库被填满后,实际的磁盘和内存使用情况","从市场安装的任何插件或 MCP 服务器在你的机器人里具体做了什么"]}}},{"id":"R_kgDOI7jmAw","slug":"langflow-ai__langflow","name":"Langflow","owner":"langflow-ai","fullName":"langflow-ai/langflow","type":"framework_infra","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"C"},"risk":"attention","signals":[{"code":"runs_python_code_without_a_sandbox_by_default","severity":"attention","evidence":"README 提到“Source code access lets you customize any component using Python”，而 pyproject.toml 把隔离描述为可选项：“Opt-in microVM sandbox backend for code-execution components (LANGFLOW_SANDBOX_BACKEND=exec-sandbox, issue #12029). Requires Python >= 3.12 and QEMU 8+ on the host; on older Pythons the extra installs nothing”。因此，你从别人那里导入的流程可能会在你的电脑上运行代码，除非你把这个选项打开。","source":"ai","title":"默认在没有沙箱的情况下运行 Python 代码"},{"code":"ssrf_protection_off_by_default","severity":"attention","evidence":".env.example：“SSRF (Server-Side Request Forgery) Protection — When enabled, blocks requests to private IP ranges, localhost, and cloud metadata endpoints ... Default: false (for backward compatibility)”。保持出厂设置的话，流程可能被指向你自己网络内部的地址。","source":"ai","title":"SSRF 防护默认关闭"},{"code":"flows_can_be_run_without_signing_in","severity":"info","evidence":".env.example 记录了“LANGFLOW_PUBLIC_FLOW_RATE_LIMIT_PER_MINUTE: Number of anonymous public-flow runs allowed per minute per IP”，默认值为 20，因此被发布为公开的流程，任何能访问到该服务器的人都可以运行，并消耗你的 API 额度。","source":"ai","title":"无需登录即可运行流程"},{"code":"login_required_in_official_builds","severity":"info","evidence":"README：“Official release packages and tagged source archives disable automatic login by default. Set `LANGFLOW_SUPERUSER_PASSWORD` before starting”，而“Development branches retain automatic login by default”。如果你从开发分支运行，在你设置密码之前是没有密码的。","source":"ai","title":"官方版本需要登录"},{"code":"very_large_dependency_list","severity":"info","evidence":"pyproject.toml 默认引入 16 个服务商组件包（lfx-ibm、lfx-openai、lfx-anthropic、lfx-google、lfx-slack 等），并带有一组专门为避开已公布漏洞而加的最低版本要求（“oauthlib>=4.0.0,<5.0.0  # CVE-2026-49264, CVE-2026-49265”、“fsspec>=2026.6.0  # CVE-2026-104851”等）。检查估算磁盘占用为 335–1005 MB。","source":"ai","title":"依赖项清单非常庞大"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","node","gpu_any"],"disk":{"minMB":335,"maxMB":1005,"basis":"installer"},"apiKeys":["OpenAI","Anthropic","Langflow (its own key, for calling your flows from other programs)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","automation","development"],"alternativeTo":[{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"make","name":"Make","confirmed":true}],"declaredAlternatives":[],"stars":156512,"downloads":230264,"downloadsWeek":3228,"downloadsDays":7,"growth":209,"growthDays":7,"history":[{"date":"2026-10-26","stars":156042},{"date":"2026-10-27","stars":156072},{"date":"2026-10-28","stars":156117},{"date":"2026-10-29","stars":156147},{"date":"2026-10-30","stars":156172},{"date":"2026-10-31","stars":156203},{"date":"2026-11-01","stars":156231},{"date":"2026-11-02","stars":156268},{"date":"2026-11-03","stars":156303},{"date":"2026-11-04","stars":156332},{"date":"2026-11-05","stars":156365},{"date":"2026-11-06","stars":156396},{"date":"2026-11-07","stars":156418},{"date":"2026-11-08","stars":156444},{"date":"2026-11-09","stars":156478},{"date":"2026-11-10","stars":156512}],"firstPublishedAt":"2026-10-27T01:10:12.653Z","commitSha":"aeb10d96bb2ea797696613af931e91ddf51f3320","verifiedAt":"2026-10-30T01:59:26.720Z","lastCommitAt":"2026-11-10T21:31:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Build AI agents and workflows by connecting boxes on a canvas, then run them as an API on your own machine.","summary":"Langflow is a platform where you draw an AI workflow instead of writing it: you drag boxes for models, databases, files and tools onto a canvas, join them, and try the result in a built-in playground. Each finished flow can then be served as an API or as an MCP tool for other programs. It is aimed at developers, but the Desktop version for Windows and macOS includes everything needed, so no Python setup is required. You still have to connect your own AI service, and most of those charge for use.","easeRationale":"On Windows and macOS there is Langflow Desktop, and the README says \"All dependencies are included, so you don't need to manage Python environments or install packages manually\" — after installing you still have to paste a key for an AI service. On Linux there is no desktop build: you install Python 3.10–3.14 and the `uv` tool, run `uv pip install langflow -U`, set a `LANGFLOW_SUPERUSER_PASSWORD` and then `uv run langflow run` before opening http://127.0.0.1:7860.","costNote":"The project is free and MIT-licensed. Flows only do something once you connect an AI service (OpenAI, Anthropic, Google and many others), and those bill you for what you use; a model running locally through Ollama is the free route.","notVerified":["Whether the Desktop app installs and runs on your version of Windows or macOS, since Radar did not run it","What a flow downloaded from someone else does when you open and run it","The real monthly cost, which depends on the AI service, vector database and search providers you connect","Real disk use after the optional document, audio and local-model extras are installed","Whether your installation ends up reachable from outside your machine, which depends on how you start it and on your network"]},"zh-Hans":{"summaryShort":"在画布上连接方框来搭建 AI 智能体和工作流,然后在你自己的电脑上把它们作为 API 运行。","summary":"Langflow 是一个让你用画而不是用写来完成 AI 工作流的平台:把代表模型、数据库、文件和工具的方框拖到画布上,把它们连起来,再在内置的试玩区里测试结果。做好的工作流可以作为 API 提供服务,也可以作为 MCP 工具供其他程序调用。它主要面向开发者,但面向 Windows 和 macOS 的 Desktop 版本已经包含所需的一切,不需要配置 Python。不过你仍然要自己接入 AI 服务,而大多数服务都是按使用量收费的。","easeRationale":"Windows 和 macOS 上有 Langflow Desktop,README 里说“所有依赖都已包含,你不需要管理 Python 环境或手动安装软件包”——安装完之后,你仍然需要粘贴一个 AI 服务的密钥。Linux 上没有桌面版:要先安装 Python 3.10–3.14 和 `uv` 工具,运行 `uv pip install langflow -U`,设置 `LANGFLOW_SUPERUSER_PASSWORD`,然后运行 `uv run langflow run`,再打开 http://127.0.0.1:7860。","costNote":"该项目免费,MIT 许可证。工作流只有在你接入 AI 服务(OpenAI、Anthropic、Google 等许多服务)后才会真正运作,这些服务会按使用量收费;通过 Ollama 在本地运行模型是免费的路线。","notVerified":["Desktop 应用是否能在你的 Windows 或 macOS 版本上安装并运行,因为 Radar 并未实际运行过","打开并运行别人分享的工作流时,它会做什么","实际的月度费用,这取决于你接入的 AI 服务、向量数据库和搜索提供商","安装可选的文档、音频和本地模型扩展后,实际占用多少磁盘空间","你的安装最终是否能从你的电脑之外访问到,这取决于你启动它的方式和你的网络环境"]}}},{"id":"R_kgDOJVxJug","slug":"langgenius__dify","name":"Dify","owner":"langgenius","fullName":"langgenius/dify","type":"framework_infra","status":"active","grade":"B","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"license_with_extra_restrictions","severity":"attention","evidence":"LICENSE：一份“modified version of the Apache License 2.0”——未经书面许可，“you may not use the Dify source code to operate a multi-tenant environment”，“you may not remove or modify the LOGO or copyright information”，并且“The producer can adjust the open-source agreement to be more strict or relaxed”。","source":"ai","title":"许可证附带额外限制"},{"code":"agents_run_commands","severity":"info","evidence":"README.md：“Autonomous agents with a sandbox of their own: they run commands, install software, and handle files to get open-ended tasks done.”","source":"ai","title":"智能体会执行命令"},{"code":"design_patent_claim","severity":"info","evidence":"LICENSE：“The interactive design of this product is protected by appearance patent.”","source":"ai","title":"声称拥有外观专利"}],"platforms":["windows","mac","linux","web"],"tools":["docker"],"buildTools":["node"],"disk":{"minMB":84,"maxMB":253,"basis":"installer"},"apiKeys":["At least one AI model provider (OpenAI, Anthropic and others) or a self-hosted model"],"cost":{"model":"freemium"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["ai_agents","automation","development"],"alternativeTo":[{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"chatbase","name":"Chatbase","confirmed":true},{"slug":"langsmith","name":"LangSmith","confirmed":true}],"declaredAlternatives":[],"stars":160877,"downloads":28036,"downloadsWeek":1921,"downloadsDays":7,"growth":534,"growthDays":7,"history":[{"date":"2026-10-26","stars":159631},{"date":"2026-10-27","stars":159731},{"date":"2026-10-28","stars":159848},{"date":"2026-10-29","stars":159933},{"date":"2026-10-30","stars":160030},{"date":"2026-10-31","stars":160110},{"date":"2026-11-01","stars":160176},{"date":"2026-11-02","stars":160262},{"date":"2026-11-03","stars":160343},{"date":"2026-11-04","stars":160407},{"date":"2026-11-05","stars":160479},{"date":"2026-11-06","stars":160568},{"date":"2026-11-07","stars":160627},{"date":"2026-11-08","stars":160681},{"date":"2026-11-09","stars":160774},{"date":"2026-11-10","stars":160877}],"firstPublishedAt":"2026-10-27T01:10:12.654Z","commitSha":"42a2184f4c1ebc72930abab1e8793cb583f8ccde","verifiedAt":"2026-10-30T01:45:06.453Z","lastCommitAt":"2026-11-10T17:07:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A web workspace for building AI chatbots, assistants and automated workflows by joining blocks on a canvas.","summary":"Dify is a platform for building apps powered by AI models. In a web page you drag blocks onto a canvas to make chatbots, workflows and AI agents, feed them your own documents (PDFs, slides and more), connect hundreds of AI models and tools, and watch the logs once they run. It is for teams and makers who want to build AI apps with little code. You can use the hosted Dify Cloud, or run it yourself with Docker on a machine with at least 2 CPU cores and 4 GB of RAM. Either way you need an AI model provider, usually a paid API key.","easeRationale":"In the browser it is easiest: Dify Cloud is hosted for you, so you only create an account and add a model key. Running it on your own Windows, Mac or Linux machine is much harder: you need Docker with Docker Compose 2.24 or later, you copy a settings file and run \"docker compose up -d\" in a terminal, and the only files in the releases are difyctl, a command-line helper.","costNote":"The self-hosted Community Edition is free. Dify Cloud has paid plans (see dify.ai/pricing) and there is a paid Enterprise edition. AI model usage is paid separately to the provider you connect.","notVerified":["Whether Docker Desktop on Windows runs the full set of Dify services without extra tuning.","Real disk and memory use once all containers, databases and documents are loaded.","Real AI cost, which depends on the model provider and how much the apps are used.","Whether the default settings in .env.example are strong enough for a server open to the internet.","What the new difyctl command-line helper in the releases installs and changes on your machine."]},"zh-Hans":{"summaryShort":"一个网页工作区,通过在画布上拼接模块来构建AI聊天机器人、助手和自动化工作流程。","summary":"Dify是一个用于构建由AI模型驱动的应用的平台。你在网页上把模块拖到画布上,制作聊天机器人、工作流程和AI代理,给它们提供你自己的文档(PDF、幻灯片等),连接数百个AI模型和工具,并在运行后查看日志。它适合想用较少代码构建AI应用的团队和开发者。你可以使用托管的Dify Cloud,也可以在至少有2个CPU核心和4GB内存的机器上用Docker自己运行。无论哪种方式,你都需要一个AI模型提供商,通常要用付费的API密钥。","easeRationale":"在浏览器里最简单:Dify Cloud已经帮你托管好了,你只需要创建一个账号并添加模型密钥。在自己的Windows、Mac或Linux电脑上运行要难得多:你需要安装Docker Compose 2.24或更新版本,复制一个配置文件,并在终端运行\"docker compose up -d\",而发布包里唯一的文件是difyctl,一个命令行辅助工具。","costNote":"自行部署的社区版是免费的。Dify Cloud有付费方案(见dify.ai/pricing),还有一个付费的企业版。AI模型的使用费用要单独支付给你连接的提供商。","notVerified":["Windows上的Docker Desktop是否能在不额外调整的情况下运行Dify的全部服务。","所有容器、数据库和文档都加载完成后,实际的磁盘和内存占用情况。","实际的AI费用,这取决于所用的模型提供商以及应用的使用量。",".env.example里的默认设置是否足够应对一个对外开放的服务器。","发布包里新增的difyctl命令行工具会在你的电脑上安装和改动哪些内容。"]}}},{"id":"R_kgDOONK_fA","slug":"lauriewired__ghidramcp","name":"GhidraMCP","owner":"LaurieWired","fullName":"LaurieWired/GhidraMCP","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"taking_other_peoples_software_apart_is_restricted","severity":"attention","evidence":"README 自己的话:\"ghidraMCP is an Model Context Protocol server for allowing LLMs to autonomously reverse engineer applications\",功能包括\"Decompile and analyze binaries\"。反编译程序对恶意软件分析和审计自己的代码来说是正常工作,但大多数商业许可证禁止这样做,一些国家也有限制,所以请确认你被允许打开什么。","source":"ai","title":"拆解他人软件受到限制"},{"code":"installs_from_the_live_branch_not_a_release","severity":"attention","evidence":"README 中每个客户端配置都运行\"uvx --from git+https://github.com/LaurieWired/GhidraMCP ghidra-mcp\",每次都会下载仓库分支的当前状态,而不是你检查过的那个发行版。最后一次推送的内容就是在你机器上运行的东西。","source":"ai","title":"从实时分支而不是发行版安装"},{"code":"plugin_opens_a_local_port_with_no_password_in_the_docs","severity":"info","evidence":"该插件运行一个\"GhidraMCP HTTP Server\",默认使用 localhost:8080,README 说\"The server IP and port are configurable and should be set to point to the target Ghidra instance\"。说明中没有提到任何密码或令牌,所以你机器上任何能访问该端口的程序都可以操控 Ghidra。","source":"ai","title":"插件开放一个本地端口,文档中没有提到密码"},{"code":"the_model_edits_your_analysis","severity":"info","evidence":"列出的一个功能是\"Automatically rename methods and data\",无头模式接受\"--project DIR to keep the Ghidra project (and save changes) between runs\"。助手会写入你的 Ghidra 项目,而不仅仅是从中读取。","source":"ai","title":"模型会修改你的分析"}],"platforms":["windows","mac","linux"],"tools":["python","java"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["A model, through the MCP client you already use (the key stays in that client, not in this project)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"ida-pro","name":"IDA Pro","confirmed":true},{"slug":"hopper","name":"Hopper","confirmed":true}],"declaredAlternatives":[],"stars":11879,"downloads":90082,"downloadsWeek":2147,"downloadsDays":4,"growth":120,"growthDays":6,"history":[{"date":"2026-11-06","stars":11790},{"date":"2026-11-07","stars":11823},{"date":"2026-11-08","stars":11843},{"date":"2026-11-09","stars":11864},{"date":"2026-11-10","stars":11879}],"firstPublishedAt":"2026-11-06T22:57:16.709Z","commitSha":"1e980252087b2479a21b9ca6d82d5829538651a7","verifiedAt":"2026-11-06T22:38:23.187Z","lastCommitAt":"2026-10-10T19:17:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant drive Ghidra to pull a program apart: decompile it, list its functions and rename them.","summary":"GhidraMCP plugs Ghidra, the free program for taking software apart, into AI assistants that speak MCP, such as Claude Code or VS Code. The assistant can then decompile a file you opened in Ghidra, list its methods, classes, imports and exports, and rename things on its own, so you can ask questions about a program in plain language. It is made for security researchers and developers who already use Ghidra: you install Ghidra 12.1.4 and this extension, then point your assistant at the plugin's local server. A headless mode can also analyze a file with no Ghidra window open.","easeRationale":"Same grade on all three systems, because the route is the same and it is long: install Ghidra 12.1.4 (which needs Java), add the extension zip from the release through File -> Install Extensions, restart Ghidra, enable the plugin under Configure -> Developer, install the uv tool, then paste a JSON block with the command and the port into your AI client. On macOS the headless mode asks for one extra step, building Ghidra's decompiler with \"support/gradle/gradlew buildNatives\".","costNote":"The extension and the server are free under Apache-2.0, and Ghidra itself is free. The AI assistant that drives it is the paid part: every decompiled function it reads is billed by your model provider or counted against your plan.","notVerified":["Whether it works with a Ghidra version other than the 12.1.4 the README pins","How much a long session costs, since each decompiled function is sent to the model","Whether the names and explanations the model produces are correct","Whether the plugin's HTTP server can be reached from outside your machine on a given network","How well the headless mode handles large files and how much memory it needs"]},"zh-Hans":{"summaryShort":"让 AI 助手操作 Ghidra 来拆解一个程序:反编译、列出其函数并重命名它们。","summary":"GhidraMCP 把 Ghidra(用于拆解软件的免费程序)接入支持 MCP 的 AI 助手,比如 Claude Code 或 VS Code。助手可以反编译你在 Ghidra 中打开的文件,列出它的方法、类、导入项和导出项,并自行重命名,这样你就能用简单的语言询问一个程序的情况。它是为已经在使用 Ghidra 的漏洞研究人员和开发者做的:你安装 Ghidra 12.1.4 和这个扩展,然后把助手指向插件的本地服务器。无头模式还可以在不打开 Ghidra 窗口的情况下分析文件。","easeRationale":"三个系统的评级相同,因为路线相同且很长:安装 Ghidra 12.1.4(需要 Java),通过 File -> Install Extensions 从发行版添加扩展 zip 包,重启 Ghidra,在 Configure -> Developer 下启用插件,安装 uv 工具,然后把包含命令和端口的 JSON 代码块粘贴到你的 AI 客户端里。在 macOS 上,无头模式还需要额外一步,用\"support/gradle/gradlew buildNatives\"来构建 Ghidra 的反编译器。","costNote":"该扩展和服务器基于 Apache-2.0 协议免费提供,Ghidra 本身也是免费的。驱动它的 AI 助手才是付费的部分:它读取的每一个反编译函数都会由你的模型提供商计费,或计入你的套餐额度。","notVerified":["是否适用于 README 固定的 12.1.4 以外的 Ghidra 版本","一次长会话要花多少钱,因为每个反编译的函数都会发送给模型","模型生成的名称和解释是否正确","在特定网络中,插件的 HTTP 服务器是否可以从你的机器之外访问到","无头模式处理大文件的效果如何,需要多少内存"]}}},{"id":"R_kgDONN8FYg","slug":"leaperone__multipost-extension","name":"MultiPost","owner":"leaperone","fullName":"leaperone/MultiPost-Extension","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"reads_and_changes_every_website_you_visit","severity":"attention","evidence":"package.json 里的 manifest 申请了\"host_permissions\": [\"https://*/*\"]以及\"tabs\"、\"scripting\"、\"clipboardWrite\"和\"clipboardRead\"这些权限。这意味着它可以读取并修改任何网站上的任何页面,也可以读取你复制的内容,而不仅仅是它发布内容的那些社交网络。","source":"ai","title":"读取并修改你访问的每一个网站"},{"code":"outside_pages_can_command_the_extension","severity":"attention","evidence":"manifest 设置了\"externally_connectable\": { \"matches\": [\"https://paiban.md/*\", \"http://localhost:2663/*\"] },所以 paiban.md 上的页面可以向你浏览器里的扩展发送指令。这个域名不是项目自己的网站(multipost.app),README 里也完全没有解释这一点。","source":"ai","title":"外部页面可以向扩展下达指令"},{"code":"the_version_in_the_stores_is_not_this_code","severity":"attention","evidence":"README 自己的\"Open-source snapshot\"部分写道:\"The MultiPost monorepo builds and publishes the official Chrome and Edge packages, which also include the proprietary Agent module. This repository maintains source code only and does not run release workflows.\"也就是说,你安装的东西是用你看不到的代码构建出来的。","source":"ai","title":"商店里的版本不是这份代码"},{"code":"posts_by_driving_your_logged_in_accounts","severity":"attention","evidence":"功能列表承诺以\"No login, no registration, no API Key required\"的方式在十多个平台上发布内容,这意味着扩展是在你的会话里直接向网站输入内容,而不是使用官方 API。各网络都把自动发帖视为违反规则的行为,可能会限制甚至关闭账号。","source":"ai","title":"通过操控你已登录的账号发帖"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"later","name":"Later","confirmed":true}],"declaredAlternatives":[],"stars":3900,"downloads":2665,"downloadsWeek":143,"downloadsDays":6,"growth":32,"growthDays":6,"history":[{"date":"2026-11-04","stars":3868},{"date":"2026-11-05","stars":3872},{"date":"2026-11-06","stars":3879},{"date":"2026-11-07","stars":3884},{"date":"2026-11-08","stars":3892},{"date":"2026-11-09","stars":3896},{"date":"2026-11-10","stars":3900}],"firstPublishedAt":"2026-11-06T22:59:23.801Z","commitSha":"5522ca01100a58572de3a652aebd7851f5fa648a","verifiedAt":"2026-11-06T22:39:13.299Z","lastCommitAt":"2026-10-24T15:29:46Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Posts the same text, pictures or video to more than ten social networks at once, from your browser.","summary":"MultiPost is a Chrome and Edge extension for people who publish the same post everywhere. You write it once in the extension's panel, pick the networks, and it fills in and sends the post on each site using the accounts you are already logged into in that browser, so there is no sign-up and no API key. The README lists over ten networks, most of them Chinese (Zhihu, Weibo, Xiaohongshu, Douyin) plus TikTok and YouTube. There is also an API so your own web app or a script can ask the extension to publish.","easeRationale":"Wherever Chrome or Edge runs, the install is one click from the browser's store, so Windows, Mac and Linux all get an A; the README links both store pages. Building it from this repository instead is a developer job with Node and pnpm, because the GitHub releases carry no files and the repository itself says it \"does not run release workflows\".","costNote":"The extension is free in the Chrome and Edge stores and this repository is Apache-2.0. The README says no login, registration or API key is needed, and points to a separate company address for \"Agent 模型账户\" (Agent model accounts), which this repository does not cover.","notVerified":["Whether each of the listed networks still works, since the extension depends on the layout of their pages","Whether the extension in the stores behaves like the code in this repository","What the proprietary Agent module does and what it sends out","Whether the RESTful API route passes your content through someone else's server","Whether your accounts stay in good standing when posts are sent this way"]},"zh-Hans":{"summaryShort":"从浏览器一次性把同一条文字、图片或视频发布到十多个社交网络。","summary":"MultiPost 是一款面向到处发布同一条内容的人群的 Chrome 和 Edge 扩展。你在扩展面板里写一次,选择要发布的网络,它就会利用你在该浏览器里已经登录的账号,在每个网站上填写并发送帖子,所以不需要注册,也不需要 API 密钥。README 列出了十多个网络,大多是中国的(知乎、微博、小红书、抖音),再加上 TikTok 和 YouTube。它还提供了一个 API,方便你自己的网页应用或脚本请求扩展发布内容。","easeRationale":"无论 Chrome 还是 Edge,安装都只需在浏览器商店里点一下,所以 Windows、Mac 和 Linux 都得到 A 级评分;README 附上了两个商店页面的链接。而从这个仓库自己构建,则是需要 Node 和 pnpm 的开发者工作,因为 GitHub 的发行版本不带任何文件,仓库自己也说它\"does not run release workflows\"。","costNote":"该扩展在 Chrome 和 Edge 商店里是免费的,这个仓库采用 Apache-2.0 协议。README 说不需要登录、注册或 API 密钥,并指向一个单独的公司地址,用于\"Agent 模型账户\",这个仓库不涉及这部分内容。","notVerified":["列出的每个网络是否仍然有效,因为该扩展依赖这些页面的布局","商店里的扩展行为是否和这个仓库里的代码一致","专有的 Agent 模块做了什么,它向外发送了什么内容","RESTful API 路由是否会让你的内容经过别人的服务器","以这种方式发帖时,你的账号是否能保持良好状态"]}}},{"id":"R_kgDOUcJK8g","slug":"leemysw__yovoice","name":"YoVoice","owner":"leemysw","fullName":"leemysw/yovoice","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"copies_a_real_persons_voice","severity":"attention","evidence":"README 提供了\"zero-shot voice cloning\"功能,以及\"clone a reference voice, design a voice from a description, adjust emotion and delivery\"这项特性。一段简短的录音就足以让那个声音说出你输入的任何内容,而未经本人同意使用其声音在多个国家都受到限制。","source":"ai","title":"复制真人的声音"},{"code":"models_are_non_commercial_only","severity":"attention","evidence":"Supported Models 部分写道:\"OmniVoice, YuE2 and Breeze TTS 2 weights are restricted to non-commercial use; Stable Audio 3 is free below USD 1M annual revenue under the Stability AI Community License\",许可说明还补充\"IndexTTS models are covered separately by their model license\"。Apache-2.0 标志只涵盖这个程序本身,并不代表每个模型都允许你拿去销售。","source":"ai","title":"部分模型仅限非商业用途"},{"code":"models_are_downloaded_after_install","severity":"info","evidence":"\"All models are downloaded inside the App or with the CLI\",再加上已经达到 785 MB 的安装程序,以及在 Linux 上通过 Settings 获取的 Vulkan GPU 运行时。应用最终占用多少空间,取决于你添加了多少个语音和音乐模型。","source":"ai","title":"安装后还需要下载模型"},{"code":"can_serve_other_machines_over_the_network","severity":"info","evidence":"\"Run yovoice serve to provide an authenticated HTTP API and MCP inference service for other machines.\"README 说这个服务需要身份验证;在你启动它之前,它是关闭的。","source":"ai","title":"可以通过网络为其他机器提供服务"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["go"],"disk":{"minMB":785,"maxMB":2356,"basis":"installer"},"apiKeys":["DeepSeek, Qwen, Kimi, OpenAI, Anthropic or a local Ollama, only for the optional assistant panel"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true}],"declaredAlternatives":[],"stars":2158,"downloads":30186,"downloadsWeek":2600,"downloadsDays":4,"growth":96,"growthDays":6,"history":[{"date":"2026-11-06","stars":2090},{"date":"2026-11-07","stars":2105},{"date":"2026-11-08","stars":2123},{"date":"2026-11-09","stars":2142},{"date":"2026-11-10","stars":2158}],"firstPublishedAt":"2026-11-06T22:59:23.808Z","commitSha":"18a7ddc8a89837e9ff3a6f1208a71565c0aa1297","verifiedAt":"2026-11-06T22:40:33.322Z","lastCommitAt":"2026-11-10T10:40:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes speech, copied voices, songs with lyrics and MIDI scores on your own computer, with no online service.","summary":"yovoice is a desktop studio for voice and music that works on your own machine. It reads your text aloud in a built-in voice, copies a voice from a short recording, dubs a story line by line with a character per line, writes songs from a style and your lyrics, and arranges scores you can export as MIDI or separate tracks. There are installers for Windows, macOS on Apple Silicon and Linux; you choose and download the models inside the app. An optional panel chats with DeepSeek, OpenAI, Anthropic or a local Ollama to write lyrics and start the work for you.","easeRationale":"Windows is the easiest route and sets the grade: run the setup .exe (a plain 78 MB one, or a 610 MB / 785 MB build that already carries the NVIDIA CUDA runtime), then do the one extra step the Quick Start asks for, \"Download a model in Settings\" — that is a B. macOS is the same with the .dmg, but only macOS 14 or newer on Apple Silicon. Linux has a .deb that also pulls in WebKitGTK and runs on the processor out of the box, with the note that \"The Linux app doesn't update itself yet\".","costNote":"The program is free under Apache-2.0 and the models run on your machine, so there is no per-character or per-minute bill. The optional assistant panel uses a key you bring yourself, and some model weights carry their own limits (see the signals).","notVerified":["Whether generation is fast enough on a machine with no NVIDIA card, since Linux runs on the processor out of the box","How much disk the app really takes once several voice and music models are downloaded","Whether the voices and songs come out like the demo page shows","Whether the Vulkan runtime offered in Settings works with your graphics card on Linux","What the optional assistant panel sends to the provider you choose"]},"zh-Hans":{"summaryShort":"在你自己的电脑上生成语音、克隆的声音、带歌词的歌曲和 MIDI 乐谱,无需联网服务。","summary":"yovoice 是一个在你自己机器上运行的语音和音乐桌面工作室。它可以用内置声音朗读你的文本,从一段简短的录音中复制一个声音,按每行一个角色逐行给故事配音,根据一种风格和你的歌词创作歌曲,并编排可以导出为 MIDI 或独立音轨的乐谱。它提供适用于 Windows、Apple 芯片版 macOS 和 Linux 的安装程序;你在应用内选择并下载模型。一个可选的面板可以和 DeepSeek、OpenAI、Anthropic 或本地的 Ollama 对话,帮你写歌词并替你开始工作。","easeRationale":"Windows 是最简单的路线,决定了评级:运行 .exe 安装程序(一个 78 MB 的简单版本,或者一个已经带有 NVIDIA CUDA 运行时的 610 MB / 785 MB 版本),然后完成 Quick Start 要求的那一个额外步骤,\"Download a model in Settings\"——这是 B 级。macOS 使用 .dmg 也是一样,但只支持 Apple 芯片上的 macOS 14 或更新版本。Linux 有一个 .deb,同时会拉取 WebKitGTK,并且开箱即用地跑在处理器上,并附有说明\"The Linux app doesn't update itself yet\"。","costNote":"该程序基于 Apache-2.0 协议免费提供,模型在你自己的机器上运行,所以没有按字符或按分钟计费。可选的助手面板使用你自己提供的密钥,一些模型权重有各自的限制(见信号部分)。","notVerified":["在没有 NVIDIA 显卡的机器上生成速度是否足够快,因为 Linux 默认跑在处理器上","下载了多个语音和音乐模型之后,应用实际占用多少磁盘空间","生成的声音和歌曲是否和演示页面展示的一样","Settings 里提供的 Vulkan 运行时是否能配合你在 Linux 上的显卡工作","可选的助手面板会把什么内容发送给你选择的提供商"]}}},{"id":"R_kgDOUsqjIA","slug":"lemomo-ai__lemo-opuscar","name":"Lemo OpusCar","owner":"lemomo-ai","fullName":"lemomo-ai/lemo-opuscar","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"37 天内获得 11185 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"agent_installs_and_downloads_on_first_use","severity":"attention","evidence":"README.md：\"On first use it downloads the guides, tools and style prompts to `~/lemo-opuscar`\" 以及 \"the agent installs the rest\"；需求文件指向 `sh plugin/skills/lemo-opuscar/scripts/setup.sh deps` 和 `sh tools/fetch.sh instruments <lib>` 来获取音乐样本库。也就是说 AI 在工作过程中会在你的电脑上获取并安装软件和媒体内容。","source":"ai","title":"代理在首次使用时安装和下载内容"},{"code":"long_runs_cost_tokens","severity":"info","evidence":"README.md「Before you start」：\"A film takes an agent about 30–60 minutes and a fair amount of tokens.\" 费用会计入你自己的 Claude 计划。","source":"ai","title":"长时间运行会消耗 token"},{"code":"tied_to_one_model","severity":"info","evidence":"README 两次提醒：\"Built for Claude. The styles are tuned for Claude Opus 5.5; other models may not reproduce them well, so use them with care.\"","source":"ai","title":"绑定单一模型"},{"code":"mixed_licences_on_the_demo_material","severity":"info","evidence":"许可部分把代码列为 MIT，演示影片列为 CC BY 4.0，但排除了三条语音轨道（macOS `say`、Microsoft Edge TTS 和 Volcengine 条款下的 Doubao TTS）：\"Don't reuse them\"。文中还说 \"You are responsible for the materials you use in your films.\"","source":"ai","title":"演示素材采用多种混合许可"}],"platforms":["windows","mac","linux"],"tools":["python","node","ffmpeg","other"],"buildTools":["python","node","ffmpeg"],"disk":{"minMB":200,"maxMB":800,"basis":"heuristic"},"apiKeys":["Anthropic (Claude Code with Opus 5.5)","Microsoft Edge TTS (optional, online voices for Chinese narration)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","ai_skills"],"alternativeTo":[{"slug":"after-effects","name":"After Effects","confirmed":true},{"slug":"remotion","name":"Remotion","confirmed":true}],"declaredAlternatives":[],"stars":14447,"downloads":429284,"downloadsWeek":51804,"downloadsDays":7,"growth":2714,"growthDays":7,"history":[{"date":"2026-10-26","stars":8079},{"date":"2026-10-27","stars":8673},{"date":"2026-10-28","stars":9101},{"date":"2026-10-29","stars":9417},{"date":"2026-10-30","stars":9775},{"date":"2026-10-31","stars":10110},{"date":"2026-11-01","stars":10623},{"date":"2026-11-02","stars":11156},{"date":"2026-11-03","stars":11733},{"date":"2026-11-04","stars":12229},{"date":"2026-11-05","stars":12622},{"date":"2026-11-06","stars":13003},{"date":"2026-11-07","stars":13356},{"date":"2026-11-08","stars":13678},{"date":"2026-11-09","stars":14043},{"date":"2026-11-10","stars":14447}],"firstPublishedAt":"2026-10-27T01:10:12.655Z","commitSha":"bebbbe1aa92a3cb4bb6f83500e61b23a17d1aa00","verifiedAt":"2026-11-03T02:42:05.272Z","lastCommitAt":"2026-11-10T01:28:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Asks Claude Code to make a short film on your topic in one of 66 ready styles, drawn entirely in code.","summary":"Lemo-Opuscar is a plugin for Claude Code with 66 film styles (watercolour, ukiyo-e, pixel RPG, news package and more). You install it, say what film you want and in which style, and the AI writes the code that draws every frame, scores the music and adds narration, ending in a finished video. No video or image generator is involved. You need Claude Code with the Opus 5.5 model, Node 20+, ffmpeg 7+ and Python 3.11+, and one film takes the agent 30 to 60 minutes.","easeRationale":"On macOS and Linux you need several tools in a terminal before anything happens: Claude Code, Node 20+, ffmpeg 7+ and Python 3.11+, then two `claude plugin` commands; Ubuntu 24.04's ffmpeg is too old and the README sends you to a manual static build. Windows is the same list, and the setup and download helpers are shell scripts (`setup.sh`, `fetch.sh`), which means Git Bash or WSL.","costNote":"The styles and code are free (MIT; demo films CC BY 4.0). The work is done by Claude Code, which needs a paid plan or API credit, and the README warns a film costs \"a fair amount of tokens\".","notVerified":["Whether a film in a given style comes out as good as the demos, which were made by the author","What the Claude usage really costs for one film","Whether it works on Windows, since the helper scripts are written for a Unix shell","What the agent downloads and installs on first use, including the music sample libraries","Whether you are free to publish the result, which depends on the voices, fonts and data you feed it"]},"zh-Hans":{"summaryShort":"让 Claude Code 用 66 种现成风格之一制作关于你主题的短片，全部由代码绘制。","summary":"Lemo-Opuscar 是 Claude Code 的一个插件，带有 66 种影片风格（水彩、浮世绘、像素 RPG、新闻集锦等）。你安装它，说出想要的影片和风格，AI 就会编写代码来绘制每一帧画面、配乐并加入旁白，最终生成一个完整的视频。整个过程不涉及任何视频或图像生成器。你需要安装好 Opus 5.5 模型的 Claude Code、Node 20+、ffmpeg 7+ 和 Python 3.11+，代理制作一部影片需要 30 到 60 分钟。","easeRationale":"在 macOS 和 Linux 上，你需要先在终端里准备好几个工具才能开始：Claude Code、Node 20+、ffmpeg 7+ 和 Python 3.11+，然后运行两个 `claude plugin` 命令；Ubuntu 24.04 自带的 ffmpeg 版本太旧，README 会让你手动静态编译。Windows 需要同样的一套工具，而安装和下载用的辅助脚本是 shell 脚本（`setup.sh`、`fetch.sh`），这意味着你需要 Git Bash 或 WSL。","costNote":"风格和代码是免费的（MIT 许可；演示影片采用 CC BY 4.0）。实际工作由 Claude Code 完成，需要付费计划或 API 额度，README 提醒制作一部影片会花费 \"a fair amount of tokens\"。","notVerified":["某种风格制作出的影片是否能达到演示效果，而演示是作者本人做的","制作一部影片实际会花费多少 Claude 用量","它是否能在 Windows 上运行，因为辅助脚本是为 Unix shell 写的","代理在首次使用时会下载和安装什么，包括音乐样本库","你是否可以自由发布制作出的作品，这取决于你提供的配音、字体和数据"]}}},{"id":"R_kgDOQIXrng","slug":"lemony-ai__cascadeflow","name":"CascadeFlow","owner":"lemony-ai","fullName":"lemony-ai/cascadeflow","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"the_savings_numbers_come_from_the_maintainers","severity":"info","evidence":"README 开头就说\"Cost Savings: 69% (MT-Bench), 93% (GSM8K), 52% (MMLU), 80% (TruthfulQA) savings, retaining 96% GPT-5 quality\",后面又说\"Result: 40-85% cost reduction, 2-10x faster responses, zero quality loss\"。仓库里带有示例和测试,但没有你可以运行来重现这些数字的基准测试。","source":"ai","title":"节省数据来自维护者自己"},{"code":"the_cheap_models_answer_is_kept_when_the_check_passes","severity":"attention","evidence":"按照设计,它\"Speculatively executes small, fast models first\",只有在验证失败时才会升级,而验证包括长度、\"Confidence scoring (logprobs analysis)\"、格式,以及一个默认值为\"semanticThreshold: 0.5\"的可选相似度分数。这些阈值能否在你的主题领域里抓住一个错误答案,取决于你自己;小模型给出的回答会被当作最终答案交付。","source":"ai","title":"检查通过时保留便宜模型的答案"},{"code":"the_cost_report_uses_prices_you_type_in","severity":"info","evidence":"价格是手工写在配置里的,例如 ModelConfig(name=\"gpt-5\", provider=\"openai\", cost=0.00562),示例由此打印出 result.total_cost。一旦某个提供商改变价格,它报告的数字就会和你的实际账单产生偏差。","source":"ai","title":"成本报告使用你自己输入的价格"},{"code":"the_published_package_is_older_than_the_code","severity":"info","evidence":"最新的发行版本是 2026-04-02 的 v1.2.0,pyproject.toml 里也写着版本 1.2.0,而仓库一直改动到 2026-09-25。\"pip install cascadeflow\"给你的是四月的那个包,而不是你在这里读到的内容。","source":"ai","title":"发布的包比代码旧"},{"code":"gateway_mode_has_no_password_in_the_instructions","severity":"info","evidence":"对于已有的应用,README 提供了一个即插即用的网关:\"python -m cascadeflow.server --mode auto --port 8084\",并加载了你的提供商密钥。说明中没有提到这个端口有任何令牌或密码,所以任何能访问到它的东西都可以用你的密钥去消费。","source":"ai","title":"网关模式在说明中没有设置密码"}],"platforms":["windows","mac","linux"],"tools":["python","node"],"buildTools":["python","node"],"disk":{"minMB":200,"maxMB":800,"basis":"heuristic"},"apiKeys":["At least one model provider: OpenAI, Anthropic, Groq, Together AI, Hugging Face, Google, Azure OpenAI, DeepSeek or OpenRouter (not needed if both models are local, in Ollama or vLLM)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"portkey","name":"Portkey","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"langsmith","name":"LangSmith","confirmed":true}],"declaredAlternatives":[],"stars":3925,"downloads":189,"downloadsWeek":16,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":3925},{"date":"2026-11-07","stars":3926},{"date":"2026-11-08","stars":3926},{"date":"2026-11-09","stars":3926},{"date":"2026-11-10","stars":3925}],"firstPublishedAt":"2026-11-06T22:59:23.815Z","commitSha":"280ac9728aae622be0f9892b7e9239786ee4e3c0","verifiedAt":"2026-11-06T22:41:51.828Z","lastCommitAt":"2026-09-25T07:32:05Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A code library that tries a cheap AI model first and only passes the question to an expensive one when needed.","summary":"cascadeflow is a library for Python and TypeScript (plus an n8n node) that sits inside the AI calls of a program you are writing. It asks a small, cheap model first, checks the answer against rules you set, and only sends the question on to a big model when the check fails. It also adds up the cost of each call, can hold a run to a budget, and can refuse a tool call or stop an agent. It talks to OpenAI, Anthropic, Groq, Together, Google, DeepSeek and to local models in Ollama or vLLM, and plugs into LangChain, CrewAI, PydanticAI and Google ADK. You have to write code to use it.","easeRationale":"C on Windows, Mac and Linux alike: \"pip install cascadeflow\" (or \"npm install @cascadeflow/core\"), a provider key in a .env file, and about ten lines of your own code listing the cheap model and the expensive one. There is nothing to open and no interface; the n8n node is the closest thing to a point-and-click route.","costNote":"The library is free under the MIT licence. What you pay is the model providers behind it, which is the whole point of the project; running the small model on your own machine with Ollama or vLLM costs nothing.","notVerified":["Whether the savings hold for your own prompts and models","Whether the quality check spots a wrong answer in your subject before it reaches your users","The real bill at the end of the month, since the figures it prints come from the prices you entered","Whether the package on PyPI and npm matches the code in this repository","The claimed overhead of under 5 milliseconds per step"]},"zh-Hans":{"summaryShort":"一个代码库,先尝试用便宜的 AI 模型,只有在需要时才把问题交给昂贵的模型。","summary":"cascadeflow 是一个面向 Python 和 TypeScript 的库(还有一个 n8n 节点),嵌入在你正在编写的程序的 AI 调用中。它先向一个小而便宜的模型提问,按你设定的规则检查答案,只有检查不通过时才把问题交给一个大模型。它还会累计每次调用的费用,可以把一次运行控制在预算之内,还可以拒绝一次工具调用或停止一个智能体。它可以对接 OpenAI、Anthropic、Groq、Together、Google、DeepSeek,以及 Ollama 或 vLLM 里的本地模型,并接入 LangChain、CrewAI、PydanticAI 和 Google ADK。你需要自己写代码才能使用它。","easeRationale":"Windows、Mac 和 Linux 都是 C 级:\"pip install cascadeflow\"(或者\"npm install @cascadeflow/core\"),在 .env 文件里放一个提供商密钥,再加上大约十行列出便宜模型和昂贵模型的自己的代码。没有任何东西可以打开,也没有界面;n8n 节点是最接近点选式操作的方式。","costNote":"该库基于 MIT 协议免费提供。你花钱的地方是它背后的模型提供商,这正是这个项目的意义所在;用 Ollama 或 vLLM 在自己的机器上运行小模型则不花一分钱。","notVerified":["这种节省是否适用于你自己的提示词和模型","质量检查能否在错误答案到达你的用户之前,在你的主题领域里发现它","月底的实际账单,因为它打印出来的数字来自你自己输入的价格","PyPI 和 npm 上的包是否和这个仓库里的代码一致","所宣称的每一步低于 5 毫秒的开销"]}}},{"id":"R_kgDOVKgi0A","slug":"leonxlnx__earshot","name":"Earshot","owner":"Leonxlnx","fullName":"Leonxlnx/earshot","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"low","signals":[{"code":"very_new_alpha_project","severity":"info","evidence":"创建于 2026-10-15（距本次检查 11 天）；pyproject.toml 中的分类标记为 'Development Status :: 3 - Alpha'。","source":"ai","title":"非常新的 alpha 项目"},{"code":"models_downloaded_on_first_use","severity":"info","evidence":"README 的 Install 部分：'The first model use downloads two open models (~1.1 GB)'，即 CLAP laion/larger_clap_general 和 MIT/ast-finetuned-audioset，两者都 'at a pinned revision with safetensors weights'。","source":"ai","title":"首次使用时下载模型"},{"code":"optional_audio_upload_to_gemini","severity":"info","evidence":"README：'`llm FILE --yes-upload` | optional audio-LLM critique (Gemini), only with your own key and explicit consent' —— 这会把音频文件发送给 Google。","source":"ai","title":"可选择把音频上传到 Gemini"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["Google Gemini (GEMINI_API_KEY, only for the optional llm critique)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","audio","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":4441,"downloads":6449,"downloadsWeek":0,"downloadsDays":7,"growth":687,"growthDays":7,"history":[{"date":"2026-10-26","stars":2914},{"date":"2026-10-27","stars":3051},{"date":"2026-10-28","stars":3148},{"date":"2026-10-29","stars":3241},{"date":"2026-10-30","stars":3320},{"date":"2026-10-31","stars":3384},{"date":"2026-11-01","stars":3450},{"date":"2026-11-02","stars":3562},{"date":"2026-11-03","stars":3754},{"date":"2026-11-04","stars":3879},{"date":"2026-11-05","stars":3986},{"date":"2026-11-06","stars":4067},{"date":"2026-11-07","stars":4163},{"date":"2026-11-08","stars":4263},{"date":"2026-11-09","stars":4361},{"date":"2026-11-10","stars":4441}],"firstPublishedAt":"2026-10-27T01:10:12.655Z","commitSha":"591fa3e028032e6dce56620a20e932e35667e8d8","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-06T17:22:43Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Lets AI coding assistants \"listen\" to sound files, describe and compare them, and make sound effects in code.","summary":"earshot turns audio into text and numbers an AI coding assistant can reason about: what a sound is, whether it fits a description, the best moments in a long recording, and a scorecard for a video's soundtrack. It can also create sound effects in code and ships a bank of 252 of them. It is for developers who use agents such as Claude Code, through a command line, a Python library or an MCP server. You need Python 3.10+, ffmpeg and a terminal. It runs on the processor (no graphics card) and downloads about 1.1 GB of models on first use.","easeRationale":"There is no installer: you install Python and ffmpeg, clone the repository, create a virtual environment and run several pip commands in a terminal.","costNote":"The optional Gemini critique uses your own Google key, billed by Google.","notVerified":["Whether the install and the MCP server work on Windows; the README commands mostly use Linux/macOS paths (.venv/bin/python).","Real speed on a typical laptop; the README says each CLI call takes about 20-40 seconds to load the models.","Real disk use after installing torch and the models; the checks estimate 950 MB to 4.5 GB.","How accurate its sound descriptions and soundtrack verdicts are outside the author's own tests."]},"zh-Hans":{"summaryShort":"让 AI 编程助手“听”声音文件、描述并比较它们,还能用代码制作音效。","summary":"earshot 把音频变成 AI 编程助手能够理解的文字和数字:一段声音是什么、是否符合某段描述、长录音中最精彩的片段,以及视频配乐的评分卡。它还能用代码生成音效,并自带包含 252 个音效的音效库。它面向使用 Claude Code 等智能体的开发者,可通过命令行、Python 库或 MCP 服务器使用。你需要 Python 3.10 及以上版本、ffmpeg 和终端。它在处理器上运行(不需要显卡),首次使用时会下载约 1.1 GB 的模型。","easeRationale":"没有安装程序:你需要先安装 Python 和 ffmpeg,克隆代码仓库,创建虚拟环境,然后在终端里运行几条 pip 命令。","costNote":"可选的 Gemini 点评功能使用你自己的 Google 密钥,由 Google 计费。","notVerified":["安装和 MCP 服务器能否在 Windows 上正常工作;README 中的命令大多使用 Linux/macOS 路径(.venv/bin/python)。","在普通笔记本电脑上的实际速度;README 称每次命令行调用加载模型大约需要 20-40 秒。","安装 torch 和模型后实际占用的磁盘空间;检查结果估计在 950 MB 到 4.5 GB 之间。","在作者自己的测试之外,它的声音描述和配乐评判有多准确。"]}}},{"id":"R_kgDORUQ5bw","slug":"leonxlnx__taste-skill","name":"Taste Skill","owner":"Leonxlnx","fullName":"Leonxlnx/taste-skill","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"each_build_burns_a_lot_of_model_tokens","severity":"attention","evidence":"README 的 \"How v2 works\" 一节：\"A build takes roughly 30 to 45 minutes and a few hundred thousand tokens.\" 这部分费用由你运行它所用的智能体或模型收取，而不是本项目。","source":"ai","title":"每次构建都会消耗大量模型 token"},{"code":"paid_referral_link_in_the_readme","severity":"info","evidence":"\"Open Source Supporters\" 表格链接到 `https://platform.kimi.ai?track_id=track-20f0d64bcaba419bb425a3e8c306f65d&aff=taste-skill`，文字为 \"Get a Kimi API key\" 和 \"taste-skill users get 10% bonus API credits\"，因此作者能从这次推荐中获得收益。","source":"ai","title":"README 中有带返利的推荐链接"},{"code":"pages_come_with_third_party_photos_under_share_alike_licences","severity":"info","evidence":"README：图片来自 \"Wikimedia Commons, Openverse and the Met\"，\"Only public domain, CC0, CC BY and CC BY-SA files are offered, and the skill puts the credits on the page.\" CC BY 和 CC BY-SA 会让法律条件附着在你发布的任何内容上，因此在商业使用某个页面之前，请先查看署名信息。","source":"ai","title":"页面附带第三方照片，许可要求相同方式共享"},{"code":"author_warns_about_fake_tokens_using_the_project_name","severity":"info","evidence":"README 页脚：\"Taste Skill has no official token, coin, or crypto project. Any token using my name, image, or project is unaffiliated and not endorsed by me.\"","source":"ai","title":"作者警告有人用项目名称发行假代币"},{"code":"very_young_repository_with_large_star_count","severity":"info","evidence":"创建于 2026-02-19，该仓库已有 104,511 个星标和 7,070 个 fork，只有 5 个未解决的 issue，并且在其他 skill 被迁到别的仓库之后，只保留了一个 skill（`design-taste-frontend`）。","source":"ai","title":"非常新的仓库，星标数量巨大"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","design_cad","development"],"alternativeTo":[{"slug":"v0-by-vercel","name":"v0 by Vercel","confirmed":true},{"slug":"figma","name":"Figma","confirmed":true}],"declaredAlternatives":[],"stars":112961,"downloads":1178,"downloadsWeek":0,"downloadsDays":7,"growth":3476,"growthDays":7,"history":[{"date":"2026-10-29","stars":105556},{"date":"2026-10-30","stars":106350},{"date":"2026-10-31","stars":107048},{"date":"2026-11-01","stars":107848},{"date":"2026-11-02","stars":108736},{"date":"2026-11-03","stars":109485},{"date":"2026-11-04","stars":109985},{"date":"2026-11-05","stars":110583},{"date":"2026-11-06","stars":111184},{"date":"2026-11-07","stars":111597},{"date":"2026-11-08","stars":111998},{"date":"2026-11-09","stars":112489},{"date":"2026-11-10","stars":112961}],"firstPublishedAt":"2026-10-29T01:54:21.380Z","commitSha":"12af471f627f866411b4bd785b19646d3672784c","verifiedAt":"2026-10-29T01:40:04.181Z","lastCommitAt":"2026-11-07T12:33:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A skill that makes Claude Code, Codex or Cursor design a web page with real photos and check its own work in a browser.","summary":"One skill you install inside a coding agent (Claude Code, Codex, Cursor and others). When you ask for a landing page, a dashboard or an app screen, it gives the agent an art direction drawn from a study of 247 sites, finds real photographs in open collections and credits them, then opens the page in a real browser at three widths and fixes the worst problems. For people who already use a coding agent and want pages that do not look generic. It needs Node 22 or newer, Chrome, Edge or Chromium, and internet access to find pictures.","easeRationale":"C on Windows, Mac and Linux: there is nothing to download from the releases, but you need a coding agent already set up, Node 22 or newer and Chrome, Edge or Chromium, and then one command (`npx skills add ...`) or two lines inside Claude Code. The alternative `./install.sh` route needs bash, which on Windows means Git Bash or WSL.","costNote":"The skill is MIT and asks for no API key of its own, but it only works inside a coding agent you pay for. The README says one build takes \"roughly 30 to 45 minutes and a few hundred thousand tokens\". The author also takes GitHub sponsorships.","notVerified":["Whether the pages it builds look like the examples in the README, since those were chosen by the author.","What one build really costs in tokens on your own model and plan.","Whether the licence and credit of every photo it picks is correct, and whether a published page meets the licence terms.","Which sites and image archives the scripts contact while running, and what they send.","Whether the skill installs and loads correctly in each listed agent (Claude Code, Codex, Cursor and the rest)."]},"zh-Hans":{"summaryShort":"一个技能,让Claude Code、Codex或Cursor设计带真实照片的网页,并在浏览器中检查自己的成果。","summary":"这是一个安装在编程智能体(Claude Code、Codex、Cursor等)中的技能。当你要求制作一个落地页、仪表盘或应用界面时,它会给智能体提供基于247个网站研究得出的美术方向,从开放图库中找到真实照片并标注出处,然后在真实浏览器中以三种宽度打开页面,并修复最严重的问题。适合已经在使用编程智能体、又想要页面不显得千篇一律的人。需要Node 22或更新版本、Chrome、Edge或Chromium,以及用于查找图片的联网环境。","easeRationale":"在Windows、Mac和Linux上都是C级:发布页没有可下载的文件,但你需要已经配置好的编程智能体、Node 22或更新版本以及Chrome、Edge或Chromium,然后运行一条命令(`npx skills add ...`)或在Claude Code里输入两行指令。另一种`./install.sh`方式需要bash,在Windows上意味着要用Git Bash或WSL。","costNote":"该技能采用MIT许可,本身不需要API密钥,但只能在你付费使用的编程智能体中运行。README提到一次构建大约\"需要30到45分钟,消耗几十万个token\"。作者也接受GitHub赞助。","notVerified":["它生成的页面是否真像README中的示例,因为那些示例是作者自己挑选的。","在你自己的模型和套餐下,一次构建实际会花费多少token。","它挑选的每张照片的许可和署名是否准确,发布后的页面是否符合这些许可条款。","脚本运行时会联系哪些网站和图片库,以及发送了什么内容。","这个技能是否能在列出的每个智能体(Claude Code、Codex、Cursor及其他)中正确安装和加载。"]}}},{"id":"R_kgDOT0PjyQ","slug":"lexmount__moli","name":"Moli","owner":"lexmount","fullName":"lexmount/moli","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B"},"risk":"low","signals":[{"code":"powershell_pipe_iex","severity":"info","evidence":"README.md，Direct installation：powershell -ExecutionPolicy ByPass -c \"irm https://github.com/lexmount/moli/releases/latest/download/moli-installer.ps1 | iex\" 会直接运行下载的脚本，并绕过 PowerShell 的脚本策略。发布页面也提供普通的 zip 文件。","source":"ai","title":"PowerShell 直接运行下载的脚本"},{"code":"agent_installs_from_repo_instructions","severity":"info","evidence":"README.md，Quick start：建议的提示词让你的 AI 智能体 \"Install the skills under https://github.com/lexmount/moli/tree/main/skills, follow their instructions to download and install the latest prebuilt Moli binary\"，也就是说智能体会照着仓库里写的说明去做。","source":"ai","title":"智能体按仓库里的说明自行安装"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":42,"maxMB":125,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","automation","ai_agents"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":23277,"downloads":44395,"downloadsWeek":6638,"downloadsDays":7,"growth":838,"growthDays":7,"history":[{"date":"2026-10-26","stars":20914},{"date":"2026-10-27","stars":21179},{"date":"2026-10-28","stars":21352},{"date":"2026-10-29","stars":21540},{"date":"2026-10-30","stars":21738},{"date":"2026-10-31","stars":21882},{"date":"2026-11-01","stars":22015},{"date":"2026-11-02","stars":22254},{"date":"2026-11-03","stars":22439},{"date":"2026-11-04","stars":22592},{"date":"2026-11-05","stars":22718},{"date":"2026-11-06","stars":22832},{"date":"2026-11-07","stars":22930},{"date":"2026-11-08","stars":23042},{"date":"2026-11-09","stars":23165},{"date":"2026-11-10","stars":23277}],"firstPublishedAt":"2026-10-27T01:10:12.656Z","commitSha":"73928db1170af114c0f913e421bc4c52c84b4aac","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T22:33:36Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"A browser with no window that AI agents and scripts use to open web pages, read their content and click through sites.","summary":"Moli is a headless browser: it loads websites, runs their JavaScript and returns the content as text, Markdown, screenshots or PDF, without showing a window. It is meant for AI agents, web crawling and test automation, and tools like Playwright, Puppeteer and Selenium can control it. It is used from the command line or by connecting other programs to it, so it is aimed at developers and people who run AI agents. Ready-made versions exist for Windows, macOS and Linux. The company behind it also sells a managed cloud version, which the README says is not needed.","easeRationale":"There is a ready Windows zip (moli-x86_64-pc-windows-msvc.zip, about 42 MB) or one PowerShell install command, but it has no window: you use it by typing commands or by connecting an AI agent or automation script to it.","costNote":"Open source under MIT or Apache-2.0. Lexmount also offers a managed cloud service (Lexmount Browser); the README says the open-source browser is fully usable without it.","notVerified":["Whether the benchmark results in the README (success rates, memory use compared with Chrome) hold on your own tasks","What the cookie-import and browser-profile parts (moli-cookie-import, moli-browser-profile folders) read from your computer","Whether the automation server started by moli serve can be reached by other devices on your network","What the agent skills under skills/ tell an AI agent to run"]},"zh-Hans":{"summaryShort":"一款没有窗口的浏览器,AI 智能体和脚本用它打开网页、读取内容并在网站中点击操作。","summary":"Moli 是一款无头浏览器:它加载网站、运行网站的 JavaScript,并以文本、Markdown、截图或 PDF 的形式返回内容,全程不显示窗口。它面向 AI 智能体、网页抓取和自动化测试,Playwright、Puppeteer 和 Selenium 等工具都可以控制它。它通过命令行使用,或让其他程序连接到它,因此主要面向开发者和运行 AI 智能体的人。它提供适用于 Windows、macOS 和 Linux 的现成版本。其背后的公司还出售托管的云端版本,README 称并不需要它。","easeRationale":"有现成的 Windows 压缩包(moli-x86_64-pc-windows-msvc.zip,约 42 MB),也可以用一条 PowerShell 安装命令,但它没有窗口:你需要输入命令,或把 AI 智能体、自动化脚本连接到它来使用。","costNote":"开源,采用 MIT 或 Apache-2.0 许可。Lexmount 也提供托管云服务(Lexmount Browser);README 称不使用该服务也能完整使用这个开源浏览器。","notVerified":["README 中的基准测试结果(成功率、与 Chrome 相比的内存占用)在你自己的任务上是否同样成立","Cookie 导入和浏览器配置文件相关部分(moli-cookie-import、moli-browser-profile 文件夹)会从你的电脑读取什么","由 moli serve 启动的自动化服务器是否可以被你网络中的其他设备访问","skills/ 下的智能体技能会让 AI 智能体运行什么"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNTk2NTgw","slug":"librecad__librecad","name":"LibreCAD","owner":"LibreCAD","fullName":"LibreCAD/LibreCAD","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":78,"maxMB":234,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"file"},"categories":["design_cad"],"alternativeTo":[{"slug":"autocad","name":"AutoCAD","confirmed":true}],"declaredAlternatives":[],"stars":6576,"downloads":1565503,"downloadsWeek":67136,"downloadsDays":7,"growth":25,"growthDays":7,"history":[{"date":"2026-10-25","stars":6511},{"date":"2026-10-26","stars":6514},{"date":"2026-10-27","stars":6515},{"date":"2026-10-28","stars":6521},{"date":"2026-10-29","stars":6528},{"date":"2026-10-30","stars":6532},{"date":"2026-10-31","stars":6535},{"date":"2026-11-01","stars":6541},{"date":"2026-11-02","stars":6546},{"date":"2026-11-03","stars":6551},{"date":"2026-11-04","stars":6553},{"date":"2026-11-05","stars":6559},{"date":"2026-11-06","stars":6564},{"date":"2026-11-07","stars":6566},{"date":"2026-11-08","stars":6568},{"date":"2026-11-09","stars":6574},{"date":"2026-11-10","stars":6576}],"firstPublishedAt":"2026-10-25T23:19:09.919Z","commitSha":"082637ba5929314788b44925cd4900550d287938","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T16:51:39Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"LibreCAD-v2.2.1.6-win64-msvc.exe","sizeBytes":84583257,"url":"https://github.com/LibreCAD/LibreCAD/releases/download/v2.2.1.6/LibreCAD-v2.2.1.6-win64-msvc.exe","releaseUrl":"https://github.com/LibreCAD/LibreCAD/releases/tag/v2.2.1.6"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Make flat 2D technical drawings, such as floor plans and part sketches, and open or save DXF and DWG files.","summary":"LibreCAD is a desktop program for 2D technical drawing (CAD). You draw lines, circles, arcs and curves with exact measurements, organize them in layers, and add dimensions and text. It opens DXF and DWG drawings and saves to DXF, DWG, PDF and SVG. It is for students, hobbyists and professionals who need flat drawings, not 3D models. It started as a copy of the free edition of QCAD and is under the GPLv2 license. Ready downloads exist for Windows, macOS and Linux, and the interface is translated into over 30 languages.","easeRationale":"The 2.2.1 release includes ready Windows installers (for example LibreCAD-v2.2.1.6-1-g05af8d052-win64-msvc.exe, about 81 MB), so you download and run it.","notVerified":["How well it opens complex DWG files made in AutoCAD; DWG support depends on how the build was made.","Whether the Windows installer is digitally signed or triggers a Windows SmartScreen warning.","How stable the \"continuous\" builds from the development branch are compared with the 2.2.1 builds."]},"zh-Hans":{"summaryShort":"绘制平面 2D 技术图纸，如户型平面图和零件草图，并可打开或保存 DXF 和 DWG 文件。","summary":"LibreCAD 是一款用于 2D 技术制图（CAD）的桌面程序。你可以用精确的尺寸绘制直线、圆、圆弧和曲线，用图层来组织它们，并添加标注和文字。它能打开 DXF 和 DWG 图纸，并可保存为 DXF、DWG、PDF 和 SVG。它适合需要平面图纸（而非 3D 模型）的学生、爱好者和专业人士。它最初由 QCAD 免费版复制而来，采用 GPLv2 许可证。提供适用于 Windows、macOS 和 Linux 的现成下载，界面已翻译成 30 多种语言。","easeRationale":"2.2.1 版本包含现成的 Windows 安装程序（例如 LibreCAD-v2.2.1.6-1-g05af8d052-win64-msvc.exe，约 81 MB），下载后运行即可。","notVerified":["它打开 AutoCAD 制作的复杂 DWG 文件的效果如何；DWG 支持取决于该版本的构建方式。","Windows 安装程序是否有数字签名，或是否会触发 Windows SmartScreen 警告。","开发分支的“continuous”构建与 2.2.1 版本相比稳定性如何。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNDAyMTI4MA==","slug":"libreoffice__core","name":"LibreOffice","owner":"LibreOffice","fullName":"LibreOffice/core","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"source_only_read_only_repo","severity":"info","evidence":"仓库描述：\"Read-only LibreOffice core repo - no pull request (use gerrit instead ...) - don't download zip\"。这个 GitHub 页面只有源代码（0 个发布版本），并不是大家要安装的程序。","source":"ai","title":"仅含源代码的只读仓库"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":358,"maxMB":1073,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[{"slug":"word","name":"Word","confirmed":true},{"slug":"excel","name":"Excel","confirmed":true},{"slug":"powerpoint","name":"PowerPoint","confirmed":true}],"declaredAlternatives":[],"stars":4624,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":32,"growthDays":7,"history":[{"date":"2026-10-25","stars":4558},{"date":"2026-10-26","stars":4561},{"date":"2026-10-27","stars":4569},{"date":"2026-10-28","stars":4575},{"date":"2026-10-29","stars":4580},{"date":"2026-10-30","stars":4583},{"date":"2026-10-31","stars":4584},{"date":"2026-11-01","stars":4590},{"date":"2026-11-02","stars":4591},{"date":"2026-11-03","stars":4592},{"date":"2026-11-04","stars":4596},{"date":"2026-11-05","stars":4603},{"date":"2026-11-06","stars":4610},{"date":"2026-11-07","stars":4617},{"date":"2026-11-08","stars":4617},{"date":"2026-11-09","stars":4624},{"date":"2026-11-10","stars":4624}],"firstPublishedAt":"2026-10-25T23:32:50.450Z","commitSha":"528ba1172ce7f2193db9297a3dac0bc4da7cd9d9","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T21:59:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"A complete office suite to write documents, make spreadsheets and build presentations, and to open most common document formats.","summary":"LibreOffice is a free office suite backed by The Document Foundation. It includes Writer for documents, Calc for spreadsheets, Impress for presentations and Draw for drawings. This GitHub repository is a read-only copy of the source code for developers. Regular users install the ready program: the official Windows installer (winget TheDocumentFoundation.LibreOffice) is a file of about 358 MB, and Radar estimates it needs 358 MB to 1 GB of disk space. The README lists Windows 10, macOS 11 or a recent Linux as the minimum.","easeRationale":"There is an official Windows installer (an MSI of about 358 MB, listed in winget as TheDocumentFoundation.LibreOffice) that you download and open; the README sets Windows 10 as the minimum.","costNote":"The README says LibreOffice is free to download, use and distribute.","notVerified":["Exact disk space after installing: Radar's range of 358 MB to about 1 GB is estimated from the installer size, not measured on a real install","How well complex Word, Excel and PowerPoint files keep their layout","Whether optional features you use need Java installed separately; the README only mentions Java for building the program","The macOS and Linux downloads, since Radar only looked at the Windows installer"]},"zh-Hans":{"summaryShort":"完整的办公套件：写文档、做电子表格和演示文稿，并能打开大多数常见文档格式。","summary":"LibreOffice 是由 The Document Foundation 支持的免费办公套件，包含用于文档的 Writer、用于电子表格的 Calc、用于演示文稿的 Impress 和用于绘图的 Draw。这个 GitHub 仓库是面向开发者的只读源代码副本。普通用户安装现成的程序：官方 Windows 安装程序（winget TheDocumentFoundation.LibreOffice）约 358 MB，Radar 估计需要 358 MB 到 1 GB 的磁盘空间。README 要求至少 Windows 10、macOS 11 或较新的 Linux。","easeRationale":"有官方 Windows 安装程序（约 358 MB 的 MSI，在 winget 中列为 TheDocumentFoundation.LibreOffice），下载后打开即可；README 规定最低为 Windows 10。","costNote":"README 说明 LibreOffice 可以免费下载、使用和分发。","notVerified":["安装后的确切磁盘占用：Radar 给出的 358 MB 到约 1 GB 是根据安装程序大小估算的，并非实际安装测得","复杂的 Word、Excel 和 PowerPoint 文件能否保持原有排版","你使用的可选功能是否需要另外安装 Java；README 只在编译程序时提到 Java","macOS 和 Linux 的下载，因为 Radar 只查看了 Windows 安装程序"]}}},{"id":"R_kgDOS-0Giw","slug":"lidge-jun__opencodex","name":"OpenCodex","owner":"lidge-jun","fullName":"lidge-jun/opencodex","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道传给 shell 执行：curl -fsSL https://bun.sh/install | bash","source":"check","title":"用 curl 管道执行 shell 脚本"},{"code":"providers_may_restrict_accounts_used_through_a_proxy","severity":"attention","evidence":"项目自己的声明：\"Some providers — notably Anthropic (Claude) — may suspend or restrict accounts that route API traffic through third-party proxies. Use at your own risk (UAYOR). Before connecting a provider, review its Terms of Service to confirm that proxy-based access is permitted.\"","source":"ai","title":"提供方可能限制通过代理使用的账号"},{"code":"pooling_accounts_runs_against_provider_limits","severity":"attention","evidence":"一个主打功能是\"manage a ChatGPT account pool for Codex auth. Add multiple ChatGPT / Codex accounts, refresh their 5h / weekly / 30d quota in the dashboard\"，并把新的会话发给用量最少的账号。README 也附上自己的提醒：\"OpenCodex does not endorse using additional accounts to circumvent provider limits or sharing account credentials between people.\"","source":"ai","title":"把多个账号凑在一起用会撞上提供方的限额"},{"code":"holds_every_provider_login_and_sees_every_prompt","severity":"info","evidence":"它把你所有提供方的凭证（\"OAuth for xAI, Anthropic, and Kimi; or forward codex login, paste a key\"）保存在 ~/.opencodex 里，你编程工具发出的每个请求都会经过它，面板里还有\"a live request log\"。在默认的本地地址下，\"opencodex binds to 127.0.0.1 and needs no extra authentication\"；只有当你把它开放到本机以外时，才会强制要求令牌。","source":"ai","title":"保存每个提供方的登录信息也看到每一条提示词"},{"code":"windows_installer_is_not_signed_yet","severity":"info","evidence":"Windows 的发行版表格写着\"Code signing via SignPath Foundation is rolling out ...; until a release is signed, SmartScreen asks once: choose More info → Run anyway\"。发行页面上每个版本文件旁边确实都有一个 .sha256。","source":"ai","title":"windows 安装包还没有签名"},{"code":"one_person_reviews_and_approves_the_code","severity":"info","evidence":"代码签名那一节列出\"Committers and reviewers: lidge-jun\"和\"Approvers: lidge-jun\"——在一个掌管你连接的每个提供方的密钥和登录信息的项目里，审核和批准都是同一个账号。","source":"ai","title":"代码由一个人审核并批准"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":64,"maxMB":192,"basis":"installer"},"apiKeys":["At least one model provider, by key or by login: OpenAI/ChatGPT, Anthropic, Google Gemini, xAI, Kimi, Azure, DeepSeek, OpenRouter and others (local Ollama needs none)","TypeSafe JEV, only for the optional automatic routing combo"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"portkey","name":"Portkey","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true}],"declaredAlternatives":[],"stars":18359,"downloads":253127,"downloadsWeek":27975,"downloadsDays":4,"growth":149,"growthDays":6,"history":[{"date":"2026-11-06","stars":18260},{"date":"2026-11-07","stars":18284},{"date":"2026-11-08","stars":18299},{"date":"2026-11-09","stars":18328},{"date":"2026-11-10","stars":18359}],"firstPublishedAt":"2026-11-06T22:59:23.823Z","commitSha":"cab7eb7d0332ae0ebb4289d7c6bac1de666b6e40","verifiedAt":"2026-11-06T22:43:03.418Z","lastCommitAt":"2026-11-10T22:27:19Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A gateway on your own machine so Codex, Claude Code and similar tools answer with the model you pick.","summary":"opencodex sits on your machine between AI coding tools and model providers. You install it, open a dashboard at localhost:10100, add providers (over 40 built in, or any OpenAI-compatible address), and then Codex, Claude Code, Claude Desktop or Grok Build keep their usual screens while the answers come from Claude, Gemini, Grok, DeepSeek, Kimi, Qwen or a local Ollama. It can also hold several ChatGPT accounts and send new sessions to the one with the most quota left. It ships installers for Windows, macOS and Linux and a command line through npm.","easeRationale":"The desktop app sets the grade on all three systems, and the README says it \"needs neither Node nor Bun\": run the .msi on Windows, the .dmg on macOS 13+, or the .AppImage/.deb on Linux, then do the one real step — add a provider with a key or a login in the dashboard at http://localhost:10100. That app is still marked beta; the other route, \"npm install -g @bitkyc08/opencodex\" plus \"ocx start\", needs Node 18+ and the terminal.","costNote":"The gateway is free under the MIT licence. You pay whoever provides the models: an API key, an existing ChatGPT or Claude plan, or nothing at all if you point it at Ollama on your own machine. Two of the providers offered in the picker (OrcaRouter, TokenLab) are paid services that sponsor the project.","notVerified":["Whether a given provider reacts to traffic coming through a proxy, and what happens to your account if it does","Whether each coding tool keeps working after its own updates, since the connection is not official","How faithfully requests are translated between the three protocols for tool calls, streaming and images","What the dashboard's request log keeps on disk and for how long","How steady the desktop app is while it is marked beta"]},"zh-Hans":{"summaryShort":"在你自己的电脑上搭建一个网关，让 Codex、Claude Code 等工具用你选择的模型来回答。","summary":"opencodex 安装在你的电脑上，处于编程 AI 工具和模型提供方之间。你安装它后，在 localhost:10100 打开一个面板，添加提供方（内置超过 40 个，或任何兼容 OpenAI 的地址），然后 Codex、Claude Code、Claude Desktop 或 Grok Build 还是保持原来的界面，回答其实来自 Claude、Gemini、Grok、DeepSeek、Kimi、Qwen 或本地的 Ollama。它还能保存多个 ChatGPT 账号，把新的会话发给配额剩得最多的那个。它提供 Windows、macOS 和 Linux 的安装包，也可以通过 npm 用命令行安装。","easeRationale":"桌面应用决定了三个系统的评级，README 说它\"needs neither Node nor Bun\"：在 Windows 上运行 .msi，在 macOS 13+ 上运行 .dmg，或者在 Linux 上运行 .AppImage/.deb，然后唯一真正要做的一步——在 http://localhost:10100 的面板里用密钥或登录添加一个提供方。这个应用还标记为测试版；另一条路，\"npm install -g @bitkyc08/opencodex\" 加上 \"ocx start\"，需要 Node 18+ 和终端。","costNote":"这个网关在 MIT 许可下免费使用。你付费的对象是提供模型的一方：一个 API 密钥、已有的 ChatGPT 或 Claude 套餐，或者如果你指向自己电脑上的 Ollama，就什么都不用付。选择器里提供的两个提供方（OrcaRouter、TokenLab）是赞助本项目的付费服务。","notVerified":["某个提供方对经代理转发的流量会有什么反应，如果有反应，你的账号会怎样","连接方式并非官方支持，每个编程工具在自身更新后是否还能继续用","三种协议之间对工具调用、流式传输和图片的转换有多准确","面板的请求日志在磁盘上保存了什么内容，保存多久","这个桌面应用还标记为测试版时，运行起来有多稳定"]}}},{"id":"R_kgDOI67vwg","slug":"lightpanda-io__browser","name":"Lightpanda","owner":"lightpanda-io","fullName":"lightpanda-io/browser","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"telemetry_on_by_default","severity":"attention","evidence":"README 'Telemetry':'By default, Lightpanda collects and sends usage telemetry. This can be disabled by setting an environment variable LIGHTPANDA_DISABLE_TELEMETRY=true.'","source":"ai","title":"遥测默认开启"},{"code":"robots_txt_only_with_flag","severity":"attention","evidence":"README 'Status' 列出了'Respect robots.txt with option --obey-robots',示例命令手动传入了这个参数,所以不加它的话,浏览器也会抓取网站要求爬虫不要碰的页面。自动化采集可能违反网站的使用条款,各国的规则也不同。","source":"ai","title":"只有加上参数才会遵守robots.txt"},{"code":"not_a_full_browser_engine","severity":"info","evidence":"README 'Why Lightpanda?':'No graphical rendering engine','--dump png > page.png'被描述为'a text-only rendering of the page',所以依赖布局或渲染的页面表现可能和Chrome不一样。","source":"ai","title":"不是一个完整的浏览器引擎"},{"code":"vendor_measured_benchmarks","severity":"info","evidence":"'16x lighter and 9x faster than Chromium'这个说法来自项目自己的基准测试表,是在一台AWS实例上针对它自己的演示仓库运行的。","source":"ai","title":"基准测试由厂商自行测量"},{"code":"copyleft_license_and_cla","severity":"info","evidence":"许可证是 AGPL-3.0,仓库顶层有 CLA.md 和 LICENSING.md,CONTRIBUTING 中写道'You must sign our CLA during the pull request process.'把修改后的版本作为公开网络服务运行,就意味着要公开你的修改。","source":"ai","title":"copyleft许可并要求签署CLA"}],"platforms":["linux","mac","windows"],"tools":["docker","node"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenAI","OpenRouter","Anthropic"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["automation","development","ai_agents"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"bright-data","name":"Bright Data","confirmed":true}],"declaredAlternatives":[],"stars":41784,"downloads":186494,"downloadsWeek":18500,"downloadsDays":7,"growth":644,"growthDays":7,"history":[{"date":"2026-11-01","stars":40232},{"date":"2026-11-02","stars":40947},{"date":"2026-11-03","stars":41140},{"date":"2026-11-04","stars":41267},{"date":"2026-11-05","stars":41400},{"date":"2026-11-06","stars":41488},{"date":"2026-11-07","stars":41563},{"date":"2026-11-08","stars":41616},{"date":"2026-11-09","stars":41693},{"date":"2026-11-10","stars":41784}],"firstPublishedAt":"2026-11-01T01:55:44.584Z","commitSha":"cebb8d51e60a68815591a5c7874823b988530aa8","verifiedAt":"2026-11-01T01:41:37.476Z","lastCommitAt":"2026-11-10T11:18:53Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A fast browser with no window, for scraping and AI agents: it loads pages and runs their JavaScript from the terminal.","summary":"Lightpanda is a browser with no window, written from scratch in Zig for automation and AI agents. From a terminal it can dump a page as HTML, Markdown or structured data, or run as a server that Puppeteer, Playwright and WebDriver scripts connect to, and it has an agent mode plus an MCP server so an AI assistant can browse for you. It is for developers and people who automate or scrape websites. It runs on Linux and macOS (Windows only inside WSL), as a downloaded binary or a Docker image. Usage telemetry is on unless you turn it off.","easeRationale":"macOS and Linux are a C: you install it with Homebrew or download one file, make it executable and then type commands such as './lightpanda serve' in a terminal; a Docker image is also offered. Windows is a D, because the README says there is no native Windows build and you have to install WSL2 and use the Linux steps inside it.","costNote":"The browser is free and open source under AGPL-3.0. Only the optional agent mode needs a paid model provider, and it can also run with local models or with --no-llm.","notVerified":["Whether the sites you care about work, since it is a new engine and not a Chromium fork","What the telemetry sends and how to see it","The speed and memory numbers on your own pages and machine","How much disk and memory a running instance really uses","What the optional agent mode costs in model calls"]},"zh-Hans":{"summaryShort":"一个没有窗口的快速浏览器,用于抓取和AI代理:从终端加载页面并运行其JavaScript。","summary":"Lightpanda 是一个没有窗口的浏览器,用 Zig 从零开始编写,面向自动化和AI代理。它可以从终端把页面导出为HTML、Markdown或结构化数据,也可以作为服务器运行,让Puppeteer、Playwright和WebDriver脚本连接,还有一个代理模式加上MCP服务器,让AI助手替你浏览。它适合开发者以及自动化或抓取网站的人。它运行在Linux和macOS上(Windows只能在WSL内运行),可以用下载的二进制文件或Docker镜像。使用遥测默认开启,除非你关闭它。","easeRationale":"macOS 和 Linux 是 C 级:你用 Homebrew 安装,或者下载一个文件、给它执行权限,然后在终端里输入像'./lightpanda serve'这样的命令;也提供了 Docker 镜像。Windows 是 D 级,因为 README 说没有原生 Windows 构建,你需要安装 WSL2,并在里面使用 Linux 的步骤。","costNote":"该浏览器是免费开源的,采用 AGPL-3.0 许可。只有可选的代理模式需要付费的模型服务商,它也可以配合本地模型或使用--no-llm运行。","notVerified":["你关心的网站是否能正常使用,因为这是一个全新的引擎,不是Chromium的分支","遥测发送了什么内容,如何查看","在你自己的页面和机器上的速度和内存数据","一个正在运行的实例实际占用多少磁盘和内存","可选的代理模式在模型调用上要花多少钱"]}}},{"id":"R_kgDORrRGBw","slug":"liketrek__trek","name":"TREK","owner":"liketrek","fullName":"liketrek/TREK","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"place_searches_leave_your_server_by_default","severity":"attention","evidence":"Compose 文件记录了默认开启的 TREK_PLACES_ENABLED：地点搜索\"asks the TREK Places API at https://places.liketrek.com by default, so the server makes outbound requests to that host. Set false ... and send nothing there. No switch in the admin panel on purpose\"。README 也确认了这一点：\"Every TREK installation in the world searches through servers that dataforest cloud sponsors.\"","source":"ai","title":"地点搜索默认会发送到你的服务器之外"},{"code":"the_quick_start_command_leaves_it_on_plain_http","severity":"attention","evidence":"\"Get started in 30 seconds\"那条命令把 COOKIE_SECURE 设为 false，README 自己解释说这\"is only for trying TREK over plain HTTP\"，并说之后要去掉；如果你不设置管理员密码，第一个密码就\"are printed to the container log (docker logs trek)\"。3000 端口会被发布出去，所以照着这条命令原样执行，会让一个未加密的登录页面在你的网络里可以被访问到。","source":"ai","title":"快速启动命令会让系统停留在明文 http 上"},{"code":"third_party_plugins_run_inside_your_installation","severity":"info","evidence":"插件可以\"from the TREK registry or sideload a zip\"来安装。项目描述了围绕插件的防护措施：\"one child process per plugin, 66 grantable permissions, an admin-edited outbound host allowlist, memory and call caps\"，注册表下载\"pinned by sha256 and checked against the author's minisign key\"，手动加载的 zip \"is marked unverified\"，TREK_PLUGINS_ENABLED=false 可以关闭整个系统。","source":"ai","title":"第三方插件在你的安装环境内部运行"},{"code":"the_ai_features_send_your_trip_data_to_the_provider_you_pick","severity":"info","evidence":"助手和\"through KItinerary or an AI model\"的预订导入功能需要\"an API key for Anthropic, OpenAI, Gemini or OpenRouter\"。README 说读取是一次性发生的，而\"every change waits as a card to apply or discard\"，\"Nothing leaves before you agree to the provider, and a file's text only with your consent\"，而且这两个功能在管理员打开之前都是关闭的。","source":"ai","title":"ai 功能会把你的行程数据发给你选择的提供方"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Anthropic, OpenAI, Gemini, OpenRouter or any OpenAI-compatible endpoint, for the AI assistant and the AI reading of booking files","Google Places or Amap, only if you prefer their place search to the built-in one","Unsplash, only for trip cover and place pictures"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["productivity"],"alternativeTo":[{"slug":"tripit","name":"TripIt","confirmed":true},{"slug":"wanderlog","name":"Wanderlog","confirmed":true},{"slug":"splitwise","name":"Splitwise","confirmed":true}],"declaredAlternatives":[],"stars":15486,"downloads":1325,"downloadsWeek":0,"downloadsDays":4,"growth":115,"growthDays":6,"history":[{"date":"2026-11-06","stars":15406},{"date":"2026-11-07","stars":15427},{"date":"2026-11-08","stars":15434},{"date":"2026-11-09","stars":15463},{"date":"2026-11-10","stars":15486}],"firstPublishedAt":"2026-11-06T22:59:23.830Z","commitSha":"dc36c21c08f681fd2806764fad8e17b270f6f175","verifiedAt":"2026-11-06T22:44:02.964Z","lastCommitAt":"2026-11-10T09:56:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Plan trips with other people on your own server: days, maps, bookings, shared costs, packing lists, journal.","summary":"TREK is a travel planner you run on your own machine or server and open in a browser; it also installs on a phone straight from the browser and keeps working offline. Several people can edit the same trip at the same time: day plans on a map, hikes and road trips with routes, 18 kinds of booking, costs split with settle-up suggestions, packing lists, a group chat and a journal you can lay out as a photo book. Sign-in can use a password, company SSO or passkeys. You start it with Docker; the AI assistant and the MCP server ship switched off.","easeRationale":"D on Windows, Mac and Linux alike, because the only route is Docker: the README's \"30 seconds\" command starts the image with two folders and a key you generate, then you open http://localhost:3000 and read the first admin password out of the container log. For real use it asks for a reverse proxy with TLS that also passes WebSocket upgrades on /ws, and the Compose file has dozens of settings to go through. Helm and Kubernetes are the other documented path.","costNote":"Free under the AGPL v3: \"Self-host freely for personal or internal company use. If you modify and offer TREK as a network service to third parties, your modifications must be open-sourced under the same licence.\" Maps, weather, currencies and place search work with no key; Google Places, Amap, Unsplash and the AI features use keys you bring. The author takes donations on Ko-fi.","notVerified":["Whether the many optional pieces (tours, Studio, Journey, plugins, MCP) work as described without running them","What the place and routing services at places.liketrek.com keep about the searches your instance sends","Real disk and memory use once photos, files and offline map tiles pile up","Whether the offline mode on a phone holds up for a whole trip with no network","What an AI model costs per booking file or per planning chat"]},"zh-Hans":{"summaryShort":"在自己的服务器上和其他人一起规划旅行：行程、地图、预订、分摊费用、打包清单、日记。","summary":"TREK 是一个旅行规划工具，你在自己的电脑或服务器上运行它，用浏览器打开；它也可以直接从浏览器安装到手机上，离线也能用。多个人可以同时编辑同一趟旅行：地图上的每日行程、带路线的徒步和自驾游、18 种预订类型、带结算建议的费用分摊、打包清单、群聊，还有可以排成相册的日记。登录可以用密码、公司的 SSO 或密钥（passkeys）。你用 Docker 启动它；AI 助手和 MCP 服务器默认是关闭的。","easeRationale":"在 Windows、Mac 和 Linux 上评级都是 D，因为唯一的途径是 Docker：README 里\"30 seconds\"那条命令用两个文件夹和一个你自己生成的密钥启动镜像，然后你打开 http://localhost:3000，从容器日志里读出第一个管理员密码。真正投入使用还需要一个支持 TLS、也能让 /ws 的 WebSocket 升级通过的反向代理，Compose 文件里有几十项设置要过一遍。Helm 和 Kubernetes 是另一条有文档说明的路径。","costNote":"在 AGPL v3 下免费：\"Self-host freely for personal or internal company use. If you modify and offer TREK as a network service to third parties, your modifications must be open-sourced under the same licence.\" 地图、天气、货币换算和地点搜索不需要密钥就能用；Google Places、高德（Amap）、Unsplash 和 AI 功能需要你自己提供密钥。作者在 Ko-fi 上接受打赏。","notVerified":["那些可选的附加部分（tours、Studio、Journey、插件、MCP）在没有实际运行的情况下是否真的如描述那样工作","places.liketrek.com 上的地点和路线服务，会保留你的实例发出的搜索的哪些信息","照片、文件和离线地图瓦片越积越多之后，实际的磁盘和内存占用情况","手机上的离线模式能不能在完全没有网络的情况下撑完整趟旅行","每个预订文件或每次规划对话，AI 模型要花多少钱"]}}},{"id":"R_kgDOK7UGYw","slug":"lipku__livetalking","name":"LiveTalking","owner":"lipku","fullName":"lipku/LiveTalking","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/drive/folders/1FOC_MD6wdogyyX_7V1d4NDIO7P9NlSAJ?usp=sharing","source":"check","title":"外部下载链接"},{"code":"models_and_ready_packages_come_from_file_sharing_links","severity":"attention","evidence":"模型文件放在夸克网盘上提供（\"夸克云盘 https://pan.quark.cn/s/83a750323ef0\"），\"Windows整合包 https://pan.quark.cn/s/a040bf5cb065\"和\"桌面客户端 https://pan.quark.cn/s/d7192d8ac19b\"也是如此。个人网盘上的文件随时可能被替换，仓库里也没有附带任何校验值。","source":"ai","title":"模型和现成安装包来自网盘分享链接"},{"code":"copies_a_real_persons_face_and_voice","severity":"attention","evidence":"Avatar 页面的\"上传视频自动生成数字人形象\"用你上传的视频生成数字人，功能二是\"支持声音克隆\"，建议用途是无人带货直播。这些加在一起，就是让一个真实的人说出他们从未说过的话，这需要本人同意，而且在不少国家是受限制的。","source":"ai","title":"复制真实人物的脸和声音"},{"code":"instructions_ask_you_to_open_every_udp_port","severity":"attention","evidence":"启动说明写着\"服务端需开放端口 TCP:8010, UDP:1-65536\"——要在服务器上开放整个 UDP 范围。之后对外提供的页面里包括用于\"实时监控会话状态与全局配置\"的\"/admin.html 管理后台\"，而 README 里完全没提到这个页面需要密码或令牌。","source":"ai","title":"说明要求打开全部udp端口"},{"code":"videos_you_publish_must_carry_the_projects_watermark","severity":"info","evidence":"声明部分写道：\"基于本项目开发并发布在B站、视频号、抖音等平台上的视频需带上 LiveTalking 水印和标识\"——用它制作、发到B站、视频号或抖音等平台的视频，必须带上 LiveTalking 的水印和标识。这个条件并不是仓库所用 Apache-2.0 许可证的一部分。","source":"ai","title":"发布的视频必须带项目的水印"},{"code":"the_dockerfile_in_the_repository_cannot_build_as_is","severity":"info","evidence":"Dockerfile 的结尾是\"Copy ../nerfstream /nerfstream\"，这个路径在构建目录之外；它还在一个 RUN 步骤里执行\"conda activate\"，而这个设置不会延续到下一步。README 的 Docker 部分也没有真的构建它，而是指向 AutoDL 和 UCloud 提供的现成镜像。","source":"ai","title":"仓库里的dockerfile本身build不起来"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","docker","gpu_any","ffmpeg"],"disk":{"minMB":510,"maxMB":1370,"basis":"heuristic"},"apiKeys":["Qwen through DashScope, Doubao or another model service, only for the chat mode (the echo mode just repeats your text)","Tencent Cloud or Azure Speech, only if you want their voices instead of the free EdgeTTS"],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","audio","social_marketing"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":10356,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":84,"growthDays":6,"history":[{"date":"2026-11-04","stars":10272},{"date":"2026-11-05","stars":10289},{"date":"2026-11-06","stars":10303},{"date":"2026-11-07","stars":10320},{"date":"2026-11-08","stars":10329},{"date":"2026-11-09","stars":10341},{"date":"2026-11-10","stars":10356}],"firstPublishedAt":"2026-11-06T22:57:16.719Z","commitSha":"c7269433a2b6ecebcd9342445d2ecc882c98b0da","verifiedAt":"2026-11-06T22:45:21.459Z","lastCommitAt":"2026-10-20T03:39:26Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Puts a talking digital person on a live stream or a web page, with the lips matched to the voice as it speaks.","summary":"LiveTalking shows a digital person who talks in real time: you send text or audio, it moves the mouth in a video of a face you provide, and sends the result to a browser over WebRTC, to a live platform over RTMP, or out as a virtual webcam. The README points it at unmanned shopping streams, avatar customer service, courses and batch short videos; it can copy a voice, be cut off mid-sentence and play a chosen video while quiet. You need Linux or Windows with an NVIDIA card, Python, and model files fetched from file-sharing folders.","easeRationale":"Linux is the tested system and still a D: create a conda environment with Python 3.12, install a PyTorch build that matches your CUDA version, pip install the requirements, download wav2lip.pth and an avatar folder from a Quark or Google Drive folder and move them into the right directories by hand, then run app.py and open the ports. Windows takes the same route; the README also mentions an all-in-one Windows package on a file-sharing site, with no instructions for it. Both need an NVIDIA card: the frame-rate table starts at an RTX 3060.","costNote":"This version is free under Apache-2.0. The README has a table comparing it with a paid \"商用版\" (commercial version) that adds the wav2lipls model and a set of extra abilities, and it also offers a hosted trial at livetalking.top. Speech and model services you plug in (Tencent Cloud, Azure, Qwen) bill separately; EdgeTTS costs nothing.","notVerified":["Whether it runs on macOS, which the badge at the top of the README claims while every instruction and figure assumes an NVIDIA card","What the files on the netdisk and Google Drive folders actually contain","The frame rate on your own card, and whether it stays above the 25 the README calls real time","Whether the admin page and the API ask for any password once the ports are open","How many people it really serves at once, and how the mouth and the interruption look in practice"]},"zh-Hans":{"summaryShort":"在直播或网页上放一个会说话的数字人，说话时嘴型和声音对得上。","summary":"LiveTalking 展示一个实时说话的数字人：你发送文字或音频，它会在你提供的人脸视频上让嘴动起来，并通过 WebRTC 把结果发到浏览器，或通过 RTMP 推到直播平台，又或者输出成虚拟摄像头。README 把它用在无人带货直播、数字人客服、课程和批量短视频上；它能克隆声音，能在说到一半时被打断，安静的时候还能播放指定的视频。你需要 Linux 或 Windows、一块 NVIDIA 显卡、Python，还有从网盘链接下载的模型文件。","easeRationale":"Linux 是经过测试的系统，评级依然是 D：用 Python 3.12 创建一个 conda 环境，安装一个和你 CUDA 版本匹配的 PyTorch，用 pip 安装依赖，从夸克网盘或 Google Drive 的文件夹下载 wav2lip.pth 和一个 avatar 文件夹，手动把它们放进正确的目录，然后运行 app.py 并打开端口。Windows 走的是同一条路；README 还提到一个放在网盘上的 Windows 整合包，但没有使用说明。两个系统都需要 NVIDIA 显卡：帧率表格从 RTX 3060 起步。","costNote":"这个版本在 Apache-2.0 下免费使用。README 有一张表，把它和付费的\"商用版\"做对比，商用版多了 wav2lipls 模型和一些额外能力，项目还在 livetalking.top 提供一个托管的试用版。你接入的语音和模型服务（腾讯云、Azure、通义千问）要单独计费；EdgeTTS 不花钱。","notVerified":["README 顶部的徽章说支持 macOS，但所有的说明和数据都是按 NVIDIA 显卡写的，实际是否能用还不确定","网盘和 Google Drive 文件夹里的文件实际包含什么内容","你自己的显卡上帧率是多少，能不能一直高于 README 所说的「实时」标准 25 帧","端口打开之后，管理页面和 API 是否会要求任何密码","它实际上能同时服务多少人，嘴型和打断效果在实际使用中是什么样子"]}}},{"id":"R_kgDOK9yGXA","slug":"liuzhao1225__youdub-webui","name":"YouDub-webui","owner":"liuzhao1225","fullName":"liuzhao1225/YouDub-webui","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖存在已知漏洞（OSV）：ajv: GHSA-2g4f-4pwh-qvx6；undici: GHSA-2m5v-pch6-74hx, GHSA-4qv8-2p9j-2685, GHSA-4w7w-5xrc-9h26；torch: GHSA-63cw-57p8-fm3p, GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47","source":"check","title":"已知漏洞"},{"code":"copies_a_real_persons_voice","severity":"attention","evidence":"主打功能是\"逐句原声克隆\"——每一句翻译后的台词，都用\"对应原文句子的音频片段作为参考\"来说出来，也就是以原说话人的片段作为参考。首页的演示就是对一段拍摄的采访这么做的。让一个人看起来说了他们从未说过的语言，需要本人同意，而且在不少国家是受限的。","source":"ai","title":"复制真实人物的声音"},{"code":"downloads_and_re_dubs_other_peoples_videos","severity":"attention","evidence":"链接导入功能\"支持 YouTube、Shorts、B 站及 b23.tv 单视频链接；使用 yt-dlp\"，没有时长或文件大小的限制，举的例子也是别人的 YouTube 视频。从 YouTube 下载违反它自己的服务条款，发布一个你并不拥有的视频的配音版本，在你发布的任何地方都是版权问题。","source":"ai","title":"下载并重新配音别人的视频"},{"code":"the_local_interface_has_no_password","severity":"info","evidence":".env.example 里写得很直白：\"Browser session settings. No access password is required.\"，YOUDUB_AUTH_COOKIE_SECURE=false。它默认监听 YOUDUB_HOST=127.0.0.1，所以默认只有你自己的机器能访问——但这个设置是可以改的，一旦改了，不会有任何东西要求输入密码。","source":"ai","title":"本地界面没有密码"},{"code":"early_version_with_parts_the_author_calls_unverified","severity":"info","evidence":"package.json 写的是 0.0.12 版本，\"当前范围\"一节列出了还没完成的地方：\"本地 workflow、Client 与 Python 插件已有验证；GitHub/npm 远端插件实装尚未验收\"，没有插件市场，插件的改动要重启之后才会生效。桌面这一层是建立在\"裁剪后的 DeepSeek Harness 框架\"之上的，一个保存在 vendor/ 里的 fork。","source":"ai","title":"早期版本中有作者自称未验证的部分"}],"platforms":["windows","mac"],"tools":["ffmpeg"],"buildTools":["python","node","gpu_nvidia","ffmpeg"],"disk":{"minMB":117,"maxMB":350,"basis":"installer"},"apiKeys":["A translation model provider, added under Settings → Models (the example settings default to gpt-4.1-mini)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"rask-ai","name":"Rask AI","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true}],"declaredAlternatives":[],"stars":5829,"downloads":6961,"downloadsWeek":789,"downloadsDays":5,"growth":29,"growthDays":6,"history":[{"date":"2026-11-05","stars":5802},{"date":"2026-11-06","stars":5809},{"date":"2026-11-07","stars":5812},{"date":"2026-11-08","stars":5817},{"date":"2026-11-09","stars":5823},{"date":"2026-11-10","stars":5829}],"firstPublishedAt":"2026-11-06T22:57:16.726Z","commitSha":"a27041552f5bff69e2ec04f1fae1402f4d7ba5f1","verifiedAt":"2026-11-06T22:46:22.159Z","lastCommitAt":"2026-11-06T10:56:02Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Translates a video and dubs it, keeping each sentence in a copy of the original speaker's voice.","summary":"YouDub Desktop is a Windows and Mac app that turns a video into another language. You import a file or paste a link, and it writes down the speech with Whisper, translates it with a model you set up, makes subtitles, and dubs it sentence by sentence in a copy of the original voice, keeping the background sound if you want. It can give you the original audio with subtitles, the dub, or both. The steps run as a workflow you can stop, correct by hand and redo. The listening and voice work happen on your machine; the translation uses a provider key.","easeRationale":"Windows and Mac both get a B: run YouDub-windows-x64-Setup.exe or the .dmg (an Apple Silicon build; there is no Intel one), and the app prepares its own Python environment at first launch. What is left is in-app: add a provider and a key for the translation model, then let it fetch the models each workflow asks for (Whisper, Demucs, VoxCPM, the aligner). The settings file notes FFmpeg has to be a build with subtitle support, and on Windows one with the av*.dll files beside ffmpeg.exe.","costNote":"The app is free under Apache-2.0 and the heavy parts (Whisper, Demucs, VoxCPM) run on your own machine. The translation step calls a model provider you configure, with gpt-4.1-mini as the default in the example settings, so a long video costs whatever that provider charges.","notVerified":["Whether the installer brings its own FFmpeg with subtitle support or you have to install one","How slow it is on a machine with no NVIDIA card, since the setup also accepts the processor alone","How much disk the app takes once the models for a workflow are downloaded","How good the translation and the sentence splitting are outside English, Chinese and Japanese","Whether the cloned voice sounds like the demos on videos of your own"]},"zh-Hans":{"summaryShort":"翻译视频并配音，每一句都用原说话人声音的克隆版本。","summary":"YouDub Desktop 是一个 Windows 和 Mac 应用，把视频转换成另一种语言。你导入一个文件或粘贴一个链接，它会用 Whisper 把语音转写出来，用你设置的模型翻译，生成字幕，并逐句用原声音的克隆版本配音，如果你想要还能保留背景声音。它可以给你带字幕的原声、配音版，或者两者都要。整个流程以工作流的形式运行，你可以中途停下、手动修正再重做。听写和配音都在你自己的电脑上完成；翻译那一步用的是你配置的提供方密钥。","easeRationale":"Windows 和 Mac 的评级都是 B：运行 YouDub-windows-x64-Setup.exe 或者 .dmg（一个 Apple Silicon 版本；没有 Intel 版），应用会在第一次启动时准备好自己的 Python 环境。剩下的都在应用里完成：添加一个翻译模型的提供方和密钥，然后让它去下载每个工作流需要的模型（Whisper、Demucs、VoxCPM、对齐器）。设置文件里提到 FFmpeg 必须是支持字幕的版本，在 Windows 上还要在 ffmpeg.exe 旁边放上 av*.dll 文件。","costNote":"这个应用在 Apache-2.0 下免费，重的部分（Whisper、Demucs、VoxCPM）都在你自己的电脑上运行。翻译这一步调用的是你配置的模型提供方，示例设置里默认用 gpt-4.1-mini，所以一个长视频花多少钱取决于那家提供方怎么收费。","notVerified":["安装包是否自带支持字幕的 FFmpeg，还是需要你自己装一个","没有 NVIDIA 显卡的机器上有多慢，因为设置里也支持只用处理器","一个工作流的模型下载完之后，应用一共占多少磁盘空间","在英语、中文和日语之外，翻译和断句的质量怎么样","克隆出来的声音用在你自己的视频上，听起来是不是和演示一样"]}}},{"id":"R_kgDOVRRpkQ","slug":"lnpandy__nx2apple","name":"NX2Apple","owner":"LNPANDY","fullName":"LNPANDY/NX2Apple","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"20 天内获得 5676 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"needs_console_keys_and_your_own_game_dumps","severity":"high","evidence":"该项目不提供游戏或主机密钥。要求写道：\"Your own game dumps (NSP or XCI) and your own `prod.keys` from your own console. Nothing is bundled, downloaded or looked up for you.\" 作者的立场写在法律声明部分：\"It does not bypass, strip or redistribute any protection — LibHac opens the user's own dumps with the user's own keys, the same operation a console performs on a cartridge.\" 同一部分还写道：\"Laws on making a personal backup copy differ by country; some places do not allow it at all.\"","source":"ai","title":"需要主机密钥和你自己转储的游戏"},{"code":"the_built_app_contains_the_whole_game","severity":"attention","evidence":"游戏列表中列出 \"The Legend of Zelda: Tears of the Kingdom ... 304,243 files, byte-checked; ~17 GB\" 都在编译出的包里。法律声明部分说你有责任 \"for never distributing game files, keys or the packages built from them — the project's outputs are for personal use on your own hardware only\"。","source":"ai","title":"编译出的应用包含完整游戏"},{"code":"unsigned_launcher_and_sideloading","severity":"attention","evidence":"限制部分：\"Unsigned launcher. The launcher is not signed for you, so you sideload it.\" 安装步骤 2：\"Open it once with right-click, then Open. The app is not signed with a paid Apple developer certificate.\" 手机应用是在 App Store 之外安装的，使用你自己的开发者团队签名，或通过 \"AltStore Classic 2.2+, Impactor\"、SideStore 或 Sideloadly 安装。","source":"ai","title":"未签名的启动器与旁加载安装"},{"code":"it_downloads_the_code_it_builds_with_and_updates_itself","severity":"attention","evidence":"安装步骤 3：\"The launcher fetches the project it builds with (a Download repository button on the setup page, a shallow clone into `~/NX2Apple` ...) checks the tools above and updates itself later.\" Windows 版本还带有 \"in-app self-update (`NX2Apple.exe`)\"。也就是说，大部分实际运行的代码是在你安装之后才获取的，而且上游引擎 \"A daily job moves the pin\"。","source":"ai","title":"会下载自己用来编译的代码并自我更新"},{"code":"built_on_switch_emulator_code","severity":"attention","evidence":"致谢：\"Ryujinx (MIT) — the shader translator that turns Maxwell shaders into SPIR-V\"，还有 \"yuzu, deko3d, libvpx, stb_vorbis, libdeflate\" 以及 \"Atmosphère ... the reference for Horizon's synchronization semantics\"。Ryujinx 和 yuzu 都曾是 Switch 模拟器，在任天堂采取行动后停止了开发，README 声明该项目 \"Not affiliated with, sponsored by or endorsed by Nintendo\"。","source":"ai","title":"构建在 Switch 模拟器代码之上"},{"code":"narrow_device_and_memory_limits","severity":"info","evidence":"系统要求和限制：需要搭载 Apple 芯片的 Mac，\"iOS or iPadOS 26 or later and a recent device. Older devices are untested\"，NVN 游戏必须使用 Xcode 27.2，只有已建立配置文件的 147 款游戏可用，并且 \"Without the Increased Memory Limit entitlement iOS gives the app about 3.4 GB, which is not enough for Tears of the Kingdom.\"","source":"ai","title":"设备和内存限制较为苛刻"},{"code":"very_new_and_changing_every_day","severity":"info","evidence":"该仓库创建于 2026-10-20，已经有 67 个发布版本，最新的是预发布版 v1.3.1-rc2（2026-11-09），还有 191 个未解决的 issue。它是 \"a community fork of nxnative\"，\"follows nxnative every day\"。","source":"ai","title":"项目非常新，每天都在变化"}],"platforms":["mac","windows"],"tools":["dotnet","other"],"buildTools":[],"disk":{"minMB":215,"maxMB":644,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[{"target":"benzinga-pro","stance":"claims","evidence":"(Xbox, PlayStation, Switch Pro, MFi) replaces them"},{"target":"dbeaver-pro","stance":"claims","evidence":"(Xbox, PlayStation, Switch Pro, MFi) replaces them"},{"target":"moz-pro","stance":"claims","evidence":"(Xbox, PlayStation, Switch Pro, MFi) replaces them"},{"target":"deepl-pro","stance":"claims","evidence":"(Xbox, PlayStation, Switch Pro, MFi) replaces them"}],"stars":5942,"downloads":361885,"downloadsWeek":0,"downloadsDays":0,"growth":2697,"growthDays":7,"history":[{"date":"2026-11-10","stars":5942}],"firstPublishedAt":"2026-11-09T23:09:09.930Z","commitSha":"e43c6d635d9a0b7335a045452ce861a250622bcc","verifiedAt":"2026-11-09T23:04:27.446Z","lastCommitAt":"2026-11-10T16:52:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Mac or Windows app that packs your own Switch game dumps into an app you can play on an iPhone, iPad or Mac.","summary":"NX2Apple is a desktop launcher for Apple silicon Macs and for Windows that turns Switch game files you dumped yourself into an app for your iPhone, iPad or Mac. The game's own code is rewritten to run on Apple chips, so it is not an emulator. You pick the game folder, choose the target and press build; your games then sit in a library with cover art, touch controls and controller support. 147 games have a profile. You need your own dumps and your own prod.keys, Xcode 27.2 on the Mac, and iOS 26 or later on the phone.","easeRationale":"On an Apple silicon Mac (macOS 13 or later) you unzip the 188 MB app and open it with right-click, then the setup page downloads the engine and installs what it can in one click, but you still have to install Xcode 27.2 from Apple's developer download page by hand and sometimes run \"sudo xcode-select -s\" in a terminal. The Windows zip builds and installs iPhone and iPad packages with a bundled engine, but games do not run on Windows itself. Either way you must supply your own game dumps and prod.keys.","costNote":"Free and MIT-licensed. The launcher is not signed, and the iPhone app is signed with your own Apple account; the README says a free Apple ID through SideStore needs a one-time App ID fix to keep the memory entitlement.","notVerified":["Whether a build really works end to end on your Mac or Windows PC, and how long it takes for a 17 GB game.","How the games behave on devices the maintainer did not test; most of the 147 profiles \"have not been tried on a phone yet\".","Total disk space needed on the computer and on the phone once a game is unpacked, rewritten and packed.","What the launcher downloads during setup and self-update, since that code is not part of the release you checked.","Whether the zips on the releases page match the source in this repository; the engine itself is not committed here.","Where prod.keys come from and what the rules are in the user's country for extracting them from a console and dumping games you own."]},"zh-Hans":{"summaryShort":"一款 Mac 或 Windows 应用，把你自己转储的 Switch 游戏打包成可以在 iPhone、iPad 或 Mac 上玩的应用。","summary":"NX2Apple 是一款适用于搭载 Apple 芯片的 Mac 和 Windows 的桌面启动器，它会把你自己转储的 Switch 游戏文件转换成可在 iPhone、iPad 或 Mac 上使用的应用。游戏本身的代码会被重写，使其运行在 Apple 芯片上，所以它不是模拟器。你选择游戏文件夹、选择目标平台并点击编译；之后你的游戏会出现在一个带封面、触控操作和手柄支持的游戏库中。已有 147 款游戏建立了配置文件。你需要自己的转储文件和自己的 prod.keys、Mac 上的 Xcode 27.2，以及手机上的 iOS 26 或更高版本。","easeRationale":"在搭载 Apple 芯片的 Mac（macOS 13 或更高版本）上，你解压这个 188 MB 的应用并用右键打开，然后设置页面会下载引擎并一键安装能装的部分，但你仍然需要手动从苹果开发者下载页面安装 Xcode 27.2，有时还要在终端里运行 \"sudo xcode-select -s\"。Windows 版压缩包会用内置的引擎编译并安装 iPhone 和 iPad 的安装包，但游戏本身不能在 Windows 上运行。无论哪种方式，你都必须提供自己的游戏转储文件和 prod.keys。","costNote":"免费，MIT 许可。启动器未签名，iPhone 应用使用你自己的 Apple 账号签名；README 说通过 SideStore 使用免费 Apple ID 时，需要做一次性的 App ID 修复才能保留内存授权（entitlement）。","notVerified":["编译结果是否真的能在你的 Mac 或 Windows 电脑上完整跑通，以及一个 17 GB 的游戏需要多长时间。","在维护者没有测试过的设备上，游戏表现如何；147 个配置文件中大多数 \"have not been tried on a phone yet\"。","游戏解包、重写并打包之后，在电脑和手机上总共需要多少磁盘空间。","启动器在设置和自我更新过程中下载了什么，因为这部分代码不在你检查的发布版本里。","发布页面上的压缩包是否与这个仓库中的源代码一致；引擎本身并没有提交到这里。","prod.keys 来自哪里，以及在用户所在国家，从主机中提取密钥并转储自己拥有的游戏，适用什么样的法律规定。"]}}},{"id":"R_kgDOSb7GsA","slug":"locoremind__locoagent","name":"LocoAgent","owner":"LocoreMind","fullName":"LocoreMind/locoagent","type":"dev_tool","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道传给 shell 执行：curl -fsSL https://bun.sh/install | bash","source":"check","title":"用 curl 管道执行 shell 脚本"},{"code":"automating_social_accounts_breaks_their_rules","severity":"attention","evidence":"这个项目存在的意义就是让一个程序代替你行动：它\"autonomously operates real accounts\"，带有覆盖点赞、回复、关注和发帖的\"37 operations for X.com\"，还有每小时评论一次的守护进程工作流\"hourly\"。X、LinkedIn 和 Reddit 都在各自条款里禁止自动化互动，并会限制或封禁相关账号。承担风险的是你自己的账号。","source":"ai","title":"自动操作社交账号违反平台规则"},{"code":"made_to_look_like_a_person_to_the_sites_checks","severity":"attention","evidence":"README 把这个设计解释为规避检测的手段：\"Social platforms detect and block headless browsers and API automation. LocoAgent runs through a real, full Chrome — same engine, same fingerprint — so it behaves like you actually do\"，驱动的是你\"actual login cookies\"。config/browser-targets.json 还给每个平台配了各自的代理地址。","source":"ai","title":"被设计成在平台检测中看起来像真人"},{"code":"most_of_the_code_is_another_product_relicensed","severity":"attention","evidence":"README 说它\"is a fork of the Claude Code CLI source tree\"，项目结构说明把 src/ 标记为\"Vendored Claude Code CLI source — treat as a dependency\"，重用了大约 40 个工具和 90 个斜杠命令；package.json 用本地桩代码替换了 @anthropic-ai/claude-agent-sdk、@anthropic-ai/sandbox-runtime 和几个 @ant/computer-use 包。覆盖这一切的 LICENSE 文件写的是\"MIT License, Copyright (c) 2026 LocoAgent Contributors\"，而 Claude Code 本身并不是在 MIT 许可下发布的。","source":"ai","title":"大部分代码是换了许可证的另一个产品"},{"code":"meant_to_run_with_the_confirmation_prompts_off","severity":"attention","evidence":"配置文档把 SKIP_PERMISSIONS=1 说成\"required for non-interactive / automated runs\"，说明里还提到它会\"makes stubs/globals.ts inject --dangerously-skip-permissions into argv so automated/headless runs don't stop on permission prompts\"。设了这个之后，智能体会在你的电脑上直接运行 shell 命令，不会再问你。","source":"ai","title":"设计成关掉确认提示运行"},{"code":"the_model_sees_your_feed_and_writes_in_your_name","severity":"info","evidence":"整个循环是先对页面做一次\"agent-browser snapshot\"，然后模型做决定，再由工具去点击或填写内容，于是你时间线和消息里的内容都会发给你配置的那家提供方，它写的文字再以你的账号发布出去。Chrome 运行在一个和你日常浏览器分开的\"a dedicated, isolated, persistent profile\"里。","source":"ai","title":"模型看得到你的动态并以你的名义发帖"},{"code":"vendors_claude_code_source_under_mit","severity":"high","evidence":"README：\"Under the hood it is a fork of the Claude Code CLI source tree, re-purposed for social automation: the battle-tested agent loop, ~40 tools, ~90 slash commands, and Ink/React terminal UI are reused\"，项目结构说明把\"src/   # Vendored Claude Code CLI source — treat as a dependency\"标记出来。覆盖这一切的 LICENSE 写的是\"MIT License, Copyright (c) 2026 LocoAgent Contributors\"，package.json 用本地桩代码替换了\"@anthropic-ai/claude-agent-sdk\"、\"@anthropic-ai/sandbox-runtime\"和\"@ant/computer-use\"系列包。Claude Code 是由 Anthropic 按自己的商业条款分发的，并不在 MIT 许可之下，而大多数国家的版权规则都限制把别人的源代码按对方未授予的许可证转发出去。","source":"ai","title":"在mit许可下搭售claude code的源代码"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["One model provider: DeepSeek, OpenAI, Anthropic, OpenRouter, AWS Bedrock, Google Vertex, or a local Ollama or LM Studio"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","ai_agents","automation"],"alternativeTo":[{"slug":"phantombuster","name":"PhantomBuster","confirmed":true},{"slug":"replyguy","name":"ReplyGuy","confirmed":true},{"slug":"linked-helper","name":"Linked Helper","confirmed":true}],"declaredAlternatives":[],"stars":1077,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":3,"growthDays":6,"history":[{"date":"2026-11-04","stars":1074},{"date":"2026-11-05","stars":1075},{"date":"2026-11-06","stars":1076},{"date":"2026-11-07","stars":1078},{"date":"2026-11-08","stars":1077},{"date":"2026-11-09","stars":1077},{"date":"2026-11-10","stars":1077}],"firstPublishedAt":"2026-11-06T22:57:16.733Z","commitSha":"c01bb3f8a7b06a0db9f697c5bea485947959d226","verifiedAt":"2026-11-06T22:47:31.800Z","lastCommitAt":"2026-06-27T01:40:53Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An agent that drives your own Chrome to like, reply, follow and post on X, LinkedIn and Reddit by itself.","summary":"LocoAgent is a terminal program that works your social accounts for you. It opens a separate Chrome where you log in once, and then an AI model reads the page and clicks: it likes posts, writes replies, follows people and publishes content on X, with scripted pipelines for LinkedIn and Reddit as well. You can ask for something in plain words, write daily tasks and limits in a file, or leave a workflow running on a schedule, and a log keeps it from liking the same post twice. Everything happens in the terminal, with Bun, Chrome and a model key.","easeRationale":"D on all three systems, because it lives only in the terminal. The one-line installer pulls in Bun and the agent-browser tool, copies the repository, writes a .env and asks for your provider key, but Chrome and Git it only looks for, and afterwards you still run \"bun run setup-chrome\", log into each social account in the window that opens, and start it with \"bun start\". Making your own workflow means writing TypeScript.","costNote":"The code is free under the MIT licence, but the agent reads pages and writes replies through a model you pay for (DeepSeek, OpenAI, Anthropic, OpenRouter and the like). Pointing it at Ollama or LM Studio on your own machine avoids that bill.","notVerified":["Whether it still works, since social sites change their pages often, there is no unit-test suite and the last commit is from June 2026","How often the model misreads a page and likes, follows or writes the wrong thing","Whether the platforms notice this kind of automation anyway, and what they do about it","What a day of running it costs with a paid model","What the agent-browser tool it installs from npm does with the pages it reads"]},"zh-Hans":{"summaryShort":"一个操纵你自己的 Chrome，自行在 X、LinkedIn 和 Reddit 上点赞、回复、关注和发帖的智能体。","summary":"LocoAgent 是一个在终端运行的程序，替你打理社交账号。它会打开一个单独的 Chrome，你只需登录一次，之后一个 AI 模型会读取页面并点击操作：在 X 上点赞帖子、写回复、关注别人、发布内容，LinkedIn 和 Reddit 也有脚本化的流程。你可以用简单的话提出要求，把每天的任务和限额写进一个文件，或者让一个工作流按计划一直运行，还有一份日志防止它给同一条帖子点两次赞。所有事情都在终端里发生，需要 Bun、Chrome 和一个模型密钥。","easeRationale":"三个系统都是 D，因为它只能在终端里跑。那个一行命令的安装程序会拉取 Bun 和 agent-browser 工具，复制仓库，写一个 .env，并要求你填入提供方密钥，但 Chrome 和 Git 它只会去检测是否存在；之后你还得运行\"bun run setup-chrome\"，在弹出的窗口里给每个社交账号登录，再用\"bun start\"启动它。要做自己的工作流，就得写 TypeScript。","costNote":"代码在 MIT 许可下免费，但智能体读取页面和写回复要靠你自己付费的模型（DeepSeek、OpenAI、Anthropic、OpenRouter 之类）。把它指向你自己电脑上的 Ollama 或 LM Studio 可以省下这笔费用。","notVerified":["它是否还能用，因为社交网站经常改版面，项目也没有单元测试，最后一次提交是 2026 年 6 月","模型读错页面、点错赞、关注错人或写错内容的频率有多高","这些平台会不会注意到这种自动化行为，又会怎么处理","用付费模型跑一整天要花多少钱","从 npm 安装的 agent-browser 工具，对它读到的页面内容做了什么"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyNjYyMzExNDk=","slug":"logseq__logseq","name":"Logseq","owner":"logseq","fullName":"logseq/logseq","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"B","web":"B"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖项存在已知漏洞（OSV）：fs: MAL-2025-21003；posthog-js: GHSA-8775-5hwv-wr6v","source":"check","title":"存在已知漏洞"},{"code":"beta_data_loss_warning","severity":"attention","evidence":"README.md：\"The DB version is in beta status while the new mobile app and RTC is in alpha. This means that **data loss is possible** so we recommend automated backups\"。","source":"ai","title":"测试版可能丢失数据的警告"},{"code":"usage_analytics_library","severity":"info","evidence":"自动检查在固定版本的依赖项中列出了分析库 \"posthog-js\"。README 没有说明它是否默认开启。","source":"ai","title":"使用情况分析库"},{"code":"linux_install_script_piped_to_shell","severity":"info","evidence":"README.md（仅 Linux）：\"curl -fsSL https://raw.githubusercontent.com/logseq/logseq/master/scripts/install-linux.sh | bash\"。Windows 和 Mac 用户使用的是普通安装程序。","source":"ai","title":"Linux 安装脚本直接交给 shell 运行"},{"code":"latest_tag_is_a_nightly_build","severity":"info","evidence":"列出的最新发布是 \"nightly\" 标签（2026-10-25），文件名为 \"2.0.2-alpha+nightly.20261025\"；最后一个带编号的版本是 2026-10-07 的 2.0.2。","source":"ai","title":"最新标签是每夜构建版"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":149,"maxMB":448,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["productivity","office"],"alternativeTo":[{"slug":"roam-research","name":"Roam Research","confirmed":true},{"slug":"notion","name":"Notion","confirmed":true}],"declaredAlternatives":[],"stars":45544,"downloads":96136040,"downloadsWeek":872822,"downloadsDays":7,"growth":78,"growthDays":7,"history":[{"date":"2026-10-26","stars":45395},{"date":"2026-10-27","stars":45400},{"date":"2026-10-28","stars":45407},{"date":"2026-10-29","stars":45416},{"date":"2026-10-30","stars":45432},{"date":"2026-10-31","stars":45437},{"date":"2026-11-01","stars":45450},{"date":"2026-11-02","stars":45464},{"date":"2026-11-03","stars":45466},{"date":"2026-11-04","stars":45478},{"date":"2026-11-05","stars":45488},{"date":"2026-11-06","stars":45501},{"date":"2026-11-07","stars":45509},{"date":"2026-11-08","stars":45518},{"date":"2026-11-09","stars":45536},{"date":"2026-11-10","stars":45544}],"firstPublishedAt":"2026-10-27T01:10:12.657Z","commitSha":"0232648eb481f4bbf34e9b00eda65265f22ff906","verifiedAt":"2026-10-30T01:45:35.593Z","lastCommitAt":"2026-11-10T13:32:58Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Logseq-win-x64-2.0.2-nsis.exe","sizeBytes":152653328,"url":"https://github.com/logseq/logseq/releases/download/2.0.2/Logseq-win-x64-2.0.2-nsis.exe","releaseUrl":"https://github.com/logseq/logseq/releases/tag/2.0.2"},{"os":"mac","file":"Logseq-darwin-arm64-2.0.2.dmg","sizeBytes":177591867,"url":"https://github.com/logseq/logseq/releases/download/2.0.2/Logseq-darwin-arm64-2.0.2.dmg","releaseUrl":"https://github.com/logseq/logseq/releases/tag/2.0.2"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A notes app where you write daily notes, link ideas, manage tasks and mark up PDFs, with the files kept on your computer.","summary":"Logseq is a desktop app for taking notes and organising what you know. You write in short bullet points, link pages to each other, keep task lists and highlight PDFs, and the app shows how your notes connect. It is for students, professionals and anyone who keeps a lot of notes. There are ready installers for Windows and Mac, an AppImage or install script for Linux, and a web version, plus plugins and themes. The new database version is in beta and the project itself warns that data loss is possible, so it recommends backups.","easeRationale":"Windows and Mac are the easiest: a ready installer (Logseq-win-x64-2.0.2-nsis.exe, about 146 MB) or a .dmg that you download, install and start writing in. On Linux the release gives an AppImage you must make runnable yourself, or the README's one-line install script. The web version at app.logseq.com needs no install but is the beta database version.","costNote":"Free and open source. The project asks for donations on Open Collective. Real-time sync and collaboration (RTC) is in alpha and needs a sign-up form.","notVerified":["Whether notes made in the older file-based version open correctly in the new database version.","Whether the app sends usage analytics, and whether that can be switched off.","How stable the 2.0 releases are on Windows with large note collections.","Whether real-time sync will stay free after the alpha."]},"zh-Hans":{"summaryShort":"一款笔记应用,可以写每日笔记、关联想法、管理任务、标注PDF,文件都保存在你自己的电脑上。","summary":"Logseq是一款用于记笔记和整理知识的桌面应用。你用简短的要点来写笔记,把页面相互关联,维护任务列表并给PDF加标注,应用会展示你的笔记之间是如何关联的。它适合学生、专业人士以及任何做大量笔记的人。Windows和Mac有现成的安装程序,Linux有AppImage或安装脚本,还有网页版,另外还有插件和主题。新的数据库版本还在测试阶段,项目方自己也提醒可能会丢失数据,因此建议做好备份。","easeRationale":"Windows和Mac最简单:下载一个现成的安装程序(Logseq-win-x64-2.0.2-nsis.exe,约146MB)或一个.dmg文件,安装后就能开始写笔记。在Linux上,发布包提供的是AppImage,你得自己把它改成可执行,或者使用README里的一行安装脚本。app.logseq.com上的网页版不需要安装,但它是测试阶段的数据库版本。","costNote":"免费且开源。项目在Open Collective上接受捐款。实时同步与协作功能(RTC)还在内测阶段,需要填写报名表。","notVerified":["旧的基于文件的版本中创建的笔记,能否在新的数据库版本中正常打开。","应用是否会发送使用数据,以及能否关闭这个功能。","在笔记数量很大的情况下,2.0版本在Windows上的稳定性如何。","内测结束后,实时同步功能是否还会继续免费。"]}}},{"id":"R_kgDOIj5GAA","slug":"lss233__kirara-ai","name":"Kirara AI","owner":"lss233","fullName":"lss233/kirara-ai","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"connects_your_personal_chat_accounts","severity":"attention","evidence":"该机器人接入的聊天平台不允许个人账号使用第三方客户端:仓库主题包括\"go-cqhttp\"、\"mirai\"、\"qq\"和\"wechat\",依赖列表中有\"wechatpy\",平台表格列出了\"OneBot\"适配器。腾讯对QQ和微信的规定禁止使用非官方客户端,这样使用的账号可能被封禁。","source":"ai","title":"连接你的个人聊天账号"},{"code":"windows_package_is_a_prerelease","severity":"info","evidence":"最新版本v3.2.1发布于2026-10-22,标注\"no platform binaries\"(没有平台二进制文件)。唯一的Windows安装包是2025-05-05发布的\"Windows-quickstart-kirara-ai-v3.3.0a2.zip\",是一个alpha版本(pyproject.toml中写着version = \"3.3.0a2\")。","source":"ai","title":"Windows安装包是预发布版本"},{"code":"interface_downloaded_at_build_time","severity":"info","evidence":"Dockerfile并不是从这个仓库构建网页界面,而是在构建时从npm获取最新版本:\"PACKAGE_INFO=$(curl -s https://registry.npmjs.org/kirara-ai-webui) && LATEST_VERSION=$(printf %s $PACKAGE_INFO | jq -r '.[\"dist-tags\"].latest')\"。因此界面代码并不固定在你所查看的那个版本上。","source":"ai","title":"界面在构建时才下载"},{"code":"packages_from_a_mirror","severity":"info","evidence":"pyproject.toml把中国的一个镜像站设为Python包的默认来源:\"[[tool.uv.index]] url = \\\"https://mirrors.ustc.edu.cn/pypi/simple\\\" default = true\"。","source":"ai","title":"软件包来自镜像站"},{"code":"builtin_tunnel_feature","severity":"info","evidence":"功能列表中包括一个内置穿透功能,可让本地服务从你的网络外部被访问到:\"内置 Frpc 内网穿透\"。该项目还提供了带/v1/chat和/v2/chat端点的HTTP API。","source":"ai","title":"内置穿透功能"},{"code":"documentation_in_chinese","severity":"info","evidence":"README、设置指南的链接(\"请移步至 快速开始\")、kirara-docs.app.lss233.com上的手册以及社区群组都是中文的;仓库里没有英文版的README。","source":"ai","title":"文档是中文的"}],"platforms":["windows","linux","mac"],"tools":["python","docker"],"buildTools":["python","docker"],"disk":{"minMB":219,"maxMB":657,"basis":"installer"},"apiKeys":["An AI model provider (OpenAI, DeepSeek, Claude, Gemini, Qwen and others)","A chat platform bot token (QQ, Telegram, WeChat, Discord)","Azure Speech (voice replies)","Midjourney or Stable Diffusion service (image replies)"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["ai_agents","automation"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"chatbase","name":"Chatbase","confirmed":true}],"declaredAlternatives":[],"stars":19277,"downloads":129040,"downloadsWeek":0,"downloadsDays":4,"growth":14,"growthDays":6,"history":[{"date":"2026-11-06","stars":19266},{"date":"2026-11-07","stars":19266},{"date":"2026-11-08","stars":19269},{"date":"2026-11-09","stars":19269},{"date":"2026-11-10","stars":19277}],"firstPublishedAt":"2026-11-07T22:56:16.833Z","commitSha":"36249e4a0aa8e3d8930acdfe4e14ec2be6484f63","verifiedAt":"2026-11-07T22:43:12.670Z","lastCommitAt":"2026-10-22T08:12:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted chatbot that answers for you on QQ, WeChat, Telegram and Discord using the AI model you choose.","summary":"Kirara AI puts an AI chatbot inside chat apps. You run it on your own machine or server, pick a model (OpenAI, DeepSeek, Claude, Gemini, Qwen, Doubao, Ollama and others), connect it to QQ, Telegram, WeChat or Discord, and set personalities, keyword triggers, workflows, image generation and voice replies from a web admin page. It is aimed at people who like tinkering: the project manual and the community are in Chinese, and you install it with Python or Docker. A plugin market adds more features.","easeRationale":"Windows is the easiest: there is a ready \"Windows-quickstart\" zip, but it belongs to a pre-release from May 2025, so for the current version you install Python and the kirara-ai package and then connect a model and a chat account in the web admin. On Linux and macOS the route is Docker (the repository ships a Dockerfile and the project publishes an image), plus the same model and chat-platform setup, and the manual is in Chinese.","costNote":"The program is free (AGPL-3.0). You pay whatever the AI model costs: the paid providers charge per use, and a local model through Ollama costs nothing. Voice replies through Azure and image models like Midjourney are paid separately.","notVerified":["Whether the 2025 Windows quickstart zip still works with the current plugins and models","Whether the QQ, WeChat and Discord adapters work today (the table marks Discord and Feishu as being rebuilt)","How much the chosen AI model costs in a busy group chat","What the plugins from the plugin market do once installed","Real disk use after the Docker image and the models are in place"]},"zh-Hans":{"summaryShort":"自托管聊天机器人,用你选择的AI模型在QQ、微信、Telegram和Discord上替你回复消息。","summary":"Kirara AI把一个AI聊天机器人放进聊天应用里。你在自己的电脑或服务器上运行它,选择一个模型(OpenAI、DeepSeek、Claude、Gemini、通义千问、豆包、Ollama等),把它连接到QQ、Telegram、微信或Discord,并在网页管理后台设置人设、关键词触发、工作流、图片生成和语音回复。它面向喜欢折腾的人:项目手册和社区都是中文的,安装方式是Python或Docker。插件市场可以增加更多功能。","easeRationale":"Windows最简单:有现成的\"Windows-quickstart\"压缩包,但它属于2025年5月的预发布版本,所以当前版本需要安装Python和kirara-ai包,再到网页管理后台连接模型和聊天账号。Linux和macOS的方式是Docker(仓库自带Dockerfile,项目也发布了镜像),加上同样的模型和聊天平台设置,手册是中文的。","costNote":"该程序免费(AGPL-3.0协议)。你需要支付AI模型本身的费用:付费服务商按使用量收费,而通过Ollama运行的本地模型不花钱。通过Azure的语音回复和Midjourney等图像模型需要另外付费。","notVerified":["2025年的Windows快速启动压缩包是否仍能配合当前的插件和模型使用","QQ、微信和Discord适配器现在是否还能用(表格标注Discord和飞书正在重建)","在活跃的群聊中,所选AI模型的费用有多高","插件市场里的插件安装后实际会做什么","Docker镜像和模型都装好后,实际占用多少磁盘空间"]}}},{"id":"R_kgDOI2kzBQ","slug":"lucidrains__musiclm-pytorch","name":"MusicLM - Pytorch","owner":"lucidrains","fullName":"lucidrains/musiclm-pytorch","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","linux":"E"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布是在2023-09-06(1158天前)","source":"check","title":"版本长期未更新"},{"code":"no_trained_model_included","severity":"info","evidence":"没有任何东西可以直接使用:README写着\"MuLaN first needs to be trained\"、\"get a ton of <sound, text> pairs and train\"以及\"After much training on all three transformers (semantic, coarse, fine), you will pass your finetuned or trained-from-scratch AudioLM\"。仓库中没有模型文件,发布版本也标注\"no platform binaries\"。","source":"ai","title":"不包含已训练好的模型"},{"code":"replication_still_incomplete","severity":"info","evidence":"该项目是对别人模型的重建,其自身的待办列表里仍有未完成事项:\"support variable lengthed audio with masking in audio transformer\"、\"add a version of mulan to open clip\"以及\"set all the proper spectrogram hyperparameters\"。README邀请大家帮忙:\"join ... if you are interested in helping out with the replication with the LAION community\"。","source":"ai","title":"复现工作仍未完成"},{"code":"no_longer_updated","severity":"info","evidence":"最后一次提交和最后一次发布都在2023-09-06,setup.py中仍标注为\"Development Status :: 4 - Beta\",版本号0.2.8。","source":"ai","title":"已经不再更新"}],"platforms":["linux","windows"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","development"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":3294,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":3294},{"date":"2026-11-06","stars":3294},{"date":"2026-11-07","stars":3294},{"date":"2026-11-08","stars":3294},{"date":"2026-11-09","stars":3294},{"date":"2026-11-10","stars":3294}],"firstPublishedAt":"2026-11-07T22:56:16.841Z","commitSha":"4a78b41feab25d0ad0f4a1be8ea46b0aaf83d23a","verifiedAt":"2026-11-07T22:43:49.565Z","lastCommitAt":"2023-09-06T16:36:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Python code that rebuilds Google's MusicLM text-to-music model, for researchers who will train it themselves.","summary":"This is a community rebuild, in PyTorch, of MusicLM, the Google model that turns a text description into music. It is a library, not a program: you install it with pip and write Python code. It ships no trained model, so it makes no music until you gather a large collection of audio with matching text and train several networks on a graphics card, which takes serious time and hardware. It is for machine learning researchers and developers; a person who just wants a song from a sentence cannot use it as it is.","easeRationale":"On Linux, the usual system for this kind of training, it is developers only: `pip install musiclm-pytorch`, then write Python, prepare a dataset of sound and text pairs and train three transformers plus the MuLaN model before anything can be generated. Windows works the same way if you have an NVIDIA card, since the example code calls .cuda(); there is no app, no interface and no ready model to download.","costNote":"The code is free (MIT). The real cost is the hardware or rented GPU time needed to train the models.","notVerified":["Whether training it really produces music close to Google's published MusicLM examples","How much audio data, GPU time and money a usable model would take","Whether the pinned dependencies (audiolm-pytorch, x-clip, torch) still install together today","Real disk use once a dataset and the training checkpoints are on disk","Which rights apply to the music collection you would need for training"]},"zh-Hans":{"summaryShort":"用Python复现谷歌文本转音乐模型MusicLM的代码,供需要自行训练的研究者使用。","summary":"这是社区用PyTorch重建的MusicLM,谷歌那个能把文字描述转成音乐的模型。它是一个库,不是一个程序:你需要用pip安装并编写Python代码。它不附带任何已训练好的模型,所以在你收集大量音频和对应文本并在显卡上训练多个网络之前,不会生成任何音乐,这需要相当多的时间和硬件。它面向机器学习研究者和开发者;只想靠一句话得到一首歌的人无法直接使用它。","easeRationale":"在Linux上(这类训练常用的系统),这完全是面向开发者的:`pip install musiclm-pytorch`,然后编写Python代码,准备一份声音和文本配对的数据集,并训练三个transformer加上MuLaN模型,之后才能生成任何内容。Windows的流程相同,只要你有NVIDIA显卡即可,因为示例代码调用了.cuda();没有应用程序,没有界面,也没有现成模型可供下载。","costNote":"代码是免费的(MIT协议)。真正的成本在于训练模型所需的硬件或租用的GPU时间。","notVerified":["训练后是否真的能生成接近谷歌公开的MusicLM示例的音乐","想要一个可用的模型,需要多少音频数据、GPU时间和资金","固定版本的依赖项(audiolm-pytorch、x-clip、torch)如今是否还能一起安装成功","数据集和训练检查点落盘后,实际占用多少磁盘空间","训练所需的音乐素材,适用哪些权利规定"]}}},{"id":"R_kgDOVNvIig","slug":"ly05010419__macnativegames","name":"MacNativeGames","owner":"ly05010419","fullName":"ly05010419/MacNativeGames","type":"needs_setup","status":"alert","grade":"E","easeByPlatform":{"mac":"E"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"10 天内获得 6041 个星标","source":"check","title":"新仓库却有大量星标"},{"code":"needs_console_keys_and_dumps","severity":"high","evidence":"README.md：\"bring your game file (XCI or NSP) and your `prod.keys`\" 以及 \"Whether dumping games, keys and system files is legal depends on where you live; check and follow your local laws.\" 获取 prod.keys 并读取 XCI 或 NSP 文件，意味着绕过 Switch 自身的保护，这在一些国家即使对你购买的游戏也是违法的。","source":"ai","title":"需要主机密钥和游戏转储文件"},{"code":"unauthorised_port_of_commercial_games","severity":"attention","evidence":"README.md 列出了 13 款商业游戏的移植版（Tears of the Kingdom、Breath of the Wild、Super Smash Bros. Ultimate、Animal Crossing 等），并声明：\"This is an independent, non-commercial project. It is not affiliated with, endorsed by, or sponsored by Nintendo.\" 版权方并未授权，因此该仓库可能被下架，项目也可能随时停止。","source":"ai","title":"未经授权移植商业游戏"},{"code":"ai_assistant_runs_the_install","severity":"attention","evidence":"README.md：\"Recommended: let an AI assistant install and test it for you. Open this repository in Claude Code, Codex, or any other AI assistant you like ... and ask it to build and install the app and to sort out whatever comes up on your Mac\"，依据的是 .claude/skills/install-zelda/SKILL.md。这意味着让助手在你的电脑上运行仓库中写好的命令。","source":"ai","title":"由 AI 助手执行安装"},{"code":"sends_data_to_the_project","severity":"info","evidence":"README.md：每新增 100 行数据，Mac 应用会在游戏启动前先询问；如果已登录 GitHub CLI，则 \"the dialog offers \\\"Submit automatically\\\" ... which sends this batch for you (a private gist and an issue) without opening the browser\"。项目称该列表只包含着色器标识符，从不包含游戏数据，并且总会先询问。","source":"ai","title":"会向项目发送数据"},{"code":"mostly_early_community_ports","severity":"info","evidence":"在游戏表格中，只有 Tears of the Kingdom 和 Breath of the Wild 标为 \"Supported\"；其余十一款游戏是 \"Community port, early\"、\"work in progress\" 或 \"experimental\"，Vision Pro 则是 \"windowed, early\"。","source":"ai","title":"多数是处于早期阶段的社区移植"}],"platforms":["mac"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":7470,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":510,"growthDays":7,"history":[{"date":"2026-10-27","stars":6038},{"date":"2026-10-28","stars":6264},{"date":"2026-10-29","stars":6442},{"date":"2026-10-30","stars":6592},{"date":"2026-10-31","stars":6686},{"date":"2026-11-01","stars":6781},{"date":"2026-11-02","stars":6881},{"date":"2026-11-03","stars":6960},{"date":"2026-11-04","stars":7024},{"date":"2026-11-05","stars":7099},{"date":"2026-11-06","stars":7185},{"date":"2026-11-07","stars":7261},{"date":"2026-11-08","stars":7326},{"date":"2026-11-09","stars":7387},{"date":"2026-11-10","stars":7470}],"firstPublishedAt":"2026-10-27T02:46:40.217Z","commitSha":"ff63a162b35d8b486b941191785d8e957c1f5df2","verifiedAt":"2026-10-28T01:38:49.068Z","lastCommitAt":"2026-11-10T21:13:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Build scripts that turn Switch games you own, such as Zelda, into an app that runs on an Apple silicon Mac.","summary":"MacNativeGames is a set of scripts that take a Nintendo Switch game file and the keys from your own console and build a Mac app (or an unsigned iPhone/iPad package) that runs the game's own code on Apple chips, with the Switch graphics rebuilt on Metal. Tears of the Kingdom and Breath of the Wild are the main cases; a dozen other games are early community ports. It ships no game content. You need an Apple silicon Mac, developer tools, a dumped game file and prod.keys, and a build that takes 20 to 40 minutes. The author suggests letting an AI assistant run the steps.","easeRationale":"On an Apple silicon Mac it is grade E: there is no ready download, you build the app from source with Python scripts, clang and Homebrew libraries, a build takes 20 to 40 minutes, and the author says so plainly: \"Installing and testing by hand is not recommended: there are many steps\" and \"I haven't provided a one-click build option\". The iPhone, iPad and Vision Pro routes are not systems Radar grades.","costNote":"The project is free (MIT). You must buy the Switch games yourself and get the files from your own console.","notVerified":["Whether it is lawful where you live to copy the game and the keys off your own console","Whether the build works with your exact game version and your own dump","How well each game actually plays, and how much it stutters while shaders compile","Real disk use, since a Switch game file plus the built app takes tens of gigabytes","What exactly the shader contribution file contains before it is sent to the project"]},"zh-Hans":{"summaryShort":"一套构建脚本,可以把你拥有的 Switch 游戏(例如《塞尔达传说》)变成能在搭载苹果芯片的 Mac 上运行的应用。","summary":"MacNativeGames 是一套脚本,它使用一份任天堂 Switch 游戏文件和你自己主机上的密钥,构建出一个 Mac 应用(或未签名的 iPhone/iPad 安装包),在苹果芯片上运行游戏本身的代码,并把 Switch 的图形重建在 Metal 之上。《王国之泪》和《旷野之息》是主要案例;另外十几款游戏还是社区早期移植版本。它不包含任何游戏内容。你需要一台搭载苹果芯片的 Mac、开发者工具、一份提取出来的游戏文件和 prod.keys,构建过程需要 20 到 40 分钟。作者建议让 AI 助手来执行这些步骤。","easeRationale":"在搭载苹果芯片的 Mac 上评级为 E:没有现成的下载包,你需要用 Python 脚本、clang 和 Homebrew 库从源代码构建应用,一次构建需要 20 到 40 分钟,作者自己也直言:“不建议手动安装和测试:步骤太多”以及“我没有提供一键构建的选项”。iPhone、iPad 和 Vision Pro 这几条路线不属于 Radar 评级的系统。","costNote":"该项目是免费的(MIT 许可)。你必须自己购买 Switch 游戏,并从自己的主机上获取文件。","notVerified":["在你所在地,从自己的主机上复制游戏和密钥是否合法","构建结果是否适用于你手上确切的游戏版本和你自己的提取文件","每款游戏实际的游玩体验如何,着色器编译时卡顿程度如何","实际占用的存储空间,因为一份 Switch 游戏文件加上构建出的应用会占用数十 GB","着色器贡献文件在发送给该项目之前具体包含哪些内容"]}}},{"id":"R_kgDOSgw_Ww","slug":"lynote-ai__humanize-text","name":"Humanize Text","owner":"lynote-ai","fullName":"lynote-ai/humanize-text","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"made_to_pass_ai_detectors","severity":"attention","evidence":"该项目的目的是让AI文本读起来像人写的,并通过AI检测工具:处理流程旨在\"break the statistical patterns typical of AI text\",使得\"no single engine's fingerprint survives\",`focus`档位增加了一个\"detection-guided feedback loop\"以实现\"Maximum restructuring\",展示表格里列出\"Detector verdict on our test\"为\"human\",置信度最高达0.9997。项目主题中包括\"undetectable-ai-alternative\"。学校、期刊和许多雇主对使用及披露AI写作有相关规定;README本身也写道\"follow your institution's policies on AI use and disclosure. Please don't use it to misrepresent authorship.\"","source":"ai","title":"专为通过AI检测而设计"},{"code":"text_sent_to_several_outside_services","severity":"info","evidence":"一次运行最多会把你的草稿发送到四个外部服务:\"Step 1 LLM ... Step 2 LLM ... Step 3 Google Translate ... Step 4 Niutrans\"。你粘贴的任何内容,包括尚未发表的作品,都会经过该模型服务商和两个翻译引擎。","source":"ai","title":"文本被发送到多个外部服务"},{"code":"promotes_the_makers_paid_service","severity":"info","evidence":"该仓库属于出售托管版本的公司,README中反复用带跟踪参数的链接指向它,例如\"Try it online\"指向\"lynote.ai/ai-humanizer?utm_source=github&utm_medium=readme...\",还有一张对比表\"This repo | Lynote.ai\",其中付费服务使用\"our own model\"。README在谈到开放的训练流程时也说:\"Model weights and full datasets are not included.\"","source":"ai","title":"为开发者的付费服务做推广"},{"code":"quality_numbers_from_the_maker","severity":"info","evidence":"README中的分数是开发者自己给出的:\"Expert evaluation on 50 text pairs\",\"Overall 9.1\"(满分10分),以及\"Key information retention: 100% (50/50 pairs)\",检测结果则是\"illustrative snapshots from one detector at one point in time, not a guarantee\"。","source":"ai","title":"质量数据来自开发者自己"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["DeepSeek or another OpenAI-compatible model provider (OpenRouter, Atlas Cloud, OrcaRouter, LiteLLM)","Niutrans (translation)","Google Translate (through deep-translator)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[{"slug":"undetectable-ai","name":"Undetectable AI","confirmed":true},{"slug":"quillbot","name":"QuillBot","confirmed":true},{"slug":"lynote-ai","name":"Lynote.ai","confirmed":false}],"declaredAlternatives":[],"stars":3497,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":16,"growthDays":6,"history":[{"date":"2026-11-06","stars":3485},{"date":"2026-11-07","stars":3485},{"date":"2026-11-08","stars":3487},{"date":"2026-11-09","stars":3490},{"date":"2026-11-10","stars":3497}],"firstPublishedAt":"2026-11-07T22:56:16.848Z","commitSha":"3be99a5e83cef690dcdedb49dc0aa656c259e55b","verifiedAt":"2026-11-07T22:44:51.886Z","lastCommitAt":"2026-10-22T09:08:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Command-line tool that rewrites AI-written text through two rewrites and two translations so it reads less like AI.","summary":"Humanize Text takes a draft written by AI and rewrites it so it sounds more like a person. It sends the text through four steps: two rewrites by a language model (DeepSeek by default) and two machine translations in different engines, hopping through Chinese, Japanese and Finnish before coming back to English. You run it as a Python command, as a Docker service with a REST endpoint, or as an n8n workflow, and you need your own model and translation keys. Every intermediate step can be printed, so you can see what changed.","easeRationale":"The same four steps on all three systems, with Linux and macOS matching the commands in the README exactly: install Python 3.10 or newer, clone the repository, run pip install, copy the example config and paste two API keys, then run one command per text. Docker is offered as an alternative for a REST endpoint. There is no installer, no window to click in, and nothing to download from the releases page.","costNote":"The code is free (MIT), but it only works with paid keys: a language model provider (DeepSeek by default, or OpenRouter and others) and a Niutrans translation key. The maker's hosted service is sold separately at \"1 credit per input word\".","notVerified":["Whether the rewritten text passes the detector your school, journal or client actually uses","How much the model and translation keys cost for a long document","How much meaning, citations and technical terms drift after four steps, which the README warns about","Whether the Docker and n8n paths work as described","What the model provider and the translation engines keep of the text you send"]},"zh-Hans":{"summaryShort":"命令行工具,通过两次改写和两次翻译重写AI写的文本,让它看起来不那么像AI写的。","summary":"Humanize Text接收一段AI写的草稿,并重写它使其更像人写的。它让文本经过四个步骤:由语言模型(默认DeepSeek)完成两次改写,再由不同引擎完成两次机器翻译,依次经过中文、日文和芬兰文,最后再译回英文。你可以把它当作Python命令运行,作为带REST接口的Docker服务运行,或作为n8n工作流运行,并且需要自己的模型和翻译密钥。每一个中间步骤都可以打印出来,让你看到发生了什么变化。","easeRationale":"三个系统的四个步骤相同,Linux和macOS与README中的命令完全一致:安装Python 3.10或更高版本,克隆仓库,运行pip install,复制示例配置并粘贴两个API密钥,然后为每段文本运行一条命令。Docker作为提供REST接口的替代方案。没有安装程序,没有可点击的窗口,发布页面上也没有任何可下载的内容。","costNote":"代码是免费的(MIT协议),但只能配合付费密钥使用:一个语言模型服务商(默认DeepSeek,或OpenRouter等)和一个Niutrans翻译密钥。开发者托管的服务另行出售,价格为\"1 credit per input word\"(每个输入单词1个积分)。","notVerified":["改写后的文本能否通过你的学校、期刊或客户实际使用的检测工具","处理一篇长文档,模型和翻译密钥要花多少钱","经过四个步骤后,含义、引文和专业术语会偏离多少,README对此有提醒","Docker和n8n的使用方式是否如文档所述那样可行","模型服务商和翻译引擎会保留你发送文本的哪些内容"]}}},{"id":"R_kgDOSGKjGA","slug":"lynpoint__cyberverse","name":"CyberVerse","owner":"Lynpoint","fullName":"Lynpoint/CyberVerse","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"makes_a_real_persons_face_speak","severity":"attention","evidence":"README提出可以用一张照片让真人\"动起来\",包括已经去世的人:\"Want to see someone you've lost again, hear their voice, watch them smile at you?\"以及\"Just one photo. CyberVerse makes them alive.\"演示角色中包括一个小说人物(小龙女),README指出这些角色\"are not bundled with CyberVerse and are not provided for commercial use\"。使用他人的脸或声音,或者别人拥有的角色,需要得到对方许可,多个国家对未经同意制作的个人形象有相关规定。","source":"ai","title":"让真人的脸说话"},{"code":"needs_a_top_end_graphics_card","severity":"info","evidence":"实时表格列出\"FlashHead 1.3B | Pro | RTX 5090 | 2 | 512×512 | 25+\"和\"LiveAct 18B | RTX PRO 6000\"。README还说\"For pure voice sessions, no local avatar GPU is required\",所以需要这类硬件或云端数字人服务的是会说话的脸这部分功能。","source":"ai","title":"需要顶级显卡"},{"code":"referral_links_in_readme","severity":"info","evidence":"README中有几个链接带着作者自己的代码:赞助板块写着\"Register through this invitation link\",指向\"passport.compshare.cn/register?referral_code=IBmJcGPVu1RF78dMihkQCX\",快速启动云镜像里重复了同样的代码,Vidu S1那一行链接到\"vidu.cn/subscription/personal?bonus-code=GWCK5M\"。","source":"ai","title":"README中含有推荐链接"},{"code":"weights_and_services_from_outside","severity":"info","evidence":"数字人模型需要单独从Hugging Face或ModelScope下载(\"hf download Soul-AILab/SoulX-FlashHead-1_3B\"、\"hf download Soul-AILab/LiveAct\"),并为中国大陆用户建议使用镜像(\"export HF_ENDPOINT=https://hf-mirror.com\"),可选的加速功能则基于其他仓库编译而成(LightX2V、CUTLASS、SageAttention、FlashAttention)。","source":"ai","title":"权重和服务来自外部"},{"code":"conversations_go_to_cloud_providers","severity":"info","evidence":"默认配置会把语音和文本发送到外部服务:config/env中需要填写\"DASHSCOPE_API_KEY\"(阿里云)、\"DOUBAO_ACCESS_TOKEN\"/\"DOUBAO_APP_ID\"(火山引擎)、\"ATLASCLOUD_API_KEY\"或\"CHEAPER_INFERENCE_API_KEY\",云端数字人则使用\"BAIDU_XILING_APP_ID\"/\"BAIDU_XILING_APP_KEY\"。角色历史记录和导入的文档保存在本地磁盘上。","source":"ai","title":"对话内容发送到云服务商"},{"code":"very_young_project","severity":"info","evidence":"创建于2026-04-18,只发布过一个版本v0.1.0(2026-05-16),pyproject.toml中仍写着version = \"0.1.0\";路线图和部分社区使用中文(语雀和一个微信群)。","source":"ai","title":"项目非常年轻"}],"platforms":["linux"],"tools":["python","node","go","gpu_nvidia","ffmpeg"],"buildTools":["python","node","gpu_nvidia","ffmpeg"],"disk":{"minMB":410,"maxMB":1370,"basis":"heuristic"},"apiKeys":["A model and voice provider (Alibaba Cloud DashScope/Qwen, Volcengine Doubao, Atlas Cloud, Cheaper Inference or another OpenAI-compatible service)","Baidu Xiling Digital Human","Xunfei Digital Human","Vidu S1 (cloud avatar)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["ai_agents","video","audio"],"alternativeTo":[{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"character-ai","name":"Character.AI","confirmed":true}],"declaredAlternatives":[],"stars":1945,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":195,"growthDays":6,"history":[{"date":"2026-11-06","stars":1851},{"date":"2026-11-07","stars":1865},{"date":"2026-11-08","stars":1892},{"date":"2026-11-09","stars":1919},{"date":"2026-11-10","stars":1945}],"firstPublishedAt":"2026-11-07T22:56:16.855Z","commitSha":"f6247428aa5efbdd60a07ab1e9306c401517d5ec","verifiedAt":"2026-11-07T22:45:49.918Z","lastCommitAt":"2026-10-21T10:28:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Self-hosted platform for an AI character you talk to by voice, with an optional talking face made from one photo.","summary":"CyberVerse runs on your own machine or rented server and gives you an AI character you speak with in real time through the browser, with interruptions like a video call. It keeps each character's memory and can read your documents to answer in character. Voice-only mode needs no graphics card, just paid voice and model services. The talking face is optional and heavy: the README's real-time table uses an RTX 5090 or RTX PRO 6000, or a cloud avatar service. Setting it up means Node, Go, Conda, Python, FFmpeg and three terminals.","easeRationale":"Linux is the only documented system and it is for developers: the prerequisites are \"Node 18+, Go 1.25 (required: protoc-gen-go, protoc-gen-go-grpc), Conda, Python 3.10+, FFmpeg, libopus-dev, libopusfile-dev, libsoxr-dev, pkg-config\", then you clone, edit two config files, run make setup and keep three terminals running. The talking face adds CUDA 12.8, a specific PyTorch build, model weights from Hugging Face and, for the optional speed-ups, compiling kernels from source.","costNote":"The code is free (GPL-3.0). The README says \"Runtime cost depends on the realtime voice/omni/LLM/TTS/ASR providers you configure\", so you pay per use for the model, speech and avatar services, or for the GPU that runs the local avatar models.","notVerified":["Whether the real-time avatar keeps up on a card weaker than the ones in the README","How much the voice, model and cloud avatar services cost per hour of conversation","Whether the full setup completes on a current Linux without manual fixes","Real disk use after the avatar model weights are downloaded","Whether the digital human looks and sounds as in the demo videos"]},"zh-Hans":{"summaryShort":"自托管平台,提供一个可语音对话的AI角色,还可选用一张照片生成会说话的脸。","summary":"CyberVerse运行在你自己的电脑或租用的服务器上,提供一个可以在浏览器中实时对话的AI角色,像视频通话一样可以被打断。它会保留每个角色的记忆,并能读取你的文档,以该角色的身份回答。纯语音模式不需要显卡,只需要付费的语音和模型服务。会说话的脸是可选且较重的功能:README的实时表格里用的是RTX 5090或RTX PRO 6000,或者云端数字人服务。搭建它需要Node、Go、Conda、Python、FFmpeg,还要开三个终端。","easeRationale":"Linux是唯一有文档说明的系统,而且是面向开发者的:前置条件是\"Node 18+, Go 1.25 (required: protoc-gen-go, protoc-gen-go-grpc), Conda, Python 3.10+, FFmpeg, libopus-dev, libopusfile-dev, libsoxr-dev, pkg-config\",然后需要克隆仓库、编辑两个配置文件、运行make setup,并保持三个终端同时运行。会说话的脸还需要CUDA 12.8、特定版本的PyTorch、来自Hugging Face的模型权重,以及(如果要启用可选的加速)从源代码编译内核。","costNote":"代码是免费的(GPL-3.0协议)。README写道\"Runtime cost depends on the realtime voice/omni/LLM/TTS/ASR providers you configure\",所以你需要为模型、语音和数字人服务按使用量付费,或者为运行本地数字人模型的GPU付费。","notVerified":["在比README中更弱的显卡上,实时数字人是否还能跟得上","语音、模型和云端数字人服务每小时对话要花多少钱","在当前的Linux上,完整安装是否无需手动修复就能完成","下载数字人模型权重后,实际占用多少磁盘空间","数字人的样子和声音是否和演示视频中一样"]}}},{"id":"R_kgDOVRtv7Q","slug":"lyramakesmusic__neuron-explorer","name":"Neuron Explorer","owner":"lyramakesmusic","fullName":"lyramakesmusic/neuron-explorer","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"bundled_agent_has_shell_access","severity":"attention","evidence":"README，Claude 标签页：\"a chat running the `claude` CLI (installed and logged in) with the explorer as an MCP server (`explorer/agent_mcp.py`)\"，它 \"has file tools and a shell in your home folder\"。","source":"ai","title":"内置智能体可以使用 shell"},{"code":"setup_script_installs_third_party_fork","severity":"info","evidence":"README，Sweep：\"bash sweep/setup_pod.sh  # fork, deps, model\"，它会安装 vLLM 的 steering fork（RhizoNymph/vllm），并在 GPU 机器（pod）上下载模型。","source":"ai","title":"安装脚本会安装第三方 fork"},{"code":"research_preview_no_releases","severity":"info","evidence":"pyproject.toml 写的是 version = \"0.1.0\"，GitHub 上没有发布版本，在被分析的提交时仓库才创建一周（创建于 2026-10-21，最后一次推送为 2026-10-28），有 1615 个星标。","source":"ai","title":"研究预览版，没有发布版本"},{"code":"ships_prebuilt_caches","severity":"info","evidence":"README，Explorer：\"the caches in `explorer/cache/` ship with the repo\" —— 这些是二进制的 .npz 文件（layout、connections），你可以加载它们，却无法阅读其内容，由维护者自己的 sweep 生成。","source":"ai","title":"自带预先生成的缓存"}],"platforms":["mac","linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["Claude CLI login, for the optional Claude tab"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","local_ai"],"alternativeTo":[{"slug":"goodfire-ember","name":"Ember","confirmed":true}],"declaredAlternatives":[],"stars":1965,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":73,"growthDays":7,"history":[{"date":"2026-10-27","stars":1476},{"date":"2026-10-28","stars":1614},{"date":"2026-10-29","stars":1703},{"date":"2026-10-30","stars":1748},{"date":"2026-10-31","stars":1776},{"date":"2026-11-01","stars":1803},{"date":"2026-11-02","stars":1859},{"date":"2026-11-03","stars":1892},{"date":"2026-11-04","stars":1912},{"date":"2026-11-05","stars":1930},{"date":"2026-11-06","stars":1935},{"date":"2026-11-07","stars":1937},{"date":"2026-11-08","stars":1944},{"date":"2026-11-09","stars":1955},{"date":"2026-11-10","stars":1965}],"firstPublishedAt":"2026-10-27T18:40:27.963Z","commitSha":"6a6f0e95bf49ad6626940021148f775d7d55d1e4","verifiedAt":"2026-10-29T01:55:55.597Z","lastCommitAt":"2026-11-01T00:46:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A 3D explorer that pushes single neurons of a small AI model and shows how its answers change.","summary":"A research tool for people who study what happens inside a language model. It runs Qwen3.5-4B on your own computer and draws its layers in 3D, one dot per neuron: you push a neuron up or down, read the text the model then writes, search the stored runs and chat with the pushed model. The heavy data comes from a sweep that needs a rented CUDA graphics pod and a modified build of vLLM. You start everything with uv and Python commands. The code is under the MIT license.","easeRationale":"Mac and Linux are the systems the README names and both land on D: there is nothing to download and open, you install Python packages with uv and start a local server from the command line, and it loads a 4-billion-parameter model on your machine (MPS, CUDA or CPU). The data sweep behind it is run on a rented CUDA GPU pod with a forked build of vLLM, which is why neither system goes higher.","costNote":"The code and the model are free to download; the sweep that produces the data is described as running on a rented CUDA 13 GPU pod, which is not.","notVerified":["Whether it runs on Windows at all: the README only names MPS (Mac), CUDA and CPU, and the sweep scripts are shell scripts","Real disk and memory use once Qwen3.5-4B, the sweep runs and the UMAP data are on your machine","What the bundled Claude tab does while it works, since the README says it has a shell in your home folder","Whether the caches shipped in explorer/cache match the model weights you download, and how long a sweep costs on a rented GPU pod"]},"zh-Hans":{"summaryShort":"一个3D探索工具,可以调动小型AI模型中的单个神经元,展示其回答如何随之改变。","summary":"这是一个供研究语言模型内部机制的人使用的工具。它在你自己的电脑上运行Qwen3.5-4B模型,并以3D方式绘制各层结构,每个神经元对应一个点:你可以把某个神经元调高或调低,查看模型随后生成的文字,搜索已保存的运行记录,并与被调整过的模型对话。背后的大量数据来自一次扫描,需要租用CUDA显卡云主机和一个改过的vLLM版本。整个过程通过uv和Python命令启动。代码采用MIT许可。","easeRationale":"README提到的系统只有Mac和Linux,两者都是D级:没有可以下载直接打开的文件,你需要用uv安装Python包,再从命令行启动本地服务器,它会在你的机器上(MPS、CUDA或CPU)加载一个40亿参数的模型。背后生成数据的扫描过程是在租用的CUDA GPU云主机上、用一个改过的vLLM版本运行的,这也是两个系统评级都上不去的原因。","costNote":"代码和模型本身可以免费下载;但生成数据所需的扫描过程据称运行在租用的CUDA 13 GPU云主机上,这部分并不是免费的。","notVerified":["它能否在Windows上运行尚不确定:README只提到了MPS(Mac)、CUDA和CPU,而扫描脚本是shell脚本。","在你的机器上加载Qwen3.5-4B、运行扫描并存放UMAP数据后,实际占用的磁盘和内存情况。","内置的Claude标签页在运行时具体做了什么,因为README提到它在你的用户目录下有一个shell。","explorer/cache中附带的缓存是否与你下载的模型权重匹配,以及在租用的GPU云主机上运行一次扫描要花多少钱。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMDIzNjc2MTU=","slug":"lyswhut__lx-music-desktop","name":"LX Music","owner":"lyswhut","fullName":"lyswhut/lx-music-desktop","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"high","signals":[{"code":"plays_paid_music_catalogues_without_a_subscription","severity":"high","evidence":"README 中项目的补充条款把内置音源指明为官方平台酷我、酷狗、咪咕等(\"酷我、酷狗、咪咕等音乐源的官方平台\"),说数据是从它们的公开服务器获取的,就像一个未登录的应用看到的那样,并要求:\"使用者务必在 24 小时内清除使用本项目的过程中所产生的版权数据\"。同一段文字还说使用\"仅用于对技术可行性的探索及研究\",\"禁止在违反当地法律法规的情况下使用本项目\",\"请尊重版权,支持正版\"。换句话说,这个应用日常的用途就是收听那些平台按订阅收费的音乐内容,而项目把法律责任放在了你身上。","source":"ai","title":"无需订阅即可播放付费音乐目录"},{"code":"music_sources_come_from_outside_the_project","severity":"attention","evidence":"README 条款 1.2:\"本项目本身没有获取某个音频数据的能力\"——项目本身无法获取音频;它把歌名和歌手交给\"自定义源\"设置里选择的\"源\",播放返回的链接,而且\"无法校验其准确性\"。这些源文件由其他人编写和分发,不是这个仓库写的,它们在应用内部运行。","source":"ai","title":"音源来自项目之外"},{"code":"development_moved_to_a_different_project","severity":"attention","evidence":"README 的说明:新项目 Any Listen 已经覆盖了 LX Music 的大部分功能,\"我们以后的开发精力将主要集中在新项目上\",邀请用户迁移过去。这里最后一次提交已经过去42天(2026-09-19),而未关闭的issue有1308个。","source":"ai","title":"开发重心已转移到另一个项目"},{"code":"optional_local_http_server_and_browser_control","severity":"info","evidence":"README:从 v2.7.0 开始,\"开放 API 支持\"功能会启动一个本地 HTTP 服务,把播放器接口暴露给其他程序;从 v1.17.0 开始,Scheme URL 功能让浏览器可以调用这个应用,配套的用户脚本发布在 Greasyfork 上。这两项都是由你自己打开的,你机器上任何能访问到那个本地服务的程序都可以控制播放器。","source":"ai","title":"可选的本地HTTP服务器和浏览器控制"},{"code":"only_github_releases_are_the_real_downloads","severity":"info","evidence":"README:\"目前本项目的原始发布地址只有 GitHub,其他渠道均为第三方转载发布,与本项目无关!\"也就是说唯一的原始发布地址就是这个 GitHub 发布页,其他渠道都是第三方转载,项目方不承担关系。在别处找到的副本不在此次检查的覆盖范围内。","source":"ai","title":"只有GitHub发布才是真正的下载渠道"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":116,"maxMB":347,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio"],"alternativeTo":[{"slug":"kugou-music","name":"Kugou Music","confirmed":true},{"slug":"kuwo-music","name":"Kuwo Music","confirmed":true},{"slug":"migu-music","name":"Migu Music","confirmed":true}],"declaredAlternatives":[],"stars":55237,"downloads":44701720,"downloadsWeek":227883,"downloadsDays":7,"growth":201,"growthDays":7,"history":[{"date":"2026-10-27","stars":54827},{"date":"2026-10-28","stars":54863},{"date":"2026-10-29","stars":54890},{"date":"2026-10-30","stars":54919},{"date":"2026-10-31","stars":54947},{"date":"2026-11-01","stars":54980},{"date":"2026-11-02","stars":55008},{"date":"2026-11-03","stars":55036},{"date":"2026-11-04","stars":55063},{"date":"2026-11-05","stars":55086},{"date":"2026-11-06","stars":55117},{"date":"2026-11-07","stars":55153},{"date":"2026-11-08","stars":55182},{"date":"2026-11-09","stars":55207},{"date":"2026-11-10","stars":55237}],"firstPublishedAt":"2026-10-27T02:46:38.189Z","commitSha":"ad95d5091c9ed689fa72b5e5c849df65f5a679ce","verifiedAt":"2026-11-01T01:45:00.350Z","lastCommitAt":"2026-11-07T02:05:42Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop music app that searches songs from Chinese streaming platforms and plays them through a source you add yourself.","summary":"LX Music (洛雪音乐助手) is an Electron desktop app that searches songs on Chinese music platforms such as Kuwo, Kugou and Migu, shows the results and plays them, with playlists, lyrics and keyboard shortcuts. It carries no music of its own: the app's own terms say it only passes the song name and artist to a \"custom source\" you set up in the settings, and plays whatever link that source returns. Ready installers exist for Windows 7+, macOS and Linux. The app and all its documentation are in Chinese.","easeRationale":"Every system has a ready download: Setup .exe files (including a Windows 7 build) and portable .7z archives for Windows, .dmg for macOS, .deb, .rpm and AppImage for Linux. What keeps it off the top grade is that nothing plays until you add a music source in the settings, and the README says the defaults and the interface are deliberately not beginner-friendly (\"为了提高使用门槛\") and asks you to read the playlist and shortcut guides first — all of which are in Chinese only.","costNote":"Completely free, Apache-2.0, with no ads: its terms say the project \"不接受任何商业（包括但不限于广告等）合作及捐赠\" — it accepts no commercial deals and no donations.","notVerified":["Whether any music source still works, since the sources are supplied by third parties and change often","What a given source file does besides returning links, which only reading and running it would show","Whether the audio that plays is really the song you asked for, which the project itself says it cannot check","Real disk use after installing, estimated at 116 to 347 MB","Whether the app works for someone who does not read Chinese, as the interface and all documentation are in Chinese"]},"zh-Hans":{"summaryShort":"一个桌面音乐应用,从中国的流媒体平台搜索歌曲,并通过你自己添加的音源播放。","summary":"LX Music(洛雪音乐助手)是一个用 Electron 开发的桌面应用,会在酷我、酷狗、咪咕等中国音乐平台搜索歌曲,显示结果并播放,带有歌单、歌词和键盘快捷键。它本身不携带任何音乐:应用自己的条款说,它只把歌名和歌手传给你在设置里配置的\"自定义源\",然后播放那个源返回的链接。Windows 7+、macOS 和 Linux 都有现成的安装包。应用本身以及全部文档都是中文。","easeRationale":"每个系统都有现成的下载:Windows 的 Setup .exe 文件(包括一个 Windows 7 版本)和便携版 .7z 压缩包,macOS 的 .dmg,以及 Linux 的 .deb、.rpm 和 AppImage。让它拿不到最高评级的原因是,在设置里添加音源之前什么都播放不了,而且 README 说默认设置和界面是故意不照顾新手的(\"为了提高使用门槛\"),要求你先读播放列表和快捷键的说明——这些说明全部只有中文。","costNote":"完全免费,Apache-2.0 许可,没有广告:其条款写道项目\"不接受任何商业(包括但不限于广告等)合作及捐赠\"——不接受任何商业合作或捐赠。","notVerified":["是否还有音源能用,因为音源由第三方提供,经常变化","某个具体的源文件除了返回链接之外还做了什么,这只有读代码并运行才能看出来","播放的音频是否真的是你要的那首歌,项目自己说它无法核实这一点","安装后实际占用的磁盘空间,估计在116到347MB之间","对不懂中文的人来说这个应用是否能用,因为界面和全部文档都是中文"]}}},{"id":"R_kgDOVT9BlA","slug":"man15h__tellyscope","name":"Tellyscope","owner":"man15h","fullName":"man15h/tellyscope","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"needs_debugging_turned_on_the_tv","severity":"attention","evidence":"使用时必须让电视的调试连接保持开启:\"In Developer options, turn on USB debugging (Fire TV: ADB debugging), and Network debugging or Wireless debugging if your TV lists it\",电视会要求你勾选 \"Always allow from this computer\"。应用本身为此增加了一个界面:\"Who has access. See every computer the TV allows over ADB, remove all but this phone\"。","source":"ai","title":"需要在电视上开启调试"},{"code":"turning_off_tv_apps_can_break_features","severity":"info","evidence":"README 在注意事项开篇写道 \"Debloating a TV can break it if done carelessly\"。项目通过设计限制损害:内置应用只会被禁用(\"Tellyscope never runs pm uninstall --user 0\"),有一份电视启动所需的受保护应用列表,还有带撤销功能的历史记录,并且 \"Every change names the apps it touches and asks before it runs\"。","source":"ai","title":"关闭电视应用可能导致功能损坏"},{"code":"installs_apps_downloaded_from_github","severity":"info","evidence":"GitHub 构建版本可能从网上获取另一个应用并安装到你的电视上:\"Fire TV has no Play Store, so Tellyscope downloads Projectivy from its developer's GitHub releases.\" 项目限制了可安装的范围:\"Which launcher APK may be installed, where it downloads from and who must have signed it come only from data/launchers.json\"。Play 商店版本 \"downloads nothing from GitHub\"。","source":"ai","title":"安装从 GitHub 下载的应用"},{"code":"behaviour_lists_come_from_another_repository","severity":"info","evidence":"应用的部分行为由它下载的文件决定:\"Tellyscope downloads two sets of public lists from GitHub, so they can change without an app release\",其中包括来自独立的 tellyscope-devices 仓库的精简列表。README 说明这些列表无法移除内置的保护机制。","source":"ai","title":"行为列表来自另一个仓库"},{"code":"very_young_project_fast_releases","severity":"info","evidence":"该仓库创建于 2026-10-22,已经有 85 个发布版本;最新的三个是 v0.16.0 和两个来自 2026-11-06 的 \"-alpha\" 构建版本,Play 商店版本处于 \"in closed testing\",需要加入测试者群组。","source":"ai","title":"项目非常年轻,发布很快"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[{"slug":"vysor","name":"Vysor","confirmed":true}],"declaredAlternatives":[],"stars":1830,"downloads":37663,"downloadsWeek":9974,"downloadsDays":5,"growth":476,"growthDays":7,"history":[{"date":"2026-11-05","stars":1541},{"date":"2026-11-06","stars":1654},{"date":"2026-11-07","stars":1715},{"date":"2026-11-08","stars":1754},{"date":"2026-11-09","stars":1801},{"date":"2026-11-10","stars":1830}],"firstPublishedAt":"2026-11-07T23:09:40.938Z","commitSha":"f8a733a0dc15c2e258331b26dbfd14102671b3f4","verifiedAt":"2026-11-07T23:04:44.314Z","lastCommitAt":"2026-11-06T17:09:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Phone app that cleans up your Android TV, installs apps, sends files and works as a remote, with no PC and no root.","summary":"Tellyscope is a 3 MB Android phone app that talks to your Android TV, Google TV, Fire TV or NVIDIA Shield over Wi-Fi using ADB, the debugging connection built into Android. From the phone you can turn off the apps the TV came with (matched against 330+ tested lists, and you see every app before it happens), install apps from files, send files both ways, watch CPU, memory and network use, run shell commands, set a different home screen and use the phone as a remote. Almost every change goes into that TV's History with an Undo, except uninstalls and Shell commands.","easeRationale":"Android phones are the only place it runs, and there it is a B: you install the signed APK from the releases page (or join the closed Play test, or use Obtainium), and then you have to prepare the TV once, pressing Build seven times in its About screen, turning on USB or network debugging in Developer options and allowing the phone's key on the TV. Computers get no grade; there is no desktop version.","costNote":"Free and open source (GPL-3.0). The README says \"No ads, tracking or accounts\".","notVerified":["Whether it works on your TV's brand and model (the README says Fire OS 8 was tested and \"NVIDIA Shield hasn't been tested on real hardware yet\")","Whether there is a debloat list for your model, and what the TV loses when those apps are off","Whether Undo and \"Restore everything\" bring the TV back in every case","Whether leaving network debugging on causes trouble on your home network","How the machine-made translations read in your language"]},"zh-Hans":{"summaryShort":"手机应用,清理你的 Android TV 预装应用、安装应用、互传文件,还能当遥控器用,不用电脑也不用 root。","summary":"Tellyscope 是一款仅 3 MB 的 Android 手机应用,通过 Wi-Fi 使用 ADB(Android 内置的调试连接)与你的 Android TV、Google TV、Fire TV 或 NVIDIA Shield 通信。在手机上,你可以关闭电视自带的应用(比对超过 330 份经过测试的列表,执行前你会看到每一个应用),从文件安装应用,双向发送文件,查看 CPU、内存和网络使用情况,运行 shell 命令,设置不同的主屏幕,还能把手机当遥控器用。除了卸载和 Shell 命令外,几乎每次改动都会记录在该电视的历史记录中,并可以撤销。","easeRationale":"它只能在 Android 手机上运行,评级为 B:你可以从发布页面安装已签名的 APK(或加入 Play 封闭测试,或使用 Obtainium),然后需要对电视做一次准备:在“关于”界面连按七次 Build,在开发者选项中打开 USB 或网络调试,并在电视上允许该手机的密钥。电脑端没有评级,因为没有桌面版本。","costNote":"免费且开源(GPL-3.0)。README 中写道:\"No ads, tracking or accounts\"。","notVerified":["是否适用于你的电视品牌和型号(README 称已测试 Fire OS 8,并且 \"NVIDIA Shield hasn't been tested on real hardware yet\")","你的型号是否有对应的精简列表,以及关闭这些应用后电视会失去哪些功能","撤销和 \"Restore everything\" 是否在所有情况下都能把电视恢复原状","一直开着网络调试是否会给你的家庭网络带来麻烦","机器翻译的文字在你的语言里读起来怎么样"]}}},{"id":"R_kgDOHt-Wyg","slug":"marcoppasini__musika","name":"Musika","owner":"marcoppasini","fullName":"marcoppasini/musika","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2;numpy: GHSA-6p56-wp2h-9hxr, GHSA-fpfv-jqm9-f5jm;scipy: PYSEC-2023-102, PYSEC-2023-114;tensorflow: GHSA-27rc-728f-x5w2, GHSA-368v-7v32-52fx, GHSA-49rq-hwc3-x77w;tqdm: GHSA-g7vv-2v7x-gj9p, PYSEC-2026-1976","source":"check","title":"已知漏洞"},{"code":"no_longer_updated","severity":"info","evidence":"最后一次提交是在 2022-10-28,也没有 GitHub 发布版本。固定的版本号停留在那个时期(tensorflow==2.10.0, numpy==1.20.3, gradio==3.3.1),安装说明要求 Python 3.9 配合 CUDA 11.2。README 还许诺了一项始终没有兑现的工作:\"A training script for custom encoders/decoders will be provided!\"","source":"ai","title":"已不再更新"},{"code":"training_data_not_disclosed","severity":"info","evidence":"README 没有说明自带模型学习自哪些音乐,只提到 misc 检查点 \"a diverse music dataset was used for training\"。仓库里没有任何曲目列表、许可信息或来源说明。","source":"ai","title":"训练数据未公开"},{"code":"trains_on_music_you_provide","severity":"info","evidence":"项目邀请你用别人的录音来训练它:\"You can train a musika system using your own custom dataset\",Colab 笔记本的介绍是 \"to train Musika on custom music, no technical skills needed!\"。使用不属于你的录音进行训练,并发布结果,取决于你是否拥有这些音乐的权利。","source":"ai","title":"会用你提供的音乐进行训练"}],"platforms":["windows","linux","web"],"tools":["python","gpu_any","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":600,"maxMB":1800,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"epidemic-sound","name":"Epidemic Sound","confirmed":true}],"declaredAlternatives":[],"stars":685,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":684},{"date":"2026-11-08","stars":684},{"date":"2026-11-09","stars":684},{"date":"2026-11-10","stars":685}],"firstPublishedAt":"2026-11-07T23:02:21.790Z","commitSha":"b1e6e79dd99b041935a142c19f2c785efff6f461","verifiedAt":"2026-11-07T22:47:31.752Z","lastCommitAt":"2022-10-28T13:41:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research tool that generates endless instrumental music on your computer, with a simple page in the browser.","summary":"Musika generates instrumental music by itself, as long as you want, and fast enough to run on a home graphics card. Two ready models come with it, techno and a mixed-genre one, so you can make tracks without training anything: one command opens a small page in your browser where you press generate. You can also train it on your own music collection. It is research code from a 2022 master's thesis: setup means conda, ffmpeg and CUDA, and there is no text prompt, you pick a model and a length.","easeRationale":"On the web the README links a ready demo page on Hugging Face Spaces, where you only press a button, plus a Colab notebook for training \"no technical skills needed\". Running it yourself is a D on both Windows and Linux: install conda and ffmpeg, create a Python 3.9 environment, install CUDA 11.2 and CuDNN for an NVIDIA card, clone the repository, install the requirements and then run a command that opens the interface; Linux needs one extra step to set a library path.","costNote":"Free and open source (MIT), models included. Training on your own music costs only the GPU time, on your card or in a rented notebook.","notVerified":["Whether the Hugging Face demo page and the Colab notebook are still online and working","Whether the 2022 versions of TensorFlow, CUDA and Gradio still install on a current system","How good the music sounds for your genre, and whether it is usable in a video or a release","How long training or finetuning takes on your own graphics card","Real disk use after the environment, the checkpoints and your encoded audio are on disk"]},"zh-Hans":{"summaryShort":"研究性工具,在你的电脑上生成无限长的纯音乐,浏览器里有个简单页面就能用。","summary":"Musika 可以自己生成纯音乐,想要多长就有多长,速度快到足以在家用显卡上运行。它自带两个现成模型,一个 techno 风格,一个混合风格,所以你不用训练任何东西就能制作曲目:一条命令就会在浏览器里打开一个小页面,点一下生成即可。你也可以用自己的音乐收藏来训练它。这是来自 2022 年一篇硕士论文的研究代码:安装需要 conda、ffmpeg 和 CUDA,而且没有文字提示词,你只需要选择一个模型和一个时长。","easeRationale":"在网页上,README 链接了 Hugging Face Spaces 上一个现成的演示页面,你只需点一个按钮,此外还有一个 Colab 笔记本用于训练,称 \"no technical skills needed\"。自己运行在 Windows 和 Linux 上都是 D 级:安装 conda 和 ffmpeg,创建 Python 3.9 环境,为 NVIDIA 显卡安装 CUDA 11.2 和 CuDNN,克隆仓库,安装依赖项,然后运行一条打开界面的命令;Linux 还需要多一步来设置库路径。","costNote":"免费且开源(MIT),包含模型。用自己的音乐训练只需要花 GPU 时间,无论是你自己的显卡还是租来的笔记本环境。","notVerified":["Hugging Face 的演示页面和 Colab 笔记本是否仍在线并可用","2022 年版本的 TensorFlow、CUDA 和 Gradio 是否还能安装在现在的系统上","生成的音乐对你的风格来说听起来怎么样,能否用于视频或正式作品","在你自己的显卡上训练或微调需要多长时间","环境、检查点和你编码后的音频都存盘之后,真实占用的磁盘空间是多少"]}}},{"id":"R_kgDOLc1V8Q","slug":"materialious__materialious","name":"Materialious","owner":"Materialious","fullName":"Materialious/Materialious","type":"ready_app","status":"alert","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A","web":"A"},"risk":"high","signals":[{"code":"third_party_youtube_client_may_break_platform_terms","severity":"attention","evidence":"README 把这个应用描述为 \"frontend for YouTube & Invidious\",具有 \"No ads\"、\"SponsorBlock built-in\" 和 \"Enjoy Materialious without an Invidious instance via our YouTube backend\" 等功能。在官方播放器之外播放 YouTube 视频并跳过广告,违反了 YouTube 的服务条款,这类客户端有时会被封锁。","source":"ai","title":"第三方 YouTube 客户端可能违反平台条款"},{"code":"playback_depends_on_third_party_servers","severity":"info","evidence":"README 说 Invidious 是可选的,还有一个 \"Local video fallback if Invidious fails loading videos\",所以视频播放依赖项目自己的后端,或者依赖可能会失效的公共 Invidious 实例。","source":"ai","title":"播放依赖第三方服务器"},{"code":"very_frequent_releases","severity":"info","evidence":"GitHub 上有 286 个发布版本,其中 1.18.17、1.18.18 和 1.18.19 三个版本都是在 2026-11-06 发布的。","source":"ai","title":"发布非常频繁"},{"code":"plays_youtube_videos_outside_the_official_player","severity":"high","evidence":"项目简介:\"Materialious is a modern material design frontend for YouTube & Invidious, focused on a clean, privacy-friendly YouTube experience. It supports local video fallback when Invidious fails\";README 中的功能:\"No ads\"、\"Enjoy Materialious without an Invidious instance via our YouTube backend\"、\"Local video fallback if Invidious fails loading videos\",致谢名单里提到 \"LuanRT for YouTube.js & SABR implementation\"。YouTube 的条款要求通过其自身服务和界面观看视频,并禁止关闭广告,许多国家的法律也限制规避控制内容访问的技术措施。","source":"ai","title":"在官方播放器之外播放 YouTube 视频"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":[],"disk":{"minMB":295,"maxMB":885,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["video"],"alternativeTo":[{"slug":"youtube-premium","name":"YouTube Premium","confirmed":true}],"declaredAlternatives":[],"stars":1243,"downloads":333339,"downloadsWeek":67652,"downloadsDays":5,"growth":4,"growthDays":6,"history":[{"date":"2026-11-05","stars":1239},{"date":"2026-11-06","stars":1240},{"date":"2026-11-07","stars":1240},{"date":"2026-11-08","stars":1240},{"date":"2026-11-09","stars":1241},{"date":"2026-11-10","stars":1243}],"firstPublishedAt":"2026-11-07T23:02:21.812Z","commitSha":"be9ea646bea9f62a28a19b836e2951b5ed77eb15","verifiedAt":"2026-11-07T22:37:23.912Z","lastCommitAt":"2026-11-10T14:30:53Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Watch YouTube in a clean app with no ads, no tracking and skipping of sponsored segments.","summary":"Materialious is a different front door for YouTube. It shows videos, channels, playlists and subscriptions in a Material design interface, with no ads and no tracking, and it can skip sponsored parts of videos (SponsorBlock), show dislike counts and change thumbnails and titles (DeArrow). It can use its own YouTube backend or an Invidious server, and there is an optional account to sync subscriptions and watch progress. Ready installers exist for Windows, macOS and Linux, plus Android, Android TV and a website version you can also self-host with Docker.","easeRationale":"On Windows the release has a ready .exe installer (151 MB), and macOS has a .dmg and Linux has AppImage, .deb and .rpm files, so it is download and open on all three. The web version is already hosted at materialio.us, so it just opens in a browser; only the optional self-hosted version needs Docker.","notVerified":["Whether video playback actually works on each platform right now, since YouTube often changes and can block third-party clients","Whether the optional Materialious account service and the hosted site at materialio.us keep any data about what you watch","Real disk use after installing (installers are 117-295 MB depending on the platform)","Whether the Android and Android TV builds behave like the desktop ones"]},"zh-Hans":{"summaryShort":"在一个干净的应用里看 YouTube,没有广告、没有追踪,还能跳过赞助片段。","summary":"Materialious 是进入 YouTube 的另一个入口。它用 Material 设计风格展示视频、频道、播放列表和订阅内容,没有广告也没有追踪,还能跳过视频里的赞助片段(SponsorBlock)、显示点踩数,以及替换缩略图和标题(DeArrow)。它可以使用自带的 YouTube 后端,也可以使用 Invidious 服务器,还有一个可选账号用来同步订阅和观看进度。Windows、macOS 和 Linux 都有现成的安装包,另外还有 Android、Android TV 版本,以及一个也可以用 Docker 自行部署的网页版本。","easeRationale":"在 Windows 上,发布版自带现成的 .exe 安装程序(151 MB),macOS 有 .dmg 文件,Linux 有 AppImage、.deb 和 .rpm 文件,三个平台都只需下载后打开即可。网页版已经托管在 materialio.us,直接在浏览器里打开就行;只有可选的自托管版本需要 Docker。","notVerified":["视频播放现在是否在每个平台上都真的能用,因为 YouTube 经常变动,可能封锁第三方客户端","可选的 Materialious 账号服务以及托管在 materialio.us 的站点是否会保留你观看内容的数据","安装之后真实占用的磁盘空间(安装包根据平台不同在 117 到 295 MB 之间)","Android 和 Android TV 版本的表现是否和桌面版一致"]}}},{"id":"R_kgDOUYU5tA","slug":"matthewdonsemail-lab__log","name":"ListeningKit Log","owner":"matthewdonsemail-lab","fullName":"matthewdonsemail-lab/log","type":"dev_tool","status":"alert","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"high","signals":[{"code":"solves_captchas_and_hides_account_location","severity":"high","evidence":".env.example在常规配置中就设置好了验证码破解服务（`CAPTCHA_KRAKEN_API_KEY`、`VLLM_BASE_URL=https://api.captchakraken.com/v1`、`CAPTCHA_HUMANIZATION=mouse # mouse | mobile | none`），还有\"One key for both: the captcha fallback (reCAPTCHA token, Arkose) and the residential proxy API\"（`2CAPTCHA_API_KEY`、`2CAPTCHA_PROXY_PASSWORD`）、通过Byteful提供的住宅和静态代理，以及用于\"accounts that hold an AdsPower profile\"的`LOG_ADSPOWER_HOMES`。README说Facebook账号\"opened once a day through a proxy in its home country\"，并且在\"Kept human\"这个标题下，每个账号都有\"daily limits, pacing and local waking hours\"。验证码破解服务用来回答平台借以区分真人和程序的验证问题，住宅代理能让连接看起来来自所选国家的一个家庭地址。X、Meta、Slack和Discord的服务条款都禁止以这种方式让账号自动化运行，并会因此封禁账号。","source":"ai","title":"解决验证码并隐藏账号所在位置"},{"code":"uses_pasted_login_cookies_and_unofficial_clients","severity":"attention","evidence":"README的Connect your accounts部分：你把登录会话cookie粘贴进log（`log twitter cookies save`、`log linkedin cookies save`、`log slack cookies save`），Platforms表格说每个平台都是通过\"its own API client\"以及\"live gateway\"/\"socket\"/\"realtime connection\"、\"no browser\"来访问的——也就是说，是私有接口，不是官方发布的API。这些cookie能完全控制账号，并保存在你机器上的`~/.logCli`里；README自己也提到\"One account was disabled after an agent read a server too much, too fast (rule 94)\"。","source":"ai","title":"使用手动粘贴的登录cookie和非官方客户端"},{"code":"checks_drafts_for_ai_phrasing_and_sends_cloned_voice_notes","severity":"attention","evidence":"README写道：\"Your voice, checked. A draft is read against how you really write to that person, for AI phrasing and for hidden characters, before it can be sent.\" 语音消息会用克隆的声音发出：.env.example说`FISH_API_KEY`是用于\"matt's cloned voice for the voice notes he asks for\"，`INWORLD_API_KEY`是\"the default voice provider for matt's cloned voice\"，还有口音设置（\"default Irish\"）、语言地区和语速。","source":"ai","title":"检查草稿是否有ai腔调，并发送克隆语音消息"},{"code":"install_pipes_script_from_a_website_to_shell","severity":"attention","evidence":"README的Quick start部分：`curl -fsSL https://log.listeningkit.com/install.sh | sh`和`irm https://log.listeningkit.com/install.ps1 | iex`。这个脚本来自作者自己的域名，不是来自仓库本身，它会\"installs the `log` and `logMcp` commands, and Python with everything the platform engines need\"，把配置写入`~/.logCli/.env`。","source":"ai","title":"安装过程把网站上的脚本直接接到shell"},{"code":"defaults_point_at_the_authors_private_server","severity":"info","evidence":".env.example默认值指向作者自己的机器：`LOG_TWENTY_BASE_URL=https://node02.tail4fcf85.ts.net/rest`，注释写着\"log-cli's Twenty is ListeningKit's on node02\"。文件里还提到了编号的内部规则（\"rule 78\"、\"rule 145\"），README里并没有解释这些规则，还有几项设置明显是围绕某一个人的账号写的。","source":"ai","title":"默认配置指向作者自己的私有服务器"},{"code":"very_small_young_project_no_releases","severity":"info","evidence":"创建于2026-09-12，36个星标，7个分支，25个未关闭的issue，GitHub上没有任何发布版本；仓库描述说它是\"Built for the Convex Hackathon\"。npm包版本是0.1.4，README说托管版本\"is planned\"。","source":"ai","title":"非常小且年轻的项目，没有发布版本"}],"platforms":["mac","linux","windows"],"tools":["node","python"],"buildTools":["node"],"disk":{"minMB":150,"maxMB":650,"basis":"heuristic"},"apiKeys":["Twenty CRM (self-hosted account and listing store)","CaptchaKraken (captcha solver)","2Captcha (captcha fallback and residential proxies)","Byteful (residential and static proxies)","AssemblyAI (voice note transcription)","Fish Audio (cloned voice)","Inworld (cloned voice, default provider)","Memelord"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","ai_agents","ai_skills"],"alternativeTo":[{"slug":"phantombuster","name":"PhantomBuster","confirmed":true},{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true}],"declaredAlternatives":[],"stars":37,"downloads":0,"downloadsWeek":0,"downloadsDays":1,"growth":1,"growthDays":1,"history":[{"date":"2026-11-09","stars":36},{"date":"2026-11-10","stars":37}],"firstPublishedAt":"2026-11-08T23:13:56.553Z","commitSha":"6527bf444f4feac0e0e280106c530ad39cf81323","verifiedAt":"2026-11-08T23:06:55.069Z","lastCommitAt":"2026-11-10T13:38:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs your X, Discord, Slack, Instagram and Facebook accounts from the terminal, or lets an AI agent run them.","summary":"log is a command-line tool plus an MCP server that works your own social accounts. Messages are only on four: on X, Discord, Slack and Instagram it reads DMs, group chats, replies and notifications and sends replies, reactions and voice notes. On Facebook the README's table lists Marketplace listings, groups and \"warm-up\"; for Reddit and LinkedIn it lists only \"session check (new)\". Everything lands in one local database that your AI assistant (Claude Code, Codex, Cursor) reads. Commands that change something are a dry run until you add `--live`.","easeRationale":"On Mac and Linux you install by piping a script from the project's website into your shell, then paste login cookies for each platform (`auth_token`, `ct0` and `twid` for X, `li_at` for LinkedIn, the `d` cookie for Slack) taken with a browser extension, register the MCP server in your agent and fill in keys for the outside services. Windows uses the same route through PowerShell. There is no graphical app and no release download.","costNote":"The code is free (MIT), but the features as shipped lean on paid accounts: a captcha solving service, residential proxies (2Captcha or Byteful), AssemblyAI for transcription and Inworld or Fish Audio for voice notes.","notVerified":["Whether the platforms ban the accounts you connect, and how quickly","Whether the private API clients still work, since the platforms change them without notice","What the captcha, proxy, transcription and voice services really cost per month","Whether the install script on the author's website matches the code in this repository","What Reddit, LinkedIn and Facebook can really do beyond the session check and the listings the README's table lists","Whether downloaded pictures and voice notes are really deleted after being turned into text, as the diagram claims"]},"zh-Hans":{"summaryShort":"通过终端操作你的X、Discord、Slack、Instagram和Facebook账号，或者让AI代理来操作。","summary":"log是一个命令行工具加一个MCP服务器，用来操作你自己的社交账号。消息功能只存在于四个平台上：在X、Discord、Slack和Instagram上，它能读取私信、群聊、回复和通知，并发送回复、表情反应和语音消息。在Facebook上，README的表格列出了Marketplace商品、群组和\"warm-up\"；对于Reddit和LinkedIn，只列出了\"session check (new)\"。所有数据都汇总到一个本地数据库里，供你的AI助手（Claude Code、Codex、Cursor）读取。会改变数据的命令默认只是演练（dry run），直到你加上`--live`才会真正执行。","easeRationale":"在Mac和Linux上，安装方式是把项目网站上的一个脚本直接接到你的shell里运行，然后为每个平台粘贴登录cookie（X用`auth_token`、`ct0`和`twid`，LinkedIn用`li_at`，Slack用`d`这个cookie），这些cookie是用浏览器扩展获取的，再把MCP服务器注册到你的代理里，并填好外部服务的密钥。Windows走的是同样的路线，只是通过PowerShell。没有图形界面应用，也没有发布版本可以下载。","costNote":"代码本身是免费的（MIT许可），但出厂自带的功能要依赖付费账号：一个验证码破解服务、住宅代理（2Captcha或Byteful）、用于转写的AssemblyAI，以及用于语音消息的Inworld或Fish Audio。","notVerified":["各平台是否会封禁你连接的账号，速度有多快","私有API客户端是否仍然能用，因为各平台会随时更改它们，不提前通知","验证码、代理、转写和语音服务每月实际花费多少","作者网站上的安装脚本是否与这个仓库里的代码一致","Reddit、LinkedIn和Facebook除了会话检查和README表格列出的功能之外，实际还能做什么","下载的图片和语音消息在转成文字后，是否真的像图示所说的那样被删除"]}}},{"id":"R_kgDORHkddg","slug":"mattpocock__skills","name":"Skills for Real Engineers","owner":"mattpocock","fullName":"mattpocock/skills","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"268 天内获得 306638 个星标（约每天 1144 个）","source":"check","title":"星标增长速度异常"},{"code":"agent_writes_and_commits_code_on_its_own","severity":"attention","evidence":"参考文档说 /implement \"drives /tdd at pre-agreed seams and closing out with /code-review before committing\"，而 /implement-spec \"implements a whole spec on one integration branch... running implementer subagents across the ready frontier for maximum concurrency\"。因此，一次运行中，多个智能体会在你的仓库里修改文件并提交 commit。","source":"ai","title":"智能体自行编写并提交代码"},{"code":"writes_into_your_issue_tracker_and_docs","severity":"info","evidence":"设置类 skill 会询问 \"which issue tracker you want to use (GitHub, GitLab, local files, or anything else you describe)\" 以及你使用哪些标签；随后 /to-spec \"publishes it to the issue tracker\"，/to-tickets 则把工单写成 \"as native blocking links on a real tracker\"。/grill-with-docs 和 /domain-modeling 等 skill 还会编辑你仓库中的 GLOSSARY.md 和架构决策记录。","source":"ai","title":"会写入你的问题跟踪器和文档"},{"code":"generates_shell_scripts_you_then_run","severity":"info","evidence":"/wizard 这个 skill \"generates an interactive bash wizard that walks a human through steps only they can perform: provisioning infrastructure, setting up credentials or CI secrets... or running a one-off migration or cutover\"。运行这个生成的脚本的人是你，所以运行前最好先读一遍。","source":"ai","title":"生成需要你自己运行的 shell 脚本"},{"code":"plugin_updates_itself","severity":"info","evidence":"README Installation：Claude Code 插件 \"Updates itself by default\"，Codex \"Updates itself at startup\"，而 Copilot 市场条目配置了 \"autoUpdate\": true。新的指令可能在你没有审阅的情况下到达你的智能体；通过 skills.sh 的方式则是复制可编辑的文件，由你手动更新。","source":"ai","title":"插件会自动更新"}],"platforms":["windows","mac","linux"],"tools":["other","node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":320785,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":9359,"growthDays":7,"history":[{"date":"2026-10-25","stars":303297},{"date":"2026-10-26","stars":304077},{"date":"2026-10-27","stars":304891},{"date":"2026-10-28","stars":305776},{"date":"2026-10-29","stars":306624},{"date":"2026-10-30","stars":307488},{"date":"2026-10-31","stars":308394},{"date":"2026-11-01","stars":309301},{"date":"2026-11-02","stars":310231},{"date":"2026-11-03","stars":311426},{"date":"2026-11-04","stars":313098},{"date":"2026-11-05","stars":315010},{"date":"2026-11-06","stars":316580},{"date":"2026-11-07","stars":317572},{"date":"2026-11-08","stars":318423},{"date":"2026-11-09","stars":319610},{"date":"2026-11-10","stars":320785}],"firstPublishedAt":"2026-10-25T23:19:10.800Z","commitSha":"a496f5513d4a4cf889bc48e189e579f4b0bdf6aa","verifiedAt":"2026-10-30T01:48:57.437Z","lastCommitAt":"2026-10-31T14:23:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds commands to an AI coding agent that interview you first, then spec, test, review and ship the work.","summary":"Matt Pocock's personal set of agent skills: about 30 short instruction files that add commands to an AI coding agent. The headline ones interview you about what you actually want (/grill-me, /grill-with-docs) before any code, then turn the conversation into a spec and tickets, build it with test-first loops, review the diff and write the pull request. Others cover debugging, codebase design and handing a session over. It is for programmers who already use Claude Code, Codex, Copilot, Gemini CLI, Cursor or a similar agent. Installing is one or two commands inside the agent.","easeRationale":"You need a supported coding agent first, then one or two commands, for example `claude plugin install mattpocock-skills@claude-plugins-official`, or `npx skills@latest add mattpocock/skills -a <agent>` for other agents, which needs Node.js. After that you run `/setup-matt-pocock-skills` once per project and answer questions about your issue tracker, labels and where documents go. The README warns not to install both the plugin and the skills.sh copies, or you get every skill twice.","costNote":"The skills are free (MIT). They only do something inside an AI coding agent, which usually needs a paid plan or paid API use. The author also runs a newsletter and a site at aihero.dev.","notVerified":["Whether the skills actually improve the result on a real project","How much extra AI usage, and therefore cost, the long interview and subagent flows consume","Which skills each agent other than Claude Code really picks up, and how they behave on Windows","What the automatic updates bring in later, since Claude Code, Codex and Copilot refresh the plugin by themselves","Whether the automatic code review and ticket writing match your team's process"]},"zh-Hans":{"summaryShort":"为AI编程助手添加一些命令,它们先采访你,然后写需求、测试、审查并交付工作成果。","summary":"Matt Pocock个人的一套代理技能:大约30个简短的指令文件,为AI编程助手添加命令。最主要的几个会先就你到底想要什么采访你(/grill-me、/grill-with-docs),然后才开始写代码,把对话变成需求说明和工单,用先写测试的方式来构建,审查改动并撰写拉取请求。其他的涉及调试、代码库设计以及把会话交接给别人。它适合已经在用Claude Code、Codex、Copilot、Gemini CLI、Cursor或类似助手的程序员。安装只需在助手里输入一两个命令。","easeRationale":"你需要先有一个受支持的编程助手,然后输入一两个命令,例如`claude plugin install mattpocock-skills@claude-plugins-official`,或者对其他助手使用`npx skills@latest add mattpocock/skills -a <agent>`,这需要Node.js。之后每个项目运行一次`/setup-matt-pocock-skills`,回答关于你的问题跟踪系统、标签和文档存放位置的问题。README提醒不要同时安装插件和skills.sh的副本,否则每个技能都会重复一份。","costNote":"这些技能是免费的(MIT协议)。它们只能在AI编程助手内部起作用,而这类助手通常需要付费方案或付费的API使用。作者还运营着一份订阅通讯和一个网站aihero.dev。","notVerified":["这些技能是否真的能在实际项目中改善结果","较长的采访流程和子代理流程会额外消耗多少AI用量,从而增加多少成本","除Claude Code之外的每个助手实际上能识别哪些技能,它们在Windows上的表现如何","由于Claude Code、Codex和Copilot会自动更新插件,以后的自动更新会带来什么变化","自动代码审查和工单撰写是否符合你团队的工作流程"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzOTc0ODcxMw==","slug":"mattvitelli__gruv","name":"GRUV","owner":"MattVitelli","fullName":"MattVitelli/GRUV","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹以及 winget 清单均未检测到许可证","source":"check","title":"没有许可证"},{"code":"obsolete_pinned_dependencies","severity":"attention","evidence":"README.md:\"Note: This code works with Keras v. 0.1.0, later versions of Keras may not work.\" 它还需要 Theano,这是一个已经停止维护的库。如今安装这些旧版本很可能会失败。","source":"ai","title":"依赖版本固定且已过时"},{"code":"unmaintained_since_2020","severity":"attention","evidence":"最后一次提交是在 2020-11-28(距检查之日 2170 天),没有发布版本,还有 27 个未关闭的议题。","source":"ai","title":"自 2020 年起无人维护"},{"code":"output_may_copy_training_songs","severity":"info","evidence":"README.md:\"Our current generation scheme uses the training / testing data as a seed sequence, which tends to produce verbatum copies of the original songs.\" 它生成的内容可能会重复你喂给它的音乐,如果那些音乐不属于你,这一点就很重要。","source":"ai","title":"生成结果可能复制训练用的歌曲"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true}],"declaredAlternatives":[],"stars":797,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":796},{"date":"2026-11-08","stars":796},{"date":"2026-11-09","stars":796},{"date":"2026-11-10","stars":797}],"firstPublishedAt":"2026-11-07T23:02:21.825Z","commitSha":"71998b14fda01d58a6ad98a82b4f65455a13a18a","verifiedAt":"2026-11-07T22:37:59.628Z","lastCommitAt":"2020-11-28T18:30:26Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research scripts that train a neural network on your own songs and then generate a new audio file.","summary":"GRUV is a small set of Python scripts from a 2015 university project. You copy your own music into a folder, run one script to convert it, another to train a neural network on the raw audio, and a third to produce a generated_song.wav file. There is no interface and no installer: you install Python plus Theano, Keras, NumPy, SciPy, h5py and the LAME and SoX audio converters, and run commands in a terminal. The README says the result often sounds like near-copies of the songs used for training. The code has not changed since 2020.","easeRationale":"On Linux, the easiest case, you still install Python and seven separate libraries and command-line tools (Theano, Keras, NumPy, SciPy, h5py, LAME, SoX) and run three commands in a terminal; training is slow and the README talks about graphics-card memory errors. Windows and macOS are the same work, with the extra difficulty that LAME, SoX and Theano are harder to set up there.","notVerified":["Whether the old Theano and Keras 0.1.0 versions can still be installed and run at all today","How long training takes and how much memory or graphics-card memory is needed for a given music library","How much disk space the converted .npy data files take for your songs","The real musical quality of the generated audio"]},"zh-Hans":{"summaryShort":"研究性脚本,用你自己的歌曲训练一个神经网络,然后生成一个新的音频文件。","summary":"GRUV 是 2015 年一个大学项目留下的一小组 Python 脚本。你把自己的音乐复制到一个文件夹里,运行一个脚本来转换它,再运行另一个脚本用原始音频训练神经网络,然后用第三个脚本生成一个 generated_song.wav 文件。没有界面也没有安装程序:你需要安装 Python,以及 Theano、Keras、NumPy、SciPy、h5py,还有 LAME 和 SoX 这两个音频转换工具,并在终端里运行命令。README 说生成的结果经常听起来像是训练用歌曲的近似拷贝。代码自 2020 年以来就没有变过。","easeRationale":"即便是最容易的 Linux,你仍要安装 Python 以及七个独立的库和命令行工具(Theano、Keras、NumPy、SciPy、h5py、LAME、SoX),并在终端里运行三条命令;训练速度很慢,README 还提到显卡显存不足的错误。Windows 和 macOS 的工作量相同,而且 LAME、SoX 和 Theano 在这些系统上更难安装。","notVerified":["老版本的 Theano 和 Keras 0.1.0 今天是否还能安装并运行","对于给定的音乐库,训练需要多长时间,需要多少内存或显存","转换后的 .npy 数据文件会占用多少磁盘空间","生成音频的真实音乐质量如何"]}}},{"id":"R_kgDOPdnatQ","slug":"meigen-ai__infinitetalk","name":"InfiniteTalk","owner":"MeiGen-AI","fullName":"MeiGen-AI/InfiniteTalk","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/drive/folders/1BNrH6GJZ2Wt5gBuNLmfXZ6kpqb9xFPjU?usp=sharing","source":"check","title":"外部下载链接"},{"code":"realistic_talking_video_of_real_people","severity":"attention","evidence":"README.md 描述了一种 \"sparse-frame video dubbing\",能从一张图片或一段视频加任意音频合成出具有 \"accurate lip synchronization\" 和 \"consistent identity preservation\" 的视频。同一份 README 写道:\"You are fully accountable for your use of the models, which must not involve sharing any content that violates applicable laws, causes harm to individuals or groups\"。它可以让真实的人看起来说了他们从未说过的话。","source":"ai","title":"真实人物的逼真说话视频"},{"code":"large_model_downloads_not_in_repo","severity":"info","evidence":"README 要求用 huggingface-cli 把 Wan2.1-I2V-14B-480P、chinese-wav2vec2-base 和 MeiGen-InfiniteTalk 下载到 ./weights 目录。这些权重文件有好几个 GB,并未计入仓库本身的大小。","source":"ai","title":"仓库之外的大型模型下载"},{"code":"cloud_api_sdk_in_requirements","severity":"info","evidence":"requirements.txt 中包含 \"dashscope\",这是阿里云 AI 服务的客户端库。即便文档中的命令都是本地运行,它仍然会被安装。","source":"ai","title":"依赖中包含云端 API 的 SDK"},{"code":"authors_point_to_newer_model","severity":"info","evidence":"README.md 的新闻部分(2026 年 5 月 21 日)宣布 LongCat-Video-Avatar-1.5 是 \"an upgraded open-source framework for audio-driven human video generation\",而这里最后一次提交是在 2026-05-22。","source":"ai","title":"作者指向了更新的模型"}],"platforms":["linux"],"tools":["python","ffmpeg","gpu_nvidia"],"buildTools":["python","ffmpeg"],"disk":{"minMB":210,"maxMB":570,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":8371,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":59,"growthDays":6,"history":[{"date":"2026-11-05","stars":8325},{"date":"2026-11-06","stars":8330},{"date":"2026-11-07","stars":8340},{"date":"2026-11-08","stars":8355},{"date":"2026-11-09","stars":8363},{"date":"2026-11-10","stars":8371}],"firstPublishedAt":"2026-11-07T23:02:21.833Z","commitSha":"50aa0a94184315407a991ae804d9b58d6d311ba8","verifiedAt":"2026-11-07T22:38:42.823Z","lastCommitAt":"2026-05-22T02:35:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a photo or a video plus an audio track into a long video of that person talking in sync.","summary":"InfiniteTalk is research code that animates a person from an audio track. Give it one photo or an existing video plus a voice recording and it produces a video where lips, head, body and expressions follow the audio, with no fixed length limit (480p or 720p). There is a command line and a simple web page (Gradio) you start yourself. To use it you set up a Python environment on Linux, install PyTorch for NVIDIA CUDA, FFmpeg and other libraries, and download several large AI models from Hugging Face. An NVIDIA graphics card is required.","easeRationale":"Linux is the only documented platform and still needs a conda environment, PyTorch built for CUDA 12.1, flash-attn compiled with ninja, FFmpeg, and three separate model downloads before the first command runs. All the examples assume an NVIDIA graphics card, and the README offers low-memory and 8-GPU variants of the same command.","notVerified":["Total disk space needed once the Wan2.1 14B and other model weights are downloaded","How much graphics-card memory is really needed, and how long one minute of video takes","Whether the install steps (flash-attn, CUDA 12.1 wheels) still work with current versions","Whether it runs on Windows or macOS at all, since neither is documented","The real quality of lip sync and the colour shift the README mentions after one minute"]},"zh-Hans":{"summaryShort":"把一张照片或一段视频加上一段音轨,变成那个人对口型说话的长视频。","summary":"InfiniteTalk 是一段研究性代码,可以根据一段音轨让人物动起来。给它一张照片或一段已有视频,再加一段语音录音,它就能生成一个嘴唇、头部、身体和表情都跟随音频变化的视频,没有固定时长限制(480p 或 720p)。它提供命令行和一个需要你自己启动的简单网页(Gradio)。使用时你需要在 Linux 上搭建 Python 环境,为 NVIDIA CUDA 安装 PyTorch,安装 FFmpeg 和其他库,并从 Hugging Face 下载几个体积很大的 AI 模型。需要一块 NVIDIA 显卡。","easeRationale":"Linux 是唯一有文档说明的平台,即便如此也需要 conda 环境、为 CUDA 12.1 编译的 PyTorch、用 ninja 编译的 flash-attn、FFmpeg,以及运行第一条命令之前的三次独立模型下载。所有示例都假定使用 NVIDIA 显卡,README 还提供了同一条命令的低显存版本和 8 卡版本。","notVerified":["下载完 Wan2.1 14B 和其他模型权重之后,总共需要多少磁盘空间","实际需要多少显存,生成一分钟视频需要多长时间","安装步骤(flash-attn、CUDA 12.1 的 wheel 包)是否还能在当前版本下正常工作","它是否能在 Windows 或 macOS 上运行,因为这两个系统都没有文档说明","口型同步的真实质量,以及 README 提到的一分钟后出现的色彩偏移问题"]}}},{"id":"R_kgDOOyDDLw","slug":"meigen-ai__multitalk","name":"MultiTalk","owner":"MeiGen-AI","fullName":"MeiGen-AI/MultiTalk","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"realistic_talking_video_of_real_people","severity":"attention","evidence":"README.md 承诺由 \"a multi-stream audio input, a reference image and a prompt\" 生成 \"Realistic Conversations\"，并且 \"consistent lip motions aligned with the audio\"。许可证部分还补充说，你的使用 \"must not involve sharing any content that violates applicable laws, causes harm to individuals or groups, disseminates personal information intended for harm, spreads misinformation\"。它可以让真实存在的人看起来说了他们从未说过的话。","source":"ai","title":"逼真的真人说话视频"},{"code":"large_model_downloads_not_in_repo","severity":"info","evidence":"README 要求你用 huggingface-cli 把 Wan2.1-I2V-14B-480P、chinese-wav2vec2-base、Kokoro-82M 和 MeiGen-MultiTalk 下载到 ./weights 目录。这些文件有好几个 GB 大小，并不包含在仓库里。","source":"ai","title":"需要下载不在仓库里的大型模型"},{"code":"install_steps_use_sudo_symlinks","severity":"info","evidence":"模型准备步骤要求你运行 \"sudo ln -s {Absolute path}/weights/MeiGen-MultiTalk/multitalk.safetensors weights/Wan2.1-I2V-14B-480P/\"，并重命名下载的基础模型里的一个文件，这意味着安装过程会以管理员权限修改文件。","source":"ai","title":"安装步骤需要用 sudo 创建符号链接"},{"code":"cloud_api_sdk_in_requirements","severity":"info","evidence":"requirements.txt 中包含 \"dashscope\"，这是阿里云 AI 服务的客户端库，尽管文档里给出的命令都是在本地运行的。","source":"ai","title":"依赖项中包含云端 API 的 SDK"},{"code":"authors_point_to_newer_models","severity":"info","evidence":"README.md 的更新说明里提到 InfiniteTalk（2025 年 8 月）和 LongCat-Video-Avatar-1.5（2026 年 5 月 21 日）是同一团队推出的更新、\"upgraded\" 框架；这个仓库最后一次提交是在 2026-05-22，720p 和 ComfyUI 在 Todo 列表里仍未完成。","source":"ai","title":"作者指向了更新的模型"}],"platforms":["linux"],"tools":["python","ffmpeg","gpu_nvidia"],"buildTools":["python","ffmpeg"],"disk":{"minMB":210,"maxMB":570,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":3033,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":4,"growthDays":6,"history":[{"date":"2026-11-06","stars":3029},{"date":"2026-11-07","stars":3030},{"date":"2026-11-08","stars":3030},{"date":"2026-11-09","stars":3032},{"date":"2026-11-10","stars":3033}],"firstPublishedAt":"2026-11-07T23:02:21.843Z","commitSha":"c1ad84009b99ed36c97ab18e4517ca5f98692438","verifiedAt":"2026-11-07T22:39:08.788Z","lastCommitAt":"2026-05-22T02:38:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes a video of one or two people talking or singing, driven by audio and a reference photo.","summary":"MultiTalk is research code (NeurIPS 2025) that turns a reference image, one or more voice tracks and a written prompt into a 480p video of people holding a conversation, singing or interacting, with lips matching the audio. It can also read text aloud itself with a built-in voice model (Kokoro). Clips are short, around 15 seconds. You run it from the terminal or from a simple web page you start yourself, after setting up Python on Linux, PyTorch for NVIDIA CUDA, FFmpeg and four separate model downloads. An NVIDIA graphics card is needed.","easeRationale":"Linux is the only documented platform: a conda environment, PyTorch and xformers built for CUDA 12.1, flash-attn compiled with ninja, FFmpeg, four model downloads from Hugging Face, and then moving and symlinking files inside the weights folder before the first command. Every example assumes an NVIDIA graphics card.","notVerified":["Total disk space after downloading the Wan2.1 14B base model and the other three model sets","How much graphics-card memory is really needed and how long a 15-second clip takes","Whether the pinned install steps (CUDA 12.1 wheels, flash-attn 2.7.4) still work today","Whether it can run on Windows or macOS, which are not documented","The real quality and the colour drift the README mentions in longer videos"]},"zh-Hans":{"summaryShort":"根据音频和一张参考照片，生成一段一人或两人说话或唱歌的视频。","summary":"MultiTalk 是一套研究代码（NeurIPS 2025），能把一张参考图片、一段或多段语音轨道和一段文字提示，转换成一段 480p 的视频，画面里的人物在交谈、唱歌或互动，嘴型与音频同步。它还能用内置的语音模型（Kokoro）自己朗读文字。片段很短，大约 15 秒。你需要先在 Linux 上配置好 Python、适配 NVIDIA CUDA 的 PyTorch、FFmpeg，并下载四个独立的模型，然后通过终端运行，或启动一个你自己搭建的简单网页。需要一块 NVIDIA 显卡。","easeRationale":"Linux 是唯一有文档说明的平台：需要一个 conda 环境，为 CUDA 12.1 编译的 PyTorch 和 xformers，用 ninja 编译的 flash-attn，FFmpeg，从 Hugging Face 下载四个模型，然后在运行第一条命令之前，把文件移动并在权重文件夹里建立符号链接。所有示例都假设你有一块 NVIDIA 显卡。","notVerified":["下载完 Wan2.1 14B 基础模型和另外三套模型后，总共占用多少磁盘空间","实际需要多少显卡显存，生成一段 15 秒的片段要花多长时间","固定版本的安装步骤（CUDA 12.1 wheels、flash-attn 2.7.4）现在是否还能用","能否在 Windows 或 macOS 上运行，这些平台没有文档说明","README 中提到的长视频里的真实画质和色彩漂移情况"]}}},{"id":"R_kgDOKG7UZg","slug":"met4citizen__talkinghead","name":"TalkingHead","owner":"met4citizen","fullName":"met4citizen/TalkingHead","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"web":"B"},"risk":"attention","signals":[{"code":"api_key_in_browser_code","severity":"attention","evidence":"新手教程告诉你 \"Simply download the file, add your Google TTS API key\"，而选项表格对 ttsApikey 发出警告：\"NOTE: I recommend that you don't use this in production and never put your API key in any client-side code.\" 这样放置的密钥，任何打开这个页面的人都能读取到。","source":"ai","title":"API 密钥写在浏览器端代码里"},{"code":"example_avatars_non_commercial_only","severity":"info","evidence":"README 写明：\"brunette.glb\" 使用 CC BY-NC 4.0 许可，而 \"avatar.glb\"（Avaturn）、\"avatarsdk.glb\" 和 \"vroid.glb\" 的创建说明是 \"for non-commercial use\"。项目本身是 MIT 许可，但随附的示例 avatar、动画和声音并非都能免费用于商业项目。","source":"ai","title":"示例 avatar 仅限非商业用途"},{"code":"test_app_sends_data_to_external_ai_services","severity":"info","evidence":"index.html 为 OpenAI 聊天、内容审核和音频转写、Vertex AI、Google TTS、ElevenLabs、Microsoft TTS 和 Grok 都设置了代理。如果你启用这些功能，你的文本和麦克风音频就会发送给这些公司。","source":"ai","title":"测试应用会把数据发送给外部 AI 服务"},{"code":"author_calls_it_a_side_project","severity":"info","evidence":"FAQ 中写道：\"This is just a small side-project for me, so I don't have any big plans for it.\" 这个仓库是活跃的（检查前 3 天还有提交，版本 v1.8.2），但只由一个人维护。","source":"ai","title":"作者称它是一个业余项目"}],"platforms":["web"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Google Cloud Text-to-Speech","ElevenLabs","Microsoft Azure Speech","OpenAI","Google Gemini","Grok (x.ai)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","development"],"alternativeTo":[{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true}],"declaredAlternatives":[],"stars":1669,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":7,"growthDays":6,"history":[{"date":"2026-11-06","stars":1665},{"date":"2026-11-07","stars":1667},{"date":"2026-11-08","stars":1667},{"date":"2026-11-09","stars":1667},{"date":"2026-11-10","stars":1669}],"firstPublishedAt":"2026-11-07T23:02:21.851Z","commitSha":"1563de184c19b310de4b636664d5835e2c74dbba","verifiedAt":"2026-11-07T22:40:03.939Z","lastCommitAt":"2026-11-04T09:49:53Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A building block for websites: a 3D person who speaks your text with matching lip movements.","summary":"TalkingHead is a JavaScript class that puts a full-body 3D character on a web page and makes it speak in real time, with lips, face and gestures following the words. It reads text aloud through a text-to-speech service (Google Cloud by default, or ElevenLabs, Azure and others) and has built-in lip-sync for English, German, French, Brazilian Portuguese, Finnish, Lithuanian and Bengali. It is meant for people who write web code: you add it from NPM or a CDN and supply your own avatar file and service keys. A test page runs online.","easeRationale":"On the web there is nothing to install: the test app runs at met4citizen.github.io/TalkingHead and the class loads from a CDN, but the README states that speech and AI \"will not work\" until you paste your own API key in the settings menu, which is the one extra step. Using it inside your own site is developer work: you also need a 3D avatar file with a Mixamo rig and ARKit and Oculus blend shapes.","costNote":"The code is MIT and free. Making the avatar talk normally needs a text-to-speech account: the README recommends Google Cloud TTS (free up to 4 million characters a month, paid above that) or Microsoft Azure / ElevenLabs. The author's separate HeadTTS module can run voices in the browser for free instead.","notVerified":["How well the lip-sync looks and sounds in each supported language","The real monthly cost once you pass Google's free text-to-speech allowance","Whether a 3D avatar you create elsewhere really has the required Mixamo rig and the 52 ARKit plus 15 Oculus blend shapes","How it performs on older phones and tablets, since 3D rendering runs in the browser","What each external AI or speech service keeps of the text and audio you send"]},"zh-Hans":{"summaryShort":"一个网站组件：一个会说出你文本内容的 3D 人物，嘴型与语音同步。","summary":"TalkingHead 是一个 JavaScript 类，可以在网页上放置一个全身 3D 角色，并让它实时说话，嘴型、表情和动作都会跟随语音内容。它通过文字转语音服务（默认是 Google Cloud，也可以用 ElevenLabs、Azure 等）朗读文本，并内置了英语、德语、法语、巴西葡萄牙语、芬兰语、立陶宛语和孟加拉语的嘴型同步。它面向会写网页代码的人：你通过 NPM 或 CDN 引入它，并提供自己的 avatar 文件和服务密钥。有一个在线的测试页面可以体验。","easeRationale":"在网页上没有任何需要安装的东西：测试应用运行在 met4citizen.github.io/TalkingHead，这个类从 CDN 加载，但 README 说语音和 AI 功能 \"will not work\"，除非你在设置菜单里粘贴自己的 API 密钥，这是唯一多出来的一步。要把它用到自己的网站里就是开发工作了：你还需要一个带 Mixamo 骨骼以及 ARKit 和 Oculus 混合形状的 3D avatar 文件。","costNote":"代码采用 MIT 许可，免费。要让 avatar 正常说话，通常需要一个文字转语音账号：README 推荐 Google Cloud TTS（每月 400 万字符以内免费，超出部分收费）或 Microsoft Azure / ElevenLabs。作者另外做的 HeadTTS 模块则可以在浏览器里免费运行语音。","notVerified":["每种支持的语言，嘴型同步的实际效果和音质如何","超过 Google 文字转语音免费额度后，每月的真实费用是多少","你在别处制作的 3D avatar 是否真的具备所需的 Mixamo 骨骼，以及 52 个 ARKit 和 15 个 Oculus 混合形状","在较旧的手机和平板上表现如何，因为 3D 渲染是在浏览器里进行的","你发送的文本和音频，各个外部 AI 或语音服务会保留哪些内容"]}}},{"id":"R_kgDOVVRzng","slug":"meunomeebero__benion","name":"Benion","owner":"meunomeebero","fullName":"meunomeebero/benion","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"instagram_automation_limited_by_meta_rules","severity":"attention","evidence":"README 写道：\"Standard Access supports owned/managed accounts with the required app role; serving unrelated customer accounts requires a different access/review path... Open source does not remove provider restrictions.\" 它还指出 \"A comment permits one initial private reply within seven days\"，并且 \"New-follower welcome DM, automatic comment likes\" 都 \"Unavailable\"。大量发送消息可能导致账号被 Meta 限制。","source":"ai","title":"instagram 自动化受 meta 规则限制"},{"code":"very_new_and_small_project","severity":"attention","evidence":"仓库创建于 2026-10-23，最后一次提交在 2026-10-24，package.json 中版本号为 0.3.0，没有发布过 release，有 12 个 star 和 5 个 fork。README 自己也说 \"No automatic CI, external monitor or production credentials are included\"，而且测试 \"do not prove delivery by Meta\"。","source":"ai","title":"项目非常新且规模很小"},{"code":"you_store_account_secrets_yourself","severity":"info","evidence":".env.example 要求填写 BENION_ADMIN_PASSWORD、BENION_SESSION_SECRET、BENION_META_APP_SECRET、BENION_META_VERIFY_TOKEN 以及一个私有的 Instagram 令牌。你的 Instagram 访问权限存放在你运行程序的那台机器上，所以这台机器的防护要靠你自己负责。","source":"ai","title":"账号密钥由你自己保管"},{"code":"ai_replies_can_be_wrong","severity":"info","evidence":"README 的能力表中写道，模型 \"drafts and reviews the reply; this reduces unsupported claims but cannot guarantee model accuracy. You can take over at any time.\" 给粉丝的自动回复仍然可能说错话。","source":"ai","title":"ai 的回复可能出错"},{"code":"interface_in_portuguese_only","severity":"info","evidence":"README 写道：\"The dashboard currently uses Portuguese, and scheduling uses São Paulo time.\"","source":"ai","title":"界面只有葡萄牙语"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram app (app secret and account token)","OpenRouter (AI replies)","Higgsfield (video generation)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"buffer","name":"Buffer","confirmed":true}],"declaredAlternatives":[{"target":"manychat","stance":"claims","evidence":"Benion is a personal Instagram alternative to a subset of ManyChat, not a promise of every ManyChat feature or provider exemption"}],"stars":12,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":12},{"date":"2026-11-08","stars":12},{"date":"2026-11-09","stars":12},{"date":"2026-11-10","stars":12}],"firstPublishedAt":"2026-11-07T23:02:21.859Z","commitSha":"c291f10213c2972eefd0da287df15ca52e820a33","verifiedAt":"2026-11-07T22:40:50.397Z","lastCommitAt":"2026-10-24T06:34:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs your own Instagram robot: sends a DM to whoever comments, schedules Reels and answers the inbox.","summary":"Benion is a dashboard you run on your own computer or server to automate your professional Instagram account: it watches comments on a Reel and sends each person a prepared direct message, posts fixed public replies, schedules Reels, and can answer the inbox with an AI model based on your product catalog, with a button to take over by hand. It is plain Node.js with no libraries to install: you start one command and open a local page. You must create your own Meta app and connect your Business or Creator account.","easeRationale":"On any of the three systems you install Node.js 20 or newer and run one command, \"node server.mjs\", then open http://127.0.0.1:5197 - the README says no npm install is needed. The hard part is outside the app: you must create your own Meta app and connect a Business or Creator account, and for incoming DMs to arrive the README asks for \"an authenticated HTTPS instance with persistent storage\", which is server work.","costNote":"The code is MIT. The README notes that \"hosting, storage, model inference and Higgsfield generation can cost money\", so the AI inbox and the video generation parts are paid services you connect yourself.","notVerified":["Whether campaigns and scheduled Reels really work end to end with a live Meta app, since the tests use synthetic transports","How much the AI inbox costs per month in model usage","How hard the Meta app review and permission steps are for a given account","What a public HTTPS deployment costs and how well the default settings protect it once exposed","Whether Instagram may restrict an account that uses this kind of automation"]},"zh-Hans":{"summaryShort":"运行你自己的 Instagram 机器人：给评论的人发私信，安排 Reels 发布，并回复私信收件箱。","summary":"Benion 是一个你在自己的电脑或服务器上运行的控制面板，用来自动化管理你的 Instagram 专业账号：它会监控某条 Reel 下的评论，给每个评论的人发送一条预设好的私信，发布固定的公开回复，安排 Reels 发布时间，还能用基于你产品目录训练的 AI 模型回复收件箱，并提供一个按钮随时手动接管。它是纯 Node.js，不需要安装任何库：你只需启动一条命令并打开一个本地页面。你必须自己创建 Meta 应用，并连接你的 Business 或 Creator 账号。","easeRationale":"在这三种系统中的任何一种上，你安装 Node.js 20 或更新版本，运行一条命令 \"node server.mjs\"，然后打开 http://127.0.0.1:5197——README 说不需要运行 npm install。真正困难的部分在应用之外：你必须自己创建 Meta 应用并连接 Business 或 Creator 账号，而且要让私信能够送达，README 要求 \"an authenticated HTTPS instance with persistent storage\"，这属于服务器方面的工作。","costNote":"代码采用 MIT 许可。README 指出 \"hosting, storage, model inference and Higgsfield generation can cost money\"，所以 AI 收件箱和视频生成这些部分都是需要你自己接入的付费服务。","notVerified":["在真实的 Meta 应用下，活动和已安排的 Reels 是否真的能端到端正常运行，因为测试用的是模拟传输","AI 收件箱在模型调用上每月要花多少钱","对于某个具体账号来说，Meta 应用审核和权限申请的步骤有多难","公开部署的 HTTPS 服务要花多少钱，默认设置在暴露后保护效果如何","Instagram 是否会限制使用这种自动化方式的账号"]}}},{"id":"R_kgDONO810w","slug":"microsoft__markitdown","name":"MarkItDown","owner":"microsoft","fullName":"microsoft/markitdown","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"can_fetch_remote_urls","severity":"info","evidence":"README.md：\"MarkItDown performs I/O with the privileges of the current process\" 以及 \"convert() method is intentionally permissive and can handle local files, remote URIs, and byte streams.\" README 要求开发者对不可信的输入进行清理（sanitize）。","source":"ai","title":"可以获取远程网址"},{"code":"third_party_plugins","severity":"info","evidence":"README.md：\"MarkItDown also supports 3rd-party plugins. Plugins are disabled by default.\" 用 --use-plugins 打开它们会运行其他作者的代码。","source":"ai","title":"第三方插件"},{"code":"cloud_options_send_documents","severity":"info","evidence":"README.md 对比表：Azure Document Intelligence 和 Content Understanding 分别是 \"Cloud layout extraction\" / \"Cloud multimodal extraction\"，并带有 \"Billable Azure API calls\"；OCR 插件通过 OpenAI 客户端使用 \"LLM Vision\"。使用这些选项时，你的文件会离开你的电脑。","source":"ai","title":"云端选项会发送文档"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Azure Document Intelligence (optional cloud conversion)","Azure Content Understanding (optional cloud conversion, audio and video)","OpenAI or a compatible service (optional image descriptions and OCR plugin)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["office","development"],"alternativeTo":[{"slug":"mathpix","name":"Mathpix","confirmed":true}],"declaredAlternatives":[],"stars":194790,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1053,"growthDays":7,"history":[{"date":"2026-10-26","stars":192382},{"date":"2026-10-27","stars":192576},{"date":"2026-10-28","stars":192735},{"date":"2026-10-29","stars":192901},{"date":"2026-10-30","stars":193083},{"date":"2026-10-31","stars":193246},{"date":"2026-11-01","stars":193426},{"date":"2026-11-02","stars":193598},{"date":"2026-11-03","stars":193737},{"date":"2026-11-04","stars":193886},{"date":"2026-11-05","stars":194046},{"date":"2026-11-06","stars":194190},{"date":"2026-11-07","stars":194314},{"date":"2026-11-08","stars":194446},{"date":"2026-11-09","stars":194600},{"date":"2026-11-10","stars":194790}],"firstPublishedAt":"2026-10-27T01:10:12.658Z","commitSha":"7482c092d9c0074a89803c9fdb63a969e7fe17fc","verifiedAt":"2026-10-29T01:42:35.966Z","lastCommitAt":"2026-11-07T00:14:01Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns PDFs, Word, Excel, PowerPoint and other files into plain Markdown text that AI tools can read.","summary":"MarkItDown is a Microsoft Python tool that converts files into Markdown, a simple text format that AI chatbots and text-analysis tools read well. It handles PDF, Word, PowerPoint, Excel, images, audio, HTML, CSV, JSON, ZIP, EPUB and YouTube links, and keeps headings, lists and tables. The README says the output is meant for software, not for faithful copies for people to read. You need Python 3.10 to 3.14 and use it from the command line or in Python code. Optional Azure and OpenAI features are paid.","easeRationale":"The work is the same on all three: install Python, run pip install 'markitdown[all]' and then type a command such as markitdown file.pdf -o document.md for each file. The README's virtual-environment commands are written for Mac and Linux shells.","costNote":"Free under the MIT license; built-in conversion runs on your computer. Azure Document Intelligence, Azure Content Understanding and OpenAI image descriptions are optional and billed by those services.","notVerified":["How good the conversion is for complex layouts, scanned PDFs and large spreadsheets without the cloud options.","Exact Windows steps: the README shows Mac/Linux commands such as source .venv/bin/activate.","Whether audio transcription needs ffmpeg installed separately on Windows.","What third-party plugins found through #markitdown-plugin actually do.","Real cost of the Azure and OpenAI options per document."]},"zh-Hans":{"summaryShort":"把PDF、Word、Excel、PowerPoint等文件转换成AI工具能读懂的纯Markdown文本。","summary":"MarkItDown是微软出品的一款Python工具,可以把文件转换成Markdown——一种AI聊天机器人和文本分析工具都能很好读取的简单文本格式。它支持PDF、Word、PowerPoint、Excel、图片、音频、HTML、CSV、JSON、ZIP、EPUB和YouTube链接,并保留标题、列表和表格。README说明这种输出是给软件使用的,不是给人阅读的忠实副本。你需要Python 3.10到3.14,通过命令行或在Python代码中使用它。可选的Azure和OpenAI功能是付费的。","easeRationale":"三个系统上的操作基本相同:安装Python,运行pip install 'markitdown[all]',然后对每个文件输入类似markitdown file.pdf -o document.md这样的命令。README中关于虚拟环境的命令是按Mac和Linux的shell写的。","costNote":"在MIT许可下免费;内置的转换功能在你自己的电脑上运行。Azure Document Intelligence、Azure Content Understanding和OpenAI的图片描述功能是可选的,按这些服务各自的收费标准计费。","notVerified":["在不使用云端选项的情况下,对复杂排版、扫描版PDF和大型电子表格的转换效果如何。","Windows上的具体操作步骤:README展示的是Mac/Linux的命令,例如source .venv/bin/activate。","音频转写在Windows上是否需要单独安装ffmpeg。","通过#markitdown-plugin找到的第三方插件实际做了什么。","Azure和OpenAI选项按文档计算的实际费用。"]}}},{"id":"R_kgDOOMjd8A","slug":"microsoft__playwright-mcp","name":"Playwright MCP","owner":"microsoft","fullName":"microsoft/playwright-mcp","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"ai_controls_browser_with_your_logins","severity":"attention","evidence":"README 写道：\"All the logged in information will be stored in the persistent profile\"，而 --extension 选项可以 \"Connect to a running browser instance (Edge/Chrome only)... leverage your logged-in sessions and browser state\"。助手决定点击或输入的任何操作，都发生在一个以你身份登录的浏览器里，而访问过的页面上的文本也可能试图引导助手。","source":"ai","title":"ai 用你的登录状态控制浏览器"},{"code":"authors_say_it_is_not_a_security_boundary","severity":"attention","evidence":"README 的相关章节写道：\"Playwright MCP is not a security boundary.\" 配置说明里对密钥也重复了这一点（\"It is a convenience and not a security feature\"），对 allowUnrestrictedFileAccess 也是如此（\"a convenience defense to catch unintended file access, not a secure boundary\"）。","source":"ai","title":"作者自己说这不是一道防护边界"},{"code":"can_run_arbitrary_javascript_and_read_files","severity":"info","evidence":"工具里包括 browser_evaluate（\"Evaluate JavaScript expression on page or element\"）和 browser_drop，带有 \"Absolute paths to files to drop onto the element\"；--allow-unrestricted-file-access 会取消默认仅限工作区文件夹的限制。","source":"ai","title":"可以执行任意 javascript 并读取文件"},{"code":"pre_release_version_and_alpha_dependency","severity":"info","evidence":"package.json 的版本号是 0.0.84，并锁定了 \"playwright\": \"1.66.0-alpha-1793916328000\"，这是一个 alpha 版本。选项和工具名称在各个版本之间仍可能发生变化（目前已有 74 个 release）。","source":"ai","title":"预发布版本及 alpha 依赖"},{"code":"docker_image_disables_browser_sandbox","severity":"info","evidence":"Dockerfile 的 ENTRYPOINT 运行 \"node /app/cli.js --headless --browser chromium --no-sandbox\"，所以在这个容器里，浏览器自身的防护层是关闭的。","source":"ai","title":"docker 镜像关闭了浏览器沙盒"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","ai_agents","development"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":39452,"downloads":95426,"downloadsWeek":0,"downloadsDays":6,"growth":314,"growthDays":7,"history":[{"date":"2026-11-04","stars":39179},{"date":"2026-11-05","stars":39226},{"date":"2026-11-06","stars":39266},{"date":"2026-11-07","stars":39306},{"date":"2026-11-08","stars":39339},{"date":"2026-11-09","stars":39387},{"date":"2026-11-10","stars":39452}],"firstPublishedAt":"2026-11-07T23:09:40.948Z","commitSha":"404be8eb89abecff77a55152b6d843d140c78372","verifiedAt":"2026-11-07T23:05:40.459Z","lastCommitAt":"2026-11-10T16:50:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant open and use a real web browser: click, type, read pages and take screenshots.","summary":"Playwright MCP is a connector (MCP server) from Microsoft that gives AI assistants such as VS Code Copilot, Claude Code, Cursor or Claude Desktop the ability to drive a real browser: open pages, click, fill forms, read the page structure and take screenshots. It reads the page as an accessibility tree instead of images, so no vision model is needed. You install Node.js 18 or newer and add a few lines to your assistant's settings, or run it in Docker. It can use a saved browser profile with your logins.","easeRationale":"The same work on Windows, macOS and Linux: install Node.js 18 or newer, then paste the standard JSON block into your assistant's settings or run one command such as \"claude mcp add playwright npx @playwright/mcp@latest\"; VS Code, Cursor and others also have one-click install links. You still need an MCP client, and the browser is downloaded on first use. The Docker route is an alternative and only supports headless Chromium.","notVerified":["How much disk the downloaded browsers take after the first run","What a given assistant actually does with the browser once it has these tools","Whether a web page can trick the assistant into unwanted clicks or logins in your profile","How well it works with each of the many listed MCP clients","Whether screenshots or page text containing personal data end up in the AI provider's logs"]},"zh-Hans":{"summaryShort":"让 AI 助手打开并使用一个真实的网页浏览器：点击、输入、读取页面内容、截屏。","summary":"Playwright MCP 是微软推出的一个连接器（MCP 服务器），让 VS Code Copilot、Claude Code、Cursor 或 Claude Desktop 这类 AI 助手具备操控真实浏览器的能力：打开页面、点击、填写表单、读取页面结构并截屏。它读取的是页面的无障碍树（accessibility tree）而不是图像，所以不需要视觉模型。你安装 Node.js 18 或更新版本，在助手的设置里加几行配置即可，也可以用 Docker 运行。它还能使用保存了你登录状态的浏览器配置文件。","easeRationale":"在 Windows、macOS 和 Linux 上步骤相同：安装 Node.js 18 或更新版本，然后把标准 JSON 配置块粘贴进助手的设置里，或者运行一条命令，比如 \"claude mcp add playwright npx @playwright/mcp@latest\"；VS Code、Cursor 等也提供一键安装链接。你仍然需要一个 MCP 客户端，浏览器会在首次使用时下载。Docker 方式是另一种选择，但只支持无界面（headless）的 Chromium。","notVerified":["首次运行后，下载的浏览器占用多少磁盘空间","某个具体的助手拿到这些工具后，实际会用浏览器做什么","网页内容是否可能诱导助手在你的浏览器配置里做出不想要的点击或登录操作","它在列出的众多 MCP 客户端里，每一个的实际效果如何","包含个人信息的截图或页面文本是否会留存在 AI 提供商的日志里"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxODQ0NTYyNTE=","slug":"microsoft__powertoys","name":"PowerToys","owner":"microsoft","fullName":"microsoft/PowerToys","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"telemetry_disclosed","severity":"info","evidence":"README，Privacy statement：\"The application logs basic diagnostic data (telemetry).\" 仓库中还附带一个 DATA_AND_PRIVACY.md 文件。","source":"ai","title":"已声明收集遥测数据"},{"code":"tools_that_touch_system_settings","severity":"info","evidence":"工具列表包括 \"Hosts File Editor\"、\"Registry Preview\"、\"Environment Variables\" 和 \"Keyboard Manager\"，它们会更改影响整台电脑的 Windows 设置。全机安装选项（`winget install --scope machine`）需要管理员权限。","source":"ai","title":"会改动系统设置的工具"},{"code":"large_install_for_small_tools","severity":"info","evidence":"每个发布版安装程序约 300 MB（PowerToysSetup-0.103.3006.0-x64.exe，303 MB），尽管各个工具本身都很小。","source":"ai","title":"小工具却要很大的安装包"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":303,"maxMB":910,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["productivity","network_system"],"alternativeTo":[{"slug":"displayfusion","name":"DisplayFusion","confirmed":true},{"slug":"snagit","name":"Snagit","confirmed":true}],"declaredAlternatives":[],"stars":140832,"downloads":151183859,"downloadsWeek":1399288,"downloadsDays":7,"growth":328,"growthDays":7,"history":[{"date":"2026-10-25","stars":139934},{"date":"2026-10-26","stars":140003},{"date":"2026-10-27","stars":140066},{"date":"2026-10-28","stars":140167},{"date":"2026-10-29","stars":140234},{"date":"2026-10-30","stars":140304},{"date":"2026-10-31","stars":140356},{"date":"2026-11-01","stars":140397},{"date":"2026-11-02","stars":140442},{"date":"2026-11-03","stars":140504},{"date":"2026-11-04","stars":140551},{"date":"2026-11-05","stars":140612},{"date":"2026-11-06","stars":140634},{"date":"2026-11-07","stars":140695},{"date":"2026-11-08","stars":140733},{"date":"2026-11-09","stars":140785},{"date":"2026-11-10","stars":140832}],"firstPublishedAt":"2026-10-25T23:19:11.665Z","commitSha":"0ccfdff939e8820c041baf2f88b097866eb76d10","verifiedAt":"2026-10-29T01:44:09.830Z","lastCommitAt":"2026-11-10T18:22:39Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"PowerToysUserSetup-0.102.3101.0-x64.exe","sizeBytes":323951160,"url":"https://github.com/microsoft/PowerToys/releases/download/v0.102.3101/PowerToysUserSetup-0.102.3101.0-x64.exe","releaseUrl":"https://github.com/microsoft/PowerToys/releases/tag/v0.102.3101"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Add over 30 handy extras to Windows: window layouts, bulk renaming, a color picker and a quick launcher.","summary":"PowerToys is a free set of more than 30 small tools from Microsoft that make Windows easier to use. Examples: arrange windows in custom layouts, rename many files at once, resize images, pick colours from the screen, copy text out of images, remap keys and launch apps from a search box. You turn each tool on or off in one settings window. It is for anyone on Windows 10 or 11 who wants more control over their desktop, and it installs with a normal Windows installer.","easeRationale":"Windows is the only system it runs on and there it is an A: the releases carry ready installers (for example PowerToysUserSetup-0.103.3006.0-x64.exe, about 303 MB), and the README also offers the Microsoft Store and WinGet.","costNote":"MIT licence, no paid version.","notVerified":["Exactly what diagnostic data is sent and whether you can turn it off in your version","Which of the 30+ tools work smoothly alongside your other apps and shortcuts","Real disk use after installing, which Radar estimates from the installer size","Whether the tools that change system files need administrator rights every time"]},"zh-Hans":{"summaryShort":"为Windows增加30多个实用小工具:窗口布局、批量重命名、取色器和快速启动器。","summary":"PowerToys是微软推出的一套免费工具集,包含30多个让Windows更好用的小工具。例如:按自定义布局排列窗口、一次性重命名多个文件、调整图片大小、从屏幕上取色、从图片中提取文字、重新映射按键,以及通过搜索框启动应用。你可以在一个设置窗口里逐个开关这些工具。它适合任何使用Windows 10或11、想对桌面有更多控制权的人,安装方式是普通的Windows安装程序。","easeRationale":"Windows是它唯一支持的系统,在这里评级为A:发布页提供了现成的安装程序(例如PowerToysUserSetup-0.103.3006.0-x64.exe,大约303MB),README还提供了通过Microsoft Store和WinGet安装的方式。","costNote":"MIT许可,没有付费版本。","notVerified":["具体发送了哪些诊断数据,以及在你使用的版本中能否关闭。","30多个工具中,哪些能与你的其他应用和快捷键顺畅共存。","安装后的实际磁盘占用,这是Radar根据安装程序大小估算出来的。","会修改系统文件的工具,是否每次都需要管理员权限。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMDAwNjA5MTI=","slug":"microsoft__terminal","name":"Windows Terminal","owner":"microsoft","fullName":"microsoft/terminal","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"the_newest_releases_here_do_not_hold_the_installer","severity":"info","evidence":"README 告诉手动安装的人\"Download the Microsoft.WindowsTerminal_<versionNumber>.msixbundle file from the Assets section\",但在最新的三个发布版本中,唯一的文件是一个2MB的 Microsoft.Windows.Console.ConPTY nupkg(一个开发者组件),其中两个还带有一个msixbundle的\"Windows10_PreinstallKit\"压缩包。现成的应用来自 Microsoft Store 或 winget,而不是这些发布附件。","source":"ai","title":"这里最新的发布版本没有安装程序"},{"code":"manual_installs_do_not_update_themselves","severity":"info","evidence":"README 提示:\"Terminal will not auto-update when new builds are released so you will need to regularly install the latest Terminal release to receive all the latest fixes and improvements!\"Canary 便携版 ZIP 同样如此,它\"will not automatically update and will not automatically check for updates\"。","source":"ai","title":"手动安装不会自动更新"},{"code":"two_install_channels_are_marked_unofficial","severity":"info","evidence":"README 自己的小标题写着\"Via Chocolatey (unofficial)\"和\"Via Scoop (unofficial)\",这些包出问题时会被引导去 Chocolatey 的处理流程或 Scoop Extras 的 issue 跟踪器,而不是 Microsoft。","source":"ai","title":"两个安装渠道被标记为非官方"},{"code":"older_windows_and_canary_limits","severity":"info","evidence":"README:\"Windows Terminal requires Windows 10 2004 (build 19041) or later\",在较旧的 Windows 10 版本上手动安装可能还需要 VC++ v14 Desktop Framework Package,而 Canary App Installer 版本\"only works on Windows 11\"。Canary 被描述为\"our least stable offering\"。","source":"ai","title":"旧版Windows和Canary版本的限制"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":2,"maxMB":5,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":105364,"downloads":41684506,"downloadsWeek":641508,"downloadsDays":7,"growth":44,"growthDays":7,"history":[{"date":"2026-10-26","stars":105246},{"date":"2026-10-27","stars":105257},{"date":"2026-10-28","stars":105265},{"date":"2026-10-29","stars":105277},{"date":"2026-10-30","stars":105292},{"date":"2026-10-31","stars":105299},{"date":"2026-11-01","stars":105305},{"date":"2026-11-02","stars":105315},{"date":"2026-11-03","stars":105320},{"date":"2026-11-04","stars":105324},{"date":"2026-11-05","stars":105327},{"date":"2026-11-06","stars":105331},{"date":"2026-11-07","stars":105343},{"date":"2026-11-08","stars":105352},{"date":"2026-11-09","stars":105357},{"date":"2026-11-10","stars":105364}],"firstPublishedAt":"2026-10-27T01:10:12.659Z","commitSha":"703ef7f969d90d3f7e132b9a5f51a8815e77d586","verifiedAt":"2026-11-01T01:47:28.636Z","lastCommitAt":"2026-11-09T21:32:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Microsoft's terminal window for Windows, with tabs, Unicode text and colour, for PowerShell, cmd and Linux shells.","summary":"This is the source code of Windows Terminal, Microsoft's window for typing commands, with tabs, panes, themes, emoji and full Unicode text, able to host PowerShell, the old cmd prompt and Linux shells through WSL. The same repository holds the classic Windows console host (conhost.exe) that ships inside Windows itself. For a normal user it is an app, not a project: Microsoft recommends installing it from the Microsoft Store, and it needs Windows 10 version 2004 (build 19041) or newer.","easeRationale":"Windows is the only platform it exists for, and installing it is a single step: the Microsoft Store (the method the README calls preferred, with automatic updates) or `winget install --id Microsoft.WindowsTerminal -e`. Nothing has to be configured before the first use, and there is no grade for macOS or Linux because the program is Windows-only.","costNote":"Free, MIT-licensed and published by Microsoft, with no paid tier.","notVerified":["Whether the app you get from the Microsoft Store was built from exactly this source","Which release file to use if you cannot reach the Store, since the newest releases here carry only developer components","Real disk use after installing, estimated at 2 to 5 MB from the published files, which is far below a full app install","How it behaves on your Windows version and graphics drivers, as rendering is hardware-accelerated"]},"zh-Hans":{"summaryShort":"微软为Windows打造的终端窗口,支持多标签、Unicode文本和颜色,可用于PowerShell、cmd和Linux shell。","summary":"这是 Windows Terminal 的源代码,这是 Microsoft 用来输入命令的窗口,支持多标签、多窗格、主题、表情符号和完整的 Unicode 文本,能够承载 PowerShell、旧版 cmd 命令提示符以及通过 WSL 运行的 Linux shell。同一个仓库里还有经典的 Windows 控制台主机(conhost.exe),它内置在 Windows 本身里。对普通用户来说它是一个应用,不是一个项目:Microsoft 建议从 Microsoft Store 安装,需要 Windows 10 版本 2004(内部版本 19041)或更高版本。","easeRationale":"Windows 是它唯一存在的平台,安装只需要一步:Microsoft Store(README 称其为首选方式,支持自动更新)或者`winget install --id Microsoft.WindowsTerminal -e`。首次使用前不需要配置任何东西,而 macOS 或 Linux 没有评级,因为这个程序只支持 Windows。","costNote":"免费,采用 MIT 许可,由 Microsoft 发布,没有付费层级。","notVerified":["你从 Microsoft Store 获取的应用是否确实是从这份代码构建的","如果你无法访问 Store,该用哪个发布文件,因为这里最新的发布版本只带有开发者组件","安装后实际占用的磁盘空间,根据已发布文件估计在2到5MB之间,远低于完整应用安装","在你的 Windows 版本和显卡驱动上表现如何,因为渲染是硬件加速的"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0MTg4MTkwMA==","slug":"microsoft__vscode","name":"Visual Studio Code","owner":"microsoft","fullName":"microsoft/vscode","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"official_build_has_different_license","severity":"info","evidence":"README.md：\"Visual Studio Code is a distribution of the Code - OSS repository with Microsoft-specific customizations released under a traditional Microsoft product license.\" GitHub 发布页（1.144.0、1.143.1、1.143.0）没有附带文件；安装程序来自 code.visualstudio.com，而不是这个仓库。","source":"ai","title":"官方版本使用不同的许可证"},{"code":"extensions_run_third_party_code","severity":"info","evidence":"README.md 描述了 \"a rich extensibility model\"，Dev Containers 链接 \"will cause VS Code to automatically install the Dev Containers extension if needed\"。应用市场里的扩展由许多作者制作，并在你的电脑上运行。","source":"ai","title":"扩展会运行第三方代码"},{"code":"built_in_ai_assistant","severity":"info","evidence":"顶层文件夹 \"copilot_shim/\" 和 package.json 脚本 \"compile-copilot\": \"npm --prefix extensions/copilot run compile\"。该编辑器自带 GitHub Copilot 集成，使用时会把代码发送到在线 AI 服务。","source":"ai","title":"内置 AI 助手"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["GitHub account for Copilot AI features"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"sublime-text","name":"Sublime Text","confirmed":true},{"slug":"jetbrains-ides","name":"IDEs","confirmed":true}],"declaredAlternatives":[],"stars":195267,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":728,"growthDays":7,"history":[{"date":"2026-10-26","stars":194352},{"date":"2026-10-27","stars":194394},{"date":"2026-10-28","stars":194397},{"date":"2026-10-29","stars":194353},{"date":"2026-10-30","stars":194390},{"date":"2026-10-31","stars":194427},{"date":"2026-11-01","stars":194461},{"date":"2026-11-02","stars":194506},{"date":"2026-11-03","stars":194539},{"date":"2026-11-04","stars":194619},{"date":"2026-11-05","stars":194727},{"date":"2026-11-06","stars":194952},{"date":"2026-11-07","stars":195138},{"date":"2026-11-08","stars":195196},{"date":"2026-11-09","stars":195235},{"date":"2026-11-10","stars":195267}],"firstPublishedAt":"2026-10-27T01:10:12.660Z","commitSha":"ac06fd8ac0eed44e3fcea3e32e7eeac1378349ad","verifiedAt":"2026-10-29T01:42:14.609Z","lastCommitAt":"2026-11-10T22:38:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free code editor from Microsoft for writing, running and fixing programs, with thousands of add-ons.","summary":"Visual Studio Code is Microsoft's widely used code editor. It helps you write and fix code in almost any programming language, with color highlighting, smart suggestions, a built-in debugger, Git support and a large library of extensions. This repository holds the open-source base, called Code - OSS. The ready-to-install Visual Studio Code from code.visualstudio.com adds Microsoft-specific parts under a Microsoft product license. It is for programmers and anyone who edits code or text files. Installers exist for Windows, macOS and Linux.","easeRationale":"Microsoft offers a ready installer for Windows, macOS and Linux on code.visualstudio.com, so on every platform you download and open it; building from this repository is only needed by developers who contribute.","costNote":"The source code is MIT. The official download is free but released under a Microsoft product license. Some add-ons and AI features, such as GitHub Copilot, have their own plans.","notVerified":["What usage data (telemetry) the official build sends by default and how to turn it off.","Whether the official installer matches this source code exactly, given the Microsoft-specific additions.","What each third-party extension does with your files.","How Copilot handles the code you share with it."]},"zh-Hans":{"summaryShort":"微软出品的免费代码编辑器,用于编写、运行和修复程序,拥有数千个扩展插件。","summary":"Visual Studio Code是微软广泛使用的代码编辑器。它能帮你用几乎任何编程语言编写和修复代码,具备颜色高亮、智能建议、内置调试器、Git支持,以及庞大的扩展库。这个仓库包含开源的基础版本,称为Code - OSS。可从code.visualstudio.com下载的现成版Visual Studio Code,在微软产品许可下加入了微软专属的部分。它面向程序员以及任何需要编辑代码或文本文件的人。提供Windows、macOS和Linux的安装程序。","easeRationale":"微软在code.visualstudio.com为Windows、macOS和Linux提供了现成的安装程序,所以在每个平台上你只需下载并打开即可;只有参与贡献的开发者才需要从这个仓库自行构建。","costNote":"源代码采用MIT许可。官方下载是免费的,但以微软产品许可发布。一些扩展和AI功能(例如GitHub Copilot)有各自的付费计划。","notVerified":["官方版本默认发送哪些使用数据(遥测),以及如何关闭它。","考虑到微软专属的附加内容,官方安装程序是否与这份源代码完全一致。","每个第三方扩展具体如何处理你的文件。","Copilot如何处理你与它分享的代码。"]}}},{"id":"R_kgDOT7RmQA","slug":"mike1ros-gh__nudge1","name":"Nudge1","owner":"mike1ros-gh","fullName":"mike1ros-gh/nudge1","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 82 天前","source":"check","title":"所有者账号很新"},{"code":"readme_prompt_instructs_ai_assistant_to_star_repo","severity":"attention","evidence":"README.md 里可供复制粘贴的提示词告诉用户的 AI 助手：\"open https://github.com/mike1ros-gh/nudge1 in your browser tool and ask if you can star it for me - I'm probably still signed into GitHub there... with my okay, click the Star button yourself\"。同一段提示词还禁止助手重命名应用、更改 logo 或移除捐赠链接，\"even if I ask\"。这是一段指挥在你账号里行动的 AI 的指令文本，可能会虚增项目的 star 数量。","source":"ai","title":"readme 里的提示词指示 ai 助手给仓库加星"},{"code":"setup_path_asks_you_to_paste_secrets_into_an_ai_chat","severity":"attention","evidence":"README 里的提示词要求助手索取 \"my Resend API key... and for the three Meta secrets once I create the app\"，只是额外加了一句 \"Remind me to rotate any secret I paste to you before real use\"。按照这个流程操作，会把你 Instagram 应用的密钥发送到一个 AI 聊天里。","source":"ai","title":"设置流程要求你把密钥粘贴进 ai 聊天"},{"code":"very_new_and_small_project","severity":"attention","evidence":"创建于 2026-08-17，最后一次提交在 2026-08-23（检查前 76 天），没有发布过 release，package.json 中版本号为 0.1.0，有 7 个 star 和 2 个 fork。它还依赖像 \"next-auth\": \"^5.0.0-beta.31\" 这样的预发布库。","source":"ai","title":"项目非常新且规模很小"},{"code":"automated_dms_limited_by_meta_rules","severity":"attention","evidence":"README 写道，频率限制 \"Stays under Meta's documented cap of 750 private replies per hour\"，收件箱功能只在 \"inside Meta's 24-hour messaging window\" 内可用，账号必须是 \"a Business or Creator account, not a personal one\"，提示词还指出 \"the app must be published for real webhooks to arrive\"（参见 META_APP_REVIEW.md）。Meta 可能会拒绝这个应用或限制这个账号。","source":"ai","title":"自动私信受 meta 规则限制"},{"code":"you_hold_instagram_tokens_and_a_shared_encryption_key","severity":"info","evidence":".env.example 要求填写 NEXTAUTH_SECRET、CRON_SECRET、ENCRYPTION_KEY、INSTAGRAM_APP_SECRET、FACEBOOK_APP_SECRET 和 WEBHOOK_VERIFY_TOKEN，设置提示词还强调 ENCRYPTION_KEY 必须 \"identical on the web app and wherever the worker runs\"。你的 Instagram 访问权限如何，取决于你把这些值保管得有多好。","source":"ai","title":"instagram 令牌和共享加密密钥由你自己保管"},{"code":"branding_and_donate_links_must_stay","severity":"info","evidence":"README 的致谢部分写道：\"please keep the name \\\"Nudge1\\\" and its gradient \\\"N1\\\" logo unchanged... Please also keep this credits section and the donate links as they are.\" AGPL 协议还要求，如果你把修改过的版本作为服务提供给其他人使用，就必须公开你的改动。","source":"ai","title":"品牌标识和捐赠链接必须保留"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram developer app (app ID and secrets)","Resend (login e-mails)"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":7,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":7},{"date":"2026-11-09","stars":7},{"date":"2026-11-10","stars":7}],"firstPublishedAt":"2026-11-07T22:56:16.876Z","commitSha":"cdf93fe3400e1d3db040a137a76da5830d839b1f","verifiedAt":"2026-11-07T22:42:21.212Z","lastCommitAt":"2026-08-23T03:18:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"When someone comments a keyword on your Instagram post, it sends that person a DM with your link.","summary":"Nudge1 is a dashboard you host yourself that watches comments on your Instagram posts and sends a private reply to anyone who comments a keyword you chose, using Meta's official API. It can also post a public reply, swap your link for a tracked one to count clicks, greet people by name, read and answer DMs, and keep logs of every send. It is a Next.js web app plus a background worker, and needs Postgres, Redis, a Meta developer app, a Resend account for login emails, and a Business or Creator Instagram account.","easeRationale":"There is no installer on any system: you clone the code, run npm install, start Postgres and Redis with docker-compose, run a database migration, and keep two processes running (the web app and the worker). On top of that you must create and publish a Meta developer app and click through Instagram's screens, which the README says to \"budget an afternoon\" for. macOS has a bundled worker app, but building it needs the Xcode command line tools.","costNote":"The code is AGPL-3.0 and free to self-host, and the README says there are no seat limits or plan caps. You pay nothing only if you stay inside free hosting tiers; the README itself warns that Railway, Render and Fly free tiers are \"a trial, not forever\" and that Upstash Redis stops at 500,000 commands a month. A commercial license is offered for organizations (COMMERCIAL.md).","notVerified":["Whether the whole chain (webhook, keyword match, worker, DM) really works on a live account","How long Meta app review takes and whether a given account is accepted","What it costs per month once free hosting tiers run out or the Redis cap is reached","Whether the hosted page at nudge1-app.vercel.app is a demo or a usable service","How the project will be maintained, given one author and no commits in over two months"]},"zh-Hans":{"summaryShort":"当有人在你的 Instagram 帖子下评论关键词时，它会给那个人发一条带着你链接的私信。","summary":"Nudge1 是一个你自己托管的控制面板，它会监控你 Instagram 帖子下的评论，使用 Meta 官方 API，给评论了你指定关键词的人发送一条私信回复。它还能发布公开回复、把你的链接换成可统计点击量的追踪链接、按名字问候对方、读取并回复私信，并记录每一次发送的日志。它是一个 Next.js 网页应用加一个后台工作进程，需要 Postgres、Redis、一个 Meta 开发者应用、一个用于登录邮件的 Resend 账号，以及一个 Instagram Business 或 Creator 账号。","easeRationale":"任何系统上都没有安装程序：你需要克隆代码、运行 npm install、用 docker-compose 启动 Postgres 和 Redis、运行一次数据库迁移，并让两个进程保持运行（网页应用和工作进程）。除此之外，你还必须创建并发布一个 Meta 开发者应用，并走完 Instagram 的审核界面，README 说这需要 \"budget an afternoon\"（预留一个下午）。macOS 自带了一个打包好的 worker 应用，但编译它需要 Xcode 命令行工具。","costNote":"代码采用 AGPL-3.0 许可，可以免费自行部署，README 说没有用户数量限制，也没有套餐上限。只有当你一直待在免费的托管额度内才完全不花钱；README 自己也提醒，Railway、Render 和 Fly 的免费额度是 \"a trial, not forever\"（只是试用，不是永久），Upstash Redis 每月达到 500,000 条命令就会停止。面向企业机构还提供商业许可（COMMERCIAL.md）。","notVerified":["整条链路（webhook、关键词匹配、worker、私信）在真实账号上是否真的能正常运作","Meta 应用审核需要多长时间，某个具体账号是否会被通过","免费托管额度用完或 Redis 上限达到后，每月要花多少钱","nudge1-app.vercel.app 上托管的页面是演示还是一个可用的服务","鉴于只有一位作者，而且已经两个多月没有提交记录，这个项目将如何继续维护"]}}},{"id":"R_kgDOT1lw9w","slug":"milind-soni__openmausbot","name":"OpenMausBot","owner":"milind-soni","fullName":"milind-soni/OpenMausBot","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：undici: GHSA-2m5v-pch6-74hx, GHSA-4qv8-2p9j-2685, GHSA-4w7w-5xrc-9h26","source":"check","title":"已知漏洞"},{"code":"windows_installer_not_code_signed","severity":"attention","evidence":"README 快速开始表格：\"The installer isn't code-signed yet, so SmartScreen shows 'unknown publisher': More info → Run anyway.\"","source":"ai","title":"Windows 安装程序没有代码签名"},{"code":"bots_can_control_your_own_computer","severity":"attention","evidence":"README，Why 部分：\"Each bot can use a cloud Linux desktop, an isolated Local VM, or—where the platform safety boundary is currently certified—your own computer\"。能力表显示 macOS 上 \"Bot control of this computer: Supported\"，而 Ubuntu Wayland 上是 \"Disabled: Wayland safety gate\"（issue #345）。","source":"ai","title":"机器人可以控制你自己的电脑"},{"code":"api_keys_stored_as_plain_text","severity":"attention","evidence":"README，Run from a terminal or on a server：\"API keys are saved as plaintext, not encrypted, in private config.json (0600 on Unix).\"","source":"ai","title":"API 密钥以明文形式保存"},{"code":"bots_create_scheduled_tasks_without_asking","severity":"attention","evidence":"README，Triggers：一个机器人 \"sets the trigger up in that turn with its trigger tool (no confirmation; the chat shows a live card with Edit, Pause and Undo)\"，也就是说一条聊天消息就能创建定时任务或 webhook。","source":"ai","title":"机器人未经询问就创建定时任务"},{"code":"optional_public_access_from_the_internet","severity":"attention","evidence":"README：webhook 接收端监听在 127.0.0.1:8800，\"For public internet delivery, proxy only this dedicated receiver through a hosted relay or a tool such as Tailscale Funnel\"；服务器模式还提供 \"--tunnel ... for a managed public address\"。","source":"ai","title":"可选择从互联网公开访问"},{"code":"part_of_the_code_is_not_open_source","severity":"info","evidence":"README，License：Apache 2.0 \"except enterprise/, which is source-available under its own license; delete that folder and what remains is the open-source edition\"。","source":"ai","title":"部分代码不是开源的"},{"code":"coins_use_the_project_name","severity":"info","evidence":"README 警告：\"No affiliation with any cryptocurrency. OpenMausBot has no token. Any coin using the OpenMausBot, Maus, or SupaMaus name is not created, endorsed, or affiliated with this project or its maintainer.\"","source":"ai","title":"有代币借用了项目的名字"},{"code":"usage_analytics_included","severity":"info","evidence":"README，Privacy：\"Usage analytics are optional and anonymous: product events and a once-a-day count of what is set up\"，可在 Settings → General 中关闭；package.json 中列出了 posthog-js 分析库。","source":"ai","title":"内置使用情况分析"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","docker"],"disk":{"minMB":248,"maxMB":744,"basis":"installer"},"apiKeys":["An account for at least one assistant: Claude, Codex (ChatGPT) or Grok","Composio (to connect Gmail, Slack, GitHub and other apps)","Boat (cloud computer for the bots)","ElevenLabs","Fish Audio","xAI (spoken replies)","OpenRouter or TypeSafe Jev (the decision model)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","productivity","automation"],"alternativeTo":[{"slug":"grok-bot","name":"Grok Bot","confirmed":true},{"slug":"manus","name":"Manus","confirmed":true},{"slug":"meta-muse","name":"Muse","confirmed":true},{"slug":"openai-dots","name":"dots","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[{"target":"invideo-ai","stance":"claims","evidence":"The open-source Grok Bot alternative: your own team of AI bots, in a chat app"},{"target":"benzinga-pro","stance":"claims","evidence":"GPT only, Pro and Business Premium plans | Several agents; your Claude, ChatGPT or Grok login; no plan required | [OpenAI dots alternative](https://www.openmausbot.com/blog/openai-dots-alternative) |"},{"target":"grok-bot","stance":"claims","evidence":"The open-source Grok Bot alternative: your own team of AI bots, in a chat app"},{"target":"manus","stance":"claims","evidence":"is an open-source Grok Bot alternative: an independent project inspired by Grok Bot, and the open-source alternative to Meta Muse, OpenAI dots and Cue by Manus — bring-your-own-agent, local-first, on"},{"target":"dbeaver-pro","stance":"claims","evidence":"GPT only, Pro and Business Premium plans | Several agents; your Claude, ChatGPT or Grok login; no plan required | [OpenAI dots alternative](https://www.openmausbot.com/blog/openai-dots-alternative) |"},{"target":"meta-muse","stance":"claims","evidence":"is an open-source Grok Bot alternative: an independent project inspired by Grok Bot, and the open-source alternative to Meta Muse, OpenAI dots and Cue by Manus — bring-your-own-agent, local-first, on"},{"target":"openai-dots","stance":"claims","evidence":"is an open-source Grok Bot alternative: an independent project inspired by Grok Bot, and the open-source alternative to Meta Muse, OpenAI dots and Cue by Manus — bring-your-own-agent, local-first, on"},{"target":"browser-use-cloud","stance":"claims","evidence":"cloud VM, US and Canada only | A team instead of one agent; runs anywhere; data in `~/.openmausbot` | [Meta Muse alternative](https://www.openmausbot.com/blog/open-source-alternative-to-meta-muse) |"},{"target":"moz-pro","stance":"claims","evidence":"GPT only, Pro and Business Premium plans | Several agents; your Claude, ChatGPT or Grok login; no plan required | [OpenAI dots alternative](https://www.openmausbot.com/blog/openai-dots-alternative) |"},{"target":"n8n-cloud","stance":"claims","evidence":"cloud VM, US and Canada only | A team instead of one agent; runs anywhere; data in `~/.openmausbot` | [Meta Muse alternative](https://www.openmausbot.com/blog/open-source-alternative-to-meta-muse) |"},{"target":"deepl-pro","stance":"claims","evidence":"GPT only, Pro and Business Premium plans | Several agents; your Claude, ChatGPT or Grok login; no plan required | [OpenAI dots alternative](https://www.openmausbot.com/blog/openai-dots-alternative) |"}],"stars":20579,"downloads":1638773,"downloadsWeek":594457,"downloadsDays":7,"growth":8888,"growthDays":7,"history":[{"date":"2026-10-30","stars":8655},{"date":"2026-10-31","stars":9630},{"date":"2026-11-01","stars":10603},{"date":"2026-11-02","stars":11225},{"date":"2026-11-03","stars":11691},{"date":"2026-11-04","stars":11992},{"date":"2026-11-05","stars":13162},{"date":"2026-11-06","stars":15053},{"date":"2026-11-07","stars":16843},{"date":"2026-11-08","stars":18103},{"date":"2026-11-09","stars":19563},{"date":"2026-11-10","stars":20579}],"firstPublishedAt":"2026-10-30T01:57:54.362Z","commitSha":"d58ef29f8e1e5f1a1a710f88a3ab3a68d92d2b59","verifiedAt":"2026-11-03T02:38:31.390Z","lastCommitAt":"2026-11-10T22:35:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A chat app where each contact is an AI helper with its own personality, model and computer to work on.","summary":"OpenMausBot looks like a messaging app, but every contact in the list is an AI helper. Each one has its own personality, its own model and its own computer (a rented Linux desktop, an isolated machine on your PC, or your own computer after you allow it), and can be connected to Gmail, Slack, GitHub and hundreds of other apps. It asks before running commands. Everything is kept in a folder on your machine. You need at least one coding assistant (Claude, Codex or Grok) installed and signed in.","easeRationale":"macOS is the smoothest platform: a signed and notarised .dmg you drag to Applications, with the server already inside. The grade is C everywhere because before the bots work you still have to install at least one assistant CLI (claude, codex or grok) in the terminal and sign in to it. Windows has a ready installer but it is not signed yet, and the Ubuntu .deb and AppImage are still called beta.","costNote":"The app is free and Apache-2.0. The AI itself comes from your own Claude, Codex or Grok account, and the README says the Boat cloud computer \"is a paid service after its trial, and using a cloud computer may incur charges\". There is also a paid hosted Cloud with higher limits, and the enterprise/ folder has its own licence.","notVerified":["Whether the bots really manage to use the cloud or local computers as the screenshots show.","Whether the approval cards catch every risky action a bot takes before it runs.","What a month with the Boat cloud computer, Composio and the voice services ends up costing.","How much disk the app uses after the bots download their browsers and files (the installers alone are 248-744 MB).","Whether the pinned undici version with the reported weaknesses is reachable in normal use.","Whether the Ubuntu beta behaves as described, since the project itself marks several features as beta or disabled there."]},"zh-Hans":{"summaryShort":"一个聊天应用，列表里的每个联系人都是有自己性格、模型和电脑的 AI 助手。","summary":"OpenMausBot 看起来像一个聊天软件，但联系人列表里的每一个联系人都是一个 AI 助手。每个助手都有自己的性格、自己的模型和自己的电脑（一台租用的 Linux 桌面、你电脑上的一台隔离机器，或者在你允许之后使用你自己的电脑），还可以连接 Gmail、Slack、GitHub 等数百个应用。它会在执行命令前先征得同意。所有内容都保存在你电脑上的一个文件夹里。你至少需要安装并登录一个编程助手（Claude、Codex 或 Grok）。","easeRationale":"macOS 是体验最顺畅的平台：一个已签名并通过苹果公证的 .dmg 文件，拖到 Applications 里即可，服务端已经包含在内。各平台评级都是 C，因为在机器人工作之前，你仍需在终端里安装至少一个助手 CLI（claude、codex 或 grok）并登录。Windows 有现成的安装程序，但还没有签名，Ubuntu 的 .deb 和 AppImage 仍被标注为 beta。","costNote":"该应用免费，采用 Apache-2.0 许可。AI 本身来自你自己的 Claude、Codex 或 Grok 账户，README 提到 Boat 云端电脑 \"is a paid service after its trial, and using a cloud computer may incur charges\"。此外还有限额更高的付费托管 Cloud，enterprise/ 文件夹有自己单独的许可证。","notVerified":["机器人是否真的能像截图那样使用云端或本地电脑。","审批卡是否能在机器人执行每一个有风险的操作之前都拦截住。","使用 Boat 云端电脑、Composio 和语音服务一个月下来到底要花多少钱。","机器人下载浏览器和文件之后，应用实际占用多少磁盘空间（光是安装程序就有 248-744 MB）。","带有已报告缺陷的固定版本 undici 在正常使用中是否会被触及。","Ubuntu 版 beta 是否像描述那样表现，因为项目自己就把其中几个功能标为 beta 或已禁用。"]}}},{"id":"R_kgDOVZ9Lng","slug":"minimax-ai__mcode-stack","name":"MiniMax Code Stack","owner":"MiniMax-AI","fullName":"MiniMax-AI/mcode-stack","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"brand_new_no_releases","severity":"info","evidence":"项目信息：创建于 2026-10-28，最近一次推送为 2026-10-29，GitHub 上有 0 个发布版本（release），1 个未关闭的议题（issue）。README 还写道 \"Everything here will keep changing as we learn\"，因此 skills 可能在更新之间发生变动或消失。","source":"ai","title":"全新项目，尚无发布版本"},{"code":"outside_contributions_restricted","severity":"info","evidence":"README 的 Contributing 部分：\"For now, pull requests for code, documentation, and stacks are accepted only from repository collaborators.\" 来自公司外部的修复需要先通过 issue 提出。","source":"ai","title":"外部贡献受限"},{"code":"mixed_licenses_inside","severity":"info","evidence":"README 的 License 部分：\"MIT, except where a stack's NOTICE.md says otherwise. The engineering-* skills are adapted from Google Engineering Practices; the adapted material stays under CC-BY 3.0.\"","source":"ai","title":"项目内部混合多种许可证"},{"code":"written_for_the_makers_own_assistant","severity":"info","evidence":"该仓库属于 MiniMax-AI，这些 skills \"install straight into MiniMax Code\"；README 还补充说 \"The skills use the common SKILL.md format, so most of them also work in other agents that support skills\"，也就是大多数，而不是全部。","source":"ai","title":"为作者自家的助手而写"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":671,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":155,"growthDays":7,"history":[{"date":"2026-10-30","stars":462},{"date":"2026-10-31","stars":484},{"date":"2026-11-01","stars":496},{"date":"2026-11-02","stars":513},{"date":"2026-11-03","stars":516},{"date":"2026-11-04","stars":535},{"date":"2026-11-05","stars":540},{"date":"2026-11-06","stars":544},{"date":"2026-11-07","stars":545},{"date":"2026-11-08","stars":547},{"date":"2026-11-09","stars":583},{"date":"2026-11-10","stars":671}],"firstPublishedAt":"2026-10-30T01:57:54.369Z","commitSha":"49a3ff4dfd1aa159c8a84579f3ac0ce38e1aaa6d","verifiedAt":"2026-10-30T01:43:37.663Z","lastCommitAt":"2026-11-10T09:52:45Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A pack of ready-made instructions that make an AI coding assistant plan, review and check its own work the way one team does.","summary":"The MiniMax Code team published the skills and checklists it uses day to day with its own AI coding assistant. Each one is a plain-text file the assistant follows: split a feature into small pull requests, decide when to hand work to a second agent, audit a long shell command, keep a long session from drifting, review a change from a clean slate, write a hand-off note. It installs with one command in MiniMax Code, Claude Code or Codex, and the files use the common SKILL.md format, so other assistants that read skills can use most of them. For programmers who already use one.","easeRationale":"You need an AI coding assistant already installed on any of the three systems, then one command, for example `claude plugin install ronny --marketplace MiniMax-AI/mcode-stack` or `codex plugin marketplace add MiniMax-AI/mcode-stack`. The manual route is a `git clone` plus copying a folder into `~/.agents/skills/`, which the README writes with shell commands aimed at Mac and Linux.","costNote":"The skills are free (MIT, with some parts under CC-BY 3.0). They only do something inside an AI coding assistant, which usually needs a paid plan or paid API use.","notVerified":["Whether these habits actually improve what your assistant produces on your own project","Which of the skills work unchanged in assistants other than MiniMax Code, Claude Code and Codex","How much extra AI usage the longer review and audit workflows consume","Whether the install commands behave the same on Windows, where the README's copy steps are written for a Unix shell"]},"zh-Hans":{"summaryShort":"一套现成的指令,让AI编程助手像某个团队那样,自己规划、审查并检查自己的工作。","summary":"MiniMax Code团队把自己日常使用的技能和检查清单发布出来,这些是他们自己的AI编程助手每天在用的东西。每一个都是一个纯文本文件,助手会照着执行:把一个功能拆成多个小的拉取请求、决定什么时候把工作交给第二个代理、审查一条很长的shell命令、防止长会话跑偏、从零开始审查一次改动、写一份交接说明。它可以在MiniMax Code、Claude Code或Codex里用一个命令安装,这些文件使用通用的SKILL.md格式,所以其他能读取技能的助手大多也能用。适合已经在用其中一种助手的程序员。","easeRationale":"你需要先在三个系统中的任意一个上装好AI编程助手,然后输入一个命令,例如`claude plugin install ronny --marketplace MiniMax-AI/mcode-stack`,或`codex plugin marketplace add MiniMax-AI/mcode-stack`。手动方式是先`git clone`,再把一个文件夹复制到`~/.agents/skills/`,README里写的这些命令是针对Mac和Linux的shell命令。","costNote":"这些技能是免费的(MIT协议,部分内容采用CC-BY 3.0协议)。它们只能在AI编程助手内部起作用,而这类助手通常需要付费方案或付费的API使用。","notVerified":["这些习惯是否真的能改善你自己项目中助手产出的结果","在MiniMax Code、Claude Code和Codex之外的助手里,哪些技能能原样生效","更长的审查和审计流程会额外消耗多少AI用量","安装命令在Windows上是否表现一致,因为README里的复制步骤是针对Unix shell写的"]}}},{"id":"R_kgDOVhdgEQ","slug":"minireel__hghub","name":"HGhub","owner":"Minireel","fullName":"Minireel/HGhub","type":"content_list","status":"alert","grade":"E","easeByPlatform":{},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"7 天内获得 8104 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"young_owner_account","severity":"attention","evidence":"作者账号创建于 57 天前","source":"check","title":"作者账号刚注册不久"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 目录和 winget 清单中都没有发现任何许可证","source":"check","title":"没有许可证"},{"code":"discontinued_and_emptied","severity":"attention","evidence":"README：\"我决定出于个人原因停止维护，并撤下本仓库中的代码、版本发布与构建记录，仅留下这份说明\"。根目录文件：只有 README.md。GitHub 发布数：0。","source":"ai","title":"已停止维护并清空"},{"code":"code_gone_but_thousands_of_forks_remain","severity":"attention","evidence":"项目数据：Forks: 3175，Stars: 8104，而仓库本身现在只剩 README.md。被删除的代码副本可能仍在各个分支和别处流传，而这些副本都无法凭本材料核对。","source":"ai","title":"代码已删，但仍留有数千个分支"},{"code":"third_party_client_for_another_service","severity":"attention","evidence":"仓库描述：\"YellowFruit third-party client\"；标签：android、drama、free、movie-web-app、reels、windows。视频服务的非官方客户端在一些地方受到限制或不被允许，而代码已被删除，无法查看它当初连接了什么、播放了什么。","source":"ai","title":"别家服务的第三方客户端"},{"code":"unofficial_client_for_a_video_service","severity":"high","evidence":"仓库描述：\"YellowFruit third-party client\"；标签：android、drama、free、movie-web-app、reels、windows。README：\"我决定出于个人原因停止维护，并撤下本仓库中的代码、版本发布与构建记录，仅留下这份说明\"，根目录文件只有 README.md，检查结果为 \"license\": {\"spdx\": null}。视频服务的条款限制非官方客户端，许多国家也有关于未经授权播放影视内容的规定；代码既已删除，就无法查看它连接了什么、播放了什么。","source":"ai","title":"某视频服务的非官方客户端"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["video"],"alternativeTo":[],"declaredAlternatives":[],"stars":8082,"downloads":0,"downloadsWeek":0,"downloadsDays":0,"growth":8082,"growthDays":7,"history":[{"date":"2026-11-05","stars":1094},{"date":"2026-11-06","stars":1580},{"date":"2026-11-07","stars":2226},{"date":"2026-11-08","stars":3173},{"date":"2026-11-09","stars":5695},{"date":"2026-11-10","stars":8082}],"firstPublishedAt":"2026-11-04T22:32:55.039Z","commitSha":"cae4645ddfdd234dd9a5e90191007dac812360d9","verifiedAt":"2026-11-10T23:55:53.966Z","lastCommitAt":"2026-11-10T13:41:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Discontinued: the author removed the code, the releases and the build records, leaving only a goodbye note.","summary":"The owner described HGhub as a third-party client for a service called YellowFruit, with topics such as drama, movie-web-app and reels. None of that is in the repository any more. The README says the author decided, for personal reasons, to stop maintaining the project and to take down the code, the releases and the build records, leaving only this note, and adds that he is fine and was not acted against by any upstream party or institution. The only top-level file is README.md, there are no GitHub releases and no licence, so there is nothing to download or install.","easeRationale":"There is nothing to install on any platform: the only file in the repository is README.md, the project facts show 0 GitHub releases and the check found no platforms from releases, so no grade can be given for Windows, Mac, Linux or web.","costNote":"Nothing is offered for sale or for download; the repository has no licence.","notVerified":["What the program actually did, since all of its code was removed from the repository","Whether copies offered in forks or on other sites are the same program the author published","Whether the reason given in the README is the whole story, as no other evidence is in the package","Which platforms the removed builds ran on and what they needed to work"]},"zh-Hans":{"summaryShort":"项目已停止：作者删除了代码、发布版本和构建记录，只留下一份告别说明。","summary":"作者曾把 HGhub 描述为某个名为 YellowFruit 的服务的第三方客户端，标签包括 drama、movie-web-app 和 reels。这些内容如今都已不在仓库里。README 写道，作者出于个人原因决定停止维护项目，并撤下代码、发布版本和构建记录，只留下这份说明，还提到自己一切安好，没有受到任何上游方或机构的处置。仓库根目录下只剩 README.md，GitHub 上没有发布版本，也没有许可证，因此没有任何可下载或安装的东西。","easeRationale":"在任何平台上都没有东西可装：仓库里唯一的文件是 README.md，项目数据显示 GitHub 发布数为 0，检查也没有从发布中识别出平台，因此 Windows、Mac、Linux 和网页版都无法给出等级。","costNote":"没有任何出售或可下载的内容；仓库没有许可证。","notVerified":["这个程序到底做了什么，因为所有代码都已从仓库中删除","分支或其他网站上提供的副本，是否就是作者当初发布的同一个程序","README 中给出的理由是否就是全部情况，因为材料里没有其他证据","被删除的构建支持哪些平台，运行时又需要什么条件"]}}},{"id":"R_kgDORWiFDA","slug":"mksglu__context-mode","name":"Context Mode","owner":"mksglu","fullName":"mksglu/context-mode","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"sandbox_does_not_isolate_your_computer","severity":"attention","evidence":"项目的风险说明写道:\"Code from ctx_execute runs on your host as your user, with your files, your network and your environment except the preload variables, so keep your agent's own permission prompts on. Here \\\"sandbox\\\" means the output stays in the sandbox and out of the context.\" 描述里的sandbox(沙盒)一词,并不代表代码与你的文件是隔离的。","source":"ai","title":"沙盒并不能隔离你的电脑"},{"code":"source_available_license_not_open_source","severity":"attention","evidence":"LICENSE文件是Elastic License 2.0:\"You may not provide the software to third parties as a hosted or managed service\"以及\"You may not move, change, disable, or circumvent the license key functionality in the software\"。GitHub把这份许可标为NOASSERTION,所以它不是标准的开源许可。","source":"ai","title":"源代码可见许可,并非开源"},{"code":"plugin_intercepts_and_blocks_assistant_tool_calls","severity":"info","evidence":"工作原理第3、4步:PreToolUse钩子会对WebFetch、curl、wget、大段读取和shell输出\"checks the call\",而且\"the hook stops the call and its reply names the exact ctx_* call to use\"。它会把钩子、一个MCP服务器和一些skills写进每个助手的配置目录。","source":"ai","title":"插件会拦截并阻止助手的工具调用"},{"code":"paid_hosted_service_routes_your_requests","severity":"info","evidence":"Gateway章节写道:\"With the Gateway on, your requests pass through its server; the plugin keeps its own data on your machine.\" 免费插件是本地运行的,但可选的付费Gateway并非如此。","source":"ai","title":"付费托管服务会转发你的请求"},{"code":"savings_numbers_are_self_measured","severity":"info","evidence":"所有类似\"99.4%\"这样的数字都来自项目自己的跑分(\"median of 5 runs\"、\"rows with 10 KB or more of raw output\"、bench/results/2026-11-01-darwin-arm64.json)。仓库里写了如何重新跑一遍,但没有包含任何独立的测量。","source":"ai","title":"节省比例的数字是项目自己测出来的"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":33808,"downloads":242,"downloadsWeek":136,"downloadsDays":5,"growth":339,"growthDays":6,"history":[{"date":"2026-11-05","stars":33554},{"date":"2026-11-06","stars":33604},{"date":"2026-11-07","stars":33647},{"date":"2026-11-08","stars":33698},{"date":"2026-11-09","stars":33758},{"date":"2026-11-10","stars":33808}],"firstPublishedAt":"2026-11-07T22:56:16.883Z","commitSha":"ee0a1f6058f6abdb5af70b05f825a997bb15615d","verifiedAt":"2026-11-07T22:43:02.743Z","lastCommitAt":"2026-11-10T03:23:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Add-on for AI coding assistants that keeps huge command and page output out of the chat.","summary":"Context Mode is a plugin for ten AI coding assistants (Claude Code, Codex, Gemini CLI, Copilot CLI, OpenCode and others). When the assistant reads a big file, runs a command or fetches a page, the plugin runs that work on your machine, stores the whole output in a local SQLite database, and hands the assistant only the part it asks for, so less of your context window is used. It also brings a session back after the chat is compacted. You install Node 24 or newer and run one command per assistant; everything stays on your computer.","easeRationale":"On macOS and Linux you install Node 24.15.0 or newer and run one or two commands, for example \"npm i -g context-mode\" then \"context-mode setup\", or your assistant's own plugin command; then you restart the assistant and run /ctx-doctor. Windows is the same plus Git for Windows, because the README says \"the shell tools run in Git Bash\". You also need one of the ten supported AI coding assistants already working.","costNote":"The plugin is described as \"free forever\" under a source-available license. The same team sells a separate hosted product, Context Mode Gateway, \"with its own account and subscription\".","notVerified":["How much context you actually save on your own work, as opposed to the project's bench rows","Whether the hooks ever block a tool call your assistant needed and change how it behaves","How large the local databases under ~/.context-mode grow over months","Whether it works with the exact version of your assistant, since the agents change often (236 releases so far)","What the Gateway subscription costs and what it keeps, if you add it"]},"zh-Hans":{"summaryShort":"面向AI编程助手的插件,把命令和网页的大量输出挡在聊天之外。","summary":"Context Mode 是一款面向十款AI编程助手(Claude Code、Codex、Gemini CLI、Copilot CLI、OpenCode等)的插件。当助手读取大文件、运行命令或抓取网页时,插件会在你的电脑上完成这项工作,把全部输出存进本地的SQLite数据库,只把助手要的那一部分交给它,这样占用的上下文窗口更少。聊天被压缩后,它还能把会话找回来。你只需安装Node 24或更新版本,给每个助手运行一条命令;所有数据都留在你的电脑上。","easeRationale":"在macOS和Linux上,你安装Node 24.15.0或更新版本,运行一两条命令,比如先\"npm i -g context-mode\"再\"context-mode setup\",或者用你的助手自带的插件安装命令;然后重启助手并运行/ctx-doctor。Windows做法相同,还要装Git for Windows,因为README写道\"the shell tools run in Git Bash\"。你还需要已经装好十款支持的AI编程助手中的一款。","costNote":"该插件被称为\"free forever\"(永久免费),使用的是源代码可见但非开源的许可。同一团队另外出售一款托管产品Context Mode Gateway,\"with its own account and subscription\"。","notVerified":["你自己的工作实际能省下多少上下文,而不是项目自己跑分的那些数据行","这些钩子是否会挡住你的助手本来需要的某次工具调用,从而改变它的行为","~/.context-mode下的本地数据库几个月后会变多大","它是否适配你助手的确切版本,因为这些智能体经常更新(到目前已有236个版本)","如果你开通Gateway订阅,费用是多少,它又会保留什么数据"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0MTE4Nzc2","slug":"mltframework__shotcut","name":"Shotcut","owner":"mltframework","fullName":"mltframework/shotcut","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"unsigned_mac_file_next_to_the_normal_one","severity":"info","evidence":"发布版 v26.10.30 同时提供 shotcut-macos-26.10.30.dmg（mac，201 MB）和 shotcut-macos-unsigned-26.10.30.dmg（mac，285 MB）；v26.9.6 也是如此。材料没有说明 \"unsigned\" 版本的用途，也没有说明两者有何区别。","source":"ai","title":"Mac 未签名文件与正常文件并列"},{"code":"files_whose_version_does_not_match_the_release","severity":"info","evidence":"发布版 v26.9.27（2026-09-28）里列出的是 shotcut-linux-x86_64-26.10.3.AppImage 和 shotcut-macos-26.9.28.dmg，发布版 v26.9.6（2026-09-06）里列出的是 shotcut-win64-26.10.5.zip。这些文件名中的版本号与所属标签并不一致，很容易下到与页面标题不符的另一个构建。","source":"ai","title":"文件版本号与发布版对不上"},{"code":"readme_points_downloads_to_another_site","severity":"info","evidence":"README 的 \"Install\"：\"Binaries are regularly built and are available at https://www.shotcut.org/download/.\" 该页面提供的内容在仓库之外，不属于本次材料。","source":"ai","title":"README 把下载指向另一个网站"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":135,"maxMB":406,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["video"],"alternativeTo":[{"slug":"premiere-pro","name":"Premiere Pro","confirmed":true},{"slug":"final-cut-pro","name":"Final Cut Pro","confirmed":true},{"slug":"camtasia","name":"Camtasia","confirmed":true}],"declaredAlternatives":[],"stars":15651,"downloads":22311968,"downloadsWeek":37143,"downloadsDays":7,"growth":62,"growthDays":7,"history":[{"date":"2026-10-25","stars":15502},{"date":"2026-10-26","stars":15501},{"date":"2026-10-27","stars":15515},{"date":"2026-10-28","stars":15528},{"date":"2026-10-29","stars":15537},{"date":"2026-10-30","stars":15541},{"date":"2026-10-31","stars":15545},{"date":"2026-11-01","stars":15562},{"date":"2026-11-02","stars":15581},{"date":"2026-11-03","stars":15589},{"date":"2026-11-04","stars":15597},{"date":"2026-11-05","stars":15612},{"date":"2026-11-06","stars":15622},{"date":"2026-11-07","stars":15630},{"date":"2026-11-08","stars":15632},{"date":"2026-11-09","stars":15642},{"date":"2026-11-10","stars":15651}],"firstPublishedAt":"2026-10-25T23:19:12.541Z","commitSha":"2f85cdff3c67fe364e806dca7f97d1a7a32018d0","verifiedAt":"2026-11-10T23:57:04.446Z","lastCommitAt":"2026-11-05T02:12:57Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"shotcut-win64-26.9.27.exe","sizeBytes":152926384,"url":"https://github.com/mltframework/shotcut/releases/download/v26.9.27/shotcut-win64-26.9.27.exe","releaseUrl":"https://github.com/mltframework/shotcut/releases/tag/v26.9.27"},{"os":"mac","file":"shotcut-macos-26.9.28.dmg","sizeBytes":210616938,"url":"https://github.com/mltframework/shotcut/releases/download/v26.9.27/shotcut-macos-26.9.28.dmg","releaseUrl":"https://github.com/mltframework/shotcut/releases/tag/v26.9.27"},{"os":"linux","file":"shotcut-linux-x86_64-26.10.3.AppImage","sizeBytes":216024256,"url":"https://github.com/mltframework/shotcut/releases/download/v26.9.27/shotcut-linux-x86_64-26.10.3.AppImage","releaseUrl":"https://github.com/mltframework/shotcut/releases/tag/v26.9.27"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free video editor for Windows, Mac and Linux, with ready installers you download and open.","summary":"Shotcut is a video editor you install and use like any other program: cut and join clips, add effects and export the result. The owner describes it as a cross-platform, open-source (GPLv3) video editor, and it has been developed since 2012. Every recent release brings ready files for the three systems, including an installer for Windows, a .dmg for Mac and an AppImage for Linux, so nothing else has to be installed to run it. The README sends people to the project's own download page, and warns that building it from source is only for beta testers and contributors.","easeRationale":"On Windows the latest release, v26.10.30 (2026-10-31), has a ready installer, shotcut-win64-26.10.30.exe of 135 MB, so you download and open it. The same release has shotcut-macos-26.10.30.dmg (201 MB) for Mac and shotcut-linux-x86_64-26.10.30.AppImage (206 MB) for Linux, both ready to run as well.","costNote":"GPLv3; the package mentions no paid version or paid tier.","notVerified":["Whether the installers really work and open on each version of Windows, Mac and Linux","What the \"unsigned\" Mac file contains and how it differs from the normal .dmg","How much disk the program takes after installing, beyond the 135 MB to 406 MB measured on the release files","Whether the files on shotcut.org/download are the same ones published in the GitHub releases","How well it performs when editing or exporting long or high-resolution videos on a normal computer"]},"zh-Hans":{"summaryShort":"一款面向 Windows、Mac 和 Linux 的免费视频剪辑软件，下载现成安装包即可使用。","summary":"Shotcut 是一款像普通软件一样安装使用的视频剪辑工具：剪切拼接片段、添加特效、导出成片。项目方把它描述为跨平台的开源（GPLv3）视频编辑器，自 2012 年起持续开发。最近的每个发布版都为三大系统提供现成文件，包括 Windows 安装程序、Mac 的 .dmg 和 Linux 的 AppImage，无需再安装其他东西即可运行。README 会把用户引导到项目自己的下载页，并提醒从源码编译只适合 Beta 测试者和贡献者。","easeRationale":"在 Windows 上，最新发布版 v26.10.30（2026-10-31）提供现成的安装程序 shotcut-win64-26.10.30.exe，大小 135 MB，下载打开即可。同一发布版还提供 Mac 的 shotcut-macos-26.10.30.dmg（201 MB）和 Linux 的 shotcut-linux-x86_64-26.10.30.AppImage（206 MB），同样可以直接运行。","costNote":"采用 GPLv3；材料中没有提到任何付费版本或付费档位。","notVerified":["这些安装包在每个 Windows、Mac 和 Linux 版本上是否都能正常安装并打开","Mac 的 \"unsigned\" 文件里装的是什么，和普通 .dmg 有什么区别","安装之后程序占用多少磁盘空间，发布文件本身测得的是 135 MB 到 406 MB","shotcut.org/download 上提供的文件，是否与 GitHub 发布页上的文件相同","在普通电脑上剪辑或导出长视频、高分辨率视频时表现如何"]}}},{"id":"R_kgDOVjiSkg","slug":"mmdoctor__google-pro-helper","name":"Google Pro Helper","owner":"mmdoctor","fullName":"mmdoctor/google-pro-helper","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"circumvents_paid_plan_billing_to_get_it_free","severity":"high","evidence":"README.md 直接写明了目标：\"免费激活 Google AI Pro 12 / 24 个月 | offer token 注入\"，控制台的步骤把\"首层 12 个月免费包\"、\"二层 6 个月\"和\"三层 6 个月\"叠加起来，以 0 美元凑够 24 个月。诱导服务商的结账流程免费发放付费套餐违反了 Google 的条款，也可能被视为通过欺骗手段获取服务。","source":"ai","title":"绕过付费套餐的扣费来免费获取"},{"code":"mass_account_automation_with_passwords_and_2fa_secrets","severity":"high","evidence":"README.md 里要求按\"每行一个\"粘贴账号，格式为\"邮箱----密码----辅助邮箱----2FA密钥\"，然后\"批量全自动执行\"会逐一登录。作者还写道\"我本人大部分号是用的是注册机批量搞的\"，并提醒\"不要用大号\"。把明文的 Google 密码和 2FA 密钥交给脚本，还一次操作大量账号，容易导致账号被封，这正是账号盗用工具的典型做法。","source":"ai","title":"批量账号自动化操作，附带密码和 2fa 密钥"},{"code":"card_number_and_cvc_typed_into_local_console","severity":"high","evidence":"README.md 要求填写卡号（16 位）、有效期、\"安全码（3位 CVC）\"、持卡人姓名，以及一个免税的美国邮编\"97201（美国俄勒冈免税州）\"，这些都会保存到本地配置文件里。把完整的卡信息加 CVC 存在硬盘上，还专门选一个能避开销售税的美国邮编，既暴露了支付数据，账单地址也是假的。","source":"ai","title":"卡号和 cvc 被输入本地控制台"},{"code":"zero_charge_claim_not_guaranteed_by_the_project","severity":"attention","evidence":"README.md 自己也收回了这个承诺：\"原说明中的'全程 0 扣费'不能作为保证，实际费用和权益以结账页面及上游结果为准\"，扩展程序部分还提醒\"极大概率需要付款首月的youtube会员\"。你的卡被真实扣费是有可能的。","source":"ai","title":"零扣费的说法并非项目的保证"},{"code":"brand_new_repository_with_no_license_and_many_forks","severity":"info","evidence":"项目信息：创建于 2026-11-05，最后一次 push 也是 2026-11-05，没有 release，却已经有 852 个 star 和 331 个 fork，也没有许可证文件。这类滥用促销的仓库通常很快会被下架，而这些 fork 意味着被改动过的副本已经在流传。","source":"ai","title":"全新仓库没有许可证却有大量 fork"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":1630,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1630,"growthDays":5,"history":[{"date":"2026-11-07","stars":1129},{"date":"2026-11-08","stars":1235},{"date":"2026-11-09","stars":1379},{"date":"2026-11-10","stars":1630}],"firstPublishedAt":"2026-11-06T22:57:16.747Z","commitSha":"618eae07fb0d752733542bce35d3e46df3338699","verifiedAt":"2026-11-06T22:38:04.275Z","lastCommitAt":"2026-11-05T09:01:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Chrome extension and Windows console that chain Google's free offers to switch on Google AI Pro for up to 24 months.","summary":"Two separate tools in one repository, both aimed at the same thing: turning on Google AI Pro (the paid Gemini plan with 5 TB of storage) for 12 to 24 months without paying. One is a Chrome extension you load by hand that injects offer tokens while you click through one.google.com/plans. The other is a Windows console where you paste lists of Google accounts (email, password, backup email, 2FA key) plus one card number, and it logs in and stacks the free offers on its own. Needs Chrome; the console also needs Windows, Python 3.10 or newer and a US proxy. The README is in Chinese.","easeRationale":"Nothing here has an installer. On Windows you get both paths: load the extension folder in Chrome with developer mode on, or install Python 3.10 or newer, set up a US proxy and run two .bat files to open a console at 127.0.0.1:8765. On Mac and Linux only the Chrome extension applies, which still means turning on developer mode, loading an unpacked folder and following a long manual checkout.","costNote":"The code is free, but a real Visa or Mastercard is required. The README says the card needs about 1 dollar available for the check and warns that an attached YouTube membership will be charged next month if you forget to cancel.","notVerified":["Whether the pasted account passwords, 2FA keys and card details stay on your computer or are sent somewhere, since the code was not run","Whether the offer stacking still works, as Google can close the gap at any time","Whether your card really stays at zero, and what the attached YouTube membership costs if you miss the cancel step","Whether Google suspends the accounts or the cards used this way","What the Chrome extension actually injects into the checkout page"]},"zh-Hans":{"summaryShort":"Chrome 扩展程序和 Windows 控制台通过串联 Google 的免费优惠，开通长达 24 个月的 Google AI Pro。","summary":"同一个仓库里有两个独立的工具，目标相同：在不付费的情况下开通 Google AI Pro（带 5 TB 存储空间的付费 Gemini 套餐）12 到 24 个月。一个是需要手动加载的 Chrome 扩展程序，在你点击 one.google.com/plans 的过程中注入优惠令牌。另一个是 Windows 控制台，你在里面粘贴一批 Google 账号（邮箱、密码、备用邮箱、2FA 密钥）再加一个卡号，它就会自己登录并叠加免费优惠。需要 Chrome；控制台还需要 Windows、Python 3.10 或更新版本，以及一个美国代理。README 是中文的。","easeRationale":"这里的东西都没有安装程序。在 Windows 上有两条路：在开启开发者模式的 Chrome 中加载扩展程序文件夹，或者安装 Python 3.10 或更新版本，配置一个美国代理，然后运行两个 .bat 文件打开一个位于 127.0.0.1:8765 的控制台。在 Mac 和 Linux 上只能用 Chrome 扩展程序，这同样需要开启开发者模式、加载一个未打包的文件夹，并完成一个很长的手动操作流程。","costNote":"代码是免费的，但需要一张真实的 Visa 或 Mastercard 卡。README 说这张卡需要有大约 1 美元余额用于验证，并提醒如果忘记取消，绑定的 YouTube 会员会在下个月扣费。","notVerified":["粘贴进去的账号密码、2FA 密钥和卡信息是留在你的电脑上，还是被发送到了别处，因为代码并没有被运行过","叠加优惠的做法是否还能用，因为 Google 随时可能堵上这个漏洞","你的卡是否真的不会被扣费，以及如果错过取消步骤，绑定的 YouTube 会员要付多少钱","Google 是否会封禁用这种方式操作的账号或银行卡","这个 Chrome 扩展程序到底往结账页面里注入了什么"]}}},{"id":"R_kgDOOQVw5Q","slug":"mobile-next__mobile-mcp","name":"Mobile MCP","owner":"mobile-next","fullName":"mobile-next/mobile-mcp","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：fast-xml-parser: GHSA-474x-hrwr-fgx4","source":"check","title":"存在已知漏洞"},{"code":"telemetry_on_by_default","severity":"attention","evidence":"README 的 Telemetry 一节写道：\"Mobile MCP collects anonymous usage telemetry via PostHog and Scarf\"，只有设置 MOBILEMCP_DISABLE_TELEMETRY=1 才会关闭。同一份 README 的 Privacy Policy 一节却说它 \"runs locally and communicates only with the devices you connect\"。","source":"ai","title":"默认开启数据上报"},{"code":"ai_gets_full_control_of_the_phone","severity":"attention","evidence":"工具列表包含 mobile_install_app、mobile_uninstall_app、mobile_clear_app_data、mobile_type_keys、mobile_clipboard（\"Read or replace the device clipboard\"）、mobile_set_location（\"Override the GPS location\"）、mobile_get_device_logs 和 mobile_start_screen_recording。在装有你账号的真实手机上，助手可以以你的身份行事。","source":"ai","title":"AI 获得手机的完全控制权"},{"code":"example_prompts_automate_app_store_ratings_and_messages","severity":"attention","evidence":"README 里的示例提示词让智能体 \"go back to the app store and rate the app 5 stars, and leave a comment how useful the app is\"、\"click on like video, after liking write a comment\"，还会给联系人发 WhatsApp 消息。在别家公司的应用里自动刷评分、评论和消息，通常违反其服务条款。","source":"ai","title":"示例提示词会自动刷应用商店评分和消息"},{"code":"http_mode_allows_unauthenticated_connections","severity":"attention","evidence":"README：使用 --listen 时，\"When [MOBILEMCP_AUTH is] unset, the server accepts unauthenticated connections and logs a warning\"，示例还监听在 0.0.0.0:3000。这样暴露后，任何能访问该端口的人都能操控所连设备。","source":"ai","title":"HTTP 模式允许无认证连接"},{"code":"version_numbers_do_not_match","severity":"info","evidence":"package.json 写的是 \"version\": \"0.0.1\"，而 GitHub 上的发布版本是 1.0.14、1.0.15 和 1.0.16。npm 安装命令都用 @latest，因此装到的是最后发布的那一版。","source":"ai","title":"版本号对不上"}],"platforms":["windows","mac","linux"],"tools":["node","docker","other"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Mobile Next Cloud account (remote devices)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","ai_agents","development"],"alternativeTo":[{"slug":"browserstack-app-automate","name":"App Automate","confirmed":true}],"declaredAlternatives":[],"stars":10476,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":181,"growthDays":6,"history":[{"date":"2026-11-07","stars":10382},{"date":"2026-11-08","stars":10406},{"date":"2026-11-09","stars":10438},{"date":"2026-11-10","stars":10476}],"firstPublishedAt":"2026-11-07T22:56:16.891Z","commitSha":"3b51952bcd83955f3a743f731f5a9beaebecdab7","verifiedAt":"2026-11-10T23:29:33.743Z","lastCommitAt":"2026-11-07T08:39:19Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant use a phone: tap, type, install apps and read the screen on iOS and Android.","summary":"Mobile MCP is a connector (MCP server) that gives AI assistants such as Claude Code, Codex, Cursor or Copilot control of iOS and Android devices: simulators, emulators or a real phone plugged in by USB. The assistant can list the elements on screen, tap, swipe, type, press Home or Back, install and remove apps, record the screen, read device logs and crash reports, change the GPS location and the clipboard. It is aimed at app testing and automation. You install Node 20 or newer plus Xcode or the Android platform tools.","easeRationale":"On macOS, where both platforms work, you install Node 20 or newer and Xcode's command line tools, boot a simulator, then run one command such as \"npx @mobilenext/mobile-mcp@latest install\" or your assistant's own install command. Windows and Linux are the same amount of work but only reach Android, since they need the Android SDK and adb; real phones also need USB debugging enabled and the device trusted.","costNote":"The code is Apache-2.0. Renting real phones through the company's separate Mobile Next Cloud service is a paid product.","notVerified":["Whether each tool works on your exact phone, simulator or Android version","What the telemetry sends and what the company keeps, since the policy is only on their website","How reliable the assistant is at long multi-step flows without breaking something on the device","What Mobile Next Cloud costs if you rent real devices","Whether automating another app's reviews, likes or messages gets your account restricted"]},"zh-Hans":{"summaryShort":"让AI助手操作手机:点按、输入、安装应用,读取iOS和Android上的屏幕内容。","summary":"Mobile MCP是一个连接器(MCP服务器),让Claude Code、Codex、Cursor或Copilot这类AI助手可以控制iOS和Android设备:iOS模拟器、Android模拟器,或者用USB连接的真机。助手可以列出屏幕上的元素、点按、滑动、输入文字、按Home键或返回键、安装和卸载应用、录屏、读取设备日志和崩溃报告、修改GPS位置和剪贴板内容。它面向应用测试和自动化场景。你需要安装Node 20或更新版本,再加上Xcode或Android平台工具。","easeRationale":"在macOS上(两个平台都能用),你安装Node 20或更新版本和Xcode命令行工具,启动一个模拟器,然后运行一条命令,比如\"npx @mobilenext/mobile-mcp@latest install\",或者用你助手自带的安装命令。Windows和Linux的工作量差不多,但只能用于Android,因为它们需要Android SDK和adb;真机还需要打开USB调试并在设备上授权。","costNote":"代码采用Apache-2.0许可。通过该公司另外提供的Mobile Next Cloud服务租用真机,是一项付费产品。","notVerified":["每个工具在你具体的手机、模拟器或Android版本上是否都能用","遥测数据发送了什么、公司保留了什么,因为相关政策只写在他们的网站上","助手在长串多步操作中有多可靠,会不会在设备上把什么东西弄坏","如果你租用真机,Mobile Next Cloud的费用是多少","自动操作别的应用的评价、点赞或消息,会不会让你的账号被限制"]}}},{"id":"R_kgDOM7KEng","slug":"modelcontextprotocol__inspector","name":"MCP Inspector","owner":"modelcontextprotocol","fullName":"modelcontextprotocol/inspector","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"secrets_can_be_stored_unencrypted","severity":"attention","evidence":"README警告：\"The Inspector manages secrets — OAuth tokens, OAuth client secrets, and stdio `env:` values — and stores them in the OS keychain, if available, by default. On a machine with no keychain — Linux without libsecret or a Secret Service, headless and SSH sessions, Termux, and containers with a mounted volume — they are saved to `~/.mcp-inspector/secrets.json` instead, unencrypted unless you supply a key.\"","source":"ai","title":"密钥可能未加密保存"},{"code":"docker_image_opens_the_backend_to_the_network","severity":"attention","evidence":"Dockerfile：`ENV HOST=0.0.0.0 DANGEROUSLY_BIND_ALL_INTERFACES=true`，注释写着\"Binding 0.0.0.0 is refused by default (it exposes the process-spawning backend to the network); a container is the sanctioned exception\"。这个后端会在机器上启动程序，所以如果把6274、6275或6278端口开放到你自己电脑之外，就等于把这个权力交给了任何能连到这些端口的人。","source":"ai","title":"docker镜像让后端暴露在网络上"},{"code":"mixed_licensing_in_the_same_repository","severity":"info","evidence":"LICENSE文件写道：\"The MCP project is undergoing a licensing transition from the MIT License to the Apache License, Version 2.0\"，以及\"Contributions made by authors who originally licensed their work under the MIT License and who have not yet granted explicit permission to relicense remain licensed under the MIT License\"；文档部分采用CC-BY-4.0许可。GitHub本身也显示该仓库没有单一的许可证。两种许可证都比较宽松，但代码并不统一属于其中一种。","source":"ai","title":"同一个仓库里混用多种许可证"},{"code":"two_versions_maintained_in_parallel","severity":"info","evidence":"README中仓库状态部分写道：开发工作在`v2/main`上进行，而\"The legacy v1 line lives on v1/main — security fixes only, published straight from that branch to the npm v1-latest tag\"。发布列表显示2.14.0和1.0.4是同一天发布的，所以你拿到的版本取决于你请求的npm标签。","source":"ai","title":"同时维护两个版本"},{"code":"written_mostly_by_ai_agents","severity":"info","evidence":"仓库在docs/ai-software-factory.md中记录了自己的流程，标题是\"MCP Inspector: Our AI Software Factory — How contributions actually happen since v2.0.0\"，AGENTS.md被称为\"the contract for changing this codebase... it applies to humans and AI agents alike\"，还在.claude/skills/中提供了代理技能。README还写道\"External contributions are accepted as issues, not pull requests\"。项目用每个文件90%的覆盖率门槛和一次强制的推送前检查来平衡这一点。","source":"ai","title":"大部分代码由ai代理编写"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["development","ai_skills"],"alternativeTo":[{"slug":"postman","name":"Postman","confirmed":true},{"slug":"insomnia","name":"Insomnia","confirmed":true}],"declaredAlternatives":[],"stars":11272,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":42,"growthDays":6,"history":[{"date":"2026-11-06","stars":11248},{"date":"2026-11-07","stars":11251},{"date":"2026-11-08","stars":11255},{"date":"2026-11-09","stars":11262},{"date":"2026-11-10","stars":11272}],"firstPublishedAt":"2026-11-08T23:01:18.761Z","commitSha":"4d3464cb3a6dde609d7b5ac24c440b947eb40f23","verifiedAt":"2026-11-08T22:51:34.479Z","lastCommitAt":"2026-11-04T22:56:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Connects to an MCP server and lets you try its tools by hand, in a browser page, a terminal screen or scripts.","summary":"MCP Inspector is the official tool for testing Model Context Protocol servers, the add-ons that give assistants like Claude Code or Cursor new tools. One command starts it in four shapes: a web page, a terminal interface, a scriptable command line for CI, and an experimental connection tool. You connect to a server, list what it offers, call a tool with your own parameters and read the answer, including OAuth sign-in. It is for people building or checking MCP servers: you need Node 22 or newer and the terminal; a Docker image is also published.","easeRationale":"Same on all three systems: install a recent Node (the README requires ^22.23.4, ^24.22.1 or >=26.12.1) and run one command, `npx @modelcontextprotocol/inspector`, which opens the web interface. There is no installer and no binary in the 75 releases, and you still need an MCP server of your own to point it at.","notVerified":["Whether a keychain is found on your machine, and therefore whether your OAuth tokens end up in a plain file","Whether the published npm package and Docker image are built from the commit analyzed here","How the experimental mcpdo daemon behaves when it is left running","Real disk use after npx downloads the package and its dependencies","Whether the web interface works in browsers other than the Chromium and Firefox the project smoke-tests"]},"zh-Hans":{"summaryShort":"连接到一个MCP服务器，让你在浏览器页面、终端界面或脚本里手动试用它的工具。","summary":"MCP Inspector是测试Model Context Protocol服务器的官方工具，这些服务器是给Claude Code或Cursor这类助手增加新工具的插件。一条命令就能以四种形式启动它：一个网页、一个终端界面、一个可用于CI的可编程命令行，以及一个实验性的连接工具。你连接到一个服务器，列出它提供的内容，用自己的参数调用某个工具并读取返回结果，也包括OAuth登录。它是给构建或检查MCP服务器的人用的：你需要Node 22或更新版本，以及终端；项目也发布了Docker镜像。","easeRationale":"三个系统上的做法相同：安装较新的Node（README要求^22.23.4、^24.22.1或>=26.12.1），运行一条命令`npx @modelcontextprotocol/inspector`，就会打开网页界面。75个发布版本里没有安装程序也没有二进制文件，而且你仍然需要自己的MCP服务器才能让它连上去。","notVerified":["你的机器上是否能找到系统密钥管理工具（keychain），从而决定你的OAuth令牌是否会落到一个普通文件里","发布的npm包和Docker镜像是否是基于这里分析的提交构建的","实验性的mcpdo守护进程长时间运行时表现如何","npx下载这个包及其依赖后，实际占用了多少磁盘空间","网页界面在项目冒烟测试所用的Chromium和Firefox之外的浏览器上是否能正常工作"]}}},{"id":"R_kgDOKdFnTQ","slug":"modelscope__funcodec","name":"FunCodec","owner":"modelscope","fullName":"modelscope/FunCodec","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"packages_installed_over_plain_http","severity":"attention","evidence":"requirements.txt只有一行：`--editable ./ -i  http://mirrors.aliyun.com/pypi/simple --trusted-host mirrors.aliyun.com`。它让pip指向一个未加密（http）的包镜像，并关闭了对该镜像的证书校验，所以安装到的内容取决于这个镜像以及中间的网络环境。","source":"ai","title":"通过未加密的http安装包"},{"code":"no_updates_since_january_2024","severity":"attention","evidence":"最近一次提交是2024-01-05（本次检查前1038天），而且完全没有发布版本，README却仍然写着\"This project is still working on progress\"，还有26个issue处于开放状态。依赖列表要求`torch>=1.12`，并锁定了`typeguard==2.13.3`，这类锁定版本正是容易在当前Python和PyTorch版本下出问题的那种。","source":"ai","title":"自2024年1月起没有更新"},{"code":"imitates_a_voice_from_a_sample","severity":"attention","evidence":"仓库的标签里包含\"voicecloning\"，README介绍\"LauraTTS... a powerful codec-based zero-shot text-to-speech synthesizer, which outperforms VALL-E in terms of semantic consistency and speaker similarity\"。Zero-shot意味着只要一段某人的简短录音，就能让一个新声音说出任何内容，这可能侵犯声音和形象权。","source":"ai","title":"从样本模仿一个声音"},{"code":"models_downloaded_from_outside_the_repository","severity":"info","evidence":"模型权重不在仓库里：`bash encoding_decoding.sh --stage 0 --model_name ${model_name} --model_hub modelscope`（或`huggingface`）会从alibaba-damo在ModelScope或Hugging Face上的页面获取它们。那些模型页面可能有自己的条款，并不是仓库的MIT许可证。","source":"ai","title":"模型要从仓库之外下载"},{"code":"license_statements_do_not_match","severity":"info","evidence":"LICENSE文件和setup.py中的`license=\"The MIT License\"`都写着MIT，但setup.py的classifiers里却列出了\"License :: OSI Approved :: Apache Software License\"。README还补充说项目\"contains various third-party components and some code modified from other repos under other open source licenses\"（Kaldi、ESPnet、EnCodec）。","source":"ai","title":"许可证声明互相矛盾"}],"platforms":["linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":449,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":449},{"date":"2026-11-09","stars":449},{"date":"2026-11-10","stars":449}],"firstPublishedAt":"2026-11-08T23:01:18.768Z","commitSha":"3435d870126a48fe61a87fac1af3b6530be388ea","verifiedAt":"2026-11-08T22:52:22.441Z","lastCommitAt":"2024-01-05T02:56:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A research toolkit that turns speech into compact codes and back into sound, and can train text-to-speech models.","summary":"FunCodec, from Alibaba's speech lab, is a research toolkit for neural audio codecs: it compresses speech into discrete codes at bitrates from 250 to 16000 bits per second and rebuilds the sound from them. Six pretrained models are published on Hugging Face and ModelScope, and the project also ships recipes for LauraTTS, a text-to-speech model that imitates a voice from a short sample. Everything runs through shell recipes on Linux with Python and one or more GPUs; there is no application, no interface and no release package.","easeRationale":"This is for researchers on Linux: you clone the repository, run `pip install --editable ./`, then drive stage-based shell scripts such as `bash encoding_decoding.sh --stage 1 --gpu_devices \"0,1\"` with Kaldi-style `wav.scp` file lists. The setup.py declares Linux only, there is no graphical or single-command path, and training needs several GPUs.","costNote":"MIT code and freely downloadable model weights; the cost is the GPU time.","notVerified":["Whether the install still completes with current Python and PyTorch versions after almost three years without commits","How much disk and video memory the models need in practice, for encoding and for training","Whether the pretrained models still download from ModelScope and Hugging Face","Whether the quality claims against VALL-E hold outside the authors' own tests","What the always-installed Alibaba Cloud storage client (oss2) and the training-time wandb package send out"]},"zh-Hans":{"summaryShort":"一个研究工具包，把语音压缩成紧凑的编码，再还原成声音，还能训练文字转语音模型。","summary":"FunCodec来自阿里巴巴语音实验室，是一个面向神经音频编解码器的研究工具包：它把语音压缩成每秒250到16000比特的离散编码，再从编码还原出声音。Hugging Face和ModelScope上发布了六个预训练模型，项目还提供了LauraTTS的训练脚本，这是一个能从一小段样本模仿声音的文字转语音模型。一切都通过Linux上的shell脚本运行，需要Python和一块或多块GPU；没有应用程序，没有界面，也没有发布包。","easeRationale":"这是给Linux上的研究者用的：你克隆仓库，运行`pip install --editable ./`，然后用Kaldi风格的`wav.scp`文件列表，驱动按阶段划分的shell脚本，例如`bash encoding_decoding.sh --stage 1 --gpu_devices \"0,1\"`。setup.py声明只支持Linux，没有图形化或一键式的途径，训练需要多块GPU。","costNote":"代码是MIT许可，模型权重可以免费下载；成本只是GPU时间。","notVerified":["在将近三年没有提交的情况下，用当前的Python和PyTorch版本，安装是否还能顺利完成","模型在实际使用中（编码和训练）到底需要多少磁盘空间和显存","预训练模型是否仍然能从ModelScope和Hugging Face下载","相对VALL-E的质量宣称，在作者自己的测试之外是否依然成立","始终会安装的阿里云存储客户端（oss2）和训练时用到的wandb包，会往外发送什么数据"]}}},{"id":"R_kgDORw4Jjw","slug":"moonfin-client__moonfin-core","name":"Moonfin","owner":"Moonfin-Client","fullName":"Moonfin-Client/Moonfin-Core","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A","web":"C"},"risk":"low","signals":[{"code":"firebase_services_bundled","severity":"info","evidence":"顶层文件列表中有 `firebase.json` 和 `PRIVACY_POLICY.md`，说明该应用接入了 Google Firebase 服务；README 中没有说明具体发送了什么内容。","source":"ai","title":"内置 Firebase 服务"},{"code":"needs_own_media_server","severity":"info","evidence":"README 中的 \"Supported Servers\" 表格要求使用 Jellyfin 10.8.0+、Emby 4.8+、Plex 1.43+ 或 Silo 服务器；若干功能（网页 PWA、主题、设置同步、动画标记）还需要在该服务器上单独安装 Moonbase 插件。","source":"ai","title":"需要自备媒体服务器"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":[],"disk":{"minMB":78,"maxMB":234,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"plex","name":"Plex","confirmed":true},{"slug":"emby","name":"Emby","confirmed":true},{"slug":"infuse","name":"Infuse","confirmed":true},{"slug":"tivimate","name":"TiviMate","confirmed":false}],"declaredAlternatives":[],"stars":986,"downloads":99982,"downloadsWeek":2716,"downloadsDays":4,"growth":30,"growthDays":6,"history":[{"date":"2026-11-06","stars":970},{"date":"2026-11-07","stars":979},{"date":"2026-11-08","stars":985},{"date":"2026-11-09","stars":987},{"date":"2026-11-10","stars":986}],"firstPublishedAt":"2026-11-08T23:01:18.777Z","commitSha":"9418bc6db97e863b4cd12b63d4f1dbbbd68de622","verifiedAt":"2026-11-08T22:42:40.589Z","lastCommitAt":"2026-11-10T21:27:25Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Moonfin_Windows_v3.0.0.exe","sizeBytes":81755272,"url":"https://github.com/Moonfin-Client/Moonfin-Core/releases/download/3.0.0/Moonfin_Windows_v3.0.0.exe","releaseUrl":"https://github.com/Moonfin-Client/Moonfin-Core/releases/tag/3.0.0"},{"os":"mac","file":"Moonfin_macOS_v3.0.0.dmg","sizeBytes":159228223,"url":"https://github.com/Moonfin-Client/Moonfin-Core/releases/download/3.0.0/Moonfin_macOS_v3.0.0.dmg","releaseUrl":"https://github.com/Moonfin-Client/Moonfin-Core/releases/tag/3.0.0"},{"os":"linux","file":"Moonfin_Linux_v3.0.0.AppImage","sizeBytes":180324856,"url":"https://github.com/Moonfin-Client/Moonfin-Core/releases/download/3.0.0/Moonfin_Linux_v3.0.0.AppImage","releaseUrl":"https://github.com/Moonfin-Client/Moonfin-Core/releases/tag/3.0.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Watch your own Jellyfin, Emby or Plex library on a computer, phone or TV, with ready installers.","summary":"Moonfin is an app that plays the films, series, music, books and live TV that are already on your own media server. It connects to Jellyfin, Emby, Plex or Silo servers, and has downloads for offline viewing, casting, parental controls with a PIN, and an admin panel for Jellyfin and Emby. There are ready installers for Windows, macOS and Linux, plus store versions for Android, Android TV, Fire TV, iPhone and Apple TV. You need your own media server already running; the app alone does not provide any content.","easeRationale":"On Windows the release has a ready .exe installer (78 MB) and there is also a WinGet command, so it is download and open; macOS has a .dmg and Linux has AppImage, deb, rpm, snap and flatpak files. The browser version is harder because, as the README says, it is only served as a PWA after the Moonbase plugin is installed in the media server.","costNote":"Free and open source (GPL-2.0); the author accepts donations through Buy Me a Coffee.","notVerified":["Whether playback, HDR and audio passthrough really work on each platform and each type of graphics card","What the bundled Firebase configuration collects and sends","How much disk space the app uses after offline downloads and retro game saves pile up","Whether the unsigned iOS and Apple TV files can be installed without a paid Apple developer account","Whether the Plex and Silo support is as complete as the README's tables claim"]},"zh-Hans":{"summaryShort":"在电脑、手机或电视上观看你自己的 Jellyfin、Emby 或 Plex 媒体库，提供现成的安装程序。","summary":"Moonfin 是一款应用，用来播放你自己媒体服务器上已有的电影、剧集、音乐、图书和直播电视。它可以连接 Jellyfin、Emby、Plex 或 Silo 服务器，支持离线下载、投屏、带 PIN 码的家长控制，以及 Jellyfin 和 Emby 的管理面板。提供适用于 Windows、macOS 和 Linux 的现成安装程序，以及 Android、Android TV、Fire TV、iPhone 和 Apple TV 的商店版本。你需要已经运行着自己的媒体服务器；应用本身不提供任何内容。","easeRationale":"在 Windows 上，发布版本带有现成的 .exe 安装程序（78 MB），也有 WinGet 命令，所以只需下载并打开；macOS 有 .dmg 文件，Linux 有 AppImage、deb、rpm、snap 和 flatpak 文件。浏览器版本更难，因为正如 README 所说，只有在媒体服务器上安装了 Moonbase 插件之后，才会以 PWA 的形式提供。","costNote":"免费且开源（GPL-2.0）；作者通过 Buy Me a Coffee 接受捐赠。","notVerified":["播放、HDR 和音频直通是否在每个平台和每种显卡上都真的能用","内置的 Firebase 配置到底收集并发送了什么","离线下载和复古游戏存档积累后，应用占用的磁盘空间有多大","未签名的 iOS 和 Apple TV 文件是否可以在没有付费 Apple 开发者账户的情况下安装","对 Plex 和 Silo 的支持是否真的像 README 表格里说的那样完整"]}}},{"id":"R_kgDONj_QTw","slug":"morettt__my-neuro","name":"my-neuro","owner":"morettt","fullName":"morettt/my-neuro","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"E"},"risk":"attention","signals":[{"code":"setup_instructions_offsite","severity":"attention","evidence":"README 把整个安装流程指向了仓库之外：\"本项目部署流程请看官网：[点我进官网](http://mynewbot.com/tutorials)\"，所以安装程序下载的内容是在 GitHub 之外决定的。","source":"ai","title":"安装说明在站外"},{"code":"downloads_code_at_runtime","severity":"attention","evidence":"顶层的 `installer.py`、`update.py` 和 `plugins-dlc/`，以及 requirements.txt 中的 `requests` 和 `pyinstaller`，说明该应用在你启动它之后会获取并安装额外的部分。","source":"ai","title":"运行时下载代码"},{"code":"controls_your_computer","severity":"attention","evidence":"功能列表中有 \"桌面控制：支持语音控制打开软件等操作\"，requirements.txt 还安装了 `pyautogui` 和一个 \"屏幕点击工具\"，所以这个 AI 能看到屏幕，并替你点击和输入。","source":"ai","title":"控制你的电脑"},{"code":"copies_another_creator_character","severity":"info","evidence":"README 把内置的角色描述为 \"肥牛（fake neuro）它是一个抄袭neuro sama的角色\" —— 一个抄袭自主播 Neuro-sama 的角色 —— 仓库的名字也来源于它。","source":"ai","title":"照搬了另一位创作者的角色"},{"code":"voice_cloning_included","severity":"info","evidence":"功能列表中包括声音模型的训练 (\"声音模型（TTS）训练支持，默认使用gpt-sovits开源项目\")，因此可以从录音中复制真人的声音。","source":"ai","title":"包含声音克隆功能"}],"platforms":["windows","linux"],"tools":["python","node","gpu_nvidia"],"buildTools":["python","node","gpu_nvidia"],"disk":{"minMB":85,"maxMB":254,"basis":"installer"},"apiKeys":["OpenAI-compatible language model API (only if you choose a closed online model instead of a local one)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","local_ai","audio"],"alternativeTo":[{"slug":"character-ai","name":"Character.AI","confirmed":true},{"slug":"replika","name":"Replika","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true}],"declaredAlternatives":[],"stars":1443,"downloads":27844,"downloadsWeek":82,"downloadsDays":3,"growth":15,"growthDays":6,"history":[{"date":"2026-11-07","stars":1435},{"date":"2026-11-08","stars":1436},{"date":"2026-11-09","stars":1438},{"date":"2026-11-10","stars":1443}],"firstPublishedAt":"2026-11-08T23:01:18.784Z","commitSha":"c9ec73f814920965554558d2f4f9ab1b59ca7bc3","verifiedAt":"2026-11-08T22:43:36.666Z","lastCommitAt":"2026-11-07T09:33:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Build a talking anime-style desktop companion that listens, answers in about a second and remembers you.","summary":"my-neuro puts an animated Live2D character on your desktop that you talk to by voice or keyboard. Everything can run on your own computer: speech recognition, the language model, the voice (trained with GPT-SoVITS), long-term memory, image recognition of your screen, and extras like singing, live streaming on Bilibili, playing games with you and opening programs when you ask. You need Windows with an NVIDIA graphics card, Python and Node, and you follow the steps on the author's site. The main README is in Chinese.","easeRationale":"Windows is the best-supported system and the one the packaged installer targets, but it still needs Python, Node and an NVIDIA card, and `requirements.txt` says the CUDA version of torch must be installed by hand (\"已注释，请手动安装CUDA版本\"), plus a row of .bat files to start speech, voice and memory services. Linux is harder: `docs/deploy/linux.md` is for running from source only.","costNote":"Free and MIT licensed; the author takes donations on Afdian. A paid online model API is optional, since the project also runs local models.","notVerified":["Whether replies really arrive in under one second, and on which graphics cards","How much video memory and disk space the local speech, voice and language models need once downloaded","What the installer on mynewbot.com and the plugins-dlc updates actually download and run","Whether the Android app, the web panel and the game features work as the list claims","How much a closed online model would cost if you use one instead of a local model"]},"zh-Hans":{"summaryShort":"打造一个会说话的动漫风格桌面伙伴，它能听你说话，约一秒内回应，还能记住你。","summary":"my-neuro 会在你的桌面上放一个 Live2D 动画角色，你可以用语音或键盘和它对话。一切都可以在你自己的电脑上运行：语音识别、语言模型、声音（用 GPT-SoVITS 训练）、长期记忆、对屏幕内容的图像识别，以及唱歌、在 Bilibili 直播、陪你玩游戏、按你的要求打开程序等额外功能。你需要配备 NVIDIA 显卡的 Windows、Python 和 Node，并按作者网站上的步骤操作。主要的 README 是中文的。","easeRationale":"Windows 是支持最好的系统，打包的安装程序也是针对它的，但仍然需要 Python、Node 和 NVIDIA 显卡，`requirements.txt` 中说 CUDA 版本的 torch 需要手动安装（\"已注释，请手动安装CUDA版本\"），还要用一系列 .bat 文件来启动语音识别、声音和记忆服务。Linux 更难：`docs/deploy/linux.md` 只支持从源码运行。","costNote":"免费且采用 MIT 许可证；作者在爱发电（Afdian）接受捐赠。付费的在线模型 API 是可选的，因为该项目也可以运行本地模型。","notVerified":["回复是否真的能在一秒以内送达，以及在哪些显卡上能做到","下载完成后，本地语音、声音和语言模型需要多少显存和磁盘空间","mynewbot.com 上的安装程序和 plugins-dlc 的更新实际下载并运行了什么","安卓应用、网页面板和游戏功能是否真的像列表里说的那样可用","如果使用闭源在线模型而不是本地模型，费用会是多少"]}}},{"id":"R_kgDOVUPmBg","slug":"morganlinton__beautifulrust","name":"BeautifulRust","owner":"morganlinton","fullName":"morganlinton/BeautifulRust","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"low","signals":[{"code":"benchmark_single_run_self_graded","severity":"info","evidence":"README 中的跨智能体基准测试：\"it was one run per configuration, graded by a Claude model\"，以及 \"They hurt in one case: Claude did worse on the Bevy feature task with them.\"","source":"ai","title":"基准测试只跑了一次，且由 AI 自行打分"},{"code":"sources_summarized_under_mixed_licenses","severity":"info","evidence":"README 的署名表：skill 中附带的五份摘要改写了条款各不相同的资料，其中包括采用 \"CC BY-NC-ND 4.0 (online edition)\" 的 Effective Rust，以及采用 MPL-2.0 的 Rust Design Patterns，而仓库本身采用 MIT。","source":"ai","title":"摘要所依据的资料许可证不一"},{"code":"skill_runs_commands_on_your_code","severity":"info","evidence":"README：rust-review skill \"runs `cargo fmt`, `clippy`, and the tests when a toolchain is available\"，port skill \"verifies with `cargo test` and `clippy`\"，也就是说智能体会在你的项目里运行构建命令。","source":"ai","title":"该 skill 会在你的代码上运行命令"}],"platforms":["windows","mac","linux","web"],"tools":["rust","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Coding agent account (Claude Code, Codex, GitHub Copilot, Cursor or Gemini CLI)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":1778,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":83,"growthDays":7,"history":[{"date":"2026-10-27","stars":1310},{"date":"2026-10-28","stars":1366},{"date":"2026-10-29","stars":1415},{"date":"2026-10-30","stars":1513},{"date":"2026-10-31","stars":1619},{"date":"2026-11-01","stars":1645},{"date":"2026-11-02","stars":1682},{"date":"2026-11-03","stars":1695},{"date":"2026-11-04","stars":1710},{"date":"2026-11-05","stars":1728},{"date":"2026-11-06","stars":1745},{"date":"2026-11-07","stars":1754},{"date":"2026-11-08","stars":1763},{"date":"2026-11-09","stars":1771},{"date":"2026-11-10","stars":1778}],"firstPublishedAt":"2026-10-27T18:40:27.972Z","commitSha":"58757ee890da9b3c8d5b785adc7a8d53cce27307","verifiedAt":"2026-10-27T18:36:00.332Z","lastCommitAt":"2026-11-08T17:21:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Add-on rules that teach AI coding assistants to write, review and port Rust the way seasoned Rust developers do.","summary":"Beautiful Rust is three Agent Skills plus a clippy configuration. Installed into a coding agent such as Claude Code, Codex, Copilot, Cursor or Gemini CLI, they steer it toward idiomatic Rust when it writes code, make it produce a prioritized review of existing Rust where each finding cites the rule it breaks, and guide it when porting C, C++, Python, Go or JavaScript to Rust. The rules are distilled from five published Rust guides, and the lint file enforces the machine-checkable ones in CI. It is for Rust developers who already use a coding agent and have cargo and clippy installed.","easeRationale":"Installing is one command such as npx skills add morganlinton/BeautifulRust or claude plugin install, but you need a coding agent set up first and a Rust toolchain with cargo and clippy for the review and the lint file to do anything.","costNote":"The skills and the lint config are MIT; the work is done by a coding agent you pay for, and the README notes that running the eval cases means \"real model runs\" that \"are billed\".","notVerified":["Whether the Rust it produces for your own project is better, since the published comparison is one run per configuration.","How well it works in agents other than Claude Code and Codex, the two the benchmark covers.","How many tokens the roughly 500-line writing skill and its reference digests add to each session.","Whether the clippy table fits an existing codebase without a long list of allows."]},"zh-Hans":{"summaryShort":"一套附加规则,教 AI 编程助手像资深 Rust 开发者那样编写、审查和移植 Rust 代码。","summary":"Beautiful Rust 由三个 Agent Skill 和一份 clippy 配置组成。装进 Claude Code、Codex、Copilot、Cursor 或 Gemini CLI 之类的编程智能体后,它们会在写代码时引导智能体写出符合惯例的 Rust,让它对已有 Rust 代码给出按优先级排序的审查结果,每条意见都标明违反了哪条规则,并在把 C、C++、Python、Go 或 JavaScript 移植到 Rust 时提供指导。这些规则是从五份已发布的 Rust 指南中提炼出来的,其中能用机器检查的部分由 lint 文件在 CI 里强制执行。它面向已经在用编程智能体、并装好了 cargo 和 clippy 的 Rust 开发者。","easeRationale":"安装只需一条命令,比如 npx skills add morganlinton/BeautifulRust 或 claude plugin install,但你需要先配置好一个编程智能体,并装好带 cargo 和 clippy 的 Rust 工具链,审查功能和 lint 文件才能发挥作用。","costNote":"这些 skill 和 lint 配置是 MIT 许可;实际工作由你付费使用的编程智能体完成,README 提到运行评测用例意味着“真实的模型调用”,而这些调用“会产生费用”。","notVerified":["它为你自己的项目产出的 Rust 代码是否真的更好,因为公开的对比里每种配置只跑了一次。","在 Claude Code 和 Codex 之外的其他智能体里效果如何,基准测试只覆盖了这两个。","约 500 行的写作 skill 以及它的参考摘要,会给每次会话增加多少 token 消耗。","clippy 规则表套用到已有代码库时,是否需要大量的例外(allow)才能用。"]}}},{"id":"R_kgDOSB6hVQ","slug":"morluto__rea","name":"REA","owner":"morluto","fullName":"morluto/rea","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"197 天内获得 397502 个星标（约每天 2018 个）","source":"check","title":"星标增长速度异常"},{"code":"reverse_engineering_may_be_restricted","severity":"attention","evidence":"README 的 Disclaimer 部分写道，REA 'is for lawful research, education, interoperability, software preservation, and analysis of software you own or are authorized to examine'，并指向 DISCLAIMER.md 了解负责任的使用方式。拆解不属于你的软件可能违反其许可条款，在一些国家还受到限制。","source":"ai","title":"逆向工程可能受到限制"},{"code":"runs_and_interacts_with_target_apps","severity":"attention","evidence":"README.md：'Runtime capture runs or interacts with the selected target using your user permissions; each runtime guide describes its effects.' 它还可以操控已连接的 Android 设备（通过 adb 处理进程、日志、屏幕截图以及 APK/文件传输）。","source":"ai","title":"会运行目标应用并与之交互"},{"code":"security_testing_tools_bundled","severity":"info","evidence":"package.json 附带了 pwntools（bridge/pwntools/）、pwndbg（bridge/pwndbg/launch.py）和 mitmproxy 流量捕获（bridge/mitmproxy/capture.py）的桥接文件；README 记录了崩溃与内存防护机制的检查，以及读取已保存的网络抓包。这些都是常见的漏洞研究工具；体积较大的需要由你自行提供，REA 不会安装。","source":"ai","title":"附带漏洞测试工具"},{"code":"readme_warns_about_copycat_repositories","severity":"info","evidence":"README.md 开头有一个 IMPORTANT 提示框：'A similar repository name or copied README does not establish that a download is official'，并列出 morluto/rea、rea.tools 和 npm 包 rea-agents 为仅有的官方来源。建议与检查中标出的异常星标增长速度一起参考。","source":"ai","title":"README 提醒提防仿冒仓库"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_skills","ai_agents"],"alternativeTo":[{"slug":"ida-pro","name":"IDA Pro","confirmed":true},{"slug":"hopper","name":"Hopper","confirmed":true}],"declaredAlternatives":[],"stars":449675,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":22677,"growthDays":7,"history":[{"date":"2026-10-25","stars":385133},{"date":"2026-10-26","stars":391703},{"date":"2026-10-27","stars":397432},{"date":"2026-10-28","stars":402591},{"date":"2026-10-29","stars":407905},{"date":"2026-10-30","stars":412365},{"date":"2026-10-31","stars":415556},{"date":"2026-11-01","stars":419035},{"date":"2026-11-02","stars":422963},{"date":"2026-11-03","stars":426998},{"date":"2026-11-04","stars":430489},{"date":"2026-11-05","stars":433989},{"date":"2026-11-06","stars":437650},{"date":"2026-11-07","stars":440698},{"date":"2026-11-08","stars":443426},{"date":"2026-11-09","stars":446599},{"date":"2026-11-10","stars":449675}],"firstPublishedAt":"2026-10-25T23:19:13.414Z","commitSha":"40b9c03f996270e944ff07cafb63be15ce903708","verifiedAt":"2026-10-28T01:37:51.228Z","lastCommitAt":"2026-11-10T22:39:53Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI agent, or you in a terminal, take apart programs you own and explain how one of their features works.","summary":"REA is a command-line tool and MCP server that connects an AI agent (Claude Code, Codex, Cursor, Gemini CLI and others) to reverse-engineering tools. You point it at a program on your own computer and ask how a feature works; it inspects native binaries, JavaScript and Electron apps, .NET assemblies, Android apps, firmware and saved network captures, and returns findings with the evidence behind them. The inspection runs on your machine. You need Node.js 22 or newer and npm; deep work on native binaries also needs Hopper, Ghidra or IDA installed.","easeRationale":"On Mac and Linux the one-line install.sh plus `npx rea-agents setup` configure it, but you first need Node.js 22.x, 24.x or 26.12+ and npm, then a separate decompiler (Hopper, Ghidra or IDA) for native binaries, and extra command-line tools per target (a procps `ps` and pwntools on Linux, a full JDK and JADX for Android, tshark for packet files). On Windows the shell installer refuses to run, so you install the npm package by hand and the README calls Ghidra support there experimental.","costNote":"The tool itself is free (MIT). Deep analysis of native binaries needs a separate decompiler: Ghidra is free, Hopper, IDA Pro and JEB are paid and you must have your own licence. The AI agent you drive it with may also be paid.","notVerified":["Whether the findings and recovered pseudocode are correct for a given program","Which features really work on Windows, where the installer is not supported and native analysis is marked experimental","What the AI agent you connect sends to its own model provider while it analyses your files","How much disk and time an analysis session takes on a large binary","What `rea update` and `npx rea-agents@latest` fetch and change on your machine at each run"]},"zh-Hans":{"summaryShort":"让 AI 智能体,或者你自己在终端中,拆解你拥有的程序,解释其中某个功能的工作原理。","summary":"REA 是一个命令行工具和 MCP 服务器,把 AI 智能体(Claude Code、Codex、Cursor、Gemini CLI 等)连接到逆向工程工具上。你把它指向自己电脑上的一个程序,询问某个功能是如何实现的;它会检查原生二进制文件、JavaScript 和 Electron 应用、.NET 程序集、安卓应用、固件以及保存下来的网络抓包文件,并返回带有证据支持的分析结果。检查过程在你自己的电脑上运行。你需要 Node.js 22 或更高版本以及 npm;对原生二进制文件的深入分析还需要安装 Hopper、Ghidra 或 IDA。","easeRationale":"在 Mac 和 Linux 上,一行命令的 install.sh 加上 `npx rea-agents setup` 就能完成配置,但你首先需要 Node.js 22.x、24.x 或 26.12+ 以及 npm,然后还要为原生二进制文件单独安装一个反编译器(Hopper、Ghidra 或 IDA),并根据分析对象安装额外的命令行工具(在 Linux 上是 procps 的 `ps` 和 pwntools,安卓需要完整的 JDK 和 JADX,抓包文件需要 tshark)。在 Windows 上,shell 安装脚本拒绝运行,所以你需要手动安装 npm 包,而且 README 把那里的 Ghidra 支持称为实验性的。","costNote":"该工具本身是免费的(MIT 许可)。对原生二进制文件进行深入分析需要单独的反编译器:Ghidra 是免费的,而 Hopper、IDA Pro 和 JEB 是付费的,你需要自备许可证。用来驱动它的 AI 智能体也可能是付费的。","notVerified":["对于某个具体程序,分析结果和还原出的伪代码是否正确","在 Windows 上(安装程序不受支持、原生分析被标记为实验性)哪些功能真正可用","你连接的 AI 智能体在分析你的文件时,会把什么信息发送给它自己的模型提供方","分析一个大型二进制文件需要多少磁盘空间和时间","`rea update` 和 `npx rea-agents@latest` 每次运行时会在你的电脑上下载和更改什么内容"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1NzYyMDE=","slug":"mrdoob__three.js","name":"three.js","owner":"mrdoob","fullName":"mrdoob/three.js","type":"framework_infra","status":"active","grade":"E","easeByPlatform":{"web":"E"},"risk":"low","signals":[{"code":"large_repository_download","severity":"info","evidence":"README.md：\"Cloning the repo with all its history results in a ~2 GB download\"，并建议改用`git clone --depth=1`。","source":"ai","title":"仓库下载体积过大"}],"platforms":["web"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","design_cad"],"alternativeTo":[{"slug":"unity","name":"Unity","confirmed":true},{"slug":"playcanvas","name":"PlayCanvas","confirmed":true}],"declaredAlternatives":[],"stars":118073,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":565,"growthDays":7,"history":[{"date":"2026-10-26","stars":117000},{"date":"2026-10-27","stars":117077},{"date":"2026-10-28","stars":117142},{"date":"2026-10-29","stars":117189},{"date":"2026-10-30","stars":117246},{"date":"2026-10-31","stars":117311},{"date":"2026-11-01","stars":117380},{"date":"2026-11-02","stars":117444},{"date":"2026-11-03","stars":117508},{"date":"2026-11-04","stars":117569},{"date":"2026-11-05","stars":117636},{"date":"2026-11-06","stars":117703},{"date":"2026-11-07","stars":117815},{"date":"2026-11-08","stars":117888},{"date":"2026-11-09","stars":117953},{"date":"2026-11-10","stars":118073}],"firstPublishedAt":"2026-10-27T01:10:12.663Z","commitSha":"fa01dd6ad164c48554f1bb9456f6500db052ec4f","verifiedAt":"2026-11-01T01:39:55.848Z","lastCommitAt":"2026-11-10T22:24:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A JavaScript library programmers use to show 3D scenes, animations and VR inside a web page.","summary":"three.js is a building block for people who write JavaScript, not a program you open. It draws 3D scenes in the browser using WebGL or WebGPU: cameras, lights, materials, loaded 3D models, sound and virtual or augmented reality. Anything built with it runs in a normal browser on any system. You add it to a project with npm or a script tag and then write code; the project site has hundreds of examples, docs and a manual. Cloning the full repository downloads about 2 GB.","easeRationale":"Developer-only: the releases contain no program to run, and the README's starting point is JavaScript code you write yourself after adding the library with npm. The result then runs in any modern browser on Windows, macOS, Linux or a phone.","costNote":"MIT license.","notVerified":["Whether a scene performs well on your graphics card, phone or browser","Whether WebGPU features work in the browser you use","How much of the library ends up in your page's final size"]},"zh-Hans":{"summaryShort":"一个JavaScript库，程序员用它在网页中展示3D场景、动画和VR。","summary":"three.js是写JavaScript的人使用的一个构建模块，不是一个可以打开的程序。它使用WebGL或WebGPU在浏览器中绘制3D场景：摄像机、灯光、材质、加载的3D模型、声音以及虚拟现实或增强现实。用它构建的任何内容都能在任何系统的普通浏览器中运行。你可以用npm或script标签把它加入项目，然后编写代码；项目网站上有数百个示例、文档和一份手册。克隆完整仓库大约要下载2 GB。","easeRationale":"仅面向开发者：发行版中不包含任何可运行的程序，README的起点是你在用npm添加这个库之后自己编写的JavaScript代码。最终结果可以在Windows、macOS、Linux或手机上的任何现代浏览器中运行。","costNote":"MIT许可证。","notVerified":["某个场景在你的显卡、手机或浏览器上是否运行流畅","WebGPU功能在你使用的浏览器中是否可用","最终会有多少库代码打包进你页面的最终体积"]}}},{"id":"R_kgDOQBjiAQ","slug":"msitarzewski__agency-agents","name":"Agency Agents","owner":"msitarzewski","fullName":"msitarzewski/agency-agents","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"recommended_app_in_other_repository","severity":"info","evidence":"README.md 推荐从 https://github.com/msitarzewski/agency-agents-app/releases/latest 安装 \"Option 1: Install the app (Recommended)\"，该应用会 \"auto-updates\"。这个应用位于另一个仓库中，不在本次分析范围内。","source":"ai","title":"推荐的应用位于另一个仓库"},{"code":"writes_into_assistant_config_folders","severity":"info","evidence":"README.md：\"./scripts/install.sh --tool claude-code\" 以及 \"cp engineering/*.md ~/.claude/agents/\"。安装程序会把文件放进 Claude Code、Cursor、Codex 和 OpenCode 等工具的配置文件夹，这些文件会改变助手的行为。","source":"ai","title":"会写入各个助手的配置文件夹"},{"code":"includes_offensive_security_roles","severity":"info","evidence":"README.md，Security Division：\"Penetration Tester | Authorized pentests, red team ops, exploitation | Finding exploitable weaknesses before attackers do\"。整体角色名单以防御为主，但这个角色会让助手去寻找攻入系统的办法，只有在你获准测试的系统上这样做才合法。","source":"ai","title":"包含攻击性的渗透测试角色"},{"code":"install_limit_in_one_host","severity":"info","evidence":"README.md：\"OpenCode's runtime currently registers only ~119 agents and silently drops the rest\"；README 建议用 --division 只安装一部分，以免超出该上限。","source":"ai","title":"某一款工具存在安装数量上限"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":167643,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":1322,"growthDays":7,"history":[{"date":"2026-10-26","stars":164270},{"date":"2026-10-27","stars":164543},{"date":"2026-10-28","stars":164882},{"date":"2026-10-29","stars":165176},{"date":"2026-10-30","stars":165502},{"date":"2026-10-31","stars":165698},{"date":"2026-11-01","stars":165915},{"date":"2026-11-02","stars":166143},{"date":"2026-11-03","stars":166321},{"date":"2026-11-04","stars":166513},{"date":"2026-11-05","stars":166727},{"date":"2026-11-06","stars":166930},{"date":"2026-11-07","stars":167102},{"date":"2026-11-08","stars":167274},{"date":"2026-11-09","stars":167491},{"date":"2026-11-10","stars":167643}],"firstPublishedAt":"2026-10-27T01:10:12.663Z","commitSha":"b37b5bb33b78f2c1d6567d98e25ca818e3259219","verifiedAt":"2026-10-30T01:44:20.950Z","lastCommitAt":"2026-11-09T00:27:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A large library of ready-made role prompts that turn AI assistants like Claude Code or Cursor into specialists.","summary":"The Agency is a big collection of instruction files (\"agents\") for AI assistants such as Claude Code, Cursor, Codex, Gemini CLI, Copilot and OpenCode. Each file gives the assistant a role — frontend developer, SEO specialist, sales coach, security auditor and many more — with a personality, working steps and expected results. It is for people who already use one of these assistants. You add the files by copying them into the assistant's folder, by running the bash install script, or with a separate desktop app the author publishes in another repository.","easeRationale":"The same on all three: you need an AI assistant already installed, then you run ./scripts/install.sh (a bash script, which on Windows means Git Bash or WSL) or copy the Markdown files by hand, as in \"cp engineering/*.md ~/.claude/agents/\".","costNote":"The files are free under the MIT licence. You need an AI assistant to use them, and that assistant may have its own subscription or usage cost.","notVerified":["Whether these role prompts actually give better results than a plain AI assistant.","Whether every agent file is free of instructions that push the assistant to run risky commands or reach outside services.","Whether scripts/install.sh and scripts/convert.sh work on Windows outside Git Bash or WSL.","What the separate desktop app installs and how its auto-update works.","How many of the many agents overlap, and how well each one is kept up to date."]},"zh-Hans":{"summaryShort":"一个庞大的现成角色提示词库,能把Claude Code或Cursor等AI助手变成各领域的专家。","summary":"The Agency是一个庞大的指令文件(\"代理\")集合,适用于Claude Code、Cursor、Codex、Gemini CLI、Copilot和OpenCode等AI助手。每个文件都给助手设定一个角色——前端开发者、SEO专家、销售教练、漏洞审计员等等——并配有性格设定、工作步骤和预期结果。它适合已经在用这些助手之一的人。你可以把文件复制到助手的文件夹里、运行bash安装脚本,或者使用作者在另一个仓库发布的独立桌面应用来添加这些文件。","easeRationale":"三个系统的情况一样:你需要先装好一个AI助手,然后运行./scripts/install.sh(一个bash脚本,在Windows上意味着要用Git Bash或WSL),或者手动复制Markdown文件,比如\"cp engineering/*.md ~/.claude/agents/\"。","costNote":"这些文件在MIT许可下是免费的。你需要一个AI助手才能使用它们,而那个助手可能有自己的订阅费用或使用成本。","notVerified":["这些角色提示词是否真的比普通AI助手带来更好的结果。","是否每个代理文件都不含会驱使助手执行高风险命令或连接外部服务的指令。","scripts/install.sh和scripts/convert.sh在Git Bash或WSL之外的Windows环境中能否正常运行。","那个独立的桌面应用会安装什么内容,它的自动更新又是如何运作的。","众多代理之间有多少重叠,每一个又维护得如何及时。"]}}},{"id":"R_kgDOIPfthA","slug":"mubertai__mubert-text-to-music","name":"Mubert Text-to-Music","owner":"MubertAI","fullName":"MubertAI/Mubert-Text-to-Music","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"web":"C"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单均未检测到任何许可证","source":"check","title":"没有许可证"},{"code":"unmaintained_since_2022","severity":"attention","evidence":"最后一次提交的日期是 2022-10-27（距本次检查 1473 天），也没有发布版本，而代码调用的是一家公司的在线服务，此后可能已经发生变化。","source":"ai","title":"自 2022 年起无人维护"},{"code":"output_use_restricted","severity":"attention","evidence":"README 写道：\"You can use the music for free with attribution to sync with images and videos, but you can't release it on DSPs (Spotify, Apple Music, Deezer etc.) as your own\"，并要求提及 @mubertapp 和 #mubert。","source":"ai","title":"输出内容的使用受限"},{"code":"depends_on_closed_online_service","severity":"info","evidence":"\"All music is generated by Mubert API – www.mubert.com/\"：没有任何部分在你的电脑上运行，所以每个提示词和生成的曲目都会经过该公司的服务器。","source":"ai","title":"依赖封闭的在线服务"},{"code":"third_party_notebook_code","severity":"info","evidence":"`Deforum_Stable_Diffusion_Mubert.ipynb` 被描述为 \"based on Deforum Stable Diffusion colab\"，所以运行它会在你的 Colab 会话中引入并运行另一个项目的代码。","source":"ai","title":"笔记本中含第三方代码"}],"platforms":["web"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Mubert API (mubert.com)"],"cost":{"model":"freemium"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["audio"],"alternativeTo":[{"slug":"epidemic-sound","name":"Epidemic Sound","confirmed":true},{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":2741,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":5,"growthDays":6,"history":[{"date":"2026-11-05","stars":2736},{"date":"2026-11-06","stars":2735},{"date":"2026-11-07","stars":2740},{"date":"2026-11-08","stars":2741},{"date":"2026-11-09","stars":2741},{"date":"2026-11-10","stars":2741}],"firstPublishedAt":"2026-11-08T23:01:18.792Z","commitSha":"fa2c4244d2ff8b1fdbd1ec1c73ed878a11a0adb9","verifiedAt":"2026-11-08T22:44:20.343Z","lastCommitAt":"2022-10-27T00:45:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Describe the music you want and get a track back, from a notebook that runs in your browser.","summary":"Two Google Colab notebooks from the company Mubert. In the first you type a description such as a mood or genre, choose a length, and the Mubert online service sends back a track; the second combines that with Deforum Stable Diffusion to make a prompt-based music video. The music is made on Mubert's servers from loops recorded by musicians, not on your computer, so you need an internet connection and a Google account. The notebooks have had no changes since October 2022.","easeRationale":"Nothing is installed: both files are Colab notebooks you open in the browser with a Google account and then run cell by cell, which still means handling code and waiting for the cells to finish. There is no desktop version and no release file.","costNote":"The notebooks are free. The README says the music can be used for free with credit to @mubertapp and the #mubert hashtag, but not released on Spotify, Apple Music or Deezer as your own; a commercial license has to be arranged with the company.","notVerified":["Whether the Mubert API calls in the notebooks still work after four years without updates","Whether the free tier with attribution is still offered and what a commercial license costs","What the service records about the prompts you send","The sound quality and length limits of the generated tracks","Whether the Stable Diffusion video notebook still runs on today's free Colab machines"]},"zh-Hans":{"summaryShort":"描述你想要的音乐，就能得到一段曲目，这一切都在浏览器里运行的笔记本中完成。","summary":"来自 Mubert 公司的两个 Google Colab 笔记本。在第一个笔记本中，你输入一个描述，比如情绪或风格，选择时长，Mubert 的在线服务就会返回一段曲目；第二个笔记本把它和 Deforum Stable Diffusion 结合起来，制作基于提示词的音乐视频。音乐是在 Mubert 的服务器上，用音乐人录制的循环片段生成的，不是在你的电脑上，所以你需要联网和一个 Google 账户。这些笔记本自 2022 年 10 月以来没有任何改动。","easeRationale":"不需要安装任何东西：这两个文件都是 Colab 笔记本，用 Google 账户在浏览器中打开，然后逐个单元格运行，这仍然意味着要处理代码并等待单元格运行完成。没有桌面版本，也没有发行文件。","costNote":"这些笔记本是免费的。README 说这些音乐可以免费使用，只要注明 @mubertapp 并带上 #mubert 标签，但不能作为你自己的作品发布到 Spotify、Apple Music 或 Deezer 上；商业授权需要和公司另行商谈。","notVerified":["笔记本中对 Mubert API 的调用，在四年没有更新之后是否还能正常工作","带署名的免费档位是否还在提供，商业授权要花多少钱","该服务会记录你发送的提示词中的哪些内容","生成曲目的音质和时长限制如何","Stable Diffusion 视频笔记本在如今的免费 Colab 机器上是否还能运行"]}}},{"id":"R_kgDOJLVrdQ","slug":"mudler__localai","name":"LocalAI","owner":"mudler","fullName":"mudler/LocalAI","type":"framework_infra","status":"active","grade":"C","easeByPlatform":{"mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unsigned_mac_installer","severity":"attention","evidence":"README 写道：\"**Note:** The DMG is not signed by Apple. After installing, run: `sudo xattr -d com.apple.quarantine /Applications/LocalAI.app`\"，这意味着要用管理员命令关闭 macOS 对这个应用的一项保护。","source":"ai","title":"未签名的 Mac 安装程序"},{"code":"downloads_backends_at_runtime","severity":"info","evidence":"\"Backend Gallery — Install/remove backends on the fly via OCI images\" 和 \"Automatic Backend Detection: LocalAI automatically detects your GPU capabilities and downloads the appropriate backend\"：60 多个引擎中的大多数，以及所有模型，都是在你启动之后从网上获取的。","source":"ai","title":"运行时下载后端"},{"code":"built_in_agent_runs_commands","severity":"info","evidence":"README 在介绍终端智能体时写道：\"It answers questions, reads your files and runs commands on your machine, asking you to approve anything that changes state.\"","source":"ai","title":"内置智能体可执行命令"},{"code":"example_configs_ship_default_passwords","severity":"info","evidence":"`docker-compose.distributed.yaml` 设置了 `POSTGRES_PASSWORD: localai`、`LOCALAI_REGISTRATION_TOKEN: \"changeme\"` 以及固定的存储密钥；文件本身也写道：\"All credentials below are examples: change them for anything beyond a local test.\"","source":"ai","title":"示例配置自带默认密码"},{"code":"face_and_voice_biometrics_included","severity":"info","evidence":"后端列表中有 `face-detect.cpp`，用于 \"face detection, recognition, demographics and anti-spoofing\"，还有 `voice-detect.cpp`，用于带年龄、性别和情绪的说话人识别，因此服务器可以根据照片和录音识别出具体的人。","source":"ai","title":"包含人脸和声音生物识别"}],"platforms":["mac","linux"],"tools":[],"buildTools":["docker","gpu_nvidia","go"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["local_ai","development","ai_agents"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":50296,"downloads":223696,"downloadsWeek":2728,"downloadsDays":6,"growth":115,"growthDays":7,"history":[{"date":"2026-11-04","stars":50204},{"date":"2026-11-05","stars":50214},{"date":"2026-11-06","stars":50231},{"date":"2026-11-07","stars":50243},{"date":"2026-11-08","stars":50257},{"date":"2026-11-09","stars":50279},{"date":"2026-11-10","stars":50296}],"firstPublishedAt":"2026-11-08T23:01:18.799Z","commitSha":"6be01747be5cd9d1729b3ba756ed10f15b741610","verifiedAt":"2026-11-08T22:45:43.946Z","lastCommitAt":"2026-11-10T22:12:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Run text, voice, image and video AI models on your own machine behind an OpenAI-style API.","summary":"LocalAI is a server you run yourself that loads AI models and answers requests in the same shape as the OpenAI, Anthropic and ElevenLabs online services, so programs written for those can point at your computer instead. It covers chat, transcription, speech, images, video, embeddings and object detection, works on NVIDIA, AMD, Intel or Apple chips and on the processor alone, and has a web interface, saved chats with agents and tools, user accounts and a cluster mode. There is a macOS installer; most people start it with Docker.","easeRationale":"The easiest route is the macOS LocalAI.dmg from the release page, but the README warns \"The DMG is not signed by Apple\" and tells you to run `sudo xattr -d com.apple.quarantine /Applications/LocalAI.app` before it opens, and you then pick and download a model. On Linux there is a launcher AppImage and a plain binary, while the README's own quickstart is a `docker run` command with a different image for each kind of graphics card.","costNote":"Free and MIT licensed; the author takes sponsorships. Running big models on your own hardware or a rented server is the real cost.","notVerified":["How much disk space and memory it needs once the backends and models are downloaded, which the release files do not include","Whether a given model really runs on your graphics card or only on the processor, and how fast","Whether the macOS and Linux launchers work as smoothly as the Docker route","Whether the many third-party backends are kept up to date and what each one downloads","Whether the \"your data never leaves your infrastructure\" claim holds for every add-on and connector"]},"zh-Hans":{"summaryShort":"在你自己的机器上运行文本、语音、图像和视频 AI 模型，通过类似 OpenAI 风格的 API 提供服务。","summary":"LocalAI 是一个你自己运行的服务器，它加载 AI 模型，并以和 OpenAI、Anthropic、ElevenLabs 在线服务相同的格式响应请求，因此原本为这些服务编写的程序可以改为指向你的电脑。它涵盖聊天、转录、语音、图像、视频、嵌入向量和物体检测，可以运行在 NVIDIA、AMD、Intel 或 Apple 芯片上，也可以只用处理器运行，还带有网页界面、保存的带有智能体和工具的对话、用户账户和集群模式。有一个 macOS 安装程序；大多数人用 Docker 启动它。","easeRationale":"最简单的方式是从发布页面下载 macOS 的 LocalAI.dmg，但 README 警告说 \"The DMG is not signed by Apple\"，并要求在打开之前运行 `sudo xattr -d com.apple.quarantine /Applications/LocalAI.app`，之后再选择并下载一个模型。在 Linux 上有一个启动器 AppImage 和一个普通的二进制文件，而 README 自己的快速开始教程是一条 `docker run` 命令，每种显卡对应不同的镜像。","costNote":"免费且采用 MIT 许可证；作者接受赞助。真正的成本在于用自己的硬件或租用的服务器来运行大型模型。","notVerified":["下载完后端和模型之后，需要多少磁盘空间和内存，而发行文件本身并不包含这些","某个模型是否真的能在你的显卡上运行，还是只能用处理器运行，速度又如何","macOS 和 Linux 的启动器是否像 Docker 方式那样顺畅","众多第三方后端是否保持更新，各自又下载了什么","\"your data never leaves your infrastructure\" 这个说法，对所有插件和连接器是否都成立"]}}},{"id":"R_kgDONuVFoA","slug":"multimodal-art-projection__yue","name":"YuE","owner":"multimodal-art-projection","fullName":"multimodal-art-projection/YuE","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"E","mac":"E","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"model_weights_non_commercial","severity":"attention","evidence":"README 的许可证表格写明代码是 Apache 2.0，但 \"**Model weights:** [CC BY-NC 4.0 with additional creator permission]\"，并且 \"Commercial use by companies: Contact us to discuss a commercial license for the model weights\"。","source":"ai","title":"模型权重为非商业用途"},{"code":"covers_of_existing_recordings","severity":"attention","evidence":"\"Cover a song\" 部分以别人的录音为起点 (\"Transcribe a source recording with SheetSage2 ... provide new lyrics or a target style\")，主题列表中还有 \"voice-cloning\" 和 \"style-transfers\"；原曲和原唱声音的权利需要由你自己去核实清楚。","source":"ai","title":"翻唱已有录音"},{"code":"heavy_hardware_and_downloads","severity":"info","evidence":"快速开始部分写道：\"NVIDIA GPU with BF16 support and 24 GB VRAM\" 以及 \"Model files download from Hugging Face on first use\"，所以真正占用磁盘和内存的是下载的内容，而不是这个仓库本身。","source":"ai","title":"硬件要求高，下载量大"},{"code":"self_reported_benchmarks","severity":"info","evidence":"将 YuE2 与 Suno、Mureka 和 MiniMax 对比的基准测试表格，是作者自己在他们自己的 WildSongBench 数据集上做的评测，README 本身也指出 \"the small gap between the highest means does not establish statistical significance\"。","source":"ai","title":"基准测试由作者自行发布"}],"platforms":["linux","windows","mac","web"],"tools":[],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true},{"slug":"epidemic-sound","name":"Epidemic Sound","confirmed":true}],"declaredAlternatives":[],"stars":13370,"downloads":9228,"downloadsWeek":530,"downloadsDays":6,"growth":285,"growthDays":6,"history":[{"date":"2026-11-04","stars":13085},{"date":"2026-11-05","stars":13114},{"date":"2026-11-06","stars":13137},{"date":"2026-11-07","stars":13164},{"date":"2026-11-08","stars":13201},{"date":"2026-11-09","stars":13284},{"date":"2026-11-10","stars":13370}],"firstPublishedAt":"2026-11-08T23:01:18.807Z","commitSha":"1116bea76032bee5668e5f097294d082fa0b84f0","verifiedAt":"2026-11-08T22:46:57.788Z","lastCommitAt":"2026-11-01T05:46:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turn lyrics and a style description into a full song with singing, and edit the score before it is sung.","summary":"YuE2 writes songs. You give it lyrics and a style, and it first makes a melody-and-chord score you can read and change, then sings and plays it as a 48 kHz stereo track. It can also take an existing recording, transcribe the melody and perform it in another style, and an included agent skill lets an AI assistant drive the whole process. Running it yourself means Linux, Python 3.12 and an NVIDIA card with 24 GB of video memory; the models download from Hugging Face the first time. A free hosted demo lets you try it in the browser.","easeRationale":"The one-click way is the free hosted demo the README links (\"Try YuE2 online for free ... No installation required\"), which runs in the browser. Doing it on your own machine is much harder: the quick start is \"Linux · Python 3.12 · NVIDIA GPU with BF16 support and 24 GB VRAM\" with a clone, a virtual environment and command-line runs, and Windows, Apple Silicon, AMD and Intel are only described as \"experimental\" paths in docs/hardware.md.","costNote":"Code and the agent skill are Apache 2.0. The model weights are CC BY-NC 4.0 with an extra permission: people, creators and musicians may use them and sell what they make, academic use is non-commercial, and companies have to ask the authors for a commercial license.","notVerified":["Whether songs really come out at the quality shown in the demos, and how long each one takes","How much disk space the YuE2-3B, VAE and SheetSage2 downloads take together","Whether the experimental Windows, Apple Silicon, AMD and Intel paths work at all","Whether the free hosted demo stays free, and what it keeps of the lyrics and audio you send","What music the model was trained on, which matters for who owns the result"]},"zh-Hans":{"summaryShort":"把歌词和风格描述变成一首完整的演唱歌曲，还能在演唱之前编辑曲谱。","summary":"YuE2 用来创作歌曲。你给它歌词和一种风格，它会先生成一份你可以阅读和修改的旋律和和弦曲谱，然后以 48 kHz 立体声音轨的形式演唱和演奏出来。它还可以接收一段已有的录音，转录出旋律，再用另一种风格演绎，内置的智能体技能还能让 AI 助手驱动整个流程。自己运行需要 Linux、Python 3.12 和一张显存 24 GB 的 NVIDIA 显卡；模型会在第一次使用时从 Hugging Face 下载。还有一个免费的在线演示，可以直接在浏览器中体验。","easeRationale":"一键体验的方式是 README 中链接的免费在线演示 (\"Try YuE2 online for free ... No installation required\")，它在浏览器中运行。在自己的机器上运行要难得多：快速开始部分写的是 \"Linux · Python 3.12 · NVIDIA GPU with BF16 support and 24 GB VRAM\"，需要克隆代码、建立虚拟环境并在命令行中运行，而 Windows、Apple Silicon、AMD 和 Intel 在 docs/hardware.md 中只被描述为 \"experimental\" 的路径。","costNote":"代码和智能体技能采用 Apache 2.0 许可证。模型权重采用 CC BY-NC 4.0 并附带额外许可：个人、创作者和音乐人可以使用它们并出售自己制作的成果，学术用途属于非商业用途，公司则需要向作者申请商业授权。","notVerified":["歌曲的质量是否真的能达到演示中展示的水平，每首又需要多长时间","YuE2-3B、VAE 和 SheetSage2 的下载加在一起占用多少磁盘空间","实验性的 Windows、Apple Silicon、AMD 和 Intel 路径是否真的能用","免费的在线演示是否会一直免费，它会保留你发送的歌词和音频中的哪些内容","模型是用什么音乐训练的，这关系到结果的归属问题"]}}},{"id":"R_kgDOKzf7EQ","slug":"myshell-ai__openvoice","name":"OpenVoice","owner":"myshell-ai","fullName":"myshell-ai/OpenVoice","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"存在已知漏洞的固定依赖项（OSV）：gradio：GHSA-26jh-r8g2-6fpr、GHSA-279j-x4gx-hfrh、GHSA-34rf-p3r3-58x2","source":"check","title":"已知漏洞"},{"code":"voice_cloning_impersonation_risk","severity":"attention","evidence":"该仓库将自己描述为 \"Instant voice cloning\"，并具有 \"Accurate Tone Color Cloning\" 功能，可以从一段参考录音复制声音，因此任何人的声音都可能被从一段样本中复制出来，并被用来说出任何内容，包括他们并不会说的语言。","source":"ai","title":"声音克隆被冒用的风险"},{"code":"unmaintained","severity":"attention","evidence":"最近一次提交是在 2025-04-19，距本次检查 568 天，没有发布过任何版本，有 317 个未解决的 issue。","source":"ai","title":"缺乏维护"},{"code":"old_pinned_dependencies","severity":"info","evidence":"requirements.txt 固定了 `numpy==1.22.0`、`gradio==3.48.0`、`librosa==0.9.1` 和 `faster-whisper==0.9.0`，这些是 2022-2023 年的版本，在当前 Python 上往往会安装失败。","source":"ai","title":"依赖版本过旧"},{"code":"audio_watermark_dependency","severity":"info","evidence":"requirements.txt 包含 `wavmark==0.0.3`，一个水印库，因此生成的音频可能带有隐藏水印。","source":"ai","title":"音频水印依赖"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI (only for the extra demo notebook; `openai` and `python-dotenv` are in requirements.txt)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true}],"declaredAlternatives":[],"stars":38192,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":58,"growthDays":7,"history":[{"date":"2026-11-04","stars":38136},{"date":"2026-11-05","stars":38142},{"date":"2026-11-06","stars":38153},{"date":"2026-11-07","stars":38173},{"date":"2026-11-08","stars":38182},{"date":"2026-11-09","stars":38191},{"date":"2026-11-10","stars":38192}],"firstPublishedAt":"2026-11-08T23:01:18.815Z","commitSha":"74a1d147b17a8c3092dd5430504bd83ef6c7eb23","verifiedAt":"2026-11-08T22:47:39.582Z","lastCommitAt":"2025-04-19T15:59:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copy a voice from a short recording and make it read any text, in six languages.","summary":"OpenVoice, from MIT researchers and the company MyShell, copies the tone of a voice from a short reference recording and then reads text in that voice, with control over emotion, accent, rhythm and pauses. Version 2 speaks English, Spanish, French, Chinese, Japanese and Korean, and the voice can speak a language the original recording was not in. It is a Python package with three demo notebooks, so you install Python, install the package and run the notebooks; everything runs on your own machine and the licence allows commercial use.","easeRationale":"The same on all three systems: install Python 3.9 or newer, install the package from source and open `demo_part1.ipynb` to `demo_part3.ipynb`, since there is no installer, no release file and no graphics card requirement. The old pinned versions make the install the hard part rather than the machine you use.","costNote":"MIT licence for both V1 and V2, free for commercial and research use according to the README.","notVerified":["Whether the install still succeeds today with those pinned versions and a current Python","Where the voice model files are downloaded from and how big they are","How close the cloned voice really sounds, and in which languages","Whether a watermark is written into every output and whether it can be read back","Whether the known gradio problems found by the checks affect the demo notebooks you run locally"]},"zh-Hans":{"summaryShort":"从一段简短录音中复制一个声音，让它用六种语言朗读任意文本。","summary":"OpenVoice 由麻省理工学院的研究人员和 MyShell 公司开发，它能从一段简短的参考录音中复制声音的音色，然后用这个声音朗读文本，并可以控制情感、口音、节奏和停顿。第二版支持英语、西班牙语、法语、中文、日语和韩语，而且这个声音可以说原始录音中没有的语言。它是一个 Python 软件包，附带三个演示笔记本，因此你需要安装 Python、安装该软件包并运行这些笔记本；一切都在你自己的电脑上运行，许可证允许商业用途。","easeRationale":"三个系统上做法相同：安装 Python 3.9 或更新版本，从源代码安装软件包，然后打开 `demo_part1.ipynb` 到 `demo_part3.ipynb`，因为没有安装程序、没有发行文件，也不需要显卡。固定的旧版本依赖让安装过程本身成为难点，而不是你使用的机器。","costNote":"V1 和 V2 均采用 MIT 许可证，根据 README，可免费用于商业和研究用途。","notVerified":["在使用这些固定版本和当前 Python 的情况下，安装今天是否仍能成功","声音模型文件从哪里下载，体积有多大","克隆出的声音到底有多像原声，在哪些语言下效果如何","每次输出是否都会写入水印，以及这个水印是否可以被读出","检查发现的 gradio 已知问题是否会影响你在本地运行的演示笔记本"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxOTMyMTU1NTQ=","slug":"n8n-io__n8n","name":"n8n","owner":"n8n-io","fullName":"n8n-io/n8n","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"licence_limits_use_and_sharing","severity":"attention","evidence":"LICENSE.md 是 \"Sustainable Use License\"。它规定 \"You may use or modify the software only for your own internal business purposes or for non-commercial or personal use\" 以及 \"You may distribute the software or provide it to others only if you do so free of charge for non-commercial purposes\"。它还规定文件名中带 \".ee.\" 或目录名带 \".ee\" 的文件需要 \"a valid n8n Enterprise License\"，并且 \"Content of branches other than the main branch (i.e. \"master\") are not licensed\"。","source":"ai","title":"许可证限制使用和分享方式"},{"code":"install_pipes_a_script_into_your_shell","severity":"attention","evidence":"README Quick Start：`curl -fsSL https://get.n8n.io | sh`。脚本被下载后立即执行，那个地址当时提供的任何内容都会在你的电脑上运行。","source":"ai","title":"安装过程把脚本直接输送到你的 shell"},{"code":"quick_start_publishes_port_5678","severity":"attention","evidence":"README 中的手动命令是 `docker run -it --rm --name n8n -p 5678:5678 -v n8n_data:/home/node/.n8n docker.n8n.io/n8nio/n8n`，它会公开 5678 端口，下一行就是 \"Access the editor at http://localhost:5678\"。说明里没有任何登录步骤。","source":"ai","title":"快速开始会公开 5678 端口"},{"code":"workflows_can_run_any_code","severity":"attention","evidence":"README：\"Combine visual building with JavaScript, Python, and npm packages for advanced AI workflows.\" README 还提供了 \"9,000+ workflow templates\"，所以你导入的工作流可能带有代码。","source":"ai","title":"工作流可以执行任意代码"},{"code":"has_a_usage_reporting_feature","severity":"info","evidence":"`package.json` 中有一个指向 PostHog 的诊断脚本：`\"dev:be:telemetry\": \"cross-env N8N_DIAGNOSTICS_ENABLED=true N8N_DIAGNOSTICS_CONFIG_BACKEND=... N8N_DIAGNOSTICS_POSTHOG_API_HOST=... N8N_DIAGNOSTICS_POSTHOG_API_KEY=local\"`。也就是说有一个由 `N8N_DIAGNOSTICS_ENABLED` 设置控制的使用情况上报功能。","source":"ai","title":"具备使用情况上报功能"},{"code":"two_version_lines_at_once","severity":"info","evidence":"2026-11-02 同时发布了三个版本：`n8n@1.123.92`、`n8n@2.46.4` 和实验性的 `n8n@2.45.7-exp.g86c6a0b7`。升级前最好先确认自己在哪条版本线上，因为 1.x 和 2.x 两条线是并行发布的。","source":"ai","title":"两条版本线同时存在"}],"platforms":["windows","mac","linux"],"tools":["docker","node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["automation","ai_agents"],"alternativeTo":[{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"make","name":"Make","confirmed":true},{"slug":"n8n-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":210014,"downloads":41034,"downloadsWeek":588,"downloadsDays":7,"growth":533,"growthDays":7,"history":[{"date":"2026-10-25","stars":208596},{"date":"2026-10-26","stars":208675},{"date":"2026-10-27","stars":208807},{"date":"2026-10-28","stars":208914},{"date":"2026-10-29","stars":209049},{"date":"2026-10-30","stars":209152},{"date":"2026-10-31","stars":209212},{"date":"2026-11-01","stars":209310},{"date":"2026-11-02","stars":209396},{"date":"2026-11-03","stars":209481},{"date":"2026-11-04","stars":209565},{"date":"2026-11-05","stars":209639},{"date":"2026-11-06","stars":209708},{"date":"2026-11-07","stars":209768},{"date":"2026-11-08","stars":209842},{"date":"2026-11-09","stars":209927},{"date":"2026-11-10","stars":210014}],"firstPublishedAt":"2026-10-25T23:19:14.287Z","commitSha":"18c045bfacb12c8163ff770970b330af9730aaab","verifiedAt":"2026-11-03T02:54:21.200Z","lastCommitAt":"2026-11-10T22:14:20Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Builds automations by joining apps and AI steps on a visual canvas, running on your own computer or server.","summary":"A tool for building automations without writing much code. You drag boxes onto a canvas, join them, and each box talks to a service, runs a condition or calls an AI model, so you can wire up things like \"when a form is filled, add a row and send a message\". The company lists more than 1500 services and thousands of ready-made examples, and you can drop in JavaScript or Python where boxes are not enough. You run it yourself with Docker and open the editor at http://localhost:5678 in a browser; the same software is also sold as a hosted service.","easeRationale":"Windows, macOS and Linux all take the same route and get the same grade: install Docker, then run `curl -fsSL https://get.n8n.io | sh` or the `docker run ... -p 5678:5678` command from the README and open http://localhost:5678. There is no installer to double-click, and keeping it running means keeping a container and a volume.","costNote":"Running it yourself costs nothing in licence fees, but the hosted n8n Cloud is paid and the files marked \".ee.\" ask for a paid Enterprise licence. Any AI models or paid services your automations call are billed separately.","notVerified":["Real disk use and memory once the container, the database and your execution history grow","Whether the connections you need are among the 1500 the company lists, and whether they still work","What a comparable amount of work costs on the paid cloud plan instead","Whether usage reporting is on in the image you downloaded, and what it would send","Whether the editor asks for a password on a first run started with the README command","How much work an upgrade between the 1.x and 2.x lines takes on an existing installation"]},"zh-Hans":{"summaryShort":"通过在可视化画布上连接应用和 AI 步骤来搭建自动化流程，运行在你自己的电脑或服务器上。","summary":"一个不用写太多代码就能搭建自动化流程的工具。你把方块拖到画布上并连接起来，每个方块对接一个服务、执行一个条件判断或调用一个 AI 模型，这样就能搭出类似「表单填写后，添加一行并发送消息」这样的流程。该公司列出了超过 1500 个服务和数千个现成示例，在方块无法满足需求时还可以插入 JavaScript 或 Python 代码。你用 Docker 自己运行它，在浏览器里打开 http://localhost:5678 使用编辑器；同一套软件也作为托管服务出售。","easeRationale":"Windows、macOS 和 Linux 走的是同一条路线，评级也相同：安装 Docker，然后运行 README 里的 `curl -fsSL https://get.n8n.io | sh` 或 `docker run ... -p 5678:5678` 命令，再打开 http://localhost:5678。没有可以双击运行的安装程序，持续运行意味着要维护一个容器和一个数据卷。","costNote":"自己运行不需要支付任何许可费用，但托管的 n8n Cloud 是收费的，标有 \".ee.\" 的文件需要付费的 Enterprise 许可证。你的自动化流程调用的任何 AI 模型或付费服务都会单独计费。","notVerified":["容器、数据库和执行历史增长后实际占用的磁盘和内存","你需要的连接是否在公司列出的 1500 个之中，以及它们是否仍能正常使用","同样的工作量放到付费云计划上要花多少钱","你下载的镜像中使用情况上报功能是否开启，以及它会发送什么内容","用 README 命令首次启动时，编辑器是否会要求设置密码","在已有安装上从 1.x 升级到 2.x 需要多少工作量"]}}},{"id":"R_kgDOUfqq7Q","slug":"nandhakishorm__laya","name":"Laya","owner":"NandhaKishorM","fullName":"NandhaKishorM/laya","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"37 天内获得 34301 个星标","source":"check","title":"新仓库却有大量星标"},{"code":"rapid_release_pace","severity":"info","evidence":"GitHub API：该仓库创建于 2026-09-18（距本次检查 37 天），已有 48 个发布版本；最近三个（v0.4.13 至 v0.4.15）发布于 2026-10-21 到 2026-10-24 之间。不同版本之间的行为可能经常变化。","source":"ai","title":"发布节奏很快"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":35354,"downloads":2445,"downloadsWeek":174,"downloadsDays":7,"growth":392,"growthDays":7,"history":[{"date":"2026-10-25","stars":34310},{"date":"2026-10-26","stars":34407},{"date":"2026-10-27","stars":34481},{"date":"2026-10-28","stars":34565},{"date":"2026-10-29","stars":34642},{"date":"2026-10-30","stars":34712},{"date":"2026-10-31","stars":34770},{"date":"2026-11-01","stars":34818},{"date":"2026-11-02","stars":34898},{"date":"2026-11-03","stars":34962},{"date":"2026-11-04","stars":35007},{"date":"2026-11-05","stars":35058},{"date":"2026-11-06","stars":35113},{"date":"2026-11-07","stars":35180},{"date":"2026-11-08","stars":35226},{"date":"2026-11-09","stars":35286},{"date":"2026-11-10","stars":35354}],"firstPublishedAt":"2026-10-25T23:19:15.150Z","commitSha":"a478221bcad24d1a9c47b361c79328ebd2b100c6","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T19:14:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"A Python library that sorts and labels text automatically, such as routing a customer email to the right team or rating its urgency.","summary":"Laya is a Python library with small AI models that answer fixed questions about a piece of text: pick one label from a list, give a score, or say yes or no, with a confidence number. A typical use is sorting support tickets or emails by department and urgency, in many languages. It is meant for developers building such features. You need Python 3.10 or newer; the models (about 2.3 GB for all three) download from Hugging Face on first use. A graphics card makes it faster but is optional, and a local test web page is included.","easeRationale":"There is no Windows installer; you install Python, then run a few commands (python -m pip install laya) or the included start.bat from a copy of the repository, and the models download on first use.","notVerified":["Whether the accuracy and speed figures in the README (for example, 33 ms per question on a T4 graphics card) hold on your own data and computer","The model files are downloaded from Hugging Face at run time, and what arrives depends on that source unless you pin a reviewed version","Real disk and memory use once the models are downloaded and loaded","How well the included start.bat works on a typical Windows computer without a graphics card"]},"zh-Hans":{"summaryShort":"一个 Python 库，可自动对文本分类和打标签，例如把客户邮件分派给合适的团队，或评估其紧急程度。","summary":"Laya 是一个 Python 库，内含小型 AI 模型，可以针对一段文本回答固定的问题：从列表中选一个标签、给出分数，或回答是或否，并附带置信度数值。典型用途是按部门和紧急程度对客服工单或邮件分类，支持多种语言。它面向开发此类功能的开发者。你需要 Python 3.10 或更新版本；模型（三个合计约 2.3 GB）在首次使用时从 Hugging Face 下载。显卡可以让它更快，但不是必需的，另附有一个本地测试网页。","easeRationale":"没有 Windows 安装程序；你需要先安装 Python，然后运行几条命令（python -m pip install laya），或在仓库副本中运行附带的 start.bat，模型会在首次使用时下载。","notVerified":["README 中的准确率和速度数据（例如在 T4 显卡上每个问题 33 毫秒）在你自己的数据和电脑上是否同样成立","模型文件在运行时从 Hugging Face 下载，除非你固定某个已审查的版本，否则下载到的内容取决于该来源","模型下载并加载后实际占用的磁盘和内存","附带的 start.bat 在没有显卡的普通 Windows 电脑上运行效果如何"]}}},{"id":"R_kgDONkPjQg","slug":"nanobrowser__nanobrowser","name":"Nanobrowser","owner":"nanobrowser","fullName":"nanobrowser/nanobrowser","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"agent_acts_in_your_logged_in_browser","severity":"attention","evidence":"该仓库的描述是 \"Use your real, logged-in browser to automate daily tasks\"，README 说控制层 \"talks to the page only through Chrome's `chrome.debugger` API\"：代理在你已经登录的会话中工作，比如邮箱、网银或社交账号。","source":"ai","title":"代理在你已登录的浏览器中操作"},{"code":"page_content_sent_to_model_provider","severity":"attention","evidence":"README：\"Pages go only to the model you choose\" ——也就是说，使用云端提供方时，你所在网页的文本，以及使用视觉模型时的截图，会离开你的电脑。","source":"ai","title":"网页内容会发送给模型提供方"},{"code":"prompt_injection_risk_acknowledged","severity":"attention","evidence":"README 自己的说明：\"The agent reads text on a web page as information, never as orders. This helps against pages that try to trick it (prompt injection), but no protection is perfect\"，以及 \"The agent may not notice every step that needs you\"。","source":"ai","title":"承认存在提示注入风险"},{"code":"anonymous_usage_telemetry","severity":"info","evidence":"`.env.example` 描述了 PostHog 分析功能：\"A production build (pnpm build) with both values sends the anonymous daily ping, and the usage counts and problem reports the user allows, to this project\"；仓库里有 PRIVACY.md，README 说网页内容和网站信息永远不会包含在内。","source":"ai","title":"匿名使用遥测"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["A cloud model provider (Anthropic, OpenAI, Google Gemini, OpenRouter, xAI Grok, DeepSeek, Mistral or any OpenAI-compatible endpoint)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","automation"],"alternativeTo":[{"slug":"browser-use-cloud","name":"Cloud","confirmed":true},{"slug":"perplexity-comet","name":"Perplexity Comet","confirmed":true},{"slug":"manus","name":"Manus","confirmed":true}],"declaredAlternatives":[],"stars":14277,"downloads":16068,"downloadsWeek":116,"downloadsDays":6,"growth":41,"growthDays":6,"history":[{"date":"2026-11-04","stars":14236},{"date":"2026-11-05","stars":14248},{"date":"2026-11-06","stars":14252},{"date":"2026-11-07","stars":14258},{"date":"2026-11-08","stars":14263},{"date":"2026-11-09","stars":14269},{"date":"2026-11-10","stars":14277}],"firstPublishedAt":"2026-11-08T23:01:18.823Z","commitSha":"ebd78afbe53a28395d4f558ac168916c174f3e68","verifiedAt":"2026-11-08T22:48:24.411Z","lastCommitAt":"2026-11-10T03:43:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An AI helper in a Chrome side panel that reads pages, clicks and types to finish tasks for you.","summary":"Nanobrowser is a Chrome and Edge extension. You write a task in a side panel and it works inside the browser you already use: it reads pages, clicks, types, moves between tabs and can save a result as a CSV or Markdown file. By default it asks before it submits, sends, deletes or uploads, and it hands passwords and payments back to you. It needs a model: either a key from Anthropic, OpenAI, Google, OpenRouter, xAI, DeepSeek or Mistral, which you pay for by use, or a model running on your own computer with Ollama or LM Studio.","easeRationale":"Same on every system that runs Chrome or Edge: install it from the Chrome Web Store with one click, then the welcome page asks you to connect a model, which means pasting an API key or pointing it at Ollama or LM Studio. Building it yourself instead needs Node 24 and pnpm and loading an unpacked folder, which is only for developers.","costNote":"The extension is free with no subscription; you pay the model provider for what you use. A local model with Ollama or LM Studio avoids that cost, and the README says local models \"work best for simple, clear tasks\".","notVerified":["Whether the Ask mode and per-site rules really stop every risky action","How much a month of use costs with a paid model provider","Whether the version in the Chrome Web Store is built from this same source","How well it works with a local model on an ordinary computer","Whether it behaves in Edge and other Chromium browsers, which the README says are not tested"]},"zh-Hans":{"summaryShort":"一个在 Chrome 侧边栏里的 AI 助手，能阅读网页、点击和输入，替你完成任务。","summary":"Nanobrowser 是一个 Chrome 和 Edge 扩展。你在侧边栏里写下一个任务，它就会在你已经在用的浏览器里工作：阅读网页、点击、输入、在多个标签页之间切换，还可以把结果保存为 CSV 或 Markdown 文件。默认情况下，它会在提交、发送、删除或上传之前先征求你的同意，并把密码和支付操作交还给你处理。它需要一个模型：可以是 Anthropic、OpenAI、Google、OpenRouter、xAI、DeepSeek 或 Mistral 的密钥，按使用量付费，也可以是用 Ollama 或 LM Studio 在你自己电脑上运行的模型。","easeRationale":"在任何能运行 Chrome 或 Edge 的系统上都一样：从 Chrome 网上应用店一键安装，然后欢迎页面会要求你连接一个模型，也就是粘贴一个 API 密钥，或者把它指向 Ollama 或 LM Studio。自己编译则需要 Node 24 和 pnpm，并加载一个未打包的文件夹，这只适合开发者。","costNote":"这个扩展是免费的，没有订阅费；你按使用量向模型提供方付费。用 Ollama 或 LM Studio 运行本地模型可以省去这笔费用，README 说本地模型\"work best for simple, clear tasks\"。","notVerified":["Ask 模式和按站点设置的规则是否真的能拦下每一个有风险的操作","使用付费模型提供方，每个月的花费大概是多少","Chrome 网上应用店里的版本是否是从这同一份源代码构建的","在普通电脑上使用本地模型效果如何","它在 Edge 和其他基于 Chromium 的浏览器上表现如何，README 说这些浏览器没有经过测试"]}}},{"id":"R_kgDOOcHu5w","slug":"nari-labs__dia","name":"Dia","owner":"nari-labs","fullName":"nari-labs/dia","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"voice_cloning_impersonation_risk","severity":"attention","evidence":"功能列表中包含从 5 到 10 秒音频进行的 \"Voice cloning\"，而项目自己的免责声明禁止 \"**Identity Misuse**: Do not produce audio resembling real individuals without permission\" 以及 \"**Deceptive Content**\"，例如虚假新闻。","source":"ai","title":"声音克隆被冒用的风险"},{"code":"superseded_by_newer_repository","severity":"attention","evidence":"README：\"**UPDATE 🚀(11/19)**: Dia2 is released on Github and HuggingFace\"，这里最近一次提交是在 2025-11-19，距本次检查 354 天，有 91 个未解决的 issue。","source":"ai","title":"已被更新的仓库取代"},{"code":"nvidia_only_no_cpu","severity":"info","evidence":"\"Dia has been tested on only GPUs (pytorch 2.0+, CUDA 12.6). CPU support is to be added soon\"，待办事项列表里仍然有 \"Docker support for ARM architecture and MacOS\"。","source":"ai","title":"仅支持 NVIDIA，不支持纯处理器"},{"code":"english_only_and_random_voices","severity":"info","evidence":"\"The model only supports English generation at the moment\" 以及 \"The model was not fine-tuned on a specific voice. Hence, you will get different voices every time you run the model\"，除非你固定随机种子或提供一段音频样本。","source":"ai","title":"仅支持英语，声音随机"}],"platforms":["windows","linux","web"],"tools":[],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[],"stars":19411,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":3,"growthDays":6,"history":[{"date":"2026-11-07","stars":19407},{"date":"2026-11-08","stars":19406},{"date":"2026-11-09","stars":19407},{"date":"2026-11-10","stars":19411}],"firstPublishedAt":"2026-11-08T23:01:18.830Z","commitSha":"876125e461a03b157ec905b0fe8b57a0f8b9e7a0","verifiedAt":"2026-11-08T22:48:57.728Z","lastCommitAt":"2025-11-19T21:11:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turn a written script into a realistic two-person conversation, with laughs, coughs and sighs.","summary":"Dia is a speech model from Nari Labs that reads a script written with [S1] and [S2] speaker tags and produces a conversation that sounds like two people talking, including sounds like laughter, coughing or sighing. You can hand it 5 to 10 seconds of recorded audio to copy a voice and keep it steady. English only. There is a free page on Hugging Face where you can try it; to run it yourself you need Python and an NVIDIA card with about 4.4 GB of video memory, since the README says the processor alone is not supported yet.","easeRationale":"The easy way is the maintainers' free Hugging Face page (\"We have a ZeroGPU Space running\"), which needs nothing installed. Running it yourself on Windows or Linux means cloning the repository, installing Python packages pinned to torch 2.6 with CUDA 12.6, and an NVIDIA card: \"Dia has been tested on only GPUs (pytorch 2.0+, CUDA 12.6). CPU support is to be added soon.\" macOS support is still on the TODO list.","costNote":"Code and weights are free (Apache 2.0, model files on Hugging Face). The README invites you to a waitlist for \"a larger version of Dia\", whose terms are not stated.","notVerified":["How good the result sounds outside the maintainers' own demo page","Total disk space after the model and the Descript Audio Codec download on first run","Whether it works on cards other than the RTX 4090 the speed table used, including the 5000 series note","Whether the free Hugging Face page stays available and what it keeps of the audio you upload","Whether the repository will get fixes now that work moved to the Dia2 repository"]},"zh-Hans":{"summaryShort":"把一段写好的剧本变成一段逼真的两人对话，还带笑声、咳嗽和叹气。","summary":"Dia 是 Nari Labs 开发的语音模型，它读取用 [S1] 和 [S2] 说话人标签写成的剧本，生成一段听起来像两个人在交谈的对话，还包括笑声、咳嗽或叹气等声音。你可以给它 5 到 10 秒的录音来复制一个声音并让它保持稳定。目前只支持英语。Hugging Face 上有一个免费页面可以试用；如果你想自己运行，需要 Python 和一块显存约 4.4 GB 的 NVIDIA 显卡，因为 README 说目前还不支持纯处理器运行。","easeRationale":"最简单的方法是维护者在 Hugging Face 上提供的免费页面（\"We have a ZeroGPU Space running\"），不需要安装任何东西。自己在 Windows 或 Linux 上运行则需要克隆仓库，安装固定为 torch 2.6、CUDA 12.6 的 Python 包，以及一块 NVIDIA 显卡：\"Dia has been tested on only GPUs (pytorch 2.0+, CUDA 12.6). CPU support is to be added soon.\" macOS 支持仍在待办事项列表里。","costNote":"代码和权重都是免费的（Apache 2.0 许可证，模型文件托管在 Hugging Face）。README 邀请你加入等候名单，以获得 \"a larger version of Dia\"，但没有说明相关条款。","notVerified":["在维护者自己的演示页面之外，生成结果听起来到底有多好","首次运行下载模型和 Descript Audio Codec 之后，总共占用多少磁盘空间","它在速度表所用的 RTX 4090 之外的显卡上是否能用，包括关于 5000 系列的说明","Hugging Face 的免费页面是否会一直可用，以及它会保留你上传的音频中的哪些内容","既然工作已经转移到 Dia2 仓库，这个仓库是否还会收到修复"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTY5NDExNTA=","slug":"natrongithub__natron","name":"Natron","owner":"NatronGitHub","fullName":"NatronGitHub/Natron","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2024-09-13（772 天前）","source":"check","title":"发布版本长期未更新"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.8/10","source":"check","title":"Scorecard 评分偏低"},{"code":"no_recent_windows_build","severity":"info","evidence":"最近的发布版本 v2.5.1-pre2（2024-09-13）和 v2.5.1-pre1（2023-04-15）只包含 macOS 的 .dmg 文件；README：\"Natron is looking for developers and maintainers!\"","source":"ai","title":"没有近期的 Windows 版本"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":256,"maxMB":768,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["video"],"alternativeTo":[{"slug":"after-effects","name":"After Effects","confirmed":true},{"slug":"nuke","name":"Nuke","confirmed":true},{"slug":"blackmagic-fusion-studio","name":"Blackmagic Fusion Studio","confirmed":true}],"declaredAlternatives":[],"stars":5691,"downloads":1167593,"downloadsWeek":4175,"downloadsDays":7,"growth":60,"growthDays":7,"history":[{"date":"2026-10-25","stars":5596},{"date":"2026-10-26","stars":5598},{"date":"2026-10-27","stars":5600},{"date":"2026-10-28","stars":5606},{"date":"2026-10-29","stars":5619},{"date":"2026-10-30","stars":5620},{"date":"2026-10-31","stars":5624},{"date":"2026-11-01","stars":5627},{"date":"2026-11-02","stars":5628},{"date":"2026-11-03","stars":5631},{"date":"2026-11-04","stars":5638},{"date":"2026-11-05","stars":5649},{"date":"2026-11-06","stars":5663},{"date":"2026-11-07","stars":5676},{"date":"2026-11-08","stars":5684},{"date":"2026-11-09","stars":5688},{"date":"2026-11-10","stars":5691}],"firstPublishedAt":"2026-10-25T23:19:16.014Z","commitSha":"3763d805d7d277d10af10025ae41af677682b3e6","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-07-24T09:48:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Combine video layers, add visual effects, cut out objects and track motion using a node-based editor.","summary":"Natron is a free video compositing program. You connect boxes (nodes) in a graph to layer footage, key out green screens, rotoscope, track motion and animate effects, then render the result. It is aimed at video effects artists and students who want something similar to Nuke or After Effects. Ready-made downloads bundle the program with four sets of effect plugins. It needs a graphics card with OpenGL 2.0 for full speed, and the README notes the project is looking for new developers and maintainers.","easeRationale":"The checks found a Windows installer in the project's releases, and the README says the standalone downloads already include the main effect plugins.","costNote":"GNU GPL version 2. Some third-party OpenFX plugins it supports are commercial.","notVerified":["Which release is the newest one with a Windows installer, and how old it is","Whether it runs well on current Windows versions and modern graphics cards","Whether commercial OpenFX plugins you own work with it"]},"zh-Hans":{"summaryShort":"用节点式编辑器合成视频图层、添加视觉特效、抠出对象并跟踪运动。","summary":"Natron 是一款免费的视频合成程序。你在图中连接方框（节点），即可叠加素材、抠除绿幕、做转描、跟踪运动并为特效添加动画，然后渲染出结果。它面向视频特效师和学生，适合想要类似 Nuke 或 After Effects 的工具的人。现成的下载包把程序与四组特效插件打包在一起。要获得完整速度，需要支持 OpenGL 2.0 的显卡；README 还提到，项目正在寻找新的开发者和维护者。","easeRationale":"检查在项目的发布页中发现了 Windows 安装程序，README 称独立下载包已包含主要的特效插件。","costNote":"GNU GPL 第 2 版。它支持的一些第三方 OpenFX 插件是商业软件。","notVerified":["哪个是带有 Windows 安装程序的最新版本，以及它有多旧","它能否在当前的 Windows 版本和新款显卡上良好运行","你已拥有的商业 OpenFX 插件能否与它配合使用"]}}},{"id":"R_kgDORsJ7Nw","slug":"ndesv21__socialclaw","name":"SocialClaw","owner":"ndesv21","fullName":"ndesv21/socialclaw","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"requires_paid_hosted_service","severity":"attention","evidence":"README：\"the CLI works against a deployed SocialClaw workspace via API key auth\" 以及 \"an API key alone is not enough for agent execution; the workspace must also have an active trial or paid plan\" ——开源部分本身无法独立发布内容。","source":"ai","title":"需要付费的托管服务"},{"code":"social_accounts_held_by_third_party","severity":"attention","evidence":"\"Users connect accounts inside the hosted SocialClaw dashboard. The dashboard and API live at `https://getsocialclaw.com`\"：你的 X、Instagram、LinkedIn、TikTok 和 YouTube 账号的访问权限，以及你上传的媒体内容，都保存在这家公司的服务器上。","source":"ai","title":"社交账号掌握在第三方手中"},{"code":"agent_publishes_on_your_behalf","severity":"info","evidence":"MCP 服务器提供了 \"17 validation-first publishing tools over your connected workspace accounts\"，Claude Code 插件的说明是让 Claude \"auto-invoke SocialClaw when a user asks to connect accounts, upload media, schedule posts\"，因此助手可以在你的真实账号上发布内容。","source":"ai","title":"代理代表你发布内容"},{"code":"heavy_keyword_listing","severity":"info","evidence":"package.json 里有一份包含 36 个条目的关键词列表（\"claude\"、\"chatgpt\"、\"cursor\"、\"codex\"、\"openclaw\"、\"clawhub\" 以及每个网络的名字），仓库的 topics 也重复了这些词，这更像是为搜索引擎写的措辞，而不是对代码内容的描述。","source":"ai","title":"关键词堆砌严重"},{"code":"repo_release_behind_published_package","severity":"info","evidence":"GitHub 上唯一的发布版本是 v0.1.6（2026-03-26），而 package.json 已经写到版本 0.1.22，说明 npm 包的进度领先于这里标记的版本，最近一次提交是在本次检查前 76 天。","source":"ai","title":"仓库发布版本落后于已发布的包"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["SocialClaw workspace API key (getsocialclaw.com, with a trial or paid plan)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","ai_skills","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"later","name":"Later","confirmed":true}],"declaredAlternatives":[],"stars":108,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":3,"growthDays":6,"history":[{"date":"2026-11-06","stars":106},{"date":"2026-11-07","stars":106},{"date":"2026-11-08","stars":106},{"date":"2026-11-09","stars":107},{"date":"2026-11-10","stars":108}],"firstPublishedAt":"2026-11-08T23:01:18.837Z","commitSha":"2071b1046251633205e894e974202ca54a79278a","verifiedAt":"2026-11-08T22:49:42.588Z","lastCommitAt":"2026-08-24T10:55:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets the command line, or an AI assistant, schedule and publish posts to 13 social networks.","summary":"SocialClaw is a command-line tool plus a skill and an MCP server for AI assistants such as Claude Code, Cursor and Codex. With it an assistant can upload images, check a schedule file, publish or queue posts and read the numbers afterwards on X, Facebook Pages, Instagram, LinkedIn, Pinterest, TikTok, Discord, Telegram, YouTube, Reddit and WordPress. The code here is only the client: you connect your accounts on the company's website and, as the README says, the workspace needs an active trial or paid plan before anything can be published.","easeRationale":"The same on all three systems: install Node 20 or newer, run `npm install -g socialclaw`, then `socialclaw login`, which opens the company's dashboard to sign in with Google, connect each social account and copy a workspace key. There is no installer and no graphical app in this repository.","costNote":"The code is MIT, but the README says \"an API key alone is not enough for agent execution; the workspace must also have an active trial or paid plan\" and points to getsocialclaw.com/pricing. The price is not in the repository.","notVerified":["What the paid plans cost and what the trial allows","Whether all 13 networks really work, since the README already warns about TikTok photo posts failing after acceptance","What the hosted service stores and for how long, including your media and account tokens","Whether the npm package contains the same code as this repository","Whether a one-person service of this size will keep running"]},"zh-Hans":{"summaryShort":"让命令行，或者一个 AI 助手，为 13 个社交网络安排和发布帖子。","summary":"SocialClaw 是一个命令行工具，外加一个 skill 和一个为 Claude Code、Cursor、Codex 等 AI 助手提供的 MCP 服务器。借助它，助手可以上传图片、检查排期文件、发布或把帖子加入队列，之后还能在 X、Facebook Pages、Instagram、LinkedIn、Pinterest、TikTok、Discord、Telegram、YouTube、Reddit 和 WordPress 上查看数据。这个仓库里的代码只是客户端：你需要在该公司的网站上连接自己的账号，而且正如 README 所说，工作空间需要有有效的试用或付费计划，才能发布任何内容。","easeRationale":"三个系统上的做法相同：安装 Node 20 或更新版本，运行 `npm install -g socialclaw`，然后运行 `socialclaw login`，它会打开该公司的控制面板，让你用 Google 登录、连接各个社交账号并复制一个工作空间密钥。这个仓库里没有安装程序，也没有图形界面应用。","costNote":"代码采用 MIT 许可证，但 README 说 \"an API key alone is not enough for agent execution; the workspace must also have an active trial or paid plan\"，并指向 getsocialclaw.com/pricing。价格信息不在这个仓库里。","notVerified":["付费计划的价格是多少，试用又能用到什么程度","13 个网络是否真的都能用，因为 README 已经提醒 TikTok 的图片发布在通过审核后会失败","托管服务会保存什么数据、保存多久，包括你的媒体文件和账号令牌","npm 包里的代码是否和这个仓库里的一致","这种由一个人维护的服务，规模这么小，是否能一直运行下去"]}}},{"id":"R_kgDOGv9OMg","slug":"neonbjb__tortoise-tts","name":"Tortoise TTS","owner":"neonbjb","fullName":"neonbjb/tortoise-tts","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"存在已知漏洞的固定依赖项（OSV）：transformers：GHSA-29pf-2h5f-8g72、GHSA-37mw-44qp-f5jm、GHSA-37q5-v5qm-c9v8；nbconvert：GHSA-9754-6rhw-gj3h、GHSA-9jmq-rx5f-8jwq、GHSA-xm59-rqc7-hhvf；tornado：GHSA-3hv7-mjh2-fv65、GHSA-3x9g-8vmp-wqvf、GHSA-753j-mpmx-qq6g；pydantic：GHSA-mr82-8j83-vxmv、PYSEC-2026-1812；deepspeed：GHSA-8cp5-3rf8-8gfh、PYSEC-2024-109","source":"check","title":"已知漏洞"},{"code":"voice_cloning_impersonation_risk","severity":"attention","evidence":"该 API 接受真实声音的 `reference_clips`（`tts.tts_with_preset(\"your text here\", voice_samples=reference_clips)`），仓库还附带了一份 `voice_customization_guide.md`，因此任何人的声音都可能从录音中被复制出来；README 中对此没有任何免责声明。","source":"ai","title":"声音克隆被冒用的风险"},{"code":"unmaintained","severity":"attention","evidence":"最近一次提交是在 2024-11-19，距本次检查 719 天，有 349 个未解决的 issue，从未发布过任何版本。","source":"ai","title":"缺乏维护"},{"code":"very_old_dependency_pins","severity":"info","evidence":"requirements.txt 固定了 `tornado==4.2`、`nbconvert==5.3.1`、`pydantic==1.9.1`、`transformers==4.31.0` 和 `deepspeed==0.8.3`，README 自己也提醒，如果不用 conda，\"you will spend a lot of time chasing dependency problems\"。","source":"ai","title":"依赖版本非常陈旧"},{"code":"contradictory_speed_claims","severity":"info","evidence":"README 中一方面说 \"this model is insanely slow ... expect to generate a medium sized sentence every 2 minutes\"，另一方面又说 \"not so slow anymore now we can get a **0.25-0.3 RTF** on 4GB vram\"，但没有说明哪种配置对应哪种结果。","source":"ai","title":"关于速度的说法互相矛盾"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","docker","ffmpeg"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"speechify","name":"Speechify","confirmed":true}],"declaredAlternatives":[],"stars":14894,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":14895},{"date":"2026-11-08","stars":14896},{"date":"2026-11-09","stars":14895},{"date":"2026-11-10","stars":14894}],"firstPublishedAt":"2026-11-08T23:01:18.845Z","commitSha":"8a2563ecabe93c4fb626f876dd0c52c966edef2f","verifiedAt":"2026-11-08T22:50:25.523Z","lastCommitAt":"2024-11-19T18:59:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Read text aloud in many voices, or copy a voice from a few short clips, on your own computer.","summary":"Tortoise is a text-to-speech program made for natural-sounding intonation and many voices. You give it a sentence or a whole text file and it writes out spoken audio; give it a few recorded clips of a voice and it will read in that voice. It is used from the command line or from Python, with scripts for one phrase, for long texts and for streaming. Running it on your machine means a conda environment, Python 3.9 and an NVIDIA card; an Apple Silicon path exists but runs on the processor. A free demo page exists on Hugging Face.","easeRationale":"The quick way is the live demo page on Hugging Face linked in the README, which needs nothing installed. Every local route is terminal work: \"If you want to use this on your own computer, you must have an NVIDIA GPU\", with miniconda, a Python 3.9 environment and a matching PyTorch build on Windows and Linux, or a Docker image; on Apple Silicon you install a nightly PyTorch, DeepSpeed is disabled and you may have to set `PYTORCH_ENABLE_MPS_FALLBACK=1`.","costNote":"Free and Apache 2.0, model weights hosted by Hugging Face. Duplicating the demo page with a graphics card, as the README suggests to skip the queue, is paid on Hugging Face.","notVerified":["How long one sentence really takes on a current graphics card, and on Apple Silicon","Whether the conda and pip installs still complete with these pinned versions","How much disk space the model files take once downloaded from Hugging Face","Whether the community-run demo page stays online and what it does with the audio you upload","Whether the known problems in the pinned packages found by the checks matter for local use"]},"zh-Hans":{"summaryShort":"在你自己的电脑上，用多种声音朗读文本，或从几段简短片段中复制一个声音。","summary":"Tortoise 是一个为自然语调和多种声音设计的文本转语音程序。给它一句话或一整个文本文件，它就会生成朗读的音频；给它几段某个声音的录音片段，它就能用那个声音朗读。可以从命令行或 Python 中使用，并提供了用于单句、长文本和流式输出的脚本。要在你自己的电脑上运行，需要一个 conda 环境、Python 3.9 和一块 NVIDIA 显卡；也有一条 Apple Silicon 的路径，但它运行在处理器上。Hugging Face 上有一个免费的演示页面。","easeRationale":"快捷的方式是 README 中链接的 Hugging Face 在线演示页面，不需要安装任何东西。每一种本地方式都要在终端里操作：\"If you want to use this on your own computer, you must have an NVIDIA GPU\"，在 Windows 和 Linux 上需要 miniconda、一个 Python 3.9 环境和与之匹配的 PyTorch 版本，或者一个 Docker 镜像；在 Apple Silicon 上你需要安装一个 nightly 版本的 PyTorch，DeepSpeed 会被禁用，还可能需要设置 `PYTORCH_ENABLE_MPS_FALLBACK=1`。","costNote":"免费且采用 Apache 2.0 许可证，模型权重托管在 Hugging Face 上。按照 README 的建议，用显卡复制一份演示页面以跳过排队，这在 Hugging Face 上是收费的。","notVerified":["在当前的显卡上，以及在 Apple Silicon 上，一句话实际需要多长时间","使用这些固定版本，conda 和 pip 的安装是否还能顺利完成","从 Hugging Face 下载后，模型文件占用多少磁盘空间","社区维护的演示页面是否还在线，以及它如何处理你上传的音频","检查发现的固定依赖包中的已知问题，是否会影响本地使用"]}}},{"id":"R_kgDON5f9ug","slug":"nerding-io__n8n-nodes-mcp","name":"n8n-nodes-mcp","owner":"nerding-io","fullName":"nerding-io/n8n-nodes-mcp","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖项存在已知漏洞（OSV）：@langchain/core：GHSA-r399-636x-v7f6","source":"check","title":"已知漏洞"},{"code":"runs_commands_you_configure","severity":"attention","evidence":"在 STDIO 传输方式中，凭据里有一个 \"Command: The command to start the MCP server\" 字段和参数；README 的示例是 Command `npx`，Arguments `-y @modelcontextprotocol/server-brave-search`。这个节点会在运行 n8n 的那台机器上启动该程序，npx 会在那一刻下载对应的包，所以无论是你还是其他能访问你 n8n 的人在那里输入了什么，都会在你的服务器上执行。","source":"ai","title":"运行你配置的命令"},{"code":"api_keys_passed_to_other_programs","severity":"info","evidence":"README：docker-compose 中以 `MCP_` 为前缀的环境变量（`MCP_BRAVE_API_KEY`、`MCP_OPENAI_API_KEY`）\"will be automatically passed to your MCP servers when they are executed\"，所以你的密钥会传到节点启动的每一个 MCP 服务器。","source":"ai","title":"api 密钥会传递给其他程序"},{"code":"third_party_security_badge","severity":"info","evidence":"README 开头展示了来自 mseep.ai 的 \"Verified on MseeP\" 徽章和 \"Security Assessment\" 徽章。这些徽章来自一个外部网站，并不是 n8n 或 Radar 做的审核。","source":"ai","title":"第三方认证徽章"},{"code":"deprecated_transport_still_shipped","severity":"info","evidence":"README：\"The Server-Sent Events (SSE) transport is deprecated and replaced by the new HTTP Streamable transport\" 以及 \"SSE is deprecated and will not receive further updates\"。SSE 的凭据仍然保留在这个包里（dist/credentials/McpClientSseApi.credentials.js）。","source":"ai","title":"仍然附带已弃用的传输方式"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["automation","ai_agents","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":3039,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":3039},{"date":"2026-11-08","stars":3039},{"date":"2026-11-09","stars":3039},{"date":"2026-11-10","stars":3039}],"firstPublishedAt":"2026-11-08T23:01:18.852Z","commitSha":"6ee215b30e514251b4334f7dba577d5a94276785","verifiedAt":"2026-11-08T22:42:40.910Z","lastCommitAt":"2026-01-02T02:49:02Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds a node to n8n so your workflows and AI agents can list and run tools on MCP servers.","summary":"This is an add-on (a \"community node\") for n8n, the workflow automation platform. Once installed, your n8n workflows can connect to MCP servers and list tools, run a tool, read resources or fetch prompts, so an AI agent inside n8n can use outside tools. It is for people who already run their own n8n, on a computer or in Docker. You also have to set the environment variable N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE to true to use it inside AI Agent nodes, and you have to provide the MCP servers yourself.","easeRationale":"There is no installer and no binary in any of the 29 releases: you first need your own n8n running (Node.js or Docker), then install the community node, set an environment variable and configure a command or URL for each MCP server. The work is the same on Windows, Mac and Linux, so all three get D.","costNote":"MIT licensed. The MCP servers you connect to may need their own paid API keys (the README uses Brave Search and OpenAI as examples).","notVerified":["Whether it still installs and works with the n8n versions released after the last commit (2026-01-02)","What the MCP servers you connect to actually do with your data or your keys","The real monthly cost of the paid APIs behind those MCP servers","Whether the known vulnerability in the pinned @langchain/core 0.3.43 can be reached through this node"]},"zh-Hans":{"summaryShort":"为 n8n 添加一个节点，让你的工作流和 AI 代理可以列出并运行 MCP 服务器上的工具。","summary":"这是面向 n8n（工作流自动化平台）的一个附加组件（一个 \"community node\"）。安装后，你的 n8n 工作流可以连接到 MCP 服务器，列出工具、运行工具、读取资源或获取提示词，这样 n8n 里的 AI 代理就能使用外部工具。它适合已经自己运行 n8n（在电脑上或 Docker 中）的人。你还需要把环境变量 N8N_COMMUNITY_PACKAGES_ALLOW_TOOL_USAGE 设为 true，才能在 AI Agent 节点中使用它，并且你需要自己提供 MCP 服务器。","easeRationale":"在全部 29 个发行版本中都没有安装程序，也没有二进制文件：你首先需要自己运行好 n8n（Node.js 或 Docker），然后安装这个 community node，设置一个环境变量，并为每个 MCP 服务器配置命令或 URL。这些步骤在 Windows、Mac 和 Linux 上都一样，所以三者都是 D。","costNote":"采用 MIT 许可证。你连接的 MCP 服务器可能需要各自的付费 API 密钥（README 中以 Brave Search 和 OpenAI 为例）。","notVerified":["它是否仍能在最后一次提交（2026-01-02）之后发布的 n8n 版本上安装和运行","你连接的 MCP 服务器到底会如何处理你的数据或密钥","这些 MCP 服务器背后付费 API 的实际月费","锁定版本 @langchain/core 0.3.43 中已知的漏洞是否可以通过这个节点被利用"]}}},{"id":"R_kgDOQfttOw","slug":"nextlevelbuilder__ui-ux-pro-max-skill","name":"UI UX Pro Max","owner":"nextlevelbuilder","fullName":"nextlevelbuilder/ui-ux-pro-max-skill","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"paid_premium_upsell","severity":"info","evidence":"README.md「Basic vs. Premium Version Comparison」：\"For more details on upgrading to the Premium tier, visit uupm.cc.\" 高级版功能（品牌身份、logo、图片生成、优先支持）不在这个仓库里。","source":"ai","title":"向付费高级版引流"},{"code":"readme_advertises_other_products_of_the_same_author","severity":"info","evidence":"README.md 开头链接到 NextLevelBuilder.io、GoClaw.sh、AgentKit.best 和 TOSE.sh，还有一个 PayPal 捐赠徽章，以及另一个技能 3Dviz Pro Max（3dviz.dev）的横幅，这些都放在说明项目本身之前。","source":"ai","title":"README 在推广同一作者的其他产品"},{"code":"demos_labelled_as_not_coming_from_the_skill","severity":"info","evidence":"README.md 把自己的示例标为未经证实：3Dviz 的横幅写着 \"An author-supplied project recorded before this skill existed ... not an English demo or a runtime test of the skill\"，仪表盘的前后对比则是 \"Illustrative, community-contributed example; not an official demo, benchmark, or runtime test of the skill.\"","source":"ai","title":"演示被标注为并非来自这个技能本身"},{"code":"install_files_come_from_npm_not_from_this_repository","severity":"info","evidence":"README.md「Architecture & Contributing」：\"Always use the CLI to install\"，使用 `npm install -g ui-ux-pro-max-cli`，因为 \"All platform-specific files ... are now generated dynamically by the CLI\"。真正安装到你电脑上的，是用仓库的 `NPM_TOKEN` 密钥发布的那个 npm 包。","source":"ai","title":"安装的文件来自 npm，不是来自这个仓库"}],"platforms":["windows","mac","linux"],"tools":["python","node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","design_cad"],"alternativeTo":[{"slug":"tailwind-plus","name":"Tailwind Plus","confirmed":true},{"slug":"v0-by-vercel","name":"v0 by Vercel","confirmed":true}],"declaredAlternatives":[{"target":"benzinga-pro","stance":"claims","evidence":"Replace `Minimalism & Swiss Style` with your preferred [catalog style](src/ui-ux-pro-max/data/styles.csv) or visual direction"},{"target":"dbeaver-pro","stance":"claims","evidence":"Replace `Minimalism & Swiss Style` with your preferred [catalog style](src/ui-ux-pro-max/data/styles.csv) or visual direction"},{"target":"moz-pro","stance":"claims","evidence":"Replace `Minimalism & Swiss Style` with your preferred [catalog style](src/ui-ux-pro-max/data/styles.csv) or visual direction"},{"target":"deepl-pro","stance":"claims","evidence":"Replace `Minimalism & Swiss Style` with your preferred [catalog style](src/ui-ux-pro-max/data/styles.csv) or visual direction"}],"stars":147765,"downloads":8557,"downloadsWeek":0,"downloadsDays":7,"growth":3003,"growthDays":7,"history":[{"date":"2026-10-26","stars":140929},{"date":"2026-10-27","stars":141586},{"date":"2026-10-28","stars":142086},{"date":"2026-10-29","stars":142550},{"date":"2026-10-30","stars":143044},{"date":"2026-10-31","stars":143424},{"date":"2026-11-01","stars":143863},{"date":"2026-11-02","stars":144287},{"date":"2026-11-03","stars":144762},{"date":"2026-11-04","stars":145177},{"date":"2026-11-05","stars":145674},{"date":"2026-11-06","stars":146133},{"date":"2026-11-07","stars":146512},{"date":"2026-11-08","stars":146880},{"date":"2026-11-09","stars":147332},{"date":"2026-11-10","stars":147765}],"firstPublishedAt":"2026-10-27T01:10:12.665Z","commitSha":"9a3f14f7423037ccb8feeb4fee7482291b9c820e","verifiedAt":"2026-10-31T01:40:45.189Z","lastCommitAt":"2026-11-10T13:03:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Gives AI coding assistants a library of design styles, colors and fonts so the screens they build look more polished.","summary":"UI UX Pro Max is a skill (a set of instructions and data) you add to AI coding assistants such as Claude Code, Cursor or Codex. When you ask the assistant to build a page or app screen, it searches a built-in catalog of styles, color palettes, font pairings and design rules and suggests a matching design system. It is for people who build websites or apps with an AI assistant. You need an AI coding assistant, Python 3 for the search script, and usually Node.js to run the installer. A paid premium version is advertised separately.","easeRationale":"C on Windows, Mac and Linux alike: none of the 41 releases carries a program file, so on each one you install Python 3 and usually Node.js, then run commands such as `npm install -g ui-ux-pro-max-cli` and `uipro init --ai claude` inside your project. You also need an AI coding assistant for it to work in.","costNote":"This repository is free (MIT). A paid Premium version with extra brand, logo and asset features is advertised at uupm.cc; no price is stated. It runs inside an AI coding assistant, which usually has its own subscription.","notVerified":["Whether the design suggestions actually improve the screens your AI assistant produces.","Whether the installer and Python search script work the same on Windows as the macOS/Linux examples in the README.","Whether the files installed by the npm CLI match the source in this repository.","Which skill features each AI assistant actually supports, since the README lists over 20 of them.","Whether the catalog numbers in the README (192 reasoning rules, 79 styles, 1,934 approved Google Fonts) match the data files that get installed."]},"zh-Hans":{"summaryShort":"为 AI 编程助手提供一套风格、配色和字体库，让它们做出来的界面更精致。","summary":"UI UX Pro Max 是一个添加到 Claude Code、Cursor 或 Codex 等 AI 编程助手上的技能（一组指令和数据）。当你让助手搭建一个页面或应用界面时，它会在内置目录里搜索风格、配色方案、字体搭配和设计规则，然后给出一套匹配的设计系统建议。它适合用 AI 助手搭建网站或应用的人。你需要一个 AI 编程助手、用于搜索脚本的 Python 3，通常还需要 Node.js 来运行安装程序。另外单独宣传了一个付费高级版。","easeRationale":"在 Windows、Mac 和 Linux 上都是 C 级：41 个发布版本里没有一个带程序文件，所以在每个系统上你都要安装 Python 3，通常还要装 Node.js，然后在项目里运行像 `npm install -g ui-ux-pro-max-cli` 和 `uipro init --ai claude` 这样的命令。你还需要一个 AI 编程助手才能让它运作起来。","costNote":"这个仓库本身免费（MIT 许可）。uupm.cc 上单独宣传了一个付费的高级版，带有额外的品牌、logo 和素材功能，没有公布价格。它运行在 AI 编程助手里面，而助手通常有自己的订阅费用。","notVerified":["这些设计建议是否真的能改善你的 AI 助手做出来的界面。","安装程序和 Python 搜索脚本在 Windows 上，是否和 README 里的 macOS/Linux 示例表现一致。","npm CLI 安装的文件是否和这个仓库里的源代码一致。","每个 AI 助手实际支持这个技能的哪些功能，因为 README 列出了 20 多种助手。","README 里的目录数字（192 条推理规则、79 种风格、1934 种经过审核的 Google 字体）是否和安装的数据文件一致。"]}}},{"id":"R_kgDOUpvcvQ","slug":"niko1221__strata","name":"Strata","owner":"Niko1221","fullName":"Niko1221/Strata","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"39 天内获得 38513 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：torch: GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47, PYSEC-2026-139；transformers: GHSA-29pf-2h5f-8g72, GHSA-69w3-r845-3855, GHSA-fgcw-684q-jj6r；pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h","source":"check","title":"已知漏洞"},{"code":"needs_a_strong_graphics_card","severity":"attention","evidence":"硬件表格只列出了 NVIDIA RTX 20/30/40/50、特定的 AMD Radeon RX 7000/9000 和 AI PRO 显卡、Linux 上的 Intel Arc、Apple Silicon M1 及更新机型，以及 Ryzen AI Max；Radeon RX 6800/6900 被标为 \"experimental\"。更老或配置较低的机器不在支持范围内。","source":"ai","title":"需要一块强力显卡"},{"code":"very_large_downloads_and_disk_use","severity":"attention","evidence":"README 中 DeepSeekV4.1 的表格：下载体积从 \"239 GB\" 到 \"274 GB\"，需要 \"about 245 GB\" 到 \"about 280 GB\" 的空闲磁盘空间；发行版中单个引擎文件可达 903 MB，draft 层还会再增加约 5 GB。","source":"ai","title":"下载量和磁盘占用都非常大"},{"code":"install_by_piping_a_script_from_the_internet","severity":"attention","evidence":"README，终端应用部分：\"curl -fsSL https://github.com/Niko1221/Strata/releases/latest/download/install.sh -o install.sh && sh install.sh\"，在 Windows 上则是 \"irm https://github.com/Niko1221/Strata/releases/latest/download/install.ps1 | iex\"。桌面安装程序可以避免这种方式。","source":"ai","title":"安装方式是从网络管道传输脚本"},{"code":"local_api_can_be_opened_to_the_network","severity":"attention","evidence":"setup.py 文档写道 \"--host 0.0.0.0 --api-key KEY (reach it over IPv4; use --host :: for IPv6)\"，Docker 示例使用 \"API_KEY=change-me-please-123\"。以这种方式开放后，任何能访问该端口的人都可以使用这个模型，除非设置了密钥。","source":"ai","title":"本地 API 可以对网络开放"},{"code":"model_files_come_from_other_people","severity":"info","evidence":"setup.py 按提交固定了来自第三方（ISTA-DASLab、unsloth、agentionai、ukisai）的 Hugging Face 仓库，并校验每个文件的 SHA-256；README 指向 \"Credits and model licenses\" 说明其条款。","source":"ai","title":"模型文件来自其他人"},{"code":"checksum_check_can_be_switched_off","severity":"info","evidence":"setup.py：\"STRATA_ALLOW_UNPINNED=1: a model file whose pinned version is gone from Hugging Face is taken from its current files instead, and is NOT checked against a SHA-256.\" 默认是关闭的，启用时会提示。","source":"ai","title":"校验码检查可以关闭"},{"code":"flagged_packages_belong_to_an_optional_step","severity":"info","evidence":"torch 和 transformers 的版本位于 requirements-nvfp4-local.txt 中，该文件说明 \"Packages only the NVIDIA FP4-Local (QAD) conversion needs ... nothing of them is shipped with Strata\"；pillow 12.3.0 位于普通的 requirements.txt 中，每次安装都会安装它。","source":"ai","title":"被标记的包属于一个可选步骤"}],"platforms":["windows","mac","linux"],"tools":["gpu_any"],"buildTools":["python","docker","gpu_nvidia"],"disk":{"minMB":108,"maxMB":323,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["local_ai","development"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"claude-code","name":"Claude Code","confirmed":false}],"declaredAlternatives":[],"stars":42839,"downloads":3276985,"downloadsWeek":974058,"downloadsDays":7,"growth":3544,"growthDays":7,"history":[{"date":"2026-10-25","stars":32840},{"date":"2026-10-26","stars":33741},{"date":"2026-10-27","stars":34578},{"date":"2026-10-28","stars":35337},{"date":"2026-10-29","stars":35998},{"date":"2026-10-30","stars":36511},{"date":"2026-10-31","stars":37090},{"date":"2026-11-01","stars":37744},{"date":"2026-11-02","stars":38457},{"date":"2026-11-03","stars":39295},{"date":"2026-11-04","stars":39889},{"date":"2026-11-05","stars":40512},{"date":"2026-11-06","stars":41024},{"date":"2026-11-07","stars":41455},{"date":"2026-11-08","stars":41950},{"date":"2026-11-09","stars":42462},{"date":"2026-11-10","stars":42839}],"firstPublishedAt":"2026-10-25T23:19:16.907Z","commitSha":"ffa80d73133ea5c548aefe57889a16e8085a938f","verifiedAt":"2026-11-03T02:43:01.468Z","lastCommitAt":"2026-11-10T15:16:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs large AI models on your own PC, with a chat window and a local address your other apps can use.","summary":"Strata installs a large AI model on your own computer and gives you a window to chat with it, read code and send pictures. It also opens an address on your machine (127.0.0.1:8080) that speaks like the OpenAI and Anthropic services, so coding tools such as Claude Code or Codex CLI can use the local model instead of a paid one. There is one installer per system, and on first start it downloads the right engine for your graphics card plus the model files, which are large.","easeRationale":"Each system has one ready file (Strata-...-windows-x64-setup.exe, a macOS .dmg, a Linux AppImage) and the README says you \"install the app, choose a model, and start chatting\". It is B because the first start still downloads the engine for your graphics card and a model of several gigabytes, and you have to pick which size fits your machine.","costNote":"MIT-licensed and free; the README asks only for a voluntary coffee on Buy Me a Coffee. The model files come free from Hugging Face, but each one carries its own licence and you pay in disk space and electricity.","notVerified":["Whether the speeds in the README's tables happen on your own machine, since they were measured by the project.","Whether the one-click install really finds and installs the right engine for every listed card.","How much disk and memory you end up using after the model, the draft layer and the engine are in place.","Whether the released installers were built from the code published here.","Whether the answers of a heavily shrunk model (Q2, IQ3) are good enough for your work.","Whether the pillow version with the reported weaknesses is reached in normal use."]},"zh-Hans":{"summaryShort":"在你自己的电脑上运行大型 AI 模型，带聊天窗口和本地地址供你的其他应用调用。","summary":"Strata 在你自己的电脑上安装一个大型 AI 模型，并提供一个窗口让你和它聊天、读代码、发图片。它还会在你的机器上开放一个地址（127.0.0.1:8080），使用方式和 OpenAI、Anthropic 的服务一样，这样 Claude Code 或 Codex CLI 之类的编程工具就能用这个本地模型，而不用付费模型。每个系统都有一个安装程序，首次启动时会下载适合你显卡的引擎和模型文件，这些文件体积很大。","easeRationale":"每个系统都有一个现成文件（Strata-...-windows-x64-setup.exe、一个 macOS 的 .dmg、一个 Linux 的 AppImage），README 说你 \"install the app, choose a model, and start chatting\"。评级是 B，因为首次启动仍需要下载适合你显卡的引擎和一个数 GB 的模型，你还得挑选适合自己机器的尺寸。","costNote":"采用 MIT 许可且免费；README 只在 Buy Me a Coffee 上请求自愿的小额赞赏。模型文件可以从 Hugging Face 免费获取，但每个文件都有自己的许可条款，你付出的是磁盘空间和电费。","notVerified":["README 表格中的速度是否能在你自己的机器上达到，因为这些数据是项目自己测量的。","一键安装是否真的能为列出的每一款显卡找到并安装正确的引擎。","模型、draft 层和引擎都安装完成后，最终会占用多少磁盘和内存。","发布的安装程序是否是从这里公开的代码构建的。","一个大幅压缩的模型（Q2、IQ3）给出的回答是否足够满足你的工作需要。","带有已报告缺陷的 pillow 版本在正常使用中是否会被触及。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTE3MTU0OA==","slug":"nlohmann__json","name":"JSON for Modern C++","owner":"nlohmann","fullName":"nlohmann/json","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":50865,"downloads":40124947,"downloadsWeek":593983,"downloadsDays":7,"growth":27,"growthDays":7,"history":[{"date":"2026-10-26","stars":50808},{"date":"2026-10-27","stars":50812},{"date":"2026-10-28","stars":50813},{"date":"2026-10-29","stars":50819},{"date":"2026-10-30","stars":50826},{"date":"2026-10-31","stars":50829},{"date":"2026-11-01","stars":50831},{"date":"2026-11-02","stars":50835},{"date":"2026-11-03","stars":50838},{"date":"2026-11-04","stars":50840},{"date":"2026-11-05","stars":50844},{"date":"2026-11-06","stars":50845},{"date":"2026-11-07","stars":50849},{"date":"2026-11-08","stars":50851},{"date":"2026-11-09","stars":50855},{"date":"2026-11-10","stars":50865}],"firstPublishedAt":"2026-10-27T01:10:12.666Z","commitSha":"9c349b1d7c393c6bc6dbbe4191e332e3bb4100b8","verifiedAt":"2026-11-01T01:38:20.440Z","lastCommitAt":"2026-11-10T21:30:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A single C++ header file that lets programmers read and write JSON data inside their own programs.","summary":"This is a building block for programmers, not an app you open. It is one header file you drop into a C++ project so the program can read, create and change JSON data, plus binary formats such as CBOR, BSON and MessagePack. To use it you need a C++11 compiler; the project is tested on Windows, macOS and Linux. There is nothing to install for a person who does not program.","easeRationale":"On every platform this is developer-only: the releases carry no program to run, and the only way to use it is to add the header to your own C++ code and compile it, optionally through CMake, Meson, Bazel or a package manager.","costNote":"MIT license; the author accepts sponsorships on GitHub.","notVerified":["Whether it compiles with your particular compiler and project settings","Whether the single-header file in the repository matches the one attached to each release","How much it slows your build or grows your program's size"]},"zh-Hans":{"summaryShort":"一个单独的C++头文件，让程序员可以在自己的程序中读写JSON数据。","summary":"这是给程序员使用的一个构建模块，不是可以打开的应用。它是一个你放进C++项目中的单独头文件，让程序能够读取、创建和修改JSON数据，还支持CBOR、BSON和MessagePack等二进制格式。要使用它你需要一个C++11编译器；该项目在Windows、macOS和Linux上进行测试。对于不编程的人来说，没有任何东西需要安装。","easeRationale":"在所有平台上，这都仅面向开发者：发行版中不包含任何可运行的程序，唯一的使用方式是把这个头文件添加到你自己的C++代码中并编译，也可以借助CMake、Meson、Bazel或某个包管理器来完成。","costNote":"MIT许可证；作者在GitHub上接受赞助。","notVerified":["它能否在你具体使用的编译器和项目设置下编译通过","仓库中的单一头文件是否与每个release中附带的文件一致","它会让你的编译变慢多少，或让程序体积增大多少"]}}},{"id":"R_kgDOJPPS-Q","slug":"nomic-ai__gpt4all","name":"GPT4All","owner":"nomic-ai","fullName":"nomic-ai/gpt4all","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"low","signals":[{"code":"development_appears_stopped","severity":"info","evidence":"最后一次提交是在 2025-05-27，最新版本 v3.10.0 发布于 2025-02-25——大约一年半没有新版本，同时有 775 个未解决的 issue。该仓库没有被归档。","source":"ai","title":"开发看起来已经停止"},{"code":"downloads_large_files_after_install","severity":"info","evidence":"README 自己的 Python 示例指出，加载 \"Meta-Llama-3-8B-Instruct.Q4_0.gguf\" 会 \"downloads / loads a 4.66GB LLM\"。桌面应用下载模型的方式也是一样，所以实际占用的磁盘和流量远大于安装程序本身。","source":"ai","title":"安装后会下载大文件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":710,"maxMB":2130,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["local_ai","productivity"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"typingmind","name":"TypingMind","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[],"stars":77353,"downloads":240199,"downloadsWeek":2574,"downloadsDays":6,"growth":5,"growthDays":7,"history":[{"date":"2026-11-04","stars":77341},{"date":"2026-11-05","stars":77339},{"date":"2026-11-06","stars":77338},{"date":"2026-11-07","stars":77343},{"date":"2026-11-08","stars":77347},{"date":"2026-11-09","stars":77353},{"date":"2026-11-10","stars":77353}],"firstPublishedAt":"2026-11-08T23:01:18.860Z","commitSha":"b666d16db5aeab8b91aaf7963adcee9c643734d7","verifiedAt":"2026-11-08T22:43:20.385Z","lastCommitAt":"2025-05-27T19:46:52Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Install one program and chat with AI language models that run on your own computer, offline.","summary":"GPT4All is a desktop chat program for AI language models that run on your own machine, so your questions do not leave the computer. There are ready installers for Windows (also ARM), macOS and Linux; after installing, you pick and download a model inside the app. A feature called LocalDocs lets you ask questions about your own files. A Python package (pip install gpt4all) is available for people who write code. No graphics card is required, but Windows and Linux need an Intel Core i3 2nd Gen / AMD Bulldozer or better, and macOS needs 12.6 or newer.","easeRationale":"Every release ships a ready installer for all three systems (gpt4all-installer-win64-v3.10.0.exe 710 MB, the macOS .dmg 305 MB, the Linux .run 699 MB), and the only extra step is choosing and downloading a model inside the app, so all three get B. The Linux build is x86-64 only, so Linux ARM machines are not covered.","costNote":"MIT licensed and, per the repository description, available for commercial use. The models you download in the app are free.","notVerified":["Whether the February 2025 build still runs well on current Windows, macOS and Linux versions","How much disk space you end up using after downloading one or more models","How fast the answers are on a computer without a graphics card","Whether the app sends any usage or analytics data to Nomic"]},"zh-Hans":{"summaryShort":"安装一个程序，就能在自己的电脑上离线与 AI 语言模型聊天。","summary":"GPT4All 是一个桌面聊天程序，背后是在你自己的设备上运行的 AI 语言模型，所以你的问题不会离开这台电脑。它为 Windows（也支持 ARM）、macOS 和 Linux 提供现成的安装程序；安装后，你在应用里挑选并下载模型。一个叫 LocalDocs 的功能可以让你针对自己的文件提问。还有一个 Python 包（pip install gpt4all）提供给写代码的人使用。不需要显卡，但 Windows 和 Linux 需要 Intel Core i3 第二代 / AMD Bulldozer 或更好的处理器，macOS 需要 12.6 或更新版本。","easeRationale":"每个发行版本都为三大系统提供现成的安装程序（gpt4all-installer-win64-v3.10.0.exe，710 MB；macOS 的 .dmg，305 MB；Linux 的 .run，699 MB），唯一的额外步骤是在应用里挑选并下载一个模型，所以三者都是 B。Linux 构建版只支持 x86-64，所以 Linux ARM 设备不在覆盖范围内。","costNote":"采用 MIT 许可证，根据仓库的描述，可用于商业用途。你在应用里下载的模型是免费的。","notVerified":["2025 年 2 月的版本是否仍能在当前的 Windows、macOS 和 Linux 版本上正常运行","下载一个或多个模型之后，最终会占用多少磁盘空间","在没有显卡的电脑上，回答速度有多快","应用是否会向 Nomic 发送任何使用或分析数据"]}}},{"id":"R_kgDOVj7lWA","slug":"notrexed__hookshot","name":"Hookshot","owner":"NotRexed","fullName":"NotRexed/Hookshot","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","linux":"C"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"5 天内获得 6465 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"needs_your_own_copy_of_a_commercial_game","severity":"attention","evidence":"readme.md：\"This repository does not contain the game, keys or an executable\"，而构建器只接受 program 0400486024DEC000、update v131072、build e0f3ccfa。也就是说，在你从自己的主机或副本中提取出这款付费任天堂游戏的文件之前，这个程序毫无用处。要取得这些文件，就必须越过主机的保护机制，即使游戏是你买的，这在不少国家也受到限制；仓库本身不提供游戏，也不提供密钥。","source":"ai","title":"需要你自己拥有的商业游戏副本"},{"code":"windows_file_downloads_the_rest_at_run_time","severity":"info","evidence":"Windows 发布文件 HookshotBuilder.exe 只有 0 MB。readme.md：\"Run it. It downloads the rest of Hookshot Builder (about 500 MB), checks it, and opens it.\" 这次下载到的内容，在发布页上是看不到的。","source":"ai","title":"Windows 文件运行时才下载其余内容"},{"code":"readme_asks_for_antivirus_exclusions","severity":"attention","evidence":"readme.md 的提醒：\"Some users report the file in releases to be a Trojan - it is not. Feel free to run it through any antivirus software\"。问题排查表在 \"Your antivirus removed part of the build tools\" 一项下还写道：\"A false alarm on unsigned programs. Restore the files from Windows Security's Protection history, or add %LOCALAPPDATA%\\HookshotBuilder to its exclusions\"。","source":"ai","title":"README 要求把它加入杀毒软件白名单"},{"code":"experimental_and_changing_fast","severity":"info","evidence":"readme.md 的 Project status：\"Experimental. Most testing so far has been around the start of the game.\" 表格列出了尚未解决的崩溃，以及离开科奇里森林之后出现的条纹阴影；AMD 一栏写着 \"Some tested\"（RX 6600、RX 7800 XT、RX 7900 XTX，其中 RX 6600 在德库树里会丢失 GPU），Intel 一栏写着 \"Few reports, many of them black screens or crashes\"。项目自 2026-11-05 创建以来已有 51 次发布。","source":"ai","title":"尚属实验性且变化极快"},{"code":"recompiles_a_commercial_game_from_console_files","severity":"high","evidence":"仓库描述：\"A Native Recompilation Of The Legend Of Zelda: Ocarina Of Time Nintendo Switch 2 Remake\"。readme.md：\"Hookshot is a static recompilation. The game's code is translated to C and compiled into a Windows or Linux program, the console's system software is reimplemented, and the Switch 2 graphics are rebuilt on Vulkan. All of this happens on your PC, from your own copy of the game\"；\"This repository does not contain the game, keys or an executable, and it does not explain how to get them. You need your own copy.\" 构建只接受 \"program 0400486024DEC000, update v131072, build e0f3ccfa\"。主机厂商的条款限制把游戏文件从主机中导出，许多国家也有法律限制绕过复制保护，哪怕游戏是你买的；游戏的代码和文件属于发行方。","source":"ai","title":"用主机文件重新编译一款商业游戏"}],"platforms":["windows","linux"],"tools":["gpu_any"],"buildTools":["python","node","ffmpeg"],"disk":{"minMB":0,"maxMB":1,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":6369,"downloads":1168043,"downloadsWeek":901903,"downloadsDays":2,"growth":6369,"growthDays":5,"history":[{"date":"2026-11-08","stars":2836},{"date":"2026-11-09","stars":4724},{"date":"2026-11-10","stars":6369}],"firstPublishedAt":"2026-11-07T22:56:16.899Z","commitSha":"9447e1de5e01ffab15a179ab4b55a1320fe14231","verifiedAt":"2026-11-10T23:55:31.624Z","lastCommitAt":"2026-11-10T22:39:01Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Builds a PC version of a Nintendo Switch 2 Zelda game from your own copy, to play on Windows or Linux with no emulator.","summary":"Hookshot translates the code of The Legend of Zelda: Ocarina of Time (Nintendo Switch 2) into a program that runs natively on Windows or Linux. A builder app does the work on your PC: you give it your own copy of the game at one exact version, it compiles for 10 to 45 minutes, and then a launcher starts and updates the game. It is for players with a recent graphics card who already own the game. You need about 60 GB free while building, 16 GB of RAM, and a card with Vulkan 1.3 and mesh shaders.","easeRationale":"On Windows, the best case, you download HookshotBuilder.exe and click through a window with nothing else to install, but you must supply your own game file at one exact version, keep about 60 GB free and wait 10 to 45 minutes for the build, on a card with Vulkan 1.3 and mesh shaders (NVIDIA RTX 20, AMD RX 6000 or Intel Arc). Linux is the same work with an AppImage you first have to allow to run as a program.","notVerified":["Whether the game really reaches and keeps 60 fps on your PC, and whether it plays past the first areas without crashing","What the 500 MB that HookshotBuilder.exe downloads at run time actually contains","The real disk use: about 60 GB is needed while building and the build files (about 4 GB) are said to be deleted at the end","How long the build takes on your CPU and memory, within the 10 to 45 minutes stated","Whether it works on AMD and Intel cards, which the project itself reports as little tested"]},"zh-Hans":{"summaryShort":"用你自己的游戏副本，构建出 Nintendo Switch 2 塞尔达游戏的 PC 版本，在 Windows 或 Linux 上免模拟器运行。","summary":"Hookshot 把《塞尔达传说：时之笛》（Nintendo Switch 2）的代码翻译成可以在 Windows 或 Linux 上原生运行的程序。一个构建器程序会在你的电脑上完成这项工作：你提供自己那份特定版本的游戏副本，它编译 10 到 45 分钟，然后由启动器打开并更新游戏。它面向已经拥有该游戏、且显卡较新的玩家。构建过程中需要约 60 GB 可用空间、16 GB 内存，以及支持 Vulkan 1.3 和网格着色器的显卡。","easeRationale":"在最顺利的 Windows 上，你下载 HookshotBuilder.exe，在窗口里一路点下去即可，不用再装别的东西，但必须自备特定版本的游戏文件，留出约 60 GB 空间，并等待 10 到 45 分钟的编译，显卡还要支持 Vulkan 1.3 和网格着色器（NVIDIA RTX 20、AMD RX 6000 或 Intel Arc）。Linux 上的工作量相同，只是先要把 AppImage 设为可执行。","notVerified":["游戏在你的电脑上是否真能达到并稳定在 60 fps，以及过了最初几个区域后会不会崩溃","HookshotBuilder.exe 运行时下载的那 500 MB 里究竟是什么","真实的磁盘占用：构建期间需要约 60 GB，据称构建文件（约 4 GB）会在结束时删除","在你的处理器和内存上编译需要多久，说明中给出的区间是 10 到 45 分钟","在 AMD 和 Intel 显卡上是否可用，项目自己也说这两类显卡测试很少"]}}},{"id":"R_kgDOPRF1Gw","slug":"nousresearch__hermes-agent","name":"Hermes Agent","owner":"NousResearch","fullName":"NousResearch/hermes-agent","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"468 天内获得 259894 个星标（约每天 555 个）","source":"check","title":"异常的星标增长速度"},{"code":"install_runs_a_script_from_the_internet","severity":"attention","evidence":"文档里的安装方式是 `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash`，Windows 上是 `iex (irm https://hermes-agent.nousresearch.com/install.ps1)`。两者都是下载一个脚本后立即运行，所以那个地址当时提供的内容会直接在你的机器上执行，之后它还会拉取 Python、Node.js、ripgrep、FFmpeg，以及在缺少 Git 时拉取一个 Git 构建。","source":"ai","title":"安装过程会运行一个来自网络的脚本"},{"code":"readme_asks_you_to_add_an_antivirus_exception","severity":"attention","evidence":"README 故障排查部分：当防病毒软件把 `%LOCALAPPDATA%\\hermes\\bin\\uv.exe` 隔离时，它说 \"this is a false positive\"，并让你运行 `Add-MpPreference -ExclusionPath \"$env:LOCALAPPDATA\\hermes\\bin\"`，以及 \"Whitelist the folder, not the file hash\"。这样一来，程序持续更新的那个文件夹就不再受防病毒软件检查。README 也提供了对照 Astral 签名发行版验证文件的命令。","source":"ai","title":"readme 要求你给防病毒软件加例外"},{"code":"acts_on_your_computer_and_unattended","severity":"attention","evidence":"README 宣传了 \"40+ tools\"、六种终端后端（包括本地和 SSH）、生成子代理，以及一个 \"Built-in cron scheduler with delivery to any platform. Daily reports, nightly backups, weekly audits — all in natural language, running unattended\"。它还会 \"creates skills from experience\" 并在使用过程中不断改进，所以它的行为会随时间变化。","source":"ai","title":"在你的电脑上无人值守地行动"},{"code":"dashboard_holds_your_keys_and_must_stay_local","severity":"attention","evidence":"`docker-compose.yml` 中的说明：\"The dashboard service binds to 127.0.0.1 by default. It stores API keys; exposing it on LAN without auth is unsafe... do NOT pass --insecure --host 0.0.0.0.\" Windows 的 compose 文件在仅限本机的端口映射后面用 `--insecure` 运行面板，Linux 的则使用 `network_mode: host`，让容器获得你机器的整个网络。","source":"ai","title":"面板保存你的密钥，必须保持本地访问"},{"code":"can_import_api_keys_from_your_other_agents","severity":"attention","evidence":"README 迁移部分：`hermes setup` 会查找 `~/.openclaw`、`~/.claude`、`~/.codex` 和 `~/.cursor`，`hermes import-agent openclaw --migrate-secrets` 会带来 \"allowlisted secrets (Telegram, OpenRouter, OpenAI, Anthropic, ElevenLabs)\"，连同你的记忆和命令审批列表一起迁移。","source":"ai","title":"可以导入你其他代理的 api 密钥"},{"code":"very_large_backlog_of_open_issues","severity":"info","evidence":"项目数据：52368 个未解决的 issue，对比 259894 个星标和 45 个发行版，最新版本号为 v0.22.6。代码注释里引用的 issue 编号甚至远高于未解决的数量（例如 Dockerfile 中的 \"See #70480\"）。","source":"ai","title":"未解决 issue 的数量非常庞大"},{"code":"self_improvement_claims_are_the_owners","severity":"info","evidence":"README 声称自己是 \"the only agent with a built-in learning loop\" 和 \"The self-improving AI agent\"。这些是项目自己对自己的说法，而不是经过测量的结果。","source":"ai","title":"自我提升的说法来自项目方自己"}],"platforms":["windows","mac","linux"],"tools":["python","node","docker","ffmpeg","gpu_any"],"buildTools":["python","node","docker","gpu_any","ffmpeg"],"disk":{"minMB":1050,"maxMB":4950,"basis":"heuristic"},"apiKeys":["A model provider key: Nous Portal, OpenRouter, OpenAI, Fireworks, Google AI Studio, Hugging Face and many others","Chat platform tokens (Telegram, Discord, Slack, WhatsApp, Signal, Microsoft Teams, Google Chat) for the messaging gateway"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","local_ai","automation"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true},{"slug":"manus","name":"Manus","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[],"stars":262415,"downloads":28178,"downloadsWeek":0,"downloadsDays":7,"growth":2163,"growthDays":7,"history":[{"date":"2026-10-25","stars":257081},{"date":"2026-10-26","stars":257416},{"date":"2026-10-27","stars":257757},{"date":"2026-10-28","stars":258078},{"date":"2026-10-29","stars":258520},{"date":"2026-10-30","stars":258884},{"date":"2026-10-31","stars":259182},{"date":"2026-11-01","stars":259529},{"date":"2026-11-02","stars":259871},{"date":"2026-11-03","stars":260252},{"date":"2026-11-04","stars":260526},{"date":"2026-11-05","stars":260844},{"date":"2026-11-06","stars":261171},{"date":"2026-11-07","stars":261468},{"date":"2026-11-08","stars":261728},{"date":"2026-11-09","stars":262038},{"date":"2026-11-10","stars":262415}],"firstPublishedAt":"2026-10-25T23:19:17.795Z","commitSha":"ad2b705b4995a958b0a795e9097be661e2d1ef63","verifiedAt":"2026-11-03T02:44:00.445Z","lastCommitAt":"2026-11-10T22:53:43Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An AI helper you run yourself that works in the terminal or in chat apps, runs commands for you and remembers past sessions.","summary":"An AI assistant you install on your own computer or on a small server. You talk to it in the terminal, or through Telegram, Discord, Slack, WhatsApp, Signal and email. It runs commands and tools for you, searches its own past conversations, writes reusable skills from what it did, and can run jobs on a schedule while you are away. You pick the model provider (Nous Portal, OpenRouter, OpenAI, your own endpoint and many more) and pay that provider. Windows, macOS and Linux are supported, plus Android through Termux, and there is a Docker image.","easeRationale":"All three systems use one command in a terminal and then a wizard: `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash` on Linux and macOS, `iex (irm https://hermes-agent.nousresearch.com/install.ps1)` in PowerShell on Windows, after which `hermes setup` asks for your provider and key. The installer pulls in Python 3.14, Node.js, ripgrep and FFmpeg itself, so you do not install those by hand.","costNote":"The program is free (MIT), but it only works with a model provider you pay for. The project sells its own subscription, Nous Portal, which bundles the model, web search, image generation, speech and a cloud browser; other listed options include a $10/month OpenCode Go plan and pay-per-use providers.","notVerified":["Whether the installer works end to end on each system, and what exactly the downloaded script does on your machine","What a month of real use costs at your chosen provider, including the scheduled jobs that run while you are away","Real disk use after Python, Node.js, a browser build and the agent's own database and skills folder","Whether the memory and skill files it writes about you stay only on your machine","How well the command approval and container isolation described in the documentation hold when the agent acts on its own","Why the star and issue counts are this high this fast, which the checks already flagged"]},"zh-Hans":{"summaryShort":"一个你自己运行的 AI 助手，可以在终端或聊天应用里工作，替你执行命令并记住之前的会话。","summary":"一个你安装在自己电脑或小型服务器上的 AI 助手。你可以在终端里和它对话，或者通过 Telegram、Discord、Slack、WhatsApp、Signal 和邮件与它交流。它会替你执行命令和工具，搜索自己过去的对话记录，从做过的事情里写出可复用的技能，还能在你不在时按计划运行任务。你可以选择模型提供方（Nous Portal、OpenRouter、OpenAI、你自己的端点等等），费用付给该提供方。支持 Windows、macOS 和 Linux，还可以通过 Termux 在 Android 上使用，另外也有 Docker 镜像。","easeRationale":"三个系统都是先在终端输入一条命令，再走一个配置向导：Linux 和 macOS 上是 `curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash`，Windows 的 PowerShell 上是 `iex (irm https://hermes-agent.nousresearch.com/install.ps1)`，之后 `hermes setup` 会询问你的提供方和密钥。安装程序会自己拉取 Python 3.14、Node.js、ripgrep 和 FFmpeg，所以这些你不用手动安装。","costNote":"程序本身免费（MIT 许可），但只能配合你自己付费的模型提供方使用。项目自己出售订阅服务 Nous Portal，打包了模型、网页搜索、图像生成、语音和云端浏览器；列出的其他选项还包括每月 10 美元的 OpenCode Go 计划和按使用量付费的提供方。","notVerified":["安装程序在每个系统上是否真的能从头跑到尾，以及下载的脚本在你的机器上到底做了什么","在你选择的提供方那里，一个月的真实使用会花多少钱，包括你不在时运行的定时任务","Python、Node.js、一个浏览器构建以及代理自身的数据库和技能文件夹加起来实际占用多少磁盘","它为你写的记忆和技能文件是否只留在你自己的机器上","当代理自行行动时，文档里描述的命令审批和容器隔离到底能撑住多少","星标数和 issue 数为什么涨得这么快，而检查环节已经指出了这一点"]}}},{"id":"R_kgDORyFzVQ","slug":"nowork-studio__notfair-plugin","name":"notfair","owner":"nowork-studio","fullName":"nowork-studio/notfair-plugin","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"your_account_data_passes_through_the_company","severity":"attention","evidence":"README.md 的 \"Privacy and support\" 部分写道：\"The plugin connects to the declared NotFair MCP endpoint through browser OAuth. Requested tool arguments go to NotFair and results from connected providers return to your authorized AI client.\" 也就是说，你的 Search Console、Analytics、Google Ads、Meta Ads、WordPress 和 GoHighLevel 数据会经过 notfair.co，而不是直接从 Google 或 Meta 到你的电脑。","source":"ai","title":"你的账户数据会经过这家公司"},{"code":"ai_can_change_live_accounts_and_spending","severity":"attention","evidence":"这些 skill 被描述为会执行操作，而不只是读取：`google-ads` \"manages supported keywords, bids, budgets, negatives, and campaigns\"，`paid-ads-x` \"audits and operates connected X Ads campaigns ... and budgets\"，WordPress 的 skill 则处理 \"settings, plugins, themes, and approved HTML or admin changes\"。一个失误就可能花掉真金白银的广告费，或者改动一个正在运行的网站。","source":"ai","title":"ai 可以改动真实账户和广告支出"},{"code":"guardrails_are_written_instructions","severity":"info","evidence":"README.md 宣称 \"Safe by design: read-only review comes before mutation, paid-media changes stay explicit\"，但也说 \"the canonical skills are plain files, not logic trapped inside one agent runtime\"。这些限制只是要求 AI 模型遵守的文字，文件本身并不能阻止任何改动的发生。","source":"ai","title":"防护规则只是书面指令"},{"code":"tells_your_agent_to_follow_remote_instructions","severity":"info","evidence":"README.md 提供了这样一段可以粘贴到助手里的内容：\"Retrieve and follow the instructions at: https://raw.githubusercontent.com/nowork-studio/notfair-plugin/main/INSTALL_FOR_AGENTS.md\"。于是你的智能体会听从一个作者可以随时修改的文件。","source":"ai","title":"让你的智能体去遵循远程指令"},{"code":"some_install_paths_not_available_yet","severity":"info","evidence":"README.md：在 Grok Bot 中，要 \"once its public listing is approved\" 才能找到 NotFair；在 Cursor 中要 \"once available\"；在 Kiro 中要 \"after its listing is approved\"；并且 \"Marketplace submission does not imply approval\"。目前只有 Claude Code、Codex 和 Gemini CLI 这几条路径被描述为可用。","source":"ai","title":"部分安装方式目前还不可用"}],"platforms":["windows","mac","linux"],"tools":["node","python","other"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["NotFair hosted MCP (browser sign-in with your Google, ad, WordPress and CRM accounts)","Google Gemini (only for the cross-model review skill)","WordPress / Strapi / Contentful / Ghost application password for the local SEO scripts"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","social_marketing"],"alternativeTo":[{"slug":"semrush","name":"Semrush","confirmed":true},{"slug":"ahrefs","name":"Ahrefs","confirmed":true},{"slug":"optmyzr","name":"Optmyzr","confirmed":true},{"slug":"surfer-seo","name":"SEO","confirmed":true},{"slug":"jasper","name":"Jasper","confirmed":false}],"declaredAlternatives":[],"stars":4137,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":25,"growthDays":6,"history":[{"date":"2026-11-07","stars":4120},{"date":"2026-11-08","stars":4124},{"date":"2026-11-09","stars":4133},{"date":"2026-11-10","stars":4137}],"firstPublishedAt":"2026-11-08T23:01:18.867Z","commitSha":"f70e36082a50db00c18a7752ba97ffa1052d3472","verifiedAt":"2026-11-08T22:44:40.871Z","lastCommitAt":"2026-11-10T00:12:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A pack of 48 marketing workflows you install into an AI assistant to audit SEO and review ad accounts.","summary":"NotFair Plugin is a set of written workflows (48 \"skills\") that you install into an AI assistant such as Claude Code, Codex, Cursor or Gemini CLI. The assistant then follows step-by-step marketing procedures: audit a site's SEO, rewrite pages to be quoted by AI search, read Google Analytics and Search Console, look for wasted ad spend on Google, Meta, X, LinkedIn, Reddit and TikTok, or operate a connected WordPress site and GoHighLevel CRM. There is nothing to open: live data arrives through one hosted \"NotFair\" connection you sign in to with your Google and ad accounts.","easeRationale":"There are no releases and nothing to download and open: on Windows, macOS and Linux alike you need an AI assistant already installed, then two commands (in Claude Code, `/plugin marketplace add nowork-studio/notfair-plugin` and `/plugin install notfair@nowork-studio`), plus a browser sign-in to the NotFair connection before any live data works. The optional local scripts also need Python.","costNote":"The workflows are free (MIT). They only do something inside an AI assistant, and the ones listed (Claude Code, Codex, Cursor, Gemini CLI) are paid or have paid plans. Live account data goes through the company's hosted NotFair connection; the repository does not state its price.","notVerified":["What NotFair stores and for how long once your ad and analytics data passes through it, and what the service costs","Whether the AI really stops at read-only review before changing a campaign, a budget or a WordPress site","How much the long workflows add to your AI assistant's usage bill","Whether the SEO and ad advice produced actually improves traffic or reduces wasted spend"]},"zh-Hans":{"summaryShort":"一套包含 48 个营销工作流的插件，安装到 AI 助手里，用来审查 SEO 和广告账户。","summary":"NotFair Plugin 是一套写好的工作流（48 个 \"skills\"），你把它们安装到像 Claude Code、Codex、Cursor 或 Gemini CLI 这样的 AI 助手里。然后助手会按步骤执行营销流程：审查网站的 SEO、改写页面以便被 AI 搜索引用、读取 Google Analytics 和 Search Console、在 Google、Meta、X、LinkedIn、Reddit 和 TikTok 上查找被浪费的广告支出，或者操作已连接的 WordPress 网站和 GoHighLevel CRM。没有任何东西需要打开：实时数据通过一个托管的 \"NotFair\" 连接传来，你用自己的 Google 账户和广告账户登录它。","easeRationale":"没有发行版本，也没有东西需要下载和打开：在 Windows、macOS 和 Linux 上都一样，你需要先装好一个 AI 助手，然后输入两条命令（在 Claude Code 里，`/plugin marketplace add nowork-studio/notfair-plugin` 和 `/plugin install notfair@nowork-studio`），再通过浏览器登录 NotFair 连接，之后才能让实时数据生效。可选的本地脚本还需要 Python。","costNote":"这些工作流本身是免费的（MIT 许可证）。它们只能在 AI 助手里起作用，而列出的那些助手（Claude Code、Codex、Cursor、Gemini CLI）是收费的或有付费方案。账户的实时数据会经过公司托管的 NotFair 连接；仓库没有说明它的价格。","notVerified":["你的广告和分析数据经过 NotFair 之后，它会保存什么内容、保存多久，以及这个服务收费多少","在改动广告系列、预算或 WordPress 网站之前，AI 是否真的只停留在只读审查阶段","这些冗长的工作流会给你的 AI 助手用量账单增加多少费用","产出的 SEO 和广告建议是否真的能提升流量或减少浪费的支出"]}}},{"id":"R_kgDOSOQ9xw","slug":"nutlope__hallmark","name":"Hallmark","owner":"Nutlope","fullName":"Nutlope/hallmark","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"made_by_a_paid_model_provider","severity":"info","evidence":"README：'Made by Together AI'，package.json 将其描述为 'A design skill for AI coding assistants. (...) Powered by Together AI.' Together AI 出售模型的使用权限，所以该项目在推广其制作方的付费服务，尽管这个 skill 本身不需要任何密钥。","source":"ai","title":"由付费模型服务商制作"},{"code":"redesign_rewrites_files_in_your_project","severity":"attention","evidence":"README：'/hallmark redesign ./src/app rebuilds a page and keeps its content and wiring'。完整的重新设计会 'changes the system and the layout' 你指定文件夹中的文件，也就是说助手会编辑你的源代码；项目自带的把关手段是 `sloplint --before` 加上 `--kept`，如果有内容丢失，它会 'fails the redesign if one went missing'。","source":"ai","title":"重新设计会改写你项目中的文件"},{"code":"installs_an_edit_time_hook_into_claude_code","severity":"info","evidence":"README 的安装选项：'`install-hook.mjs` adds an edit-time Floor check to Claude Code'，也就是说其中一种安装方式会给助手接入一段在你编辑文件时运行的代码。","source":"ai","title":"在 Claude Code 中安装编辑时触发的 hook"},{"code":"pulls_photographs_from_the_internet_into_your_pages","severity":"info","evidence":"README：'755 openly licensed photos, picked per subject and credited on the page'，许可说明指出 `scripts/photos.json` 'lists photographs hosted on Wikimedia Commons and holds no image files. Each keeps its own licence (CC0, public domain, CC BY or CC BY-SA)'。其中一些许可证要求你在发布的页面上保留署名。","source":"ai","title":"把互联网上的照片引入你的页面"},{"code":"no_releases_install_pulls_the_latest_main","severity":"info","evidence":"项目信息：GitHub 上有 0 个发布版本，安装命令是 `npx skills add nutlope/hallmark -g`，它会获取当时仓库里的任何内容，而不是某个固定版本。最近一次提交与收集数据包是在同一天。","source":"ai","title":"没有发布版本，安装会拉取 main 的最新内容"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","design_cad"],"alternativeTo":[{"slug":"tailwind-plus","name":"Tailwind Plus","confirmed":true},{"slug":"v0-by-vercel","name":"v0 by Vercel","confirmed":true}],"declaredAlternatives":[],"stars":52513,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":5555,"growthDays":7,"history":[{"date":"2026-10-27","stars":39014},{"date":"2026-10-28","stars":40727},{"date":"2026-10-29","stars":42103},{"date":"2026-10-30","stars":42981},{"date":"2026-10-31","stars":43931},{"date":"2026-11-01","stars":44950},{"date":"2026-11-02","stars":46010},{"date":"2026-11-03","stars":46958},{"date":"2026-11-04","stars":47817},{"date":"2026-11-05","stars":48663},{"date":"2026-11-06","stars":49525},{"date":"2026-11-07","stars":50177},{"date":"2026-11-08","stars":50861},{"date":"2026-11-09","stars":51699},{"date":"2026-11-10","stars":52513}],"firstPublishedAt":"2026-10-27T18:40:27.982Z","commitSha":"8f6969f8922caabde9de1784959eae11ddf68073","verifiedAt":"2026-10-29T01:40:17.919Z","lastCommitAt":"2026-11-10T18:01:18Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes the web pages your AI coding assistant builds look designed instead of generic, and can redesign pages you already have.","summary":"Hallmark is a design skill you add to an AI coding assistant such as Claude Code, Codex, Cursor or OpenCode. When the assistant builds a web page, Hallmark picks one of fifty visual styles, lays the page out around what the page is for, then checks the result against a list of tells that make pages look machine-made. It can also redesign or audit a page you already have. You need Node 18 or newer and one of those assistants; the browser-based checks and screenshots also need Chrome and puppeteer-core. The output is self-contained HTML and CSS.","easeRationale":"Same on Windows, macOS and Linux: have Node 18 or newer and a supported assistant, then one command (`npx skills add nutlope/hallmark -g`) and you call it with `/hallmark`. For the browser checks and screenshots you also install Chrome and `npm i -g puppeteer-core`; without them only the static checks run.","costNote":"MIT licensed and free. It only works inside an AI coding assistant, which normally needs a paid plan or paid API use. The README says it is made by Together AI, a paid model provider.","notVerified":["Whether the generated pages really look better than what the assistant produces on its own","How much extra AI usage, and therefore cost, one build with its reviews consumes","Whether the redesign keeps every link, form field and handler on a real project","What the edit-time hook does on each run inside Claude Code","Real disk use after the skill, the theme references and Chrome are installed","Whether the photo credits it writes satisfy every licence in the list"]},"zh-Hans":{"summaryShort":"让AI编程助手搭建的网页看起来经过设计,而不是千篇一律,还能重新设计你已有的页面。","summary":"Hallmark 是一项添加到AI编程助手(如Claude Code、Codex、Cursor或OpenCode)中的设计技能。当助手搭建网页时,Hallmark会从五十种视觉风格中选一种,按页面用途安排布局,再对照一份让网页显得像机器制作的特征清单检查结果。它还能重新设计或审查你已有的页面。你需要Node 18或更新版本,以及上述助手之一;基于浏览器的检查和截图还需要Chrome和puppeteer-core。输出结果是自包含的HTML和CSS。","easeRationale":"在Windows、macOS和Linux上一样:安装Node 18或更新版本以及一个受支持的助手,然后执行一条命令(`npx skills add nutlope/hallmark -g`),再用`/hallmark`调用它。要进行浏览器检查和截图,还需要安装Chrome和`npm i -g puppeteer-core`;没有它们的话,只会运行静态检查。","costNote":"采用MIT许可,免费。它只能在AI编程助手内运行,而助手通常需要付费套餐或付费API使用。README说它由Together AI制作,这是一家收费的模型提供商。","notVerified":["生成的页面是否真的比助手自行生成的效果更好","带有审查的一次构建会多消耗多少AI用量,从而带来多少额外费用","重新设计是否在真实项目中保留了每个链接、表单字段和处理程序","编辑时触发的钩子在Claude Code里每次运行时做了什么","安装该技能、主题参考资料和Chrome后,实际占用的磁盘空间","它写的图片版权说明是否满足清单中的每一项许可要求"]}}},{"id":"R_kgDORGJUeg","slug":"nuviomedia__nuviotv","name":"NuvioTV","owner":"NuvioMedia","fullName":"NuvioMedia/NuvioTV","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"no_content_of_its_own_sources_are_up_to_you","severity":"attention","evidence":"README.md：\"Bring your own sources. Nuvio turns them into a library with artwork, ratings, subtitles, and your place saved on every screen.\" 仓库从没说明什么是 \"source\"，也没说去哪里获取，所以你看到的内容是否合法，完全取决于你自己连接的东西，而在一些国家，人们给这类应用添加的片源并不合法。","source":"ai","title":"本身没有内容，片源全靠你自己"},{"code":"newest_releases_are_betas","severity":"info","evidence":"证据包里最新的三个发行版本都带有 beta 标签，相隔一两天发布：1.1.0-beta.13（2026-11-08）、1.1.0-beta.12（2026-11-06）和 1.1.0-beta.11（2026-11-05）。该仓库有 855 个未解决的 issue。","source":"ai","title":"最新版本都是测试版"},{"code":"apk_outside_the_store","severity":"info","evidence":"README 的 \"Get Nuvio TV\" 部分在 Google Play 链接旁边，还提供了一个指向发行页的 \"Android TV APK\" 链接。通过 APK 文件手动安装的应用不会经过 Google Play 的审查，也不会自动更新。","source":"ai","title":"商店之外安装的 apk"},{"code":"bundles_its_own_player_libraries","severity":"info","evidence":"仓库中包含 `libmpv-android` 和一个 `ffmpeg-decoder-downmix/` 模块，还有一个 `DV7/` 文件夹，所以这个应用自带了视频解码代码，而不是只依赖电视自带的播放器。","source":"ai","title":"自带播放器相关库"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["video","hardware"],"alternativeTo":[{"slug":"plex","name":"Plex","confirmed":true},{"slug":"emby","name":"Emby","confirmed":true},{"slug":"infuse","name":"Infuse","confirmed":true}],"declaredAlternatives":[],"stars":3464,"downloads":3051920,"downloadsWeek":162635,"downloadsDays":6,"growth":102,"growthDays":6,"history":[{"date":"2026-11-04","stars":3362},{"date":"2026-11-05","stars":3373},{"date":"2026-11-06","stars":3386},{"date":"2026-11-07","stars":3409},{"date":"2026-11-08","stars":3432},{"date":"2026-11-09","stars":3446},{"date":"2026-11-10","stars":3464}],"firstPublishedAt":"2026-11-08T23:13:56.599Z","commitSha":"12d96bced9821bdf4bc233039ffb1ccd1795e55b","verifiedAt":"2026-11-08T23:07:58.090Z","lastCommitAt":"2026-11-10T20:15:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android TV app that turns video sources you add yourself into a library with artwork, ratings and subtitles.","summary":"Nuvio TV is a media app for the television. It brings no films or series of its own: as the README puts it, you \"bring your own sources\", and the app turns them into a library with artwork, ratings, subtitles and your place saved on every screen. The owner calls this the \"Official Nuvio Android TV Repository\", and the only downloads offered here are for Android TV: the Google Play listing, or the \"Android TV APK\" from the releases page. The README also names phone and desktop, which this repository does not offer.","easeRationale":"What this repository offers is an Android TV app, so none of the computer systems get a grade. On an Android TV the README gives two ways in, the Google Play listing or the \"Android TV APK\" link to the releases page; the step that makes it a B is that nothing plays until you add your own sources.","costNote":"GPL-3.0 and free on Google Play. The README links to a \"Support Nuvio\" page for donations.","notVerified":["What kind of \"sources\" the app expects and whether using them is legal where you live","Whether there is an installable file on the releases page: the three releases in the evidence package are listed with no platform binaries","Whether the app on Google Play is built from this source code","Which Android TV versions and devices it actually works on","Whether the phone and desktop versions the README mentions exist anywhere, since this repository only offers Android TV","What the app sends to the internet while it looks up artwork, ratings and subtitles","How much space the library and cached artwork take on the TV"]},"zh-Hans":{"summaryShort":"一个 Android TV 应用，能把你自己添加的视频来源整理成带有封面、评分和字幕的库。","summary":"Nuvio TV 是一款电视媒体应用。它本身不提供任何电影或剧集：正如 README 所说，你需要 \"bring your own sources\"（自己带来片源），应用会把它们整理成一个带封面、评分、字幕的库，并在每个屏幕上保存你的观看进度。作者把它称为 \"Official Nuvio Android TV Repository\"，这里提供的下载只有 Android TV 版：Google Play 页面，或者发行页上的 \"Android TV APK\"。README 还提到了手机版和桌面版，但这个仓库并不提供它们。","easeRationale":"这个仓库提供的是一个 Android TV 应用，所以电脑端的各个系统都没有评级。在 Android TV 上，README 给出了两种方式：Google Play 页面，或者指向发行页的 \"Android TV APK\" 链接；让它评为 B 的原因是，在你添加自己的片源之前什么都播放不了。","costNote":"采用 GPL-3.0 许可证，在 Google Play 上免费。README 链接到一个 \"Support Nuvio\" 页面接受捐赠。","notVerified":["应用所期待的 \"sources\"（片源）到底是什么，在你所在的地方使用它们是否合法","发行页上是否有可安装的文件：证据包里列出的三个发行版本都没有任何平台的二进制文件","Google Play 上的应用是否是基于这份源代码构建的","它实际能在哪些 Android TV 版本和设备上运行","README 提到的手机版和桌面版是否存在于别的地方，因为这个仓库只提供 Android TV 版","应用在查找封面、评分和字幕时会向互联网发送什么内容","库和缓存的封面会在电视上占用多少空间"]}}},{"id":"MDEwOlJlcG9zaXRvcnk3MjMxMTY4MA==","slug":"nuxt__vue-meta","name":"vue-meta","owner":"nuxt","fullName":"nuxt/vue-meta","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发行版本是在 2021-06-06（1981 天前）","source":"check","title":"版本长期未更新"},{"code":"archived","severity":"info","evidence":"该仓库已被归档（只读，不再维护）","source":"check","title":"已归档"},{"code":"readme_says_project_is_abandoned","severity":"attention","evidence":"README.md 的第一行就是 \"# This project is not actively maintained anymore.\"，并指向 issue #808 查看详情。该仓库已被归档，所以没有人再能提交 issue 或 pull request。","source":"ai","title":"readme 表明项目已被放弃"},{"code":"vue3_support_never_left_alpha","severity":"attention","evidence":"README.md 写道 \"Check the next branch for Vue3 support\"，而最新的三个发行版本是 v3.0.0-alpha.8（2021-06-06）、v3.0.0-alpha.3 和 v3.0.0-alpha.2——Vue 3 版本从未完成。这里的稳定包是面向 Vue 2 的 2.4.0 版本（devDependency \"vue\": \"^2.6.11\"）。","source":"ai","title":"vue3 支持从未脱离 alpha 阶段"},{"code":"readme_loads_floating_version_from_cdn","severity":"info","evidence":"README.md 提供了这样的代码：`<script src=\"https://unpkg.com/vue-meta/dist/vue-meta.min.js\"></script>`。这个网址没有版本号，所以使用它的页面会运行 unpkg 当时提供的任何文件，这个文件来自第三方服务器。","source":"ai","title":"readme 从 cdn 加载的是浮动版本"},{"code":"very_old_toolchain_and_dependencies","severity":"info","evidence":"最后一次提交是在 2023-07-26，package.json 里仍然锁定着 webpack ^4.43.0、jest ^26、typescript ^3.9.5、eslint ^7 和 chromedriver ^86。项目自那之后再没有更新过，所以这些包里任何新发现的问题都一直没有修复。","source":"ai","title":"工具链和依赖非常陈旧"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":4068,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":4068},{"date":"2026-11-08","stars":4068},{"date":"2026-11-09","stars":4068},{"date":"2026-11-10","stars":4068}],"firstPublishedAt":"2026-11-08T23:01:18.881Z","commitSha":"96d8ce4c38011343d41c005098f42e0f5018818a","verifiedAt":"2026-11-08T22:47:22.395Z","lastCommitAt":"2023-07-26T16:22:29Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Vue 2 plugin for developers that sets page titles and meta tags from inside components. No longer maintained.","summary":"Vue Meta is a code library for people building websites with Vue.js. Each component can declare a `metaInfo` block with the page title, meta tags and html attributes, and the library writes them into the page, including on the server so search engines see them. It is installed with npm or yarn into a project. There is nothing for a non-programmer to open. The repository is archived and the README says in its first line that the project is not actively maintained anymore; Nuxt.js, Gridsome and other frameworks used to build on it.","easeRationale":"This is a library, not a program: on Windows, macOS and Linux alike the only install instructions are `yarn add vue-meta` or `npm install vue-meta` inside an existing Vue project, plus a few lines of code (`Vue.use(VueMeta, ...)`), so it is E everywhere.","costNote":"MIT licensed.","notVerified":["Whether it still works with current Vue 2 builds and current Node versions","Whether the unfinished Vue 3 branch works in a real project","Whether the old build and test dependencies contain known security problems","Which replacement a project should move to, since the repository does not name one in the README"]},"zh-Hans":{"summaryShort":"一个面向开发者的 Vue 2 插件，可以在组件内部设置页面标题和 meta 标签。已不再维护。","summary":"Vue Meta 是一个面向使用 Vue.js 构建网站的人的代码库。每个组件都可以声明一个 `metaInfo` 代码块，里面有页面标题、meta 标签和 html 属性，库会把它们写入页面，包括在服务器端写入，这样搜索引擎也能看到。它通过 npm 或 yarn 安装到项目里。不会编程的人没有什么可以打开的东西。这个仓库已被归档，README 第一行就写着这个项目已经不再积极维护；Nuxt.js、Gridsome 等框架曾经基于它构建。","easeRationale":"这是一个库，不是一个程序：在 Windows、macOS 和 Linux 上都一样，唯一的安装说明是在现有的 Vue 项目里运行 `yarn add vue-meta` 或 `npm install vue-meta`，再加几行代码（`Vue.use(VueMeta, ...)`），所以各平台都是 E。","costNote":"采用 MIT 许可证。","notVerified":["它是否仍能配合当前的 Vue 2 版本和当前的 Node 版本工作","未完成的 Vue 3 分支是否能在真实项目里使用","旧的构建和测试依赖是否存在已知漏洞","一个项目应该迁移到哪个替代方案，因为仓库在 README 里并没有指出一个"]}}},{"id":"R_kgDORsnMJw","slug":"nvidia__skillspector","name":"SkillSpector","owner":"NVIDIA","fullName":"NVIDIA/SkillSpector","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"alpha_software_that_issues_install_verdicts","severity":"attention","evidence":"pyproject.toml 声明 \"Development Status :: 3 - Alpha\"，但输出的却是人们会据此行动的裁定：51-100 分会打印 \"DO NOT INSTALL\"，0-20 分会给出绿色推荐，MCP 工具还会返回是否安装的字段。因此，扫描工具漏掉的问题就会被当成已经通过。","source":"ai","title":"Alpha 阶段软件却给出安装裁定"},{"code":"scanned_files_are_sent_to_an_ai_provider","severity":"info","evidence":"更深入的扫描需要用到 LLM：\"For the best results, configure an OpenAI-compatible LLM endpoint for semantic analysis\"，默认提供方是 `nv_build`（build.nvidia.com）。除非你使用 `--no-llm` 或本地 Ollama 模型，否则你扫描的任何技能——包括私有代码或客户代码——都会发送给该提供方。","source":"ai","title":"被扫描的文件会发送给某个 AI 提供方"},{"code":"research_numbers_cannot_be_checked_here","severity":"info","evidence":"README 开头写着 \"In the 31,132-skill analyzed subset of the research dataset, 26.1% of skills contain vulnerabilities and 5.2% show likely malicious intent.\" 该数据集并不在仓库中，所以这些数字只是 NVIDIA 自己的说法，无法通过这份代码来验证。","source":"ai","title":"研究数据无法在此核实"},{"code":"mcp_http_mode_has_no_login","severity":"info","evidence":"README 自己的风险提示写道：\"The HTTP transport ships without authentication. Any caller that can reach the port can invoke scan_skill.\" 它被限制在回环地址（127.0.0.1 或 ::1），本地路径在 HTTP 下会被拒绝；它还提到，对于 stdio 传输方式，\"the initialize hang reported in issue #199 still applies\"。","source":"ai","title":"MCP 的 HTTP 模式没有登录验证"},{"code":"it_downloads_what_it_inspects","severity":"info","evidence":"`skillspector scan https://github.com/user/my-skill` 会克隆仓库并解压你指定的 zip 文件，所以不可信的文件会先落到你的磁盘上，之后才进行分析。它设有限制：`INGEST_MAX_BYTES` 为 100 MiB，`INGEST_MAX_ZIP_MEMBERS` 为 10,000；安装提示还警告说 \"This project will download and install additional third-party open source software projects.\"","source":"ai","title":"它会先下载要检查的内容"},{"code":"one_provider_is_disabled_for_safety","severity":"info","evidence":".env.example 和提供方列表说明 `codex_cli` 被\"registered but disabled: its read-only sandbox permits host-file reads\"，因此这条路径是故意被屏蔽的。","source":"ai","title":"出于谨慎停用了一个提供方"}],"platforms":["windows","mac","linux"],"tools":["python","docker"],"buildTools":["python","node","docker","gpu_nvidia"],"disk":{"minMB":200,"maxMB":800,"basis":"heuristic"},"apiKeys":["NVIDIA Build (NVIDIA_INFERENCE_KEY, the default provider)","OpenAI","Anthropic","Azure OpenAI","Any OpenAI-compatible gateway (SKILLSPECTOR_COMPAT_API_KEY)","LangSmith tracing (LANGCHAIN_API_KEY, optional)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[{"slug":"snyk","name":"Snyk","confirmed":true},{"slug":"semgrep","name":"Semgrep","confirmed":true}],"declaredAlternatives":[],"stars":25199,"downloads":51908,"downloadsWeek":6659,"downloadsDays":4,"growth":850,"growthDays":6,"history":[{"date":"2026-11-06","stars":24660},{"date":"2026-11-07","stars":24790},{"date":"2026-11-08","stars":24905},{"date":"2026-11-09","stars":25063},{"date":"2026-11-10","stars":25199}],"firstPublishedAt":"2026-11-08T23:01:18.888Z","commitSha":"e48d771a26f7398b36256ceb8d8f94adce3d5c44","verifiedAt":"2026-11-08T22:49:02.802Z","lastCommitAt":"2026-11-10T17:21:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An NVIDIA command-line scanner that checks an AI agent skill or MCP server for risky patterns before you install it.","summary":"SkillSpector is a terminal tool from NVIDIA that inspects the skills and MCP servers people install into AI assistants such as Claude Code, Codex CLI or Gemini CLI. You point it at a folder, a SKILL.md file, a zip or a Git address and it looks for 71 patterns in 17 groups (prompt injection, data exfiltration, credential access, obfuscated code, vulnerable dependencies), then prints a 0-100 risk score and a recommendation. A fast pass runs offline with `--no-llm`; the deeper pass sends the files to an AI model you choose. It can also run as an MCP server to check a skill before installing it.","easeRationale":"There is no installer: on macOS and Linux you install Python 3.12+ and uv (`uv tool install git+https://github.com/NVIDIA/skillspector.git`) or build the included Dockerfile, then run commands in a terminal and set environment variables for an AI provider, so D. Windows is D as well but rougher: every example in the README is written for a Unix shell (`source .venv/bin/activate`, `$PWD`), so Docker or WSL is the practical route. A graphics card is only needed if you choose to run the AI model locally with Ollama.","costNote":"Apache-2.0 and free to run. The static pass (`--no-llm`) needs no account; the second, semantic pass needs an AI provider — NVIDIA Build, OpenAI, Anthropic, Azure OpenAI, Bedrock or Gemini all bill per use, and a local Ollama model is the free option.","notVerified":["How often it misses a genuinely malicious skill, and how many harmless skills it flags","What one scan costs at the paid AI providers, since a full workflow can run up to 18 minutes and many analysis batches","Whether the documented install works on Windows without Docker or WSL","What the chosen AI provider stores of the files you send for analysis","Whether a low score means much for skills whose risky behaviour only appears while they run"]},"zh-Hans":{"summaryShort":"NVIDIA 的命令行扫描工具，在安装前检查 AI 代理技能或 MCP 服务器中的风险模式。","summary":"SkillSpector 是 NVIDIA 推出的终端工具，用于检查人们安装到 Claude Code、Codex CLI 或 Gemini CLI 等 AI 助手中的技能和 MCP 服务器。你指定一个文件夹、一个 SKILL.md 文件、一个 zip 包或一个 Git 地址，它会在 17 组共 71 种模式中查找问题（提示注入、数据外泄、凭证访问、混淆代码、存在漏洞的依赖），然后给出 0 到 100 的风险评分和建议。使用 `--no-llm` 可离线快速扫描；更深入的扫描会把文件发送给你选择的 AI 模型。它也可以作为 MCP 服务器，在安装技能前先检查。","easeRationale":"没有安装程序：在 macOS 和 Linux 上，你需要安装 Python 3.12+ 和 uv（`uv tool install git+https://github.com/NVIDIA/skillspector.git`），或者自己构建附带的 Dockerfile，然后在终端中运行命令并设置 AI 提供方的环境变量，因此评级为 D。Windows 同样是 D，但更麻烦：README 里的所有示例都是为 Unix shell 写的（`source .venv/bin/activate`、`$PWD`），所以 Docker 或 WSL 才是实用的方式。只有在你选择用 Ollama 本地运行 AI 模型时才需要显卡。","costNote":"采用 Apache-2.0 协议，运行免费。静态扫描（`--no-llm`）不需要账号；第二步的语义扫描需要一个 AI 提供方——NVIDIA Build、OpenAI、Anthropic、Azure OpenAI、Bedrock 或 Gemini 都按使用量计费，本地运行的 Ollama 模型是免费的选项。","notVerified":["它多常漏掉真正恶意的技能，又有多少无害的技能被误判","在付费 AI 提供方上扫描一次要花多少钱，因为一次完整流程可能要跑到 18 分钟，分很多批次分析","文档中的安装方法在没有 Docker 或 WSL 的 Windows 上是否真的可行","所选的 AI 提供方会如何保存你发送去分析的文件","对于那些风险行为只在运行时才出现的技能，低分数能说明多少问题"]}}},{"id":"R_kgDOP_gcWw","slug":"obra__superpowers","name":"Superpowers","owner":"obra","fullName":"obra/superpowers","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"383 天内获得 306455 个星标（约每天 800 个）","source":"check","title":"星标增长速度异常"},{"code":"agent_works_long_stretches_and_deletes_code","severity":"attention","evidence":"README：'It's not uncommon for your agent to work autonomously for a couple hours at a time'，test-driven-development skill 'Deletes code written before tests'，finishing-a-development-branch 'presents options (merge/PR/keep/discard), cleans up worktree'。这些 skills 被描述为 'Mandatory workflows, not suggestions'，因此智能体会被引导自行删除并重写已有的工作成果。","source":"ai","title":"智能体会长时间工作并删除代码"},{"code":"loads_an_image_from_the_authors_site_by_default","severity":"info","evidence":"README 的 'Visual companion telemetry' 部分：默认情况下，brainstorming 的 visual companion 中的 Prime Radiant 徽标是从他们的网站加载的，并带有你正在使用的 Superpowers 版本号；他们说其中不包含任何关于你的项目、提示词或智能体的信息，并且设置 SUPERPOWERS_DISABLE_TELEMETRY（或 Claude Code 的 DISABLE_TELEMETRY）即可关闭。","source":"ai","title":"默认会从作者网站加载一张图片"},{"code":"injects_instructions_at_every_session_start","severity":"info","evidence":"该仓库附带 hooks/ 以及面向十几种智能体的 plugin 文件夹（.claude-plugin/、.codex-plugin/、.cursor-plugin/、.devin-plugin/、.muse-plugin/ 等）；README 说 `using-superpowers` 引导内容会通过原生的 SessionStart hook 注入，并在压缩之后再次注入，因此它的指令会在你没有要求的情况下加入每一次对话。","source":"ai","title":"每次会话开始时都会注入指令"},{"code":"one_install_path_fetches_instructions_from_a_url","severity":"info","evidence":"对于 OpenCode，README 让你说：'Fetch and follow instructions from https://raw.githubusercontent.com/obra/superpowers/refs/heads/main/.opencode/INSTALL.md'，这会让智能体下载一个文件，并照做其中当时所写的任何内容。","source":"ai","title":"有一种安装方式会从网址获取指令"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":315690,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":3835,"growthDays":7,"history":[{"date":"2026-10-25","stars":305095},{"date":"2026-10-26","stars":305618},{"date":"2026-10-27","stars":306447},{"date":"2026-10-28","stars":307448},{"date":"2026-10-29","stars":308553},{"date":"2026-10-30","stars":309439},{"date":"2026-10-31","stars":310067},{"date":"2026-11-01","stars":310665},{"date":"2026-11-02","stars":311251},{"date":"2026-11-03","stars":311855},{"date":"2026-11-04","stars":312376},{"date":"2026-11-05","stars":312960},{"date":"2026-11-06","stars":313555},{"date":"2026-11-07","stars":314055},{"date":"2026-11-08","stars":314532},{"date":"2026-11-09","stars":315088},{"date":"2026-11-10","stars":315690}],"firstPublishedAt":"2026-10-25T23:19:18.673Z","commitSha":"2ce8aa15d05bd57213e6f746f953aafb7ded7a8b","verifiedAt":"2026-10-28T01:42:13.326Z","lastCommitAt":"2026-11-03T22:17:07Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A pack of skills that makes an AI coding agent brainstorm, plan, write tests first and review its own work.","summary":"Superpowers is a set of written skills plus a small plugin that changes how an AI coding agent works. Instead of jumping into code, the agent asks what you are really trying to build, writes a design you approve, turns it into small tasks, then works through them with tests first and a code review after each step. It installs as a plugin in Claude Code, Codex, Cursor, Gemini CLI, Copilot CLI, Grok, Kimi, OpenCode, Devin and others, usually with one command inside the agent. It is for programmers who already use one of those agents.","easeRationale":"You need one of the supported coding agents installed first, then one or two commands inside it, for example `/plugin install superpowers@claude-plugins-official` in Claude Code or `gemini extensions install https://github.com/obra/superpowers`. Each agent has its own install path and you repeat it per agent; on some you also restart the session or approve the hooks.","costNote":"The skills are free (MIT). They only work inside an AI coding agent, which usually needs a paid plan or paid API use. The README offers paid commercial support for companies through sales@primeradiant.com.","notVerified":["Whether the workflow actually improves results on a real project","How much extra AI usage, and therefore cost, the subagent-driven workflow consumes","What the bundled hooks do in each of the supported agents","How well agents other than Claude Code pick the skills up, especially on Windows","Whether the git worktree steps behave well in an existing repository with uncommitted work"]},"zh-Hans":{"summaryShort":"一套技能包,让AI编程智能体先头脑风暴、做计划、先写测试,再检查自己的工作。","summary":"Superpowers 是一套技能文档加一个小插件,改变AI编程智能体的工作方式。智能体不会直接动手写代码,而是先问你到底想做什么,写出一份你认可的设计,把它拆成小任务,然后按任务逐一处理,每一步都先写测试、再做代码检查。它以插件形式安装到 Claude Code、Codex、Cursor、Gemini CLI、Copilot CLI、Grok、Kimi、OpenCode、Devin 等工具中,通常在对应智能体里敲一条命令即可。它适合已经在用这些智能体之一的程序员。","easeRationale":"你需要先安装好一个受支持的编程智能体,然后在其中执行一两条命令,例如在 Claude Code 里执行 `/plugin install superpowers@claude-plugins-official`,或者执行 `gemini extensions install https://github.com/obra/superpowers`。每个智能体的安装方式不同,需要逐个重复操作;在有些工具里还要重启会话或批准钩子(hooks)。","costNote":"这些技能是免费的(MIT 许可)。它们只能在AI编程智能体内使用,而这类工具通常需要付费套餐或付费的API用量。README 中提到可通过 sales@primeradiant.com 为企业提供付费的商业支持。","notVerified":["这种工作流程是否真的能在实际项目中提升效果","这种由子智能体驱动的工作流会消耗多少额外的AI用量,进而增加多少成本","随附的钩子(hooks)在每个受支持的智能体里具体做了什么","除 Claude Code 之外的其他智能体对这些技能的支持效果如何,尤其是在 Windows 上","git worktree 相关步骤在一个已有未提交改动的仓库里是否能正常工作"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMzIzMzE1OA==","slug":"obsproject__obs-studio","name":"OBS Studio","owner":"obsproject","fullName":"obsproject/obs-studio","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"low","signals":[{"code":"linux_build_single_distro","severity":"info","evidence":"33.0.1 版本只有一个 Linux 文件，\"OBS-Studio-33.0.1-Ubuntu-26.04-x86_64.deb\"。其他 Linux 系统不在官方下载的覆盖范围内。","source":"ai","title":"Linux 版本只支持单一发行版"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":178,"maxMB":534,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"xsplit-broadcaster","name":"XSplit Broadcaster","confirmed":true},{"slug":"camtasia","name":"Camtasia","confirmed":true}],"declaredAlternatives":[],"stars":79267,"downloads":103580786,"downloadsWeek":203774,"downloadsDays":7,"growth":363,"growthDays":7,"history":[{"date":"2026-10-25","stars":78214},{"date":"2026-10-26","stars":78291},{"date":"2026-10-27","stars":78395},{"date":"2026-10-28","stars":78489},{"date":"2026-10-29","stars":78572},{"date":"2026-10-30","stars":78634},{"date":"2026-10-31","stars":78711},{"date":"2026-11-01","stars":78781},{"date":"2026-11-02","stars":78842},{"date":"2026-11-03","stars":78904},{"date":"2026-11-04","stars":78963},{"date":"2026-11-05","stars":79005},{"date":"2026-11-06","stars":79064},{"date":"2026-11-07","stars":79122},{"date":"2026-11-08","stars":79184},{"date":"2026-11-09","stars":79226},{"date":"2026-11-10","stars":79267}],"firstPublishedAt":"2026-10-25T23:19:19.537Z","commitSha":"65c320b46e1af5c7a555161aea6182cdf7a8c524","verifiedAt":"2026-10-30T01:42:23.339Z","lastCommitAt":"2026-11-10T21:02:58Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"OBS-Studio-33.0.1-Windows-x64-Installer.exe","sizeBytes":186662440,"url":"https://github.com/obsproject/obs-studio/releases/download/33.0.1/OBS-Studio-33.0.1-Windows-x64-Installer.exe","releaseUrl":"https://github.com/obsproject/obs-studio/releases/tag/33.0.1"},{"os":"mac","file":"OBS-Studio-33.0.1-macOS-Apple.dmg","sizeBytes":209628971,"url":"https://github.com/obsproject/obs-studio/releases/download/33.0.1/OBS-Studio-33.0.1-macOS-Apple.dmg","releaseUrl":"https://github.com/obsproject/obs-studio/releases/tag/33.0.1"},{"os":"linux","file":"OBS-Studio-33.0.1-Ubuntu-26.04-x86_64.deb","sizeBytes":159354224,"url":"https://github.com/obsproject/obs-studio/releases/download/33.0.1/OBS-Studio-33.0.1-Ubuntu-26.04-x86_64.deb","releaseUrl":"https://github.com/obsproject/obs-studio/releases/tag/33.0.1"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Record your screen, webcam and games, or stream them live to sites like YouTube and Twitch.","summary":"OBS Studio is a free program for recording and live streaming video. You build \"scenes\" from sources such as your screen, a game, a webcam, images and microphones, mix the audio, and then record to a file or broadcast live to a streaming service. It is widely used by streamers, teachers and anyone making screen recordings. On Windows you download the installer from the releases and run it; ready downloads for macOS (.dmg) and for Ubuntu Linux (.deb) are published too.","easeRationale":"On Windows you download and run a ready installer (OBS-Studio-33.0.1-Windows-x64-Installer.exe, 178 MB); macOS has ready .dmg files for Apple and Intel chips, and Linux has a ready package, but only one built for Ubuntu 26.04 (OBS-Studio-33.0.1-Ubuntu-26.04-x86_64.deb), so other Linux systems may need another source.","notVerified":["How well recording and streaming perform on your computer's graphics card and processor.","Whether game capture works with the specific games or apps you use.","Real disk use once you start saving recordings, which can grow quickly.","Whether the Ubuntu package installs on other Linux systems."]},"zh-Hans":{"summaryShort":"录制你的屏幕、摄像头和游戏画面,或者直播到YouTube、Twitch等平台。","summary":"OBS Studio是一款免费的录屏和直播软件。你用屏幕、游戏画面、摄像头、图片和麦克风等素材搭建\"场景\",混合音频,然后录制成文件,或者直播到流媒体平台。主播、教师以及任何需要录屏的人都在广泛使用它。在Windows上,你从发布包里下载安装程序并运行;macOS(.dmg)和Ubuntu Linux(.deb)也发布了现成的下载包。","easeRationale":"在Windows上,你下载并运行一个现成的安装程序(OBS-Studio-33.0.1-Windows-x64-Installer.exe,178MB);macOS有适用于Apple芯片和Intel芯片的现成.dmg文件,Linux也有现成的安装包,但只构建了Ubuntu 26.04版本(OBS-Studio-33.0.1-Ubuntu-26.04-x86_64.deb),所以其他Linux系统可能需要另找来源。","notVerified":["录制和直播在你电脑的显卡和处理器上表现如何。","游戏捕获功能是否适用于你实际使用的那些游戏或应用。","开始保存录像之后实际的磁盘占用情况,这可能会迅速增长。","Ubuntu安装包能否在其他Linux系统上安装。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyOTExMzc=","slug":"ohmyzsh__ohmyzsh","name":"Oh My Zsh","owner":"ohmyzsh","fullName":"ohmyzsh/ohmyzsh","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"install_script_from_internet","severity":"info","evidence":"README.md 中的安装命令：sh -c \"$(curl -fsSL https://raw.githubusercontent.com/ohmyzsh/ohmyzsh/master/tools/install.sh)\"。README 本身建议先下载并阅读 install.sh：\"It's a good idea to inspect the install script from projects you don't yet know.\"","source":"ai","title":"从网上下载安装脚本"},{"code":"replaces_shell_settings","severity":"info","evidence":"README.md：\"any previous `.zshrc` will be renamed to `.zshrc.pre-oh-my-zsh`\"；此外，除非使用 --unattended 运行，安装程序还会更改你的默认 shell。","source":"ai","title":"会替换 shell 设置"},{"code":"pulls_updates_from_main_branch","severity":"info","evidence":"README.md：\"By default, you will be prompted to check for updates every 2 weeks\" 以及 \"By default, updates always pull the latest changes.\" 设置 `zstyle ':omz:update' cooldown 10` 可以推迟更新。","source":"ai","title":"更新来自主分支"},{"code":"plugins_run_on_every_terminal","severity":"info","evidence":"README.md：插件列在 `~/.zshrc` 中，并在 shell 启动时加载，而且 \"If you want to override any of the default behaviors, just add a new file (ending in `.zsh`) in the `custom/` directory.\" 你在那里添加的任何内容，每次打开终端时都会运行。","source":"ai","title":"每次打开终端都会运行插件"},{"code":"no_versioned_releases","severity":"info","evidence":"项目信息：\"GitHub releases: 0\"，检查结果也没有给出最近一次发布的日期。你安装的是当时 master 分支上的内容，而更新程序也从同一个分支拉取。","source":"ai","title":"没有带版本号的发布"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","network_system","productivity"],"alternativeTo":[],"declaredAlternatives":[],"stars":190433,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":130,"growthDays":7,"history":[{"date":"2026-10-26","stars":190179},{"date":"2026-10-27","stars":190192},{"date":"2026-10-28","stars":190215},{"date":"2026-10-29","stars":190233},{"date":"2026-10-30","stars":190254},{"date":"2026-10-31","stars":190271},{"date":"2026-11-01","stars":190283},{"date":"2026-11-02","stars":190291},{"date":"2026-11-03","stars":190303},{"date":"2026-11-04","stars":190317},{"date":"2026-11-05","stars":190327},{"date":"2026-11-06","stars":190333},{"date":"2026-11-07","stars":190350},{"date":"2026-11-08","stars":190375},{"date":"2026-11-09","stars":190392},{"date":"2026-11-10","stars":190433}],"firstPublishedAt":"2026-10-27T01:10:12.667Z","commitSha":"17a2b4c3a5855524ad4772a17ada0dc474e2fdf3","verifiedAt":"2026-10-30T01:43:14.758Z","lastCommitAt":"2026-11-09T12:36:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds themes, shortcuts and hundreds of plugins to the zsh terminal so the command line looks nicer and is quicker to use.","summary":"Oh My Zsh is a set of ready-made settings for zsh, a program for typing commands in a terminal. It adds 300+ optional plugins (for git, Docker, Python and more), 140+ themes that change how the command prompt looks, and an update tool. It is for people who spend time in the terminal, mostly programmers. It runs on macOS and Linux; on Windows it only works inside WSL2 (Linux running on Windows). You need zsh, git and curl or wget, then you run one install command and edit a settings file to pick plugins and themes.","easeRationale":"Mac sets the grade: zsh is already the shell there, so you paste one install command into Terminal and then edit ~/.zshrc by hand to choose plugins and a theme. Linux is the same work plus installing zsh with your package manager first. Windows is harder because the README's compatibility table only lists \"Windows (WSL2)\", so you must set up WSL2 and a Linux system before you start.","costNote":"Free and open source (MIT). The README links to optional merchandise.","notVerified":["How smoothly it works inside WSL2 on a given Windows setup.","How much the chosen plugins and themes slow down opening a new terminal.","Whether a theme needs an extra font (Powerline or Nerd Font) to display correctly on your machine.","What the install script actually changes on your machine, since it is fetched at install time.","How much disk space the clone and its plugins take, which the checks could not determine."]},"zh-Hans":{"summaryShort":"为zsh终端添加主题、快捷方式和数百个插件,让命令行界面更好看、用起来更快。","summary":"Oh My Zsh是一套为zsh准备好的配置,zsh是一个在终端里输入命令的程序。它提供300多个可选插件(用于git、Docker、Python等),140多种可以改变命令提示符外观的主题,还有一个更新工具。它适合经常使用终端的人,主要是程序员。它可以在macOS和Linux上运行;在Windows上只能在WSL2(在Windows里运行的Linux)里使用。你需要先有zsh、git和curl或wget,然后运行一条安装命令,再编辑一个配置文件来选择插件和主题。","easeRationale":"Mac决定了这个评级:在Mac上zsh已经是默认shell,所以你只需要把一条安装命令粘贴到终端里,然后手动编辑~/.zshrc来选择插件和主题。Linux的工作量差不多,只是要先用你的包管理器安装zsh。Windows更难一些,因为README的兼容性表格里只列出了\"Windows (WSL2)\",所以你得先搭建好WSL2和一个Linux系统才能开始。","costNote":"免费且开源(MIT协议)。README里有可选周边商品的链接。","notVerified":["在某个具体的Windows环境里,它在WSL2内部运行得有多顺畅。","选用的插件和主题会让打开新终端变慢多少。","某个主题是否需要额外的字体(Powerline或Nerd Font)才能在你的电脑上正确显示。","安装脚本实际会在你的电脑上改动什么,因为它是在安装时才下载的。","克隆仓库和其插件占用多少磁盘空间,检查未能确定这一点。"]}}},{"id":"R_kgDOJ0Z1Pg","slug":"ollama__ollama","name":"Ollama","owner":"ollama","fullName":"ollama/ollama","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"C"},"risk":"low","signals":[{"code":"readme_alternatives_belong_to_other_projects","severity":"info","evidence":"\"Community Integrations\" 列表介绍的是其他人的项目，而不是 Ollama 本身：\"[Void] - Open source AI code editor, Cursor alternative\"、\"[twinny] - Copilot and Copilot chat alternative\"、\"[AppFlowy] - AI collaborative workspace, self-hostable Notion alternative\"、\"[Perplexica] - AI-powered search engine, open-source Perplexity alternative\"。Ollama 本身只是这些应用所连接的引擎。","source":"ai","title":"README 里的替代品属于其他项目"},{"code":"local_api_without_a_password","severity":"info","evidence":"README：`curl http://localhost:11434/api/chat -d ...` 不需要密钥或密码，社区列表中还包括 \"[Ollama Fortress] - Security proxy for Ollama\"。任何能访问该端口的程序都可以使用你的模型；它通常只在你自己的电脑上监听。","source":"ai","title":"本地接口没有密码"},{"code":"install_script_runs_from_website","severity":"info","evidence":"README 的 Download 部分：Mac 和 Linux 使用 `curl -fsSL https://ollama.com/install.sh | sh`，Windows 使用 `irm https://ollama.com/install.ps1 | iex`。这些命令会运行当时从 ollama.com 获取的脚本；README 也提供了手动安装程序的链接。","source":"ai","title":"安装脚本直接从网站运行"},{"code":"models_have_their_own_licences","severity":"info","evidence":"项目描述列出了从 ollama.com/library 拉取的 \"Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models\"。Ollama 自身的 MIT 许可证并不涵盖这些模型权重，它们带有各自的条款。","source":"ai","title":"模型有各自的许可证"},{"code":"large_download_and_disk_use","severity":"info","evidence":"发布文件本身就很大（OllamaSetup.exe 为 1598 MB，ollama-windows-amd64.zip 为 1528 MB），之后你每添加一个模型，通常还要再多占用几 GB。","source":"ai","title":"下载量大，占用磁盘多"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","go"],"disk":{"minMB":1598,"maxMB":4795,"basis":"installer"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["local_ai","development"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true}],"declaredAlternatives":[{"target":"cursor","stance":"claims","evidence":"- [Void](https://github.com/voideditor/void) - Open source AI code editor, Cursor alternative"},{"target":"github-copilot","stance":"claims","evidence":"- [twinny](https://github.com/rjmacarthy/twinny) - Copilot and Copilot chat alternative"},{"target":"invideo-ai","stance":"claims","evidence":"- [Perplexica](https://github.com/ItzCrazyKns/Perplexica) - AI-powered search engine, open-source Perplexity alternative"},{"target":"notion","stance":"claims","evidence":"- [AppFlowy](https://github.com/AppFlowy-IO/AppFlowy) - AI collaborative workspace, self-hostable Notion alternative"},{"target":"perplexity","stance":"claims","evidence":"- [Perplexica](https://github.com/ItzCrazyKns/Perplexica) - AI-powered search engine, open-source Perplexity alternative"}],"stars":186564,"downloads":242298241,"downloadsWeek":3721029,"downloadsDays":7,"growth":748,"growthDays":7,"history":[{"date":"2026-10-25","stars":184767},{"date":"2026-10-26","stars":184901},{"date":"2026-10-27","stars":185027},{"date":"2026-10-28","stars":185141},{"date":"2026-10-29","stars":185256},{"date":"2026-10-30","stars":185382},{"date":"2026-10-31","stars":185479},{"date":"2026-11-01","stars":185611},{"date":"2026-11-02","stars":185721},{"date":"2026-11-03","stars":185816},{"date":"2026-11-04","stars":185924},{"date":"2026-11-05","stars":186024},{"date":"2026-11-06","stars":186129},{"date":"2026-11-07","stars":186271},{"date":"2026-11-08","stars":186372},{"date":"2026-11-09","stars":186443},{"date":"2026-11-10","stars":186564}],"firstPublishedAt":"2026-10-25T23:19:20.421Z","commitSha":"6db7be24cc3b7a53506640e88856746e27d60caa","verifiedAt":"2026-10-29T01:41:34.766Z","lastCommitAt":"2026-11-10T22:43:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Download and chat with open AI models like DeepSeek, Qwen and Gemma on your own computer.","summary":"Ollama runs open AI language models on your own computer. You install it, pick a model, and chat with it in a window or a terminal, or connect it to other apps and coding tools through its local web interface. Models run on your machine, so your conversations stay there unless you choose Ollama's optional cloud models. Models are several gigabytes each, big ones need a lot of memory, and a good graphics card makes them much faster.","easeRationale":"Windows has a ready installer (OllamaSetup.exe) and Mac a .dmg, and after installing you only pick and download a model. On Linux the README gives only a terminal install command (`curl -fsSL https://ollama.com/install.sh | sh`) and you use it by typing commands.","costNote":"Running models on your own machine is free. The README offers access \"locally or in the cloud\"; the cloud option is a paid service from Ollama and the package does not state prices.","notVerified":["How fast models run on your specific computer","Real disk use after downloading models, which are often several GB each","What the optional cloud models cost and what data they send","Whether the installer leaves a background service running and listening after you close the app"]},"zh-Hans":{"summaryShort":"在自己的电脑上下载并与DeepSeek、Qwen、Gemma等开放AI模型聊天。","summary":"Ollama 在你自己的电脑上运行开放的AI语言模型。你安装它、选一个模型,然后在窗口或终端里跟它聊天,也可以通过它的本地网页接口连接到其他应用和编程工具。模型在你的机器上运行,所以你的对话都留在本地,除非你选择Ollama的可选云端模型。每个模型有几个GB大小,大模型需要很多内存,一块好显卡会让它们快得多。","easeRationale":"Windows有现成的安装程序(OllamaSetup.exe),Mac有.dmg文件;安装后你只需选择并下载一个模型。在Linux上README只给出一条终端安装命令(`curl -fsSL https://ollama.com/install.sh | sh`),之后靠输入命令来使用。","costNote":"在自己的机器上运行模型是免费的。README提供\"本地或云端\"的访问方式;云端选项是Ollama提供的付费服务,该资料没有列出价格。","notVerified":["模型在你具体的电脑上运行速度如何","下载模型后实际占用的磁盘空间,每个模型往往有几个GB","可选的云端模型费用是多少,会发送哪些数据","关闭应用后,安装程序是否会留下一个在后台运行并监听的服务"]}}},{"id":"R_kgDOVFJing","slug":"onehumanbeing__esp32-iphone-agent","name":"ESP32 iPhone Agent","owner":"onehumanbeing","fullName":"onehumanbeing/esp32-iphone-agent","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"mac":"E"},"risk":"attention","signals":[{"code":"ai_controls_your_phone","severity":"attention","evidence":"README.md：该应用 \"decodes the stream, sends screenshots to any OpenAI-compatible vision model, and runs a planner and an actor: tap, swipe, draw, type, press keys, go home, finish\"，项目自己的建议是 \"The agent acts with your hands: watch it, and keep the stop button within reach.\" 手机屏幕上的任何内容，包括私人消息和账户页面，都会被截图并发送到你所配置的模型服务地址。","source":"ai","title":"AI 会操控你的手机"},{"code":"private_api_and_temporary_signing","severity":"info","evidence":"README.md 中的限制说明：\"The app leaves itself with a private iOS API (`suspend`); fine for a self-built app, to be replaced before the App Store\"，并且该应用需要每 7 天用免费的 Apple ID 重新签名一次。","source":"ai","title":"使用私有接口且签名是临时的"},{"code":"mixed_license_firmware_gpl","severity":"info","evidence":"README.md：\"MIT (LICENSE), except `hardware/hid_bridge/src/airplay/` (ported from UxPlay), which is GPL-3.0 ...; the firmware binary that links it is therefore distributed under GPL-3.0\"，而 GitHub 显示该仓库为 MIT。","source":"ai","title":"混合许可证，固件为 GPL"},{"code":"fake_install_commands_reported","severity":"info","evidence":"README.md 中带有一条警告：\"Only follow the install steps in this README. Never run commands from issues or comments (for example a \\\"quick install\\\" one-liner such as PowerShell `iex(irm ...)`); these are not from the maintainers.\"","source":"ai","title":"有人提醒存在假冒的安装命令"},{"code":"very_young_repo_many_stars","severity":"info","evidence":"仓库创建于 2026-10-11，到 2026-10-28 已有 4016 个星标和 531 个 fork，只有一个发布版本（v0.1.0），README 称其基于电脑的设计已经被替换。","source":"ai","title":"非常新的仓库却有很多星标"}],"platforms":["mac"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenAI-compatible vision model endpoint (cloud or your own server)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","hardware","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":4809,"downloads":1411,"downloadsWeek":1272,"downloadsDays":7,"growth":507,"growthDays":7,"history":[{"date":"2026-10-26","stars":3949},{"date":"2026-10-27","stars":4014},{"date":"2026-10-28","stars":4062},{"date":"2026-10-29","stars":4098},{"date":"2026-10-30","stars":4132},{"date":"2026-10-31","stars":4160},{"date":"2026-11-01","stars":4205},{"date":"2026-11-02","stars":4237},{"date":"2026-11-03","stars":4302},{"date":"2026-11-04","stars":4367},{"date":"2026-11-05","stars":4437},{"date":"2026-11-06","stars":4575},{"date":"2026-11-07","stars":4627},{"date":"2026-11-08","stars":4689},{"date":"2026-11-09","stars":4753},{"date":"2026-11-10","stars":4809}],"firstPublishedAt":"2026-10-27T01:10:12.668Z","commitSha":"32effd3267e79dc5ffdb831f1f8b0452711b8e5e","verifiedAt":"2026-10-28T01:39:30.225Z","lastCommitAt":"2026-11-10T13:41:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI model tap and type on a real iPhone using a $7 ESP32 board as its eyes and hands.","summary":"This project turns a $7 ESP32-S3 board into the eyes and hands of an AI agent on an iPhone: the phone mirrors its screen to the board over the board's own Wi-Fi, the board acts as a USB mouse and keyboard, and an app on the same phone sends screenshots to a vision model and carries out the taps and swipes. It is for tinkerers: you need the board, an iPhone on iOS 18 or later, a Mac with Xcode to flash the board and build the app once, and an endpoint for an OpenAI-compatible vision model with a key.","easeRationale":"The only computer route is a Mac and it is grade E: you install arduino-cli plus two board libraries, run hardware/flash.sh to flash the board, then generate an Xcode project with XcodeGen and build and sign the iPhone app yourself (\"A free Apple ID can sign it (re-sign every 7 days)\"). There is no ready download, and the device that is actually automated is an iPhone, which Radar does not grade.","costNote":"The code is free. You buy the ESP32-S3 board (about $7) and pay for the vision model, unless you run your own model server (vLLM, llama.cpp or Ollama are listed as options).","notVerified":["Whether it works with your board model, your iPhone and your cellular plan","What the vision model calls cost for a real task, since every step sends a screenshot","Whether the model provider keeps the screenshots of your phone","How often the agent taps the wrong thing, which matters because it acts inside your own apps and accounts","Whether the exported run traces, kept on the phone, include sensitive screen content"]},"zh-Hans":{"summaryShort":"让AI模型用一块7美元的ESP32开发板作为“眼睛和手”,在真实iPhone上点击和输入。","summary":"这个项目把一块7美元的ESP32-S3开发板变成iPhone上AI智能体的“眼睛和手”:手机通过开发板自带的Wi-Fi把屏幕投屏给开发板,开发板扮演USB鼠标和键盘的角色,同一台手机上的一个App把截图发给视觉模型,再执行点击和滑动操作。这适合喜欢自己动手折腾的人:你需要这块开发板、一台运行iOS 18或更高版本的iPhone、一台装有Xcode的Mac(用来给开发板刷机并编译一次App),以及一个兼容OpenAI接口的视觉模型端点和对应的密钥。","easeRationale":"唯一的电脑途径是Mac,等级为E:你要安装arduino-cli和两个开发板库,运行hardware/flash.sh给开发板刷机,然后用XcodeGen生成一个Xcode项目,自己编译并签名这个iPhone App(“免费的Apple ID也能签名,但每7天要重新签一次”)。没有现成的下载包,而真正被自动操作的设备是iPhone,Radar不对其评级。","costNote":"代码是免费的。你需要购买ESP32-S3开发板(约7美元),并为视觉模型付费,除非你自己运行模型服务器(文档中列出的选项有vLLM、llama.cpp或Ollama)。","notVerified":["是否适配你具体的开发板型号、你的iPhone以及你的移动流量套餐","在实际任务中,视觉模型的调用费用是多少,因为每一步都会发送一张截图","模型提供商是否会保留你手机上的这些截图","智能体点错东西的频率有多高,这一点很重要,因为它是在你自己的App和账户内操作","保存在手机上导出的运行记录,是否包含敏感的屏幕内容"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2OTg2NzI5OA==","slug":"onlyoffice__desktopeditors","name":"ONLYOFFICE Desktop Editors","owner":"ONLYOFFICE","fullName":"ONLYOFFICE/DesktopEditors","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":286,"maxMB":858,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["office"],"alternativeTo":[{"slug":"word","name":"Word","confirmed":true},{"slug":"excel","name":"Excel","confirmed":true},{"slug":"powerpoint","name":"PowerPoint","confirmed":true},{"slug":"acrobat","name":"Acrobat","confirmed":true}],"declaredAlternatives":[],"stars":5715,"downloads":16577865,"downloadsWeek":381876,"downloadsDays":7,"growth":26,"growthDays":7,"history":[{"date":"2026-10-25","stars":5633},{"date":"2026-10-26","stars":5641},{"date":"2026-10-27","stars":5651},{"date":"2026-10-28","stars":5656},{"date":"2026-10-29","stars":5659},{"date":"2026-10-30","stars":5671},{"date":"2026-10-31","stars":5678},{"date":"2026-11-01","stars":5681},{"date":"2026-11-02","stars":5689},{"date":"2026-11-03","stars":5689},{"date":"2026-11-04","stars":5693},{"date":"2026-11-05","stars":5692},{"date":"2026-11-06","stars":5698},{"date":"2026-11-07","stars":5703},{"date":"2026-11-08","stars":5707},{"date":"2026-11-09","stars":5713},{"date":"2026-11-10","stars":5715}],"firstPublishedAt":"2026-10-25T23:19:21.287Z","commitSha":"daf8c6948148827201b2907ac1134d222fbfa7b4","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-10-22T17:54:33Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"DesktopEditors_x64.exe","sizeBytes":299860896,"url":"https://github.com/ONLYOFFICE/DesktopEditors/releases/download/v10.0.0/DesktopEditors_x64.exe","releaseUrl":"https://github.com/ONLYOFFICE/DesktopEditors/releases/tag/v10.0.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Create and edit documents, spreadsheets, presentations and PDFs on your computer, with good Word, Excel and PowerPoint file support.","summary":"ONLYOFFICE Desktop Editors is a free office suite with a word processor, spreadsheet, presentation and PDF editor, plus a form creator and a diagram viewer. It works offline on files stored on your computer and is built to open and save Microsoft Office formats (.docx, .xlsx, .pptx). You can optionally connect it to cloud services such as Nextcloud, Dropbox or ONLYOFFICE's own cloud to work on files with others. It is for anyone who needs an office suite without a subscription.","easeRationale":"The latest release includes ready Windows installers (DesktopEditors_x64.exe, plus arm64 and 32-bit versions), so you download one and open it.","costNote":"GNU AGPL version 3. Cloud services you connect it to may have their own prices.","notVerified":["How exactly complex Word, Excel and PowerPoint files keep their layout after editing","Which AI services the AI assistant feature uses and whether they need a paid account or key","What data is sent when you connect a cloud account"]},"zh-Hans":{"summaryShort":"在你的电脑上创建和编辑文档、电子表格、演示文稿和 PDF，对 Word、Excel 和 PowerPoint 文件支持良好。","summary":"ONLYOFFICE Desktop Editors 是一套免费的办公软件，包含文字处理、电子表格、演示文稿和 PDF 编辑器，还有表单创建工具和图表查看器。它可以离线处理保存在你电脑上的文件，并且能够打开和保存 Microsoft Office 格式（.docx、.xlsx、.pptx）。你还可以选择把它连接到 Nextcloud、Dropbox 或 ONLYOFFICE 自家的云服务，与他人一起处理文件。它适合所有需要不用订阅的办公软件的人。","easeRationale":"最新版本包含现成的 Windows 安装程序（DesktopEditors_x64.exe，另有 arm64 和 32 位版本），下载一个并打开即可。","costNote":"采用 GNU AGPL 第 3 版许可证。你连接的云服务可能有各自的价格。","notVerified":["复杂的 Word、Excel 和 PowerPoint 文件在编辑后能在多大程度上保持版式","AI 助手功能使用哪些 AI 服务，以及它们是否需要付费账号或密钥","连接云账号时会发送哪些数据"]}}},{"id":"R_kgDOIpvk-w","slug":"oobabooga__textgen","name":"Text Generation WebUI","owner":"oobabooga","fullName":"oobabooga/textgen","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"flags_can_put_it_on_the_internet","severity":"attention","evidence":"命令行参数列表包含 `--listen`（\"Make the web UI reachable from your local network\"）、`--share`（\"Create a public URL\"）和 `--public-api`（\"Create a public URL for the API using Cloudflare\"）。密码是另外单独的选项（`--gradio-auth`、`--api-key`），所以如果开了公开地址却没设密码，任何拿到链接的人都能打开你的模型和聊天记录。","source":"ai","title":"某些参数会把它暴露到公网上"},{"code":"option_runs_code_shipped_with_a_model","severity":"attention","evidence":"`--trust-remote-code` 这个参数的说明是 \"Set trust_remote_code=True while loading the model. Necessary for some models.\" 开启后，下载的模型里自带的 Python 代码会在你的电脑上运行；另外还有一个参数专门用于强制使用能 \"prevents arbitrary code execution\" 的文件格式。","source":"ai","title":"某个选项会运行模型自带的代码"},{"code":"privacy_claim_is_conditional","severity":"info","evidence":"README 写着 \"100% offline and private, with zero telemetry, external resources, or remote update requests.\" 但当你使用文档中介绍的功能时，应用仍会联网：你需要从 Hugging Face 下载模型，内置的工具调用会执行 \"web search, page fetching, and math\"。","source":"ai","title":"隐私承诺是有前提的"},{"code":"full_install_downloads_a_lot_at_run_time","severity":"info","evidence":"完整安装\"Requires ~10GB disk space and downloads PyTorch\"，一键脚本则\"uses Miniforge to set up a Conda environment in installer_files/\"，所以大部分实际运行的内容是在你启动脚本之后才下载的，并不是来自你下载的发行文件本身。","source":"ai","title":"完整安装会在运行时下载大量内容"},{"code":"portable_builds_lag_the_code","severity":"info","evidence":"最新版本是 2026-05-20 发布的 v4.9，而最后一次提交是在 2026-08-17——现成的安装包比代码落后了好几个月，8 月以来再没有新提交，目前还有 867 个未处理的 issue。","source":"ai","title":"便携版落后于最新代码"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":1361,"maxMB":4082,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["local_ai","development"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"midjourney","name":"Midjourney","confirmed":false}],"declaredAlternatives":[],"stars":47734,"downloads":437757,"downloadsWeek":2666,"downloadsDays":6,"growth":3,"growthDays":7,"history":[{"date":"2026-11-04","stars":47731},{"date":"2026-11-05","stars":47736},{"date":"2026-11-06","stars":47737},{"date":"2026-11-07","stars":47736},{"date":"2026-11-08","stars":47734},{"date":"2026-11-09","stars":47732},{"date":"2026-11-10","stars":47734}],"firstPublishedAt":"2026-11-08T23:08:23.049Z","commitSha":"c93f8871239550de2ccfe1e95d469aa82616f07e","verifiedAt":"2026-11-08T22:49:54.086Z","lastCommitAt":"2026-08-17T17:38:15Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Download, unzip and double-click to chat with AI language models running on your own computer.","summary":"TextGen (formerly text-generation-webui) is a desktop app for AI language models that run on your machine. Portable packages for Windows, macOS and Linux include everything; you add a GGUF model file to the user_data/models folder and chat. It handles images in messages, PDF and Word attachments, message editing and branching, model tool-calling with web search, a tab for generating pictures, LoRA training, and an OpenAI/Anthropic-compatible API other programs can call. Each system has builds for CPU only, NVIDIA (CUDA), AMD (ROCm) and Vulkan.","easeRationale":"All three systems are B: the README says \"Download, unzip, double-click textgen. A window opens\", and each release has a portable package for them (for example textgen-portable-4.9-windows-cuda12.4.zip, 936 MB, or textgen-portable-4.9-macos-arm64.tar.gz, 272 MB), but you still have to fetch a GGUF model from Hugging Face and drop it into user_data/models. Extra backends, training and image generation need the full install instead, with Conda, PyTorch and about 10 GB of disk.","costNote":"Free under AGPL-3.0, a copyleft licence. The author states that Andreessen Horowitz gave a grant for the work in August 2023.","notVerified":["Whether the portable build starts on your particular graphics card, and which of the CUDA, ROCm, Vulkan or CPU packages you need","Real disk use after downloading models, which are usually several gigabytes each","Whether the app truly sends nothing out, which can only be seen by watching the network while it runs","How fast answers are on a computer without a graphics card","Whether the image generation and training tabs work as described"]},"zh-Hans":{"summaryShort":"下载、解压、双击即可与在你自己电脑上运行的 AI 语言模型聊天。","summary":"TextGen（原名 text-generation-webui）是一款桌面应用，用于在你自己的电脑上运行 AI 语言模型。Windows、macOS 和 Linux 的便携包已包含所需的一切；你只需把 GGUF 模型文件放进 user_data/models 文件夹就可以开始聊天。它支持消息中的图片、PDF 和 Word 附件、消息编辑与分支、带网页搜索的模型工具调用、一个生成图片的选项卡、LoRA 训练，以及一个兼容 OpenAI/Anthropic 的 API，供其他程序调用。每个系统都提供仅 CPU、NVIDIA（CUDA）、AMD（ROCm）和 Vulkan 版本。","easeRationale":"三个系统的评级都是 B：README 写着 \"Download, unzip, double-click textgen. A window opens\"，每个版本都有对应的便携包（例如 936 MB 的 textgen-portable-4.9-windows-cuda12.4.zip，或 272 MB 的 textgen-portable-4.9-macos-arm64.tar.gz），但你仍需要从 Hugging Face 下载一个 GGUF 模型文件并放进 user_data/models。额外的后端、训练和图像生成功能则需要完整安装，涉及 Conda、PyTorch 和约 10 GB 的磁盘空间。","costNote":"采用 AGPL-3.0 协议，免费使用，这是一种 copyleft 许可证。作者表示 Andreessen Horowitz 在 2023 年 8 月为这项工作提供了一笔资助。","notVerified":["便携版能否在你那块具体的显卡上启动，以及你需要 CUDA、ROCm、Vulkan 还是 CPU 版本","下载模型后实际占用的磁盘空间，这些模型通常各有好几 GB","这个应用是否真的不对外发送任何数据，这只能靠在运行时监控网络来确认","在没有显卡的电脑上回答速度有多快","图像生成和训练这两个选项卡是否真的像说明那样可用"]}}},{"id":"R_kgDOQ8h7kg","slug":"op7418__humanizer-zh","name":"Humanizer-zh","owner":"op7418","fullName":"op7418/Humanizer-zh","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"purpose_is_to_remove_signs_of_ai_writing","severity":"attention","evidence":"仓库描述说这个技能 \"aims to erase the traces of AI generation in a text\"（旨在消除文本中 AI 生成的痕迹）。用它把 AI 写的文本冒充成自己写的，可能会违反学校、出版方或平台的规定。README 本身也提醒 \"cannot prove who wrote an article, and does not guarantee passing any AI detector\"。","source":"ai","title":"目的是消除 AI 写作的痕迹"},{"code":"it_rewrites_your_files_directly","severity":"info","evidence":"文档给出的用法是 \"请用 humanizer-zh 润色 article.md 中的正文\"——助手会直接编辑文件本身。README 说如果你只想要一份审阅意见，就必须说 \"give suggestions, do not change the file\"（给出建议，不修改文件）。","source":"ai","title":"它会直接改写你的文件"},{"code":"the_file_is_an_instruction_for_your_agent","severity":"info","evidence":"README 写道，这是 \"editing guidance read and executed by an agent, not a standalone detection program\"。仓库里只有 SKILL.md、README.md、CHANGELOG.md 和 tests/，所以这个文件一旦更新，你的助手改写文本的方式也会跟着变。","source":"ai","title":"这个文件是给你的 agent 的指令"},{"code":"effect_is_not_measured","severity":"info","evidence":"测试部分描述了 18 个简短案例、一个 Markdown 样本和一个结构检查脚本，然后说明单一且有限的样本无法代表所有模型和风格，字数的减少或模型自己打出的分数也无法证明效果。","source":"ai","title":"效果并没有被实际衡量"},{"code":"translation_of_another_project","severity":"info","evidence":"来源说明中致谢 \"blader/humanizer v3.0.0\" 为原始项目，也是 A-E 分类、语气校准和文件模式的来源，额外的中文检查点则来自本仓库的 PR #39。","source":"ai","title":"改编自另一个项目"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","office"],"alternativeTo":[{"slug":"quillbot","name":"QuillBot","confirmed":true},{"slug":"undetectable-ai","name":"Undetectable AI","confirmed":true},{"slug":"grammarly","name":"Grammarly","confirmed":true}],"declaredAlternatives":[],"stars":22492,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":512,"growthDays":6,"history":[{"date":"2026-11-06","stars":22152},{"date":"2026-11-07","stars":22209},{"date":"2026-11-08","stars":22284},{"date":"2026-11-09","stars":22378},{"date":"2026-11-10","stars":22492}],"firstPublishedAt":"2026-11-08T23:08:23.071Z","commitSha":"f4518a8eab97b8bfebc66a89d34320a89bef6930","verifiedAt":"2026-11-08T22:50:54.115Z","lastCommitAt":"2026-09-23T02:24:26Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Chinese-language skill for Claude Code that rewrites text to cut empty phrases and template-sounding sentences.","summary":"Humanizer-zh is a single instruction file (SKILL.md) you install into Claude Code. You paste Chinese text or point it at a file, and the assistant rewrites it following 31 checkpoints: filler openings, forced rule-of-three rhythm, inflated wording, decorative formatting, chatbot leftovers and six checks specific to Chinese. The rules tell it to keep facts, hedging and the author's voice. The README says plainly that it is guidance read by an agent, cannot prove who wrote a text and does not guarantee passing any AI detector.","easeRationale":"There is nothing to download and open: on Windows, macOS and Linux alike you need Claude Code already installed and then one command, `npx skills add https://github.com/op7418/Humanizer-zh.git` (which needs Node) or a git clone into `~/.claude/skills/`, followed by a restart. The texts and the rules are in Chinese, so it is of little use for other languages.","costNote":"The file is free (MIT). It only does something inside Claude Code, which is a paid assistant.","notVerified":["Whether the rewritten text reads better to a Chinese reader, or just different","Whether facts, hedging and numbers really survive every rewrite","How much the long instruction file adds to your Claude Code usage","Whether it changes anything for any particular AI detector"]},"zh-Hans":{"summaryShort":"一个面向 Claude Code 的中文技能，用来改写文本，去掉空话和模板化的句子。","summary":"Humanizer-zh 是一个单独的指令文件（SKILL.md），你把它安装进 Claude Code。你粘贴一段中文文本或指向一个文件，助手就会按照 31 个检查点来改写：开头的套话、刻意的\"三段式\"节奏、夸大的用词、装饰性的排版、聊天机器人式的残留痕迹，以及六项中文特有的检查。规则要求它保留事实、保留留有余地的说法，以及作者本人的语气。README 明确说明，这只是给 agent 阅读的指导，不能证明文章是谁写的，也不保证能通过任何 AI 检测器。","easeRationale":"没有需要下载打开的东西：在 Windows、macOS 和 Linux 上，你都需要先装好 Claude Code，然后只需一条命令，`npx skills add https://github.com/op7418/Humanizer-zh.git`（需要 Node），或者在 `~/.claude/skills/` 下执行 git clone，再重启一次即可。文本和规则都是中文写的，所以对其他语言用处不大。","costNote":"该文件是免费的（MIT 协议）。它只能在 Claude Code 内部起作用，而 Claude Code 是付费助手。","notVerified":["改写后的文本对中文读者来说是真的更好，还是只是变得不一样","事实、留有余地的说法和数字是否真的在每次改写中都保留下来","这个长长的指令文件会给你的 Claude Code 用量增加多少","它对某个具体的 AI 检测器是否真的有影响"]}}},{"id":"R_kgDOKxW6xA","slug":"open-llm-vtuber__open-llm-vtuber","name":"Open-LLM-VTuber","owner":"Open-LLM-VTuber","fullName":"Open-LLM-VTuber/Open-LLM-VTuber","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖中存在已知漏洞（OSV）：aiohttp: GHSA-2fqr-mr3j-6wp8, GHSA-2g87-pp6c-29x5, GHSA-2vrm-gr82-f7m5; brotli: GHSA-2qfp-q593-8484, GHSA-9v8h-qqgm-4m35, PYSEC-2026-2401; mcp: GHSA-5h93-6whr-6q8j, GHSA-84m7-p3x7-pcfv, GHSA-9h52-p55h-vw2f; requests: GHSA-gc5v-m9x4-r6x2, PYSEC-2026-2275; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p; torch: GHSA-63cw-57p8-fm3p, GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47; aiohttp: GHSA-2fqr-mr3j-6wp8, GHSA-2g87-pp6c-29x5, GHSA-2vrm-gr82-f7m5; anyio: GHSA-5p39-cfhj-2xmp, GHSA-82r6-8w77-94w6, PYSEC-2026-4024; certifi: GHSA-6q48-6c7w-9gw2, PYSEC-2026-4320; cryptography: GHSA-537c-gmf6-5ccf, GHSA-g6cj-pr64-35w5, GHSA-jwv3-5hgf-82ww; dnspython: GHSA-45c3-73f7-pv4m; filelock: GHSA-69hj-f68j-2879, GHSA-8mx5-7m5m-r77g; fsspec: GHSA-27vj-qcqg-25rc, PYSEC-2026-4217; idna: GHSA-65pc-fj4g-8rjx, PYSEC-2026-215; mcp: GHSA-5h93-6whr-6q8j, GHSA-84m7-p3x7-pcfv, GHSA-hvrp-rf83-w775; multidict: GHSA-54p9-h82j-f925, PYSEC-2026-4244; protobuf: GHSA-7gcm-g887-7qv7, PYSEC-2026-1805; pydantic-settings: GHSA-4xgf-cpjx-pc3j; pygments: GHSA-5239-wwwm-4pmq, PYSEC-2026-2987; pyjwt: GHSA-2gx3-rcp4-g85q, GHSA-42vr-xj54-vc7v, GHSA-752w-5fwx-jx9f; python-dotenv: GHSA-mf9w-mj56-hr94, PYSEC-2026-2270; python-multipart: GHSA-5rvq-cxj2-64vf, GHSA-6jv3-5f52-599m, GHSA-mj87-hwqh-73pj; requests: GHSA-gc5v-m9x4-r6x2, PYSEC-2026-2275; setuptools: GHSA-h35f-9h28-mq5c, GHSA-xmj3-9gf7-h52w, PYSEC-2026-3447; soupsieve: GHSA-2wc2-fm75-p42x, GHSA-836r-79rf-4m37, GHSA-gjv8-xp57-g29c; starlette: GHSA-82w8-qh3p-5jfq, GHSA-86qp-5c8j-p5mr, GHSA-jp82-jpqv-5vv3; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p; torch: GHSA-rrmf-rvhw-rf47, PYSEC-2026-139; urllib3: GHSA-8988-9cw3-xx77, GHSA-gh4c-6fx4-qh6g, GHSA-mf9v-mfxr-j63j; virtualenv: GHSA-394h-qwmc-48pf, GHSA-43mp-35q3-x3fq, GHSA-494p-63j8-28p3","source":"check","title":"已知漏洞"},{"code":"it_can_watch_your_camera_and_screen","severity":"attention","evidence":"README 的功能列表写着：\"Visual perception, supporting camera, screen recording and screenshots, allowing your AI companion to see you and your screen\"。如果你配置的是云端模型而不是本地模型，这些关于你本人和你屏幕的画面就会被发送给那家公司。","source":"ai","title":"它能看到你的摄像头和屏幕"},{"code":"included_characters_are_not_under_the_mit_licence","severity":"attention","evidence":"LICENSE 和 README 说明：Live2D 示例模型\"are licensed separately under the Live2D Free Material License Agreement ... They are not covered by the MIT license of this project\"，并且\"For commercial use, especially by medium or large-scale enterprises, the use of these Live2D sample models may be subject to additional licensing requirements.\"","source":"ai","title":"内置角色不适用 MIT 协议"},{"code":"version_1_is_on_hold_for_a_rewrite","severity":"info","evidence":"README 写道：\"We are focusing on Open-LLM-VTuber v2.0 — a complete rewrite of the codebase ... We kindly ask you to refrain from opening new issues or pull requests for feature requests on v1.\" 它还提醒 \"This project is in its early stages\"。最新版本是 2025-08-26 发布的 v1.2.1，最后一次提交是在 2026-05-15。","source":"ai","title":"v1 版本暂停维护，等待重写"},{"code":"downloads_models_outside_the_project_folder","severity":"info","evidence":"卸载说明提到，通过\"via ModelScope or Hugging Face may also be in MODELSCOPE_CACHE or HF_HOME\"获取的模型并不在项目自己的模型目录里，所以卸载应用后仍可能留下好几 GB 的文件。","source":"ai","title":"模型会下载到项目文件夹之外"},{"code":"sold_as_a_virtual_girlfriend_or_boyfriend","severity":"info","evidence":"README 写道：\"whether you want a virtual girlfriend, boyfriend, cute pet, or any other character, it can meet your expectations\"，仓库的标签里有 ai-waifu 和 ai-companion，用户评价部分还引用了 \"This girlfriend has been used over 100,000 times\"。角色设定其实是你自己写的提示词，所以角色说什么取决于你选择的模型和提示词。","source":"ai","title":"被当作虚拟女友或男友来宣传"}],"platforms":["windows","mac","linux"],"tools":["python","docker","ffmpeg","gpu_nvidia"],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":96,"maxMB":289,"basis":"installer"},"apiKeys":["OpenAI or any OpenAI-compatible service","Anthropic (Claude)","Azure Speech (recognition and voices)","ElevenLabs, Cartesia or Fish Audio voices","Groq"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["local_ai","audio","video"],"alternativeTo":[{"slug":"character-ai","name":"Character.AI","confirmed":true},{"slug":"replika","name":"Replika","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true}],"declaredAlternatives":[],"stars":14667,"downloads":42390,"downloadsWeek":496,"downloadsDays":3,"growth":83,"growthDays":6,"history":[{"date":"2026-11-07","stars":14627},{"date":"2026-11-08","stars":14643},{"date":"2026-11-09","stars":14660},{"date":"2026-11-10","stars":14667}],"firstPublishedAt":"2026-11-08T23:08:23.082Z","commitSha":"992309c0aa19845960228f880013d4685fde93b5","verifiedAt":"2026-11-08T22:51:50.257Z","lastCommitAt":"2026-05-15T07:18:03Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A voice AI companion with a Live2D cartoon avatar that talks with you on your desktop and can run offline.","summary":"Open-LLM-VTuber is an AI companion you talk to out loud. A Live2D cartoon character answers with a cloned or synthetic voice, shows expressions, can be interrupted mid-sentence and can sit on your desktop as a transparent pet window. You choose the parts: a local model through Ollama or LM Studio, or a paid cloud API; several speech-recognition and text-to-speech engines. The Windows and macOS downloads are the desktop client only — the server behind it is installed with Python, uv and ffmpeg, or with Docker, and configured in a conf.yaml file.","easeRationale":"All three are D. The releases only hold the desktop client (open-llm-vtuber-1.2.1-setup.exe, 96 MB, and the 165 MB .dmg); the part that does the work is a Python server you install with uv plus ffmpeg, or the included Dockerfile, then edit conf.yaml to pick a language model, a speech-recognition engine and a voice. Linux has no download at all and the README warns that reaching the server from another device needs https set up behind a reverse proxy.","costNote":"MIT licensed, except the included Live2D sample characters. It can run entirely on local models at no cost; the cloud options it supports (OpenAI, Claude, Azure Speech, ElevenLabs, Fish Audio, Groq) are paid per use.","notVerified":["Whether the voice conversation really keeps up on a computer without a graphics card","Total disk space once speech, voice and language models are downloaded","Whether the installed desktop client talks to the server without extra fiddling on each system","What the cloud providers keep of your voice, camera and screen images if you choose them","Whether the many listed speech and voice engines all still work in version 1.2.1"]},"zh-Hans":{"summaryShort":"一个带 Live2D 卡通形象的语音 AI 陪伴伙伴，可以在你的桌面上和你对话，也能离线运行。","summary":"Open-LLM-VTuber 是一个你可以开口交谈的 AI 陪伴伙伴。一个 Live2D 卡通角色会用克隆或合成的声音回答你，带有表情变化，可以在说话中途被打断，还能以透明桌面宠物窗口的形式停留在你的桌面上。各个部分都由你自己选择：通过 Ollama 或 LM Studio 使用本地模型，或者用付费的云端 API；还可以选择多种语音识别和文字转语音引擎。Windows 和 macOS 的下载包只是桌面客户端——背后的服务器需要用 Python、uv 和 ffmpeg 安装，或者用 Docker 安装，并在 conf.yaml 文件中配置。","easeRationale":"三个系统的评级都是 D。发布的版本只包含桌面客户端（96 MB 的 open-llm-vtuber-1.2.1-setup.exe，以及 165 MB 的 .dmg），真正干活的部分是一个 Python 服务器，需要你用 uv 加 ffmpeg 自己安装，或者使用附带的 Dockerfile，然后编辑 conf.yaml 来选择语言模型、语音识别引擎和声音。Linux 完全没有下载包，README 还提醒，要从另一台设备访问这个服务器，需要在反向代理后面配置好 https。","costNote":"采用 MIT 协议，但内置的 Live2D 示例角色不在此列。它完全可以用本地模型免费运行；它支持的云端选项（OpenAI、Claude、Azure Speech、ElevenLabs、Fish Audio、Groq）都按使用量收费。","notVerified":["在没有显卡的电脑上，语音对话是否真的能跟得上","下载完语音、音色和语言模型之后，总共会占用多少磁盘空间","安装好的桌面客户端是否能在各个系统上直接连上服务器，不用额外折腾","如果你选择云端服务，这些提供方会如何保存你的语音、摄像头画面和屏幕截图","列出的那么多语音识别和语音合成引擎，是否在 1.2.1 版本里都还能用"]}}},{"id":"R_kgDOKdDCcw","slug":"open-webui__open-webui","name":"Open WebUI","owner":"open-webui","fullName":"open-webui/open-webui","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"license_forbids_removing_branding","severity":"attention","evidence":"LICENSE 文件并不是普通的 BSD：第 4 条规定被许可人 \"are strictly prohibited from altering, removing, obscuring, or replacing any 'Open WebUI' branding\"，除非是 30 天内最多 \"fifty (50)\" 名用户的部署、取得书面许可，或持有企业许可证。GitHub 将该许可证显示为 NOASSERTION，pyproject.toml 将其归类为 \"License :: Other/Proprietary License\"。","source":"ai","title":"许可证禁止移除品牌标识"},{"code":"insecure_network_defaults_in_example","severity":"attention","evidence":".env.example 自带 `CORS_ALLOW_ORIGIN='*'` 和 `FORWARDED_ALLOW_IPS='*'`，文件自己的警告是：\"WARNING: * trusts forwarded headers from every connection. Use only behind a trusted proxy\"。如果你按示例设置原样把它放到公开的互联网上，这条警告就适用于你。","source":"ai","title":"示例中的网络默认设置过于宽松"},{"code":"plugins_install_their_own_code","severity":"attention","evidence":".env.example 描述了一个 \"Master switch for internal Tools/Functions and external OpenAPI/MCP/Open Terminal plugins\"，默认 `ENABLE_PLUGINS=true`，而 `ENABLE_TOOLS` 控制 \"workspace Tools and their dependency installation\"。因此，你从社区添加的工具可能会在你的电脑上引入并运行额外的代码。","source":"ai","title":"插件会安装自己的代码"},{"code":"optional_ai_terminal_access","severity":"attention","evidence":"README：Open Terminal \"Give your agents a terminal and filesystem to carry out multi-step tasks. Let them analyze data, run scripts, fix errors\"。这个配套组件是可选的，但一旦连接，AI 就可以在运行它的机器上执行命令。","source":"ai","title":"可选的 AI 终端访问"},{"code":"contributor_agreement_required","severity":"info","evidence":"LICENSE：\"By contributing to this project, you agree to the project's Contributor License Agreement\"，仓库中有 CONTRIBUTOR_LICENSE_AGREEMENT 文件，以及记录早期代码曾使用不同条款的 LICENSE_HISTORY。","source":"ai","title":"需要签署贡献者协议"},{"code":"telemetry_off_in_shipped_settings","severity":"info","evidence":".env.example 和 Dockerfile 都设置了 `SCARF_NO_ANALYTICS=true`、`DO_NOT_TRACK=true` 和 `ANONYMIZED_TELEMETRY=false`，并且 Dockerfile 会向内置向量数据库的遥测文件写入一个全零的标识。","source":"ai","title":"自带设置中已关闭遥测"}],"platforms":["windows","mac","linux"],"tools":["python","docker"],"buildTools":["python","node","docker"],"disk":{"minMB":250,"maxMB":950,"basis":"heuristic"},"apiKeys":["OpenAI"],"cost":{"model":"freemium"},"license":{"spdx":"BSD-3-Clause","category":"permissive","detectedFrom":"file"},"categories":["local_ai","ai_agents","productivity"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"typingmind","name":"TypingMind","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[],"stars":158318,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":631,"growthDays":7,"history":[{"date":"2026-10-26","stars":156857},{"date":"2026-10-27","stars":156976},{"date":"2026-10-28","stars":157085},{"date":"2026-10-29","stars":157179},{"date":"2026-10-30","stars":157298},{"date":"2026-10-31","stars":157407},{"date":"2026-11-01","stars":157488},{"date":"2026-11-02","stars":157584},{"date":"2026-11-03","stars":157687},{"date":"2026-11-04","stars":157781},{"date":"2026-11-05","stars":157862},{"date":"2026-11-06","stars":157944},{"date":"2026-11-07","stars":158026},{"date":"2026-11-08","stars":158125},{"date":"2026-11-09","stars":158227},{"date":"2026-11-10","stars":158318}],"firstPublishedAt":"2026-10-27T01:10:12.668Z","commitSha":"f7ac7ee10962ddec0017f5c29b6740d9b9c5faca","verifiedAt":"2026-10-30T01:41:18.277Z","lastCommitAt":"2026-11-08T23:48:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A chat page you run on your own computer for talking to AI models, either local ones or paid ones through an API key.","summary":"Open WebUI is a chat interface you install on your own computer or server and then open in a browser. It talks to models running locally through Ollama and to any OpenAI-compatible service, so your conversations and uploaded documents stay on the machine you run it on. Besides chat it has document search, web search, notes, a calendar, voice calls, image generation and user accounts with permissions. You install it with one command using Python (pip) or Docker, and you still need either a local model or an API key to get answers.","easeRationale":"There is no installer: on all three systems you install Python 3.12 and run `pip install open-webui` then `open-webui serve`, or use the one-line Docker command the README gives. After that you open http://localhost:8080 in a browser, create the first account and point it at Ollama or paste an API key.","costNote":"Free to install and run yourself. The project also sells an Enterprise Plan, and the license only lets you change or remove the \"Open WebUI\" branding below 50 users unless you buy an enterprise license. Any paid AI service you connect is billed separately.","notVerified":["Whether it actually installs and runs on each system, since Radar did not run it","Real disk use, which grows a lot once the speech and document-search models are downloaded on first use (the Dockerfile pulls Whisper and sentence-transformers models)","What a community Tool, Function or MCP server does once you install it, and what extra packages it pulls in","The real monthly cost, which depends on whatever AI service, web search or storage provider you connect","Whether your installation ends up reachable from the internet, which depends on how you set up Docker, ports and any proxy"]},"zh-Hans":{"summaryShort":"一个在自己电脑上运行的聊天页面,用来和AI模型对话,可以是本地模型,也可以通过API密钥使用付费模型。","summary":"Open WebUI 是一个安装在自己电脑或服务器上、然后在浏览器里打开的聊天界面。它可以连接通过 Ollama 在本地运行的模型,也可以连接任何兼容 OpenAI 的服务,因此你的对话和上传的文档都保留在运行它的这台机器上。除了聊天功能,它还有文档搜索、网页搜索、笔记、日历、语音通话、图像生成和带权限管理的用户账户。你可以用 Python(pip)或 Docker 的一条命令完成安装,但仍然需要本地模型或 API 密钥才能获得回答。","easeRationale":"没有安装程序:三个系统上都要先安装 Python 3.12,运行 `pip install open-webui`,再运行 `open-webui serve`,或者使用 README 给出的一行 Docker 命令。之后在浏览器里打开 http://localhost:8080,创建第一个账户,再连接 Ollama 或粘贴 API 密钥。","costNote":"自己安装和运行是免费的。该项目还出售企业版方案,许可证规定用户数低于50人时才能更改或移除\"Open WebUI\"品牌标识,否则需要购买企业许可证。你连接的任何付费AI服务都会单独计费。","notVerified":["它是否真的能在各个系统上安装并运行,因为 Radar 没有实际运行过它","真实的磁盘占用,首次使用时下载语音和文档搜索模型后会大幅增加(Dockerfile 会拉取 Whisper 和 sentence-transformers 模型)","安装后社区提供的 Tool、Function 或 MCP 服务器会做什么,以及它们会额外引入哪些软件包","真实的每月费用,这取决于你连接的AI服务、网页搜索或存储提供商","你的安装是否最终可以从互联网访问,这取决于你如何配置 Docker、端口和任何代理"]}}},{"id":"R_kgDOOYsS4Q","slug":"openai__codex","name":"Codex CLI","owner":"openai","fullName":"openai/codex","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道传给 shell：curl -fsSL https://chatgpt.com/codex/install.sh | sh","source":"check","title":"curl 管道到 shell"},{"code":"irm_pipe_iex","severity":"attention","evidence":"README.md：Windows 安装时直接在 PowerShell 中运行远程脚本，并关闭了执行策略：powershell -ExecutionPolicy ByPass -c \"irm https://chatgpt.com/codex/install.ps1 | iex\"","source":"ai","title":"irm 管道到 iex"},{"code":"downloads_outside_github","severity":"info","evidence":"README.md：\"The standalone installers download from `https://releases.openai.com/codex` by default and fall back to GitHub Releases\"，所以除非你把 CODEX_INSTALLER_USE_RELEASES_OPENAI_COM 设为 false，你拿到的文件并不是 GitHub 上发布的那些。","source":"ai","title":"下载来源不是 github"},{"code":"agent_edits_files_and_runs_commands_on_your_computer","severity":"attention","evidence":"该仓库把自己描述为 \"Lightweight coding agent that runs in your terminal\"，README 也说 Codex \"runs locally on your computer\"；这个代理会读取你的项目、修改文件并运行命令，所以一条错误的指令，或者藏在它读取的文件里的提示词，都可能改动或删除你的工作成果。","source":"ai","title":"代理会在你的电脑上编辑文件并运行命令"},{"code":"pre_1_0_with_very_fast_release_churn_and_a_big_issue_backlog","severity":"info","evidence":"创建于 2025-04-13，项目目前版本为 rust-v0.166.0，共有 1200 个 GitHub 发行版（最近三个中有两个是 2026-10-28 和 2026-10-29 的 alpha 构建），open issue 有 25,736 个。","source":"ai","title":"版本未到 1.0，发行更新很快，积压的 issue 很多"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":355,"maxMB":1064,"basis":"installer"},"apiKeys":["OpenAI API (optional instead of signing in with ChatGPT)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true},{"slug":"claude-code","name":"Claude Code","confirmed":true}],"declaredAlternatives":[],"stars":134758,"downloads":58226205,"downloadsWeek":2323380,"downloadsDays":7,"growth":1748,"growthDays":7,"history":[{"date":"2026-10-26","stars":131030},{"date":"2026-10-27","stars":131276},{"date":"2026-10-28","stars":131445},{"date":"2026-10-29","stars":131570},{"date":"2026-10-30","stars":131971},{"date":"2026-10-31","stars":132500},{"date":"2026-11-01","stars":132753},{"date":"2026-11-02","stars":132887},{"date":"2026-11-03","stars":133010},{"date":"2026-11-04","stars":133183},{"date":"2026-11-05","stars":133337},{"date":"2026-11-06","stars":133477},{"date":"2026-11-07","stars":133585},{"date":"2026-11-08","stars":133792},{"date":"2026-11-09","stars":134336},{"date":"2026-11-10","stars":134758}],"firstPublishedAt":"2026-10-27T01:10:12.669Z","commitSha":"67f0994b532b8740f14b5b738af7b7b2a915b1d6","verifiedAt":"2026-10-31T01:42:28.519Z","lastCommitAt":"2026-11-10T22:32:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An AI helper from OpenAI that reads, writes and runs code for you from a terminal window on your own computer.","summary":"Codex CLI is a coding assistant made by OpenAI. You type requests in a terminal, and it reads your project files, edits code and runs commands on your computer. There are also editor extensions (VS Code, Cursor, Windsurf) and a desktop app you start with `codex app`. It is aimed at people who write software. To use it you sign in with a paid ChatGPT plan (Plus, Pro, Business, Edu or Enterprise) or set up an OpenAI API key. It installs on Windows with one PowerShell command, on Mac or Linux with one shell command, or through npm or Homebrew.","easeRationale":"B on all three: one command installs a ready binary (a PowerShell line on Windows, a shell line on Mac and Linux) and then you sign in with ChatGPT, which is the extra step. You can instead download the binary straight from the releases, where the files are named by platform (for example `codex-x86_64-pc-windows-msvc.exe`); Node.js is only needed if you pick the npm route.","costNote":"The program is free (Apache-2.0), but it needs a paid ChatGPT plan (Plus, Pro, Business, Edu or Enterprise) or a paid OpenAI API key.","notVerified":["What the install scripts at chatgpt.com actually download and change on the computer at the time you run them.","How much use each ChatGPT plan allows before hitting limits, and the real cost when using an API key.","Which commands and file changes the agent makes on its own, and how well its sandbox limits them on each system.","Real disk use after install (the checks estimate 355 to 1064 MB from release files).","Whether the desktop app (`codex app`) and the editor extensions behave like the terminal version."]},"zh-Hans":{"summaryShort":"一个来自 OpenAI 的 AI 助手，在你自己电脑的终端窗口里帮你读取、编写并运行代码。","summary":"Codex CLI 是 OpenAI 制作的编程助手。你在终端里输入请求，它会读取你的项目文件、修改代码并在你的电脑上运行命令。还有编辑器扩展（VS Code、Cursor、Windsurf）和一个用 `codex app` 启动的桌面应用。它面向会写程序的人。使用时你需要用付费的 ChatGPT 套餐（Plus、Pro、Business、Edu 或 Enterprise）登录，或者配置一个 OpenAI API 密钥。在 Windows 上用一条 PowerShell 命令安装，在 Mac 或 Linux 上用一条 shell 命令安装，也可以通过 npm 或 Homebrew 安装。","easeRationale":"三个系统都是 B 级：一条命令就能安装好一个现成的二进制文件（Windows 上是一行 PowerShell，Mac 和 Linux 上是一行 shell），然后用 ChatGPT 登录，这是额外的一步。你也可以直接从 releases 页面下载二进制文件，文件按平台命名（例如 `codex-x86_64-pc-windows-msvc.exe`）；只有选择 npm 方式时才需要 Node.js。","costNote":"程序本身是免费的（Apache-2.0），但需要付费的 ChatGPT 套餐（Plus、Pro、Business、Edu 或 Enterprise）或付费的 OpenAI API 密钥。","notVerified":["chatgpt.com 上的安装脚本在你运行时究竟会下载什么、对电脑做了哪些改动。","每个 ChatGPT 套餐在达到限额前允许多少用量，以及使用 API 密钥时的实际花费。","代理自己执行哪些命令、改动哪些文件，以及它的沙箱在每个系统上能限制到什么程度。","安装后的实际磁盘占用（检测估算为 355 到 1064 MB，基于发行文件）。","桌面应用（`codex app`）和编辑器扩展是否和终端版本表现一致。"]}}},{"id":"R_kgDOVRkL5Q","slug":"openai__php2shell-checker","name":"php2shell Checker","owner":"openai","fullName":"openai/php2shell-checker","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_skills"],"alternativeTo":[],"declaredAlternatives":[],"stars":1469,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":9,"growthDays":7,"history":[{"date":"2026-10-27","stars":1438},{"date":"2026-10-28","stars":1443},{"date":"2026-10-29","stars":1448},{"date":"2026-10-30","stars":1453},{"date":"2026-10-31","stars":1456},{"date":"2026-11-01","stars":1459},{"date":"2026-11-02","stars":1458},{"date":"2026-11-03","stars":1460},{"date":"2026-11-04","stars":1462},{"date":"2026-11-05","stars":1464},{"date":"2026-11-06","stars":1465},{"date":"2026-11-07","stars":1464},{"date":"2026-11-08","stars":1467},{"date":"2026-11-09","stars":1468},{"date":"2026-11-10","stars":1469}],"firstPublishedAt":"2026-10-27T18:40:27.992Z","commitSha":"ffcd3fc82e5b404f2a04a95b2805ca7639d530fa","verifiedAt":"2026-10-27T18:32:07.724Z","lastCommitAt":"2026-10-22T22:55:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Checks whether a website or local PHP install is running a version affected by a known PHP vulnerability.","summary":"This is a checking tool, not an exploit: it reads the PHP version a website or local install reports (via HTTP headers or a local check) and compares it to the versions fixed by an October 2026 advisory. It warns 'potentially vulnerable' for affected ranges and EOL branches, or 'unknown' when the version is unclear; the README states it sends no exploit probes. Ships as a Node.js script plus two packaged skills for the Codex AI assistant. For developers and site admins checking their own PHP exposure.","easeRationale":"There is no installer: running it needs Node.js installed first, then typing `node phpcheck.js <url>` in a terminal, or asking Codex to install the packaged skill.","notVerified":["Whether version detection via HTTP headers stays accurate on sites that hide or spoof their Server/X-Powered-By headers.","Whether the listed upstream fixed versions stay accurate as further patches ship.","What exactly the Codex skill installer executes when asked to install the skills from this repository."]},"zh-Hans":{"summaryShort":"检查某个网站或本地 PHP 安装所用的版本是否受某个已知 PHP 漏洞影响。","summary":"这是一个检测工具,不是利用工具:它读取网站或本地 PHP 安装报告的版本号(通过 HTTP 响应头或本地检查),再和 2026 年 10 月某公告中修复的版本对比。对受影响的版本范围和已停止支持(EOL)的分支,它会提示“可能存在漏洞”;版本不清楚时提示“未知”。README 声明它不会发出任何利用探测。它以一个 Node.js 脚本的形式提供,外加两个打包给 Codex AI 助手用的 skill。面向想核查自己 PHP 暴露面的开发者和网站管理员。","easeRationale":"没有安装程序:运行前需要先装好 Node.js,然后在终端里输入 `node phpcheck.js <url>`,或者让 Codex 安装打包好的 skill。","notVerified":["在隐藏或伪造 Server/X-Powered-By 响应头的网站上,通过 HTTP 响应头做的版本检测是否仍然准确。","随着后续补丁发布,列出的上游修复版本是否仍然准确。","被要求安装本仓库的 skill 时,Codex 的 skill 安装器到底执行了什么操作。"]}}},{"id":"R_kgDOIAstBQ","slug":"openai__whisper","name":"Whisper","owner":"openai","fullName":"openai/whisper","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"low","signals":[{"code":"no_release_in_over_a_year","severity":"info","evidence":"GitHub 上的最新发布版本是 v20250625（2025-06-26）；最近一次提交是在 2026-08-31。","source":"ai","title":"一年多没有新版本"},{"code":"may_need_rust_to_install","severity":"info","evidence":"README.md：\"You may need `rust` installed as well, in case tiktoken does not provide a pre-built wheel for your platform.\"","source":"ai","title":"安装时可能需要 Rust"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","productivity"],"alternativeTo":[{"slug":"otter-ai","name":"Otter.ai","confirmed":true},{"slug":"google-cloud-speech-to-text","name":"Cloud Speech-to-Text","confirmed":true}],"declaredAlternatives":[],"stars":112371,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":479,"growthDays":7,"history":[{"date":"2026-10-26","stars":111344},{"date":"2026-10-27","stars":111415},{"date":"2026-10-28","stars":111486},{"date":"2026-10-29","stars":111546},{"date":"2026-10-30","stars":111607},{"date":"2026-10-31","stars":111654},{"date":"2026-11-01","stars":111738},{"date":"2026-11-02","stars":111808},{"date":"2026-11-03","stars":111892},{"date":"2026-11-04","stars":111953},{"date":"2026-11-05","stars":112033},{"date":"2026-11-06","stars":112102},{"date":"2026-11-07","stars":112184},{"date":"2026-11-08","stars":112248},{"date":"2026-11-09","stars":112302},{"date":"2026-11-10","stars":112371}],"firstPublishedAt":"2026-10-27T01:10:12.670Z","commitSha":"86098128c0b4f24f0e2aa2994de830614b474227","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-08-31T17:19:19Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Turn speech in audio or video files into text, in many languages, or translate it into English, on your own computer.","summary":"Whisper is OpenAI's speech recognition model. It turns spoken audio into text in many languages, can detect which language is spoken, and can translate speech into English. You run it by typing a command in a terminal or from Python code, and it runs on your own computer once a model is downloaded. You need Python and ffmpeg; a graphics card makes it much faster, especially for the larger models, which need about 5–10 GB of video memory. Accuracy varies a lot from one language to another.","easeRationale":"There is no installer: you install Python and ffmpeg (the README suggests Chocolatey or Scoop on Windows), run pip install -U openai-whisper, and then use it by typing commands.","costNote":"Free. The README says both the code and the model weights are under the MIT license.","notVerified":["How large the model download is and where it is stored on first use","How slow it is on a computer without a graphics card","How accurate it is for your language and your recordings (the README shows large differences between languages)","Whether pip installs a graphics-card-enabled version of PyTorch on Windows without extra steps"]},"zh-Hans":{"summaryShort":"在你自己的电脑上，把音频或视频文件里的语音转成文字，支持多种语言，也可翻译成英文。","summary":"Whisper 是 OpenAI 的语音识别模型。它能把多种语言的口语音频转成文字，能判断说的是哪种语言，还能把语音翻译成英文。你通过在终端输入命令或用 Python 代码来运行它，下载模型后就在你自己的电脑上运行。你需要 Python 和 ffmpeg；有显卡会快很多，尤其是较大的模型，它们需要约 5–10 GB 的显存。不同语言之间的准确度差别很大。","easeRationale":"没有安装程序：你要安装 Python 和 ffmpeg（README 建议在 Windows 上用 Chocolatey 或 Scoop），运行 pip install -U openai-whisper，然后通过输入命令来使用。","costNote":"免费。README 说代码和模型权重都采用 MIT 许可证。","notVerified":["模型下载有多大，以及首次使用时保存在哪里","在没有显卡的电脑上有多慢","对你的语言和你的录音准确度如何（README 显示不同语言之间差异很大）","在 Windows 上 pip 是否无需额外步骤就能安装支持显卡的 PyTorch 版本"]}}},{"id":"R_kgDOVZP6Hg","slug":"openbmb__minicpm-ocr","name":"MiniCPM-OCR","owner":"OpenBMB","fullName":"OpenBMB/MiniCPM-OCR","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"linux":"D","web":"B"},"risk":"attention","signals":[{"code":"skill_tells_agent_to_execute_a_web_page","severity":"attention","evidence":"README 的 Skill 部分：交给编程智能体的指令是 \"Read and execute https://github.com/OpenBMB/MiniCPM-OCR/blob/main/docs/SKILL_en.md\"。随后智能体会根据一个内容随时可能改变的网页，在你的电脑上运行配置命令；README 也提供了本地副本 ./docs/SKILL_en.md，你可以先阅读它。","source":"ai","title":"技能（skill）让智能体去执行一个网页"},{"code":"demo_server_without_encryption","severity":"attention","evidence":"README 中的 \"Basic Demo (CN)\" 链接是一个不带 HTTPS 的裸地址：http://82.157.64.212:10111/playground/document-parse。你上传到那里的文档会以未加密的方式传输，而且没有说明这台机器的运营者是谁。","source":"ai","title":"演示服务器没有加密"},{"code":"documents_leave_your_computer_on_the_hosted_paths","severity":"info","evidence":"除了本地运行，README 还提供了演示和位于 platform.modelbest.ai / platform.modelbest.cn 的 \"Free API\"。使用它们意味着要把文档上传到该公司的服务器；仓库没有说明之后会如何处理这些文档。","source":"ai","title":"使用托管方式时，文档会离开你的电脑"},{"code":"benchmark_scores_are_the_makers_own","severity":"info","evidence":"README 声称 \"a SOTA score of 93.68 on Real5-OmniDocBench\"，吞吐量是 PaddleOCR-VL-1.6 的 \"1.49x\"、MinerU2.5 Pro 的 \"3.62x\"。这些都是作者自己的测量结果；仓库附有复现指南（docs/benchmark_reproduction_en.md），而雷达图自己的说明写着 \"some values remain to be confirmed\"。","source":"ai","title":"评测成绩是开发者自己给出的"},{"code":"brand_new_repository","severity":"info","evidence":"项目信息：创建于 2026-10-27，最近一次推送是 2026-10-29，只有一个发布版本（sdk-v1.0.2），没有各平台的二进制文件。README 把 MiniCPM-OCR 1.0 称为 \"the first OCR-specific model in the MiniCPM-V/o family\"。","source":"ai","title":"全新的仓库"}],"platforms":["linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["office","image","development"],"alternativeTo":[{"slug":"abbyy-finereader","name":"FineReader","confirmed":true},{"slug":"mathpix","name":"Mathpix","confirmed":true},{"slug":"acrobat","name":"Acrobat","confirmed":true}],"declaredAlternatives":[],"stars":872,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":80,"growthDays":7,"history":[{"date":"2026-10-30","stars":676},{"date":"2026-10-31","stars":721},{"date":"2026-11-01","stars":746},{"date":"2026-11-02","stars":770},{"date":"2026-11-03","stars":792},{"date":"2026-11-04","stars":811},{"date":"2026-11-05","stars":823},{"date":"2026-11-06","stars":838},{"date":"2026-11-07","stars":846},{"date":"2026-11-08","stars":851},{"date":"2026-11-09","stars":863},{"date":"2026-11-10","stars":872}],"firstPublishedAt":"2026-10-30T01:57:54.407Z","commitSha":"1269e4f81fb82e5a301ae42ee8548b2f3336beec","verifiedAt":"2026-10-30T01:46:15.331Z","lastCommitAt":"2026-11-09T15:09:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns photos, scans and PDFs into text, tables and formulas you can edit, and scores which parts may be wrong.","summary":"Two small AI models from the MiniCPM family, released with their weights. MiniCPM-OCR 1.0 (1.3B) reads Chinese and English documents and PDFs into Markdown or JSON, including tables, formulas, charts, stamps and handwriting, and copes with scans, photos, skewed and curved pages. MiniCPM-OCR Jev (2B) looks at a result and gives each region a score from 0 to 1, so you only check the parts that are likely wrong. It is for developers: a pip package, a free hosted API, and local running with vLLM, SGLang, llama.cpp or Ollama. There are demo pages to try first.","easeRationale":"The easiest path is the web: the maker's demo pages (a console on platform.modelbest.ai, which needs an account, and a Hugging Face Space that does not) let you drop a document in and read the result with nothing installed. Running it yourself is a Linux job: Python 3.11-3.13, an NVIDIA card with CUDA, several GB of model weights, and an installer that fetches pinned vLLM and transformers sources and compiles them (pyproject.toml, scripts/install.sh).","costNote":"Code, documentation and both model weights are Apache 2.0. The maker also offers a free hosted API and demo pages; their limits are not stated in the repository.","notVerified":["Whether the quality holds on your own documents, languages other than Chinese and English, or your handwriting","How much graphics memory each model and backend really needs","The real disk use of the weights and the environments (the 950 MB to 4.5 GB estimate is a guess)","Whether the llama.cpp and Ollama route works on Windows and Mac, which the repository does not document","The limits of the free API and what the company does with the documents sent to it","What score threshold to use with Jev, which the README says must be checked on your own batch first"]},"zh-Hans":{"summaryShort":"把照片、扫描件和PDF转成可编辑的文字、表格和公式,并标出哪些部分可能有误。","summary":"MiniCPM 系列中的两个小型AI模型,随权重一起发布。MiniCPM-OCR 1.0(1.3B)可以把中英文文档和PDF转成Markdown或JSON,包括表格、公式、图表、印章和手写内容,并能处理扫描件、照片以及倾斜或弯曲的页面。MiniCPM-OCR Jev(2B)会检查结果,给每个区域打出0到1的分数,这样你只需要检查可能出错的部分。它面向开发者:提供pip包、免费的托管API,也可以用vLLM、SGLang、llama.cpp或Ollama在本地运行。还有演示页面可以先试用。","easeRationale":"最简单的方式是通过网页:开发者提供的演示页面(platform.modelbest.ai上的控制台需要账号,Hugging Face Space则不需要)可以直接拖入文档查看结果,无需安装任何东西。自己运行则是Linux上的工作:需要Python 3.11-3.13、带CUDA的NVIDIA显卡、好几GB的模型权重,以及一个会拉取固定版本的vLLM和transformers源码并编译的安装脚本(pyproject.toml、scripts/install.sh)。","costNote":"代码、文档以及两个模型的权重都采用Apache 2.0许可。开发者还提供免费的托管API和演示页面;其限制没有在仓库中说明。","notVerified":["在你自己的文档、中英文以外的语言或你的手写字迹上,效果是否依然可靠","每个模型和后端实际需要多少显存","权重和运行环境的真实磁盘占用(950MB到4.5GB只是估算)","llama.cpp和Ollama这条路线是否能在Windows和Mac上运行,仓库中没有说明","免费API的限制,以及该公司如何处理发送给它的文档","使用Jev时该采用什么分数阈值,README说这需要先在自己的数据批次上验证"]}}},{"id":"R_kgDOPwqvLQ","slug":"openbmb__voxcpm","name":"VoxCPM","owner":"OpenBMB","fullName":"OpenBMB/VoxCPM","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"voice_cloning_can_imitate_real_people","severity":"attention","evidence":"README 有一个 \"Risks and Limitations\" 部分写道：\"Potential for Misuse: VoxCPM's voice cloning can generate highly realistic synthetic speech. It is strictly forbidden to use VoxCPM for impersonation, fraud, or disinformation. We strongly recommend clearly marking any AI-generated content.\" 同一份 README 还展示了如何从一段简短录音复刻声音，例如命令 \"voxcpm clone --text \\\"This is a voice cloning demo.\\\" --reference-audio path/to/voice.wav --output out.wav\"。许多国家都有法律规定，未经本人许可不得使用其声音。","source":"ai","title":"声音克隆可以模仿真实的人"},{"code":"model_weights_downloaded_on_first_use","severity":"info","evidence":"README 的第一个示例调用了 VoxCPM.from_pretrained(\"openbmb/VoxCPM2\")，在你第一次运行时会从 Hugging Face（或者通过 snapshot_download(\"OpenBMB/VoxCPM2\") 从 ModelScope）拉取这个 20 亿参数的权重文件，所以下载是在仓库之外发生的。","source":"ai","title":"模型权重在首次使用时才下载"},{"code":"package_marked_alpha_by_the_authors","severity":"info","evidence":"pyproject.toml 为已发布的 \"voxcpm\" 包列出了分类标签 \"Development Status :: 3 - Alpha\"。","source":"ai","title":"作者自己把这个包标记为 Alpha 阶段"},{"code":"accuracy_varies_a_lot_by_language","severity":"info","evidence":"在 README 的 Minimax-MLS-test WER 表格中，VoxCPM2 在英语上的得分是 2.289，荷兰语是 0.913，但粤语是 38.584，捷克语是 24.132，罗马尼亚语是 21.577，印地语是 19.699，说明在列出的 30 种语言之间质量并不均衡。README 自己的说明还补充道 \"Voice Design and Controllable Voice Cloning results can vary between runs\"。","source":"ai","title":"不同语言之间的准确率差异很大"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"speechify","name":"Speechify","confirmed":true}],"declaredAlternatives":[],"stars":42206,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":1201,"growthDays":7,"history":[{"date":"2026-11-05","stars":41098},{"date":"2026-11-06","stars":41124},{"date":"2026-11-07","stars":41306},{"date":"2026-11-08","stars":41729},{"date":"2026-11-09","stars":42100},{"date":"2026-11-10","stars":42206}],"firstPublishedAt":"2026-11-08T23:13:56.618Z","commitSha":"4b8dd4440d33396b3f429201f8af0529b4d2c601","verifiedAt":"2026-11-08T23:09:48.670Z","lastCommitAt":"2026-11-09T07:36:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turn text into very realistic speech on your own computer, in 30 languages, and copy a voice from a short recording.","summary":"VoxCPM2 turns written text into speech on your own machine. It can invent a voice from a written description, or copy a voice from a short recording. The README lists 30 languages and 48kHz audio output. It is for people comfortable with a terminal: you install it with \"pip install voxcpm\", and the README asks for Python 3.10 or newer, PyTorch 2.5.0 or newer and CUDA 12.0 or newer. The model table lists about 8 GB of VRAM for VoxCPM2. There is also a browser page you start yourself with \"python app.py\". Code and model weights are under Apache-2.0.","easeRationale":"The best grade is D and it is the same on Windows, macOS and Linux: the 14 GitHub releases carry no ready-to-run file, so you install Python yourself, run \"pip install voxcpm\" and then type commands, and the README's model table lists about 8 GB of VRAM for VoxCPM2. macOS is not easier: the README only says that \"python app.py --device auto\" uses MPS on Apple Silicon, so the same install steps apply.","costNote":"License is Apache-2.0 and the README says the weights and code are \"free for commercial use\". No price or paid plan is mentioned.","notVerified":["Whether the install really works on Windows, macOS and Linux, since no release carries a ready-to-run file.","Whether it runs usefully without an NVIDIA card: the README asks for CUDA 12.0 or newer, while \"python app.py --device\" also accepts cpu and mps.","The real disk space used after the VoxCPM2 weights are downloaded (the checks estimate 950-4450 MB by heuristic).","Whether the speed figures in the README (RTF ~0.30 on an RTX 4090, ~0.13 with Nano-vLLM) and the benchmark tables hold on your own machine.","Whether about 8 GB of VRAM is enough in practice for long texts or several requests at once."]},"zh-Hans":{"summaryShort":"在你自己的电脑上把文本转换成非常逼真的语音，支持 30 种语言，还能从一段简短录音中复刻声音。","summary":"VoxCPM2 可以在你自己的电脑上把书面文本转换成语音。它可以根据一段文字描述自己创造一个声音，也可以从一段简短录音中复刻某个声音。README 列出了 30 种语言和 48kHz 的音频输出。它适合熟悉终端操作的人：你可以用 \"pip install voxcpm\" 安装，README 要求 Python 3.10 及以上、PyTorch 2.5.0 及以上、CUDA 12.0 及以上。模型表里 VoxCPM2 大约需要 8 GB 显存。还有一个浏览器页面，你可以自己用 \"python app.py\" 启动。代码和模型权重都采用 Apache-2.0 协议。","easeRationale":"最好的评级也只是 D，Windows、macOS 和 Linux 都一样：GitHub 上的 14 个发行版都没有可直接运行的文件，你需要自己安装 Python，运行 \"pip install voxcpm\"，然后输入命令；README 的模型表里 VoxCPM2 大约需要 8 GB 显存。macOS 也不会更简单：README 只是说 \"python app.py --device auto\" 在 Apple Silicon 上会使用 MPS，安装步骤仍然一样。","costNote":"协议是 Apache-2.0，README 说模型权重和代码 \"free for commercial use\"。没有提到任何价格或付费方案。","notVerified":["安装是否真的能在 Windows、macOS 和 Linux 上正常使用，因为没有任何一个发行版带有可直接运行的文件。","没有 NVIDIA 显卡是否也能正常使用：README 要求 CUDA 12.0 及以上，不过 \"python app.py --device\" 也接受 cpu 和 mps。","下载 VoxCPM2 权重之后实际占用的磁盘空间（检查结果凭经验估计在 950 到 4450 MB 之间）。","README 里的速度数据（在 RTX 4090 上 RTF 约 0.30，用 Nano-vLLM 约 0.13）和基准测试表格在你自己的电脑上是否成立。","大约 8 GB 显存在实际使用中，对长文本或同时多个请求是否足够。"]}}},{"id":"R_kgDOQb6kRw","slug":"openclaw__openclaw","name":"OpenClaw","owner":"openclaw","fullName":"openclaw/openclaw","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"339 天内获得 393387 个星标（约每天 1160 个）","source":"check","title":"星标增长速度异常"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖项中存在已知漏洞（OSV）：fast-uri：GHSA-333x-w4r6-6crp、GHSA-w74x-v6f9-j3xg、GHSA-wjr9-j795-jxcr","source":"check","title":"存在已知漏洞"},{"code":"agent_tools_run_on_host","severity":"attention","evidence":"README 的 Security 部分：\"Tools run on the host for the main session unless you configure sandboxing.\" 同一部分还说要把收到的消息当作不可信的输入，因此，连接的聊天中收到的一条消息，就可能左右一个正在你电脑上运行命令的助手。","source":"ai","title":"智能体的工具直接在你的电脑上运行"},{"code":"gateway_listens_on_the_local_network","severity":"attention","evidence":"docker-compose.yml 使用 `--bind ${OPENCLAW_GATEWAY_BIND:-lan}` 启动网关，并把 18789、18790 和 3978 端口发布到主机上。.env 示例说明，一旦网关绑定到回环地址以外，就需要令牌，如果留空则会自动生成，所以这个服务可以从你的网络访问，令牌是唯一的屏障。","source":"ai","title":"网关在本地网络中监听"},{"code":"remote_script_install","severity":"info","evidence":"README 的 Install：`curl -fsSL https://openclaw.ai/install.sh | bash` 和 `iwr -useb https://openclaw.ai/install.ps1 | iex` 会从项目网站下载脚本并立即运行；npm 方式使用 `npm install -g openclaw@latest --allow-scripts=openclaw`，这允许软件包自带的安装脚本运行。","source":"ai","title":"通过远程脚本安装"},{"code":"web_session_cookie_passthrough","severity":"info","evidence":"docker-compose.yml 把 CLAUDE_AI_SESSION_KEY、CLAUDE_WEB_SESSION_KEY 和 CLAUDE_WEB_COOKIE 传入两个容器，这表明使用的是聊天网站的登录会话，而不是 API 密钥。这样做是否符合服务提供方的条款，由该服务提供方决定。","source":"ai","title":"网页登录会话的 Cookie 被传入容器"},{"code":"daily_version_check_by_default","severity":"info","evidence":"README：\"by default OpenClaw itself phones home for nothing but a daily version check, anonymous feature statistics are opt-in, and `update.checkOnStart: false` disables both\"。compose 文件还接入了 OpenTelemetry 的追踪、指标和日志导出器，在你设置目标地址之前它们都保持关闭。","source":"ai","title":"默认每天检查一次版本"},{"code":"very_fast_moving_project","severity":"info","evidence":"项目信息：共 276 个发布版本，最新一个就在分析当天，有 7875 个未解决的 issue，并且 v2026.10.10-beta.4 这样的预发布版本与稳定版本一起发布。请预期会有频繁的更新，以及更新之间的变化。","source":"ai","title":"项目变化非常快"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","docker","ffmpeg"],"disk":{"minMB":123,"maxMB":368,"basis":"installer"},"apiKeys":["AI model provider (OpenAI, Anthropic, Gemini, OpenRouter or similar)","Gateway token (OPENCLAW_GATEWAY_TOKEN), generated on first start or set by you","Chat app bot tokens (Telegram, Discord, Slack, Mattermost, Twitch)","Optional tools and voices (Brave Search, Perplexity, Firecrawl, ElevenLabs, Deepgram, Inworld)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["local_ai","ai_agents","automation"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[],"stars":394300,"downloads":2740369,"downloadsWeek":70066,"downloadsDays":7,"growth":436,"growthDays":7,"history":[{"date":"2026-10-25","stars":393059},{"date":"2026-10-26","stars":393161},{"date":"2026-10-27","stars":393221},{"date":"2026-10-28","stars":393293},{"date":"2026-10-29","stars":393383},{"date":"2026-10-30","stars":393480},{"date":"2026-10-31","stars":393560},{"date":"2026-11-01","stars":393645},{"date":"2026-11-02","stars":393754},{"date":"2026-11-03","stars":393864},{"date":"2026-11-04","stars":393916},{"date":"2026-11-05","stars":393983},{"date":"2026-11-06","stars":394060},{"date":"2026-11-07","stars":394138},{"date":"2026-11-08","stars":394204},{"date":"2026-11-09","stars":394247},{"date":"2026-11-10","stars":394300}],"firstPublishedAt":"2026-10-25T23:19:22.162Z","commitSha":"0b81496b64abb2f5de689b9b20f207d96c3ea27e","verifiedAt":"2026-10-30T01:48:24.669Z","lastCommitAt":"2026-11-10T22:59:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Run your own AI assistant on your computer and talk to it from WhatsApp, Telegram, Slack, Discord and other apps.","summary":"OpenClaw is a personal AI assistant that runs on your own computer. A background service (the Gateway) connects the AI model of your choice to chat apps such as WhatsApp, Telegram, Slack, Discord and iMessage, and the assistant can use tools, skills and plugins to do things for you. There are companion apps for Mac, Linux, Windows, iOS and Android, and the same setup can serve a whole team. You install it with one command (the installer brings its own Node.js), finish an onboarding wizard and supply a model key or a local model. Settings, memory and keys stay on your machine.","easeRationale":"All three systems are the same kind of work and land on C: one install command (`curl -fsSL https://openclaw.ai/install.sh | bash`, or `iwr -useb https://openclaw.ai/install.ps1 | iex` on Windows), which brings its own Node.js, then an onboarding wizard, a model provider key and terminal commands such as `openclaw gateway status`. Mac and Linux also get a ready companion app in the latest release (a .dmg of 226-334 MB, an AppImage or a .deb), but that app still needs the Gateway set up first.","costNote":"MIT licence; the README says the project has no paid tier, hosted service or token and is run by a non-profit foundation. You pay the AI provider you connect, unless you run a local model.","notVerified":["What the install.sh and install.ps1 scripts actually do, since they are fetched from openclaw.ai at install time","The real monthly cost of the model provider, which depends on how much the assistant is used","What plugins and skills from ClawHub can do on your computer once installed","Whether the pre-releases (for example v2026.10.10-beta.4) are stable enough for daily use","How well it works on Windows compared with Mac and Linux, which are the platforms with companion app files in the latest release"]},"zh-Hans":{"summaryShort":"在自己电脑上运行专属AI助手,通过WhatsApp、Telegram、Slack、Discord等应用和它对话。","summary":"OpenClaw 是一个在你自己电脑上运行的个人AI助手。一个后台服务(Gateway)把你选择的AI模型连接到WhatsApp、Telegram、Slack、Discord和iMessage等聊天应用,助手还可以使用工具、技能和插件来帮你做事。它为Mac、Linux、Windows、iOS和Android提供配套应用,同一套安装也可以供整个团队使用。安装只需一条命令(安装程序会自带Node.js),完成引导向导后提供模型密钥或使用本地模型即可。设置、记忆和密钥都保存在你自己的机器上。","easeRationale":"三个系统所需的工作量相当,评级都是C:一条安装命令(`curl -fsSL https://openclaw.ai/install.sh | bash`,Windows上则是`iwr -useb https://openclaw.ai/install.ps1 | iex`)会自带Node.js,接着是引导向导、模型提供商密钥,以及类似`openclaw gateway status`的终端命令。Mac和Linux在最新版本中还有现成的配套应用(226-334 MB的.dmg、AppImage或.deb),但这个应用仍需要先配置好Gateway。","costNote":"MIT许可;README说该项目没有付费档位、托管服务或代币,由一个非营利基金会运营。除非使用本地模型,否则需要向你连接的AI提供商付费。","notVerified":["install.sh和install.ps1脚本实际做了什么,因为它们是在安装时从openclaw.ai下载的","模型提供商的真实月费,这取决于助手的使用量","来自ClawHub的插件和技能安装后能在你的电脑上做什么","预发布版本(例如v2026.10.10-beta.4)是否足够稳定,能用于日常使用","它在Windows上的表现与Mac和Linux相比如何,后两者是最新版本中有配套应用文件的平台"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMTg1Mjc5","slug":"openemu__openemu","name":"OpenEmu","owner":"OpenEmu","fullName":"OpenEmu/OpenEmu","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"mac":"A"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2023-12-30（距今 1044 天）","source":"check","title":"发布版本已过时"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到许可证","source":"check","title":"没有许可证"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.3/10","source":"check","title":"scorecard 评分偏低"},{"code":"you_must_supply_your_own_game_files","severity":"attention","evidence":"这个仓库提供的是模拟引擎，不是游戏：README 列出了 Atari 2600、NES、Game Boy、任天堂 64、任天堂 DS、GameCube、世嘉 Genesis、世嘉 Saturn、索尼 PlayStation 和索尼 PSP 的引擎，顶层文件夹就是这些引擎（Stella、Nestopia、Mednafen、Mupen64Plus、DeSmuME、SNES9x）。里面不包含任何游戏文件，需要用户自己提供游戏机游戏，而且各国对复制游戏的规定也不一样。","source":"ai","title":"需要你自备游戏文件"},{"code":"download_is_not_in_the_github_releases","severity":"attention","evidence":"项目信息：GitHub 上共有 27 个发布版本，最新的三个（2023-12-30 的 v2.4.1 和 v2.4，2021-01-09 的 v2.3.3）都被记录为 \"no platform binaries\"。软件包里唯一的下载地址是主页 https://openemu.org，所以你安装的文件并非来自这个仓库，也无法与这里的代码进行比对。","source":"ai","title":"下载文件不在github发布页里"},{"code":"updates_itself_in_the_background","severity":"info","evidence":"README：\"One third-party library example is Sparkle, which is used for auto-updating.\" 这款应用可以自行从项目控制的服务器获取并安装新版本。","source":"ai","title":"会在后台自动更新自己"},{"code":"built_from_dozens_of_outside_emulators","severity":"info","evidence":"这个仓库里有一个 .gitmodules 文件，以及 25 个以他人模拟器命名的顶层文件夹（4DO、Atari800、BSNES、Bliss、CrabEmu、DeSmuME、FCEU、Gambatte、GenesisPlus、Mednafen、Mupen64Plus、Nestopia、Reicast、SNES9x、mGBA 等等），README 里每一个都链接到各自的项目。你运行的代码来自许多不同的团队，各有各的许可证和修复节奏。","source":"ai","title":"由几十个外部模拟器构建而成"},{"code":"quiet_for_about_a_year","severity":"info","evidence":"最近一次提交在 2025-10-22（本次检查前 382 天），最新发布在 2023-12-30，有 434 个未关闭的 issue，对应 17866 个星标。这个项目已经 15 年了（创建于 2010-12-20），并未被归档。","source":"ai","title":"将近一年没有动静"}],"platforms":["mac"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":17866,"downloads":17543749,"downloadsWeek":5191,"downloadsDays":3,"growth":4,"growthDays":6,"history":[{"date":"2026-11-07","stars":17867},{"date":"2026-11-08","stars":17866},{"date":"2026-11-09","stars":17866},{"date":"2026-11-10","stars":17866}],"firstPublishedAt":"2026-11-08T23:08:23.100Z","commitSha":"1d205104640d8410659d321809889cbfd06b99a9","verifiedAt":"2026-11-08T22:43:43.578Z","lastCommitAt":"2025-10-22T21:39:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Mac app that plays old console games — NES, Game Boy, Nintendo 64, Sega, PlayStation and more — in one game library.","summary":"OpenEmu is a macOS program that plays games from more than 30 old consoles — Atari, NES, Game Boy, Nintendo 64, Nintendo DS, Sega, PlayStation, PSP and others — inside one familiar Mac window, by loading a different emulation engine for each console as a plugin. It needs macOS Mojave 10.14.4 or newer. The games themselves are not included: you supply your own game files. The GitHub releases list no download file, so the app comes from openemu.org. The last release is v2.4.1 (2023-12-30) and the last commit here is from 2025-10-22.","easeRationale":"macOS is the only platform and sits at grade A: this is a finished Mac app you open after downloading, and the README says it \"leverages modern macOS technologies\" and uses \"Sparkle, which is used for auto-updating\". The one catch is where the file comes from: the three latest GitHub releases carry no download file, so the app is fetched from openemu.org, and building it from this repository instead needs \"Xcode 14.3 and macOS Ventura\".","costNote":"Nothing in the package asks for payment. No license file was detected, so the terms for reusing the code are not stated.","notVerified":["Whether the app from openemu.org matches the code in this repository, since no checksum or signature is given","Whether a given game runs correctly, as each console is handled by a different engine with its own quirks","Whether it still works on current macOS versions, given the last commit is from October 2025 and the last release from December 2023","Real disk use, which the package does not state and which grows with every game file you add","Whether any console needs extra system files copied from the real machine before its games will start"]},"zh-Hans":{"summaryShort":"一款 Mac 应用，可在同一个游戏库里运行老式游戏机的游戏——NES、Game Boy、任天堂 64、世嘉、PlayStation 等等。","summary":"OpenEmu 是一款 macOS 程序，可以在一个熟悉的 Mac 窗口里运行超过 30 种老式游戏机的游戏——Atari、NES、Game Boy、任天堂 64、任天堂 DS、世嘉、PlayStation、PSP 等等，方法是为每种游戏机加载一个不同的模拟引擎作为插件。它需要 macOS Mojave 10.14.4 或更新版本。游戏本身不包含在内：你需要自备游戏文件。GitHub 的发布列表里没有任何下载文件，所以这个应用来自 openemu.org。最新版本是 v2.4.1（2023-12-30），这里最近一次提交是在 2025-10-22。","easeRationale":"macOS 是唯一支持的平台，评级为 A：这是一款下载后直接打开就能用的成品 Mac 应用，README 里说它 \"leverages modern macOS technologies\"，并使用 \"Sparkle, which is used for auto-updating\"。唯一要注意的是文件的来源：GitHub 上最新的三个发布版本都没有附带下载文件，所以这个应用是从 openemu.org 获取的，而如果改从这个仓库自己编译，则需要 \"Xcode 14.3 and macOS Ventura\"。","costNote":"这个包里没有任何收费项目。没有检测到许可证文件，所以重用代码的条款没有写明。","notVerified":["openemu.org 上的应用是否与这个仓库中的代码一致，因为没有给出任何校验值或签名","某个游戏是否能正常运行，因为每种游戏机都由不同的引擎处理，各有各的小毛病","它在当前版本的 macOS 上是否仍然可用，因为最近一次提交是 2025 年 10 月，而最新发布是 2023 年 12 月","实际占用的磁盘空间，这一点软件包没有说明，而且会随着你添加的每个游戏文件而增长","是否有游戏机需要先从真实主机复制额外的系统文件，游戏才能启动"]}}},{"id":"R_kgDOVLpWZg","slug":"openmoss__moss-image","name":"MOSS-Image","owner":"OpenMOSS","fullName":"OpenMOSS/MOSS-Image","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D"},"risk":"attention","signals":[{"code":"runs_code_downloaded_from_model_host","severity":"attention","evidence":"README.md 中的 Python 示例：DiffusionPipeline.from_pretrained(\"OpenMOSS-Team/MOSS-Image-Base\", custom_pipeline=\"pipeline_moss_image\", trust_remote_code=True) — 这会在你的电脑上运行从 Hugging Face 下载的 Python 代码。","source":"ai","title":"会运行从模型托管站下载的代码"},{"code":"large_model_download","severity":"info","evidence":"README.md 介绍了一个 50 亿（5B）参数的模型，其权重需要另外从 Hugging Face 或 ModelScope 下载；仓库本身不包含权重。","source":"ai","title":"模型下载量大"},{"code":"very_new_project","severity":"info","evidence":"项目信息：仓库创建于 2026-10-16，没有 GitHub 发布版本，有 25 个未解决的 issue。","source":"ai","title":"非常新的项目"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["image","local_ai"],"alternativeTo":[{"slug":"midjourney","name":"Midjourney","confirmed":true},{"slug":"ideogram","name":"Ideogram","confirmed":true},{"slug":"adobe-firefly","name":"Firefly","confirmed":true}],"declaredAlternatives":[],"stars":2785,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":234,"growthDays":7,"history":[{"date":"2026-10-27","stars":2098},{"date":"2026-10-28","stars":2169},{"date":"2026-10-29","stars":2234},{"date":"2026-10-30","stars":2306},{"date":"2026-10-31","stars":2357},{"date":"2026-11-01","stars":2430},{"date":"2026-11-02","stars":2508},{"date":"2026-11-03","stars":2551},{"date":"2026-11-04","stars":2596},{"date":"2026-11-05","stars":2638},{"date":"2026-11-06","stars":2686},{"date":"2026-11-07","stars":2715},{"date":"2026-11-08","stars":2737},{"date":"2026-11-09","stars":2766},{"date":"2026-11-10","stars":2785}],"firstPublishedAt":"2026-10-27T18:40:28.000Z","commitSha":"2b46b1e0e13b5aaec253879e0d006a42564b8b27","verifiedAt":"2026-10-27T18:29:42.000Z","lastCommitAt":"2026-11-08T13:04:50Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Creates images from a written description on your own computer, including images with readable text inside them.","summary":"MOSS-Image is an open image generator you run yourself. You write a description and it makes a picture, and it is built to write clear text inside the image and to place elements where you ask using a JSON layout. There is a simple web screen (Gradio) that opens in your browser. To use it you install Python, download the project and let it download the model files (several gigabytes), and you need a good graphics card. It also works inside ComfyUI.","easeRationale":"There is no installer: you clone the repository, create a Python environment, install the dependencies and run commands, and you need a graphics card with enough memory for a 5B model.","costNote":"Code and model weights are free; the authors also sell a paid API on the MOSI platform for people who do not want to run it locally.","notVerified":["Whether the Gradio app and the install steps work on Windows, since the README shows Linux/macOS commands only","How much graphics memory is really needed for the recommended 1456x1456 images","Real disk use after the model weights are downloaded","What the remote pipeline code downloaded with trust_remote_code does","The price of the MOSI API platform mentioned in the README"]},"zh-Hans":{"summaryShort":"在你自己的电脑上根据文字描述生成图片,包括图中带有可读文字的图片。","summary":"MOSS-Image 是一个你自己运行的开放图像生成器。你写一段描述,它就生成一张图片;它专门针对在图像里写出清晰文字,并能用一份 JSON 布局把元素放到你指定的位置。它有一个简单的网页界面(Gradio),会在你的浏览器里打开。使用前要安装 Python、下载项目,并让它下载模型文件(有几个 GB 大),还需要一块性能不错的显卡。它也能在 ComfyUI 里使用。","easeRationale":"没有安装程序:你需要克隆仓库、创建 Python 环境、安装依赖并运行命令,还需要一块显存足够跑 50 亿参数模型的显卡。","costNote":"代码和模型权重都是免费的;作者还在 MOSI 平台上出售付费 API,给不想在本地运行的人用。","notVerified":["Gradio 应用和安装步骤能否在 Windows 上跑通,因为 README 只给出了 Linux/macOS 的命令。","生成推荐的 1456x1456 图片实际需要多大的显存。","下载完模型权重后,实际占用的磁盘空间有多大。","用 trust_remote_code 下载的远程流水线代码到底做了什么。","README 里提到的 MOSI API 平台的价格。"]}}},{"id":"R_kgDOR_Oz7Q","slug":"openmoss__moss-tts-nano","name":"MOSS-TTS-Nano","owner":"OpenMOSS","fullName":"OpenMOSS/MOSS-TTS-Nano","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：torch：GHSA-3749-ghw9-m3mg、GHSA-63cw-57p8-fm3p、GHSA-887c-mr87-cxwp；transformers：GHSA-29pf-2h5f-8g72、GHSA-69w3-r845-3855、GHSA-fgcw-684q-jj6r","source":"check","title":"已知漏洞"},{"code":"copies_a_voice_from_a_short_clip","severity":"attention","evidence":"README 把声音克隆称为 \"the main recommended workflow for MOSS-TTS-Nano\"，示例命令是 `python infer.py --prompt-audio-path assets/audio/zh_1.wav --text \"...\"`，所以你手上任何一段录音都能变成模型可以用来说话的声音。这个软件包里没有提到任何同意规则、免责声明，也没有在生成的音频上做标记。","source":"ai","title":"从一段简短录音中复制声音"},{"code":"install_step_pulls_code_from_a_personal_fork","severity":"attention","evidence":"推荐的 Conda 安装过程会执行 `pip install git+https://github.com/WhizZest/WeTextProcessing.git`，这是文本规范化包的一份拷贝，保存在个人账号下，而不是原项目或 PyPI，README 还指向 \"a community-tested example, see Issue #6\" 来寻找 `pynini` 的 wheel 包。","source":"ai","title":"安装步骤会拉取个人fork的代码"},{"code":"readme_license_section_contradicts_the_license_file","severity":"info","evidence":"README 的 License 部分写道：\"This repository will follow the license specified in the root `LICENSE` file. If you are reading this before that file is published, please treat the repository as **not yet licensed for redistribution**.\" 而根目录的 LICENSE 文件实际上是 Apache License 2.0，GitHub 也报告为 Apache-2.0，所以这看起来是第一次发布时遗留下来的文字。","source":"ai","title":"readme的许可证说明与许可证文件相互矛盾"},{"code":"downloads_models_on_first_run","severity":"info","evidence":"README：\"If `--model-dir` is omitted, the script automatically checks `./models`. When the model files are missing, it downloads them on first run from\" Hugging Face 上的 MOSS-TTS-Nano-100M-ONNX 和 MOSS-Audio-Tokenizer-Nano-ONNX 仓库。检查结果估计会占用 950 MB 到 4450 MB 的磁盘空间。","source":"ai","title":"首次运行时会下载模型"},{"code":"speed_and_quality_numbers_come_from_the_authors","severity":"info","evidence":"\"nearly 2x\" 的效率提升、在 MacBook Air M4 单核上的结果，以及 \"best overall reconstruction quality\" 的表格，全部都是作者自己在 README 和论文（arXiv 2603.18090 和 2602.10934）中给出的测量结果。README 还提到该模型只有 0.1B 参数，远小于同系列 8B 的旗舰模型 MOSS-TTS。","source":"ai","title":"速度和质量数据来自作者自己"},{"code":"young_project_at_version_0_1_0","severity":"info","evidence":"创建于 2026-04-10，pyproject.toml 显示版本号为 \"0.1.0\"，GitHub 上完全没有发布版本，最近一次提交是在 2026-09-06（本次检查前 63 天），有 28 个未关闭的 issue，对应 4645 个星标。","source":"ai","title":"项目较新版本号为0 1 0"}],"platforms":["windows","mac","linux","web"],"tools":["python"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":4657,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":44,"growthDays":6,"history":[{"date":"2026-11-06","stars":4635},{"date":"2026-11-07","stars":4642},{"date":"2026-11-08","stars":4645},{"date":"2026-11-09","stars":4651},{"date":"2026-11-10","stars":4657}],"firstPublishedAt":"2026-11-08T23:08:23.109Z","commitSha":"8b7bcc9341b3b4ef3a3a58ba1338a7d85ff133eb","verifiedAt":"2026-11-08T22:45:19.818Z","lastCommitAt":"2026-09-06T06:19:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from a short clip and reads your text in it, in 20 languages, on a normal computer with no graphics card.","summary":"MOSS-TTS-Nano is a small text-to-speech model (0.1B parameters) you run yourself. Give it a few seconds of someone speaking and it reads your text in that voice, in 20 languages, as 48 kHz stereo audio. The authors say it is fast enough for real time on a 4-core processor with no graphics card, and that the ONNX version runs on a single core of a MacBook Air M4. You install Python packages and open a page on your own machine, or try the authors' free online demo first. The model files download on first run.","easeRationale":"The easiest way in is the web at grade A: the authors keep a free online demo and a Hugging Face Space, so you open a page and type text. Installing it is C on Windows, macOS and Linux: Python 3.10 or newer, a Conda environment, `pip install -r requirements.txt` and `pip install -e .`, then `python app_onnx.py` and open http://127.0.0.1:18083 — no graphics card needed. The README itself warns that one dependency often fails (\"If `WeTextProcessing` or `pynini` fails to install\"), which can turn those few commands into a longer fight.","costNote":"Free software under Apache-2.0 and free model files. The cost is your own machine plus roughly 1 to 4.5 GB of disk for the downloaded models.","notVerified":["How natural the cloned voice sounds for your language, since the authors' own tests focus on Chinese and English","Whether it really keeps up in real time on your processor, as the CPU claims are the authors' measurements on their machines","Real disk use after the model files are downloaded, estimated at 950 MB to 4450 MB but not stated by the project","What the authors' online demo and Hugging Face Space do with the reference audio and text you upload","Whether the files fetched from Hugging Face on first run match the model the paper describes, as no checksum is given"]},"zh-Hans":{"summaryShort":"从一段简短的录音中复制声音，用它读出你的文本，支持 20 种语言，在没有显卡的普通电脑上也能运行。","summary":"MOSS-TTS-Nano 是一个你可以自己运行的小型文本转语音模型（0.1B 参数）。给它几秒钟某人说话的录音，它就能用那个声音读出你的文本，支持 20 种语言，输出 48 kHz 立体声音频。作者说它在没有显卡的 4 核处理器上也能快到满足实时需求，而 ONNX 版本在 MacBook Air M4 的单核上就能运行。你需要安装 Python 包并在自己的电脑上打开一个页面，或者先试试作者提供的免费在线演示。模型文件会在第一次运行时下载。","easeRationale":"最简单的入门方式是网页版，评级为 A：作者提供了免费的在线演示和一个 Hugging Face Space，你只需打开页面输入文字。在 Windows、macOS 和 Linux 上自行安装的评级是 C：需要 Python 3.10 或更新版本、一个 Conda 环境、`pip install -r requirements.txt` 和 `pip install -e .`，然后运行 `python app_onnx.py` 并打开 http://127.0.0.1:18083——不需要显卡。README 自己也提醒有一个依赖经常安装失败（\"If `WeTextProcessing` or `pynini` fails to install\"），这可能会让这几条命令变成一场更长的折腾。","costNote":"软件基于 Apache-2.0 协议免费提供，模型文件也是免费的。成本在于你自己的电脑，以及下载模型所需的大约 1 到 4.5 GB 磁盘空间。","notVerified":["克隆出来的声音在你的语言里听起来有多自然，因为作者自己的测试主要集中在中文和英文上","在你的处理器上是否真的能跟上实时速度，因为 CPU 方面的说法是作者在自己机器上测得的数据","下载模型文件后实际占用的磁盘空间，估计在 950 MB 到 4450 MB 之间，但项目本身没有说明","作者的在线演示和 Hugging Face Space 会如何处理你上传的参考音频和文本","第一次运行时从 Hugging Face 获取的文件是否与论文中描述的模型一致，因为没有给出任何校验值"]}}},{"id":"R_kgDOO88x-g","slug":"openmoss__moss-ttsd","name":"MOSS-TTSD","owner":"OpenMOSS","fullName":"OpenMOSS/MOSS-TTSD","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：torch：GHSA-63cw-57p8-fm3p、GHSA-qfhq-4f3w-5fph、GHSA-rrmf-rvhw-rf47；transformers：GHSA-29pf-2h5f-8g72、GHSA-fgcw-684q-jj6r、GHSA-x9r9-c232-4q39；orjson：GHSA-hx9q-6w63-j58v、PYSEC-2026-107；gradio：GHSA-39mp-8hj3-5c49、GHSA-6655-8ph2-63j3、GHSA-7hp7-4p35-3cx2","source":"check","title":"已知漏洞"},{"code":"copies_voices_from_short_samples","severity":"attention","evidence":"项目简介承诺提供 \"zero-shot voice cloning\"，推荐模式是 `--mode voice_clone_and_continuation`，输入 `prompt_audio_speaker1` 到 `prompt_audio_speaker5`。README 自己的 Usage Disclaimer 写道：\"Users must not use this model for unauthorized voice cloning, impersonation, fraud, scams, deepfakes, or any illegal activities\"——这个愿望写在了文档里，但代码本身并不会检查是否取得同意，也不会在生成的音频上做标记。","source":"ai","title":"从简短样本中复制声音"},{"code":"readme_example_invents_a_talk_between_real_people","severity":"attention","evidence":"Quick Start 示例从 `asset/reference_02_s1.wav` 和 `asset/reference_02_s2.wav` 克隆了两个声音，它们的文本分别是 \"In short, we embarked on a mission to make America great again for all Americans\" 和 \"NVIDIA reinvented computing for the first time after 60 years\"，然后让它们围绕中国进行一场编造的争论，结尾是 \"It's a disaster, Jensen. A total disaster.\" 这个项目的第一个演示，就是把话硬塞进可识别的公众人物嘴里。","source":"ai","title":"readme的示例编造了真实人物之间的对话"},{"code":"runs_code_downloaded_with_the_model","severity":"attention","evidence":"文档中给出的调用是 `AutoProcessor.from_pretrained(..., trust_remote_code=True)` 和 `AutoModel.from_pretrained(..., trust_remote_code=True)`，服务器启动命令是 `sglang serve ... --trust-remote-code`。这个设置允许保存在 Hugging Face 权重旁边的 Python 文件在你的机器上运行。","source":"ai","title":"运行随模型一起下载的代码"},{"code":"the_fast_server_listens_on_every_network_address","severity":"attention","evidence":"SGLang 的说明是 `sglang serve --model-path <path-to-fused-model> --delay-pattern --trust-remote-code --port 30000 --host 0.0.0.0`。主机地址 0.0.0.0 意味着任何能连到你机器的设备都可以向这个端口发送请求，而且没有提到任何密码或令牌。","source":"ai","title":"快速服务器监听所有网络地址"},{"code":"speed_setup_uses_a_forked_engine","severity":"info","evidence":"为了走更快的路径，README 让你执行 `git clone https://github.com/OpenMOSS/sglang -b moss-ttsd-v1.0-with-cat` 和 `pip install ./sglang/python[all]`——这是项目自己的 SGLang 分支，他们称之为 \"the deeply extended SGLang from OpenMOSS\"，而不是官方发布版本。","source":"ai","title":"加速方案使用了一个分支版本的引擎"},{"code":"the_comparison_tables_are_the_authors_own","severity":"info","evidence":"与 Eleven V3、gemini-2.5-pro-preview-tts、Doubao_Podcast 和 Vibevoice 的对比分数，来自作者自己的 TTSD-eval 评测框架和他们自己的标注员（\"we compute Elo ratings\"），与论文 arXiv:2603.19739 一同发布。没有引用任何独立测试。","source":"ai","title":"对比表格来自作者自己"},{"code":"no_versions_published_on_github","severity":"info","evidence":"GitHub 上完全没有发布版本；版本（v0.5、v0.7、v1.0）只以 Hugging Face 模型仓库和 News 条目的形式存在，旧代码保存在一个 `legacy/` 文件夹里。最近一次提交是在 2026-09-06（本次检查前 63 天），有 53 个未关闭的 issue，对应 1418 个星标。","source":"ai","title":"github上没有发布过版本"}],"platforms":["windows","linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"notebooklm","name":"NotebookLM","confirmed":true},{"slug":"descript","name":"Descript","confirmed":true}],"declaredAlternatives":[],"stars":1420,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":4,"growthDays":6,"history":[{"date":"2026-11-07","stars":1417},{"date":"2026-11-08","stars":1418},{"date":"2026-11-09","stars":1419},{"date":"2026-11-10","stars":1420}],"firstPublishedAt":"2026-11-08T23:08:23.117Z","commitSha":"46973e425da227bfa3b315dcd7a4aba8b6be5574","verifiedAt":"2026-11-08T22:46:16.329Z","lastCommitAt":"2026-09-06T06:19:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a written dialogue into a spoken conversation with up to five voices, each copied from a short audio sample.","summary":"MOSS-TTSD turns a written dialogue into spoken conversation: 1 to 5 speakers, up to 60 minutes in one session, in 20 languages, with each voice copied from a short reference recording. The authors aim it at podcasts, audiobooks, dubbing and sports or esports commentary. The model has 8 billion parameters and the documented setup expects Python 3.12, an NVIDIA graphics card and commands in a terminal; the authors also host a free online demo you can try first. Free under Apache-2.0, with the model files downloaded from Hugging Face.","easeRationale":"The easiest way in is the web at grade A: the authors keep a free Hugging Face Space, so you open a page and listen. Running it yourself is D on Linux and Windows: `conda create -n moss_ttsd python=3.12`, `pip install -r requirements.txt` and `pip install flash-attn`, where requirements.txt pins the CUDA builds `torch==2.9.1+cu128` and `torchaudio==2.9.1+cu128`, so an NVIDIA card is assumed; then you write a Python script or a JSONL file and run `python inference.py`. macOS is not graded because the pinned CUDA packages and flash-attn are not for Apple machines.","costNote":"Free software under Apache-2.0 and free model files. The cost is the hardware: an 8B model needs a graphics card with enough memory, and the README warns about \"VRAM fragmentation\" if \"GPU memory is tight\".","notVerified":["Real disk and graphics-memory use, since the model is 8 billion parameters and the estimate in the checks is a heuristic, not a measurement","Whether 60-minute audio really stays stable, as that figure is the authors' own claim","How good the result sounds in the 18 languages beyond Chinese and English, which the published tables do not cover","What the authors' Hugging Face Space does with the reference audio and the script you upload","Which Python files come down with the weights and run on your machine through trust_remote_code"]},"zh-Hans":{"summaryShort":"把一段文字对话变成最多五个声音的真人对话，每个声音都从一小段音频样本复制而来。","summary":"MOSS-TTSD 能把一段写好的对话变成有声对话：1 到 5 个说话人，单次会话最长 60 分钟，支持 20 种语言，每个声音都从一段简短的参考录音复制而来。作者把它用在播客、有声书、配音以及体育或电竞解说等场景。模型有 80 亿参数，文档中的配置要求 Python 3.12、一张 NVIDIA 显卡和终端命令；作者还提供了一个免费在线演示可以先试用。基于 Apache-2.0 协议免费提供，模型文件从 Hugging Face 下载。","easeRationale":"最简单的入门方式是网页版，评级为 A：作者提供了一个免费的 Hugging Face Space，你只需打开页面收听即可。在 Linux 和 Windows 上自己运行的评级是 D：`conda create -n moss_ttsd python=3.12`、`pip install -r requirements.txt` 和 `pip install flash-attn`，其中 requirements.txt 固定了 CUDA 版本的 `torch==2.9.1+cu128` 和 `torchaudio==2.9.1+cu128`，所以默认假设你有一张 NVIDIA 显卡；之后你要编写一个 Python 脚本或一个 JSONL 文件，再运行 `python inference.py`。macOS 没有评级，因为固定的 CUDA 依赖包和 flash-attn 都不适用于苹果电脑。","costNote":"软件基于 Apache-2.0 协议免费提供，模型文件也是免费的。成本在于硬件：80 亿参数的模型需要一张显存足够的显卡，README 还提醒，如果 \"GPU memory is tight\" 可能出现 \"VRAM fragmentation\"。","notVerified":["实际占用的磁盘和显存，因为这个模型有 80 亿参数，而检查里的估算只是一种经验推测，不是实测数据","60 分钟的音频是否真的能保持稳定，因为这个数字是作者自己的说法","除了中文和英文之外的 18 种语言，效果听起来如何，这一点已发布的表格并未涉及","作者的 Hugging Face Space 会如何处理你上传的参考音频和脚本","哪些 Python 文件会随着权重一起下载，并通过 trust_remote_code 在你的机器上运行"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzNjY0ODMwMQ==","slug":"openshot__openshot-qt","name":"OpenShot","owner":"OpenShot","fullName":"OpenShot/openshot-qt","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","ffmpeg"],"disk":{"minMB":216,"maxMB":648,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"file"},"categories":["video"],"alternativeTo":[{"slug":"premiere-pro","name":"Premiere Pro","confirmed":true},{"slug":"final-cut-pro","name":"Final Cut Pro","confirmed":true}],"declaredAlternatives":[],"stars":6764,"downloads":37242327,"downloadsWeek":68407,"downloadsDays":7,"growth":24,"growthDays":7,"history":[{"date":"2026-10-25","stars":6693},{"date":"2026-10-26","stars":6694},{"date":"2026-10-27","stars":6699},{"date":"2026-10-28","stars":6702},{"date":"2026-10-29","stars":6705},{"date":"2026-10-30","stars":6707},{"date":"2026-10-31","stars":6715},{"date":"2026-11-01","stars":6724},{"date":"2026-11-02","stars":6738},{"date":"2026-11-03","stars":6740},{"date":"2026-11-04","stars":6742},{"date":"2026-11-05","stars":6746},{"date":"2026-11-06","stars":6750},{"date":"2026-11-07","stars":6752},{"date":"2026-11-08","stars":6757},{"date":"2026-11-09","stars":6760},{"date":"2026-11-10","stars":6764}],"firstPublishedAt":"2026-10-25T23:19:23.038Z","commitSha":"856fa5e31495067b893cc38b214a735dda184438","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T05:41:16Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"OpenShot-v4.1.1-x86_64.exe","sizeBytes":248497059,"url":"https://github.com/OpenShot/openshot-qt/releases/download/v4.1.1/OpenShot-v4.1.1-x86_64.exe","releaseUrl":"https://github.com/OpenShot/openshot-qt/releases/tag/v4.1.1"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Edit videos on your computer: cut clips, add music, titles, transitions and effects, then export.","summary":"OpenShot is a desktop video editor for beginners and hobbyists. You drag clips onto a timeline with unlimited tracks, trim and cut them, add transitions, titles, music and effects such as chroma key (green screen), change speed, and animate with keyframes. It reads and exports many video, audio and image formats. Windows, macOS and Linux have ready installers; on Windows the download is around 216 MB and includes everything needed. It is free under the GPL license.","easeRationale":"The v4.0.1 release includes ready Windows installers (OpenShot-v4.0.1-x86_64.exe, about 216 MB, plus a 32-bit version), so you download it and run it; Python and FFmpeg are only needed when building from source.","notVerified":["How smoothly it plays back and exports high-resolution or long videos on an average computer.","Whether the experimental hardware encoding and decoding work with your graphics card.","How stable the new 4.0 version is for larger projects."]},"zh-Hans":{"summaryShort":"在电脑上剪辑视频：剪切片段，添加音乐、标题、转场和特效，然后导出。","summary":"OpenShot 是一款面向初学者和爱好者的桌面视频编辑器。你把片段拖到轨道数量不限的时间轴上，进行裁剪和切割，添加转场、标题、音乐和特效（如色度键抠像，即绿幕），调整速度，并用关键帧制作动画。它可读取和导出多种视频、音频和图片格式。Windows、macOS 和 Linux 都有现成的安装程序；Windows 版下载约 216 MB，已包含所需的一切。它在 GPL 许可证下免费。","easeRationale":"v4.0.1 版本包含现成的 Windows 安装程序（OpenShot-v4.0.1-x86_64.exe，约 216 MB，另有 32 位版本），下载后运行即可；只有从源代码构建时才需要 Python 和 FFmpeg。","notVerified":["在普通电脑上播放和导出高分辨率或较长视频是否流畅。","实验性的硬件编码和解码能否在你的显卡上正常工作。","新的 4.0 版本在较大项目中稳定性如何。"]}}},{"id":"R_kgDOH_j-rA","slug":"opentalker__video-retalking","name":"VideoReTalking","owner":"OpenTalker","fullName":"OpenTalker/video-retalking","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2023-03-15（距今 1334 天）","source":"check","title":"发布版本已过时"},{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/drive/folders/18rhjMpxK8LVVxf7PI6XwOidt8Vouv_H0?usp=share_link","source":"check","title":"外部下载链接"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：basicsr：GHSA-86w8-vhw6-q9qq、PYSEC-2026-1215","source":"check","title":"已知漏洞"},{"code":"makes_a_real_person_appear_to_say_other_words","severity":"attention","evidence":"README：\"a new system to edit the faces of a real-world talking head video according to input audio, producing a high-quality and lip-syncing output video even with a different emotion\"，还配有一个 \"face enhancement network\" 用来 \"improve the photo-realism\"。项目自己的 Disclaimer 禁止 \"fraud, deception, infringement of others' portrait rights, reputation rights\"，但代码里没有任何部分会检查视频里是谁，也不会在结果上做标记。","source":"ai","title":"让真实的人看起来说了别的话"},{"code":"the_model_files_come_from_a_shared_google_drive","severity":"attention","evidence":"README 的 Pretrained Models 部分写道：\"Please download our pre-trained models and put them in `./checkpoints`\"，指向一个 Google Drive 文件夹（drive.google.com/drive/folders/18rhjMpxK8LVVxf7PI6XwOidt8Vouv_H0）。没有这些文件什么都跑不起来，这个文件夹在 GitHub 之外，也没有给出任何大小、校验值或签名。","source":"ai","title":"模型文件来自一个共享的google drive"},{"code":"code_frozen_on_an_old_software_stack","severity":"attention","evidence":"最近一次提交是在 2023-11-04（本次检查前 1100 天），说明文档固定了 Python 3.8、`torch==1.9.0+cu111` 和 `dlib==19.24.0`，有 215 个未关闭的 issue，对应 7298 个星标。这些版本都是 2021 年的，现在的显卡驱动或 Python 可能已经不接受它们，而且没有人更新过这份安装方法。","source":"ai","title":"代码停留在一套老旧的软件栈上"},{"code":"not_an_official_tencent_product","severity":"info","evidence":"Disclaimer：\"This is not an official product of Tencent\"，第 5 条还补充说 \"This open-source code does not have the ability to directly provide services to end-users\"。有几位作者的署名单位是腾讯 AI Lab，但这个仓库是研究代码，不是有支持的产品。","source":"ai","title":"不是腾讯的官方产品"},{"code":"built_on_six_other_research_projects","severity":"info","evidence":"仓库里有一个 `third_part/` 文件夹，Acknowledgement 部分感谢了 Wav2Lip、PIRenderer、GFP-GAN、GPEN、ganimation_replicate 和 STIT \"for sharing their code\"。这些部分各有自己的许可证和条件，这个仓库里的 Apache-2.0 文件并不涵盖它们。","source":"ai","title":"建立在另外六个研究项目之上"},{"code":"limits_stated_by_the_authors","severity":"info","evidence":"README：\"You can test any talking face videos without manual alignment. But it is worth noting that DNet cannot handle extreme poses.\" 预告片里的质量数据来自作者自己的论文（arXiv:2211.14758）。","source":"ai","title":"作者自己说明的限制"}],"platforms":["windows","linux","web"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","ffmpeg"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"rask-ai","name":"Rask AI","confirmed":true}],"declaredAlternatives":[],"stars":7300,"downloads":461156,"downloadsWeek":945,"downloadsDays":6,"growth":1,"growthDays":6,"history":[{"date":"2026-11-04","stars":7299},{"date":"2026-11-05","stars":7300},{"date":"2026-11-06","stars":7299},{"date":"2026-11-07","stars":7300},{"date":"2026-11-08","stars":7298},{"date":"2026-11-09","stars":7298},{"date":"2026-11-10","stars":7300}],"firstPublishedAt":"2026-11-08T23:08:23.126Z","commitSha":"d32e8e58248255e2d243eeaf3cba545dbe505ca8","verifiedAt":"2026-11-08T22:47:21.050Z","lastCommitAt":"2023-11-04T04:36:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Edits a video of someone talking so their mouth matches a new audio track, and can change the expression too.","summary":"VideoReTalking takes an existing video of a person talking and reworks the face so the mouth matches a different audio track, with an option to change the expression (neutral, smile, surprise, angry). It comes from a SIGGRAPH Asia 2022 paper by Xidian University, Tencent AI Lab and Tsinghua University. To run it you set up Python 3.8 and an NVIDIA card on CUDA 11.1, install ffmpeg, download the model files from a Google Drive folder and run one command. There is also a Colab notebook and a paid page run by other people.","easeRationale":"The gentlest way in is the web at grade B: the README offers a Colab notebook (quick_demo.ipynb) and a Replicate page, where you open the page and press run instead of installing anything. Installing it is D on Linux and Windows: `conda create -n video_retalking python=3.8`, `conda install ffmpeg`, then a graphics-card-specific PyTorch line the README itself limits (\"This installation command only works on CUDA 11.1\"), `pip install -r requirements.txt`, and a manual download of the pre-trained models into `./checkpoints` before `python3 inference.py` will run.","costNote":"The code is free under Apache-2.0. The README also links a Replicate page run by someone else (cjwbw/video-retalking); services like that charge for the machine time, which the README does not explain.","notVerified":["Whether the recipe still installs today, given the CUDA 11.1 and Python 3.8 pins and three years without commits","What exactly is inside the Google Drive checkpoints, since no checksum or file list is published","Real disk use, as the checks estimate 100 to 300 MB from a heuristic and the model files are not counted","How long one video takes on your graphics card, which the README never states","How convincing the result looks on your own footage, especially with side views or fast movement"]},"zh-Hans":{"summaryShort":"编辑一段某人说话的视频，让嘴型配合新的音轨，还可以改变表情。","summary":"VideoReTalking 拿一段已有的人物说话视频，重新处理面部，让嘴型配合另一段音轨，还可以选择改变表情（中立、微笑、惊讶、生气）。它来自西安电子科技大学、腾讯 AI Lab 和清华大学在 SIGGRAPH Asia 2022 发表的一篇论文。要运行它，你需要配置 Python 3.8 和一张支持 CUDA 11.1 的 NVIDIA 显卡，安装 ffmpeg，从一个 Google Drive 文件夹下载模型文件，再运行一条命令。另外还有一个 Colab notebook 和一个由其他人运营的付费页面。","easeRationale":"最温和的入门方式是网页版，评级为 B：README 提供了一个 Colab notebook（quick_demo.ipynb）和一个 Replicate 页面，你只需打开页面点击运行，而不用安装任何东西。自己安装的评级是 D，适用于 Linux 和 Windows：`conda create -n video_retalking python=3.8`、`conda install ffmpeg`，然后是一行针对特定显卡的 PyTorch 安装命令，README 自己也限定了范围（\"This installation command only works on CUDA 11.1\"），接着 `pip install -r requirements.txt`，还要手动把预训练模型下载到 `./checkpoints`，之后 `python3 inference.py` 才能运行。","costNote":"代码基于 Apache-2.0 协议免费提供。README 还链接了一个由他人运营的 Replicate 页面（cjwbw/video-retalking）；这类服务按机器运行时间收费，README 并未说明这一点。","notVerified":["考虑到 CUDA 11.1 和 Python 3.8 这些固定版本，以及三年没有提交，这份安装方法现在是否仍然能装上","Google Drive 的 checkpoints 里到底是什么，因为没有公布任何校验值或文件清单","实际占用的磁盘空间，因为检查里的估算是基于经验推测的 100 到 300 MB，并未计入模型文件","在你的显卡上处理一段视频要花多长时间，README 从未说明","在你自己的素材上，尤其是侧脸或快速动作时，效果看起来有多逼真"]}}},{"id":"R_kgDORMI4oA","slug":"osmantic__ods","name":"ODS","owner":"Osmantic","fullName":"Osmantic/ODS","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接通过管道传给 shell：curl -fsSL https://install.osmantic.com/ods.sh | bash","source":"check","title":"curl管道直通shell"},{"code":"the_installers_fetch_the_moving_main_branch","severity":"attention","evidence":"README Status：\"The installers below install the current `main` branch, which receives fixes continuously and is not a signed release\"，以及 \"The hosted Linux/macOS endpoint proxies the current bootstrap from repository `main`. Reviewed merges reach it automatically after edge-cache refresh.\" 所以你安装的是今天刚合并的内容，而不是 v3.0.0 这个标签版本，README 还补充说 \"no release has qualified for the verified channel yet\"。","source":"ai","title":"安装程序会获取不断变化的main分支"},{"code":"windows_setup_turns_off_the_script_block","severity":"attention","evidence":"Windows 的安装命令里包含 `Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass`，在运行 `install.ps1` 之前，先把 `https://github.com/Osmantic/ODS/archive/refs/heads/main.zip` 下载到一个临时文件夹。README 解释说这 \"allows the installer script to run in the current session. It does not change your system-wide policy\"，但这在 Windows 上相当于直接运行一个从网上下载、未经检查的脚本。","source":"ai","title":"windows安装过程关闭了脚本限制"},{"code":"the_project_tells_you_to_read_its_security_advisories_first","severity":"attention","evidence":"README Status：\"`v3.0.0` is the latest published source release; check the [security advisories](https://github.com/Osmantic/ODS/security/advisories) before pinning it.\" 这个仓库还附带了 SECURITY.md 和 SECURITY_AUDIT.md，有 779 个未关闭的 issue，对应 7797 个星标，项目创建于 2026-02-09。","source":"ai","title":"项目提醒你先阅读它的风险通告"},{"code":"part_of_the_bundled_code_is_not_open_source","severity":"info","evidence":"README License：\"ODS code is Apache-2.0 except the bundled Pixel source, which has a separate ODS-only use and distribution grant\"，并指向 ods/LICENSING.md 和 ods/vendor/pixel/LICENSE.md。Pixel 是 Windows 和 macOS 推荐安装方式会装上的代理运行时，所以 Apache-2.0 通常给的自由并不覆盖全部内容。","source":"ai","title":"打包代码的一部分不是开源的"},{"code":"many_services_listening_on_your_machine","severity":"info","evidence":"README 记录了 Open WebUI 使用 http://localhost:3000、Portal 控制面板使用 http://localhost:3001、llama-server 使用 11434 或 8080，并把 Qdrant、SearXNG、n8n、Whisper、Kokoro、ComfyUI、LiteLLM 和 Langfuse 作为容器安装。安装程序还支持 `-Lan` / `--lan` 开关，以及带有 \"QR onboarding, first-boot setup, AP mode, mDNS\" 的无屏模式，这些都会让这些页面超出你自己电脑的范围。","source":"ai","title":"你的机器上有很多服务在监听"},{"code":"wants_a_lot_of_disk_and_downloads_models_by_itself","severity":"info","evidence":"Windows 安装程序会 \"Checks free disk space (40 GB) and that hardware virtualization is on\"。Bootstrap 模式会 \"downloads Qwen3.5 2B (about 1.3 GB) and starts it\"，同时 \"the full model for your hardware downloads in the background\"——在一张 24 GB 的 NVIDIA 显卡上，默认选择是 Qwen3.5 27B，这是一个大得多的文件。","source":"ai","title":"需要大量磁盘空间并会自行下载模型"},{"code":"agents_can_run_tools_on_your_computer","severity":"info","evidence":"方框列表里包括 \"Hermes Agent — independent general-purpose agent ... includes memory, skills, and a proxy with optional owner-card gating; direct access by default\"、\"APE — Agent Policy Engine for auditing and governing autonomous tool calls\" 以及 \"OpenCode — browser-based AI coding assistant wired to the local stack\"。默认情况下，这个代理在你没有自己设置策略关卡的情况下就会直接行动。","source":"ai","title":"代理可以在你的电脑上运行工具"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenAI, Anthropic or Together (only for cloud or hybrid mode)","Brave Search"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["local_ai","ai_agents","automation"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"n8n-cloud","name":"Cloud","confirmed":true},{"slug":"midjourney","name":"Midjourney","confirmed":true}],"declaredAlternatives":[],"stars":7809,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":49,"growthDays":6,"history":[{"date":"2026-11-05","stars":7773},{"date":"2026-11-06","stars":7782},{"date":"2026-11-07","stars":7789},{"date":"2026-11-08","stars":7796},{"date":"2026-11-09","stars":7802},{"date":"2026-11-10","stars":7809}],"firstPublishedAt":"2026-11-08T23:01:18.902Z","commitSha":"891f722ab7cca7715fa719cff5f8b3ac3b09c1ac","verifiedAt":"2026-11-08T22:48:22.907Z","lastCommitAt":"2026-11-03T21:45:36Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Installs a whole private AI server on your own computer: local chat, voice, image generation, workflows and a dashboard.","summary":"ODS (Osmantic Deployment System) turns one computer into a private AI server. A single installer command wires together local model serving, an Open WebUI chat page, a control dashboard, speech to text and text to speech, n8n workflows, local search, document retrieval and ComfyUI image generation, and picks a model that fits your graphics card or memory. It runs on Linux, on Windows through WSL2 and on Apple Silicon Macs, needs Docker and checks for 40 GB of free disk. This is a V3 pre-release.","easeRationale":"All three systems are D because everything happens in a terminal and Docker is required. Linux and macOS run one line, `curl -fsSL https://install.osmantic.com/ods.sh | bash`, but \"Docker must be installed and running\" first and macOS needs Apple Silicon plus Docker Desktop. Windows is the longest path even though it is guided: you paste a PowerShell block that downloads the source ZIP, then the installer enables WSL2, installs Docker Desktop with winget, asks for administrator permission, restarts the computer once, downloads Ubuntu 24.04 and asks you to invent an Ubuntu username and password.","costNote":"The software costs nothing and local mode needs no account. Optional extras do cost: cloud mode runs on \"OpenAI/Anthropic/Together APIs\" and the list of services includes \"Brave Search — optional paid Brave Search API integration\".","notVerified":["Whether the privacy promises hold in practice (\"ODS collects no telemetry\", online only for models, images, release checks and searches) — that is the project's own statement","Whether the one-line install really finishes on your distribution, hardware and driver version","Real disk use, since the checks could not measure it and the 40 GB check does not include the big model files","How fast the chosen model answers on your machine, which the README asks you to measure yourself","What the script served from install.osmantic.com contains at the moment you run it, as it tracks the main branch and is not signed"]},"zh-Hans":{"summaryShort":"在你自己的电脑上安装一整套私有 AI 服务器：本地聊天、语音、图像生成、工作流和控制面板。","summary":"ODS（Osmantic Deployment System）能把一台电脑变成私有 AI 服务器。一条安装命令就能把本地模型服务、Open WebUI 聊天页面、控制面板、语音转文字和文字转语音、n8n 工作流、本地搜索、文档检索和 ComfyUI 图像生成连接在一起，并根据你的显卡或内存自动选一个合适的模型。它可以在 Linux、通过 WSL2 的 Windows 以及搭载 Apple Silicon 的 Mac 上运行，需要 Docker，并会检查是否有 40 GB 的空闲磁盘空间。这是 V3 的预发布版本。","easeRationale":"三个系统的评级都是 D，因为一切都要在终端里完成，而且必须安装 Docker。Linux 和 macOS 只需一行命令，`curl -fsSL https://install.osmantic.com/ods.sh | bash`，但前提是 \"Docker must be installed and running\"，而 macOS 还需要 Apple Silicon 加 Docker Desktop。Windows 是最长的一条路，即便有引导：你要粘贴一段 PowerShell 代码下载源码 ZIP 包，然后安装程序会启用 WSL2，用 winget 安装 Docker Desktop，请求管理员权限，重启一次电脑，下载 Ubuntu 24.04，并要求你自己想一个 Ubuntu 用户名和密码。","costNote":"这个软件本身不收费，本地模式也不需要账号。可选的扩展功能是收费的：云模式运行在 \"OpenAI/Anthropic/Together APIs\" 上，服务列表里还包括 \"Brave Search — optional paid Brave Search API integration\"。","notVerified":["隐私方面的承诺（\"ODS collects no telemetry\"，只有模型、图像、版本检查和搜索才联网）在实际使用中是否真的成立——这只是项目自己的说法","一条命令的安装方式在你的发行版、硬件和驱动版本上是否真的能顺利完成","实际占用的磁盘空间，因为检查无法测量它，而 40 GB 的检查也没有把大型模型文件计算在内","选定的模型在你的机器上响应速度有多快，README 让你自己去测量","install.osmantic.com 提供的脚本在你运行的那一刻里面具体是什么内容，因为它跟随的是main分支，并没有签名"]}}},{"id":"R_kgDOOD8xTg","slug":"other-blowsnow__n8n-i18n-chinese","name":"n8n-i18n-chinese","owner":"other-blowsnow","fullName":"other-blowsnow/n8n-i18n-chinese","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"archived","severity":"info","evidence":"该仓库已归档（只读，不再维护）","source":"check","title":"已归档"},{"code":"repackages_someone_elses_program_with_no_license","severity":"attention","evidence":"Dockerfile 内容是 `FROM n8nio/n8n:${VERSION}` 加上 `COPY ./editor-ui-dist /usr/local/lib/node_modules/n8n/node_modules/n8n-editor-ui/dist`，README 把重新构建的结果作为镜像 `blowsnow/n8n-chinese` 提供出来。这个仓库里没有检测到任何许可证文件，也没有在任何地方引用 n8n 自己的条款，所以修改并再分发 n8n 到底被允许做到什么程度，没有答案。","source":"ai","title":"在没有许可证的情况下重新打包别人的程序"},{"code":"every_command_turns_off_the_secure_cookie","severity":"attention","evidence":"三种安装方式都设置了 `N8N_SECURE_COOKIE=false`（两条 `docker run` 命令和 docker-compose.yml 都是）。这个开关让 n8n 不再要求登录 Cookie 必须通过加密连接传输，这样在笔记本电脑上用起来很方便，但如果同一套环境是从网络上访问的，登录页面就会因此变得更脆弱。","source":"ai","title":"每条命令都关掉了 Cookie 的加密保护"},{"code":"the_chinese_build_comes_from_a_personal_account","severity":"attention","evidence":"第一条推荐命令拉取的是 `blowsnow/n8n-chinese`，这是某个人在 Docker Hub 上的账号里的镜像，而不是 n8n 的官方镜像；另一种方式下载的 `editor-ui.tar.gz` 也是同一个人构建的。两者都没有提供校验值、签名或构建日志，而这些文件最终会放进 n8n 自己的程序文件夹里。","source":"ai","title":"中文版本来自个人账号"},{"code":"each_pack_only_fits_one_n8n_version","severity":"attention","evidence":"仓库里有 182 个按 n8n 版本命名的发行版（release/2.33.7、release/2.33.6、release/2.33.5），README 说要 \"download the matching version editor-ui.tar.gz\"，而仓库已经归档，最后一个发行版发布于 2026-08-07。任何比 2.33.7 更新的 n8n 都没有对应的安装包，混用版本就意味着把一个 n8n 版本的界面换成另一个版本的界面。","source":"ai","title":"每个安装包只适配一个 n8n 版本"},{"code":"the_translation_is_produced_by_a_script","severity":"info","evidence":"package.json 里只有一个脚本，`\"i18n:translate\": \"node ./script/translate.js\"`，依赖里包含 `dotenv` 和 `qs`，这说明翻译是由 .env 文件配置的自动化流程完成的，而不是人工检查过的文字。字符串都放在 `languages/` 里，README 解释的方法是：把 `zh-CN.json` 加到编辑器的 locales 文件夹里，然后重新构建。","source":"ai","title":"翻译内容是脚本生成的"},{"code":"readme_carries_a_sponsor_advertisement","severity":"info","evidence":"README 的最后一节标题是 广告（\"advertisement\"），说项目的 CDN 和防护由腾讯云 EdgeOne 赞助，并附有一个带追踪参数的链接（edgeone.ai/zh?from=github），邀请读者领取免费套餐。","source":"ai","title":"README 里带有赞助商广告"},{"code":"documentation_only_in_chinese","severity":"info","evidence":"仓库的描述是 \"n8n 配置多语言\"，整篇 README（包括安装步骤）都是中文，没有英文版本。不懂中文的人得先把说明翻译出来才能照着做。","source":"ai","title":"文档只有中文"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["automation"],"alternativeTo":[{"slug":"n8n-cloud","name":"Cloud","confirmed":true},{"slug":"zapier","name":"Zapier","confirmed":true}],"declaredAlternatives":[],"stars":2169,"downloads":68504,"downloadsWeek":58,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":2173},{"date":"2026-11-08","stars":2173},{"date":"2026-11-09","stars":2172},{"date":"2026-11-10","stars":2169}],"firstPublishedAt":"2026-11-08T23:01:18.909Z","commitSha":"794087808ddd942c068ba93cc6012d5f76dace02","verifiedAt":"2026-11-08T22:49:28.030Z","lastCommitAt":"2026-08-21T17:31:03Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes the n8n automation editor speak Chinese, through a ready Docker image or by swapping its interface files.","summary":"n8n, the tool for connecting apps into automatic workflows, ships without a Chinese interface. This project supplies one. The short route is the author's ready-made Docker image (blowsnow/n8n-chinese), which is n8n with the Chinese files already inside; the longer route is to download the translated editor files for your exact n8n version, mount them over n8n's own and set N8N_DEFAULT_LOCALE=zh-CN. It is written for Chinese-speaking n8n users and the whole README is in Chinese. The repository is archived and the last pack matches n8n 2.33.7.","easeRationale":"All three systems are D because every route goes through Docker and the terminal: the shortest is one long `docker run -it --rm --name n8ntest -p 15678:5678 -v ~/.n8n:/home/node/.n8n -e N8N_SECURE_COOKIE=false blowsnow/n8n-chinese`, and the alternative is mounting a downloaded folder over a path inside the n8n container. The third route, for n8n started with npx, means finding a folder such as `C:\\Users\\xxxxxx\\AppData\\Local\\npm-cache\\_npx\\n8n\\node_modules\\n8n-editor-ui\\dist`, unpacking a tar.gz over it and setting an environment variable, which the README itself tells you to ask an AI how to do.","costNote":"Nothing is charged, and no license file says what you may do with the files.","notVerified":["How complete the Chinese translation is, since no coverage figure is given and the text is generated by a script","What exactly is inside the blowsnow/n8n-chinese image compared with the official n8n image of the same version","Whether the trick still works on n8n versions released after 2.33.7, as the repository is archived","Real disk use, estimated by the checks at 100 to 500 MB from a heuristic, not measured","Whether replacing files inside n8n breaks anything else in the editor"]},"zh-Hans":{"summaryShort":"让 n8n 自动化编辑器显示中文，可以用现成的 Docker 镜像，也可以替换它的界面文件。","summary":"n8n 是把各种应用连接成自动化工作流的工具，但自带的界面没有中文。这个项目补上了这一块。最简单的方式是作者做好的 Docker 镜像（blowsnow/n8n-chinese），它就是已经内置中文文件的 n8n；更长的方式是下载与你的 n8n 版本完全对应的翻译后编辑器文件，覆盖安装到 n8n 自己的文件上，并设置 N8N_DEFAULT_LOCALE=zh-CN。这个项目是为中文用户写的，README 全文都是中文。仓库已归档，最后一个安装包对应 n8n 2.33.7。","easeRationale":"三个系统都是 D 级，因为每条路径都要经过 Docker 和终端：最短的是一条很长的命令 `docker run -it --rm --name n8ntest -p 15678:5678 -v ~/.n8n:/home/node/.n8n -e N8N_SECURE_COOKIE=false blowsnow/n8n-chinese`，另一种做法是把下载的文件夹挂载到 n8n 容器内的某个路径上。第三条路径针对用 npx 启动的 n8n，需要找到类似 `C:\\Users\\xxxxxx\\AppData\\Local\\npm-cache\\_npx\\n8n\\node_modules\\n8n-editor-ui\\dist` 的文件夹，把一个 tar.gz 解压覆盖到上面，再设置一个环境变量，README 自己也建议你去问 AI 该怎么做。","costNote":"不收取任何费用，但没有许可证文件说明你可以如何使用这些文件。","notVerified":["中文翻译到底有多完整，因为没有给出覆盖率数据，而且文字是脚本生成的","blowsnow/n8n-chinese 镜像里到底有什么内容，和同版本的官方 n8n 镜像相比有什么差别","由于仓库已归档，这个办法在 2.33.7 之后发布的 n8n 版本上是否还能用","实际占用的磁盘空间，检查只是用启发式方法估算在 100 到 500 MB 之间，并没有实测","替换 n8n 内部文件是否会破坏编辑器里的其他功能"]}}},{"id":"R_kgDOQy5Vlw","slug":"othmanadi__planning-with-files","name":"Planning with Files","owner":"OthmanAdi","fullName":"OthmanAdi/planning-with-files","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"installs_scripts_that_run_by_themselves_in_every_session","severity":"attention","evidence":"快速安装方式是 `npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g`，直接从 GitHub 仓库做全局安装，项目对自己的描述是 \"Not a prompt it might follow. A hook that fires every turn\"。仓库里有 `hooks/` 和 `scripts/` 文件夹，外加 17 个助手的插件文件夹，所以执行完一条命令之后，这些脚本就会自动在你助手的会话里运行，处理你的文件。","source":"ai","title":"装上之后每个会话都会自己运行脚本"},{"code":"the_benchmark_numbers_are_the_authors_own","severity":"info","evidence":"\"benchmark-96.7% pass (29/30)\" 和 \"blind A/B 3/3 wins\" 这两个徽章，以及计划 \"turned a 13.3-turn re-orientation into 5.0 turns\"、\"One hook fire measures 289ms wall-clock\" 这些说法，全部都来自项目自己的 docs/evals.md。README 本身也只是指向 \"numbers and limits\"，并没有引用任何独立测试。","source":"ai","title":"基准测试的数据来自作者自己"},{"code":"readme_promotes_a_paid_sponsor_to_connect_to_your_agent","severity":"info","evidence":"README 靠前的位置有一个 Sponsor 区块，宣传 \"Scrape.ly by evomi\"，带有追踪链接（scrape.ly/?utm_source=github&utm_medium=readme&utm_campaign=planning-with-files），邀请你 \"Connect the Scrape.ly MCP server to Claude Code, Cursor or any MCP client\"。另外还有一个给 CodeRabbit 的 \"OSS code review support\" 区块。这些推荐和安装步骤写在同一份 README 里。","source":"ai","title":"README 里推广一个付费赞助商来接入你的智能体"},{"code":"the_plan_files_are_thrown_away_by_design","severity":"info","evidence":"FAQ: \"They are working memory, not a tracked deliverable. `task_plan.md`, `findings.md`, `progress.md`, and the `.planning/` directory are gitignored by default and are not archived automatically: the next task overwrites the root plan.\" 想保留的笔记需要手动挪到别处，或者用 `set-active-plan.sh --archive <slug> --yes` 来归档。","source":"ai","title":"计划文件设计上就是用完即丢"},{"code":"what_you_get_depends_on_your_assistant","severity":"info","evidence":"README 写道 \"Commands and model-callable tools depend on the host; Qoder CLI loads the skill and lifecycle hooks without the Claude-specific command prompts\"，列出了 13 个带原生插件的宿主，并通过 Agent Skills 文件标准达到 \"60+\" 和 \"70+ others\" 这些数字，\"the `npx skills` installer alone targets 71 agents\"。其中只有一部分是经过测试的集成。","source":"ai","title":"你能用到的功能取决于你用的助手"},{"code":"very_fast_release_pace","severity":"info","evidence":"仓库创建于 2026-01-03，已经有 127 个发行版，最新的 v3.28.1 发布于 2026-11-07（本次检查前一天），README 里记录了在 v3.6.0、v3.7.0、v3.10.0、v3.11.0 和 v2.36.0 中发生变化的行为。27953 个星标和 17 个未关闭的 issue 说明这个项目很活跃，但一个会全局重装自己的 skill 变动得也很频繁。","source":"ai","title":"发布节奏非常快"},{"code":"project_site_on_a_generated_address","severity":"info","evidence":"仓库的主页字段是 https://gilded-laurel-c7nj.here.now/，这是一个看起来随机生成的托管子域名，而不是属于项目自己的名字，安装次数则是通过第三方徽章服务（skills.sh、skillsplayground.com、skill-history.com）展示出来的。","source":"ai","title":"项目网站用的是自动生成的地址"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","productivity"],"alternativeTo":[{"slug":"manus","name":"Manus","confirmed":true}],"declaredAlternatives":[],"stars":27989,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":105,"growthDays":6,"history":[{"date":"2026-11-06","stars":27925},{"date":"2026-11-07","stars":27940},{"date":"2026-11-08","stars":27953},{"date":"2026-11-09","stars":27971},{"date":"2026-11-10","stars":27989}],"firstPublishedAt":"2026-11-08T23:01:18.917Z","commitSha":"2362ca385ad74f061aa0e7b73f56543a4b7e6df2","verifiedAt":"2026-11-08T22:50:38.306Z","lastCommitAt":"2026-11-10T18:28:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Keeps your AI coding assistant's plan in three files on disk, so the work survives a reset, a crash or /clear.","summary":"Planning with Files is a skill you add to an AI coding assistant. It keeps the plan for the current job in three plain text files in your project (task_plan.md, findings.md, progress.md) and feeds them back to the assistant at the start of every turn, so the work carries on after /clear, a crash or a context reset. It can also hold the assistant back from declaring the job finished until the plan says it is. One command installs it in Claude Code, Codex, Cursor, Copilot and dozens of other assistants. The skill is free; the assistant usually is not.","easeRationale":"C on all three systems: you must already have an AI coding assistant installed, then run one command, `npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g`, which needs Node.js. The moving parts are shell scripts (the repository's main language is Shell and the completion gate is `check-complete.sh` matching with `grep -F`), so on Windows they expect a Bash environment; the Qoder route is documented as needing \"Bash and Python 3.10+\".","costNote":"The skill is free under the MIT license. It only does anything inside an AI coding assistant such as Claude Code, Codex or Cursor, and those usually need a paid plan or paid API usage.","notVerified":["Whether the plan really survives /clear and compaction in your assistant, since every host integrates differently","Whether the hooks slow your sessions down in practice, as the 289ms figure is the authors' own measurement","What the global install writes into your home folder and which hook files each assistant ends up running","Whether the 60+ supported agents all behave as described, which the README itself qualifies","How much the extra text injected every turn adds to the cost of a paid assistant"]},"zh-Hans":{"summaryShort":"把 AI 编程助手的计划保存在磁盘上的三个文件里，这样工作内容能在重置、崩溃或 /clear 之后继续保留。","summary":"Planning with Files 是一个加到 AI 编程助手上的 skill。它把当前工作的计划保存在项目里三个纯文本文件中（task_plan.md、findings.md、progress.md），并在每一轮开始时把内容交还给助手，这样工作在 /clear、崩溃或上下文重置之后还能继续。它还能在计划没有标记完成之前，拦住助手不让它宣布任务结束。一条命令就能把它装进 Claude Code、Codex、Cursor、Copilot 和几十个其他助手里。这个 skill 是免费的，但助手本身通常不是。","easeRationale":"三个系统都是 C 级：你本来就要先装好一个 AI 编程助手，然后运行一条命令，`npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g`，这需要 Node.js。真正干活的部分是 shell 脚本（这个仓库的主要语言是 Shell，完成的判定由 `check-complete.sh` 用 `grep -F` 做匹配），所以在 Windows 上它们期望有一个 Bash 环境；Qoder 这条路径的文档里写明需要 \"Bash and Python 3.10+\"。","costNote":"这个 skill 在 MIT 许可证下是免费的。它只能在像 Claude Code、Codex 或 Cursor 这样的 AI 编程助手里面起作用，而这些助手通常需要付费套餐或付费的 API 用量。","notVerified":["计划是否真的能在你的助手里撑过 /clear 和压缩，因为每个宿主的集成方式都不一样","这些 hook 在实际使用中是否会拖慢你的会话速度，因为 289 毫秒这个数字是作者自己测的","全局安装会往你的主目录写入什么内容，每个助手最终又会运行哪些 hook 文件","README 自己都留有余地的那 60 多个受支持的智能体是否真的都像描述的那样表现","每一轮都注入的额外文字会给一个付费助手增加多少成本"]}}},{"id":"R_kgDORXS4VA","slug":"panniantong__agent-reach","name":"Agent Reach","owner":"Panniantong","fullName":"Panniantong/Agent-Reach","type":"dev_tool","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"reads_logged_in_pages","severity":"high","evidence":"README 的平台表格把 xiaohongshu、boss、reddit、twitter、facebook、instagram 和 linkedin 标为需要 \"登录\"，并且 README 写道 \"第一次用某个网站时，会把你 Chrome 里这个网站的登录复制一份过去\"。也就是说，这个工具用你自己的会话读取登录后的页面，而不是使用各平台的官方 API。这些平台的使用条款限制对登录后页面的自动化访问，README 自己也警告 \"建议用小号登录。复用过来的通常是主账号，而平台可能限制自动访问的账号。\"，并把 twitter、facebook 和 instagram 标为 \"实验\"：\"能用，但平台反爬更严，可能随时失效\"。","source":"ai","title":"读取需要登录的页面"},{"code":"copies_your_browser_login","severity":"attention","evidence":"README：\"第一次用某个网站时，会把你 Chrome 里这个网站的登录复制一份过去\"。也就是说，第一次使用某个网站时，它会把你 Chrome 中该网站的登录状态复制到自己的浏览器配置文件里。它说自己不读取你的密码库，也不动原来的浏览器，但确实会为 Xiaohongshu、Reddit、Twitter、Facebook、Instagram、LinkedIn 和 Boss 等网站取得一个可用的会话。","source":"ai","title":"复制你的浏览器登录状态"},{"code":"install_by_letting_ai_run_remote_instructions","severity":"attention","evidence":"快速上手的方法是告诉你的助手：\"帮我安装 Agent Reach: https://raw.githubusercontent.com/.../docs/install.md\"，并打开 \"允许它执行命令\"。实际执行的步骤来自当时才下载的一个文件，而可选渠道 \"需要的软件会自动装好\"，也就是会自行安装额外软件，包括 README 中注明来源的第三方命令行工具（bili-cli、boss-agent-cli、OpenCLI）。","source":"ai","title":"让 AI 执行远程指令来安装"},{"code":"same_name_package_elsewhere_is_not_this_project","severity":"attention","evidence":"README 警告：\"PyPI 上叫 `agent-reach` 的包不是本项目，请勿从 PyPI 安装\"。也就是说，PyPI 上名为 `agent-reach` 的包不是这个项目，不能安装。按常规方式搜索的人可能会装错东西。","source":"ai","title":"别处的同名软件包并不是本项目"},{"code":"very_fast_star_growth","severity":"info","evidence":"一个创建于 2026-02-24、只有 9 个版本发布的仓库，已有 107,663 个星标和 9,600 个 fork，并带有 Trendshift \"#1 Repository of the Day\" 徽章。README 还有一个 \"防骗提醒\" 部分，说明该项目没有代币、币或钱包，使用其名称的加密货币项目与它无关。","source":"ai","title":"星标增长极快"},{"code":"paid_sponsors_in_readme","severity":"info","evidence":"README 的赞助商表格带有付费代理和网页抓取服务的推荐链接（Bright Data、RapidProxy、BrowserAct、CoreClaw、AstraFlow），而项目描述却承诺 \"zero API fees\"。","source":"ai","title":"README 里有付费赞助商"},{"code":"depends_on_sites_that_change","severity":"info","evidence":"README 把 twitter、facebook 和 instagram 标为 \"实验\"（实验性）：\"能用，但平台反爬更严，可能随时失效\"。也就是说，这些功能能用，但这些网站对自动化访问的限制更严，功能可能随时失效。","source":"ai","title":"依赖会变化的网站"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["Groq","OpenAI"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","automation","development"],"alternativeTo":[{"slug":"bright-data","name":"Bright Data","confirmed":true},{"slug":"x-twitter-api","name":"X (Twitter) API","confirmed":true},{"slug":"browserbase","name":"Browserbase","confirmed":true}],"declaredAlternatives":[],"stars":110893,"downloads":96283,"downloadsWeek":31896,"downloadsDays":7,"growth":2457,"growthDays":7,"history":[{"date":"2026-10-27","stars":104929},{"date":"2026-10-28","stars":105636},{"date":"2026-10-29","stars":106301},{"date":"2026-10-30","stars":106830},{"date":"2026-10-31","stars":107229},{"date":"2026-11-01","stars":107642},{"date":"2026-11-02","stars":108092},{"date":"2026-11-03","stars":108436},{"date":"2026-11-04","stars":108773},{"date":"2026-11-05","stars":109175},{"date":"2026-11-06","stars":109652},{"date":"2026-11-07","stars":109974},{"date":"2026-11-08","stars":110273},{"date":"2026-11-09","stars":110596},{"date":"2026-11-10","stars":110893}],"firstPublishedAt":"2026-10-27T18:40:28.010Z","commitSha":"7a38047acf26a729cbb27851e3cb81639da86b29","verifiedAt":"2026-11-02T03:06:01.185Z","lastCommitAt":"2026-11-08T02:17:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A command-line tool that lets your AI assistant read and search X, Reddit, YouTube, GitHub and Chinese sites like Bilibili.","summary":"Agent Reach is a single command you install on your computer so an AI assistant such as Claude Code can look things up on sites it normally cannot reach. It searches and reads posts, profiles, job ads, video subtitles and stock quotes, and hands the result back as tidy data. For sites that need an account it runs its own background browser and copies the login for that one site out of your Chrome. It only reads; it never posts. You install it from the terminal and the main documentation is in Chinese.","easeRationale":"Same on all three systems: there is no installer or ready binary, so you install the `uv` helper and run the install steps from the terminal (the README suggests pasting a link and letting your AI assistant run the commands for you). Afterwards each site that needs an account has its own sign-in step, such as scanning a QR code for Bilibili subtitles.","costNote":"The tool is free and MIT-licensed. Turning speech into text on YouTube or the Xiaoyuzhou podcast site needs a Groq or OpenAI key, which those services charge for.","notVerified":["Whether each of the 14 sites still works today, since the project itself says features break when a site changes","Exactly what is copied out of your Chrome profile and what the private browser keeps on disk","Whether using it gets your account on any of those sites limited or closed","What the install file and the per-channel third-party tools do at the moment they are downloaded and run","Real disk use once the private browser and each channel's tools are installed","The real cost of speech-to-text, which depends on your Groq or OpenAI usage"]},"zh-Hans":{"summaryShort":"一个命令行工具,让你的AI助手能读取和搜索X、Reddit、YouTube、GitHub以及B站等中文网站。","summary":"Agent Reach 是一条安装在你电脑上的命令,让像Claude Code这样的AI助手能够查找它通常无法访问的网站上的信息。它可以搜索和读取帖子、个人资料、招聘信息、视频字幕和股票报价,并把结果整理成结构化数据返回。对于需要账号的网站,它会运行自己的后台浏览器,并从你的Chrome中复制该网站的登录信息。它只读取,从不发帖。你需要在终端里安装它,主要文档是中文的。","easeRationale":"三个系统上情况相同:没有安装程序或现成的二进制文件,所以你需要安装`uv`工具并在终端里执行安装步骤(README建议复制一个链接,让你的AI助手替你运行这些命令)。之后,每个需要账号的网站都有各自的登录步骤,比如扫码登录以获取B站字幕。","costNote":"该工具免费,采用MIT许可。要在YouTube或小宇宙播客网站上把语音转成文字,需要Groq或OpenAI的密钥,这些服务是收费的。","notVerified":["这14个网站是否每一个今天仍然能用,因为项目方自己也说网站一改版功能就会失效","究竟从你的Chrome账户中复制了什么,以及这个私有浏览器在磁盘上保留了什么","使用它是否会导致你在这些网站上的账号被限制或封禁","安装文件和各个渠道的第三方工具在下载和运行的那一刻会做什么","安装好私有浏览器和各渠道工具后的真实磁盘占用","语音转文字的真实费用,这取决于你的Groq或OpenAI用量"]}}},{"id":"R_kgDOVY7csg","slug":"patch-gg__anymove","name":"AnyMove","owner":"patch-gg","fullName":"patch-gg/anymove","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 23 天前","source":"check","title":"所有者账号很新"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖项中存在已知漏洞（OSV）：torch：GHSA-rrmf-rvhw-rf47；pillow：GHSA-22hg-r758-rrw9、GHSA-24rc-8hj2-v98c、GHSA-254f-3936-288h","source":"check","title":"存在已知漏洞"},{"code":"licence_of_what_you_make_is_not_only_apache","severity":"attention","evidence":"README 的许可证部分：\"MiniMax H3's open weights carry a community licence and hosted use on fal goes through fal's terms: check both before shipping what you make with them\"，已发布的狼 LoRA \"is a LoRA on MiniMax H3, so MiniMax H3's community licence applies to it, not Apache-2.0.\" Patch 的名称和标志也不在该许可证的授权范围内。","source":"ai","title":"你创作内容的许可证不只是 Apache"},{"code":"downloads_models_and_example_files_on_first_use","severity":"info","evidence":"requirements.txt：DINOv2 主干网络 `facebook/dinov2-with-registers-base` 标注为 \"downloaded on first use\"；`tools/fetch_examples.sh` 会拉取发布版本中的文件（\"13 MB mesh, 355 MB reader, 4 MB clip\"），PyTorch 则来自 `--index-url https://download.pytorch.org/whl/cu128`。","source":"ai","title":"首次使用时会下载模型和示例文件"},{"code":"author_lists_cases_that_do_not_work","severity":"info","evidence":"README 的 \"limits.\" 部分：八条腿的行走不适用（\"The two views disagree about which legs are up (0.36-0.49)\"），躺下的动作会失败，\"Video models drift\"（未经处理的 MiniMax H3 在 5 个狼的片段中有 2 个把四肢颜色弄混了），而且地面接触项只认识狼和龙风格的爪子名称。","source":"ai","title":"作者列出了行不通的情况"},{"code":"repository_is_days_old","severity":"info","evidence":"创建于 2026-10-27，只有一个发布版本（`examples-v1`，同一天发布），已有 359 个星标；自动检查也标出所有者账号只有 23 天。","source":"ai","title":"仓库只有几天历史"}],"platforms":["linux"],"tools":["python","gpu_nvidia","other"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["fal.ai"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["design_cad","games","ai_skills"],"alternativeTo":[{"slug":"move-ai","name":"Move AI","confirmed":true},{"slug":"deepmotion","name":"DeepMotion","confirmed":true}],"declaredAlternatives":[],"stars":570,"downloads":1726,"downloadsWeek":372,"downloadsDays":7,"growth":73,"growthDays":7,"history":[{"date":"2026-10-29","stars":419},{"date":"2026-10-30","stars":441},{"date":"2026-10-31","stars":448},{"date":"2026-11-01","stars":460},{"date":"2026-11-02","stars":488},{"date":"2026-11-03","stars":497},{"date":"2026-11-04","stars":505},{"date":"2026-11-05","stars":542},{"date":"2026-11-06","stars":544},{"date":"2026-11-07","stars":555},{"date":"2026-11-08","stars":560},{"date":"2026-11-09","stars":563},{"date":"2026-11-10","stars":570}],"firstPublishedAt":"2026-10-29T01:54:21.438Z","commitSha":"7a82a0a8fadffddd5c24fabf0697d9b6c814ce6e","verifiedAt":"2026-10-29T01:41:57.185Z","lastCommitAt":"2026-11-09T09:48:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns one AI-made video of a creature into a 3D animation file you can load into a game engine.","summary":"A set of scripts and Claude Code skills that make animations for rigged 3D creatures (wolf, dragon, spider, demon) without a motion-capture suit. A video model films the creature from the front and the side in one frame, a vision model trained on that single creature finds its joints in both views, and the pair of views places each joint in 3D so the rig can be fitted and exported as a GLB. For technical 3D and game people: you need Linux-style commands, Python 3.11, Blender 5.x, an NVIDIA graphics card with CUDA (24 GB to train) and a paid fal.ai key for the clips.","easeRationale":"D on Linux, the only system the instructions fit: you clone the repository, build a virtualenv with Python 3.11, install a CUDA build of PyTorch, put Blender 5.x on your PATH, add a fal.ai key and run a chain of `tools/*.sh` shell scripts; training the joint reader wants an NVIDIA card with 24 GB. Skeletons are still placed by hand on grid views. One smaller step works without a graphics card: the README shows fitting the example dragon's roar on the CPU.","costNote":"The code is Apache-2.0. The paid parts are listed in the README (\"Prices on fal as of October 2026\"): about $0.26 for a 5-second two-camera clip, $0.40 for a mesh from Tripo, and about $12 for training the optional MiniMax H3 LoRA. A new move for an existing creature is about $0.52.","notVerified":["Whether it runs on Windows (with WSL) or on a Mac; the commands and shell scripts assume a Linux-style system and an NVIDIA card.","Real disk use after PyTorch, the DINOv2 weights and the example files land on your machine.","Whether the animation quality matches the README's figures on your own creature and your own mesh.","What the whole process costs on fal for a project with several moves, and whether the October 2026 prices still hold.","How much hand work placing the skeleton takes for a creature that is not close to the four examples."]},"zh-Hans":{"summaryShort":"把一段AI生成的生物视频转换成可以导入游戏引擎的3D动画文件。","summary":"一套脚本和Claude Code技能,用于为带骨架的3D生物(狼、龙、蜘蛛、恶魔)制作动画,不需要动作捕捉服。一个视频模型在一帧画面里从正面和侧面拍摄该生物,一个针对这只生物单独训练的视觉模型在两个视角中找出关节位置,再由这两个视角把每个关节定位到3D空间,这样骨架就能被适配并导出为GLB文件。适合3D和游戏技术人员:你需要类Linux命令行、Python 3.11、Blender 5.x、带CUDA的NVIDIA显卡(训练需要24GB)以及一个用于生成片段的付费fal.ai密钥。","easeRationale":"D,仅适用于Linux这一系统,说明书也只针对它:你需要克隆仓库、用Python 3.11搭建虚拟环境、安装带CUDA的PyTorch版本、把Blender 5.x加入PATH、添加fal.ai密钥,然后依次运行一串`tools/*.sh`脚本;训练关节识别模型需要一块24GB显存的NVIDIA显卡。骨架仍需在网格视图上手动摆放。有一个较小的步骤不需要显卡:README展示了如何在CPU上适配示例龙吼叫的动作。","costNote":"代码采用Apache-2.0许可。付费部分在README中列出(\"截至2026年10月fal上的价格\"):一个5秒双镜头片段约0.26美元,一个来自Tripo的网格约0.40美元,训练可选的MiniMax H3 LoRA约12美元。为已有生物添加一个新动作约0.52美元。","notVerified":["能否在Windows(配合WSL)或Mac上运行;这些命令和shell脚本都假定使用类Linux系统和NVIDIA显卡。","PyTorch、DINOv2权重和示例文件装进机器后,实际占用的磁盘空间。","在你自己的生物和网格上,动画质量是否能达到README中给出的数字。","对于有多个动作的项目,整个流程在fal上要花多少钱,以及2026年10月的价格是否仍然有效。","对于与四个示例差别较大的生物,手动摆放骨架需要多少工作量。"]}}},{"id":"R_kgDOVcsWsA","slug":"patch-gg__pixelfx","name":"PixelFX","owner":"patch-gg","fullName":"patch-gg/pixelfx","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"10 天内获得 6312 个星标","source":"check","title":"年轻的仓库却有很多星标"},{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 35 天前","source":"check","title":"账号年龄很短的所有者"},{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道传给 shell：curl -fsSL https://patch.gg/install.sh | sh","source":"check","title":"用 curl 管道到 shell 来安装"},{"code":"repository_very_new","severity":"attention","evidence":"这个仓库创建于 2026-10-30，检查记录为 \"6312 stars in 10 days\"。目前完全没有 GitHub 发行版，package.json 中写着 \"version\": \"0.3.1\"，低于 1.0。代码仍然每天都在改动（检查显示最近一次提交是在 2026-11-09），所以这里的一切都还没有经过其他人长时间的使用。","source":"ai","title":"仓库非常新"},{"code":"font_outside_the_mit_license","severity":"info","evidence":"README：\"the Patch Pixel font in `gallery/fonts` is Patch's own and not covered: see NOTICE.md\"。特效和代码是 MIT 许可，这款字体不是。","source":"ai","title":"一款字体不在 MIT 许可证范围内"},{"code":"adapts_code_from_another_author","severity":"info","evidence":"README：\"`src/core/sim.js` and the effects that say so at the top adapt his code\"，改编自 Hugo Duprez 在 Sprite Fusion 中的 MIT 特效，许可说明部分还写道 \"Parts are adapted from Hugo Duprez's MIT effects and from Hollow Crown's MIT source\"。LICENSE 中有两行版权声明（\"Copyright (c) 2026 Patch\"、\"Copyright (c) 2026 Hugo Duprez\"），并指出 NOTICE.md 是详细说明的地方。","source":"ai","title":"改编了另一位作者的代码"},{"code":"installs_a_skill_that_runs_commands","severity":"info","evidence":"README：先执行 `/plugin marketplace add patch-gg/pixelfx`，再执行 `/plugin install pixelfx@pixelfx`，或者把 `skills/pixelfx` 复制到 `~/.claude/skills/`。安装后，这个 skill 会让 Claude Code 在你的电脑上运行仓库自带的脚本（`tools/new.mjs`、`tools/render.mjs`、`tools/check.mjs`）。","source":"ai","title":"安装一个会执行命令的 skill"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games","image","ai_skills"],"alternativeTo":[{"slug":"aseprite","name":"Aseprite","confirmed":true}],"declaredAlternatives":[],"stars":6640,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":2530,"growthDays":7,"history":[{"date":"2026-10-31","stars":1421},{"date":"2026-11-01","stars":2675},{"date":"2026-11-02","stars":3507},{"date":"2026-11-03","stars":4110},{"date":"2026-11-04","stars":4660},{"date":"2026-11-05","stars":5237},{"date":"2026-11-06","stars":5581},{"date":"2026-11-07","stars":5812},{"date":"2026-11-08","stars":6063},{"date":"2026-11-09","stars":6302},{"date":"2026-11-10","stars":6640}],"firstPublishedAt":"2026-10-31T01:54:33.085Z","commitSha":"044e12a45dc9cc312cec50ee606dfc98afe5adc2","verifiedAt":"2026-11-09T23:06:01.208Z","lastCommitAt":"2026-11-09T09:48:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free collection of 244 pixel art effects for games — fire, explosions, magic, weather — ready to drop in or export.","summary":"pixelfx is a set of 244 pixel art animations (fire, explosions, magic, weather, hits, pickups) written as small pieces of JavaScript instead of drawings. The 61 effects with a character and the 12 \"moves\" act the scene out: a wind-up, a freeze on contact, a recovery, and a target that rocks back. A browser game can import them directly; for Godot, Unity or LÖVE one command writes a sprite sheet PNG plus a JSON description, and any effect takes your own palette. It also ships a Claude Code skill to find, export or invent effects. You need Node.js 18 or newer.","easeRationale":"The same on all three systems: install Node.js 18+, fetch the code with `npm i github:patch-gg/pixelfx` or copy the files, then run commands such as `node tools/render.mjs ice/frost-nova --sheet` to get a sprite sheet. The project's own gallery page shows every effect playing in a browser, but to put one in a game you write code or run the export command.","costNote":"MIT, including commercial use; the Patch Pixel font in gallery/fonts is excluded (NOTICE.md).","notVerified":["Whether the 244 effects really look as good as the montage in the README","Whether the acted moves line up with your own character animations","Whether the exported sprite sheets load without changes in Godot, Unity or Phaser","How much disk the full export to `sheets/` takes, since every effect becomes a PNG","Whether the Claude Code skill produces usable new effects, which also depends on your Claude subscription"]},"zh-Hans":{"summaryShort":"一套免费的 244 个游戏像素特效合集——火焰、爆炸、魔法、天气——可以直接接入,也可以导出使用。","summary":"pixelfx 是一套 244 个像素艺术动画(火焰、爆炸、魔法、天气、打击、拾取物),它们不是画出来的,而是用一小段 JavaScript 写成的。其中 61 个带角色的特效和 12 个\"moves\"会演出完整的动作:预备动作、接触时的定帧、恢复动作,以及目标被击退的后仰。浏览器游戏可以直接导入它们;对于 Godot、Unity 或 LÖVE,一条命令就能生成一张精灵表 PNG 加一份 JSON 描述,并且任何特效都可以套用你自己的调色板。它还附带一个 Claude Code skill,用来查找、导出或创造新特效。你需要 Node.js 18 或更高版本。","easeRationale":"三个系统上都一样:安装 Node.js 18+,用 `npm i github:patch-gg/pixelfx` 获取代码,或者直接复制文件,然后运行类似 `node tools/render.mjs ice/frost-nova --sheet` 的命令来得到一张精灵表。项目自己的画廊页面能在浏览器里播放每个特效,但要把特效用进游戏里,还是得写代码或运行导出命令。","costNote":"MIT 许可,包括商业使用;gallery/fonts 中的 Patch Pixel 字体不在此范围内(见 NOTICE.md)。","notVerified":["这 244 个特效实际效果是否真有 README 里的合集展示那么好","演出式的 moves 是否能和你自己的角色动画对得上","导出的精灵表能否在 Godot、Unity 或 Phaser 中原样加载","全部导出到 `sheets/` 要占多少磁盘空间,因为每个特效都会变成一张 PNG","Claude Code skill 生成的新特效是否真的可用,这还取决于你的 Claude 订阅"]}}},{"id":"R_kgDOQWgufQ","slug":"pbakaus__impeccable","name":"Impeccable","owner":"pbakaus","fullName":"pbakaus/impeccable","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"hooks_run_outside_model_approval","severity":"attention","evidence":"README.md 的 \"Good to know\"：\"Hooks run outside model approval. In Claude Code, installed command hooks run even when you deny the model's own commands, so the first edit can download the engine. Review installed hooks before unattended runs, or pass `--settings '{\"disableAllHooks\": true}'` to turn them off for a run.\"","source":"ai","title":"钩子（hooks）不经过模型审批就会运行"},{"code":"downloads_a_program_at_run_time","severity":"attention","evidence":"README.md 的 Installation：\"The skill itself runs a self-contained engine binary that ships next to it or downloads once into `~/.impeccable/bin/`.\" package.json 中附带了 `skill/scripts/engine-checksums.txt`，用于这次下载。","source":"ai","title":"使用过程中会下载程序"},{"code":"contacts_the_vendor_even_with_telemetry_off","severity":"attention","evidence":"README.md 的 \"Telemetry\"：\"When a new design starts, the skill fetches a set of visual directions from the impeccable.style catalog. Turning telemetry off doesn't stop that request; it stops the anonymous record of which directions were shown and which one you picked.\" 它还补充说 \"Your project name, files, code, and `PRODUCT.md` are never sent\"，并且可以用 `npx impeccable telemetry off`、`DO_NOT_TRACK=1` 或 `\"telemetry\": false` 关闭这项记录。","source":"ai","title":"即使关闭遥测，也会联系开发方"},{"code":"writes_into_your_coding_tools_settings","severity":"info","evidence":"README.md 的 Installation：\"The installer detects your coding tools, asks whether to install for this project or globally, and offers the design hook where your tool supports one.\" 仓库中为每个受支持的工具都带有一个设置文件夹（.claude/、.cursor/、.codex/、.github/、.gemini/、.grok/、.hermes/ 等）。","source":"ai","title":"会写入你的编程工具的设置"},{"code":"runs_a_project_script_in_your_shell","severity":"info","evidence":"README.md 的 Live mode：\"Applying copy edits runs your `package.json`'s optional `impeccable:manual-edit-validate` script with your permissions, so review it before using live mode in an unfamiliar repo.\"","source":"ai","title":"会在你的 shell 中运行项目脚本"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","rust"],"disk":{"minMB":19,"maxMB":58,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","design_cad"],"alternativeTo":[],"declaredAlternatives":[],"stars":107516,"downloads":4364234,"downloadsWeek":569347,"downloadsDays":7,"growth":5998,"growthDays":7,"history":[{"date":"2026-10-27","stars":90538},{"date":"2026-10-28","stars":92686},{"date":"2026-10-29","stars":95340},{"date":"2026-10-30","stars":97074},{"date":"2026-10-31","stars":98291},{"date":"2026-11-01","stars":99417},{"date":"2026-11-02","stars":100485},{"date":"2026-11-03","stars":101518},{"date":"2026-11-04","stars":102666},{"date":"2026-11-05","stars":103802},{"date":"2026-11-06","stars":104723},{"date":"2026-11-07","stars":105430},{"date":"2026-11-08","stars":106130},{"date":"2026-11-09","stars":106800},{"date":"2026-11-10","stars":107516}],"firstPublishedAt":"2026-10-27T18:40:28.019Z","commitSha":"5e745ace7e151feb2a93d360d0281fe39448b857","verifiedAt":"2026-10-29T01:43:03.696Z","lastCommitAt":"2026-11-10T21:06:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds design commands to AI coding tools, plus a checker that flags ugly or broken web pages.","summary":"Impeccable is a pack of design instructions and 24 commands you install inside AI coding assistants such as Claude Code, Cursor, Codex and Copilot. After installing you type things like /impeccable audit or /impeccable polish and the assistant reviews or improves the look of your web pages. It also ships a command-line checker with 62 fixed rules that scans files or a website address with no AI and no key, and a Chrome extension that shows the same findings on any page. It is for people who already build websites with an AI assistant; you need Node installed to run the installer.","easeRationale":"On every system it is the same work: Node 18+ (package.json \"engines\": {\"node\": \">=18.19.0\"}), then `npx impeccable install` in a terminal, then `/impeccable init` inside your AI coding tool before anything happens. There is no installer in the releases, only the engine binaries the skill uses internally.","costNote":"Apache-2.0. The README says the detector needs \"No LLM, no API key\"; the design commands run inside an AI assistant you pay for separately.","notVerified":["Whether the installer and the downloaded engine really work on a normal Windows setup","What the impeccable.style catalog request actually sends and stores, beyond what the README describes","Whether the engine downloaded on first run matches the checksums shipped with the skill","How good the design results are in practice, since the 62 detector rules only check fixed patterns","Whether the hook slows down your coding tool on every file edit"]},"zh-Hans":{"summaryShort":"为AI编程工具添加设计命令,还附带一个能找出丑陋或损坏网页的检查工具。","summary":"Impeccable 是一套设计指令和24个命令,安装在Claude Code、Cursor、Codex和Copilot等AI编程助手里。安装后你输入像/impeccable audit或/impeccable polish这样的命令,助手就会检查或改善你网页的外观。它还附带一个带62条固定规则的命令行检查工具,可以在不使用AI、不需要密钥的情况下扫描文件或网站地址,以及一个Chrome扩展,能在任何页面上显示同样的检查结果。它适合已经在用AI助手搭建网站的人;运行安装程序需要先装好Node。","easeRationale":"在所有系统上步骤都一样:先装Node 18+(package.json中的\"engines\":{\"node\":\">=18.19.0\"}),然后在终端里运行`npx impeccable install`,再在你的AI编程工具里运行`/impeccable init`,之后才能开始使用。发布页里没有安装程序,只有该技能内部使用的引擎二进制文件。","costNote":"采用Apache-2.0许可。README说检测工具\"无需LLM,无需API密钥\";设计命令则运行在你另外付费使用的AI助手里。","notVerified":["安装程序和下载的引擎在普通Windows环境下是否真的能用","除了README描述的内容,impeccable.style目录请求实际发送和存储了什么","首次运行时下载的引擎是否与该技能自带的校验值匹配","由于62条检测规则只检查固定模式,实际设计效果到底有多好","该钩子是否会在每次编辑文件时拖慢你的编程工具"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyODEwMjky","slug":"pbatard__rufus","name":"Rufus","owner":"pbatard","fullName":"pbatard/rufus","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"it_erases_the_drive_you_point_it_at","severity":"attention","evidence":"README 中的第一项功能是 'Format USB, flash card and virtual drives to FAT/FAT32/NTFS/UDF/exFAT/ReFS/ext2/ext3'，项目自称 'The Reliable USB Formatting Utility'。写入镜像或格式化会替换所选驱动器上的全部内容，所以选错驱动器就会毁掉上面的数据。","source":"ai","title":"会清除你所指定的驱动器"},{"code":"can_skip_windows_11_hardware_requirements","severity":"info","evidence":"README 中的功能：'Create Windows 11 installation drives for PCs that don't have TPM or Secure Boot' 以及 'Improve Windows installation experience by automatically setting up OOBE parameters (local account, privacy options, etc.)'。这些是微软放在其安装程序里的检查和设置界面，以这种方式安装的电脑不在微软的支持范围之内。","source":"ai","title":"可以绕过 Windows 11 的硬件要求"},{"code":"downloads_windows_images_from_microsoft","severity":"info","evidence":"README 中的功能：'Download official Microsoft Windows 8, Windows 10 or Windows 11 retail ISOs'。这些文件来自微软，但使用 Windows 仍然需要微软的许可证。","source":"ai","title":"从微软下载 Windows 镜像"},{"code":"no_release_for_four_months_while_code_keeps_moving","severity":"info","evidence":"检查结果：最近一次发布是 2026-06-30，而最近一次提交是 2026-10-28，所以已发布的 4.15 版本比当前代码旧。目前只有 10 个未解决的 issue。","source":"ai","title":"四个月没有发布新版本，代码却一直在更新"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":5,"maxMB":16,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system","hardware"],"alternativeTo":[{"slug":"wintousb","name":"WinToUSB","confirmed":true}],"declaredAlternatives":[],"stars":38310,"downloads":273651229,"downloadsWeek":948088,"downloadsDays":7,"growth":77,"growthDays":7,"history":[{"date":"2026-10-26","stars":38128},{"date":"2026-10-27","stars":38139},{"date":"2026-10-28","stars":38147},{"date":"2026-10-29","stars":38163},{"date":"2026-10-30","stars":38176},{"date":"2026-10-31","stars":38193},{"date":"2026-11-01","stars":38209},{"date":"2026-11-02","stars":38220},{"date":"2026-11-03","stars":38233},{"date":"2026-11-04","stars":38241},{"date":"2026-11-05","stars":38255},{"date":"2026-11-06","stars":38261},{"date":"2026-11-07","stars":38272},{"date":"2026-11-08","stars":38284},{"date":"2026-11-09","stars":38298},{"date":"2026-11-10","stars":38310}],"firstPublishedAt":"2026-10-27T01:10:12.671Z","commitSha":"2a8ecf282801668f70945f99eb1f78a979ded5aa","verifiedAt":"2026-10-29T01:45:55.252Z","lastCommitAt":"2026-11-09T23:50:14Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"rufus-4.15.exe","sizeBytes":1989992,"url":"https://github.com/pbatard/rufus/releases/download/v4.15/rufus-4.15.exe","releaseUrl":"https://github.com/pbatard/rufus/releases/tag/v4.15"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a USB stick into a bootable drive from an ISO file, and formats memory cards and USB drives on Windows.","summary":"Rufus is a small Windows program that writes an ISO or disk image onto a USB stick so you can boot a computer from it, for example to install Windows or a Linux system. It also formats USB sticks and memory cards, checks them for bad blocks and fake capacity, calculates checksums, can download official Microsoft Windows ISO files, and can build Windows To Go drives that run from the stick. It is a single 2 MB file with nothing to install, speaks 38 languages, and is free software under the GPL.","easeRationale":"On Windows you download one file from the release (rufus-4.15.exe, 2 MB, with separate x86 and ARM64 builds) and open it; the README says 'Small footprint. No installation required. Portable.' There are no Mac or Linux builds.","notVerified":["Whether a drive it creates really boots on your particular computer or firmware","Whether the downloaded Windows ISO matches what Microsoft publishes today","Whether the bad-block and fake-capacity checks catch every bad USB stick","What the program does on a machine where you are not an administrator","Real disk use while a large ISO is being written"]},"zh-Hans":{"summaryShort":"把USB闪存盘从ISO文件制作成可启动驱动器,并在Windows上格式化内存卡和USB驱动器。","summary":"Rufus 是一款小巧的Windows程序,能把ISO或磁盘镜像写入USB闪存盘,让你用它启动电脑,比如安装Windows或Linux系统。它还能格式化USB闪存盘和内存卡,检测坏块和虚假容量,计算校验值,可以下载微软官方的Windows ISO文件,还能制作从闪存盘运行的Windows To Go驱动器。它是一个仅2MB的单文件程序,无需安装,支持38种语言,是GPL许可下的自由软件。","easeRationale":"在Windows上,你只需从发布页下载一个文件(rufus-4.15.exe,2MB,分x86和ARM64版本)并打开它;README写道\"Small footprint. No installation required. Portable.\"没有Mac或Linux版本。","notVerified":["它制作的驱动器在你具体的电脑或固件上是否真的能启动","下载的Windows ISO是否与微软今天发布的版本一致","坏块和虚假容量检测是否能发现所有有问题的USB闪存盘","在你不是管理员的电脑上,这个程序会做什么","写入大型ISO时实际占用的磁盘空间"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNTM3OTYyMA==","slug":"pcsx2__pcsx2","name":"PCSX2","owner":"PCSX2","fullName":"PCSX2/pcsx2","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"high","signals":[{"code":"needs_system_software_copied_from_a_real_ps2","severity":"attention","evidence":"README 的 System Requirements 部分写道：\"Please note that a BIOS dump from a legitimately-owned PS2 console is required to use the emulator.\" 这个文件是索尼的软件，并不包含在下载里，复制它意味着要从你自己拥有的主机里取出来；下载别人提供的副本是另外一回事，各个国家的规定也不一样。","source":"ai","title":"需要从真实的 ps2 主机上复制系统软件"},{"code":"you_must_supply_your_own_games","severity":"attention","evidence":"这个项目提供的是模拟器，不是游戏：\"This allows you to play PS2 games on your PC\"，兼容性列表在 pcsx2.net/compat/ 上写着 \"more than 2500 titles tested\"。游戏文件由你自己提供，从你拥有的光盘上合法复制内容的规定在不同国家也不一样。","source":"ai","title":"游戏需要你自己提供"},{"code":"the_github_downloads_are_nightly_builds","severity":"info","evidence":"仓库里有 1200 个发行版，最新的三个（v2.9.162、v2.9.160、v2.9.159）日期都是 2026-11-06 或 2026-11-07。README 区分了两个渠道：\"Installers and binaries for both stable and nightly builds are available from our website\"（pcsx2.net/downloads/）。直接从 GitHub 拿到的是这一周的开发版本。","source":"ai","title":"GitHub 上的下载是每日构建版"},{"code":"whether_your_game_runs_is_tracked_game_by_game","severity":"info","evidence":"README 写道：\"newer versions can run most games at full speed, including popular titles such as Final Fantasy X and Devil May Cry 3\"，并有一份公开的兼容性列表覆盖 2500 多款游戏。809 个未关闭的 issue 对应 16021 个星标，这和一个每款游戏都有自己独特问题的项目是相符的。","source":"ai","title":"你的游戏能不能跑，是逐个游戏跟踪的"},{"code":"built_from_many_outside_libraries","severity":"info","evidence":"仓库里有一个 `3rdparty/` 文件夹和一个 .gitmodules 文件，说明程序的部分内容来自其他项目，有各自的许可证，而项目本身采用 GPL-3.0（COPYING.GPLv3）。","source":"ai","title":"依赖大量外部库构建"},{"code":"no_checksum_published_with_the_downloads","severity":"info","evidence":"每个版本的发行版里都有五个文件（AppImage、Flatpak、tar.xz、两个 7z 压缩包），README 里没有任何校验值或签名信息，所以这个安装包里没有任何东西可以用来核对下载到的文件。","source":"ai","title":"下载文件没有公布校验值"},{"code":"needs_bios_copied_from_the_console","severity":"high","evidence":"README 的 System Requirements 写道：\"Please note that a BIOS dump from a legitimately-owned PS2 console is required to use the emulator.\" 这个文件是索尼的系统软件，下载包里并不包含它，游戏也同样需要你自己提供：README 的 Project Details 部分写道 \"This allows you to play PS2 games on your PC\"，项目的兼容性列表上写着 \"more than 2500 titles tested\"。很多国家的法律限制复制主机的系统软件和光盘内容，各国的规定也不一样。","source":"ai","title":"需要从主机上复制 BIOS"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":30,"maxMB":91,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":16036,"downloads":25586929,"downloadsWeek":23755,"downloadsDays":3,"growth":43,"growthDays":6,"history":[{"date":"2026-11-07","stars":16015},{"date":"2026-11-08","stars":16021},{"date":"2026-11-09","stars":16028},{"date":"2026-11-10","stars":16036}],"firstPublishedAt":"2026-11-08T23:01:18.925Z","commitSha":"695e26713be06daf8b9d05978602946b82008716","verifiedAt":"2026-11-08T22:51:24.392Z","lastCommitAt":"2026-11-10T15:42:26Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Plays PlayStation 2 games on your computer, often sharper than on the console, if you supply the games yourself.","summary":"PCSX2 plays PlayStation 2 games on a Windows, Mac or Linux computer, usually looking better than on the console. Every release has a ready file for all three systems (about 30 MB for Windows, 23 MB for macOS, a 58 MB AppImage or a 27 MB Flatpak for Linux). Two things are not included: the games, and the PS2 BIOS, which the project says must be copied from a console you own. More than 2500 games have been tested and the result varies from game to game. The project is over 20 years old and is still updated daily.","easeRationale":"All three systems sit at B: the download is ready to run (pcsx2-v2.9.162-windows-x64-Qt.7z at 30 MB, pcsx2-v2.9.162-macos-Qt.tar.xz at 23 MB, an AppImage or Flatpak for Linux, 58 and 27 MB), but one extra step stands between you and a game, because the README says \"a BIOS dump from a legitimately-owned PS2 console is required to use the emulator\". Windows users also have to pick the right file: the `-symbols.7z` next to it is debugging data, not the program.","costNote":"Free and GPL-3.0. The hidden cost is the hardware: you need a PlayStation 2 console to take the BIOS from, and the project's own requirements page lists what a PC needs.","notVerified":["Whether the game you want runs well, which the project tracks per title on its own compatibility list","How it performs on your machine, since speed depends on the processor and graphics card","Real disk use beyond the 30 to 91 MB of the downloads, once games, BIOS and save states are added","Whether a downloaded file matches the published source, as no checksum or signature is given here","Whether taking a BIOS copy from your own console is allowed where you live"]},"zh-Hans":{"summaryShort":"在电脑上玩 PlayStation 2 游戏，画面常常比主机更清晰，但游戏要你自己提供。","summary":"PCSX2 可以在 Windows、Mac 或 Linux 电脑上玩 PlayStation 2 游戏，画面通常比主机上看起来更好。每个发行版都为三个系统各提供一个可以直接用的文件（Windows 大约 30 MB，macOS 23 MB，Linux 的 AppImage 58 MB 或 Flatpak 27 MB）。有两样东西不包含在内：游戏本身，以及 PS2 的 BIOS，项目说这个 BIOS 必须从你自己拥有的主机上复制过来。已经测试过 2500 多款游戏，效果因游戏而异。这个项目已经超过 20 年历史，现在仍然每天都有更新。","easeRationale":"三个系统都是 B 级：下载下来就能直接运行（Windows 版 pcsx2-v2.9.162-windows-x64-Qt.7z 30 MB，macOS 版 pcsx2-v2.9.162-macos-Qt.tar.xz 23 MB，Linux 的 AppImage 或 Flatpak 分别是 58 MB 和 27 MB），但在你能玩上游戏之前还有一步，因为 README 说 \"a BIOS dump from a legitimately-owned PS2 console is required to use the emulator\"。Windows 用户还要选对文件：旁边那个 `-symbols.7z` 是调试数据，不是程序本体。","costNote":"免费，采用 GPL-3.0 许可证。隐藏的成本是硬件：你需要一台 PlayStation 2 主机来取出 BIOS，项目自己的需求页面也列出了电脑需要满足的条件。","notVerified":["你想玩的游戏运行是否顺畅，这是项目在自己的兼容性列表里按游戏逐一跟踪的","在你自己的电脑上表现如何，因为速度取决于处理器和显卡","除了下载文件的 30 到 91 MB 之外，加上游戏、BIOS 和存档状态后实际占用的磁盘空间","下载的文件是否和发布的源代码一致，因为这里没有给出任何校验值或签名","从你自己的主机上取出 BIOS 副本，在你所在的地方是否被允许"]}}},{"id":"R_kgDOMtqknw","slug":"pdfmathtranslate__pdfmathtranslate","name":"PDFMathTranslate","owner":"PDFMathTranslate","fullName":"PDFMathTranslate/PDFMathTranslate","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"your_document_is_sent_to_a_translation_service","severity":"attention","evidence":"README 第 4.1 节写道：\"Use Google as the default translation service\"，依赖里包括 googletrans、deepl、openai、azure-ai-translation-text 和 tencentcloud-sdk-python-tmt。所以默认情况下，你 PDF 里的文字会被发送到 Google，只有 ollama 和 xinference-client 这两个选项能让内容留在你自己的电脑上。","source":"ai","title":"你的文档会被发送到一个翻译服务"},{"code":"one_flag_puts_the_page_on_the_public_internet","severity":"attention","evidence":"选项表里列出了 `--share`，功能是 \"Public link\"（`pdf2zh -i --share`），这是 Gradio 的功能，会给出一个任何人都能打开的地址，而登录验证是另外需要单独打开的选项（`--authorized users.txt`）。Docker 的说明里也用 `docker run -d -p 7860:7860 byaidu/pdf2zh` 把界面公开出来。","source":"ai","title":"一个参数就能把页面公开到互联网上"},{"code":"the_ready_windows_file_is_from_july_2025","severity":"attention","evidence":"最新的发行版是 2025-07-11 发布的 v1.9.11，而最后一次提交是在 2026-11-08，pyproject.toml 里已经写着版本号 1.9.12。同一份 pyproject 在 `gradio>=6.16.0` 旁边有一条注释 \"# >=6.16.0 fixes the gradio 5.x advisories (e.g. CVE-2026-28414)\"，这个修复不可能出现在十六个月前打包的版本里。","source":"ai","title":"现成的 Windows 安装文件停留在 2025 年 7 月"},{"code":"version_2_lives_in_another_repository","severity":"info","evidence":"README 写道：\"2.0 Moved to a new repository under the organization: PDFMathTranslate/PDFMathTranslate-next\" 和 \"Version 2.0 official release has been published\"，而第 4.3 节说这个仓库是 \"the present and the original project for stable release\"，另一个分支 \"is intended solely for development and does not address compatibility issues and is not designed for community-contributions\"。两个在用同一个名字的项目都还活跃着。","source":"ai","title":"2.0 版本放在另一个仓库里"},{"code":"downloads_a_layout_model_and_ocr_data_on_first_use","severity":"info","evidence":"README 写道：\"The current program relies on the AI model (`wybxc/DocLayout-YOLO-DocStructBench-onnx`)\"，在 OCR 模式下 \"The first scanned page downloads the requested language data from Tesseract's `tessdata_fast` 4.1.0 release into `~/.cache/pdf2zh/tessdata/4.1.0`\"。Windows 发行版有两个大小，308 MB 和 521 MB 的 \"with-assets\" 版，差的就是这部分内容。","source":"ai","title":"首次使用时会下载排版模型和 OCR 数据"},{"code":"a_mirror_is_suggested_for_the_model_download","severity":"info","evidence":"针对 \"network difficulties when loading the AI model\" 的情况，README 建议设置 `set HF_ENDPOINT=https://hf-mirror.com`，这会从第三方镜像站而不是 Hugging Face 本身获取模型。","source":"ai","title":"建议用镜像站下载模型"},{"code":"sponsors_in_the_readme_and_rewards_for_contributors","severity":"info","evidence":"README 开头是一条 Warp 赞助横幅，链接到 go.warp.dev/PDFMathTranslate，致谢部分写道 \"Immersive Translation sponsors monthly Pro membership redemption codes for active contributors to this project\"。Immersive Translate 的 BabelDOC 也是推荐的在线演示之一，还是这个项目的 \"new backend\"。","source":"ai","title":"README 里有赞助商，也给贡献者提供奖励"},{"code":"experimental_modes_with_stated_limits","severity":"info","evidence":"README 把 OCR、`--mode precise` 和 `--ultrafast` 标为实验性功能，并自己列出了限制：\"The initial implementation targets white-background scans: partial scans on pages that already contain text are skipped, and handwritten text or inline equations may be recognized incorrectly\"，ultrafast 模式 \"is best suited to PDFs that already contain native text; use normal mode for complex tables or formulas\"。","source":"ai","title":"实验性模式，带有明确说明的限制"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["python","docker","gpu_nvidia"],"disk":{"minMB":521,"maxMB":1563,"basis":"installer"},"apiKeys":["OpenAI (only if you choose it as the translation service)","DeepL","Azure Translator","Tencent Machine Translation"],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["office","productivity","ai_skills"],"alternativeTo":[{"slug":"deepl-pro","name":"Pro","confirmed":true},{"slug":"acrobat","name":"Acrobat","confirmed":true},{"slug":"mathpix","name":"Mathpix","confirmed":true}],"declaredAlternatives":[],"stars":38349,"downloads":147915,"downloadsWeek":1403,"downloadsDays":5,"growth":179,"growthDays":7,"history":[{"date":"2026-11-05","stars":38226},{"date":"2026-11-06","stars":38257},{"date":"2026-11-07","stars":38275},{"date":"2026-11-08","stars":38293},{"date":"2026-11-09","stars":38320},{"date":"2026-11-10","stars":38349}],"firstPublishedAt":"2026-11-08T23:01:18.931Z","commitSha":"d81a33ce52b66de74b33969e922476f0eb15e04e","verifiedAt":"2026-11-08T22:52:40.012Z","lastCommitAt":"2026-11-10T21:15:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Translates a PDF into another language while keeping the layout, formulas, tables and figures where they were.","summary":"PDFMathTranslate (pdf2zh) translates PDF documents into another language and keeps the original layout: formulas, charts, tables, contents and notes stay in place, and it can also produce a bilingual file with both languages. It is built for scientific papers. On Windows you unzip the release and double-click pdf2zh.exe; on Mac and Linux you install it with pip or run the Docker image; the authors also keep free online pages. Translation goes through Google by default, or DeepL, OpenAI, Azure and local Ollama models.","easeRationale":"Windows and the web share grade A. On Windows you \"Download pdf2zh-version-win64.zip from release page\" (308 MB, or 521 MB with assets), \"Unzip and double-click `pdf2zh.exe` to run\", with everything bundled; the README only warns that some machines first need Microsoft's vc_redist.x64.exe. On the web the authors run a free service at pdf2zh.com plus Hugging Face and ModelScope pages. macOS and Linux are C: there is no ready file for them, so you install Python 3.11 or 3.12, then `pip install pdf2zh` and run `pdf2zh -i` to open the page at http://localhost:7860/.","costNote":"The program is free (AGPL-3.0) and the default Google service needs no account. Picking DeepL, OpenAI, Azure or Tencent instead means paying that service for what you translate.","notVerified":["How good the translation reads, since the wording comes from whichever outside service you pick","Whether the layout really survives on your own PDFs, especially scans, complex tables and inline formulas","What each translation service keeps of the text you send it","Real disk use beyond the 521 MB to 1.5 GB measured from the installers, once models and OCR data are cached","Whether the 2025 Windows build still behaves like the current code, which has moved on by sixteen months"]},"zh-Hans":{"summaryShort":"把 PDF 翻译成另一种语言，同时保持排版，公式、表格和图片都留在原来的位置。","summary":"PDFMathTranslate（pdf2zh）把 PDF 文档翻译成另一种语言，同时保留原来的排版：公式、图表、表格、目录和注释都留在原位，还可以生成一份中英对照的双语文件。它是为学术论文设计的。在 Windows 上，你只需解压发行版并双击 pdf2zh.exe；在 Mac 和 Linux 上用 pip 安装或运行 Docker 镜像；作者还提供免费的在线页面。翻译默认走 Google，也可以选择 DeepL、OpenAI、Azure 以及本地的 Ollama 模型。","easeRationale":"Windows 和网页版都是 A 级。在 Windows 上，你要 \"Download pdf2zh-version-win64.zip from release page\"（308 MB，带 assets 的是 521 MB），\"Unzip and double-click `pdf2zh.exe` to run\"，所有东西都打包好了；README 只提醒有些电脑要先装微软的 vc_redist.x64.exe。在网页上，作者维护着一个免费服务 pdf2zh.com，还有 Hugging Face 和 ModelScope 的页面。macOS 和 Linux 是 C 级：没有现成的文件，你得先装 Python 3.11 或 3.12，再 `pip install pdf2zh`，然后运行 `pdf2zh -i` 打开 http://localhost:7860/ 这个页面。","costNote":"这个程序是免费的（AGPL-3.0 许可证），默认使用的 Google 服务不需要账号。如果改用 DeepL、OpenAI、Azure 或腾讯，就要为你翻译的内容向那个服务付费。","notVerified":["翻译读起来效果怎样，因为具体措辞取决于你选的那个外部服务","排版在你自己的 PDF 上是否真的能保持住，尤其是扫描件、复杂表格和行内公式","每个翻译服务会保留你发送文本中的哪些内容","除了安装包测得的 521 MB 到 1.5 GB 之外，缓存模型和 OCR 数据之后实际占用的磁盘空间","2025 年的 Windows 构建版本是否还和已经往前走了十六个月的当前代码表现一致"]}}},{"id":"R_kgDOVQrvmA","slug":"pengchujin__aerial-tv","name":"Aerial TV","owner":"pengchujin","fullName":"pengchujin/aerial-tv","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"12 天内获得 8706 个星标","source":"check","title":"新仓库，星标很多"},{"code":"powerful_permissions_requested","severity":"attention","evidence":"README.md 中的安装命令授予应用写入受保护系统设置的权限、通知监听器权限和始终显示在最上层的权限：\"adb shell pm grant app.orchard.tv android.permission.WRITE_SECURE_SETTINGS\"、\"cmd notification allow_listener app.orchard.tv/...guard.RevivalListener\"、\"appops set app.orchard.tv SYSTEM_ALERT_WINDOW allow\"。README 称无障碍服务 \"only looks at which window is in front\"，通知监听器 \"does not read notifications\"，这一点从外部无法核实。","source":"ai","title":"申请了高权限"},{"code":"binaries_hosted_outside_github","severity":"attention","evidence":"GitHub 上的 70 个版本发布都不带文件；README.md 把下载指向作者自己的网站 (https://aerial.qust.me/download/aerial-standard.apk)，快速安装方式是 \"curl -fsSL https://aerial.qust.me/install.sh | KEEP_ADB=1 TV=... bash\"，即运行来自该网站的脚本。你安装的内容并不是这里检查过的代码。","source":"ai","title":"安装文件托管在 GitHub 之外"},{"code":"keeps_adb_debugging_open","severity":"attention","evidence":"推荐的安装方式使用 KEEP_ADB=1，README.md 的解释是：Xiaomi 在待机后会关闭 ADB，加上这个参数后，Aerial 会把它重新打开。让 ADB 调试一直开着，意味着同一网络里的任何人都可以在电视上安装应用。","source":"ai","title":"让 ADB 调试一直开着"},{"code":"works_only_on_some_tvs","severity":"info","evidence":"README.md 的支持表用红叉标出了许多型号，例如 Skyworth 7T641 和 75A6E（\"black screen, or loses focus and is killed by the system\"）以及 FFALCON 75S78A Plus（\"the firmware wipes the accessibility service about once a minute\"）。LG webOS、Samsung Tizen 和 Huawei Harmony 电视完全不受支持。","source":"ai","title":"只适用于部分电视"},{"code":"third_party_video_content","severity":"info","evidence":"README.md：114 部航拍屏保影片归 Apple 所有，\"copyright belongs to Apple, played directly from Apple's servers\"，每日壁纸来自 Bing。项目声明与 Apple、Xiaomi 或 TCL 没有关联。","source":"ai","title":"第三方视频内容"},{"code":"very_young_and_fast_moving","severity":"info","evidence":"仓库创建后的 12 天里发布了 70 个版本（v1.75、v1.76 和 v1.77 在两天内发布），面向老电视的 Lite 版本被描述为首个尚未充分测试的版本：\"only tested on Android 4.0.4 / 4.4.4 virtual machines\"。","source":"ai","title":"项目很新且更新很快"},{"code":"install_delegated_to_ai_agent","severity":"info","evidence":"README.md 推荐的方法是把一段文字粘贴给能在你电脑上运行命令的 AI 代理（\"WorkBuddy, Doubao desktop, Qwen desktop, Trae, Claude Code, Codex\"），让它读取 https://aerial.qust.me/agent.md 并自己运行命令。","source":"ai","title":"把安装交给 AI 代理"}],"platforms":[],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[{"slug":"projectivy-launcher","name":"Launcher","confirmed":true}],"declaredAlternatives":[],"stars":11091,"downloads":130044,"downloadsWeek":33391,"downloadsDays":7,"growth":1752,"growthDays":7,"history":[{"date":"2026-10-25","stars":5112},{"date":"2026-10-26","stars":6090},{"date":"2026-10-27","stars":6760},{"date":"2026-10-28","stars":7387},{"date":"2026-10-29","stars":7743},{"date":"2026-10-30","stars":8057},{"date":"2026-10-31","stars":8325},{"date":"2026-11-01","stars":8698},{"date":"2026-11-02","stars":9014},{"date":"2026-11-03","stars":9339},{"date":"2026-11-04","stars":9615},{"date":"2026-11-05","stars":9954},{"date":"2026-11-06","stars":10332},{"date":"2026-11-07","stars":10554},{"date":"2026-11-08","stars":10731},{"date":"2026-11-09","stars":10894},{"date":"2026-11-10","stars":11091}],"firstPublishedAt":"2026-10-25T23:19:23.908Z","commitSha":"e8c251e607689db7b80709c791d8b309185f8979","verifiedAt":"2026-11-02T02:53:02.854Z","lastCommitAt":"2026-11-10T13:37:20Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Replaces the home screen of a Chinese-brand Android TV with an Apple TV-style one, with no ads or recommendation rows.","summary":"Aerial is a home screen for Android TV boxes, projectors and TVs sold in China (Xiaomi, TCL, Skyworth, Hisense and others). It shows only your own apps, folders, wallpapers and Apple's aerial screensaver videos, instead of the maker's ad and series rows, and can start automatically when the TV turns on. You can also arrange the icons from your phone by scanning a QR code. Installing it is the hard part: you turn on ADB debugging on the TV and run several commands from a computer. Android 5.0 or newer; a Lite version covers Android 4.0-4.4.","easeRationale":"It is an app for Android TV, so no desktop system gets a grade. To really replace the home screen you turn on hidden developer options on the TV with remote-control key sequences, install Android's ADB tool on a computer (macOS and Linux have a one-line script, Windows does not) and run around six commands, including pairing on newer TVs. Copying the APK to a USB drive is easier, but then the app opens only as a normal app and does not take over the home button.","notVerified":["Whether it works on your exact TV model: most of the support table comes from user reports, not from testing by the author","What the accessibility service and the notification listener actually read, since both have wide access on the TV","Whether the APK and install script on the author's website are built from the code in this repository","Whether a firmware update from the TV maker breaks the home-screen takeover","How much space the app uses on the TV (the releases carry no files, so no size could be measured)"]},"zh-Hans":{"summaryShort":"把中国品牌安卓电视的主屏幕换成苹果电视风格的主屏幕，没有广告，也没有推荐内容行。","summary":"Aerial 是一款面向在中国销售的安卓电视盒子、投影仪和电视（Xiaomi、TCL、Skyworth、Hisense 等）的主屏幕。它只显示你自己的应用、文件夹、壁纸和苹果的航拍屏保视频，取代厂商的广告行和剧集行，并且可以在电视开机时自动启动。你也可以用手机扫二维码来整理图标。安装是最难的部分：需要在电视上打开 ADB 调试，并在电脑上运行好几条命令。要求 Android 5.0 或更高版本；另有 Lite 版本支持 Android 4.0-4.4。","easeRationale":"这是一款安卓电视应用，所以不对桌面系统评级。要真正替换主屏幕，需要先用遥控器按键序列在电视上打开隐藏的开发者选项，在电脑上安装 Android 的 ADB 工具（macOS 和 Linux 有一行脚本，Windows 没有），再运行大约六条命令，较新的电视还需要配对。把 APK 复制到 U 盘里更简单，但这样应用只会作为普通应用打开，不会接管主页键。","notVerified":["它能否在你的具体电视型号上运行：支持表里的大部分内容来自用户反馈，而不是作者的测试","无障碍服务和通知监听器实际读取了什么，因为这两者在电视上都有很广的访问权限","作者网站上的 APK 和安装脚本是否由这个仓库里的代码构建","电视厂商的固件更新是否会破坏对主屏幕的接管","这个应用在电视上占多少空间（发布版本不带文件，所以无法测量大小）"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0ODc2OTI0MQ==","slug":"penpot__penpot","name":"Penpot","owner":"penpot","fullName":"penpot/penpot","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"linux":"D","web":"A"},"risk":"low","signals":[{"code":"no_desktop_app","severity":"info","evidence":"最近三个发布版本（2.18.5、2.18.4、2.18.3）都没有各平台的二进制文件，README 只提供托管服务或 \"Docker, Kubernetes, Elestio or other options\"。它是一个浏览器工具，而不是需要安装的程序。","source":"ai","title":"没有桌面应用"},{"code":"hosted_version_holds_your_files","severity":"info","evidence":"README.md 把位于 design.penpot.app 的 \"our SAAS\" 作为简便途径，所以走这条路时，你的设计保存在该公司（Kaleidos）的服务器上。项目重点推荐的替代方案是自行托管：\"full ownership of your design infrastructure\"。","source":"ai","title":"托管版本会保存你的文件"},{"code":"paid_tier_for_some_features","severity":"info","evidence":"README.md：\"Penpot Enterprise is our paid plan ... Manage teams and access from a centralized Admin Console, configure advanced permissions, and connect your identity provider through SSO\"，云端和自行托管的部署都可使用。","source":"ai","title":"部分功能需要付费套餐"}],"platforms":["web","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Penpot access token (only for the API, webhooks and the MCP server)"],"cost":{"model":"freemium"},"license":{"spdx":"MPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["design_cad","development"],"alternativeTo":[{"slug":"figma","name":"Figma","confirmed":true},{"slug":"sketch","name":"Sketch","confirmed":true},{"slug":"miro","name":"Miro","confirmed":true}],"declaredAlternatives":[],"stars":62260,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":231,"growthDays":7,"history":[{"date":"2026-10-25","stars":61670},{"date":"2026-10-26","stars":61720},{"date":"2026-10-27","stars":61758},{"date":"2026-10-28","stars":61801},{"date":"2026-10-29","stars":61845},{"date":"2026-10-30","stars":61887},{"date":"2026-10-31","stars":61932},{"date":"2026-11-01","stars":61967},{"date":"2026-11-02","stars":61998},{"date":"2026-11-03","stars":62029},{"date":"2026-11-04","stars":62070},{"date":"2026-11-05","stars":62112},{"date":"2026-11-06","stars":62140},{"date":"2026-11-07","stars":62165},{"date":"2026-11-08","stars":62189},{"date":"2026-11-09","stars":62216},{"date":"2026-11-10","stars":62260}],"firstPublishedAt":"2026-10-25T23:19:24.784Z","commitSha":"f6f9df9a1a94087ecc9fd30337f908598ada983c","verifiedAt":"2026-10-28T01:40:12.815Z","lastCommitAt":"2026-11-10T19:06:46Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Design and prototype screens in the browser, with real-time teamwork, and keep the files on your own server.","summary":"Penpot is a design and prototyping tool for app and website screens that runs in the browser. Several people can work on the same file at once, and developers can read the SVG, CSS and HTML behind a design, use design tokens, plugins and an API. You can open the company's hosted version with an account, or install it on your own machine or server with Docker, which takes a terminal and some setup. There is a free plan and a paid Enterprise plan for larger organisations.","easeRationale":"The best grade is on the web: the hosted version at design.penpot.app opens in a browser with an account and nothing to install. Running your own copy is grade D, because the documented route is Docker or Kubernetes on a server you set up yourself; none of the 92 releases carries a program file for Windows, Mac or Linux.","costNote":"The software is free and open source (MPL-2.0) and self-hosting costs nothing beyond your server. Penpot Enterprise is a paid plan for admin console, advanced permissions and single sign-on; the price is not in the repository.","notVerified":["How well files from Figma or Sketch really import, and what is lost","The real server size, memory and disk a self-hosted Penpot needs for a team","What the free hosted plan limits (projects, team members, storage)","Whether the browser tool stays smooth on very large design files"]},"zh-Hans":{"summaryShort":"在浏览器里设计和制作界面原型,支持团队实时协作,文件可以保存在你自己的服务器上。","summary":"Penpot 是一款在浏览器中运行的App和网站界面设计与原型制作工具。多人可以同时编辑同一个文件,开发者还能查看设计背后的SVG、CSS和HTML代码,使用设计令牌(design tokens)、插件和API。你可以用账号打开官方提供的托管版本,也可以用Docker把它装在自己的电脑或服务器上,这需要用到终端并做一些配置。它有免费版,也有面向大型组织的付费Enterprise版。","easeRationale":"最好的评级出现在网页端:design.penpot.app 上的托管版本只需用账号登录浏览器打开,不用安装任何东西。自己运行一份评级为D,因为官方文档给出的方式是在自己搭建的服务器上用Docker或Kubernetes部署;92个发布版本里没有一个是面向Windows、Mac或Linux的程序安装包。","costNote":"该软件免费且开源(MPL-2.0),自己搭建服务器运行除了服务器本身没有其他费用。Penpot Enterprise是一个付费套餐,提供管理控制台、高级权限和单点登录;具体价格没有写在代码仓库里。","notVerified":["Figma或Sketch的文件实际导入效果如何,又会丢失哪些内容","团队自行部署的Penpot实际需要多大的服务器配置、内存和磁盘空间","免费托管版有哪些限制(项目数量、团队成员数、存储空间)","在非常大的设计文件下,这个浏览器工具是否还能保持流畅"]}}},{"id":"R_kgDOJMaHRQ","slug":"picsart-ai-research__text2video-zero","name":"Text2Video-Zero","owner":"Picsart-AI-Research","fullName":"Picsart-AI-Research/Text2Video-Zero","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 里有一个指向 https://www.dropbox.com/s/uv90mi2z598olsq/Text2Video-Zero.MP4?dl=0 的链接","source":"check","title":"外部下载链接"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定版本的依赖存在已知漏洞（OSV）：accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; basicsr: GHSA-86w8-vhw6-q9qq, PYSEC-2026-1215; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2; opencv_python: GHSA-qr4w-53vh-m672, PYSEC-2023-183; opencv-contrib-python: GHSA-cxjf-x6jp-p7mc, PYSEC-2023-181; pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h; pytorch_lightning: GHSA-2vj5-px25-gjrp, GHSA-4cv3-v7pv-rfhf, GHSA-75m9-98v2-hjpm; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-5pcm-hx3q-hm94; tqdm: GHSA-g7vv-2v7x-gj9p, PYSEC-2026-1976; transformers: GHSA-282v-666c-3fvg, GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm","source":"check","title":"存在已知漏洞"},{"code":"research_code_untouched_since_april_2023","severity":"attention","evidence":"最后一次提交是在 2023-04-14（本次检查前 1304 天），完全没有任何发行版，这份安装方法锁定的还是那个时期的版本：`torch==1.13.1`、`diffusers==0.14.0`、`gradio==3.23.0`、`transformers==4.26.0`，environment.yaml 要求 python=3.8.5 和 cudatoolkit=11.3，而 README 又要求 Python 3.9。像 \"We plan to continue optimizing our code\" 这样的承诺之后并没有再出现新的提交。","source":"ai","title":"研究代码自 2023 年 4 月起就没再变动"},{"code":"the_license_limits_how_you_may_use_it","severity":"attention","evidence":"GitHub 没能识别出许可证类型（NOASSERTION），因为这个文件是 \"CreativeML Open RAIL-M dated March 28, 2023\"，它说自己 \"added use-based restrictions not permitting the use of the Model in very specific scenarios\"，而且任何衍生作品都必须带有 \"at minimum - the same use-based restrictions\"。这不是一份普通的宽松许可证，README 还补充说原始的 Stable Diffusion 代码也遵循同样的条款。","source":"ai","title":"许可证限制了你可以怎样使用它"},{"code":"one_flag_puts_the_page_on_the_public_internet","severity":"attention","evidence":"README 写道：\"To access the app remotely, run this shell command: `python app.py --public_access`\"，后面还写着 \"For security information about public access we refer to the documentation of gradio\"。这个页面本身没有登录验证，所以它给出的地址谁拿到都能打开。","source":"ai","title":"一个参数就能把页面公开到互联网上"},{"code":"loads_model_files_from_community_sites","severity":"attention","evidence":"代码通过 `get_model_list()` 列出并加载模型，\"any base model for stable diffusion and any dreambooth model hosted on huggingface can now be loaded\"，README 还建议 \"Dreambooth models can be obtained, for instance, from CIVITAI\"。这些都是陌生人上传的文件，你的 Python 进程随后会把它们加载进来，而项目并没有提供任何校验或审核。","source":"ai","title":"从社区网站加载模型文件"},{"code":"the_bundled_styles_copy_commercial_franchises","severity":"info","evidence":"内置的 Dreambooth 模型分别叫 \"Anime DB\"、\"Avatar DB\"、\"GTA-5 DB\" 和 \"Arcane DB\"，由关键词 `gtav style`、`arcane style` 和 `avatar style` 触发。它们的存在就是为了重现属于其他公司的电影和游戏的视觉风格，用这些内容能发布什么，取决于你所在的地方。","source":"ai","title":"内置的风格模仿商业特许作品"},{"code":"needs_a_large_nvidia_card","severity":"info","evidence":"2023 年 3 月 28 日的更新说明写道：\"We added a new low-memory setup. Minimum required GPU VRAM is currently **12 GB**\"，2023 年 4 月 6 日的说明提到用 Token Merging \"and chunk size set to `2`, our code can run with **less than 7 GB VRAM**\"，代价是画质下降（\"too high values will decrease the image quality\"）。","source":"ai","title":"需要一块较大的 NVIDIA 显卡"},{"code":"clips_are_very_short_by_default","severity":"info","evidence":"文档里的示例调用用的是 `\"video_length\": 8` 帧，`fps = 4`，也就是一段两秒的片段，公开的结果展示也是 GIF。后来的一条更新说明生成的视频 \"can have **arbitrary length**\"，但默认值和示例仍然停留在八帧。","source":"ai","title":"默认情况下视频片段非常短"},{"code":"one_feature_is_missing_from_the_code","severity":"info","evidence":"README 的 Ablation Study 部分写道：\"Adding `smooth_bg=True` activates background smoothing. However, our code does not include the salient object detector necessary to run that code.\" 仓库里还提到同样的方法也在 Diffusers 库里（\"available in Diffusers, starting from version `0.15.0`\"），维护得比这份代码好。","source":"ai","title":"代码里缺了一个功能"}],"platforms":["windows","linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":4247,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":4246},{"date":"2026-11-07","stars":4247},{"date":"2026-11-08","stars":4247},{"date":"2026-11-09","stars":4247},{"date":"2026-11-10","stars":4247}],"firstPublishedAt":"2026-11-08T23:01:18.939Z","commitSha":"fd76734e06aadb1dee83d2b7a368b14bdaa35565","verifiedAt":"2026-11-08T22:53:42.924Z","lastCommitAt":"2023-04-14T08:02:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes a short video from a text prompt, or restyles a clip you already have, on your own graphics card.","summary":"Text2Video-Zero makes short videos from a written prompt by driving an image model (Stable Diffusion) frame by frame, with no video training involved. It can also follow the poses or edges of an existing clip, or restyle one from an instruction such as \"make it Van Gogh Starry Night\". This is research code from a 2023 paper: Python 3.9, CUDA and an NVIDIA card with at least 12 GB of memory, or about 7 GB with compression turned up. The authors keep a free Hugging Face demo. The code has not changed since April 2023 and the method now also ships inside the Diffusers library.","easeRationale":"The web is the easy way in at grade A: the authors keep a free Hugging Face Space (PAIR/Text2Video-Zero) where you type a prompt. Doing it yourself is D on Linux and Windows: \"Install requirements using Python 3.9 and CUDA >= 11.6\", create a virtualenv, `pip install -r requirements.txt` with 35 pinned packages from 2023, and have an NVIDIA card with 12 GB of memory before `python app.py` opens the page at 127.0.0.1:7860. macOS is not graded because the setup requires CUDA.","costNote":"No charge for the code or the models. The license is not a plain open-source one: it is the CreativeML Open RAIL-M, which adds rules about how you may use it.","notVerified":["Whether the 2023 recipe still installs on a current graphics driver, Python or CUDA version","How good or steady the video looks on your own prompts, since the published examples were chosen by the authors","Real disk use, estimated by the checks at 950 MB to 4450 MB from a heuristic, once Stable Diffusion and Dreambooth files are downloaded","How long one clip takes on your card, which the README never states","What is inside a Dreambooth file taken from a community site before you load it"]},"zh-Hans":{"summaryShort":"根据一段文字生成短视频，或者给你已有的一段视频换个风格，用你自己的显卡运行。","summary":"Text2Video-Zero 根据一段文字提示，通过逐帧驱动一个图像模型（Stable Diffusion）来生成短视频，完全不涉及任何视频训练。它还可以跟随一段现有视频的姿势或边缘，或者根据一句指令改变它的风格，比如 \"make it Van Gogh Starry Night\"。这是 2023 年一篇论文配套的研究代码：需要 Python 3.9、CUDA，以及一块至少 12 GB 显存的 NVIDIA 显卡，把压缩调高后大约 7 GB 也可以。作者维护着一个免费的 Hugging Face 演示。这份代码自 2023 年 4 月以来没有再变过，这种方法现在也已经收录进 Diffusers 库里。","easeRationale":"网页版是最简单的方式，评级 A：作者维护着一个免费的 Hugging Face Space（PAIR/Text2Video-Zero），你只需输入一个提示词。自己动手安装在 Linux 和 Windows 上是 D 级：\"Install requirements using Python 3.9 and CUDA >= 11.6\"，创建一个 virtualenv，`pip install -r requirements.txt` 里有 35 个锁定在 2023 年版本的依赖包，还要有一块 12 GB 显存的 NVIDIA 显卡，然后 `python app.py` 才会打开 127.0.0.1:7860 这个页面。macOS 没有评级，因为这个安装过程需要 CUDA。","costNote":"代码和模型都不收费。但这个许可证不是普通的开源许可证：它是 CreativeML Open RAIL-M，对你可以怎样使用它附加了一些规则。","notVerified":["2023 年的这套安装方法在当前的显卡驱动、Python 或 CUDA 版本上是否还能装上","用你自己的提示词生成的视频效果好不好、稳不稳定，因为公开的示例是作者自己挑选的","实际占用的磁盘空间，检查只是用启发式方法估算在 950 MB 到 4450 MB 之间，这还是在下载了 Stable Diffusion 和 Dreambooth 文件之后","在你的显卡上生成一段视频要花多长时间，README 完全没有说明","从社区网站下载的 Dreambooth 文件在加载之前里面到底有什么内容"]}}},{"id":"R_kgDORLtfbQ","slug":"pingdotgg__t3code","name":"T3 Code","owner":"pingdotgg","fullName":"pingdotgg/t3code","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B","web":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令将一个远程脚本通过管道传给shell执行：curl -fsSL https://vite.plus | bash","source":"check","title":"curl管道直接传给shell"},{"code":"install_pipes_remote_script_to_shell","severity":"attention","evidence":"README的'Installation'部分：'curl -fsSL https://t3.codes/install.sh | sh'，在Windows上则是'irm https://t3.codes/install.ps1 | iex'——主安装过程会运行一个当时下载的脚本。","source":"ai","title":"安装过程将远程脚本直接传给shell执行"},{"code":"agent_acts_on_your_computer","severity":"attention","evidence":"README：T3 Code 'enables control of the agents on your machine'，还有名为'Permission modes'和'Remote access from a phone or another machine'的文档页面，因此代理会在你电脑上运行命令并修改文件，甚至可以通过手机操作。","source":"ai","title":"代理会在你的电脑上执行操作"},{"code":"remote_control_through_vendor_relay","severity":"attention","evidence":".env.example：'T3CODE_RELAY_URL=https://relay.t3.codes'，加上'Production Clerk instance'的密钥——默认情况下，手机和浏览器端的控制都要经过该项目托管的中继和登录服务，不过文件中提到'infra/relay'可以自行托管。","source":"ai","title":"通过厂商中继服务进行远程控制"},{"code":"early_version_many_open_issues","severity":"attention","evidence":"README的'Some notes'部分：'We are very very early in this project. Expect bugs.'最新标签是v0.1.6，GitHub显示有4048个未解决的issue。","source":"ai","title":"版本非常早期且有大量未解决的issue"},{"code":"telemetry_settings_present","severity":"info","evidence":".env.example中有一段注释掉的配置'Public, ingest-only mobile OpenTelemetry configuration'，包含'T3CODE_MOBILE_OTLP_TRACES_URL=https://api.axiom.co/v1/traces'，因此移动应用可以把使用情况的追踪数据发送到外部服务。","source":"ai","title":"存在遥测设置"}],"platforms":["windows","mac","linux","web"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":277,"maxMB":831,"basis":"installer"},"apiKeys":["An AI coding agent account: Claude Code, Codex, Cursor, Grok Build, OpenCode or Google Antigravity"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","development"],"alternativeTo":[{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"conductor","name":"Conductor","confirmed":true}],"declaredAlternatives":[],"stars":55581,"downloads":76132026,"downloadsWeek":5954290,"downloadsDays":7,"growth":11234,"growthDays":7,"history":[{"date":"2026-11-01","stars":38197},{"date":"2026-11-02","stars":41943},{"date":"2026-11-03","stars":44347},{"date":"2026-11-04","stars":46348},{"date":"2026-11-05","stars":48127},{"date":"2026-11-06","stars":49540},{"date":"2026-11-07","stars":51074},{"date":"2026-11-08","stars":52349},{"date":"2026-11-09","stars":53919},{"date":"2026-11-10","stars":55581}],"firstPublishedAt":"2026-11-01T01:56:46.019Z","commitSha":"8b74a69019e3d0f5be62dc118c526669002fc195","verifiedAt":"2026-11-01T01:39:49.389Z","lastCommitAt":"2026-11-10T22:35:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop, phone and web control panel for the AI coding agents already on your computer, such as Claude Code or Codex.","summary":"T3 Code is a control panel for AI coding agents that are already installed on your computer. You install the desktop app (Windows, macOS or Linux) or the command-line server, connect a provider such as Claude Code, Codex, Cursor, Grok Build, OpenCode or Google Antigravity, and then start, watch and steer agent sessions from the desktop app, a browser, or the iOS and Android app. It brings no AI model of its own, so you use your own provider account. The README says the project is very early and to expect bugs.","easeRationale":"On Windows, macOS and Linux there are ready installers (.exe, .dmg, .deb, AppImage, winget, Homebrew), and the README says Antigravity needs no command line: you enable it in Settings and sign in with Google, which is one extra step after installing. The other providers are harder, because each needs its own command-line tool and a login command. Using the browser app is a C, because the server still has to be installed and running on your own computer.","costNote":"The app is free. It controls agents you pay for separately (Claude Code, Codex, Cursor, Grok Build, OpenCode); the README says Google Antigravity only needs a Google sign-in.","notVerified":["What the hosted relay at relay.t3.codes sees of your code and agent sessions","Whether the permission modes really stop an agent from running a command you did not approve","Whether each listed provider works in practice, and with which versions","What you end up paying for the provider subscription behind it","Disk and memory use once the desktop app and the providers are installed"]},"zh-Hans":{"summaryShort":"面向已经安装在你电脑上的AI编程代理（如Claude Code或Codex）的桌面、手机和网页控制面板。","summary":"T3 Code是一个面向已安装在你电脑上的AI编程代理的控制面板。你可以安装桌面应用（Windows、macOS或Linux）或命令行服务器，连接一个提供商，例如Claude Code、Codex、Cursor、Grok Build、OpenCode或Google Antigravity，然后通过桌面应用、浏览器，或iOS和Android应用来启动、查看和操控代理会话。它本身不带任何AI模型，所以你需要使用自己的提供商账号。README说该项目处于非常早期的阶段，要做好遇到bug的准备。","easeRationale":"在Windows、macOS和Linux上都有现成的安装程序（.exe、.dmg、.deb、AppImage、winget、Homebrew），README说Antigravity不需要命令行：在Settings中启用并用Google登录即可，这只是安装后多一步的操作。其他提供商更麻烦，因为每一个都需要自己的命令行工具和登录命令。使用浏览器应用是C级，因为服务器仍然需要安装并运行在你自己的电脑上。","costNote":"该应用是免费的。它控制的是你单独付费使用的代理（Claude Code、Codex、Cursor、Grok Build、OpenCode）；README说Google Antigravity只需要用Google账号登录。","notVerified":["托管在relay.t3.codes上的中继服务，能看到你代码和代理会话中的哪些内容","权限模式是否真的能阻止代理运行你未批准的命令","列出的每个提供商在实际使用中是否可用，以及支持哪些版本","最终你要为背后的提供商订阅支付多少费用","安装完桌面应用和各提供商之后的磁盘和内存占用情况"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMzU4NzEz","slug":"pintaproject__pinta","name":"Pinta","owner":"PintaProject","fullName":"PintaProject/Pinta","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":58,"maxMB":174,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"photoshop","name":"Photoshop","confirmed":false}],"declaredAlternatives":[],"stars":4135,"downloads":1812750,"downloadsWeek":9746,"downloadsDays":7,"growth":9,"growthDays":7,"history":[{"date":"2026-10-25","stars":4111},{"date":"2026-10-26","stars":4112},{"date":"2026-10-27","stars":4115},{"date":"2026-10-28","stars":4115},{"date":"2026-10-29","stars":4115},{"date":"2026-10-30","stars":4117},{"date":"2026-10-31","stars":4120},{"date":"2026-11-01","stars":4120},{"date":"2026-11-02","stars":4125},{"date":"2026-11-03","stars":4126},{"date":"2026-11-04","stars":4128},{"date":"2026-11-05","stars":4130},{"date":"2026-11-06","stars":4134},{"date":"2026-11-07","stars":4133},{"date":"2026-11-08","stars":4133},{"date":"2026-11-09","stars":4136},{"date":"2026-11-10","stars":4135}],"firstPublishedAt":"2026-10-25T23:19:25.641Z","commitSha":"3dc77194b8f8fdbcc83207b0cd559bcd8b1e9f40","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-08T22:23:48Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Pinta-x86_64.exe","sizeBytes":60723896,"url":"https://github.com/PintaProject/Pinta/releases/download/3.1.2/Pinta-x86_64.exe","releaseUrl":"https://github.com/PintaProject/Pinta/releases/tag/3.1.2"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Draw, crop, resize and touch up pictures with a simple paint program that has layers and effects.","summary":"Pinta is a simple, free drawing and image editing program modeled on Paint.NET. You can draw and paint, crop and resize photos, work with layers, undo many steps and apply effects and color adjustments. It is meant for everyday image edits rather than professional photo work. Ready-made downloads exist for Windows, Mac and Linux. The README says the Windows build is code-signed and that the program does not send information over the network unless you ask it to.","easeRationale":"The latest release has ready Windows installers (Pinta-x86_64.exe and Pinta-arm64.exe), so you download one and open it.","costNote":"MIT license.","notVerified":["How it handles very large images or many layers on your computer","Which file formats beyond the common ones open correctly","Whether the privacy statement in the README matches what the program actually does"]},"zh-Hans":{"summaryShort":"用一款带图层和特效的简易绘图程序来绘画、裁剪、调整大小和修饰图片。","summary":"Pinta 是一款免费、简洁的绘图和图像编辑程序，以 Paint.NET 为蓝本。你可以绘画、裁剪和调整照片大小、使用图层、撤销多步操作，并应用特效和色彩调整。它面向日常的图片编辑，而不是专业摄影后期。它提供适用于 Windows、Mac 和 Linux 的现成下载。README 称 Windows 版本经过代码签名，并且除非你要求，否则程序不会通过网络发送信息。","easeRationale":"最新版本提供现成的 Windows 安装程序（Pinta-x86_64.exe 和 Pinta-arm64.exe），下载其中一个并打开即可。","costNote":"MIT 许可。","notVerified":["在你的电脑上处理超大图片或大量图层时表现如何","除常见格式外，还有哪些文件格式能正常打开","README 中的隐私声明是否与程序的实际行为一致"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzNDM4MzQ4OTc=","slug":"pkhungurn__talking-head-anime-2-demo","name":"Talking Head Anime 2 Demo","owner":"pkhungurn","fullName":"pkhungurn/talking-head-anime-2-demo","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"D","linux":"D","web":"C"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://www.dropbox.com/s/tsl04y5wvg73ij4/talking-head-anime-2-model.zip?dl=0","source":"check","title":"外部下载链接"},{"code":"abandoned_and_pinned_to_old_versions","severity":"attention","evidence":"最近一次提交是在 2022-06-29，比这次检查早了四年多，也没有任何发行版。environment.yml 固定了 \"python=3.8.5\"、\"pytorch=1.7.1=py3.8_cuda102_cudnn7_0\" 和 \"cudatoolkit=10.2.89\"，这些版本对现在的 NVIDIA 显卡来说太旧了；README 自己也说，你可能需要把这些版本换成适合自己电脑的版本。","source":"ai","title":"已经放弃维护，并固定在旧版本上"},{"code":"the_author_warns_the_rights_may_be_claimed_by_his_employer","severity":"attention","evidence":"README 的免责声明：\"While the author is an employee of Google Japan, this software is not Google's product and is not supported by Google. The copyright of this software belongs to me as I have requested it using the IARC process. However, Google might claim the rights to the intellectual property of this invention.\"","source":"ai","title":"作者提醒版权可能会被他的雇主主张"},{"code":"a_paid_iphone_app_is_needed_for_the_live_face_mode","severity":"info","evidence":"README：\"the `ifacialmocap_puppeteer` requires iFacialMocap, which is available in the App Store for 980 yen. You also need to install the paired desktop application on your PC or Mac. (Linux users, I'm sorry!)\" 手机需要有 TrueDepth 前置摄像头（iPhone X 或更新机型），并且要和电脑处于同一网络。","source":"ai","title":"实时人脸模式需要一款付费的 iPhone 应用"},{"code":"strict_limits_on_the_picture_and_the_output","severity":"info","evidence":"README：输入图片 \"must have an alpha channel\"，\"must contain only one humanoid anime character\"，正对镜头，头部位于居中的 128x128 方框内，并且 \"the programs will always resize it to 256x256 and will always output a 256x256 image. This is an intrinsic limitation of the system.\"","source":"ai","title":"对图片和输出有严格限制"}],"platforms":["windows","linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","image"],"alternativeTo":[{"slug":"live2d-cubism","name":"Cubism","confirmed":true}],"declaredAlternatives":[],"stars":1154,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":1154},{"date":"2026-11-08","stars":1154},{"date":"2026-11-09","stars":1154},{"date":"2026-11-10","stars":1154}],"firstPublishedAt":"2026-11-09T23:02:44.176Z","commitSha":"385a82b3e63e9184169fae89367982eb56ddd15e","verifiedAt":"2026-11-09T22:53:37.613Z","lastCommitAt":"2022-06-29T20:26:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Animates the face of a single anime drawing: pose it by hand, or drive it live with your own face from an iPhone.","summary":"This is the demo code for a research project that animates an anime character from one drawing. One program gives you sliders to change the character's expression and head rotation; the other takes your own face, captured by a paid iPhone app, and makes the drawing copy your movements, which is how VTubers work. You run it with Python commands on Windows or Linux and need a powerful NVIDIA graphics card, or you can try the slider version in your browser through Google Colab. The output is always 256x256 pixels.","easeRationale":"The easiest way in is the web: the repository ships a Google Colab notebook you open in the browser and run, with nothing to install, which is a C. On Windows and Linux it is a D: you install Anaconda, build a Python 3.8 environment with PyTorch and the CUDA toolkit, download the model files from a Dropbox link by hand and run the programs from a shell, and the README says both need \"a recent and powerful Nvidia GPU\" such as an RTX 2080 or better. Macs get no grade: there is no NVIDIA card for them, even though the face-capture companion app runs there.","costNote":"The code is MIT and the model files are Creative Commons Attribution 4.0, which the README says allows commercial use if you credit the author. Driving the character with your own face needs the iFacialMocap iPhone app, \"available in the App Store for 980 yen\", plus its free desktop companion.","notVerified":["Whether the Dropbox link still works and what exactly the model zip contains.","Whether it still runs on a current graphics card and driver, given the 2021-era CUDA and PyTorch pins.","Whether the Google Colab notebook still runs, and whether the free Colab tier gives a graphics card big enough.","How smooth the live face-tracking is in practice, and how much delay there is between your face and the drawing.","Whether an anime drawing you did not make yourself can be used this way without the artist's permission."]},"zh-Hans":{"summaryShort":"让单张动漫画面中的角色动起来：用手动调整表情，或用 iPhone 实时用你自己的脸来驱动它。","summary":"这是一个研究项目的演示代码，可以让动漫角色仅凭一张画动起来。一个程序提供滑块，用来调整角色的表情和头部转动；另一个程序用付费 iPhone 应用捕捉你自己的脸，让画面模仿你的动作，这也是 VTuber 的工作原理。你需要在 Windows 或 Linux 上用 Python 命令运行它，并需要一块强大的 NVIDIA 显卡，也可以通过 Google Colab 在浏览器中试用滑块版本。输出图像始终是 256x256 像素。","easeRationale":"最简单的入门方式是通过网页：仓库自带一个 Google Colab 笔记本，你在浏览器中打开并运行即可，不需要安装任何东西，这属于 C 级。在 Windows 和 Linux 上则是 D 级：你需要安装 Anaconda，搭建带有 PyTorch 和 CUDA 工具包的 Python 3.8 环境，手动从 Dropbox 链接下载模型文件，再从命令行运行这些程序，README 说两者都需要 \"a recent and powerful Nvidia GPU\"，比如 RTX 2080 或更好的显卡。Mac 不予评级：Mac 没有 NVIDIA 显卡，尽管用于捕捉面部的配套应用可以在 Mac 上运行。","costNote":"代码采用 MIT 许可，模型文件采用 Creative Commons Attribution 4.0 许可，README 说只要注明作者就可以商业使用。用自己的脸来驱动角色需要 iFacialMocap 这款 iPhone 应用，\"available in the App Store for 980 yen\"，还需要配套的免费桌面端程序。","notVerified":["Dropbox 链接是否仍然有效，模型压缩包里到底装了什么","考虑到 2021 年前后的 CUDA 和 PyTorch 版本限定，它是否还能在现在的显卡和驱动上运行","Google Colab 笔记本是否仍能运行，免费版 Colab 提供的显卡是否够用","实际使用时实时人脸跟踪有多流畅，你的脸和画面之间有多大延迟","如果一张动漫画作不是你自己画的，是否可以在未经画师许可的情况下这样使用"]}}},{"id":"R_kgDOHlQDQA","slug":"pkhungurn__talking-head-anime-3-demo","name":"Talking Head Anime 3 Demo","owner":"pkhungurn","fullName":"pkhungurn/talking-head-anime-3-demo","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"D","linux":"D","web":"C"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://www.dropbox.com/s/y7b8jl4n2euv8xe/talking-head-anime-3-models.zip?dl=0","source":"check","title":"外部下载链接"},{"code":"abandoned_and_pinned_to_old_versions","severity":"attention","evidence":"最近一次提交是在 2023-01-21，比这次检查早了将近四年，也没有任何发行版。environment.yml 固定了 \"python=3.8.13\"、\"pytorch=1.11.0=py3.8_cuda11.3_cudnn8_0\" 和 \"cudatoolkit=11.3.1\"。README 自己也提醒：\"This particular setup might not work in the future because you may find that this particular PyTorch package does not work with your new computer.\"","source":"ai","title":"已经放弃维护，并固定在旧版本上"},{"code":"the_author_warns_the_rights_may_be_claimed_by_his_employer","severity":"attention","evidence":"README 的免责声明：\"While the author is an employee of Google Japan, this software is not Google's product and is not supported by Google. The copyright of this software belongs to me as I have requested it using the IARC process. However, Google might claim the rights to the intellectual property of this invention.\"","source":"ai","title":"作者提醒版权可能会被他的雇主主张"},{"code":"a_paid_iphone_app_is_needed_for_the_live_face_mode","severity":"info","evidence":"README：\"you will also need to an iOS software called iFacialMocap (a 980 yen purchase in the App Store)\"。手机需要有 TrueDepth 前置摄像头（iPhone X 或更新机型），而且 \"Your iOS and your computer must use the same network\"——你把手机的 IP 地址输入程序，程序就会通过你的 Wi-Fi 接收你脸部的动作。","source":"ai","title":"实时人脸模式需要一款付费的 iPhone 应用"},{"code":"strict_limits_on_the_picture","severity":"info","evidence":"README：输入图片 \"should be of resolution 512 x 512\"（其他尺寸会被调整），\"must have an alpha channel\"，\"must contain only one humanoid character\"，站立并正对镜头，双手 \"below and far from the head\"，背景完全透明。","source":"ai","title":"对图片有严格限制"},{"code":"the_example_pictures_have_their_own_licences","severity":"info","evidence":"README 的免责声明：\"Please see the README.md file in the ``data/images`` directory for the licenses for the images there.\" 示例文件名为 crypko_00.png 到 crypko_07.png 以及 lambda_00.png，说明它们来自别处，不受该项目 MIT 许可证的约束。","source":"ai","title":"示例图片有各自的许可证"}],"platforms":["windows","linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","image"],"alternativeTo":[{"slug":"live2d-cubism","name":"Cubism","confirmed":true}],"declaredAlternatives":[],"stars":1050,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":1050},{"date":"2026-11-08","stars":1050},{"date":"2026-11-09","stars":1050},{"date":"2026-11-10","stars":1050}],"firstPublishedAt":"2026-11-09T23:02:44.186Z","commitSha":"8946939ec7b417f443d7b0e3fcd97384313fcdb8","verifiedAt":"2026-11-09T22:54:09.111Z","lastCommitAt":"2023-01-21T04:23:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Animates an anime drawing's face and upper body: pose it by hand, or drive it live with your own face from an iPhone.","summary":"This is the demo code for the third version of a research project that animates an anime character from one drawing. One program gives you sliders for the character's expression, head and body rotation and even breathing; the other takes your own face, captured by a paid iPhone app on your wifi, and makes the drawing copy you, which is how VTubers work. You run it with Python commands on Windows or Linux with a powerful NVIDIA graphics card, or try the slider version in your browser through Google Colab. Pictures are handled at 512x512.","easeRationale":"The easiest way in is the web: the repository ships a Google Colab notebook you open in the browser and run, with nothing to install, which is a C. On Windows and Linux it is a D: you update your graphics driver, install the CUDA Toolkit and Anaconda, build a Python 3.8 environment with PyTorch, download the model files from a Dropbox link by hand and start the programs from a shell, and the README asks for \"a recent and powerful Nvidia GPU\" such as an RTX 2080 or better. Macs get no grade, since there is no NVIDIA card for them.","costNote":"The code is MIT and the model files are Creative Commons Attribution 4.0, which the README says allows commercial use if you credit the author. Driving the character with your own face needs the iFacialMocap iPhone app, \"a 980 yen purchase in the App Store\".","notVerified":["Whether the Dropbox link still works and what exactly the model zip contains.","Whether it still runs on a current graphics card and driver, given the 2022-era CUDA and PyTorch pins.","Whether the Google Colab notebook still runs, and whether the free Colab tier gives a graphics card big enough.","How smooth the live face-tracking is, and how much delay there is between your face and the drawing.","How much video memory each of the four model variants needs, since the README only compares them in general terms.","Whether an anime drawing you did not make yourself can be used this way without the artist's permission."]},"zh-Hans":{"summaryShort":"让动漫画面中角色的脸和上半身动起来：用手动调整，或用 iPhone 实时用你的脸驱动它。","summary":"这是一个研究项目第三个版本的演示代码，可以让动漫角色仅凭一张画动起来。一个程序提供滑块，用来调整角色的表情、头部和身体的转动，甚至呼吸；另一个程序通过你 Wi-Fi 上的付费 iPhone 应用捕捉你自己的脸，让画面模仿你，这也是 VTuber 的工作原理。你需要在配有强大 NVIDIA 显卡的 Windows 或 Linux 上用 Python 命令运行它，也可以通过 Google Colab 在浏览器中试用滑块版本。图片按 512x512 处理。","easeRationale":"最简单的入门方式是通过网页：仓库自带一个 Google Colab 笔记本，你在浏览器中打开并运行即可，不需要安装任何东西，这属于 C 级。在 Windows 和 Linux 上则是 D 级：你需要更新显卡驱动，安装 CUDA Toolkit 和 Anaconda，搭建带有 PyTorch 的 Python 3.8 环境，手动从 Dropbox 链接下载模型文件，再从命令行启动这些程序，README 要求 \"a recent and powerful Nvidia GPU\"，比如 RTX 2080 或更好的显卡。Mac 不予评级，因为 Mac 没有 NVIDIA 显卡。","costNote":"代码采用 MIT 许可，模型文件采用 Creative Commons Attribution 4.0 许可，README 说只要注明作者就可以商业使用。用自己的脸来驱动角色需要 iFacialMocap 这款 iPhone 应用，\"a 980 yen purchase in the App Store\"。","notVerified":["Dropbox 链接是否仍然有效，模型压缩包里到底装了什么","考虑到 2022 年前后的 CUDA 和 PyTorch 版本限定，它是否还能在现在的显卡和驱动上运行","Google Colab 笔记本是否仍能运行，免费版 Colab 提供的显卡是否够用","实时人脸跟踪实际有多流畅，你的脸和画面之间有多大延迟","四种模型变体各自需要多少显存，因为 README 只做了笼统的比较","如果一张动漫画作不是你自己画的，是否可以在未经画师许可的情况下这样使用"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMzA0MzQzMTM=","slug":"pkhungurn__talking-head-anime-demo","name":"Talking Head Anime","owner":"pkhungurn","fullName":"pkhungurn/talking-head-anime-demo","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"D","linux":"D","web":"C"},"risk":"attention","signals":[{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/open?id=1ajHViqyLDKFKfBtGPE5cbSGcMNa8rz8k","source":"check","title":"外部下载链接"},{"code":"abandoned_and_pinned_to_very_old_versions","severity":"attention","evidence":"最近一次提交是在 2022-06-29，比这次检查早了四年多，也没有任何发行版。environment.yml 固定了 2019 年的 \"python=3.8.1\"、\"pytorch=1.4.0=py3.8_cuda101_cudnn7_0\"、\"cudatoolkit=10.1.243\" 和 \"dlib=19.19\"，甚至还保留着作者自己的文件夹路径（\"prefix: C:\\tools\\Anaconda3\\envs\\talking-head-anime\"）。","source":"ai","title":"已经放弃维护，并固定在非常旧的版本上"},{"code":"the_author_released_it_without_his_employers_approval","severity":"attention","evidence":"README 的免责声明：\"one of the condition for the release of this source code is that the publication of the \\\"Talking Head Anime from a Single Image\\\" be approved by the internal publication approval process. I requested approval on 2019/11/17. It has been reviewed by a researcher, but has not been formally approved by a manager in my product area (Google Maps). I have decided to release this code, bearing all the risks that it may incur.\"","source":"ai","title":"作者在未获雇主批准的情况下发布了它"},{"code":"the_author_points_to_newer_versions","severity":"info","evidence":"README 第一行：\"You may want to check out the much more capable Version 2 of the same software: http://github.com/pkhungurn/talking-head-anime-2-demo\"。还有一个第三版，也能让身体动起来。","source":"ai","title":"作者指向了更新的版本"},{"code":"it_watches_you_through_the_webcam","severity":"info","evidence":"README：\"The *puppeteer* makes an anime character imitate the head movement of the human capture by a webcam feed\" 以及 \"the peppeteer tool requires a webcam\"。人脸跟踪使用了 dlib 的 shape_predictor_68_face_landmarks.dat 和 \"a face tracker code implemented by KwanHua Lee\"；README 中没有任何地方说视频会离开你的电脑。","source":"ai","title":"它通过摄像头观察你"},{"code":"strict_limits_on_the_picture","severity":"info","evidence":"README：图片 \"must be in PNG format\"，\"must be of size 256 x 256\"，头部必须在居中的 128x128 方框内，需要 4 个通道（RGBA）和完全透明的背景。data/illust 中附带了五张示例图片。","source":"ai","title":"对图片有严格限制"}],"platforms":["windows","linux","web"],"tools":["python","gpu_nvidia"],"buildTools":["python"],"disk":{"minMB":160,"maxMB":420,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","image"],"alternativeTo":[{"slug":"live2d-cubism","name":"Cubism","confirmed":true}],"declaredAlternatives":[],"stars":2027,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":2027},{"date":"2026-11-06","stars":2027},{"date":"2026-11-07","stars":2027},{"date":"2026-11-08","stars":2027},{"date":"2026-11-09","stars":2027},{"date":"2026-11-10","stars":2027}],"firstPublishedAt":"2026-11-09T23:02:44.196Z","commitSha":"bfbf7425748c4559451b95cdc39873ae609d8590","verifiedAt":"2026-11-09T22:54:47.246Z","lastCommitAt":"2022-06-29T20:27:15Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Animates an anime drawing's face: pose it with sliders, or let it copy your head movements through your webcam.","summary":"This is the demo code for the first version of a research project that animates an anime character from a single drawing. One program poses the character's face with sliders; the other uses your webcam so the drawing copies your head movements, which is how VTubers work. You run it with Python commands on Windows or Linux and need a powerful NVIDIA graphics card, or you can try the slider version in your browser through Google Colab. Pictures must be 256x256 PNG files with a transparent background, and the author points to newer versions of the project.","easeRationale":"The easiest way in is the web: the repository ships a Google Colab notebook you open in the browser and run, with nothing to install, which is a C. On Windows and Linux it is a D: you install Anaconda and build a Python environment with PyTorch, dlib and OpenCV (\"this will require several gigabytes of your storage\"), download the models from a Google Drive link plus a face-landmark file from another repository, and start the programs from a shell; the README says it needs \"a recent and powerful Nvidia GPU\". Macs get no grade, since there is no NVIDIA card for them.","costNote":"The code is MIT and the model files are Creative Commons Attribution 4.0. No paid parts: the live mode uses an ordinary webcam.","notVerified":["Whether the Google Drive link still works and what exactly the model zip contains.","Whether it still runs on a current graphics card and driver, given the 2019-era CUDA, PyTorch and dlib pins.","Whether the Google Colab notebook still runs, and whether the free Colab tier gives a graphics card big enough.","How smooth the webcam tracking is, and how much delay there is between your head and the drawing.","Whether an anime drawing you did not make yourself can be used this way without the artist's permission."]},"zh-Hans":{"summaryShort":"让动漫画面中的脸动起来：用滑块调整，或通过摄像头让它模仿你的头部动作。","summary":"这是一个研究项目第一个版本的演示代码，可以让动漫角色仅凭一张画动起来。一个程序用滑块来摆出角色的表情；另一个程序用你的摄像头，让画面模仿你的头部动作，这也是 VTuber 的工作原理。你需要在 Windows 或 Linux 上用 Python 命令运行它，并需要一块强大的 NVIDIA 显卡，也可以通过 Google Colab 在浏览器中试用滑块版本。图片必须是 256x256、背景透明的 PNG 文件，作者也提到了这个项目更新的版本。","easeRationale":"最简单的入门方式是通过网页：仓库自带一个 Google Colab 笔记本，你在浏览器中打开并运行即可，不需要安装任何东西，这属于 C 级。在 Windows 和 Linux 上则是 D 级：你需要安装 Anaconda，搭建带有 PyTorch、dlib 和 OpenCV 的 Python 环境（\"this will require several gigabytes of your storage\"），从 Google Drive 链接下载模型，再加上另一个仓库里的人脸关键点文件，然后从命令行启动这些程序；README 说需要 \"a recent and powerful Nvidia GPU\"。Mac 不予评级，因为 Mac 没有 NVIDIA 显卡。","costNote":"代码采用 MIT 许可，模型文件采用 Creative Commons Attribution 4.0 许可。没有付费部分：实时模式用的是普通摄像头。","notVerified":["Google Drive 链接是否仍然有效，模型压缩包里到底装了什么","考虑到 2019 年前后的 CUDA、PyTorch 和 dlib 版本限定，它是否还能在现在的显卡和驱动上运行","Google Colab 笔记本是否仍能运行，免费版 Colab 提供的显卡是否够用","摄像头跟踪实际有多流畅，你的头部和画面之间有多大延迟","如果一张动漫画作不是你自己画的，是否可以在未经画师许可的情况下这样使用"]}}},{"id":"R_kgDORchPVA","slug":"pku-yuangroup__helios","name":"Helios","owner":"PKU-YuanGroup","fullName":"PKU-YuanGroup/Helios","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"linux":"D","web":"A"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):torch: GHSA-rrmf-rvhw-rf47, PYSEC-2026-139; transformers: GHSA-fgcw-684q-jj6r, GHSA-x9r9-c232-4q39, GHSA-xrqw-3rrv-vx5w; sentence-transformers: GHSA-jhr6-gm9c-rqjv, PYSEC-2026-4164; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; torch: GHSA-63cw-57p8-fm3p, GHSA-qfhq-4f3w-5fph, GHSA-rrmf-rvhw-rf47","source":"check","title":"已知漏洞"},{"code":"the_install_script_removes_packages_and_deletes_folders","severity":"attention","evidence":"install.sh会运行\"pip uninstall triton torchao xformers wandb tensorflow tensorflow-cpu -y\",然后运行\"rm -rf ~/.triton/cache/\"和\"rm -rf /tmp/torchinductor_*\"。它会从当前激活的Python环境中卸载六个软件包,包括TensorFlow,还会删除项目之外的文件夹,导致同一台机器上的其他AI项目可能无法正常运行。","source":"ai","title":"安装脚本会卸载软件包并删除文件夹"},{"code":"the_speed_claim_needs_a_data_centre_card","severity":"info","evidence":"README写道:\"minute-scale, high-quality video synthesis at 19.5 FPS on a single H100 GPU (about 10 FPS on a single Ascend NPU)\",以及\"fitting up to four 14B models within 80 GB of GPU memory\"。低内存方案则说\"The Helios model below requires ~6GB of VRAM\",通过group offloading实现,一条更新说明还补充\"real-time generation performance depends not only on the GPU, but also on the CPU, memory, CUDA driver version\"。","source":"ai","title":"宣称的速度需要数据中心级显卡"},{"code":"the_models_are_downloaded_separately_and_are_large","severity":"info","evidence":"仓库本身不包含权重:你需要用\"huggingface-cli download BestWishYSH/Helios-Base\"从Hugging Face或ModelScope获取Helios-Base、Helios-Mid或Helios-Distilled。这些是140亿参数的视频模型,所以实际占用的磁盘空间远远超过检查从仓库估算出的1到4.5GB。","source":"ai","title":"模型需要单独下载,而且体积很大"},{"code":"research_code_with_parts_still_unfinished","severity":"info","evidence":"没有发布版本,有47个未关闭的issue,交互模式会提示\"⚠️ This feature is still under development — results may not always meet expectations\",Helios-Mid检查点\"may not meet expected quality\",一条更新说明还提到作者\"identified several ways to further improve Helios's performance, such as fixing the i2v train-inference inconsistency\"。","source":"ai","title":"研究代码仍有部分未完成"},{"code":"install_steps_pull_code_straight_from_git_branches","severity":"info","evidence":"有几条安装路径会从主分支安装尚未发布的代码:\"pip install git+https://github.com/huggingface/diffusers.git\"、\"pip install git+https://github.com/vllm-project/vllm-omni.git\"和\"pip install git+https://github.com/sgl-project/sglang.git\"。你安装当天得到的代码内容会不一样。","source":"ai","title":"安装步骤直接从git分支拉取代码"}],"platforms":["linux","web"],"tools":[],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":1010,"maxMB":4570,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"kling","name":"Kling","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":2258,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":11,"growthDays":6,"history":[{"date":"2026-11-08","stars":2256},{"date":"2026-11-09","stars":2258},{"date":"2026-11-10","stars":2258}],"firstPublishedAt":"2026-11-09T23:02:44.206Z","commitSha":"f1a8aadd7c42df29968b9ed73b5c3bc6ae2980c4","verifiedAt":"2026-11-09T22:55:50.329Z","lastCommitAt":"2026-11-01T09:00:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes minute-long videos from a text prompt, a photo or another video, with a model that runs on your own graphics card.","summary":"Helios is the code and the model weights from a university research paper on long video generation. You give it a text prompt, a picture or a video and it produces a clip that can run for about a minute, at 384x640, in chunks of 33 frames. The authors report 19.5 frames per second on a single H100 card, and a slower low-memory mode that fits in about 6 GB. You can try it in your browser on the authors' Hugging Face demo page, or install it yourself with conda and shell commands on a Linux machine with an NVIDIA card or a Huawei Ascend NPU.","easeRationale":"The quickest look is the authors' own Hugging Face demo page linked at the top of the README, which needs nothing installed, so the web is an A. Running it yourself is a D: you clone the repository, create a conda environment, pick the PyTorch build that matches your CUDA version, run \"bash install.sh\", download tens of gigabytes of weights with the Hugging Face or ModelScope command-line tool and start shell scripts. Windows and Mac get no grade: every step is a Linux shell script and the only pointer for an ordinary PC is an unofficial community video.","costNote":"The code is Apache-2.0 and the three model versions are free downloads from Hugging Face or ModelScope. Running it fast needs an expensive data-centre card, and the README points at CNAPS.AI, an outside online service that runs Helios for you.","notVerified":["How long one minute of video really takes, and how slow the 6 GB low-memory mode is on an ordinary gaming card.","Real disk space needed for the three model versions plus the Python environment.","Whether the Hugging Face demo page is up and how long the queue is.","Whether the install works at all on Windows or a Mac, since no official steps exist for them.","Picture quality with your own prompts, since the examples and the benchmark are the authors' own.","What the CNAPS.AI online service charges and what it does with the videos you make there."]},"zh-Hans":{"summaryShort":"根据文字提示、一张照片或另一段视频,生成约一分钟的视频,模型在你自己的显卡上运行。","summary":"Helios 是一篇大学研究论文关于长视频生成的代码和模型权重。你给它一段文字提示、一张照片或一段视频,它就能生成约一分钟长的片段,分辨率为384x640,每33帧一组。作者报告在单张H100显卡上可达每秒19.5帧,还有一个更慢的低内存模式,大约6GB即可运行。你可以在浏览器中试用作者在Hugging Face上的演示页面,也可以在带有NVIDIA显卡或华为Ascend NPU的Linux机器上,用conda和shell命令自己安装。","easeRationale":"最快的方式是作者在README顶部链接的Hugging Face演示页面,不需要安装任何东西,所以网页版得A。自己运行是D:你需要克隆仓库、创建conda环境、选择匹配你CUDA版本的PyTorch版本、运行\"bash install.sh\"、用Hugging Face或ModelScope的命令行工具下载数十GB的权重,并启动shell脚本。Windows和Mac没有评级:每一步都是Linux shell脚本,普通PC唯一的参考只有一个非官方的社区视频。","costNote":"代码采用Apache-2.0许可,三个模型版本都可以从Hugging Face或ModelScope免费下载。想要运行得快需要一块昂贵的数据中心显卡,README还提到了CNAPS.AI,一个代为运行Helios的外部在线服务。","notVerified":["生成一分钟视频实际需要多长时间,以及6GB低内存模式在普通游戏显卡上有多慢。","三个模型版本加上Python环境实际占用的磁盘空间。","Hugging Face演示页面是否可用,排队要多久。","在Windows或Mac上是否能安装成功,因为官方没有提供相应步骤。","用你自己的提示词生成的画面质量如何,因为示例和基准测试都是作者自己做的。","CNAPS.AI在线服务收费多少,以及它如何处理你在那里生成的视频。"]}}},{"id":"R_kgDOICwyOg","slug":"playvoice__whisper-vits-svc","name":"Whisper-VITS-SVC","owner":"PlayVoice","fullName":"PlayVoice/whisper-vits-svc","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布是在2023-10-05(1131天前)","source":"check","title":"版本发布已过时"},{"code":"external_download_links","severity":"attention","evidence":"README链接到https://drive.google.com/drive/folders/15oeBYf6Qn1edONkVLXe82MzdIi3O_9m3","source":"check","title":"外部下载链接"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-34rf-p3r3-58x2","source":"check","title":"已知漏洞"},{"code":"project_no_longer_updated","severity":"attention","evidence":"README.md写道\"No More Upgrade\";最后一次提交是在2024-04-23(距检查时930天前)。","source":"ai","title":"项目已不再更新"},{"code":"voice_cloning_of_real_people","severity":"attention","evidence":"该仓库的描述是\"Core Engine of Singing Voice Conversion & Singing Voice Clone\",README中列出了\"SNAC | Samsung | One Shot Clone of VITS\";在一些国家,复制真人歌手的声音受到限制,可能需要本人许可。","source":"ai","title":"克隆真人声音"},{"code":"third_party_package_mirror","severity":"info","evidence":"README.md中的安装步骤:\"pip install -i https://pypi.tuna.tsinghua.edu.cn/simple -r requirements.txt\"会从中国的一个镜像源安装Python软件包,而不是官方源。","source":"ai","title":"第三方软件包镜像"},{"code":"models_from_file_sharing_and_external_sites","severity":"info","evidence":"README.md要求你从一个Google Drive文件夹获取best_model.pth.tar,从openaipublic.azureedge.net获取large-v2.pt,从另一个GitHub仓库获取full.pth;没有人替你检查这些文件。","source":"ai","title":"模型来自文件分享和外部网站"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_nvidia","ffmpeg"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"kits-ai","name":"Kits AI","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true}],"declaredAlternatives":[],"stars":2864,"downloads":15010,"downloadsWeek":33,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":2864},{"date":"2026-11-07","stars":2864},{"date":"2026-11-08","stars":2864},{"date":"2026-11-09","stars":2864},{"date":"2026-11-10","stars":2864}],"firstPublishedAt":"2026-11-09T23:02:44.216Z","commitSha":"b95a84954d74f651982c0ecea5f7eb67f9c02d46","verifiedAt":"2026-11-09T22:45:58.045Z","lastCommitAt":"2024-04-23T10:50:32Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Changes a recorded singing voice into another singer's voice, using voice models you train on your own computer.","summary":"An engine for singing voice conversion: you record a song, and it re-sings it with the voice of a singer whose model you trained from audio files. It also mixes voices to create a new one. It is aimed at people learning deep learning: the README says basic Python and PyTorch knowledge are required and that the authors will not make a one-click package. You need an NVIDIA graphics card (6 GB of video memory for training), Python, FFmpeg, several model files downloaded by hand, and a long list of commands. It does not work in real time.","easeRationale":"There is no installer on any platform: on Windows and Linux alike you install Python and PyTorch, download five separate model files by hand (speaker encoder, whisper large-v2, HuBERT, crepe, pretrained model) and run many commands to prepare data, train and convert. Training needs an NVIDIA card with at least 6 GB of video memory, and the README states no one-click package will be made.","costNote":"No paid service is needed; the models and code are free.","notVerified":["Whether the five model downloads listed in the README are still available, since the project stopped being updated in 2024.","Real disk use: the checks estimate 150-450 MB for the code, but the whisper large-v2 file alone is a few gigabytes and the prepared training data grows with your audio.","Whether training really fits in 6 GB of video memory, and how many hours it takes on a normal home graphics card.","Whether it works on Windows: the included environment.yml lists Linux packages only.","The quality of the converted singing voice, which depends on your own recordings."]},"zh-Hans":{"summaryShort":"把一段录好的演唱转换成另一位歌手的声音,使用你在自己电脑上训练的语音模型。","summary":"一个歌声转换引擎:你录一段歌,它用你用音频文件训练出的某个歌手的模型重新演唱。它还能混合声音来创造一个新的声音。它面向正在学习深度学习的人:README说需要基础的Python和PyTorch知识,作者也不会做一键安装包。你需要一块NVIDIA显卡(训练需要6GB显存)、Python、FFmpeg、几个手动下载的模型文件,以及一长串命令。它不支持实时运行。","easeRationale":"任何平台都没有安装程序:无论是Windows还是Linux,你都要安装Python和PyTorch,手动下载五个独立的模型文件(speaker encoder、whisper large-v2、HuBERT、crepe、预训练模型),并运行大量命令来准备数据、训练和转换。训练需要一块至少有6GB显存的NVIDIA显卡,README中说明不会制作一键安装包。","costNote":"不需要任何付费服务;模型和代码都是免费的。","notVerified":["README中列出的五个模型下载是否仍然有效,因为该项目在2024年就停止更新了。","实际磁盘占用:检查估算代码部分为150-450MB,但仅whisper large-v2文件就有几个GB,准备好的训练数据还会随你的音频增长。","训练是否真的能在6GB显存内完成,在普通家用显卡上要花多少个小时。","是否能在Windows上运行:附带的environment.yml只列出了Linux的软件包。","转换后的演唱质量如何,这取决于你自己的录音。"]}}},{"id":"R_kgDOQKxyzA","slug":"pnnbao97__vieneu-tts","name":"VieNeu-TTS","owner":"pnnbao97","fullName":"pnnbao97/VieNeu-TTS","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接传给shell执行:curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"curl管道直接执行shell"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; aiohttp: GHSA-2fqr-mr3j-6wp8, GHSA-2g87-pp6c-29x5, GHSA-2vrm-gr82-f7m5; anyio: GHSA-5p39-cfhj-2xmp, GHSA-82r6-8w77-94w6, PYSEC-2026-4024; brotli: GHSA-2qfp-q593-8484, GHSA-9v8h-qqgm-4m35, PYSEC-2026-2401; certifi: GHSA-6q48-6c7w-9gw2, PYSEC-2026-4320; datasets: GHSA-379c-qx7v-6h59, PYSEC-2026-3716; filelock: GHSA-69hj-f68j-2879, GHSA-8mx5-7m5m-r77g, GHSA-qmgc-5h2g-mvrw; fsspec: GHSA-27vj-qcqg-25rc, PYSEC-2026-4217; gradio: GHSA-39mp-8hj3-5c49, GHSA-6655-8ph2-63j3, GHSA-7hp7-4p35-3cx2; idna: GHSA-65pc-fj4g-8rjx, PYSEC-2026-215; lxml: GHSA-vfmq-68hx-4jfw, PYSEC-2026-87; msgpack: GHSA-6v7p-g79w-8964, GHSA-j586-36cw-2gc2, PYSEC-2026-3625; multidict: GHSA-54p9-h82j-f925, PYSEC-2026-4244; orjson: GHSA-hx9q-6w63-j58v, PYSEC-2026-107; pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h; protobuf: GHSA-7gcm-g887-7qv7, PYSEC-2026-1805; pyarrow: GHSA-rgxp-2hwp-jwgg, PYSEC-2026-113; pygments: GHSA-5239-wwwm-4pmq, PYSEC-2026-2987; python-multipart: GHSA-5rvq-cxj2-64vf, GHSA-6jv3-5f52-599m, GHSA-mj87-hwqh-73pj; ray: GHSA-hhrp-gw25-jr43, GHSA-mw35-8rx3-xf9r, GHSA-q5fh-2hc8-f6rq; starlette: GHSA-82w8-qh3p-5jfq, GHSA-86qp-5c8j-p5mr, GHSA-jp82-jpqv-5vv3; transformers: GHSA-29pf-2h5f-8g72, GHSA-69w3-r845-3855, GHSA-fgcw-684q-jj6r","source":"check","title":"已知漏洞"},{"code":"voice_cloning_of_real_people","severity":"attention","evidence":"README宣传\"Instant voice cloning from a 3-8s clip\",作者没有将v4版本开源,理由是\"Due to the strength of its voice-cloning capabilities and the potential for misuse\";pyproject.toml显示音频水印是可选的(\"audio watermarking is opt-in via pip install vieneu[watermark]\"),所以生成的音频通常不带水印。","source":"ai","title":"克隆真人声音"},{"code":"optional_remote_code_execution_setting","severity":"attention","evidence":".env.example中有VIENEU_TRUST_REMOTE_CODE=1,说明写的是\"Chỉ bật nếu tin tưởng repo nhập ở Custom Model: cho phép repo đó chạy code Python riêng\"(只有在信任你在Custom Model中填写的仓库时才打开:它会允许该仓库运行自己的Python代码)。","source":"ai","title":"可选的远程代码执行设置"},{"code":"local_web_ui_without_password_by_default","severity":"info","evidence":".env.example中登录设置是被注释掉的(\"# VIENEU_WEB_AUTH=admin:doi-mat-khau-nay\"),所以除非你自己设置密码,否则7860端口上的Gradio页面和流式API都没有密码;如果你在Docker中运行或在共享网络上运行,这一点需要注意。","source":"ai","title":"本地网页界面默认没有密码"},{"code":"newest_version_not_open_source","severity":"info","evidence":"README写道:\"VieNeu-TTS v4 ... is proprietary and will not be open-sourced. It is available exclusively through the VieNeu API / vieneu.io\",以及\"Dubbing, Lecture, Tiktok Reel and Audiobook features are exclusively available in the official VieNeu App\"。","source":"ai","title":"最新版本并非开源"},{"code":"optional_public_tunnel_in_old_docker_image","severity":"info","evidence":"已弃用的v2服务器说明中运行了\"docker run --gpus all -p 23333:23333 ... pnnbao/vieneu-tts:latest --tunnel\",这会打开一个内置的bore隧道,让你的服务器在互联网上拥有一个公开地址。","source":"ai","title":"旧版docker镜像中的可选公开隧道"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":90,"maxMB":270,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":3188,"downloads":1837,"downloadsWeek":127,"downloadsDays":4,"growth":89,"growthDays":6,"history":[{"date":"2026-11-06","stars":3134},{"date":"2026-11-07","stars":3155},{"date":"2026-11-08","stars":3165},{"date":"2026-11-09","stars":3176},{"date":"2026-11-10","stars":3188}],"firstPublishedAt":"2026-11-09T23:02:44.225Z","commitSha":"22dd2ef2a89bba24a77da6dbc84c75f7f4fda05b","verifiedAt":"2026-11-09T22:47:12.752Z","lastCommitAt":"2026-11-09T14:42:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Reads Vietnamese text aloud in 25 ready-made voices and can copy a voice from a 3-8 second clip, on your own computer.","summary":"A Vietnamese text-to-speech program: you paste text and it speaks it with one of 25 built-in voices (northern, central and southern accents), or with a voice copied from a 3-8 second recording. It also does two-speaker conversations and can stream audio to other apps through an OpenAI-style API. Windows has a ready installer; on Mac and Linux you install the uv tool and run a few commands to open a local web page. It runs on a plain processor (about twice real time) and much faster with an NVIDIA card. Text and voices stay on your machine.","easeRationale":"On Windows there is a standalone installer in the releases (VieNeu_0.18.22_x64-setup.exe, 90 MB), so you install and open it; the 90 MB file cannot hold the voice models, so the app fetches them from Hugging Face on first use. On Mac and Linux the README says the installer is not ready yet: you install the uv tool, run \"uv sync\" and \"uv run vieneu-web\" in a terminal and then open a local web page.","costNote":"Everything in this repository is free under Apache-2.0. The newer v4 model is not open source and is offered only through the author's own service, vieneu.io.","notVerified":["Whether the Windows installer really works without any extra steps, and how big the voice model download is on first use (the checks count only the 90 MB installer).","How close a cloned voice sounds to the original, and how good the English parts of bilingual text are.","The speed numbers in the README (first audio in about 115 ms, 16 streams on an RTX 3060) were measured by the author on one machine.","Whether the Mac and Linux terminal route works today, since there is no installer for them yet.","Whether the app or the installer contacts the author's service (vieneu.io) while you use it."]},"zh-Hans":{"summaryShort":"用25种内置声音朗读越南语文字,还能用一段3到8秒的录音复刻一个声音,全部在你自己的电脑上运行。","summary":"一款越南语文字转语音程序:粘贴文字后,它会用25种内置声音之一(北部、中部、南部口音)朗读出来,也可以用一段3到8秒的录音复刻出的声音朗读。它还支持双人对话,并能通过类似OpenAI的API把音频传给其他应用。Windows有现成的安装程序;在Mac和Linux上,你需要安装uv工具并运行几条命令来打开一个本地网页。它在普通处理器上运行(大约是实时速度的两倍),配合NVIDIA显卡会快得多。文字和声音都留在你自己的设备上。","easeRationale":"在Windows上,releases里有一个独立安装程序(VieNeu_0.18.22_x64-setup.exe,90MB),安装打开即可;这90MB的文件装不下语音模型,所以应用会在首次使用时从Hugging Face下载它们。在Mac和Linux上,README说安装程序还没准备好:你需要安装uv工具,在终端运行\"uv sync\"和\"uv run vieneu-web\",然后打开一个本地网页。","costNote":"这个仓库里的一切都是在Apache-2.0许可下免费提供的。更新的v4模型并非开源,只能通过作者自己的服务vieneu.io使用。","notVerified":["Windows安装程序是否真的无需额外步骤就能运行,以及首次使用时语音模型下载有多大(检查只统计了90MB的安装程序)。","克隆出的声音与原声有多接近,以及双语文本中英语部分的效果如何。","README中的速度数据(首个音频约115毫秒,RTX 3060上16路并发)是作者在一台机器上测出来的。","Mac和Linux的终端安装方式现在是否真的能用,因为目前还没有给它们的安装程序。","你在使用时,应用或安装程序是否会联系作者的服务(vieneu.io)。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1MDU2NTQzMA==","slug":"pnpm__pnpm","name":"pnpm","owner":"pnpm","fullName":"pnpm/pnpm","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"part_of_the_repository_is_not_open_source","severity":"attention","evidence":"README的License部分：\"MIT, except the pnpr/ directory, which is source-available under the PolyForm Shield License 1.0.0\"。pnpr文件夹里是一个包注册服务器（Cargo.toml中列出的pnpr crate）；它的代码可以查看，但PolyForm Shield不是开源许可证，并限制与该项目竞争的使用方式。","source":"ai","title":"仓库的一部分不是开源的"},{"code":"it_downloads_runtimes_and_package_code_by_itself","severity":"attention","evidence":"包管理器会从注册表获取其他人的代码，并在你的机器上运行他们的安装脚本。这个仓库自己的package.json就展示了这种模式：devEngines中声明了\"packageManager\"为pnpm 12.16.1、\"runtime\"为node 26.13.0，两者都带有\"onFail\": \"download\"，所以当缺少这些组件时，工具会自动下载包管理器和Node.js运行时。","source":"ai","title":"自行下载运行时和包代码"},{"code":"two_major_lines_and_an_in_progress_rust_rewrite","severity":"info","evidence":"README提到项目\"Includes pacquet, an experimental port of the CLI written in Rust\"，该仓库的主要语言现在是Rust，并且在2026-10-30同时发布了v12.17.0和v11.28.10。版本12的下载包大小约为22 MB，而版本11约为50 MB，所以在1200个release中，你拿到的程序会因所选版本线而有很大差异。","source":"ai","title":"存在两条主线版本以及正在进行的Rust重写"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","rust"],"disk":{"minMB":23,"maxMB":69,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":37059,"downloads":46213358,"downloadsWeek":2034551,"downloadsDays":7,"growth":57,"growthDays":7,"history":[{"date":"2026-10-26","stars":36929},{"date":"2026-10-27","stars":36934},{"date":"2026-10-28","stars":36948},{"date":"2026-10-29","stars":36957},{"date":"2026-10-30","stars":36970},{"date":"2026-10-31","stars":36977},{"date":"2026-11-01","stars":36988},{"date":"2026-11-02","stars":36997},{"date":"2026-11-03","stars":37002},{"date":"2026-11-04","stars":37005},{"date":"2026-11-05","stars":37007},{"date":"2026-11-06","stars":37016},{"date":"2026-11-07","stars":37025},{"date":"2026-11-08","stars":37035},{"date":"2026-11-09","stars":37045},{"date":"2026-11-10","stars":37059}],"firstPublishedAt":"2026-10-27T01:10:12.674Z","commitSha":"b003905dd9ad40d9e9bacf1be635f7527eb2b907","verifiedAt":"2026-11-01T01:45:49.916Z","lastCommitAt":"2026-11-10T22:53:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A faster package manager for JavaScript projects that saves disk space by keeping one copy of each file for all projects.","summary":"pnpm is a command-line package manager for JavaScript and Node.js projects: it installs a project's dependencies, keeps a lock file so everyone gets the same versions, and manages groups of projects in one repository. Instead of copying the same files into every project, it stores each file once and links it, which the project says saves gigabytes and makes installs up to twice as fast. It is for developers; it runs in a terminal on Windows, macOS and Linux, and recent versions can also install Node.js itself.","easeRationale":"Each release has a ready standalone program for Windows (.zip), macOS and Linux (.tar.gz), around 21 to 27 MB, so there is nothing to compile; the extra step is putting it where your terminal can find it, or following the install page at pnpm.io, and from then on everything happens with typed commands inside a code project. Installing it the other common way, through npm, assumes Node.js is already on the machine.","costNote":"Free and MIT-licensed, funded through sponsors on OpenCollective. The pnpr/ folder in the same repository is under a different, non-open-source licence.","notVerified":["Whether installs really are up to twice as fast, which depends on your project, disk and network","How much disk space you save, since that depends on how many projects share the same dependencies","What the packages it installs do once their install scripts run","Disk use of the program itself, estimated at 23 to 69 MB, not counting the shared store it builds","Whether the experimental Rust version behaves the same as the older one on your project"]},"zh-Hans":{"summaryShort":"一个更快的JavaScript项目包管理器，通过为所有项目保留每个文件的单一副本来节省磁盘空间。","summary":"pnpm是一个面向JavaScript和Node.js项目的命令行包管理器：它安装项目的依赖，保留一个锁文件以确保每个人得到相同的版本，并在一个仓库中管理多个项目组。它不会把相同的文件复制到每个项目中，而是只存储每个文件一次并链接它，项目方表示这样可以节省数GB空间，并让安装速度最多提升一倍。它是面向开发者的工具；可以在Windows、macOS和Linux的终端中运行，最近的版本还能自行安装Node.js。","easeRationale":"每个release都有一个现成的独立程序，适用于Windows（.zip）、macOS和Linux（.tar.gz），大小约21到27 MB，因此没有任何需要编译的东西；额外的步骤是把它放到你的终端能找到的位置，或者按照pnpm.io上的安装页面操作，从那以后，一切都通过在代码项目中键入命令来完成。另一种常见的安装方式是通过npm安装，这默认机器上已经装有Node.js。","costNote":"免费且采用MIT许可证，通过OpenCollective上的赞助者获得资金支持。同一仓库中的pnpr/文件夹采用另一种不同的、非开源的许可证。","notVerified":["安装速度是否真的能提升到两倍，这取决于你的项目、磁盘和网络","你能节省多少磁盘空间，这取决于有多少项目共享相同的依赖","它安装的包在运行安装脚本后会做什么","程序本身的磁盘占用，估计为23到69 MB，不包括它建立的共享存储区","实验性的Rust版本在你的项目中是否和旧版本表现一致"]}}},{"id":"R_kgDOJCp3Ww","slug":"pot-app__pot-desktop","name":"Pot","owner":"pot-app","fullName":"pot-app/pot-desktop","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"B","linux":"A"},"risk":"attention","signals":[{"code":"archived","severity":"info","evidence":"该仓库已被归档(只读,不再维护)","source":"check","title":"已归档"},{"code":"no_longer_maintained","severity":"attention","evidence":"该仓库已被归档,最新版本3.0.7发布于2025-05-10,所以这里不会修复漏洞或失效的翻译服务。","source":"ai","title":"不再维护"},{"code":"local_http_control_port","severity":"attention","evidence":"README写道:\"Pot 提供了完整的 HTTP 接口\",默认在127.0.0.1的60828端口上,有像GET /ocr_recognize和GET /ocr_translate这样的接口;同一台电脑上运行的任何程序都可以调用它们,让pot截取你屏幕的一部分。","source":"ai","title":"本地HTTP控制端口"},{"code":"text_sent_to_the_chosen_online_service","severity":"info","evidence":"支持的服务包括OpenAI、Gemini、智谱AI、阿里巴巴、百度、腾讯、火山引擎、有道、DeepL、Google、Bing和Yandex;你选中或截图的内容会发送到你选择的服务,除非你使用系统自带的离线OCR、Tesseract或Ollama。","source":"ai","title":"文字会发送到你选择的在线服务"},{"code":"plugins_installed_from_outside_the_app","severity":"info","evidence":"额外的服务以.potext插件文件的形式提供,你需要从插件仓库下载并手动安装;故障排查说明(\"找不到指定的模块\"、\"不是有效的 Win32 应用程序\")表明这些插件包含为你的系统编译的程序。","source":"ai","title":"插件来自应用之外"},{"code":"linux_graphics_crash_warning","severity":"info","evidence":"README警告说,在Webkit2Gtk 2.42.0和Nvidia专有驱动下,应用会启动失败或崩溃,并要求设置WEBKIT_DISABLE_DMABUF_RENDERER=1。","source":"ai","title":"Linux图形崩溃警告"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":178,"maxMB":534,"basis":"installer"},"apiKeys":["OpenAI","Google Gemini","DeepL","Baidu Translate / Baidu OCR","Tencent Translate / Tencent OCR"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[{"slug":"deepl-pro","name":"Pro","confirmed":true},{"slug":"immersive-translate","name":"Immersive Translate","confirmed":true},{"slug":"abbyy-finereader","name":"FineReader","confirmed":true}],"declaredAlternatives":[],"stars":19334,"downloads":1783364,"downloadsWeek":19168,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":19337},{"date":"2026-11-08","stars":19334},{"date":"2026-11-09","stars":19333},{"date":"2026-11-10","stars":19334}],"firstPublishedAt":"2026-11-09T23:02:44.235Z","commitSha":"594d32ede96acd106b0256deaa8bb440ffcdff40","verifiedAt":"2026-11-09T22:48:03.071Z","lastCommitAt":"2026-07-04T11:01:21Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"pot_3.0.7_x64-setup.exe","sizeBytes":34105974,"url":"https://github.com/pot-app/pot-desktop/releases/download/3.0.7/pot_3.0.7_x64-setup.exe","releaseUrl":"https://github.com/pot-app/pot-desktop/releases/tag/3.0.7"},{"os":"linux","file":"pot_3.0.7_amd64.AppImage","sizeBytes":116344000,"url":"https://github.com/pot-app/pot-desktop/releases/download/3.0.7/pot_3.0.7_amd64.AppImage","releaseUrl":"https://github.com/pot-app/pot-desktop/releases/tag/3.0.7"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Select text anywhere on your computer, press a shortcut and see the translation; it also reads text out of screenshots.","summary":"A small desktop program for translating and reading text. You select words in any app and press a shortcut, type text in its window, or grab part of the screen so it reads the text in the picture and translates it. You choose which service does the work: free ones like Google and Bing, paid ones like DeepL or OpenAI with your own key, the offline text reader built into Windows and Mac, or a local model through Ollama. Installers exist for Windows, Mac and Linux. The repository was archived, so it is no longer being updated.","easeRationale":"On Windows you download the .exe installer (or run \"winget install Pylogmon.pot\") and open it, and on Linux there are .deb, .rpm, AppImage and Flathub packages, so both are download-and-open. Mac costs one extra step: the app is not signed, so the README tells you to allow it in Privacy & Security or run \"sudo xattr -d com.apple.quarantine /Applications/pot.app\", and to grant Accessibility permission for select-and-translate.","costNote":"The program is free. Some of the translation and text-reading services it can use are paid and billed by those companies.","notVerified":["Whether the free services (Google, Bing, Yandex) still work inside the app, since it stopped being updated.","Whether the installers still run on current Windows, macOS and Linux versions.","What each translation company keeps of the text you send it.","Whether the plugins listed on pot-app.com are still available and who built them.","Real disk use after the app stores its settings, history and any offline text-reading data."]},"zh-Hans":{"summaryShort":"在电脑上任意选中文字,按下快捷键即可看到翻译;它还能从截图中读取文字。","summary":"一个用来翻译和朗读文字的小型桌面程序。你可以在任意应用中选中文字并按下快捷键,在它的窗口中输入文字,或者截取屏幕的一部分让它读出图片中的文字并翻译。你可以选择由哪个服务来完成工作:免费的像Google和Bing,付费的像DeepL或用你自己密钥的OpenAI,Windows和Mac自带的离线文字识别,或者通过Ollama使用的本地模型。Windows、Mac和Linux都有安装程序。该仓库已被归档,不再更新。","easeRationale":"在Windows上,你下载.exe安装程序(或运行\"winget install Pylogmon.pot\")然后打开它;在Linux上有.deb、.rpm、AppImage和Flathub包,所以两者都是下载后直接打开。Mac多了一步:应用没有签名,所以README告诉你要在系统的隐私设置中允许它,或者运行\"sudo xattr -d com.apple.quarantine /Applications/pot.app\",并授予辅助功能权限以支持选中翻译。","costNote":"该程序本身是免费的。它可以使用的一些翻译和文字识别服务是付费的,由相应的公司收费。","notVerified":["免费服务(Google、Bing、Yandex)在应用中是否仍然可用,因为它已经停止更新。","安装程序在当前版本的Windows、macOS和Linux上是否仍能运行。","每家翻译公司会保留你发送给它的文字中的哪些内容。","pot-app.com上列出的插件是否仍然可用,以及是谁开发的。","应用存储设置、历史记录以及任何离线文字识别数据后的实际磁盘占用。"]}}},{"id":"R_kgDOMNgjyw","slug":"praj2408__smart-marketing-assistant-crew-ai","name":"Smart Marketing Assistant","owner":"praj2408","fullName":"praj2408/Smart-Marketing-Assistant-Crew-AI","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub没有检测到许可证,license文件、LICENSES文件夹和winget清单中也都没有","source":"check","title":"没有许可证"},{"code":"instructions_do_not_match_the_repository","severity":"attention","evidence":"readme.md让你克隆\"https://github.com/yourusername/Smart-Marketing-Assistant-using-Ai-Agents.git\",运行\"python main.py\"并打开\"http://localhost:5000\",但顶层文件只有.gitignore、docs/、pyproject.toml、readme.md、requirements.txt和src/,而且pyproject.toml是空的。","source":"ai","title":"说明与仓库内容不一致"},{"code":"promised_features_without_supporting_code","severity":"attention","evidence":"README承诺了\"Post Scheduling\"、\"Performance Analytics\"以及对评论和消息的自动回复,但requirements.txt里只列出了\"crewai\"和\"exa_py\",完全没有Instagram、排期或数据分析相关的软件包。","source":"ai","title":"承诺的功能没有对应代码"},{"code":"license_claimed_but_no_license_file","severity":"attention","evidence":"readme.md声称\"This project is licensed under the MIT License\",但没有LICENSE文件,GitHub也没有检测到任何许可证,所以你在法律上能对这段代码做什么并不清楚。","source":"ai","title":"声称有许可证但没有许可证文件"},{"code":"social_network_automation_against_platform_rules","severity":"info","evidence":"该项目描述提供自动化\"Instagram marketing workflow\",包括自动发帖和对评论、消息的自动回复;在官方工具之外这样做,可能导致账户被Instagram或X限制或封禁。","source":"ai","title":"违反平台规则的社交网络自动化"},{"code":"dependencies_without_versions","severity":"info","evidence":"requirements.txt没有固定任何版本(只有\"crewai\"和\"exa_py\"),所以每次安装都会拉取当时最新的版本,可能导致以后无法正常运行。","source":"ai","title":"依赖没有指定版本"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI","Exa","LangChain / LangSmith tracing","XQuik (X post search)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","ai_agents"],"alternativeTo":[{"slug":"jasper","name":"Jasper","confirmed":true},{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"manychat","name":"ManyChat","confirmed":false}],"declaredAlternatives":[],"stars":37,"downloads":0,"downloadsWeek":0,"downloadsDays":1,"growth":0,"growthDays":6,"history":[{"date":"2026-11-09","stars":37},{"date":"2026-11-10","stars":37}],"firstPublishedAt":"2026-11-09T23:02:44.244Z","commitSha":"b95f9903fae2e5736e8042ae9ab7981f056a30f9","verifiedAt":"2026-11-09T22:48:54.101Z","lastCommitAt":"2026-06-29T22:34:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A small set of AI agents meant to write Instagram posts, pick hashtags and research trends, run from your terminal.","summary":"A hobby project built on CrewAI: AI agents that are supposed to write Instagram posts, suggest hashtags and research what competitors are posting. You install Python, install two packages and put your own OpenAI key (plus optional Exa, LangChain and X keys) in a .env file. It is very early: the repository has only a src folder, a two-line requirements file and an empty pyproject.toml, and the README's commands mention files that are not there. 37 stars, no releases, last change in June 2026.","easeRationale":"On all three systems you work in a terminal: install Python, run pip install, create a .env file with up to five keys, and then find the right script yourself, because the README tells you to run \"python main.py\" and edit \"config.py\" while the repository has no file with either name at the top level.","costNote":"The code is free, but the agents call OpenAI (the README suggests gpt-3.5-turbo-0125) and you pay OpenAI for every run.","notVerified":["Whether the project runs at all, since the file the README asks you to run is not at the top level of the repository.","Which of the advertised features really exist inside the src folder.","How much a run costs at OpenAI, which depends on how many agent steps it takes.","Whether the X search part works, as it depends on a third-party service (XQuik) with its own account.","Whether the project touches your Instagram account directly or only writes text for you to post."]},"zh-Hans":{"summaryShort":"一组小型AI代理,旨在撰写Instagram帖子、挑选话题标签和研究趋势,通过终端运行。","summary":"一个基于CrewAI构建的业余项目:AI代理本应撰写Instagram帖子、建议话题标签并研究竞争对手发布的内容。你需要安装Python,安装两个软件包,并在.env文件中填入你自己的OpenAI密钥(还有可选的Exa、LangChain和X密钥)。这个项目还处于很早期的阶段:仓库只有一个src文件夹、一个两行的requirements文件和一个空的pyproject.toml,README中的命令提到的文件其实并不存在。37个星标,没有发布版本,最后一次改动是在2026年6月。","easeRationale":"在这三种系统上你都要在终端里操作:安装Python,运行pip install,创建一个最多包含五个密钥的.env文件,然后还要自己找到正确的脚本,因为README让你运行\"python main.py\"并编辑\"config.py\",而仓库顶层并没有这两个名字的文件。","costNote":"代码是免费的,但这些代理会调用OpenAI(README建议使用gpt-3.5-turbo-0125),每次运行你都要向OpenAI付费。","notVerified":["这个项目是否真的能运行,因为README要求运行的文件并不在仓库的顶层。","宣传的功能中有哪些真的存在于src文件夹里。","在OpenAI上运行一次要花多少钱,这取决于代理执行了多少步骤。","X搜索部分是否能用,因为它依赖一个有自己账户的第三方服务(XQuik)。","这个项目是直接操作你的Instagram账户,还是只是写好文字让你自己发布。"]}}},{"id":"R_kgDOO9AkNg","slug":"profullstack__social-poster","name":"Social Poster","owner":"profullstack","fullName":"profullstack/social-poster","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget manifest 都没有检测到许可证","source":"check","title":"没有许可证"},{"code":"works_around_platform_rules","severity":"attention","evidence":"项目把自己描述为 \"No restrictive APIs - just browser automation that works like a human\"，README 写着 \"Uses Puppeteer to login like a human - no API restrictions\"，故障排查部分还提示你 \"Verify platform isn't blocking automated browsers\"；这种自动发布方式违反 X、TikTok、Facebook 和 LinkedIn 的条款，可能导致账号被限制或关闭。","source":"ai","title":"绕开平台规则"},{"code":"stores_your_logged_in_sessions","severity":"attention","evidence":"README：登录会话保存在 ~/.config/social-poster/sessions.json 中，包含 \"Browser cookies, Local storage data, Session storage data\"；任何复制到这个文件的人都能以你的身份发帖。README 声称会话被 \"securely stored\" 和 \"encrypted\"，但同一份 README 里的示例 config.json 却以明文显示 OpenAI 密钥（\"openaiApiKey\": \"sk-...\"）。","source":"ai","title":"保存你已登录的会话"},{"code":"license_claimed_but_no_license_file","severity":"attention","evidence":"package.json 写着 \"license\": \"MIT\"，README 也指向一个 LICENSE 文件，但顶层文件中没有 LICENSE 文件，GitHub 也没有检测到许可证。","source":"ai","title":"声称有许可证却没有许可证文件"},{"code":"half_of_the_platforms_unfinished","severity":"info","evidence":"平台表格里只有 X、TikTok 和 Pinterest 被标记为 \"Ready\"；LinkedIn、Reddit 和 Facebook 是 \"In Progress\"，Stacker News、Primal 和 Hacker News 是 \"Planned\"，尽管项目描述承诺支持 \"LinkedIn, Reddit, Facebook, and more\"。","source":"ai","title":"一半平台尚未完成"},{"code":"ai_text_sent_to_a_peer_network_by_default","severity":"info","evidence":"README：\"AI generation runs on the c0mpute peer network first\"，通过 @profullstack/ai-providers 实现，OpenAI 只是备用方案，所以除非你设置 C0MPUTE_DISABLED=1，你的提示词就会发给那个网络中的其他计算机。","source":"ai","title":"AI 文案默认发给一个对等网络"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["OpenAI","c0mpute / Infernet"],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"phantombuster","name":"PhantomBuster","confirmed":true}],"declaredAlternatives":[],"stars":59,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":58},{"date":"2026-11-08","stars":58},{"date":"2026-11-09","stars":58},{"date":"2026-11-10","stars":59}],"firstPublishedAt":"2026-11-09T23:02:44.254Z","commitSha":"27affb74d84c91856265eaeef01caf5aa34fc218","verifiedAt":"2026-11-09T22:49:37.712Z","lastCommitAt":"2026-11-07T10:06:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Posts the same message to X, TikTok, Pinterest and other networks from the terminal, by driving a browser logged in as you.","summary":"A command-line tool that publishes a text, link, image or video to several social networks at once. Instead of the networks' official programming interfaces it opens a real browser with Puppeteer, you log in by hand once, and it reuses that session to click through the posting screens. It can also write the post text for you with AI. You need Node.js 20 or newer and the terminal; there is no window with buttons. X, TikTok and Pinterest are marked ready, while LinkedIn, Reddit and Facebook are still in progress.","easeRationale":"Same work on all three systems: install Node.js 20 or newer, run \"npm install -g @profullstack/social-poster\", then run \"sp login\" and log in by hand in the browser it opens (the README suggests --headless=false for the first login) before every \"sp post\" command. Everything happens in a terminal.","costNote":"The tool is free. AI text generation uses the c0mpute network (free up to 10 calls a day) and can fall back to OpenAI, which you pay for.","notVerified":["Whether posting still works: the tool clicks on page elements of each site, and the sites change their screens often (the project ships a 'selector recording' script for exactly that reason).","Whether the saved sessions are really encrypted on disk, as the README claims.","How each network reacts: your account may be asked for a verification code, limited or blocked.","Real disk use, since Puppeteer downloads its own copy of Chromium.","The quality and speed of the AI text, which the README itself says can take 60 to 240 seconds per post on c0mpute."]},"zh-Hans":{"summaryShort":"从终端把同一条消息发到 X、TikTok、Pinterest 等多个社交网络，操作的是已登录你账号的浏览器。","summary":"这是一个命令行工具，可以把文字、链接、图片或视频同时发布到多个社交网络。它不使用各网络官方的编程接口，而是用 Puppeteer 打开一个真实的浏览器，你只需手动登录一次，之后它会重复使用这个登录会话，自动点击完成发布流程。它还可以用 AI 帮你写发布文案。你需要 Node.js 20 或更新版本以及终端；没有带按钮的窗口。X、TikTok 和 Pinterest 已标记为可用，LinkedIn、Reddit 和 Facebook 仍在开发中。","easeRationale":"三个系统上的操作相同：安装 Node.js 20 或更新版本，运行 \"npm install -g @profullstack/social-poster\"，然后运行 \"sp login\"，在打开的浏览器里手动登录（README 建议首次登录时加上 --headless=false），每次运行 \"sp post\" 命令之前都要这样做。所有操作都在终端里完成。","costNote":"这个工具本身免费。AI 文案生成使用 c0mpute 网络（每天 10 次以内免费），超出后可以切换到需要付费的 OpenAI。","notVerified":["发布功能是否还能正常使用：这个工具会点击各个网站页面上的元素，而这些网站经常改版（项目专门为此提供了一个“选择器录制”脚本）。","保存的登录会话是否真的像 README 所说那样在磁盘上加密。","各个网络会如何反应：你的账号可能被要求输入验证码、受到限制，或者被封禁。","实际占用的磁盘空间，因为 Puppeteer 会下载自己的一份 Chromium。","AI 文案的质量和速度，README 自己也说在 c0mpute 上每条发布可能需要 60 到 240 秒。"]}}},{"id":"R_kgDOTFKWJA","slug":"qaxlabs__autodmx","name":"AutoDMX","owner":"Qaxlabs","fullName":"Qaxlabs/AutoDMX","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"web":"E"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：next：GHSA-2464-fhr4-w9xw、GHSA-2xp9-vwfh-vxw4、GHSA-36qx-fr4f-26g5","source":"check","title":"存在已知漏洞"},{"code":"whole_dashboard_behind_one_shared_password","severity":"attention","evidence":".env.example 里只有一个 DASHBOARD_PASSWORD，README 把它称作 \"Dashboard Password Protection - Single-password gate\"；同一套部署里还保存着 SUPABASE_SERVICE_ROLE_KEY、META_APP_SECRET 和一个长期有效的 Instagram 访问令牌，所以谁知道这一个密码，就能控制从你账号发出的消息。","source":"ai","title":"整个后台只靠一个共享密码保护"},{"code":"very_early_stage","severity":"attention","evidence":"package.json 写着 \"version\": \"0.1.0\" 和 \"private\": true，没有发行版，只有 14 个 star，仓库顶层还留着开发时期的文件（dev_server.log、supabase_out.log、mock-trigger.js、test-routes.js、test_supabase.js）。","source":"ai","title":"处于非常早期的阶段"},{"code":"instagram_automation_depends_on_meta_approval","severity":"attention","evidence":"快速开始部分要求有 \"A Meta Developer App with Instagram Graph API access\"，以及订阅了 \"comments\" 和 \"messages\" 的 webhook；Meta 只有在审核应用之后才会授予这些权限，自动发送的私信也要遵守它的规则和速率限制，所以这项自动化可能随时停止运作，甚至让你失去账号的访问权限。","source":"ai","title":"Instagram 自动化依赖 Meta 的审批"},{"code":"follow_gating_feature","severity":"info","evidence":"README：\"Follow-Gating - Optionally require users to follow your account before receiving the DM\"；用内容交换关注，正是 Meta 平台规则所限制的那类激励方式。","source":"ai","title":"带有关注才能解锁的功能"},{"code":"security_claims_not_checkable_from_outside","severity":"info","evidence":"README 声称令牌用 \"AES-256-GCM\" 加密，速率限制 \"Respects Meta's published limits out-of-the-box\"；这些都是作者关于 lib/crypto.ts 和数据库中代码的说法，不是检查项能确认的事。","source":"ai","title":"防护方面的说法无法从外部核实"}],"platforms":["web"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram Graph API","Supabase"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[{"target":"manychat","stance":"claims","evidence":"auto-reply, lead-generation, instagram-marketing, social-media-automation, meta-graph-api, manychat-alternative, self-hosted, open-source, nextjs, typescript, supabase, tailwindcss, webhooks"}],"stars":14,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":14},{"date":"2026-11-08","stars":14},{"date":"2026-11-09","stars":14},{"date":"2026-11-10","stars":14}],"firstPublishedAt":"2026-11-09T23:02:44.264Z","commitSha":"ee4189fdc714b865a3ee9c5939e638ac78d34cc7","verifiedAt":"2026-11-09T22:51:04.257Z","lastCommitAt":"2026-09-04T06:30:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Watches comments on your Instagram posts for a keyword and answers each one with an automatic private message.","summary":"A website you host yourself that turns Instagram comments into private messages: when someone comments a keyword such as \"PRICE\", it sends that person a prepared DM with a link, coupon or file, and shows clicks and conversions on a dashboard. It uses Meta's official Instagram interface, so you must create a Meta developer app, a Supabase database and a webhook before anything works, then publish the site on Netlify or another Node.js host. It is version 0.1.0 with 14 stars and no releases yet.","easeRationale":"There is no installer and no hosted version for you to sign into: you clone the repository, install Node.js 18 or newer, create a Supabase project and apply the SQL migrations, create a Meta developer app with Instagram access, fill in ten environment variables and point Meta's webhook at your own domain. That is service configuration work, which is why the only grade is E on the web.","costNote":"The code is free (MIT) and the README says the free tiers of Supabase and Netlify are enough; the Instagram side needs a Meta developer app.","notVerified":["Whether Meta approves a small developer app for the comment and message permissions, which is what makes or breaks the whole tool.","Whether the encryption and the rate limits work as the README describes.","Whether the free Supabase, Netlify and GitHub Actions tiers hold up for a busy account.","Whether the project still works with Instagram's current interface, since the last change was in September 2026.","What the demo site autodmx.netlify.app does with any data you type into it."]},"zh-Hans":{"summaryShort":"监控你 Instagram 帖子下的评论，发现某个关键词后，自动给对方发一条私信回复。","summary":"这是一个你自己搭建的网站，把 Instagram 评论转换成私信：当有人评论一个关键词，比如 \"PRICE\"，它就会给那个人发一条准备好的私信，里面带链接、优惠码或文件，并在后台面板上显示点击量和转化数据。它使用 Meta 官方的 Instagram 接口，所以在一切生效之前，你需要创建一个 Meta 开发者应用、一个 Supabase 数据库和一个 webhook，然后把网站发布到 Netlify 或其他 Node.js 托管服务上。目前是 0.1.0 版本，有 14 个 star，还没有发行版。","easeRationale":"没有安装程序，也没有可以直接登录的托管版本：你需要克隆仓库，安装 Node.js 18 或更新版本，创建一个 Supabase 项目并执行 SQL 迁移，创建一个带 Instagram 权限的 Meta 开发者应用，填好十个环境变量，并把 Meta 的 webhook 指向你自己的域名。这些都是服务配置工作，所以在网页端的评分只有 E。","costNote":"代码本身免费（MIT 许可），README 说 Supabase 和 Netlify 的免费额度就够用；Instagram 这一侧需要一个 Meta 开发者应用。","notVerified":["Meta 是否会批准一个小型开发者应用获得评论和消息权限，这一点决定了整个工具能不能用。","加密和速率限制是否真的像 README 描述的那样运作。","Supabase、Netlify 和 GitHub Actions 的免费额度能不能撑住一个活跃账号的用量。","这个项目是否还能适配 Instagram 当前的界面，因为上一次改动是在 2026 年 9 月。","演示站点 autodmx.netlify.app 会怎样处理你在里面输入的任何数据。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0MTU3OTI0","slug":"qcad__qcad","name":"QCAD","owner":"qcad","fullName":"qcad/qcad","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"multiple_license_files","severity":"info","evidence":"GitHub 将许可证报告为 NOASSERTION。顶层目录有 LICENSE.txt、gpl-3.0.txt、gpl-3.0-exceptions.txt、lgpl.txt 和 cc-by-3.0.txt，README 称脚本扩展和 C++ 插件 \"are released under their respective licenses\"。","source":"ai","title":"有多个许可证文件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["design_cad"],"alternativeTo":[{"slug":"autocad","name":"AutoCAD","confirmed":true}],"declaredAlternatives":[],"stars":1929,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":7,"growthDays":7,"history":[{"date":"2026-10-25","stars":1912},{"date":"2026-10-26","stars":1914},{"date":"2026-10-27","stars":1914},{"date":"2026-10-28","stars":1915},{"date":"2026-10-29","stars":1917},{"date":"2026-10-30","stars":1917},{"date":"2026-10-31","stars":1919},{"date":"2026-11-01","stars":1920},{"date":"2026-11-02","stars":1921},{"date":"2026-11-03","stars":1922},{"date":"2026-11-04","stars":1922},{"date":"2026-11-05","stars":1924},{"date":"2026-11-06","stars":1925},{"date":"2026-11-07","stars":1926},{"date":"2026-11-08","stars":1925},{"date":"2026-11-09","stars":1925},{"date":"2026-11-10","stars":1929}],"firstPublishedAt":"2026-10-25T23:32:51.327Z","commitSha":"e6a6bcf770f3f707b111d0cd62a76f0a2cd1a5c2","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T20:46:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Draw 2D technical plans such as floor plans, machine parts and diagrams, and save them as DXF files.","summary":"QCAD is a 2D drawing program (CAD) for technical drawings such as building plans, interiors, mechanical parts, schematics and diagrams. It works with DXF files; opening DWG files needs a separate proprietary plugin. It is for hobbyists, students and professionals who need precise 2D drawings, and it runs on Windows, macOS and Linux. This GitHub repository holds only the source code, but the official site qcad.org offers a ready Windows installer for the free QCAD Community Edition.","easeRationale":"qcad.org offers a ready Windows installer for the free QCAD Community Edition that you download and open; the compiling instructions linked in the README are only for developers.","costNote":"QCAD Community Edition and the QCAD 3 source code are free under GPLv3. DWG support comes from a proprietary plugin; the package does not state its price.","notVerified":["The Windows installer on qcad.org itself: it is not on GitHub or in winget, so Radar could not check its size, contents or disk use.","How the free Community Edition differs from any paid edition offered on qcad.org.","The price and terms of the proprietary DWG plugin.","Which license covers each add-on or plugin."]},"zh-Hans":{"summaryShort":"绘制 2D 技术图纸，如平面图、机械零件和示意图，并保存为 DXF 文件。","summary":"QCAD 是一款 2D 绘图（CAD）程序，用于建筑平面图、室内设计、机械零件、原理图和示意图等技术图纸。它使用 DXF 文件；打开 DWG 文件需要单独的专有插件。适合需要精确 2D 图纸的爱好者、学生和专业人士，可在 Windows、macOS 和 Linux 上运行。这个 GitHub 仓库只有源代码，但官方网站 qcad.org 提供免费 QCAD Community Edition 的现成 Windows 安装程序。","easeRationale":"qcad.org 提供免费 QCAD Community Edition 的现成 Windows 安装程序，下载后打开即可；README 中链接的编译说明只面向开发者。","costNote":"QCAD Community Edition 和 QCAD 3 源代码基于 GPLv3 免费提供。DWG 支持来自专有插件；资料中未说明其价格。","notVerified":["qcad.org 上的 Windows 安装程序本身：它不在 GitHub 或 winget 上，因此 Radar 无法检查其大小、内容或磁盘占用。","免费的 Community Edition 与 qcad.org 可能提供的付费版本有何区别。","专有 DWG 插件的价格和条款。","每个插件或扩展适用哪种许可证。"]}}},{"id":"R_kgDOMRSMsg","slug":"qwenaudio__cosyvoice","name":"CosyVoice","owner":"QwenAudio","fullName":"QwenAudio/CosyVoice","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：diffusers：GHSA-7wx4-6vff-v64p、GHSA-98h9-4798-4q5v、PYSEC-2026-2446；gdown：GHSA-76hw-p97h-883f、PYSEC-2026-2158；gradio：GHSA-39mp-8hj3-5c49、GHSA-5cpq-9538-jm2j、GHSA-6655-8ph2-63j3；hydra-core：GHSA-2cp2-2r3c-7p7r、GHSA-c3wx-c55w-pxjq、PYSEC-2026-3850；lightning：GHSA-cgwc-qvrx-rf7f、GHSA-mr7h-w2qc-ffc2、GHSA-qqmf-gpg7-g8gw；modelscope：GHSA-fhhq-h4hg-549x、PYSEC-2026-2663；onnx：GHSA-3r9x-f23j-gc73、GHSA-48m4-24cq-37qf、GHSA-538c-55jv-c5g9；protobuf：GHSA-7gcm-g887-7qv7、GHSA-8qvm-5x2c-j2w7、PYSEC-2026-1805；pyarrow：GHSA-rgxp-2hwp-jwgg、PYSEC-2026-113；torch：GHSA-3749-ghw9-m3mg、GHSA-53q9-r3pm-6pq6、GHSA-63cw-57p8-fm3p；transformers：GHSA-29pf-2h5f-8g72、GHSA-37mw-44qp-f5jm、GHSA-4w7r-h757-3r74","source":"check","title":"存在已知漏洞"},{"code":"voice_cloning_of_real_people","severity":"attention","evidence":"项目宣传自己支持 \"multi-lingual/cross-lingual zero-shot voice cloning\"，只需一段短样本，它自己的声明也写着 \"The content provided above is for academic purposes only\"，并请求如果有例子侵犯了某人的权利请联系说明；复刻一个人的声音在不少国家是受限制的。","source":"ai","title":"克隆真实人物的声音"},{"code":"repository_is_a_copy_of_the_original_project","severity":"info","evidence":"这个仓库是 QwenAudio/CosyVoice，但 README 里所有链接都指向 FunAudioLLM/CosyVoice，安装步骤里写的是 \"git clone --recursive https://github.com/FunAudioLLM/CosyVoice.git\"，所以这些说明实际构建的是另一个仓库，不是这个。","source":"ai","title":"这个仓库是原项目的复制品"},{"code":"install_from_a_mirror_with_certificate_check_off","severity":"info","evidence":"README 的安装命令是：\"pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple/ --trusted-host=mirrors.aliyun.com\"——这些包来自阿里云的镜像，--trusted-host 参数关闭了对该站点的证书校验。","source":"ai","title":"从关闭证书校验的镜像安装"},{"code":"large_model_downloads_from_outside_github","severity":"info","evidence":"README 要求从 ModelScope 或 Hugging Face 获取 Fun-CosyVoice3-0.5B、CosyVoice2-0.5B、CosyVoice-300M、-SFT、-Instruct 和 ttsfrd 资源，可选的 ttsfrd 步骤还会安装两个随该下载内容附带的 .whl 文件。","source":"ai","title":"从 GitHub 之外下载大体量模型"},{"code":"no_releases_and_many_open_issues","severity":"info","evidence":"这个仓库在 GitHub 上没有发行版，有 640 个未解决的 issue，最后一次提交是在 2026-05-25（检查时的 168 天前），所以你安装到的始终是当时 main 分支上的内容。","source":"ai","title":"没有发行版，且有很多未解决的 issue"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_nvidia","other"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":1250,"maxMB":5250,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":24356,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":64,"growthDays":6,"history":[{"date":"2026-11-06","stars":24318},{"date":"2026-11-07","stars":24327},{"date":"2026-11-08","stars":24332},{"date":"2026-11-09","stars":24346},{"date":"2026-11-10","stars":24356}],"firstPublishedAt":"2026-11-09T23:02:44.274Z","commitSha":"074ca6dc9e80a2f424f1f74b48bdd7d3fea531cc","verifiedAt":"2026-11-09T22:51:50.016Z","lastCommitAt":"2026-05-25T18:15:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Reads text aloud in nine languages and can copy a voice from a short recording, on your own computer.","summary":"A voice generation model from the FunAudioLLM team: it turns text into speech in nine languages and more than eighteen Chinese dialects, copies a voice from a short sample without training, and can follow instructions about emotion, speed or accent. You install Conda and Python, download model files of a few gigabytes from ModelScope or Hugging Face, then run a sample script or a local web page. An NVIDIA card is expected for decent speed, and the faster server setups (vLLM, TensorRT, Docker) are Linux only.","easeRationale":"On every system it is terminal work: clone with submodules, install Conda, create an environment, pip install a long list of packages, download several model folders by script and only then run \"python example.py\" or \"python3 webui.py\". Linux is the best supported one: the deepspeed, onnxruntime-gpu, TensorRT and Docker pieces in requirements.txt and in the README only apply there, and sox may need a system package.","costNote":"Code under Apache-2.0 and model files free to download; the electricity and the graphics card are yours.","notVerified":["Whether it runs on Windows or Mac in practice, and how slow it is there without an NVIDIA card.","Real disk use: the checks estimate 1.2-5.2 GB, and that depends on how many of the model folders you download.","The quality and speed claims (150 ms latency, the scores in the evaluation table) were measured by the authors.","Who can reach the web page started by webui.py, which has no login of its own.","How good the voices are in languages other than Chinese and English."]},"zh-Hans":{"summaryShort":"能用九种语言朗读文字，还能从一段短录音复刻一个声音，全部在你自己的电脑上运行。","summary":"这是 FunAudioLLM 团队发布的语音生成模型：可以把文字转换成九种语言和十八种以上中文方言的语音，无需训练就能从一段短样本复刻一个声音，还能按指令调整情绪、语速或口音。你需要安装 Conda 和 Python，从 ModelScope 或 Hugging Face 下载几 GB 的模型文件，再运行示例脚本或本地网页。要获得不错的速度需要一块 NVIDIA 显卡，更快的服务器方案（vLLM、TensorRT、Docker）只支持 Linux。","easeRationale":"在每个系统上都是终端操作：带子模块克隆仓库，安装 Conda，创建环境，用 pip 安装一长串依赖包，用脚本下载好几个模型文件夹，最后才运行 \"python example.py\" 或 \"python3 webui.py\"。Linux 的支持最好：requirements.txt 和 README 里的 deepspeed、onnxruntime-gpu、TensorRT 和 Docker 相关部分只适用于 Linux，sox 还可能需要安装一个系统软件包。","costNote":"代码采用 Apache-2.0 协议，模型文件可以免费下载；电费和显卡费用自己承担。","notVerified":["在 Windows 或 Mac 上是否真的能跑，以及没有 NVIDIA 显卡时到底有多慢。","实际占用的磁盘空间：检查结果估算为 1.2 到 5.2 GB，具体取决于你下载了多少个模型文件夹。","质量和速度方面的说法（150 毫秒延迟、评测表里的分数）都是作者自己测出来的。","谁能访问 webui.py 打开的那个网页，它本身没有登录功能。","除中文和英文之外，其他语言的语音效果到底怎么样。"]}}},{"id":"R_kgDOQ-D4bA","slug":"qwenlm__qwen3-tts","name":"Qwen3-TTS","owner":"QwenLM","fullName":"QwenLM/Qwen3-TTS","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"voice_cloning_of_real_people","severity":"attention","evidence":"Base 模型的介绍是能做到 \"3-second rapid voice clone from user audio input\"，而 ref_audio \"can be a local file path, a URL, a base64 string\"，所以从网上拿到某人三秒钟的声音就足够了；复刻一个人的声音在不少国家是受限制的。","source":"ai","title":"克隆真实人物的声音"},{"code":"demo_server_opens_to_the_network_without_a_password","severity":"attention","evidence":"README 自己给出的命令是 \"qwen-tts-demo Qwen/Qwen3-TTS-12Hz-1.7B-Base --ip 0.0.0.0 --port 8000\"，以及用于 HTTPS 的自签名证书加 \"--no-ssl-verify\"；0.0.0.0 意味着任何能在那个端口连到你机器的人都能使用这个没有登录功能的网页。","source":"ai","title":"演示服务器不设密码就对外开放"},{"code":"voice_design_example_describes_a_childlike_voice","severity":"info","evidence":"README 里一个 Voice Design 示例要求 \"撒娇稚嫩的萝莉女声\"，另一个要求 \"Male, 17 years old\" 的声音，说明这个模型可以被要求生成听起来像未成年人的声音。","source":"ai","title":"声音设计示例描述了一个像孩子的声音"},{"code":"model_files_download_on_first_use","severity":"info","evidence":"README：\"During model loading in the qwen-tts package or vLLM, model weights will be automatically downloaded based on the model name\"，来源是 Hugging Face 或 ModelScope；检查里给出的 150-450 MB 估算只是代码体积，不包含这些文件。","source":"ai","title":"模型文件在首次使用时下载"},{"code":"no_releases_and_quiet_since_march","severity":"info","evidence":"GitHub 上没有发行版，pyproject.toml 里的包版本是 0.1.1，最后一次提交是在 2026-03-17（检查时的 237 天前），而 README 却说报告里提到的其他模型 \"will be released in the near future\"。","source":"ai","title":"没有发行版，自 3 月起没有动静"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia"],"buildTools":["python"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":14343,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":128,"growthDays":6,"history":[{"date":"2026-11-05","stars":14230},{"date":"2026-11-06","stars":14248},{"date":"2026-11-07","stars":14271},{"date":"2026-11-08","stars":14291},{"date":"2026-11-09","stars":14314},{"date":"2026-11-10","stars":14343}],"firstPublishedAt":"2026-11-09T23:02:44.284Z","commitSha":"022e286b98fbec7e1e916cb940cdf532cd9f488e","verifiedAt":"2026-11-09T22:52:46.031Z","lastCommitAt":"2026-03-17T06:38:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns text into speech in ten languages, invents a voice from a written description, or copies one from 3 seconds of audio.","summary":"Alibaba's Qwen team released these speech models with a Python package. You can read text aloud in ten languages with nine ready voices, describe the voice you want in plain words and let it invent one, or give it a three-second recording and have it speak in that voice. You install Python and the qwen-tts package, and the model files (0.6B or 1.7B) download on first use. The examples all load the model on an NVIDIA card. One command opens a local web page for trying it without writing code; Alibaba also sells the same thing as an online service.","easeRationale":"On Linux and Windows it is terminal work plus specific hardware: create a Python environment, \"pip install -U qwen-tts\", let a model of a few gigabytes download, then run \"qwen-tts-demo <model>\" for the local web page. Every example in the README loads the model with device_map=\"cuda:0\" and bfloat16 and recommends building flash-attn, so an NVIDIA card is expected; Linux is the easier of the two because flash-attn and vLLM are built for it. Mac is not mentioned at all.","costNote":"Code and model weights are free under Apache-2.0; Alibaba Cloud's DashScope offers the same models as a paid online service.","notVerified":["Whether it runs at all without an NVIDIA card, since the README shows no CPU or Apple Silicon path.","How much disk and video memory the 0.6B and 1.7B models really take after downloading.","The headline numbers (97 ms first audio, the error rates in the tables) were measured by the authors.","How good the Portuguese and Spanish voices are compared with Chinese and English, which get most of the attention.","Whether building flash-attn works on a normal home machine; the README warns about memory and CPU cores during that step."]},"zh-Hans":{"summaryShort":"能用十种语言朗读文字，可以根据文字描述凭空创造一个声音，也能从 3 秒音频复刻一个声音。","summary":"阿里巴巴的 Qwen 团队发布了这些语音模型，并配套提供一个 Python 包。你可以用九种现成的声音，以十种语言朗读文字；也可以用一句话描述你想要的声音，让模型凭空创造一个；或者给它一段三秒钟的录音，让它用那个声音说话。你需要安装 Python 和 qwen-tts 包，模型文件（0.6B 或 1.7B）会在第一次使用时自动下载。README 里的所有示例都是把模型加载到 NVIDIA 显卡上运行。一条命令就能打开一个本地网页，不写代码也能试用；阿里巴巴还把同样的功能作为付费在线服务出售。","easeRationale":"在 Linux 和 Windows 上都是终端操作，外加特定的硬件要求：创建一个 Python 环境，运行 \"pip install -U qwen-tts\"，等几个 GB 的模型下载完成，再运行 \"qwen-tts-demo <model>\" 打开本地网页。README 里的每个示例都用 device_map=\"cuda:0\" 和 bfloat16 加载模型，并建议编译 flash-attn，所以需要一块 NVIDIA 显卡；Linux 比 Windows 更省心，因为 flash-attn 和 vLLM 都是为它构建的。README 完全没有提到 Mac。","costNote":"代码和模型权重在 Apache-2.0 协议下免费；阿里云的 DashScope 把同样的模型作为付费在线服务出售。","notVerified":["没有 NVIDIA 显卡的情况下到底能不能跑，因为 README 完全没有给出 CPU 或 Apple Silicon 的路径。","0.6B 和 1.7B 模型下载之后实际占用多少磁盘和显存。","几个主要指标（首次出声 97 毫秒、表格里的错误率）都是作者自己测出来的。","葡萄牙语和西班牙语的语音效果和得到更多关注的中文、英文相比怎么样。","在普通家用电脑上编译 flash-attn 是否真的可行；README 在这一步提醒要注意内存和 CPU 核心数。"]}}},{"id":"R_kgDOVWezCw","slug":"qybaihe__bla","name":"BLA","owner":"qybaihe","fullName":"qybaihe/bla","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"rewrites_your_build_output","severity":"attention","evidence":"README 的 Quick Start：`bla harden .` 会 \"auto-discovers build outputs (dist/, build/, .asar, .wasm, .sqlite, .safetensors, native binaries, Python/C/Rust/Go)\" 并就地改写；MCP 流程还会改动源码（`bla_encrypt_strings`、`bla_bind_scope`）。体积会被刻意大幅撑大，例如 Electron 的 .asar 从 607 B 变成 16.9 MiB，Go 二进制的一个段从 50 KiB 变成 1.16 MiB。","source":"ai","title":"会改写你的构建产物"},{"code":"equivalence_checked_only_on_bundled_samples","severity":"attention","evidence":"README 顶部表格第三列：\"14/14 bundled benchmark targets pass input/output equivalence tests (`bla verify` checks Counter-REA structural gates; run your project's test suite for custom artifacts)\"。也就是说，没有任何证据表明你自己的程序在加固后行为不变。","source":"ai","title":"等价性只在自带样例上验证过"},{"code":"can_plant_instructions_aimed_at_ai_readers","severity":"attention","evidence":"README 的 Security & Configuration Defaults：\"Opt-In --llm-directives — Aggressive in-band prompt decoys are off by default; pass --llm-directives (or llmDirectives: true) to enable.\" 防御表还提到会植入 \"plausible shadow state machines\" 和 \"synthetic rabbit holes\"，以误导阅读该文件的语言模型。","source":"ai","title":"可以埋入针对 AI 读者的指令"},{"code":"anti_debugging_and_fake_telemetry_at_runtime","severity":"attention","evidence":"README 的 \"Runtime Truman Show\"：加固后的程序会 \"detects CDP / Node Inspector / Frida / WebDriver attachment (runtime-shield.ts)\"，并 \"serves believable synthetic telemetry during live debugging\"。这类行为也正是杀毒扫描工具在可疑软件中寻找的特征。","source":"ai","title":"运行时反调试并伪造遥测数据"},{"code":"dual_use_anti_analysis","severity":"info","evidence":"README 称该工具 \"Defensive Use Only\"，仅用于 \"your own\" 的软件（DISCLAIMER.md），但它所做的事——让反编译器和分析智能体看不清二进制里有什么——对任何不想让别人检查自己程序的人同样有效。","source":"ai","title":"可两用的反分析手段"},{"code":"numbers_are_the_authors_own_estimates","severity":"info","evidence":"README 的 Benchmark 一节：\"Token and USD columns below are internal estimates derived from BLA's catalog of REA limits\"，而总计的 \"~639.5x token amplification\" 也 \"limited to the benchmark fixtures\"。另一张表把包括 Ghidra、Frida、JADX 和固件在内的防护列为 \"not empirically run on this machine\"。","source":"ai","title":"这些数字是作者自己的估算"},{"code":"very_new_project_fast_releases","severity":"info","evidence":"仓库创建于 2026-10-25，到 2026-11-10 已有 4991 个星标、599 个分支和 36 次发布，16 天内版本号升到 3.2.4。","source":"ai","title":"项目非常新且发布极为频繁"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_skills"],"alternativeTo":[{"slug":"jscrambler","name":"Jscrambler","confirmed":true}],"declaredAlternatives":[],"stars":4991,"downloads":0,"downloadsWeek":0,"downloadsDays":0,"growth":2903,"growthDays":7,"history":[],"firstPublishedAt":"2026-11-10T23:53:36.462Z","commitSha":"3e53a1e5b63f1267cc94560a3411d0bb7775313d","verifiedAt":"2026-11-10T23:36:13.895Z","lastCommitAt":"2026-11-10T18:25:01Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Command-line tool that rewrites your built files so AI code-reading agents run out of budget trying to reverse engineer them.","summary":"BLA is a Node command-line tool and MCP server for software authors who do not want AI agents reading their released builds. Pointed at a folder, it rewrites build outputs of 19 kinds, including JavaScript bundles, Electron .asar files, WebAssembly, SQLite databases, model files, Windows .exe and native binaries, padding them with decoys and structures that hit the size and complexity limits of automatic analysis tools. It also plugs into Claude Code, Cursor, Codex and other coding assistants, and into Vite, Rollup, Webpack and esbuild builds.","easeRationale":"Same on all three systems, with Linux as good a case as any: install Node 22.23.4 or newer, then run `npx bla-agents harden .` in a terminal inside your project, or `npx bla-agents setup` to register it with your coding assistant. There is no window and no installer, and it only makes sense if you already build software.","costNote":"The code is MIT and the npm package is free. The project's website links to an Enterprise page whose terms are not in the repository.","notVerified":["whether a program still works correctly after hardening, on anything other than the 14 samples that ship with it","whether the protection holds against analysis tools and agents other than the one it was tested with, or against a newer version of them","how much bigger and slower your release gets, since the published growth figures come from tiny test files","what the MCP tools change in your own source files when a coding assistant runs the weave workflow","whether the bla-agents package on npm is built from the code in this repository"]},"zh-Hans":{"summaryShort":"命令行工具，会改写你的构建产物，让读代码的 AI 智能体在逆向分析时耗尽预算。","summary":"BLA 是一个 Node 命令行工具兼 MCP 服务器，面向不希望 AI 智能体读取自己已发布构建的软件作者。把它指向一个目录，它就会改写 19 类构建产物，包括 JavaScript 打包文件、Electron 的 .asar、WebAssembly、SQLite 数据库、模型文件、Windows .exe 和原生二进制，向其中填入诱饵和结构，使自动分析工具撞上体积与复杂度上限。它还能接入 Claude Code、Cursor、Codex 等编程助手，以及 Vite、Rollup、Webpack 和 esbuild 的构建流程。","easeRationale":"三个系统上的做法相同，以 Linux 为例：安装 Node 22.23.4 或更高版本，然后在项目目录的终端里运行 `npx bla-agents harden .`，或用 `npx bla-agents setup` 把它注册到编程助手里。没有图形界面，也没有安装程序，只有本来就在开发软件的人才用得上。","costNote":"代码采用 MIT 许可，npm 包免费。项目网站链接到一个 Enterprise 页面，其条款不在仓库中。","notVerified":["除了随附的 14 个样例之外，程序在加固之后是否仍能正确运行","面对与测试对象不同的分析工具和智能体，或它们更新的版本时，这层保护是否还有效","你的发布版本会变大、变慢多少，因为公布的膨胀数字来自极小的测试文件","当编程助手运行 weave 流程时，MCP 工具会在你自己的源码文件里改动什么","npm 上的 bla-agents 包是否由本仓库的代码构建而来"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzMjY2NTcxOA==","slug":"railsware__upterm","name":"Upterm","owner":"railsware","fullName":"railsware/upterm","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发行是在 2018-05-29（3086 天前）","source":"check","title":"发行版很久没更新"},{"code":"archived","severity":"info","evidence":"该仓库已归档（只读，不再维护）","source":"check","title":"已归档"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞（OSV）：csv-parse：GHSA-582f-p4pg-xc74、GHSA-8cw4-87c7-c6xx；immutable：GHSA-v56q-mh7h-f735、GHSA-wf6x-7x77-mvgw、GHSA-xvcm-6775-5m9r；lodash：GHSA-29mw-wpgm-hmr9、GHSA-35jh-r3h4-6jhm、GHSA-jf85-cpcp-j695；react-dom：GHSA-mvjj-gqq2-p4hw","source":"check","title":"存在已知漏洞"},{"code":"declared_dead_by_the_author","severity":"attention","evidence":"README 开头就是 \"# Deprecated - Upterm is deprecated. Project had lost maintainer... I am no longer accepting pull requests and issues. I recomended to check Hyper instead.\"，仓库已归档，最后一次提交是在 2019 年 5 月。","source":"ai","title":"作者已宣布项目死亡"},{"code":"built_on_a_very_old_browser_engine","severity":"attention","evidence":"package.json 固定了 \"electron\": \"2.0.10\" 和 Node 8，都是 2018 年的版本，早已不再收到漏洞修复，检查结果还列出了固定版本的 lodash、immutable、csv-parse 和 react-dom 中的已知漏洞。","source":"ai","title":"构建在非常老旧的浏览器引擎上"},{"code":"linux_download_older_than_the_mac_one","severity":"info","evidence":"最新发行版（v0.4.4，2018-05-29）只提供 upterm-0.4.4-macOS.dmg；最后一个 Linux AppImage 来自 2017 年 9 月的 v0.4.3。","source":"ai","title":"Linux 下载版本比 Mac 的还旧"},{"code":"it_runs_your_shell_commands","severity":"info","evidence":"这个项目既是终端模拟器，也是交互式 shell（依赖 node-pty），所以你在里面输入的任何内容都会以你自己用户的权限运行，就像任何终端一样。","source":"ai","title":"会执行你的 shell 命令"}],"platforms":["mac","linux"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":19045,"downloads":158919,"downloadsWeek":24,"downloadsDays":4,"growth":3,"growthDays":6,"history":[{"date":"2026-11-06","stars":19043},{"date":"2026-11-07","stars":19043},{"date":"2026-11-08","stars":19042},{"date":"2026-11-09","stars":19043},{"date":"2026-11-10","stars":19045}],"firstPublishedAt":"2026-11-09T23:02:44.294Z","commitSha":"6db8d5b89625fca9a3058e1e5956084867225e9c","verifiedAt":"2026-11-09T22:53:22.348Z","lastCommitAt":"2019-05-20T17:42:14Z","isMirror":false,"collectFailing":false,"installers":[{"os":"mac","file":"upterm-0.4.4-macOS.dmg","sizeBytes":60985039,"url":"https://github.com/railsware/upterm/releases/download/v0.4.4/upterm-0.4.4-macOS.dmg","releaseUrl":"https://github.com/railsware/upterm/releases/tag/v0.4.4"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"An alternative terminal window that suggests commands as you type; the author stopped the project and recommends Hyper.","summary":"Upterm (once called Black Screen) is a terminal window built with web technology that tries to behave like a code editor: it shows an autocompletion box as you type, with command descriptions and the value of aliases, and it displays some output, such as JSON, in a friendlier way. There is a Mac .dmg and a Linux AppImage you just open; Windows was never supported. The author declared the project deprecated, stopped taking issues and pull requests, and points people to Hyper instead; the last release is from 2018.","easeRationale":"Both platforms are download-and-open: on Mac you install the upterm-0.4.4-macOS.dmg from the releases page (or \"brew cask install upterm\"), and on Linux you download the AppImage and open it. Windows has no grade because the README says it \"is not officially supported\".","notVerified":["Whether a 2018 build still opens on today's macOS and Linux versions, and whether macOS blocks it as an unidentified developer.","Whether the autocompletion still matches current versions of the commands it suggests.","How it behaves with full-screen programs such as vim, emacs or htop, which the README itself asks people to report glitches about.","Real disk use after installation (the releases are 58-61 MB downloads).","Whether any of the 220 open issues breaks everyday use, since nobody is answering them."]},"zh-Hans":{"summaryShort":"一个会在你输入时给出命令建议的替代终端窗口；作者已停止该项目，并推荐使用 Hyper。","summary":"Upterm（以前叫 Black Screen）是一个用网页技术构建的终端窗口，试图表现得像代码编辑器：你输入时会出现一个自动补全框，带有命令说明和别名的值，还会把一些输出，比如 JSON，显示得更友好。有一个 Mac 的 .dmg 文件和一个 Linux 的 AppImage，打开就能用；Windows 从来没有被支持过。作者宣布这个项目已经废弃，不再处理 issue 和 pull request，并建议大家改用 Hyper；最后一个发行版是 2018 年的。","easeRationale":"两个平台都是下载后直接打开：在 Mac 上，你从发行版页面安装 upterm-0.4.4-macOS.dmg（或者用 \"brew cask install upterm\"），在 Linux 上，你下载 AppImage 并打开它。Windows 没有评分，因为 README 说它 \"is not officially supported\"。","notVerified":["2018 年的构建是否还能在今天的 macOS 和 Linux 版本上打开，macOS 是否会把它当作未知开发者而拦截。","自动补全是否还跟得上它所建议命令的当前版本。","它在 vim、emacs 或 htop 这类全屏程序下表现如何，README 自己也请大家反馈这方面的问题。","安装之后实际占用的磁盘空间（发行版下载大小是 58 到 61 MB）。","220 个未解决的 issue 里有没有影响日常使用的问题，因为没有人在回复它们。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0MDUzOTAwNQ==","slug":"rawtherapee__rawtherapee","name":"RawTherapee","owner":"RawTherapee","fullName":"RawTherapee/RawTherapee","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":45,"maxMB":135,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"lightroom","name":"Lightroom","confirmed":true},{"slug":"adobe-camera-raw","name":"Camera Raw","confirmed":true},{"slug":"capture-one","name":"Capture One","confirmed":true}],"declaredAlternatives":[],"stars":4286,"downloads":226855,"downloadsWeek":12153,"downloadsDays":7,"growth":14,"growthDays":7,"history":[{"date":"2026-10-25","stars":4251},{"date":"2026-10-26","stars":4253},{"date":"2026-10-27","stars":4255},{"date":"2026-10-28","stars":4256},{"date":"2026-10-29","stars":4257},{"date":"2026-10-30","stars":4259},{"date":"2026-10-31","stars":4260},{"date":"2026-11-01","stars":4262},{"date":"2026-11-02","stars":4265},{"date":"2026-11-03","stars":4272},{"date":"2026-11-04","stars":4275},{"date":"2026-11-05","stars":4276},{"date":"2026-11-06","stars":4280},{"date":"2026-11-07","stars":4280},{"date":"2026-11-08","stars":4282},{"date":"2026-11-09","stars":4284},{"date":"2026-11-10","stars":4286}],"firstPublishedAt":"2026-10-25T23:19:26.533Z","commitSha":"b2ee504c47e5e6d5e054395aa12bf1fe43bd1c3c","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-09T01:10:22Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"RawTherapee_5.13_win64_x86_64_release.exe","sizeBytes":47304123,"url":"https://github.com/RawTherapee/RawTherapee/releases/download/5.13/RawTherapee_5.13_win64_x86_64_release.exe","releaseUrl":"https://github.com/RawTherapee/RawTherapee/releases/tag/5.13"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Turn RAW photos from your digital camera into finished pictures, with fine control over color, light and detail.","summary":"RawTherapee is a desktop program for developing RAW photo files from digital cameras, and it also edits JPEG, TIFF and PNG images. It offers detailed controls for exposure, color, sharpness, noise and lens fixes, and keeps your original files unchanged. It is for photography enthusiasts and semi-professionals who are willing to learn; the project says it does not try to be an all-in-one tool and lacks photo-library management and printing. Windows (including Windows on Arm), macOS and Linux have ready downloads; the Windows installer is about 45 MB.","easeRationale":"The 5.13 release includes a ready Windows installer (RawTherapee_5.13_win64_x86_64_release.exe, about 45 MB) and a Windows on Arm version, so you download it and run it.","notVerified":["Whether RAW files from your specific camera are supported well.","How fast editing and export are on an average computer with large RAW files.","How steep the learning curve is for someone new to RAW editing."]},"zh-Hans":{"summaryShort":"把数码相机的 RAW 照片处理成成品图片，可精细控制色彩、光线和细节。","summary":"RawTherapee 是一款用于处理数码相机 RAW 照片文件的桌面程序，也能编辑 JPEG、TIFF 和 PNG 图像。它提供曝光、色彩、锐度、降噪和镜头校正等精细控制，并保持原始文件不变。它面向愿意学习的摄影爱好者和半专业人士；项目自称不想做成一体化工具，缺少照片库管理和打印功能。Windows（包括 Windows on Arm）、macOS 和 Linux 都有现成的下载；Windows 安装程序约 45 MB。","easeRationale":"5.13 版本包含现成的 Windows 安装程序（RawTherapee_5.13_win64_x86_64_release.exe，约 45 MB）和 Windows on Arm 版本，下载后运行即可。","notVerified":["你所用相机的 RAW 文件是否得到良好支持。","在普通电脑上处理大尺寸 RAW 文件时，编辑和导出的速度如何。","对刚接触 RAW 编辑的人来说，学习曲线有多陡。"]}}},{"id":"R_kgDOVVhSrQ","slug":"rddmwmddr__android-ad-cleaner","name":"Android Ad Cleaner","owner":"rddmwmddr","fullName":"rddmwmddr/android-ad-cleaner","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"6 天内获得 5069 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"very_new_repository_growing_fast","severity":"attention","evidence":"创建于 2026-10-24，最近一次推送是 2026-10-30，在这段时间里获得了 5069 个星标、462 个 fork、130 个 open issue 和 11 个发行版。版本号仍停留在 0.2.x。","source":"ai","title":"仓库非常新，增长很快"},{"code":"requires_usb_debugging","severity":"attention","evidence":"使用时需要在手机上打开开发者模式和 USB 调试（docs/00-打开开发者模式.md），这会在开启期间降低手机自身的防护。工具在最后一步提供重新关闭它的选项：\"python3 tool/run.py -s <序列号> dev-off --yes # 从电脑关闭开发者模式\"。","source":"ai","title":"需要 usb 调试"},{"code":"suspends_vendor_ad_packages","severity":"attention","evidence":"README 表格：\"广告组件 | com.opos.ads、com.miui.systemAdSolution 等 | pm suspend 挂起，要机主同意\"——它会挂起厂商的广告组件（需要机主同意，并可通过一条命令恢复），所以厂商软件的一部分会停止按出厂状态运行。","source":"ai","title":"挂起厂商的广告组件"},{"code":"device_support_varies_and_partly_unverified","severity":"info","evidence":"已测试机型表格中有多项标记为尚未验证，例如 \"OPPO | Find X9 Ultra（PMA120） | ColorOS 16.0.10 | ❌ 页面和 N6 不同，待补\" 和 \"魅族 21 Pro ... ⚠️ 从开启状态自动关闭待验证\"。README 还提到设置页面在不同系统版本间会变化，入口缺失的步骤会被跳过。","source":"ai","title":"各机型支持情况不同，部分尚未验证"},{"code":"sends_diagnostics_with_feedback","severity":"info","evidence":"README：客户端的反馈按钮 \"会自动附上客户端版本、手机型号、系统版本和最近的日志（序列号已替换，不含应用列表和通知内容）\"，0.2.9 发行说明还链接到 docs/12-错误日志上报.md 介绍错误日志上报。","source":"ai","title":"随反馈一起发送诊断信息"},{"code":"no_automatic_updates","severity":"info","evidence":"v0.2.9 的 README：\"旧版不会自动更新，0.2.7 起的客户端左下角会提示新版；请重新下载\"——旧版本不会自己更新，需要手动下载新版本。","source":"ai","title":"不会自动更新"}],"platforms":["windows","mac"],"tools":[],"buildTools":["python"],"disk":{"minMB":29,"maxMB":87,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","network_system","automation"],"alternativeTo":[],"declaredAlternatives":[],"stars":7380,"downloads":112189,"downloadsWeek":23389,"downloadsDays":7,"growth":1309,"growthDays":7,"history":[{"date":"2026-10-27","stars":3313},{"date":"2026-10-28","stars":4299},{"date":"2026-10-29","stars":4733},{"date":"2026-10-30","stars":5064},{"date":"2026-10-31","stars":5365},{"date":"2026-11-01","stars":5665},{"date":"2026-11-02","stars":5900},{"date":"2026-11-03","stars":6071},{"date":"2026-11-04","stars":6303},{"date":"2026-11-05","stars":6541},{"date":"2026-11-06","stars":6736},{"date":"2026-11-07","stars":6922},{"date":"2026-11-08","stars":7109},{"date":"2026-11-09","stars":7244},{"date":"2026-11-10","stars":7380}],"firstPublishedAt":"2026-10-27T18:40:28.027Z","commitSha":"564f574029431f09e67e1b4d0c572147a3909a83","verifiedAt":"2026-10-31T02:00:30.454Z","lastCommitAt":"2026-11-10T22:54:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop app that connects to a Chinese-brand Android phone by cable and switches off system ads and marketing pushes.","summary":"Android Ad Cleaner is a Mac and Windows app for Chinese-brand Android phones (HyperOS, ColorOS, OriginOS, MagicOS and others). You plug the phone in by cable, it detects the model, then flips the phone's own settings to turn off personalised ads, ad tracking, trending searches, news feeds, store promotions and marketing notifications, while keeping order, delivery and weather alerts. No root and no reflashing, and changes can be undone. You first switch on developer mode and USB debugging on the phone. There is also a Python command line. The app and the documentation are in Chinese.","easeRationale":"Best on Mac: AndroidAdCleaner-0.2.9-mac-arm64.dmg (with AndroidAdCleaner-0.2.9-mac-intel.dmg for Intel Macs) is described in the README as signed and notarised by Apple, so it opens with a double click, and you need neither Python nor an AI tool — only a cable and one extra step on the phone, turning on developer mode and USB debugging. Windows works the same way with AndroidAdCleaner-0.2.9-windows-setup.exe or the portable zip, but the README warns that Windows shows \"Windows protected your PC\", so you must click \"More info\" and \"Run anyway\".","costNote":"Free and open source under the MIT licence, with ready installers in the releases.","notVerified":["Whether it works on your exact phone model and system version: the README's own table shows some scripts still waiting for verification.","Whether the automatic screen reading taps only the intended switches on a phone that is not in the table.","Whether the undo steps put everything back, and whether a system update brings the ads back.","What the error-log reporting actually sends to the developer.","The real disk space used: the checks estimate 29 to 87 MB for the installers.","Whether suspending the vendor ad packages affects other features of the phone over time."]},"zh-Hans":{"summaryShort":"桌面应用，通过数据线连接国产品牌 Android 手机，关闭系统广告和营销推送。","summary":"Android Ad Cleaner 是一款面向国产品牌 Android 手机（HyperOS、ColorOS、OriginOS、MagicOS 等）的 Mac 和 Windows 应用。用数据线连接手机后，它会识别机型，然后修改手机自身的设置，关闭个性化广告、广告追踪、热搜、资讯推送、商店促销和营销通知，同时保留订单、物流和天气提醒。不需要 root，也不需要刷机，改动都可以撤销。使用前要先在手机上打开开发者模式和 USB 调试。还提供一个 Python 命令行工具。应用和文档都是中文的。","easeRationale":"在 Mac 上体验最好：README 说 AndroidAdCleaner-0.2.9-mac-arm64.dmg（Intel Mac 用 AndroidAdCleaner-0.2.9-mac-intel.dmg）经过 Apple 签名和公证，双击就能打开，既不需要 Python 也不需要 AI 工具——只要一根数据线，外加在手机上多做一步：打开开发者模式和 USB 调试。Windows 用 AndroidAdCleaner-0.2.9-windows-setup.exe 或便携版 zip 同样方便，但 README 提醒 Windows 会显示“Windows 已保护你的电脑”，需要点击“更多信息”再点击“仍要运行”。","costNote":"免费开源，MIT 许可，发行版里有现成的安装包。","notVerified":["是否适用于你具体的手机型号和系统版本：README 自带的表格显示有些脚本还在等待验证。","在表格之外的手机上，自动读屏操作是否只点到预期的开关。","撤销步骤是否能把一切还原，系统更新后广告是否会卷土重来。","错误日志上报到底把什么发给了开发者。","实际占用的磁盘空间：检测估算安装包为 29 到 87 MB。","挂起厂商的广告组件是否会长期影响手机的其他功能。"]}}},{"id":"R_kgDOM_ZVxw","slug":"readest__readest","name":"Readest","owner":"readest","fullName":"readest/readest","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A","web":"A"},"risk":"low","signals":[{"code":"paid_plans_around_the_free_app","severity":"info","evidence":"Dockerfile 说明,在自托管镜像中 \"premium features are unlocked by default\",并且 \"An operator running a hosted service that sells plans sets SELF_HOSTED=false\";构建还接受 NEXT_PUBLIC_STORAGE_FIXED_QUOTA 和 NEXT_PUBLIC_TRANSLATION_FIXED_QUOTA 参数,所以官方服务对存储和翻译有限制。","source":"ai","title":"免费应用周边有付费计划"},{"code":"sync_uploads_your_books_to_a_service","severity":"info","evidence":"README:\"Sync across Platforms - Synchronize book files, reading progress, notes, and bookmarks across all supported platforms\",而 Dockerfile 的构建参数(NEXT_PUBLIC_SUPABASE_URL、NEXT_PUBLIC_OBJECT_STORAGE_TYPE)表明文件会进入运营方管理的 Supabase 账户和对象存储。","source":"ai","title":"同步会把你的书上传到服务端"},{"code":"web_clipping_reuses_your_browser_session","severity":"info","evidence":"README:\"Open websites, sign in, and clip pages with From Web Browser. From Web Novel can reuse your browser session to import selected chapters and their images\",所以该功能在你已经登录的网站上才能生效。","source":"ai","title":"网页剪辑会复用你的浏览器会话"},{"code":"protected_store_books_are_not_covered","severity":"info","evidence":"功能列表里没有提到打开受保护的文件;唯一和 DRM 相关的说法是 \"pair a reflowable EPUB locally with DRM-free MP3, M4A, or M4B narration\",所以被 Kindle 或其他书店锁定的图书大概无法打开。","source":"ai","title":"不支持书店的加密保护图书"},{"code":"some_advertised_features_are_still_being_built","severity":"info","evidence":"计划功能表中,\"AI-Powered Summarization\" 和 \"Advanced Reading Stats\" 被标记为开发中,\"Handwriting Annotations\" 被标记为计划中。","source":"ai","title":"部分宣传的功能仍在开发中"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["node","docker","rust"],"disk":{"minMB":93,"maxMB":278,"basis":"installer"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["productivity","office"],"alternativeTo":[{"slug":"speechify","name":"Speechify","confirmed":true},{"slug":"deepl-pro","name":"Pro","confirmed":true},{"slug":"audible","name":"Audible","confirmed":false}],"declaredAlternatives":[],"stars":26162,"downloads":2050903,"downloadsWeek":20205,"downloadsDays":4,"growth":207,"growthDays":6,"history":[{"date":"2026-11-06","stars":26012},{"date":"2026-11-07","stars":26043},{"date":"2026-11-08","stars":26083},{"date":"2026-11-09","stars":26127},{"date":"2026-11-10","stars":26162}],"firstPublishedAt":"2026-11-09T23:02:44.304Z","commitSha":"3ee24ebd668fbcb047718c8bd630e71afa76b096","verifiedAt":"2026-11-09T22:54:17.539Z","lastCommitAt":"2026-11-10T17:11:17Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Readest_0.12.20_x64-setup.exe","sizeBytes":62213579,"url":"https://github.com/readest/readest/releases/download/v0.12.20/Readest_0.12.20_x64-setup.exe","releaseUrl":"https://github.com/readest/readest/releases/tag/v0.12.20"},{"os":"mac","file":"Readest_0.12.20_universal.dmg","sizeBytes":71442588,"url":"https://github.com/readest/readest/releases/download/v0.12.20/Readest_0.12.20_universal.dmg","releaseUrl":"https://github.com/readest/readest/releases/tag/v0.12.20"},{"os":"linux","file":"Readest_0.12.20_amd64.AppImage","sizeBytes":213462688,"url":"https://github.com/readest/readest/releases/download/v0.12.20/Readest_0.12.20_amd64.AppImage","releaseUrl":"https://github.com/readest/readest/releases/tag/v0.12.20"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Reads EPUB, PDF, MOBI and more, with highlights, dictionary, translation and read-aloud, on computer, phone or browser.","summary":"A polished ebook reader for EPUB, PDF, MOBI, AZW3, FB2, CBZ, TXT, HTML and Markdown files. You can highlight and take notes, search a whole shelf, look words up in a dictionary or Wikipedia, read two books side by side, have the text read aloud, play audiobooks, or translate sentences with DeepL and Yandex. There are installers for Windows, Mac and Linux, apps on the App Store and Google Play, and a browser version at web.readest.com. An optional account syncs files, progress and notes between devices.","easeRationale":"Every platform is download-and-open: a setup.exe or a portable .exe on Windows, the App Store or a universal .dmg on Mac, AppImage, .deb, .rpm or Flathub on Linux, and nothing at all for the browser version at web.readest.com. Two platform quirks can cost an extra step: Windows needs Microsoft's Edge WebView2 Runtime, and Linux builds need XWayland in a Wayland session.","costNote":"The app is free software (AGPL-3.0) and asks for donations; the project's own hosted service has paid plans, and the self-hosted Docker image unlocks the premium features by default.","notVerified":["Real disk use once your library, covers and offline dictionaries are stored (the installers are 54-204 MB downloads).","How much free storage the official sync gives you, and who can read the files you upload.","Whether the read-aloud voices sound good in your language, since they come from the system or the browser.","Whether the browser version keeps your books only in the browser or sends them to the service.","How the phone apps behave, which this card does not grade."]},"zh-Hans":{"summaryShort":"支持 EPUB、PDF、MOBI 等格式,带高亮、词典、翻译和朗读功能,可在电脑、手机或浏览器上使用。","summary":"一款精致的电子书阅读器,支持 EPUB、PDF、MOBI、AZW3、FB2、CBZ、TXT、HTML 和 Markdown 文件。你可以高亮文字并做笔记,搜索整个书架,在词典或维基百科中查词,并排阅读两本书,让文字被朗读出来,播放有声书,或用 DeepL 和 Yandex 翻译句子。提供 Windows、Mac 和 Linux 的安装程序,App Store 和 Google Play 上的应用,以及 web.readest.com 的浏览器版本。可选账户可在设备之间同步文件、进度和笔记。","easeRationale":"在每个平台上都是下载即用:Windows 上是 setup.exe 或便携版 .exe,Mac 上是 App Store 或通用的 .dmg,Linux 上是 AppImage、.deb、.rpm 或 Flathub,而 web.readest.com 的浏览器版本完全不需要安装。两个平台上的小麻烦可能多花一步:Windows 需要微软的 Edge WebView2 Runtime,Linux 构建版在 Wayland 会话中需要 XWayland。","costNote":"该应用是自由软件(AGPL-3.0),接受捐赠;项目自家的托管服务有付费计划,而自托管的 Docker 镜像默认解锁高级功能。","notVerified":["存储你的书库、封面和离线词典后实际占用的磁盘空间(安装包下载大小为 54-204 MB)。","官方同步给你多少免费存储空间,以及谁能读取你上传的文件。","朗读的声音在你的语言中听起来是否自然,因为它们来自系统或浏览器。","浏览器版本是只把你的书留在浏览器里,还是会发送到服务端。","手机应用的表现如何,这张卡片不对此评分。"]}}},{"id":"R_kgDOU0THuw","slug":"rehan-remade__universal-modder","name":"Universal Modder","owner":"rehan-remade","fullName":"rehan-remade/universal-modder","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"26 天内获得 9019 个星标","source":"check","title":"新仓库，星标很多"},{"code":"reverse_engineering_and_input_control","severity":"attention","evidence":"README.md：逆向工程 skill 使用 \"ILSpy / Cpp2IL / Vineflower / Ghidra and IDA over MCP / Cheat Engine / Frida / RenderDoc\"，`um win drive` 会向游戏发送输入；README 称它 \"asks before driving your mouse and keyboard, installing loaders into game folders\"。","source":"ai","title":"逆向工程与输入控制"},{"code":"game_terms_may_forbid_modding","severity":"info","evidence":"README.md 面向 GTA V、Halo 和 Elden Ring 等游戏，并称 \"It refuses to inject into online games with anti-cheat... or bypass anti-cheat, DRM or ownership checks\"；但修改游戏仍可能违反游戏自身的条款。","source":"ai","title":"游戏条款可能禁止改装（mod）"},{"code":"powershell_with_embedded_code","severity":"info","evidence":"README.md：\"Two no-build Windows tools ship inside the package (`um/ps1/`): WinDrive input and ProcLoopback game-only audio, both PowerShell with embedded C#.\"","source":"ai","title":"PowerShell 中内嵌代码"},{"code":"plugin_hook_changes_path","severity":"info","evidence":"README.md：\"The Claude Code plugin puts it on PATH (its SessionStart hook)\"。","source":"ai","title":"插件的 hook 会修改 PATH"}],"platforms":["windows","linux"],"tools":["python","ffmpeg","other"],"buildTools":["python","ffmpeg"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["AI coding assistant (Claude Code, Codex, Cursor, Gemini CLI or similar)","fal","Sprixen","Voidless"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","games","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":9588,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":201,"growthDays":7,"history":[{"date":"2026-10-26","stars":9020},{"date":"2026-10-27","stars":9089},{"date":"2026-10-28","stars":9154},{"date":"2026-10-29","stars":9197},{"date":"2026-10-30","stars":9248},{"date":"2026-10-31","stars":9289},{"date":"2026-11-01","stars":9320},{"date":"2026-11-02","stars":9348},{"date":"2026-11-03","stars":9387},{"date":"2026-11-04","stars":9420},{"date":"2026-11-05","stars":9443},{"date":"2026-11-06","stars":9472},{"date":"2026-11-07","stars":9503},{"date":"2026-11-08","stars":9529},{"date":"2026-11-09","stars":9557},{"date":"2026-11-10","stars":9588}],"firstPublishedAt":"2026-10-27T01:10:12.675Z","commitSha":"c1dff21663665dd17af289744a301b66e369605e","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-06T08:37:53Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Add-on that lets an AI coding assistant build mods for PC games you own, including new art, 3D models and sound.","summary":"Universal Modder is a set of instructions, tools and notes for AI coding assistants such as Claude Code, Codex or Cursor. You ask the assistant to mod a PC game you own, and it identifies the game engine, reads the game's code, writes the mod, makes art and sound with the paid fal service, tests it in the running game and records a video. It is for gamers and modders comfortable with a terminal. You need an AI coding assistant, Git, Python 3.10+ and FFmpeg; Blender is needed for 3D renders. The README says it refuses to bypass anti-cheat, DRM or ownership checks.","easeRationale":"There is no installer: you add it as a plugin inside an AI coding assistant, install the `um` tool with uv or pipx, and need Git, Python, FFmpeg and optionally Blender.","costNote":"The toolkit is free (MIT); the coding assistant and asset services (fal, Sprixen, Voidless) are paid separately. A local ComfyUI server can make images without a key.","notVerified":["Whether a given game can actually be modded this way, and how many attempts it takes.","Real cost in AI assistant usage and fal credits for one mod.","What the PowerShell tools with embedded C# do on your system.","Whether a mod could break a game install or save files despite the backup step.","Whether modding a specific game breaks that game's terms of use."]},"zh-Hans":{"summaryShort":"一个插件，让 AI 编程助手为你拥有的 PC 游戏制作模组，包括新的美术、3D 模型和音效。","summary":"Universal Modder 是一套面向 Claude Code、Codex、Cursor 等 AI 编程助手的说明、工具和笔记。你让助手为你拥有的某款 PC 游戏制作模组，它会识别游戏引擎、阅读游戏代码、编写模组，用付费的 fal 服务生成美术和音效，在运行中的游戏里测试，并录制视频。它适合习惯使用终端的玩家和模组作者。你需要一个 AI 编程助手、Git、Python 3.10+ 和 FFmpeg；3D 渲染还需要 Blender。README 称它会拒绝绕过反作弊、DRM 或所有权验证。","easeRationale":"没有安装程序：你要把它作为插件添加到 AI 编程助手中，用 uv 或 pipx 安装 `um` 工具，还需要 Git、Python、FFmpeg，以及可选的 Blender。","costNote":"该工具包免费（MIT）；编程助手和素材服务（fal、Sprixen、Voidless）需另外付费。本地 ComfyUI 服务器无需密钥即可生成图片。","notVerified":["某款游戏是否真的能用这种方式制作模组，以及需要尝试多少次。","制作一个模组实际需要花费多少 AI 助手用量和 fal 额度。","内嵌 C# 的 PowerShell 工具会在你的系统上做什么。","尽管有备份步骤，模组是否仍可能损坏游戏安装或存档文件。","为某款游戏制作模组是否违反该游戏的使用条款。"]}}},{"id":"R_kgDOOeP4SQ","slug":"resemble-ai__chatterbox","name":"Chatterbox","owner":"resemble-ai","fullName":"resemble-ai/chatterbox","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"declared_retired_by_authors","severity":"attention","evidence":"README.md:\"As of August 2026, Chatterbox is retired research and is no longer maintained.\" 以及 \"The code stays public as an archive under the MIT license, with no support, updates, or paid tier.\" 该仓库在 GitHub 上并未被标记为存档,检查时的最后一次提交是 2026-11-04,距检查时间五天,但自 2025-06-13 的 v0.1.2 以来再无新版本发布,且有 381 个未关闭的 issue。","source":"ai","title":"作者已宣布项目退役"},{"code":"voice_cloning_of_real_people","severity":"attention","evidence":"使用示例会从一段简短样本中克隆声音:`wav = model.generate(text, audio_prompt_path=\"your_10s_ref_clip.wav\")`,注释写着 \"requires a reference clip for voice cloning\"。在一些国家,未经许可复制他人声音受法律限制。作者给出的唯一提示是免责声明 \"Don't use this model to do bad things.\"","source":"ai","title":"克隆真实人物的声音"},{"code":"output_audio_is_watermarked","severity":"info","evidence":"README.md:\"Every audio file generated by Chatterbox includes Resemble AI's Perth (Perceptual Threshold) Watermarker - imperceptible neural watermarks that survive MP3 compression, audio editing, and common manipulations while maintaining nearly 100% detection accuracy.\" README 中还展示了从文件中读取出水印的脚本。","source":"ai","title":"输出的音频带有水印"},{"code":"dependency_installed_from_a_git_branch","severity":"info","evidence":"pyproject.toml 直接从 GitHub 安装一个依赖项,且没有固定版本:\"resemble-perth @ git+https://github.com/resemble-ai/Perth.git@master\",所以安装时该分支上的内容就是你得到的版本。","source":"ai","title":"依赖项直接从 git 分支安装"},{"code":"pinned_library_versions","severity":"info","evidence":"pyproject.toml 固定了精确版本,如 \"torch==2.6.0\"、\"transformers==5.2.0\"、\"diffusers==0.29.0\" 和 \"gradio==6.8.0\";README 解释说这样固定是 \"to ensure consistency\",但这可能与同一环境中的其他 Python 程序发生冲突。","source":"ai","title":"库版本被固定"}],"platforms":["windows","mac","linux","web"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true}],"declaredAlternatives":[],"stars":27748,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":181,"growthDays":6,"history":[{"date":"2026-11-06","stars":27618},{"date":"2026-11-07","stars":27650},{"date":"2026-11-08","stars":27682},{"date":"2026-11-09","stars":27715},{"date":"2026-11-10","stars":27748}],"firstPublishedAt":"2026-11-09T23:02:44.314Z","commitSha":"ab55923c27757c90cfcc407baa57dc25a7d25217","verifiedAt":"2026-11-09T22:46:43.489Z","lastCommitAt":"2026-11-04T18:48:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Reads typed text aloud on your own computer, and can copy a voice from a 10-second recording in 23 languages.","summary":"A family of text-to-speech models from Resemble AI. You type text and it speaks it, either with its own voice or copying the voice in a short recording you provide, in English or in 23 languages. Small models can even run on a normal processor. You install it with Python (`pip install chatterbox-tts`) and run one of the included example scripts or the small web pages that come with the project; a graphics card makes it much faster. There are also free online demo pages made by the authors. The README says the project was retired in August 2026 and is no longer maintained.","easeRationale":"On the web it is an A: the authors keep free demo pages on Hugging Face Spaces where you type text and listen, with nothing to install. On Windows, Mac and Linux it is a C: you install Python, run `pip install chatterbox-tts`, then run one of the included scripts or web pages (`example_tts.py`, `gradio_tts_app.py`), and the model files download on first use; the project was developed and tested on Python 3.11 on Linux, and a graphics card is assumed in the examples (`device=\"cuda\"`), with a separate `example_for_mac.py` for Macs.","costNote":"No price: MIT code, models published on Hugging Face and free demo pages. The README says there is \"no support, updates, or paid tier\" and points to Resemble AI's own paid speech service for anyone who needs to scale.","notVerified":["How the voices really sound, and how well the voice copying works for each of the 23 languages","Whether it runs on a Mac or on a computer with no graphics card, and how slow it is (the README claims the small Nano model is 3x faster than realtime on 8 processor cores)","Real disk space used after the model files download on first use","Which licence the model files on Hugging Face carry, which can differ from the MIT licence of the code","Whether the free demo pages and the Hugging Face model downloads stay online now that the authors say the project is retired"]},"zh-Hans":{"summaryShort":"在你自己的电脑上朗读输入的文字,还能从 10 秒录音中复制一个声音,支持 23 种语言。","summary":"来自 Resemble AI 的一系列文本转语音模型。你输入文字,它就会读出来,可以用自带的声音,也可以复制你提供的一段简短录音中的声音,支持英语或 23 种语言。小模型甚至可以在普通处理器上运行。你用 Python 安装它(`pip install chatterbox-tts`),然后运行项目自带的示例脚本或小型网页;有显卡会快得多。作者还提供了免费的在线演示页面。README 说该项目已于 2026 年 8 月退役,不再维护。","easeRationale":"在网页上是 A 级:作者在 Hugging Face Spaces 上维护免费演示页面,你只需输入文字就能听到结果,无需安装任何东西。在 Windows、Mac 和 Linux 上是 C 级:你要安装 Python,运行 `pip install chatterbox-tts`,再运行项目自带的脚本或网页之一(`example_tts.py`、`gradio_tts_app.py`),模型文件会在第一次使用时下载;项目是在 Linux 上用 Python 3.11 开发和测试的,示例默认假设有显卡(`device=\"cuda\"`),Mac 则有单独的 `example_for_mac.py`。","costNote":"没有定价:MIT 代码,模型发布在 Hugging Face 上,还有免费的演示页面。README 说该项目 \"no support, updates, or paid tier\",需要扩展规模的人可以使用 Resemble AI 自己的付费语音服务。","notVerified":["这些声音实际听起来如何,以及语音复制功能在 23 种语言中各自的效果如何。","它能否在 Mac 或没有显卡的电脑上运行,速度有多慢(README 声称小型 Nano 模型在 8 核处理器上比实时速度快 3 倍)。","模型文件首次使用时下载后,实际占用的磁盘空间。","Hugging Face 上的模型文件采用哪种许可证,它可能与代码的 MIT 许可证不同。","既然作者说项目已退役,免费演示页面和 Hugging Face 的模型下载是否还能继续使用。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxOTIwNTg5Ng==","slug":"restic__restic","name":"restic","owner":"restic","fullName":"restic/restic","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"lost_password_means_lost_backup","severity":"attention","evidence":"README自己的快速入门部分写道：\"Please note that knowledge of your password is required to access the repository. Losing your password means that your data is irrecoverably lost.\"如果忘记密码，没有任何恢复途径。","source":"ai","title":"丢失密码就意味着丢失备份"},{"code":"reproducible_release_builds","severity":"info","evidence":"README：\"The binaries released with each restic version starting at 0.6.1 are reproducible, which means that you can reproduce a byte identical version from the source code for that release\"，具体说明在项目的builder仓库中。","source":"ai","title":"可复现的发行版构建"},{"code":"release_older_than_last_commit","severity":"info","evidence":"最新的发行版v0.19.1发布于2026-07-05，而最后一次提交是在2026-10-18，所以7月以来的修复只存在于源代码中，并未包含在可下载的二进制文件里。","source":"ai","title":"发行版比最新提交更旧"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["go"],"disk":{"minMB":11,"maxMB":32,"basis":"installer"},"apiKeys":["Amazon S3 or S3-compatible storage","Backblaze B2","Microsoft Azure Blob Storage","Google Cloud Storage"],"cost":{"model":"free"},"license":{"spdx":"BSD-2-Clause","category":"permissive","detectedFrom":"github"},"categories":["network_system","productivity"],"alternativeTo":[{"slug":"arq-backup","name":"Arq Backup","confirmed":true},{"slug":"dropbox","name":"Dropbox","confirmed":true}],"declaredAlternatives":[],"stars":37131,"downloads":37168146,"downloadsWeek":297120,"downloadsDays":7,"growth":147,"growthDays":7,"history":[{"date":"2026-10-26","stars":36819},{"date":"2026-10-27","stars":36847},{"date":"2026-10-28","stars":36863},{"date":"2026-10-29","stars":36882},{"date":"2026-10-30","stars":36892},{"date":"2026-10-31","stars":36911},{"date":"2026-11-01","stars":36934},{"date":"2026-11-02","stars":36958},{"date":"2026-11-03","stars":36984},{"date":"2026-11-04","stars":36999},{"date":"2026-11-05","stars":37013},{"date":"2026-11-06","stars":37035},{"date":"2026-11-07","stars":37059},{"date":"2026-11-08","stars":37082},{"date":"2026-11-09","stars":37105},{"date":"2026-11-10","stars":37131}],"firstPublishedAt":"2026-10-27T01:10:12.676Z","commitSha":"dd09b6bd01af220b1d5d95c0d4d7d472e05e05c4","verifiedAt":"2026-11-01T01:43:07.726Z","lastCommitAt":"2026-11-06T20:42:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Backs up your files, encrypted and without storing the same data twice, to a disk, a server or cloud storage.","summary":"restic is a backup program you run from the terminal. It creates an encrypted \"repository\" on a local disk, an SSH server, Amazon S3, Backblaze B2, Azure, Google Cloud Storage or anything rclone reaches, then saves snapshots of your folders; repeated data is stored only once, so later backups are small. You can list snapshots, restore files or mount a backup as a folder to browse it. One ready binary of about 10 MB per system, no other software needed.","easeRationale":"Equal on the three platforms: each release has a single ready binary (restic_0.19.1_windows_amd64.zip, _darwin_arm64.bz2, _linux_amd64.bz2) with nothing else to install, but there is no window or buttons — you type commands such as `restic init --repo` and `restic backup ~/work` and set a password. Scheduling it to run every day is up to you.","costNote":"BSD 2-Clause, free. You still pay whatever the storage costs if you back up to a cloud service.","notVerified":["Whether a restore brings back everything on your own system, especially Windows permissions and special files","How long a first backup takes and how much space it uses with your data","What a cloud provider will charge for the storage and the transfers","Whether the encryption holds up to review, which needs a security audit"]},"zh-Hans":{"summaryShort":"将你的文件加密备份到磁盘、服务器或云存储，且不会重复存储相同的数据。","summary":"restic是一个从终端运行的备份程序。它会在本地磁盘、SSH服务器、Amazon S3、Backblaze B2、Azure、Google Cloud Storage或任何rclone能访问的地方创建一个加密的\"仓库\"，然后保存你文件夹的快照；重复的数据只存储一次，因此之后的备份体积很小。你可以列出快照、恢复文件，或把一个备份挂载为文件夹来浏览。每个系统只有一个约10 MB的现成二进制文件，不需要其他软件。","easeRationale":"三个平台表现一致：每个release都有一个现成的独立二进制文件（restic_0.19.1_windows_amd64.zip、_darwin_arm64.bz2、_linux_amd64.bz2），不需要安装其他任何东西，但没有窗口也没有按钮——你要输入诸如`restic init --repo`和`restic backup ~/work`这样的命令，并设置一个密码。要让它每天自动运行，需要你自己来设置计划任务。","costNote":"BSD 2-Clause许可证，免费。如果你备份到云服务，仍然需要支付相应的存储费用。","notVerified":["恢复操作能否找回你系统上的所有内容，尤其是Windows的权限和特殊文件","首次备份需要多长时间，以及用你的数据会占用多少空间","云服务商会为存储和传输收取多少费用","这套加密方案是否经得起检验，这需要专门的审计来确认"]}}},{"id":"MDEwOlJlcG9zaXRvcnk1MTQzODA5","slug":"retropie__retropie-setup","name":"RetroPie","owner":"RetroPie","fullName":"RetroPie/RetroPie-Setup","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布于 2022-03-14(1701 天前)","source":"check","title":"版本发布已过时"},{"code":"needs_your_own_game_files","severity":"attention","evidence":"README 开头写着 \"Shell script to setup the Raspberry Pi, Vero4K, ODroid-C1 or a PC running Ubuntu with many emulators and games\",但仓库顶层只有 retropie_setup.sh、retropie_packages.sh 和 scriptmodules/,README 从未列出这些脚本会带来哪些游戏。老主机的游戏文件和启动(BIOS)文件受版权保护,许多国家限制复制它们。","source":"ai","title":"需要你自己提供游戏文件"},{"code":"runs_as_root_and_installs_software","severity":"attention","evidence":"README 让你用 \"sudo ./retropie_setup.sh\" 启动,并提醒 \"When you first run the script it may install some additional packages that are needed\",所以该脚本会以管理员身份修改你的系统,并从这个仓库之外获取软件。","source":"ai","title":"以 root 权限运行并安装软件"}],"platforms":["linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"file"},"categories":["games","hardware"],"alternativeTo":[],"declaredAlternatives":[],"stars":10455,"downloads":11004179,"downloadsWeek":1144,"downloadsDays":2,"growth":5,"growthDays":6,"history":[{"date":"2026-11-08","stars":10452},{"date":"2026-11-09","stars":10452},{"date":"2026-11-10","stars":10455}],"firstPublishedAt":"2026-11-09T23:08:44.829Z","commitSha":"427e76216633591861791cfb28807ac1039c5ea8","verifiedAt":"2026-11-09T23:04:11.172Z","lastCommitAt":"2026-11-08T04:11:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a Raspberry Pi, ODroid or Ubuntu PC into a retro game machine, installing emulators from a text menu.","summary":"RetroPie-Setup is a menu-driven shell script that installs emulators and the EmulationStation front end on a Raspberry Pi, a Vero4K, an ODroid-C1 or a PC running Ubuntu. You update your packages, install git, copy the repository with git and run the setup script with sudo, then pick what to install from a text menu. The README says a Raspberry Pi can install ready-made binaries, while the other devices build everything from source, which \"can take a long time\". The repository itself holds only scripts.","easeRationale":"Linux is the only platform, and the README's own steps are several terminal commands: apt update, apt dist-upgrade, installing git, cloning the repository and running \"sudo ./retropie_setup.sh\", followed by choices in a text menu. The three latest releases list no ready files to download, and outside the Raspberry Pi every emulator is compiled on your machine.","notVerified":["Which games, if any, the scripts in scriptmodules/ actually install, since the README does not list them","Whether the setup works on each listed device (Raspberry Pi, Vero4K, ODroid-C1, Ubuntu PC)","How much disk space and how many hours a full install or source build takes","Which servers the binary and source packages are downloaded from at run time, and what those third-party emulators contain"]},"zh-Hans":{"summaryShort":"把 Raspberry Pi、ODroid 或 Ubuntu 电脑变成一台复古游戏机,通过文字菜单安装模拟器。","summary":"RetroPie-Setup 是一个菜单式的 shell 脚本,可以在 Raspberry Pi、Vero4K、ODroid-C1 或运行 Ubuntu 的 PC 上安装模拟器和 EmulationStation 前端。你需要更新软件包、安装 git、用 git 复制仓库,再用 sudo 运行安装脚本,然后从文字菜单中选择要安装的内容。README 说 Raspberry Pi 可以安装现成的二进制文件,而其他设备则要从源代码编译一切,这 \"can take a long time\"。仓库本身只包含脚本。","easeRationale":"Linux 是唯一支持的平台,README 自己给出的步骤就有好几条终端命令:apt update、apt dist-upgrade、安装 git、克隆仓库,然后运行 \"sudo ./retropie_setup.sh\",之后在文字菜单中做选择。最近的三个版本都没有列出可直接下载的现成文件,除了 Raspberry Pi 之外,每个模拟器都要在你自己的机器上编译。","notVerified":["scriptmodules/ 里的脚本实际安装了哪些游戏(如果有的话),因为 README 并未列出。","该安装流程是否适用于列出的每种设备(Raspberry Pi、Vero4K、ODroid-C1、Ubuntu 电脑)。","完整安装或从源代码编译需要多少磁盘空间和多少小时。","运行时二进制包和源代码包是从哪些服务器下载的,以及这些第三方模拟器包含什么内容。"]}}},{"id":"R_kgDOVMk9fQ","slug":"revivedx__savagewindrecomp","name":"SavageWindRecomp","owner":"Revivedx","fullName":"Revivedx/SavageWindRecomp","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C"},"risk":"high","signals":[{"code":"needs_game_files_you_dump_yourself","severity":"attention","evidence":"README：\"Savage Wind Recomp needs your own legally obtained copy of the game, updated to version 1.9.0, dumped from your own Nintendo Switch\"，以及\"Dumping a game you own for personal use is legal in some countries and not in others; that is your responsibility.\"该项目不提供任何游戏文件、密钥或转储工具。","source":"ai","title":"需要你自己转储的游戏文件"},{"code":"unofficial_nintendo_fan_project","severity":"attention","evidence":"README的\"Respect copyright\"部分：\"This is a fan project. It is not affiliated with, endorsed by or sponsored by Nintendo.\"你生成的构建包含任天堂的代码，\"for your own use only: don't share them, upload them or ship them.\"","source":"ai","title":"非官方的任天堂同人项目"},{"code":"written_almost_entirely_by_ai","severity":"info","evidence":"README的第一条说明：\"This project is made almost entirely with AI. Nearly all of the code, tooling, reverse engineering and documentation here is written by an AI (Anthropic's Claude, through Claude Code), directed, tested and reviewed by a human.\"","source":"ai","title":"几乎完全由AI编写"},{"code":"alpha_only_opening_played","severity":"attention","evidence":"README警告：\"Nobody has played past the Great Plateau on Savage Wind yet, so expect bugs or crashes further on. This is an alpha, a preview for testers.\"状态表显示\"Tested in game\"为50%，DLC和mod为0%。","source":"ai","title":"仅测试过开场的alpha版本"},{"code":"possible_illegal_use","severity":"high","evidence":"主要用途依赖于从主机提取的文件，生成的程序包含任天堂的代码。README：\"Savage Wind Recomp needs your own legally obtained copy of the game, updated to version 1.9.0, dumped from your own Nintendo Switch\"；\"Follow the copyright laws of your country. Dumping a game you own for personal use is legal in some countries and not in others; that is your responsibility.\"；\"The game code you generate and the build made from it contain Nintendo's code. They are for your own use only: don't share them, upload them or ship them.\"","source":"ai","title":"可能存在的违法使用"}],"platforms":["windows","linux"],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":86,"maxMB":258,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":3802,"downloads":144704,"downloadsWeek":51102,"downloadsDays":7,"growth":823,"growthDays":7,"history":[{"date":"2026-10-27","stars":2086},{"date":"2026-10-28","stars":2218},{"date":"2026-10-29","stars":2315},{"date":"2026-10-30","stars":2431},{"date":"2026-10-31","stars":2561},{"date":"2026-11-01","stars":2743},{"date":"2026-11-02","stars":2891},{"date":"2026-11-03","stars":2979},{"date":"2026-11-04","stars":3077},{"date":"2026-11-05","stars":3201},{"date":"2026-11-06","stars":3315},{"date":"2026-11-07","stars":3421},{"date":"2026-11-08","stars":3567},{"date":"2026-11-09","stars":3696},{"date":"2026-11-10","stars":3802}],"firstPublishedAt":"2026-10-27T18:40:28.036Z","commitSha":"d71294213781a982f13004518d91d4e37d228764","verifiedAt":"2026-10-27T18:35:21.886Z","lastCommitAt":"2026-11-09T01:26:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your own dumped copy of Zelda: Breath of the Wild into a Windows or Linux program running at 60 fps.","summary":"Savage Wind Recomp rebuilds The Legend of Zelda: Breath of the Wild (Switch version 1.9.0) as a native Windows or Linux program, with PC graphics options, keyboard and mouse, any controller and 60 fps or more. No game files come with it: you supply a copy you extracted yourself from your own Switch, and the program rebuilds the game code on your PC once, about 15 minutes, 35 GB of free space and a Vulkan 1.3 graphics card. It is an early alpha: only the opening and the Great Plateau have been played through. Dumping a game you own is allowed in some countries and not in others.","easeRationale":"The Windows release is a zip you extract and run with nothing to install, but it only starts once you point it at game files you extracted from your own console, and the first run rebuilds the game code on your machine, about 15 minutes with 12 threads, 35 GB of free space and a GPU with Vulkan 1.3.","costNote":"The program is free. You need the game itself, bought on a Nintendo Switch and extracted by you; the project supplies no game files, keys or dumping tools.","notVerified":["How far the game really plays past the Great Plateau on your own machine.","How long the one-time build takes on a CPU with fewer threads or without AVX2, where it runs in a slower compatible mode.","Whether the ARM AppImage works at all: the README says it has never run on real ARM hardware.","Whether the released zip matches the code in the repository, since the binaries are built by the project."]},"zh-Hans":{"summaryShort":"把你自己提取的《塞尔达传说:旷野之息》副本变成一个在Windows或Linux上以60帧运行的程序。","summary":"Savage Wind Recomp 将《塞尔达传说:旷野之息》(Switch 1.9.0版本)重建为可在Windows或Linux上原生运行的程序,支持PC画面选项、键盘鼠标、任意手柄,帧率可达60帧或更高。它不附带任何游戏文件:你需要自己从自己的Switch上提取一份游戏拷贝,程序会在你的电脑上把游戏代码重新编译一次,大约15分钟,需要35GB空闲空间和一块支持Vulkan 1.3的显卡。目前是早期alpha版本:只有开场和初始高地部分被完整通关测试过。提取自己拥有的游戏在一些国家是合法的,在另一些国家则不是。","easeRationale":"Windows版本是一个压缩包,解压后直接运行,无需安装任何东西,但只有在你指定从自己主机提取的游戏文件后才能启动,首次运行会在你的电脑上重新编译游戏代码,使用12线程大约需要15分钟,还需要35GB空闲空间和一块支持Vulkan 1.3的显卡。","notVerified":["在你自己的设备上,游戏在初始高地之后到底能玩到多远。","在线程数较少或不支持AVX2的CPU上,一次性编译需要多长时间,此时会以较慢的兼容模式运行。","ARM版AppImage是否真的能用:README说它从未在真实ARM硬件上运行过。","发布的压缩包是否与仓库中的代码一致,因为二进制文件是由项目方自行编译的。"]}}},{"id":"R_kgDOIeLh3g","slug":"riffusion__riffusion-app-hobby","name":"Riffusion","owner":"riffusion","fullName":"riffusion/riffusion-app-hobby","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖项存在已知漏洞(OSV):next: GHSA-2464-fhr4-w9xw, GHSA-2xp9-vwfh-vxw4, GHSA-36qx-fr4f-26g5","source":"check","title":"存在已知漏洞"},{"code":"declared_no_longer_maintained","severity":"attention","evidence":"README.md 第一行:\":no_entry: This project is no longer actively maintained.\" 检查时的最后一次提交是 2024-07-22,距检查时间 840 天,没有任何版本发布,且有 25 个未关闭的 issue。","source":"ai","title":"已声明不再积极维护"},{"code":"half_of_the_app_only","severity":"attention","evidence":"README.md:\"To actually generate model outputs, we need a model backend that responds to inference requests via API\",指向另一个仓库来提供推理服务器。这个仓库只包含 Next.js 页面部分:`pages/`、`components/`、`shaders.js`、`samplePrompts.ts`。","source":"ai","title":"只是应用的一半"},{"code":"frozen_old_dependencies","severity":"info","evidence":"package.json 固定了 2022 年底的版本,比如 \"next\": \"13.0.4\"、\"react\": \"18.2.0\" 和 \"eslint\": \"8.27.0\";这正是自动检查标记为 `next` 已知漏洞的原因,而且没有人在更新这些版本。","source":"ai","title":"依赖项被冻结在旧版本"},{"code":"page_sends_usage_data","severity":"info","evidence":"package.json 包含 \"@vercel/analytics\": \"^0.1.6\",这是 Vercel 的访客分析库,当应用部署在 Vercel 上时会报告页面访问情况。","source":"ai","title":"页面会发送使用数据"}],"platforms":["windows","mac","linux"],"tools":["node","python","gpu_any"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":2671,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":2671},{"date":"2026-11-07","stars":2671},{"date":"2026-11-08","stars":2671},{"date":"2026-11-09","stars":2671},{"date":"2026-11-10","stars":2671}],"firstPublishedAt":"2026-11-09T23:02:44.331Z","commitSha":"3a72a8ff41e8c9c84774bd75638c1176508106d3","verifiedAt":"2026-11-09T22:48:25.573Z","lastCommitAt":"2024-07-22T18:01:36Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"The web page behind Riffusion, which turns text prompts into music loops, plus a separate AI server you must run yourself.","summary":"This is only the web page of Riffusion, the project that makes music from text prompts using an image-generation model on sound pictures. You install Node 18 or newer, run `npm install` and `npm run dev`, and open localhost:3000. On its own it produces nothing: the README says you must also clone and run a second project, the inference server, on \"a large GPU that can run stable diffusion in under five seconds\", and point this page at it in a `.env.local` file. The README says the project is no longer actively maintained.","easeRationale":"A D on Windows, Mac and Linux alike: there is no installer, you need Node 18 or newer, `npm install` and `npm run dev` for the page, and then a second project (the Flask inference server at hmartiro/riffusion-inference) on a graphics card strong enough to \"run stable diffusion in under five seconds\", wired up through a `RIFFUSION_FLASK_URL` entry in a `.env.local` file you create yourself.","costNote":"No price: MIT code. The cost is the computer or rented graphics card needed to run the separate inference server.","notVerified":["Whether the hosted page at riffusion.vercel.app listed as the project homepage is still online","Whether `npm install` still succeeds with these 2022 package versions on current Node releases","Whether the separate inference server project still runs, and with which model files","What graphics card and how much video memory are really needed for music to come out in near real time","The quality and length of the music it produces today"]},"zh-Hans":{"summaryShort":"Riffusion 背后的网页,能把文字提示变成音乐循环,还需要你自己运行一个独立的 AI 服务器。","summary":"这只是 Riffusion 的网页部分,Riffusion 是一个用图像生成模型处理声音图片、从文字提示生成音乐的项目。你需要安装 Node 18 或更新版本,运行 `npm install` 和 `npm run dev`,然后打开 localhost:3000。它单独运行什么都做不出来:README 说你还必须克隆并运行第二个项目,也就是推理服务器,运行在 \"a large GPU that can run stable diffusion in under five seconds\" 的显卡上,并在 `.env.local` 文件里把这个网页指向它。README 说该项目已不再积极维护。","easeRationale":"在 Windows、Mac 和 Linux 上都是 D 级:没有安装程序,你需要 Node 18 或更新版本,为网页运行 `npm install` 和 `npm run dev`,然后还要有第二个项目(位于 hmartiro/riffusion-inference 的 Flask 推理服务器),运行在足够强大、能 \"run stable diffusion in under five seconds\" 的显卡上,并通过你自己创建的 `.env.local` 文件中的 `RIFFUSION_FLASK_URL` 配置项连接起来。","costNote":"没有定价:MIT 代码。成本在于运行独立推理服务器所需的电脑或租用的显卡。","notVerified":["作为项目主页列出的 riffusion.vercel.app 托管页面是否仍在线。","`npm install` 在当前的 Node 版本上是否还能用这些 2022 年的包版本成功安装。","独立的推理服务器项目是否还能运行,使用的是哪些模型文件。","要让音乐接近实时生成,实际需要什么显卡和多少显存。","它今天生成的音乐的质量和长度。"]}}},{"id":"R_kgDOIgQlGg","slug":"riffusion__riffusion-hobby","name":"Riffusion","owner":"riffusion","fullName":"riffusion/riffusion-hobby","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布于 2022-12-27(1413 天前)","source":"check","title":"版本发布已过时"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖项存在已知漏洞(OSV):diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446","source":"check","title":"存在已知漏洞"},{"code":"declared_no_longer_maintained","severity":"attention","evidence":"README.md 第二行:\":no_entry: This project is no longer actively maintained.\" 检查时的最后一次提交是 2024-07-22,距检查时间 840 天,唯一的发布版本是 2022-12-27 的 v0.3.0,且有 71 个未关闭的 issue。","source":"ai","title":"已声明不再积极维护"},{"code":"pinned_library_is_hard_to_update","severity":"info","evidence":"requirements.txt 固定了 \"diffusers==0.9.0\"(自动检查标记为存在已知漏洞的版本),README 说 \"If you have an issue, try upgrading diffusers... Tested with 0.9 - 0.11\",所以更新的版本从未与这份代码一起测试过。","source":"ai","title":"固定的库版本难以更新"},{"code":"real_time_needs_a_strong_graphics_card","severity":"info","evidence":"README 的 Backends 部分写道:\"`cpu` is supported but is quite slow\",而要实现实时音频,\"you need a GPU that can run stable diffusion with approximately 50 steps in under five seconds, such as a 3090 or A10G\"。Apple Silicon 后端 \"is not deterministic\"。","source":"ai","title":"实时生成需要强力显卡"},{"code":"model_files_come_from_outside","severity":"info","evidence":"该仓库没有任何模型文件;README 指向一个单独的下载地址,\"Model checkpoint: https://huggingface.co/riffusion/riffusion-model-v1\",服务器允许你传入 \"--checkpoint with your own directory or huggingface ID\"。","source":"ai","title":"模型文件来自仓库之外"},{"code":"windows_help_is_a_community_link","severity":"info","evidence":"README 中唯一的 Windows 说明是一个外部链接:\"Simple Install Guide for Windows\",指向一个 Reddit 帖子,而不是仓库内的文档。","source":"ai","title":"Windows 帮助只是社区链接"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_any","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":3907,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":4,"growthDays":6,"history":[{"date":"2026-11-05","stars":3904},{"date":"2026-11-06","stars":3905},{"date":"2026-11-07","stars":3907},{"date":"2026-11-08","stars":3907},{"date":"2026-11-09","stars":3907},{"date":"2026-11-10","stars":3907}],"firstPublishedAt":"2026-11-09T23:02:44.340Z","commitSha":"94c29abbdc1de60b3e03131715ea8c3da12bd933","verifiedAt":"2026-11-09T22:49:11.587Z","lastCommitAt":"2024-07-22T18:00:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes short music clips from text prompts on your own computer, by turning pictures of sound into audio.","summary":"The core code of Riffusion: it writes music by generating images of sound with a Stable Diffusion model and turning them back into audio. It is a Python library with a command-line tool, an interactive page you start yourself, and a small server the Riffusion web app can talk to. You need Python 3.9 or 3.10, a list of Python packages, ffmpeg for formats other than WAV, and for real-time results a graphics card \"that can run stable diffusion with approximately 50 steps in under five seconds, such as a 3090\". The README says it is no longer actively maintained.","easeRationale":"D everywhere: there is no installer. On Linux and Mac you create a conda environment, run `python -m pip install -r requirements.txt`, install ffmpeg, possibly `libsndfile`, and then start the command line, the streamlit page or the Flask server; Apple Silicon works through the `mps` backend but \"some operations fall back to CPU\" and may need `PYTORCH_ENABLE_MPS_FALLBACK=1`. Windows is the same work with no in-repo instructions: the only Windows guide the README offers is a link to a Reddit post.","costNote":"No price: MIT code and the model checkpoint published on Hugging Face. What it costs you is the graphics card or rented machine to run it.","notVerified":["Whether the 2022 package list still installs on current Python and PyTorch versions","How the generated music actually sounds, and how long the clips are","Real disk space after the model checkpoint downloads","Whether it runs on a Mac or on a computer with no dedicated graphics card, and how slowly","Whether the riffusion.com site mentioned in the README still relates to this code"]},"zh-Hans":{"summaryShort":"在你自己的电脑上,通过把声音图片转换成音频,从文字提示生成简短的音乐片段。","summary":"Riffusion 的核心代码:它用 Stable Diffusion 模型生成声音图片,再把它们转换回音频,从而创作音乐。这是一个 Python 库,带有命令行工具、一个你自己启动的交互页面,以及一个 Riffusion 网页应用可以访问的小型服务器。你需要 Python 3.9 或 3.10、一系列 Python 软件包、处理非 WAV 格式所需的 ffmpeg,以及要实现实时效果所需的显卡,README 要求它 \"that can run stable diffusion with approximately 50 steps in under five seconds, such as a 3090\"。README 说该项目已不再积极维护。","easeRationale":"各平台都是 D 级:没有安装程序。在 Linux 和 Mac 上,你要创建一个 conda 环境,运行 `python -m pip install -r requirements.txt`,安装 ffmpeg,可能还需要 `libsndfile`,然后启动命令行、streamlit 页面或 Flask 服务器;Apple Silicon 通过 `mps` 后端运行,但 \"some operations fall back to CPU\",可能需要设置 `PYTORCH_ENABLE_MPS_FALLBACK=1`。Windows 需要同样的工作量,而仓库里没有相应说明:README 给出的唯一 Windows 指南是一个指向 Reddit 帖子的链接。","costNote":"没有定价:MIT 代码,模型检查点发布在 Hugging Face 上。花费在于运行它所需的显卡或租用的机器。","notVerified":["这份 2022 年的软件包列表能否在当前的 Python 和 PyTorch 版本上成功安装。","生成的音乐实际听起来如何,片段有多长。","模型检查点下载后实际占用的磁盘空间。","它能否在 Mac 或没有独立显卡的电脑上运行,速度有多慢。","README 中提到的 riffusion.com 网站是否仍与这份代码相关。"]}}},{"id":"R_kgDOUd-wKA","slug":"robbietilton__compositor","name":"Compositor","owner":"robbietilton","fullName":"robbietilton/Compositor","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"mac":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"41 天内获得 18067 个星标","source":"check","title":"新仓库，星标很多"},{"code":"needs_newest_macos_and_apple_silicon","severity":"attention","evidence":"README.md 的 Requirements 部分：\"macOS 26.0 or later on a Mac with Apple silicon\"。Intel 芯片的 Mac 和较旧版本的 macOS 完全无法运行它。","source":"ai","title":"需要最新的 macOS 和 Apple 芯片"},{"code":"single_maintainer","severity":"info","evidence":"README.md 关于翻译的说明：\"It's maintained by one person, every new string would need translating from then on\"。该应用只有英文版，不接受本地化的 pull request。","source":"ai","title":"只有一位维护者"},{"code":"very_fast_release_cadence","severity":"info","evidence":"从仓库创建的 2026-09-16 到 2026-10-27 共发布了 58 个版本，其中三个（v1.5.10 至 v1.5.12）出现在两天之内。该应用会自动更新，所以改动很快就会到你手上，无论是否经过测试。","source":"ai","title":"发布节奏非常快"},{"code":"unofficial_ports_for_other_systems","severity":"info","evidence":"README.md 列出了由其他人制作的 Windows 和 Linux 移植版，并说：\"These unofficial ports are made by others, not reviewed or supported by me\"。这些下载来自本项目无法控制的仓库。","source":"ai","title":"面向其他系统的非官方移植版"}],"platforms":["mac"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"photoshop","name":"Photoshop","confirmed":true},{"slug":"pixelmator-pro","name":"Pixelmator Pro","confirmed":true},{"slug":"adobe-camera-raw","name":"Camera Raw","confirmed":true},{"slug":"lightroom","name":"Lightroom","confirmed":true}],"declaredAlternatives":[],"stars":18589,"downloads":128825,"downloadsWeek":8883,"downloadsDays":7,"growth":174,"growthDays":7,"history":[{"date":"2026-10-26","stars":17950},{"date":"2026-10-27","stars":18065},{"date":"2026-10-28","stars":18126},{"date":"2026-10-29","stars":18176},{"date":"2026-10-30","stars":18225},{"date":"2026-10-31","stars":18280},{"date":"2026-11-01","stars":18345},{"date":"2026-11-02","stars":18380},{"date":"2026-11-03","stars":18415},{"date":"2026-11-04","stars":18437},{"date":"2026-11-05","stars":18467},{"date":"2026-11-06","stars":18494},{"date":"2026-11-07","stars":18515},{"date":"2026-11-08","stars":18538},{"date":"2026-11-09","stars":18557},{"date":"2026-11-10","stars":18589}],"firstPublishedAt":"2026-10-27T01:10:12.677Z","commitSha":"26922f88b8b728875477bd8402f7aca28918bdef","verifiedAt":"2026-10-28T01:40:43.783Z","lastCommitAt":"2026-11-08T01:27:10Z","isMirror":false,"collectFailing":false,"installers":[{"os":"mac","file":"Compositor.dmg","sizeBytes":8867874,"url":"https://github.com/robbietilton/Compositor/releases/download/v1.6.6/Compositor.dmg","releaseUrl":"https://github.com/robbietilton/Compositor/releases/tag/v1.6.6"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A free Mac image editor with layers, masks, selections and filters, built to work like Photoshop.","summary":"Compositor is a free image editor for Mac, written to feel like Photoshop: layers and folders, layer masks, adjustment and filter layers, selections with Select Subject and content-aware fill, brush and clone tools, a camera raw panel and the same keyboard shortcuts. It opens JPEG, PNG, HEIC, TIFF, WebP, SVG, PDF, animated GIF, camera RAW and Photoshop PSD or PSB files, and exports PNG, JPEG, PDF and GIF. You download a signed 8 MB installer. It needs macOS 26 or later on an Apple silicon Mac, and it is English only.","easeRationale":"Grade A on Mac: every release ships a ready Compositor.dmg (8 MB) that the README says is signed and notarized and updates itself, and there is also a Homebrew cask. It runs nowhere else, as the author states: \"Compositor is macOS only\".","costNote":"Free and open source (MIT), with no account and no paid tier mentioned.","notVerified":["Whether the ready .dmg was built from the source code in this repository","How faithfully Photoshop files open, beyond the stated limits (8-bit RGB only, no CMYK, some vectors and vertical text become pixels)","How it behaves with very large documents and how much memory and disk it uses","Whether the pace of releases brings new problems, since the app updates itself"]},"zh-Hans":{"summaryShort":"一款免费的Mac图像编辑器,带图层、蒙版、选区和滤镜,使用方式仿照Photoshop。","summary":"Compositor 是一款免费的Mac图像编辑器,操作体验仿照Photoshop:支持图层和文件夹、图层蒙版、调整图层和滤镜图层、“选择主体”和内容感知填充的选区功能、画笔和仿制工具、相机原始格式(camera raw)面板,以及相同的键盘快捷键。它可以打开JPEG、PNG、HEIC、TIFF、WebP、SVG、PDF、动态GIF、相机RAW以及Photoshop的PSD或PSB文件,并能导出PNG、JPEG、PDF和GIF。你下载的是一个经过签名的8 MB安装包。需要搭载Apple芯片的Mac,且系统为macOS 26或更高版本,界面仅有英文。","easeRationale":"在Mac上的评级为A:每个发布版本都提供一个现成的Compositor.dmg(8 MB),README中说它经过签名并通过了notarization认证,并且会自动更新,还提供了Homebrew cask。它不能在其他任何系统上运行,正如作者所说:“Compositor仅支持macOS”。","costNote":"免费且开源(MIT许可),不需要账号,文档中也没有提到任何付费套餐。","notVerified":["这个现成的.dmg安装包是否确实是用本仓库里的源代码编译出来的","打开Photoshop文件的还原度如何,除了已说明的限制之外(仅支持8位RGB、不支持CMYK,部分矢量图形和竖排文字会被转成像素)","处理非常大的文档时表现如何,会占用多少内存和磁盘空间","既然这个App会自动更新,发布节奏是否会带来新的问题"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2NTIwNzg0","slug":"rpcs3__rpcs3","name":"RPCS3","owner":"RPCS3","fullName":"RPCS3/rpcs3","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"no_game_files_in_the_repository","severity":"attention","evidence":"所有者将该仓库描述为 \"PlayStation 3 emulator and debugger\"，顶层文件列表只有源代码、构建文件和配置文件（rpcs3/、3rdparty/、CMakeLists.txt、BUILDING.md 等），没有游戏。要完成可用的安装，README 把你指向别处：\"Check our friendly quickstart guide to make sure your computer meets the minimum system requirements to run RPCS3\"。PlayStation 3 游戏及主机软件由其所有者销售并受版权保护，各国对复制这些内容的规定各不相同。","source":"ai","title":"仓库中没有游戏文件"},{"code":"setup_instructions_live_outside_the_repository","severity":"info","evidence":"README 的 \"Running\" 部分只有三行，链接到 https://rpcs3.net/quickstart；仓库内唯一的说明是 BUILDING.md，内容是如何搭建构建环境。逐个游戏的信息则在另外的 wiki 上，地址是 https://wiki.rpcs3.net/。","source":"ai","title":"安装说明在仓库之外"},{"code":"game_compatibility_varies","severity":"info","evidence":"该项目请非程序员 \"test games and make bug reports\"，并把兼容性报告保存在论坛和 wiki 上；本次检查快照中有 1053 个未解决的 issue，因此某款具体游戏能否顺利运行要看游戏本身。","source":"ai","title":"游戏兼容性各不相同"},{"code":"high_hardware_requirements","severity":"info","evidence":"README 让你使用快速入门指南 \"to make sure your computer meets the minimum system requirements to run RPCS3\"，并保持显卡驱动更新；topics 中列出 \"vulkan\" 和 \"opengl\" 作为图形后端。","source":"ai","title":"硬件要求较高"}],"platforms":["windows","linux","mac"],"tools":[],"buildTools":[],"disk":{"minMB":7,"maxMB":20,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[{"slug":"playstation-plus-premium","name":"PlayStation Plus Premium","confirmed":true}],"declaredAlternatives":[],"stars":20388,"downloads":55721,"downloadsWeek":0,"downloadsDays":4,"growth":52,"growthDays":6,"history":[{"date":"2026-11-06","stars":20353},{"date":"2026-11-07","stars":20365},{"date":"2026-11-08","stars":20373},{"date":"2026-11-09","stars":20380},{"date":"2026-11-10","stars":20388}],"firstPublishedAt":"2026-11-09T23:08:44.841Z","commitSha":"8183f6e328f4d9903b29325978db59226e5e7177","verifiedAt":"2026-11-09T23:05:24.710Z","lastCommitAt":"2026-11-10T11:40:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs PlayStation 3 games on a Windows, Linux or Mac computer, with a debugger for studying the console.","summary":"An emulator that lets a normal computer run PlayStation 3 games, with a debugger for people who study how the console works. The README says it works on Windows, Linux, macOS and FreeBSD, and sends you to the quickstart guide on rpcs3.net to check that your computer is powerful enough. The program itself is small: the Windows download in the releases is 7-20 MB. Windows users also need the Visual C++ 2022 runtime and up-to-date graphics drivers.","easeRationale":"Windows is a B: the GitHub releases carry a ready Windows build (a 7-20 MB download), but before playing you still have to install the Visual C++ 2022 runtime, update your graphics driver and follow the quickstart guide on rpcs3.net. Linux and Mac are also a B: the README says the emulator runs there and points to the same quickstart guide for the download and setup, rather than to a build in this repository.","costNote":"No price, GPL-2.0 code. The README asks for voluntary support: \"Support the Lead Developers on Patreon\".","notVerified":["What else you have to provide before a game will start, since the README only points to the quickstart guide on rpcs3.net and the package does not say","How many games actually run, and at what speed, on your own computer","How well it works on a Mac and on FreeBSD, which the README mentions but the releases in this snapshot do not cover","How much disk space a real setup takes (the emulator download itself is 7-20 MB)","Whether the builds on rpcs3.net match the code in this repository"]},"zh-Hans":{"summaryShort":"在 Windows、Linux 或 Mac 电脑上运行 PlayStation 3 游戏，并带有调试器供研究主机使用。","summary":"这是一款模拟器，让普通电脑也能运行 PlayStation 3 游戏，并带有调试器，供研究主机运作原理的人使用。README 说它可在 Windows、Linux、macOS 和 FreeBSD 上运行，并引导你前往 rpcs3.net 上的快速入门指南，检查你的电脑性能是否足够。程序本身很小：releases 中的 Windows 下载包只有 7-20 MB。Windows 用户还需要 Visual C++ 2022 运行库和最新的显卡驱动。","easeRationale":"Windows 评级 B：GitHub releases 中有现成的 Windows 版本（7-20 MB 的下载包），但开始游戏前仍需安装 Visual C++ 2022 运行库、更新显卡驱动，并按照 rpcs3.net 上的快速入门指南操作。Linux 和 Mac 同样是 B：README 说模拟器在这些系统上也能运行，但下载和配置都指向同一份快速入门指南，而不是本仓库中的某个版本。","costNote":"免费，代码采用 GPL-2.0 许可。README 请求自愿支持：\"Support the Lead Developers on Patreon\"。","notVerified":["游戏启动前你还需要准备什么，因为 README 只指向 rpcs3.net 上的快速入门指南，软件包本身没有说明","在你自己的电脑上，实际能运行多少游戏、速度如何","在 Mac 和 FreeBSD 上的运行效果如何——README 提到了这两个系统，但本次抓取的 releases 并未覆盖","一次真实安装占用多少磁盘空间（模拟器下载包本身只有 7-20 MB）","rpcs3.net 上的构建版本是否与本仓库中的代码一致"]}}},{"id":"R_kgDOLrNzLw","slug":"rssnext__folo","name":"Folo","owner":"RSSNext","fullName":"RSSNext/Folo","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B"},"risk":"low","signals":[{"code":"cloud_account_service","severity":"info","evidence":"README.md：Folo \"keeps your subscriptions, read status and stars in sync across all of them\"，并提供 \"An inbox with its own email address for newsletters\"，这意味着你的数据保存在 Folo 的服务器上。","source":"ai","title":"云端账号服务"},{"code":"built_in_updater","severity":"info","evidence":"桌面版发布包含 latest-linux.yml 和 latest-mac.yml 更新文件，README 中的徽章从 https://ota.folo.is/versions 读取版本号。","source":"ai","title":"内置更新程序"},{"code":"license_exception_icons","severity":"info","evidence":"README.md：\"All content in the `icons/mgc` directory is copyrighted by https://mgc.mingcute.com/ and cannot be redistributed.\"","source":"ai","title":"图标适用许可例外"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["node"],"disk":{"minMB":158,"maxMB":475,"basis":"installer"},"apiKeys":["AI model provider of your choice (optional, to use your own key)"],"cost":{"model":"freemium"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["productivity"],"alternativeTo":[{"slug":"feedly","name":"Feedly","confirmed":true},{"slug":"inoreader","name":"Inoreader","confirmed":true}],"declaredAlternatives":[],"stars":39267,"downloads":31023963,"downloadsWeek":25240,"downloadsDays":7,"growth":47,"growthDays":7,"history":[{"date":"2026-10-26","stars":39175},{"date":"2026-10-27","stars":39176},{"date":"2026-10-28","stars":39186},{"date":"2026-10-29","stars":39195},{"date":"2026-10-30","stars":39201},{"date":"2026-10-31","stars":39205},{"date":"2026-11-01","stars":39207},{"date":"2026-11-02","stars":39215},{"date":"2026-11-03","stars":39220},{"date":"2026-11-04","stars":39225},{"date":"2026-11-05","stars":39235},{"date":"2026-11-06","stars":39241},{"date":"2026-11-07","stars":39244},{"date":"2026-11-08","stars":39246},{"date":"2026-11-09","stars":39254},{"date":"2026-11-10","stars":39267}],"firstPublishedAt":"2026-10-27T01:10:12.678Z","commitSha":"68edee907003a90877f2ffa6bd9ed9bedababfd4","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T16:50:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Follow blogs, news, newsletters, podcasts and videos in one timeline, with AI summaries and translations.","summary":"Folo is a news and feed reader. You subscribe to websites, newsletters, podcasts, video channels and social feeds, and read them all in one place. AI can summarize articles, translate them, answer questions about what you follow and write daily briefings. Your subscriptions sync between the website, phone apps and the desktop apps for Windows, Mac and Linux. It works through Folo's online service, so your feeds and reading history are kept there. You can use its AI models or bring your own API key.","easeRationale":"Windows has a code-signed .exe installer and a Microsoft Store listing, and there is a web version; the extra step is signing in so your feeds sync. Node is only needed to build it from source.","costNote":"The README does not list prices. AI features use Folo's hosted models or your own API key.","notVerified":["Which features are free and which need a paid plan.","Whether you can use the desktop app without creating an account.","What reading data Folo's servers keep and for how long.","Real API cost if you connect your own AI key."]},"zh-Hans":{"summaryShort":"在同一条时间线上关注博客、新闻、简报邮件、播客和视频，并提供 AI 摘要和翻译。","summary":"Folo 是一款新闻和信息源阅读器。你可以订阅网站、邮件简报、播客、视频频道和社交媒体信息流，并在同一个地方阅读。AI 可以总结文章、翻译文章、回答关于你所关注内容的问题，并撰写每日简报。你的订阅会在网页端、手机应用以及 Windows、Mac、Linux 桌面应用之间同步。它通过 Folo 的在线服务运行，因此你的订阅源和阅读记录都保存在那里。你可以使用它的 AI 模型，也可以使用自己的 API 密钥。","easeRationale":"Windows 版有带代码签名的 .exe 安装程序，并在 Microsoft Store 上架，另有网页版；多出的一步是登录账号，以便同步你的订阅源。只有从源码构建时才需要 Node。","costNote":"README 没有列出价格。AI 功能使用 Folo 托管的模型或你自己的 API 密钥。","notVerified":["哪些功能免费，哪些需要付费套餐。","不创建账号能否使用桌面应用。","Folo 的服务器保存哪些阅读数据，保存多久。","如果接入你自己的 AI 密钥，API 的实际费用是多少。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxODM0ODM5ODg=","slug":"ruffle-rs__ruffle","name":"Ruffle","owner":"ruffle-rs","fullName":"ruffle-rs/ruffle","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A","web":"A"},"risk":"low","signals":[{"code":"license_files_say_mit_or_apache","severity":"info","evidence":"LICENSE.md 写道：\"Ruffle is licensed under either of - Apache License, Version 2.0 ... - MIT license ... at your option\"，Cargo.toml 中 license = \"MIT OR Apache-2.0\"，而 GitHub 显示该仓库没有单一许可证（NOASSERTION）。README 还补充说 \"Ruffle depends on third-party libraries under compatible licenses\"，如果许可证对你很重要，请查看文件头部信息。","source":"ai","title":"许可证文件写的是 mit 或 apache"},{"code":"daily_builds_instead_of_stable_versions","severity":"info","evidence":"最新的三个 release 标签分别是 nightly-2026-11-08、nightly-2026-11-07 和 nightly-2026-11-06，共有 1200 个 release，README 链接到 \"Nightly builds\"。Cargo.toml 中的版本号仍是 0.7.1。","source":"ai","title":"每日构建版本而非稳定版本"},{"code":"flash_support_is_incomplete","severity":"info","evidence":"README 的 Project status 写道：\"Ruffle supports ActionScript 1, 2 and 3 pretty well, but it's still not finished by any means. Please report any issues in the Issue Tracker.\" 本次检查快照中有 5757 个未解决的 issue。","source":"ai","title":"flash 支持尚不完整"},{"code":"browser_version_installed_from_stores","severity":"info","evidence":"README 中的徽章把浏览器版本指向 addons.mozilla.org、Chrome 网上应用店、Flathub、npm 和 Arch user repository，因此大多数人安装的是在别处构建的软件包，而不是本仓库中的文件。","source":"ai","title":"浏览器版本来自应用商店安装"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["rust"],"disk":{"minMB":21,"maxMB":62,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"file"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":18846,"downloads":19447299,"downloadsWeek":83143,"downloadsDays":4,"growth":40,"growthDays":6,"history":[{"date":"2026-11-06","stars":18826},{"date":"2026-11-07","stars":18829},{"date":"2026-11-08","stars":18835},{"date":"2026-11-09","stars":18841},{"date":"2026-11-10","stars":18846}],"firstPublishedAt":"2026-11-09T23:02:44.358Z","commitSha":"c8a03ccd335e30031fa51f698d55c5402d450578","verifiedAt":"2026-11-09T22:51:11.337Z","lastCommitAt":"2026-11-10T19:33:34Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Opens old Flash games and animations on Windows, Mac, Linux or in the browser, after Flash Player was retired.","summary":"A replacement for Adobe Flash Player, written from scratch in Rust. You download the small program for Windows, Mac or Linux and open a .swf file, or use the web version and the Firefox and Chrome extensions so Flash content embedded in old pages plays again. There is also a demo page where you pick a file and it plays in the browser. The README says Flash support is good but \"still not finished by any means\", and the project publishes a new build almost every day instead of stable versions.","easeRationale":"An A on every platform: each daily release has ready programs to unpack and open — Windows (20-21 MB zip), a universal Mac build (32 MB) and Linux builds for 64-bit, 32-bit and ARM — plus a web version, extensions in the Firefox and Chrome stores and a Flathub package. On the web it is also an A: the demo page plays a file you choose with nothing installed. Compiling from source would need Rust and Java, but no end user has to do that.","costNote":"No price. The project is paid for by sponsors (GitHub Sponsors, with sites such as Newgrounds, Kongregate and Cool Math Games listed as diamond sponsors).","notVerified":["Whether a specific Flash game or animation of yours plays correctly and at full speed","Whether the Mac and Linux builds open without extra security prompts, since they are plain archives","How the daily builds behave: a build that worked yesterday may change today","Whether the browser extensions in the Firefox and Chrome stores match this code","What the extensions are allowed to see in the pages you visit"]},"zh-Hans":{"summaryShort":"在 Flash Player 停止维护后，仍可在 Windows、Mac、Linux 或浏览器中打开旧的 Flash 游戏和动画。","summary":"一个用 Rust 从零编写的 Adobe Flash Player 替代品。你可以下载 Windows、Mac 或 Linux 的小程序并打开 .swf 文件，也可以使用网页版，或安装 Firefox 和 Chrome 扩展，让旧网页中嵌入的 Flash 内容重新播放。还有一个演示页面，你选择一个文件，它就会在浏览器中播放。README 说 Flash 支持已经不错，但 \"still not finished by any means\"，而且项目几乎每天都会发布新的构建版本，而不是稳定版本。","easeRationale":"在所有平台上都是 A：每个每日发布版都有现成的程序可以解压打开——Windows（20-21 MB 的 zip）、通用的 Mac 版本（32 MB），以及 64 位、32 位和 ARM 的 Linux 版本——此外还有网页版、Firefox 和 Chrome 应用商店中的扩展，以及 Flathub 软件包。网页端同样是 A：演示页面可以直接播放你选择的文件，什么都不用安装。从源代码编译需要 Rust 和 Java，但普通用户不需要这样做。","costNote":"免费。项目由赞助商资助（GitHub Sponsors，Newgrounds、Kongregate 和 Cool Math Games 等网站被列为钻石级赞助商）。","notVerified":["你某个具体的 Flash 游戏或动画是否能正确播放、速度是否正常","Mac 和 Linux 版本打开时是否会弹出额外的系统警告，因为它们只是普通压缩包","每日构建版本的表现如何：昨天还能用的版本今天可能就变了","Firefox 和 Chrome 商店中的浏览器扩展是否与此代码一致","这些扩展被允许在你访问的网页中看到什么内容"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyOTkzNTQyMDc=","slug":"rustdesk__rustdesk","name":"RustDesk","owner":"rustdesk","fullName":"rustdesk/rustdesk","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道传给 shell：curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh","source":"check","title":"curl 管道传输到 shell"},{"code":"can_be_used_to_take_over_another_persons_computer","severity":"attention","evidence":"README 开头就有一份 \"Misuse Disclaimer\"：\"The developers of RustDesk do not condone or support any unethical or illegal use of this software. Misuse, such as unauthorized access, control or invasion of privacy, is strictly against our guidelines.\" 任何拿到一台机器的编号和密码的人都可以操控它。","source":"ai","title":"可被用来控制他人的电脑"},{"code":"unattended_access_and_background_service","severity":"attention","evidence":"该程序会安装一个系统服务，并可以在没有人坐在机器前的情况下接受连接；发行版中甚至有一个名为 \"rustdesk-unattended-wayland-1.5.1-x86_64.deb\" 的构建版本，Cargo.toml 在 Windows 相关部分列出了 windows-service、impersonate-system 和 virtual_display。","source":"ai","title":"无人值守访问和后台服务"},{"code":"connections_pass_through_the_projects_servers_by_default","severity":"attention","evidence":"README：\"You can use our rendezvous/relay server, set up your own, or write your own rendezvous/relay server\"。在你搭建自己的服务器之前，配对和中继都由该公司运营的机器完成。同一段中声称你拥有 \"full control of your data, with no concerns about security\"，这是项目自己的说法。","source":"ai","title":"连接默认经过项目自己的服务器"},{"code":"many_parts_come_from_the_teams_own_forks","severity":"info","evidence":"Cargo.toml 直接从 github.com/rustdesk-org 下的分支（rdev、arboard、cpal、kcp-sys、wezterm、clipboard-master 等）拉取了大约十几个依赖，构建说明还会从 raw githubusercontent.com 的一个直接链接下载 Sciter 界面库。","source":"ai","title":"许多部分来自团队自己的分支"},{"code":"nightly_tag_mixes_versions","severity":"info","evidence":"发行版列表显示一个日期为 2026-07-10 的 \"nightly\" 标签同时携带了标注为 1.4.9、1.5.0 和 1.5.1 的文件；稳定下载版本是那些带编号的发行版。","source":"ai","title":"nightly 标签混合了不同版本"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","rust"],"disk":{"minMB":25,"maxMB":76,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system","productivity"],"alternativeTo":[{"slug":"teamviewer","name":"TeamViewer","confirmed":true},{"slug":"anydesk","name":"AnyDesk","confirmed":true},{"slug":"parsec","name":"Parsec","confirmed":true}],"declaredAlternatives":[],"stars":128916,"downloads":82082334,"downloadsWeek":1541385,"downloadsDays":7,"growth":850,"growthDays":7,"history":[{"date":"2026-10-26","stars":127172},{"date":"2026-10-27","stars":127296},{"date":"2026-10-28","stars":127377},{"date":"2026-10-29","stars":127507},{"date":"2026-10-30","stars":127627},{"date":"2026-10-31","stars":127733},{"date":"2026-11-01","stars":127812},{"date":"2026-11-02","stars":127942},{"date":"2026-11-03","stars":128066},{"date":"2026-11-04","stars":128196},{"date":"2026-11-05","stars":128314},{"date":"2026-11-06","stars":128455},{"date":"2026-11-07","stars":128557},{"date":"2026-11-08","stars":128665},{"date":"2026-11-09","stars":128786},{"date":"2026-11-10","stars":128916}],"firstPublishedAt":"2026-10-27T01:10:12.679Z","commitSha":"beeeef14f0a61bd622aacf042b4396bdeaed9b61","verifiedAt":"2026-11-03T02:42:05.631Z","lastCommitAt":"2026-11-09T08:04:24Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"rustdesk-1.5.1-x86_64.exe","sizeBytes":26211184,"url":"https://github.com/rustdesk/rustdesk/releases/download/1.5.1/rustdesk-1.5.1-x86_64.exe","releaseUrl":"https://github.com/rustdesk/rustdesk/releases/tag/1.5.1"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"See and use another computer's screen over the internet, with files and sound, from your own machine.","summary":"RustDesk lets you see and control another computer over the internet: you open the program on both machines, type the number and password shown on the other one, and its screen appears on yours. It also copies files, carries sound and can tunnel connections. There are installers for Windows, macOS and Linux and apps for phones. Connections are matched by servers run by the project, or by a server you set up yourself. It is free software under AGPL-3.0.","easeRationale":"On Windows you download the .exe or .msi, open it, and the connection number and password are already on screen (the README says it \"works out of the box with no configuration required\"). macOS is B because the system asks you to allow screen recording and control by hand, and Linux is B because the right package depends on your system (.deb, .rpm, Flatpak or AppImage) and Wayland desktops have a separate build.","costNote":"The program and the project's matching servers cost nothing. The README links \"RustDesk Server Pro - Advanced Features\" to rustdesk.com/pricing.html, a paid version for people who run their own server.","notVerified":["Whether the released installers were built from the code published here.","What the project's matching and relay servers record about a connection.","How well the picture and sound hold up on a slow link or across countries.","Whether the Wayland and macOS permission paths work as described on your own system.","Whether using it to reach a work computer is allowed where you are, which depends on your employer's rules."]},"zh-Hans":{"summaryShort":"通过互联网查看并使用另一台电脑的屏幕，支持文件传输和声音，就在你自己的机器上操作。","summary":"RustDesk 让你通过互联网查看并控制另一台电脑：你在两台机器上都打开这个程序，输入对方屏幕上显示的编号和密码，对方的屏幕就会出现在你这边。它还能传输文件、传递声音，并可以建立连接隧道。提供 Windows、macOS 和 Linux 的安装程序，以及手机应用。连接的配对由项目运营的服务器完成，或者由你自己搭建的服务器完成。它是 AGPL-3.0 许可下的自由软件。","easeRationale":"在 Windows 上，你下载 .exe 或 .msi 并打开，连接编号和密码已经显示在屏幕上（README 说它 \"works out of the box with no configuration required\"）。macOS 评级为 B，因为系统需要你手动允许屏幕录制和控制；Linux 也是 B，因为合适的包取决于你的系统（.deb、.rpm、Flatpak 或 AppImage），Wayland 桌面还需要单独的版本。","costNote":"程序本身和项目运营的配对服务器都是免费的。README 把 \"RustDesk Server Pro - Advanced Features\" 链接到 rustdesk.com/pricing.html，这是为自建服务器的用户提供的付费版本。","notVerified":["发布的安装程序是否是从这里公开的代码构建的。","项目的配对和中继服务器对一次连接会记录什么内容。","在慢速网络或跨国连接下，画面和声音能保持多好的质量。","Wayland 和 macOS 的权限流程在你自己的系统上是否真的像描述那样工作。","用它访问工作用电脑在你所在地是否被允许，这取决于你雇主的规定。"]}}},{"id":"R_kgDOO07yiQ","slug":"ruvnet__ruflo","name":"Ruflo","owner":"ruvnet","fullName":"ruvnet/ruflo","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接通过管道传给 shell 执行：curl -fsSL https://cdn.jsdelivr.net/gh/ruvnet/ruflo@main/scripts/install.sh | bash","source":"check","title":"curl 直接管道到 shell 执行"},{"code":"known_vulnerabilities","severity":"attention","evidence":"被固定版本的依赖项存在已知漏洞（OSV）：zod: GHSA-mrhx-c7xc-m3w5","source":"check","title":"已知漏洞"},{"code":"plugins_run_unsandboxed_with_your_permissions","severity":"attention","evidence":"README.md 中两次提到：\"Mods run with your account's permissions, so review their code before installing.\" 以及 \"mods run with your account's permissions and are not sandboxed, so read the code first\"。安装路径是先执行 `/plugin marketplace add ruvnet/ruflo`，再对 48 个插件逐一执行 `/plugin install ...`，另外还有 3 个来自其他仓库的外部插件。","source":"ai","title":"插件在无沙箱环境下以你的权限运行"},{"code":"assistant_can_drive_the_tool_itself","severity":"info","evidence":"README.md 写道：\"Let Claude drive. You set the limits. Enable Settings → Claude control to navigate and run console actions. Choose `read`, `write`, `manage` or `full`, with `ask` or `auto` approval. The current default is read + ask; raising the level is your choice.\" 在更高权限等级下，AI 会自行创建任务并执行控制台操作。","source":"ai","title":"助手可以自行操作该工具"},{"code":"agents_can_talk_to_other_machines","severity":"info","evidence":"联邦功能会把你的智能体连接到其他地方的智能体：`federation join wss://team-b.example.com:8443`，被描述为 \"shared workspaces across trust boundaries\"。README 说 \"PII is stripped automatically before it leaves\"，并列出了 mTLS 和 ed25519 身份验证，但这仍然是离开你电脑的网络流量。","source":"ai","title":"智能体可以与其他机器通信"},{"code":"promotional_numbers_in_the_readme","severity":"info","evidence":"README 中有一个下载徽章，其自身说明写着 \"Illustrative download counter starts at 98,000,000 and adds varied amounts each second in a 30 second loop\"，还有诸如 \"ruflo wins cold start, single turn, RSS by 1.3×–1953×\"、\"Intelligent routing (89% accuracy)\" 和 \"The original agent harness\" 这样的说法。还有几处说明文字标注这些动画使用的是 \"sample data\"，并且 \"not evidence of completed swarm work or memory recall\"。","source":"ai","title":"readme 中的宣传性数字"},{"code":"very_large_surface_for_one_tool","severity":"info","evidence":"README.md 写道：\"You don't need to learn 406 MCP tools or 61 CLI commands\"，另外还有 \"100+ Agents\"、48 个插件、\"12 auto-triggered workers\" 和 27 个 hooks。共有 1200 个 release，v3.74.1 和 v3.73.0 都在 2026-11-09 同一天发布，还有 618 个未解决的 issue。","source":"ai","title":"单一工具的功能面极其庞大"},{"code":"package_identity_is_mixed","severity":"info","evidence":"package.json 中仍然写着 \"name\": \"claude-flow\" 和 \"repository\": \"git+https://github.com/ruvnet/claude-flow.git\"，而 README 却用 `npx ruflo@latest` 安装；示例中两者混用（`npx claude-flow@latest federation init`）。安装过程还会拉取同一作者的许多软件包，包括 `agentdb`、`ruvector`、`agentic-flow` 和 `better-sqlite3` 等可选项。","source":"ai","title":"软件包身份混乱"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","rust"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Anthropic Claude (through Claude Code or an API key)","Other model providers: OpenAI/Codex, Gemini, Cohere","Cognitum Music account (ruflo-music plugin)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","ai_agents","development"],"alternativeTo":[{"slug":"mem0","name":"Mem0","confirmed":true},{"slug":"zep","name":"Zep","confirmed":true},{"slug":"langsmith","name":"LangSmith","confirmed":true}],"declaredAlternatives":[],"stars":81428,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":630,"growthDays":7,"history":[{"date":"2026-11-04","stars":80870},{"date":"2026-11-05","stars":80989},{"date":"2026-11-06","stars":81074},{"date":"2026-11-07","stars":81163},{"date":"2026-11-08","stars":81262},{"date":"2026-11-09","stars":81350},{"date":"2026-11-10","stars":81428}],"firstPublishedAt":"2026-11-09T23:02:44.367Z","commitSha":"f6fe4199f7663cf31446d4c8d7f15b376a996655","verifiedAt":"2026-11-09T22:52:10.284Z","lastCommitAt":"2026-11-10T22:16:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds teams of AI agents, shared memory, plugins and a console to Claude Code or Codex with one command.","summary":"A large add-on for AI coding assistants, formerly called Claude Flow. One command, `npx ruflo@latest init`, puts agents, hooks, memory and an MCP server into your project, and Claude Code or Codex then uses them: work is split between many agents, results are stored in a vector database, and a console shows runs, tokens and cost. There are 48 plugins for memory, security, testing, trading and more, and a network feature that lets agents on different machines exchange work. You need Node 20 or newer and a paid AI assistant or model key.","easeRationale":"C on all three systems: you need Node 20 or newer, then `npx ruflo@latest init` (or the wizard) in your project folder and a restart of Claude Code; the lighter route is typing `/plugin marketplace add ruvnet/ruflo` inside Claude Code 2.1.287 or later. Mac and Linux also have the one-line `curl ... | bash` installer, which the README says needs a POSIX shell on Windows (Git-Bash, WSL or MSYS), so Windows users take the `npx` route.","costNote":"MIT code with no price, but it is an add-on for Claude Code, Codex or another model provider, which you pay for; the README's own table lists \"Claude, GPT, Gemini, Cohere, Ollama\" as providers and the project ships a cost tracker for token spending.","notVerified":["Whether the agent teams, the learning and the memory actually improve results on your own work","The benchmark and accuracy numbers in the README, which come from the author","How many tokens, and how much money, a swarm run really costs","What the 48 plugins and the external plugins do once installed, since they run with your own permissions","What the federation feature sends over the network and whether personal data is really removed first"]},"zh-Hans":{"summaryShort":"只需一条命令，就能为 Claude Code 或 Codex 添加 AI 智能体团队、共享内存、插件和控制台。","summary":"这是一个面向 AI 编程助手的大型扩展，曾叫作 Claude Flow。只需一条命令 `npx ruflo@latest init`，就能把智能体、hooks、内存和一个 MCP 服务器加入你的项目，之后 Claude Code 或 Codex 就会使用它们：任务会拆分给多个智能体，结果保存在向量数据库中，控制台会显示运行记录、token 用量和费用。项目提供 48 个插件，涉及内存、防护、测试、交易等领域，还有一个网络功能，让不同机器上的智能体互相交换工作。你需要 Node 20 或更新版本，以及一个付费的 AI 助手或模型密钥。","easeRationale":"三大系统评级均为 C：你需要 Node 20 或更新版本，然后在项目文件夹里运行 `npx ruflo@latest init`（或使用向导），并重启 Claude Code；更轻量的方式是在 Claude Code 2.1.287 或更高版本中输入 `/plugin marketplace add ruvnet/ruflo`。Mac 和 Linux 还可以使用一行命令 `curl ... | bash` 安装，README 说在 Windows 上需要 POSIX shell（Git-Bash、WSL 或 MSYS），所以 Windows 用户走 `npx` 这条路。","costNote":"代码遵循 MIT 许可，本身免费，但它是 Claude Code、Codex 或其他模型提供方的扩展，而这些是你要另外付费的；README 自带的表格列出了 \"Claude, GPT, Gemini, Cohere, Ollama\" 等提供方，项目还自带一个用于跟踪 token 花费的费用追踪器。","notVerified":["智能体团队、学习机制和内存是否真的能提升你自己工作的结果","README 中的基准测试和准确率数字，这些数字来自作者本人","一次 swarm 运行实际会消耗多少 token、花费多少钱","48 个插件以及外部插件安装后会做什么，因为它们是以你自己的权限运行的","联邦功能通过网络发送了什么内容，以及个人数据是否真的会先被移除"]}}},{"id":"R_kgDOO3hF2A","slug":"ruvnet__ruview","name":"RuView","owner":"ruvnet","fullName":"ruvnet/RuView","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把远程脚本通过管道交给 shell 执行：curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh","source":"check","title":"用 curl 管道把脚本直接交给 shell 执行"},{"code":"health_and_safety_readings_are_not_validated","severity":"attention","evidence":"项目自己的 MetaHarness 防护清单写着 \"Device-reported vitals are labelled as unvalidated\"，但模块目录里却提供 `sleep-apnea`、`seizure-detect`、`cardiac-arrhythmia` 和 `respiratory-distress`；`smoke-fire` 模块声明它 \"Not a UL-listed replacement for code-required smoke alarms.\" 这里的任何东西都不是医疗设备，也不是保障生命的设备。","source":"ai","title":"健康与防护类读数未经验证"},{"code":"advertised_pose_feature_is_not_wired_up","severity":"attention","evidence":"README 的 \"Model weights: what's real, what's not\" 部分：仓库中提交的设备端姿态模型得分为 \"PCK@20 = 3.0%\"，低于项目自己设定的 ≥35% 目标，并且 \"Its runtime path in `cog-pose-estimation/src/inference.rs` is still a centred-skeleton stub returning `confidence=0` — the weights are not yet wired in.\" 已发布的编码器文件也无法加载（\"the published file's header is NUL-padded, which the reference `safetensors.torch.load_file` rejects\"），此前的 \"100% presence\" 数字也已被撤回。","source":"ai","title":"宣传的姿态识别功能尚未接入"},{"code":"watching_people_who_have_not_agreed_to_it","severity":"attention","evidence":"README：\"See through walls with WiFi ... Detect people, measure breathing and heart rate ... with no cameras or wearables\"、\"no app on the user's phone\"，以及 \"uses your neighbors' routers as free radar illuminators\" 的多信道扫描。监控他人，或借用他人的网络，在许多国家受隐私法限制；在把它对准某个房间之前，请先查清你所在地的规定。","source":"ai","title":"监视没有同意的人"},{"code":"example_configuration_ships_wide_open","severity":"attention","evidence":"`example.env` 设置了 `HOST=0.0.0.0`、`ALLOWED_HOSTS=*`、`CORS_ORIGINS=*`、`DEBUG=true`、`ENABLE_AUTHENTICATION=false`、`ENABLE_TEST_ENDPOINTS=true` 和 `SECRET_KEY=your-secret-key-here-change-for-production`。该文件列出了生产环境需要修改的十项内容，但原样复制的服务器在你的网络里完全没有登录验证。","source":"ai","title":"示例配置完全敞开"},{"code":"very_large_claim_surface_and_release_churn","severity":"info","evidence":"GitHub 上有 367 个发布版本（其中三个标记于 2026-10-26）、504 个未解决的 issue，README 还提供一个 \"105-cog catalog\"，从生命体征到 `ghost-hunter` 和 `weapon-detect` 应有尽有。项目自己维护着说明哪些内容经过测量、哪些没有的诚实度表格，这也说明它宣称的内容有多少。","source":"ai","title":"宣称的内容极多，版本更替频繁"}],"platforms":["windows","mac","linux"],"tools":["docker","python","other"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":6,"maxMB":17,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","automation","network_system"],"alternativeTo":[{"slug":"origin-wireless","name":"Origin Wireless","confirmed":true}],"declaredAlternatives":[],"stars":103862,"downloads":112739,"downloadsWeek":0,"downloadsDays":7,"growth":1612,"growthDays":7,"history":[{"date":"2026-10-29","stars":101576},{"date":"2026-10-30","stars":101767},{"date":"2026-10-31","stars":101882},{"date":"2026-11-01","stars":102015},{"date":"2026-11-02","stars":102114},{"date":"2026-11-03","stars":102250},{"date":"2026-11-04","stars":102393},{"date":"2026-11-05","stars":102549},{"date":"2026-11-06","stars":102714},{"date":"2026-11-07","stars":102847},{"date":"2026-11-08","stars":103048},{"date":"2026-11-09","stars":103374},{"date":"2026-11-10","stars":103862}],"firstPublishedAt":"2026-10-29T01:54:56.463Z","commitSha":"3686bc770f77eb6161d5379861dcd09d7d9a1e4b","verifiedAt":"2026-10-29T01:45:20.082Z","lastCommitAt":"2026-11-10T22:37:33Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Uses cheap ESP32 boards and ordinary WiFi signals to sense people, movement and breathing in a room, with no camera.","summary":"A WiFi sensing platform. Small ESP32 boards (about $9 each) read how radio waves bounce around a room, and the software turns that into presence, movement, breathing and heart-rate readings, plus alerts like falls, and publishes them to Home Assistant, Apple Home, Google Home or Alexa. There is a Docker image with simulated data to try it without hardware. For tinkerers: you buy and flash boards, run a server and calibrate. Free and MIT-licensed; the project itself labels several of its models and vital-sign readings as unvalidated.","easeRationale":"D on Windows, Mac and Linux alike: the quickest look is a Docker image fed with simulated data (and it refuses to start without a token you generate yourself), while anything real means buying ESP32 boards, flashing firmware with esptool or ESP-IDF, provisioning WiFi, running a server and calibrating the room. A laptop with no CSI-capable board only gets coarse presence from signal strength.","costNote":"The software is MIT and runs locally, with no cloud account. You pay for hardware: the README quotes about $9 for one ESP32-S3 node, around $54 for a mesh of 3-6 nodes, and about $140 for a full kit with a Cognitum Seed.","notVerified":["Whether presence, breathing and heart rate work in a normal home, outside the author's measurements.","Whether through-wall sensing reaches the \"up to ~5 m\" the README mentions in your own walls.","Whether firmware flashing and WiFi provisioning on an ESP32-S3 or C6 go as described.","Real disk and memory use once models and the Python stack are installed.","Whether the signed edge modules do what their one-line descriptions say, and what they send to `seed.cognitum.one`.","Whether using it on a given room, or on neighbouring networks, is legal where you live."]},"zh-Hans":{"summaryShort":"用廉价的ESP32开发板和普通WiFi信号感知房间里的人、动作和呼吸,无需摄像头。","summary":"一个WiFi感知平台。小型ESP32开发板(约9美元一块)读取无线电波在房间里反射的方式,软件把这些数据转换成存在感、动作、呼吸和心率读数,以及摔倒等警报,并发布到Home Assistant、Apple Home、Google Home或Alexa。有一个带模拟数据的Docker镜像,可以不用硬件就先试试。适合喜欢动手的人:你需要购买并刷写开发板、运行服务器并进行校准。免费且采用MIT许可;项目本身也标明其中几个模型和生命体征读数尚未经过验证。","easeRationale":"D,在Windows、Mac和Linux上情况相同:最快的体验方式是用模拟数据驱动的Docker镜像(而且如果没有你自己生成的令牌,它会拒绝启动),而要做真正的部署就需要购买ESP32开发板、用esptool或ESP-IDF刷写固件、配置WiFi、运行服务器并对房间进行校准。没有支持CSI的板子的笔记本电脑,只能通过信号强度获得粗略的存在感检测。","costNote":"该软件采用MIT许可,在本地运行,无需云端账户。你需要为硬件付费:README报价一个ESP32-S3节点约9美元,一个3到6节点的网状网络约54美元,一套包含Cognitum Seed的完整套件约140美元。","notVerified":["存在感、呼吸和心率检测在普通家庭环境中(而不是作者自己的测量条件下)是否有效。","穿墙感知在你自己的墙壁条件下,是否能达到README中提到的\"最多约5米\"。","在ESP32-S3或C6上刷写固件和配置WiFi是否真的像描述那样顺利。","安装好模型和Python技术栈后,实际占用的磁盘和内存空间。","签名的边缘模块是否真的像一行简介所说的那样运作,以及它们向`seed.cognitum.one`发送了什么数据。","在你所在地区,在某个房间使用它,或者对邻居的网络使用它,是否合法。"]}}},{"id":"R_kgDOLEzKcw","slug":"rvc-boss__gpt-sovits","name":"GPT-SoVITS","owner":"RVC-Boss","fullName":"RVC-Boss/GPT-SoVITS","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"B","mac":"D","linux":"D","web":"A"},"risk":"attention","signals":[{"code":"voice_cloning_from_a_short_sample","severity":"attention","evidence":"README 的 Features 部分写道：\"Zero-shot TTS: Input a 5-second vocal sample and experience instant text-to-speech conversion\" 以及 \"Few-shot TTS: Fine-tune the model with just 1 minute of training data\"。README 中没有任何地方标明生成结果是机器制作的，而未经许可复制真人声音在不少国家受到法律限制。","source":"ai","title":"仅凭一小段样本即可克隆声音"},{"code":"models_and_images_come_from_personal_accounts","severity":"info","evidence":"install.sh 和 install.ps1 从 \"https://huggingface.co/XXXXRT/GPT-SoVITS-Pretrained/...\"（或 hf-mirror.com / modelscope.cn）下载模型包，Windows 整合包来自 \"huggingface.co/lj1995/GPT-SoVITS-windows-package\"，docker-compose.yaml 则从 Docker Hub 拉取 \"xxxxrt666/gpt-sovits:latest-cu126\" 等类似镜像——这些都是个人账户，而不是 GitHub 的 releases。","source":"ai","title":"模型和镜像来自个人账户"},{"code":"docker_setup_shares_the_folder_and_opens_ports","severity":"info","evidence":"docker-compose.yaml 挂载了整个项目文件夹（`- .:/workspace/GPT-SoVITS`），并开放了五个端口（9871-9874 以及作为 API 的 9880）。README 警告说：\"Docker Compose will mount all files in the current directory\"，并要求 `shm_size: 16g` 的共享内存。","source":"ai","title":"docker 配置会共享文件夹并开放端口"},{"code":"many_versions_with_manual_model_downloads","severity":"info","evidence":"README 为 v2、v3、v4、v2Pro 和 v5 分别提供了发布说明和模型下载，每个版本都要求你从 Hugging Face 把文件取到 `GPT_SoVITS/pretrained_models` 中；V5 甚至要求从另一个分支（`cuda_graph_accel_v5`）\"Clone the latest code\"。最新打标签的 release 是 20250606v2pro，而代码却每天都在变化。","source":"ai","title":"版本繁多且需手动下载模型"},{"code":"mac_results_are_worse","severity":"info","evidence":"README 的 macOS 部分写道：\"The models trained with GPUs on Macs result in significantly lower quality compared to those trained on other devices, so we are temporarily using CPUs instead.\" 实测速度表显示，M4 处理器上的 RTF 为 0.526，而 4090 显卡上只有 0.014。","source":"ai","title":"mac 上的效果较差"},{"code":"main_guides_are_outside_the_repository","severity":"info","evidence":"用户指南链接指向 yuque.com（中文）和 rentry.co（英文），而不是仓库内的文档；中国用户会被引导到一个语雀页面去下载模型和 Windows 整合包。","source":"ai","title":"主要指南在仓库之外"}],"platforms":["windows","mac","linux","web"],"tools":["python","docker","gpu_nvidia","ffmpeg"],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":450,"maxMB":1250,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true}],"declaredAlternatives":[],"stars":64720,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":0,"growthDays":7,"history":[{"date":"2026-11-04","stars":64847},{"date":"2026-11-05","stars":64943},{"date":"2026-11-06","stars":64404},{"date":"2026-11-07","stars":64507},{"date":"2026-11-08","stars":64567},{"date":"2026-11-09","stars":64647},{"date":"2026-11-10","stars":64720}],"firstPublishedAt":"2026-11-09T23:02:44.378Z","commitSha":"59c44686f5cb9ffca9d8f1dec8b9c6836a129754","verifiedAt":"2026-11-09T22:53:06.618Z","lastCommitAt":"2026-11-09T13:59:04Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from a 5-second clip and reads your text aloud in it, in Chinese, English, Japanese, Korean or Cantonese.","summary":"A voice cloning and text-to-speech workshop with a page in your browser. You give it a 5-second sample for an instant copy of a voice, or about one minute of audio to train a better one, and it then reads your text in that voice, including in a language the sample was not in. The page also helps you prepare the training audio: separating voice from music, cutting clips and writing out what is said. Windows users can download a ready package and double-click `go-webui.bat`; otherwise it is conda, Python and an Nvidia card, or Docker. There is a free online demo.","easeRationale":"The web is an A: the README links a free demo page where you type and listen, plus Colab notebooks for training. Windows is a B: you download the \"integrated package\" (a 7z file from Hugging Face), unpack it and double-click `go-webui.bat`. Linux and Mac are a D: you create a conda environment and run `bash install.sh --device <CU126|CU128|ROCM|MPS|CPU> --source <HF|HF-Mirror|ModelScope>`, which installs ffmpeg, cmake and a compiler and downloads several model packs; the full feature set assumes an Nvidia card, and Docker images need an Nvidia runtime.","costNote":"No price: MIT code, trained models published openly on Hugging Face and ModelScope, and a free online demo on Hugging Face.","notVerified":["How good the cloned voice really sounds from 5 seconds or from one minute of audio","Real disk space after the pretrained models, the speech-recognition models and the voice-separation models download","Whether the Windows \"integrated package\" (dated 20250228) matches the current code and the newer versions","Whether it runs usefully without an Nvidia card, and how slow it is","Whether the free online demo stays available","What is inside the third-party model packs and Docker images that the installer downloads"]},"zh-Hans":{"summaryShort":"只需 5 秒的语音片段即可复制一个声音，并用它朗读你的文字，支持中文、英语、日语、韩语或粤语。","summary":"一个语音克隆与文本转语音的工作台，有一个浏览器页面。你给它一个 5 秒的样本即可立即克隆一个声音，或者用大约一分钟的音频训练出更好的效果，之后它就能用这个声音朗读你的文字，甚至可以朗读样本中没有出现过的语言。该页面还能帮你准备训练音频：把人声从音乐中分离、剪辑片段、并写出其中说的内容。Windows 用户可以下载现成的整合包，双击 `go-webui.bat` 即可；其他情况则需要 conda、Python 和一块 Nvidia 显卡，或者使用 Docker。还有一个免费的在线演示。","easeRationale":"网页端是 A：README 链接了一个免费演示页面，你可以在那里输入文字并试听，还有用于训练的 Colab 笔记本。Windows 是 B：下载来自 Hugging Face 的 \"整合包\"（一个 7z 文件），解压后双击 `go-webui.bat` 即可。Linux 和 Mac 是 D：你需要创建一个 conda 环境，并运行 `bash install.sh --device <CU126|CU128|ROCM|MPS|CPU> --source <HF|HF-Mirror|ModelScope>`，该命令会安装 ffmpeg、cmake 和一个编译器，并下载若干模型包；完整功能默认需要 Nvidia 显卡，Docker 镜像也需要 Nvidia 运行时。","costNote":"免费：代码采用 MIT 许可，训练好的模型在 Hugging Face 和 ModelScope 上公开发布，并提供 Hugging Face 上的免费在线演示。","notVerified":["仅用 5 秒或一分钟音频克隆出的声音实际听起来效果如何","下载预训练模型、语音识别模型和人声分离模型之后，实际占用多少磁盘空间","Windows 的 \"整合包\"（日期为 20250228）是否与当前代码和更新的版本一致","在没有 Nvidia 显卡的情况下是否能有效运行，速度会慢多少","免费在线演示是否会一直保持可用","安装器下载的第三方模型包和 Docker 镜像里面到底是什么内容"]}}},{"id":"R_kgDOTxc1Qg","slug":"ryanlaiyanip-ctrl__chatmany","name":"ChatMany","owner":"ryanlaiyanip-ctrl","fullName":"ryanlaiyanip-ctrl/chatmany","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"one_token_guards_all_your_contacts","severity":"attention","evidence":"README 第 2.7 步关于 OWNER_TOKEN 写道：\"This single value is the only thing protecting your contacts, emails, and campaigns from anyone who finds your web address.\" .env.example 也允许把它放进网址本身（\"send it as `Authorization: Bearer <OWNER_TOKEN>` or `?token=<OWNER_TOKEN>`\"），而网址很容易被记录在日志和浏览器历史记录中。","source":"ai","title":"一个 token 守护你所有的联系人"},{"code":"depends_on_meta_rules_and_your_own_published_app","severity":"attention","evidence":"该项目的运作方式是发布你自己的 Meta 应用：\"The app MUST be Published (switched from Development to Live)\"，以及 \"A 'Meta app' is just a permission slip... It's free. Nobody reviews it.\" 自动消息功能受 Meta 平台规则约束，Meta 可以随时对某个应用或账户更改或执行这些规则；开场消息是 \"a private reply to the comment ... valid up to 7 days, once per comment\"。","source":"ai","title":"依赖于 meta 的规则和你自己发布的应用"},{"code":"you_become_the_keeper_of_personal_data","severity":"info","evidence":"控制面板有一个 \"Contacts — everyone who entered a campaign, their funnel status, captured email, with Export CSV\" 页面，设置过程还要求你发布自己的 \"Privacy Policy URL\"、\"Terms of Service URL\" 和 \"Data deletion instructions URL\"。这些邮箱和联系人记录都保存在你自己的 Cloudflare 数据库中，随之而来的责任也由你承担。","source":"ai","title":"你成为个人数据的保管者"},{"code":"follow_check_is_not_real","severity":"info","evidence":"README 的 \"How it works\" 部分写道：\"Follow-gate is a nudge, not a check. No Instagram API can tell you whether a specific person follows you — there is no follower list and no relationship edge in the scopes this app holds.\" README 还说通过 API 给评论点赞是不可能的，所以那个开关是灰色不可用的。","source":"ai","title":"关注验证并非真实核实"},{"code":"young_project_with_no_releases","severity":"info","evidence":"该仓库创建于 2026-08-07，有 69 个 star，没有任何 GitHub release，package.json 中的版本号是 \"0.1.0\"，本次检查快照中的最后一次提交是在 2026-09-28，距检查时间 42 天。更新操作需要先 `git pull` 再做数据库迁移，README 警告顺序很重要：\"DATABASE FIRST, DEPLOY SECOND\"，因为顺序错了会 \"silently\" 失败。","source":"ai","title":"年轻项目且无任何 release"},{"code":"readme_asks_you_to_let_an_ai_run_the_setup","severity":"info","evidence":"README 提供了两段可以直接粘贴到 ChatGPT 或 Codex 的内容——\"or Codex, if you'd rather it run the commands for you\"——这意味着在安装过程中让助手运行终端命令，并处理你的 Meta 应用密钥。","source":"ai","title":"readme 建议让 ai 来完成安装"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta Instagram app (APP_ID and APP_SECRET)","Owner token you invent (OWNER_TOKEN)","Webhook verify token you invent (WEBHOOK_VERIFY_TOKEN)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true}],"declaredAlternatives":[],"stars":69,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":4,"growthDays":6,"history":[{"date":"2026-11-04","stars":65},{"date":"2026-11-05","stars":65},{"date":"2026-11-06","stars":68},{"date":"2026-11-07","stars":68},{"date":"2026-11-08","stars":68},{"date":"2026-11-09","stars":69},{"date":"2026-11-10","stars":69}],"firstPublishedAt":"2026-11-09T23:02:44.388Z","commitSha":"285ddd546aeb88db8e7900687a087162ef877a6a","verifiedAt":"2026-11-09T22:54:10.437Z","lastCommitAt":"2026-09-28T18:29:11Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Your own Instagram robot that answers a keyword in the comments with a direct message, a follow nudge and a link.","summary":"When someone comments a keyword on your Instagram post or reel, this sends them a direct message, can ask them to follow you, can ask for their email, then delivers your link. You run your own copy: you create a free Meta developer app, connect your Instagram Professional account and put the code on Cloudflare's free plan, so no outside company holds your contacts. It has a dashboard in the browser for building campaigns, seeing numbers and exporting contacts. You need Node.js 22, a terminal and about 45 minutes of clicking in Meta's dashboard.","easeRationale":"A D on Windows, Mac and Linux: the README itself says \"Setup takes about 45 minutes. It has six parts, and they must be done in this order\". You install Node.js 22 and git, clone the repository, run `npm install`, log in to Cloudflare, create a database, hand-edit two lines in `wrangler.toml`, set four secrets and deploy — and in between you create a Meta app, register the same address in four places and switch the app to Live. Every command is written out to copy, and the README even offers a prompt so an AI assistant walks you through it.","costNote":"MIT code and the README says \"Free ($0/month) at single-creator scale\" on Cloudflare's free plan, with a free Meta developer account. The free plan has limits: the README notes that Cloudflare allows 5 scheduled jobs per account and each install uses 2, so two installs fit.","notVerified":["Whether Meta keeps allowing this kind of automated messaging for your app and account","How many messages a day really get through, and what Instagram's limits are for you","Whether the measured claims about tokens lasting 60 days and renewing daily hold over time","Whether the Cloudflare free plan is enough once a campaign gets busy","Whether the project keeps working as Instagram's API changes, given it is young and has one open issue and no releases","What happens to your contacts and emails if your own token leaks"]},"zh-Hans":{"summaryShort":"你自己的 Instagram 机器人，当评论中出现关键词时，会自动发送私信、提示关注并附上链接。","summary":"当有人在你的 Instagram 帖子或 reel 下评论某个关键词时，这个程序会给对方发私信，可以要求对方关注你，还可以索要对方的邮箱，然后发出你的链接。你运行的是自己的一份副本：创建一个免费的 Meta 开发者应用，连接你的 Instagram 专业账户，并把代码部署在 Cloudflare 的免费套餐上，这样就没有外部公司保存你的联系人信息。它有一个浏览器中的控制面板，用于搭建活动、查看数据和导出联系人。你需要 Node.js 22、一个终端，以及大约 45 分钟在 Meta 后台里点击操作的时间。","easeRationale":"在 Windows、Mac 和 Linux 上都是 D：README 自己写道 \"Setup takes about 45 minutes. It has six parts, and they must be done in this order\"。你需要安装 Node.js 22 和 git，克隆仓库，运行 `npm install`，登录 Cloudflare，创建数据库，手动编辑 `wrangler.toml` 中的两行，设置四个 secrets 并部署——期间还要创建一个 Meta 应用，在四个地方注册同一个地址，并把应用切换为 Live。每一条命令都写好了可以直接复制，README 甚至还提供了一个提示词，方便 AI 助手带你完成整个流程。","costNote":"代码采用 MIT 许可，README 说在 Cloudflare 的免费套餐上 \"Free ($0/month) at single-creator scale\"，并使用免费的 Meta 开发者账户。免费套餐有限制：README 提到 Cloudflare 每个账户允许 5 个定时任务，每次安装占用 2 个，所以最多可以装两套。","notVerified":["Meta 是否会一直允许你的应用和账户使用这类自动消息功能","每天实际能发出多少条消息，以及 Instagram 对你的限制是什么","关于 token 可持续 60 天并每天续期的实测说法是否长期成立","活动变得繁忙后，Cloudflare 的免费套餐是否够用","随着 Instagram API 的变化，该项目是否能持续可用——毕竟它比较新，只有一个未解决的 issue，也没有任何 release","如果你自己的 token 泄露，你的联系人和邮箱会面临什么后果"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyNzU3NDQxOA==","slug":"ryanoasis__nerd-fonts","name":"Nerd Fonts","owner":"ryanoasis","fullName":"ryanoasis/nerd-fonts","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"low","signals":[{"code":"mixed_licenses","severity":"info","evidence":"LICENSE：\"Nerd Fonts source fonts, patched fonts, and folders with explict OFL SIL files are licensed under SIL OPEN FONT LICENSE Version 1.1\"，脚本是 MIT 许可，而且 \"Many other licenses are present in this project\"，还有单独的 license-audit.md。GitHub 本身把这个仓库的许可证标记为 NOASSERTION。","source":"ai","title":"许可证混杂"},{"code":"third_party_remote_script_installer","severity":"info","evidence":"readme.md 列出了一个第三方 PowerShell 安装程序，一步就从另一个域名下载并运行脚本：\"& ([scriptblock]::Create((iwr 'https://to.loredo.me/Install-NerdFont.ps1')))\"。Chocolatey 和 Scoop 的包也被标注为非官方。这些都是可选的；release 压缩包才是主要的下载方式。","source":"ai","title":"第三方远程脚本安装程序"},{"code":"reserved_font_names","severity":"info","evidence":"打了补丁的字体表格中，很多字体家族被标记为 \"RFN = Reserved Font Name\"（Anonymous Pro、IBM Plex Mono、Cascadia Code、Source Code Pro 等），这也是打补丁后的版本要改名的原因。如果你自己打补丁并分享字体，OFL 关于保留名称的规则同样适用于你。","source":"ai","title":"保留字体名称"},{"code":"repository_itself_is_not_the_download","severity":"info","evidence":"readme.md 重要提示：\"cloning this repository is not recommended (due to Repo size)\"，选项 8：\"Will not work to get newer fonts as they are not inside the repo anymore.\" 字体来自 release 压缩包，最新的发行版（v3.5.1）没有按平台提供的二进制文件，只有字体压缩包。","source":"ai","title":"仓库本身不是下载来源"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["development","design_cad"],"alternativeTo":[],"declaredAlternatives":[],"stars":65187,"downloads":47355439,"downloadsWeek":221523,"downloadsDays":7,"growth":63,"growthDays":7,"history":[{"date":"2026-10-26","stars":65012},{"date":"2026-10-27","stars":65040},{"date":"2026-10-28","stars":65052},{"date":"2026-10-29","stars":65065},{"date":"2026-10-30","stars":65078},{"date":"2026-10-31","stars":65086},{"date":"2026-11-01","stars":65095},{"date":"2026-11-02","stars":65108},{"date":"2026-11-03","stars":65124},{"date":"2026-11-04","stars":65130},{"date":"2026-11-05","stars":65133},{"date":"2026-11-06","stars":65139},{"date":"2026-11-07","stars":65152},{"date":"2026-11-08","stars":65166},{"date":"2026-11-09","stars":65172},{"date":"2026-11-10","stars":65187}],"firstPublishedAt":"2026-10-27T01:10:12.680Z","commitSha":"adbd16ab7ca911eaca7ce3b0f2b7994207c7ecae","verifiedAt":"2026-10-31T01:40:30.975Z","lastCommitAt":"2026-10-17T22:03:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Get popular programming fonts with thousands of extra icons built in, for nicer terminals and code editors.","summary":"Nerd Fonts takes well-known programming fonts (Hack, JetBrains Mono, Fira Code, Source Code Pro and about 67 families in total) and adds more than 10,000 small icons to them, so terminals, shell prompts and code editors can show folder, file-type, Git and status symbols. You download the font archive for the family you want and install the font files like any other font. A separate script (font-patcher) lets you add the same icons to a font of your own, but that one needs FontForge and Python, or Docker.","easeRationale":"The same on all three: there is no installer, so you download a font archive from the releases page (for example JetBrainsMono.tar.xz), unzip it, install the font files the way your system does it, and then pick the font in your terminal or editor settings. Mac and Linux also have a one-command route (brew install font-hack-nerd-font, or the project's install.sh).","notVerified":["Whether the icons show correctly in your specific terminal or editor on Windows.","Disk space used if you install many or all of the 67 font families.","What the unofficial Chocolatey, Scoop and third-party PowerShell installers actually download and run.","Whether the font patcher runs on Windows, since it needs FontForge with Python, or Docker."]},"zh-Hans":{"summaryShort":"获取流行的编程字体，内置数千个额外图标，让终端和代码编辑器更好看。","summary":"Nerd Fonts 把常见的编程字体（Hack、JetBrains Mono、Fira Code、Source Code Pro，总共约 67 个字体家族）加上了超过 10,000 个小图标，这样终端、shell 提示符和代码编辑器就能显示文件夹、文件类型、Git 和状态符号。你下载想要的字体家族压缩包，然后像安装其他字体一样安装这些字体文件。另外还有一个单独的脚本（font-patcher），可以把同样的图标加到你自己的字体上，但它需要 FontForge 和 Python，或者 Docker。","easeRationale":"三个系统都一样：没有安装程序，所以你从 releases 页面下载一个字体压缩包（例如 JetBrainsMono.tar.xz），解压，用你系统自带的方式安装字体文件，然后在终端或编辑器设置里选中这个字体。Mac 和 Linux 还有一条命令就能完成的方式（brew install font-hack-nerd-font，或者项目自带的 install.sh）。","notVerified":["图标在你 Windows 上具体使用的终端或编辑器里是否能正常显示。","如果安装很多个甚至全部 67 个字体家族，会占用多少磁盘空间。","非官方的 Chocolatey、Scoop 和第三方 PowerShell 安装程序到底下载和运行了什么。","字体打补丁工具能否在 Windows 上运行，因为它需要带 Python 的 FontForge，或者 Docker。"]}}},{"id":"R_kgDOUIZL3w","slug":"ryze-ai-adgent__open-seo-mcp-skills","name":"Open SEO MCP Skills","owner":"Ryze-AI-Adgent","fullName":"Ryze-AI-Adgent/open-seo-mcp-skills","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"your_site_and_ads_data_flow_through_a_company_server","severity":"attention","evidence":"这些技能只描述要做的工作；数据来自厂商托管的服务器：\"claude mcp add ryze --transport http https://connector.get-ryze.ai/mcp\" 以及 \"Sign in, pick your workspace, connect your Google Search Console / GA4 / ads accounts once.\" README 最后写道：\"Data access runs through your own Ryze workspace and your own connected accounts.\" 这家公司具体保留或记录了什么，本仓库并未说明。","source":"ai","title":"你网站和广告的数据会经过一家公司的服务器"},{"code":"repository_is_the_front_door_of_a_paid_product","severity":"info","evidence":"几乎每个链接都带着指向厂商网站的追踪标签（`?utm_source=github&utm_content=open-seo-mcp-skills`），README 开头写着 \"Skip the setup: use the hosted Ryze MCP\"，还有一个\"Connect Ryze in one click\"按钮，并宣传其他产品（\"Google Ads MCP\"、\"Meta Ads MCP\"）。对比表格以及诸如 \"It is the only one we know of\" 这样的句子都是厂商自己的说法。","source":"ai","title":"这个仓库是一个付费产品的入口"},{"code":"plugin_install_is_not_pinned_to_a_version","severity":"info","evidence":"GitHub 上没有任何 release，所以 `claude plugin marketplace add Ryze-AI-Adgent/open-seo-mcp-skills` 会安装那一刻仓库里的任何内容。这些文件是你的助手要遵循的指令，而这个仓库很新：创建于 2026-08-29，最近一次提交在 2026-10-27，距检查日仅 13 天。","source":"ai","title":"插件安装没有锁定版本"},{"code":"installer_overwrites_skills_with_the_same_name","severity":"info","evidence":"install.sh 会在未征得同意的情况下把每个文件夹复制进你的全局技能目录：`DEST=\"$HOME/.claude/skills\"` 以及 `cp -R \"$d\" \"$DEST/$name\"`。如果你已有同名的技能，就会被替换，而且这些技能会安装到所有项目中，而不只是一个项目。","source":"ai","title":"安装程序会覆盖同名技能"},{"code":"many_stars_in_a_short_time","severity":"info","evidence":"这个仓库创建于 2026-08-29，大约两个月后就有 9048 个星标和 395 个 fork，内容只有一个 README、一份许可证、一个 `install.sh`、一个插件清单和 `skills/` 文件夹，有 4 个未关闭的 issue，也没有任何 release。","source":"ai","title":"短时间内获得大量星标"},{"code":"features_depend_on_your_workspace","severity":"info","evidence":"README 的\"Notes for skill authors\"写道：\"Tool availability can vary by workspace; every skill here states its fallback.\" 所以这八项技能中哪些真正能返回数据，取决于厂商为你开放了什么，以及你连接了哪些 Google 账户。","source":"ai","title":"功能取决于你的工作区"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Ryze workspace (hosted connector at connector.get-ryze.ai)","Your Google accounts: Search Console, Analytics 4, Google Ads (linked through the Ryze sign-in)","Ahrefs or Semrush"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","ai_skills"],"alternativeTo":[{"slug":"semrush","name":"Semrush","confirmed":true},{"slug":"ahrefs","name":"Ahrefs","confirmed":true},{"slug":"moz-pro","name":"Pro","confirmed":true},{"slug":"surfer-seo","name":"SEO","confirmed":true}],"declaredAlternatives":[],"stars":9105,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":345,"growthDays":7,"history":[{"date":"2026-11-04","stars":8806},{"date":"2026-11-05","stars":8865},{"date":"2026-11-06","stars":8875},{"date":"2026-11-07","stars":8915},{"date":"2026-11-08","stars":8965},{"date":"2026-11-09","stars":9043},{"date":"2026-11-10","stars":9105}],"firstPublishedAt":"2026-11-09T23:02:44.398Z","commitSha":"f587b0a69dc1ed92b65cdb241e4f14985b1031fd","verifiedAt":"2026-11-09T22:55:26.286Z","lastCommitAt":"2026-10-27T08:57:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Eight SEO jobs for Claude - audits, keywords, rankings, backlinks - reading your real Search Console and Analytics data.","summary":"A pack of eight written skills that let Claude do search-engine work on your own site: a site audit, keyword research, rank movements, competitor keyword gaps, backlinks, which AI engines send you traffic, content briefs, and paid-versus-organic overlap. The Markdown files are free, but on their own they do nothing: the numbers come from the company's hosted connector at connector.get-ryze.ai, where you sign in and link your Google Search Console, Analytics 4 and ads accounts. You need Claude Code, Claude Desktop, Cursor or ChatGPT to use them.","easeRationale":"The same C on all three systems: you must already have Claude Code (or Claude Desktop, Cursor or ChatGPT), then run two commands — `claude plugin marketplace add Ryze-AI-Adgent/open-seo-mcp-skills` and `/plugin install open-seo-mcp-skills@ryze` — or clone the repository and copy the `skills/` folders into `~/.claude/skills`. Before anything returns data you also add the connector by hand and sign in to link your Google accounts in the browser.","costNote":"The skills are MIT and the README says the connector is \"free to add\" with \"No subscription for the tool. No markup on API calls\". You still pay for the AI assistant that runs them, and the data flows through a Ryze workspace whose own terms and prices are not in this repository; Ahrefs and Semrush can be connected \"if you already pay for them\".","notVerified":["What the hosted connector does with your Search Console, Analytics and ads data, and where it is stored","Whether the connector stays free, and what the Ryze workspace costs once limits are reached","Whether the paid-source data (DataForSEO backlinks, competitor keywords, SERPs) really comes with no extra bill","How good the eight skills' output is compared with Semrush or Ahrefs","Whether the skills work as described in Cursor, ChatGPT or Claude Desktop, as the README claims","How much of your Claude plan a full audit consumes"]},"zh-Hans":{"summaryShort":"面向 Claude 的八项 SEO 任务——站点审查、关键词、排名、反向链接——读取你真实的 Search Console 和 Analytics 数据。","summary":"这是一套由八个 Markdown 技能组成的包，让 Claude 在你自己的网站上完成搜索引擎相关工作：站点审查、关键词调研、排名变化、与竞争对手的关键词差距、反向链接、哪些 AI 引擎给你带来流量、内容简报，以及付费流量与自然流量的重叠情况。这些 Markdown 文件本身是免费的，但单独使用什么也做不了：数据来自该公司托管的连接器 connector.get-ryze.ai，你需要登录并关联你的 Google Search Console、Analytics 4 和广告账户。要使用它们，你需要 Claude Code、Claude Desktop、Cursor 或 ChatGPT。","easeRationale":"三个系统上的易用度都是 C：你必须已经安装 Claude Code（或 Claude Desktop、Cursor、ChatGPT），然后运行两条命令——`claude plugin marketplace add Ryze-AI-Adgent/open-seo-mcp-skills` 和 `/plugin install open-seo-mcp-skills@ryze`——或者克隆仓库并把 `skills/` 文件夹复制到 `~/.claude/skills`。在返回任何数据之前，你还要手动添加连接器，并在浏览器里登录以关联你的 Google 账户。","costNote":"这些技能采用 MIT 许可，README 说连接器\"free to add\"，并且\"No subscription for the tool. No markup on API calls\"。但你仍要为运行它们的 AI 助手付费，数据会经过一个 Ryze 工作区，其条款和价格并不在本仓库中；Ahrefs 和 Semrush 可以连接，\"if you already pay for them\"。","notVerified":["托管连接器会如何处理你的 Search Console、Analytics 和广告数据，以及数据存放在哪里","连接器是否会一直保持免费，超出限额后 Ryze 工作区要收多少费用","付费数据源（DataForSEO 的反向链接、竞争对手关键词、SERP）是否真的不产生额外费用","这八项技能的结果与 Semrush 或 Ahrefs 相比质量如何","这些技能是否真的像 README 所说那样，在 Cursor、ChatGPT 或 Claude Desktop 中也能正常使用","一次完整审查会消耗你 Claude 套餐的多少额度"]}}},{"id":"R_kgDON29OPQ","slug":"saiyan-world__goku","name":"Goku","owner":"Saiyan-World","fullName":"Saiyan-World/goku","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单均未检测到许可证","source":"check","title":"没有许可证"},{"code":"known_vulnerabilities","severity":"attention","evidence":"被锁定版本的依赖存在已知漏洞（OSV）：diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446；transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74","source":"check","title":"存在已知漏洞"},{"code":"no_usage_instructions","severity":"info","evidence":"README.md 里只有论文链接、项目主页、基准测试表和 BibTeX；整篇文档里没有任何安装、下载或推理命令。","source":"ai","title":"没有使用说明"},{"code":"no_model_weights_published","severity":"info","evidence":"根目录下的文件只有 .gitignore、README.md、configs/、goku/、requirements.txt 和 tools/，README 也没有链接任何 checkpoint 或权重下载，所以这份代码没有可加载的模型。","source":"ai","title":"没有发布模型权重"},{"code":"no_activity_since_paper_release","severity":"info","evidence":"创建于 2025-02-10，最后一次推送在 2025-02-19（距本次检查 628 天），没有任何 release：这个仓库看起来更像是配合 arXiv 论文的一次性发布。","source":"ai","title":"自论文发布后没有新动态"}],"platforms":["linux"],"tools":["python","gpu_nvidia"],"buildTools":["python"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["video","image","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":2913,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":2915},{"date":"2026-11-08","stars":2914},{"date":"2026-11-09","stars":2913},{"date":"2026-11-10","stars":2913}],"firstPublishedAt":"2026-11-09T23:02:44.407Z","commitSha":"e29a217bdb9195fa26b2db6add9f1848472a1bf9","verifiedAt":"2026-11-09T22:46:15.112Z","lastCommitAt":"2025-02-19T14:32:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code for a model family that makes videos and images from a text description, released with a paper.","summary":"Goku is the published code of a research project from HKU and ByteDance for generating video from text, video from an image and images from text. It is aimed at researchers and developers: the repository holds the model code, configuration files and a Python requirements list, but the README has no installation or run steps and no model files to download, so there is nothing a normal user can open. Running it would mean Python and a strong NVIDIA graphics card on Linux. The repository has no license, and nothing has changed since February 2025.","easeRationale":"Linux with Python and an NVIDIA card is the only realistic setup, and even there the README offers only the paper, benchmark tables and a citation, with no install or generation commands, so it is for developers who will read the code.","notVerified":["whether the code can generate anything at all, since no model files are published","how much graphics memory and disk space a generation really needs","whether it can be made to run on Windows or macOS","whether the benchmark scores in the README can be reproduced"]},"zh-Hans":{"summaryShort":"一个模型家族的研究代码，能根据文字描述生成视频和图像，随论文一起发布。","summary":"Goku 是香港大学（HKU）和字节跳动一个研究项目公开发布的代码，用于根据文本生成视频、根据图像生成视频以及根据文本生成图像。它面向研究者和开发者：仓库里有模型代码、配置文件和一份 Python 依赖列表，但 README 没有安装或运行步骤，也没有可下载的模型文件，普通用户没有任何东西可以直接打开使用。要运行它需要在 Linux 上配备 Python 和一块性能较强的 NVIDIA 显卡。这个仓库没有许可证，而且自 2025 年 2 月以来没有任何改动。","easeRationale":"在 Linux 上配备 Python 和 NVIDIA 显卡是唯一现实的配置方式，即便如此，README 也只提供了论文、基准测试表和引用格式，没有安装或生成命令，所以这是留给愿意读代码的开发者的项目。","notVerified":["这份代码到底能不能生成任何内容，因为没有公开任何模型文件","一次生成实际需要多少显存和磁盘空间","能否让它在 Windows 或 macOS 上运行","README 里的基准测试分数能否被复现"]}}},{"id":"R_kgDOLKJXyA","slug":"sakethsripada__flask-socialmedia-automation","name":"Flask Social Media Automation","owner":"SakethSripada","fullName":"SakethSripada/Flask-SocialMedia-Automation","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"被锁定版本的依赖存在已知漏洞（OSV）：anyio: GHSA-5p39-cfhj-2xmp, GHSA-82r6-8w77-94w6, PYSEC-2026-4024；certifi: GHSA-248v-346w-9cwc, GHSA-6q48-6c7w-9gw2, PYSEC-2024-230；dnspython: GHSA-3rq5-2g8h-59hc, GHSA-45c3-73f7-pv4m, PYSEC-2026-1307；flask: GHSA-68rp-wp8r-4726, PYSEC-2026-2151；h11: GHSA-vqfr-h8mv-ghfj, PYSEC-2026-348；idna: GHSA-65pc-fj4g-8rjx, GHSA-jjg7-2v4v-x38h, PYSEC-2024-60；instagrapi: GHSA-ggxf-37hm-9wqf；jinja2: GHSA-cpwx-vrp4-4pq7, GHSA-gmj6-6f8f-6699, GHSA-h75v-3vvj-5mfj；mako: GHSA-2h4p-vjrc-8xpq, GHSA-5639-2j2p-m4mx, GHSA-v92g-xgxw-vvmm；oauthlib: GHSA-7984-qgvg-3j9c, GHSA-hj66-6f7g-4r5v, GHSA-xpv3-w29h-x7cv；pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h；pygments: GHSA-5239-wwwm-4pmq, PYSEC-2026-2987；requests: GHSA-9hjg-9r4m-mvj7, GHSA-9wx4-h78v-vm56, GHSA-gc5v-m9x4-r6x2；soupsieve: GHSA-2wc2-fm75-p42x, GHSA-836r-79rf-4m37, GHSA-gjv8-xp57-g29c；tqdm: GHSA-g7vv-2v7x-gj9p, PYSEC-2026-1976；urllib3: GHSA-2xpw-w6gg-jr37, GHSA-34jh-p97f-mpxf, GHSA-38jv-5279-wg99；werkzeug: GHSA-29vq-49wr-vm6x, GHSA-2g68-c3qc-8985, GHSA-87hc-h4r5-73f7","source":"check","title":"存在已知漏洞"},{"code":"instagram_unofficial_api","severity":"attention","evidence":"requirements.txt 锁定了 instagrapi==2.0.3，仓库说明里写着它 'uses instagrapi to automate posting at scheduled times'；instagrapi 会在官方 API 之外用你的 Instagram 账户登录，这是 Instagram 条款不允许的做法，可能导致账户被限制。","source":"ai","title":"使用非官方的 instagram api"},{"code":"credentials_in_plain_file","severity":"attention","evidence":"设置步骤 3 要求创建 secret.py，把 MAIL_PASSWORD、API_KEY（OpenAI）、五个 Twitter 令牌、OAuth 客户端密钥和 NEWS_API_KEY 以明文形式写在项目文件夹里。","source":"ai","title":"凭据保存在明文文件里"},{"code":"handles_other_people_accounts","severity":"info","evidence":"README.md 记录了 `flask delete-all-users`、`flask delete-users` 和 `flask list-users`，说明里还提到注册和邮件验证：这个应用把其他人的账户存放在本地的 SQLite 数据库里。","source":"ai","title":"处理其他人的账户"},{"code":"no_recent_activity","severity":"info","evidence":"最后一次推送在 2024-12-25（距本次检查 684 天），没有任何 release，而 README 里仍然列着 'Upcoming Features'。","source":"ai","title":"近期没有更新"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI","Gmail app password (login and password-reset emails)","X (Twitter) API","NewsAPI"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"later","name":"Later","confirmed":true}],"declaredAlternatives":[],"stars":38,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":38},{"date":"2026-11-09","stars":38},{"date":"2026-11-10","stars":38}],"firstPublishedAt":"2026-11-09T23:02:44.416Z","commitSha":"fd479f4bb8c74319efa3cd973a5aa89ffc29a1aa","verifiedAt":"2026-11-09T22:47:01.292Z","lastCommitAt":"2024-12-25T09:01:40Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A website you run on your own computer to schedule Instagram posts and tweets, with text and images made by AI.","summary":"A small Flask website you install and run yourself. After logging in you can upload an image or write a text, schedule it for Instagram or X (Twitter), or ask OpenAI to write the tweet and draw the picture; it can also post daily news headlines and send scheduled emails from your Gmail account. For people comfortable with a terminal: you need Python, keys from four services (OpenAI, X, NewsAPI, Gmail app password) written into a secret.py file, and a few database commands before the site starts.","easeRationale":"Same work on Windows, macOS and Linux: install Python and the packages, sign up for up to four services, hand-write a secret.py with thirteen values, then run three database commands before `flask run`. There is no installer and no hosted version.","costNote":"The code is free (MIT). The AI posts go through the paid OpenAI API, and posting to X needs a developer account on its paid tiers.","notVerified":["whether Instagram still accepts logins through instagrapi, or blocks the account","how much the OpenAI calls and the X developer account cost per post in practice","whether scheduled posts go out when the computer is off, since the scheduler runs inside the app","whether the pinned package versions still install on current Python releases"]},"zh-Hans":{"summaryShort":"一个你在自己电脑上运行的网站，用来安排 Instagram 发帖和推文，文字和图片都由 AI 生成。","summary":"一个你自己安装并运行的小型 Flask 网站。登录后，你可以上传一张图片或写一段文字，安排它发到 Instagram 或 X（Twitter），也可以让 OpenAI 帮你写推文并画图；它还能发布每日新闻标题，并通过你的 Gmail 账户发送定时邮件。适合熟悉终端的人：你需要 Python，还需要把来自四个服务（OpenAI、X、NewsAPI、Gmail 应用专用密码）的密钥写进一个 secret.py 文件，并在网站启动前运行几条数据库命令。","easeRationale":"在 Windows、macOS 和 Linux 上工作量相同：安装 Python 和相关包，最多要在四个服务上注册账号，手写一个包含十三个值的 secret.py，然后在运行 `flask run` 之前执行三条数据库命令。没有安装程序，也没有托管版本。","costNote":"代码是免费的（MIT）。AI 生成的帖子要走 OpenAI 的付费 API，发到 X 上还需要一个使用付费层级的开发者账户。","notVerified":["Instagram 是否仍然接受通过 instagrapi 的登录，还是会封锁账户","实际使用中，OpenAI 调用和 X 开发者账户每条帖子要花多少钱","电脑关机时，定时发布的内容是否还会发出去，因为调度器是在这个应用内部运行的","锁定版本的依赖包是否还能在当前版本的 Python 上安装"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMDMzNTA4MzI=","slug":"salu133445__musegan","name":"MuseGAN","owner":"salu133445","fullName":"salu133445/musegan","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"被锁定版本的依赖存在已知漏洞（OSV）：setuptools: GHSA-5rjg-fvgr-3xxf, GHSA-cx63-2mw6-8hw5, GHSA-h35f-9h28-mq5c；grpcio: GHSA-496j-2rq6-j6cc, PYSEC-2026-1424；markdown: GHSA-5wmx-573v-2qwq, PYSEC-2026-89；numpy: GHSA-5545-2q6w-2gh6, GHSA-6p56-wp2h-9hxr, GHSA-9fq2-x9r6-wfmf；pillow: GHSA-254f-3936-288h, GHSA-27h8-gf7j-v5hc, GHSA-332p-7hrc-92wf；protobuf: GHSA-7gcm-g887-7qv7, GHSA-8gq9-2x98-w8hf, GHSA-8qvm-5x2c-j2w7；pyyaml: GHSA-8q59-q68h-6hv4, GHSA-rprw-h62v-c2w7, PYSEC-2018-49；scipy: PYSEC-2023-102, PYSEC-2023-114；tensorflow-gpu: GHSA-23hm-7w47-xw72, GHSA-2475-53vw-vp25, GHSA-247x-2f9f-5wp7；werkzeug: GHSA-29vq-49wr-vm6x, GHSA-2g68-c3qc-8985, GHSA-87hc-h4r5-73f7","source":"check","title":"存在已知漏洞"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2/10","source":"check","title":"scorecard 分数很低"},{"code":"obsolete_framework_pinned","severity":"attention","evidence":"requirements.txt 锁定了 2018 年的版本：tensorflow-gpu==1.10.1、tensorboard==1.10.0、numpy==1.14.5 和 scipy==1.1.0；TensorFlow 1.x 需要旧版本的 Python 和 CUDA，而且已经不再维护。","source":"ai","title":"锁定了过时的框架版本"},{"code":"downloads_data_from_outside_github","severity":"info","evidence":"README.md 中的 `./scripts/download_data.sh` 和 `./scripts/download_models.sh` 会从个人的 OneDrive 账户（ucsdcloud-my.sharepoint.com）下载训练数据和预训练模型，而不是从仓库本身下载。","source":"ai","title":"从 github 之外下载数据"},{"code":"no_audio_output","severity":"info","evidence":"Outputs 部分列出的是 .npy、.png 和 .npz 格式的 piano-roll 文件，并展示了用 Pypianoroll 写出 .mid 文件的 Python 代码：没有音频渲染，也没有播放器。","source":"ai","title":"没有音频输出"},{"code":"no_recent_activity","severity":"info","evidence":"最后一次推送在 2024-06-07（距本次检查 885 天），没有任何 release，21 个未关闭的 issue，README 还指向另一个仓库获取 PyTorch 版本。","source":"ai","title":"近期没有更新"}],"platforms":["linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"suno","name":"Suno","confirmed":true},{"slug":"udio","name":"Udio","confirmed":true}],"declaredAlternatives":[],"stars":2045,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":2045},{"date":"2026-11-07","stars":2045},{"date":"2026-11-08","stars":2045},{"date":"2026-11-09","stars":2045},{"date":"2026-11-10","stars":2045}],"firstPublishedAt":"2026-11-09T23:02:44.426Z","commitSha":"3c9a0dc71ab743b58f4e0620832f21dbe023408e","verifiedAt":"2026-11-09T22:47:35.956Z","lastCommitAt":"2024-06-07T10:22:36Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that generates short pop-style arrangements for five instruments as MIDI, not as finished audio.","summary":"MuseGAN is the code behind a 2018 research paper on music generation. It trains a model on a piano-roll dataset and then writes short multi-track phrases (bass, drums, guitar, piano, strings), which come out as numpy arrays, images and piano-roll files you convert to MIDI yourself; there is no audio and no interface. It is for researchers and developers on Linux with Python and a graphics card, and the pinned packages are the 2018 versions of TensorFlow 1 and numpy, so the environment has to be rebuilt by hand.","easeRationale":"Linux is the only documented path and it is developer work: every step is a shell script (download_data.sh, run_inference.sh), and requirements.txt pins tensorflow-gpu==1.10.1, numpy==1.14.5 and setuptools==39.1.0 from 2018, which no longer install on current Python without rebuilding an old environment.","notVerified":["whether the OneDrive links for the dataset and the pretrained models still work","how the generated phrases actually sound once turned into MIDI and played with instruments","whether the old TensorFlow 1 code can be made to run on a current graphics card","how long training takes and how much disk the dataset needs"]},"zh-Hans":{"summaryShort":"一份研究代码，为五种乐器生成简短的流行风格编曲，输出 MIDI，而不是成品音频。","summary":"MuseGAN 是 2018 年一篇音乐生成研究论文背后的代码。它在一个 piano-roll 数据集上训练模型，然后生成简短的多轨乐句（低音、鼓、吉他、钢琴、弦乐），输出形式是 numpy 数组、图片和 piano-roll 文件，需要你自己转换成 MIDI；没有音频，也没有界面。它面向在 Linux 上使用 Python 和显卡的研究者和开发者，锁定的依赖是 2018 年版本的 TensorFlow 1 和 numpy，所以环境需要手动重建。","easeRationale":"Linux 是唯一有文档说明的路径，而且是开发者的活儿：每一步都是一个 shell 脚本（download_data.sh、run_inference.sh），requirements.txt 锁定了 2018 年的 tensorflow-gpu==1.10.1、numpy==1.14.5 和 setuptools==39.1.0，在当前的 Python 上已经无法直接安装，需要重建一个旧环境。","notVerified":["数据集和预训练模型的 OneDrive 链接是否还能用","生成的乐句转换成 MIDI 并用乐器演奏后实际听起来怎样","这份老旧的 TensorFlow 1 代码能否在当前的显卡上运行起来","训练需要多长时间，数据集占用多少磁盘空间"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyNDcxNjc2NTQ=","slug":"salu133445__muspy","name":"MusPy","owner":"salu133445","fullName":"salu133445/muspy","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最新一次 release 在 2022-04-16（1668 天前）","source":"check","title":"release 已经过时"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.1/10","source":"check","title":"scorecard 分数很低"},{"code":"downloads_third_party_datasets","severity":"info","evidence":"Features 列表第一项就是 'Dataset management system for commonly used datasets'，Disclaimer 中写道 'This is a utility library that downloads and prepares public datasets'；音乐数据来自其他人的服务器，这次检查估计会占用 1.4-5.8 GB 的磁盘空间。","source":"ai","title":"下载第三方数据集"},{"code":"dataset_license_is_your_responsibility","severity":"attention","evidence":"README.md 的 Disclaimer 写道：'We do not host or distribute these datasets, vouch for their quality or fairness, or claim that you have license to use the dataset. It is your responsibility to determine whether you have permission to use the dataset under the dataset's license.'","source":"ai","title":"数据集许可由你自行负责"},{"code":"beta_status_declared","severity":"info","evidence":"setup.py 的 classifiers 声明 'Development Status :: 4 - Beta'，最新的 release 是 2022-04-16 的 v0.5.0，有 26 个未关闭的 issue。","source":"ai","title":"声明为测试版状态"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","gpu_any"],"disk":{"minMB":1400,"maxMB":5800,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":530,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":529},{"date":"2026-11-08","stars":529},{"date":"2026-11-09","stars":529},{"date":"2026-11-10","stars":530}],"firstPublishedAt":"2026-11-09T23:02:44.435Z","commitSha":"2e1dc660dde6974c1693147f528d7437019d9580","verifiedAt":"2026-11-09T22:48:54.921Z","lastCommitAt":"2024-01-25T09:40:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Python library for people building music-generation models: it loads datasets, converts MIDI and measures results.","summary":"MusPy is a Python library for researchers and developers who build systems that compose music as notes (not as recorded audio). It downloads and prepares well-known music datasets, reads and writes MIDI, MusicXML and ABC files, turns pieces into the formats models expect, and offers evaluation tools with score and piano-roll pictures and audio rendering. It installs with `pip install muspy` and is used from your own Python code; there is no app or interface. The last release is from 2022.","easeRationale":"This is a library, not a program: on Windows, macOS and Linux the only instruction is `pip install muspy` and then writing Python code that calls it, so it is E everywhere.","notVerified":["whether the dataset download links inside the library still work","how much disk and time each dataset really takes","whether it still installs cleanly next to current versions of music21, PyTorch and TensorFlow","whether the audio rendering needs extra software installed on the machine"]},"zh-Hans":{"summaryShort":"一个面向音乐生成模型开发者的 Python 库：加载数据集、转换 MIDI、评估结果。","summary":"MusPy 是一个面向研究者和开发者的 Python 库，适用于构建以音符（而不是录制音频）形式创作音乐的系统。它可以下载并准备常用的音乐数据集，读写 MIDI、MusicXML 和 ABC 文件，把乐曲转换成模型所需的格式，并提供带乐谱和 piano-roll 图像以及音频渲染的评估工具。安装方式是 `pip install muspy`，在你自己的 Python 代码里调用；没有应用程序或界面。最新一个 release 来自 2022 年。","easeRationale":"这是一个库，不是一个程序：在 Windows、macOS 和 Linux 上唯一的说明就是 `pip install muspy`，然后编写调用它的 Python 代码，所以各平台的易用度都是 E。","notVerified":["库内部的数据集下载链接是否还能用","每个数据集实际需要多少磁盘空间和时间","它是否还能和当前版本的 music21、PyTorch、TensorFlow 顺利一起安装","音频渲染是否需要在机器上额外安装软件"]}}},{"id":"R_kgDOOs99Gg","slug":"sam42a__dune","name":"DUNE","owner":"Sam42a","fullName":"Sam42a/DUNE","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{},"risk":"attention","signals":[{"code":"archived","severity":"info","evidence":"该仓库已被归档(只读,不再维护)","source":"check","title":"已归档"},{"code":"unofficial_fork_frozen_on_an_old_base","severity":"attention","evidence":"README:\"DUNE is a modified version of the official Jellyfin Android TV client\" 以及 \"This is an unofficial fork not affiliated with the Jellyfin project\"。该仓库已被归档(最后一次提交在 2026-01-19),因此它停留在分叉时的 Jellyfin SDK 版本上,而官方客户端仍在不断变化。","source":"ai","title":"非官方分支停留在旧版本上"},{"code":"app_updates_itself_from_an_unnamed_source","severity":"attention","evidence":"Key Features, Technical Improvements:\"Built in automatic updates\"。README 没有说明应用从哪里获取这些更新,也没说如何检查更新,而且该项目已被归档。","source":"ai","title":"应用从未具名的来源自我更新"},{"code":"no_app_store_and_no_checksum_for_the_files","severity":"attention","evidence":"项目资料显示 \"GitHub releases: 8\",最近三个版本(v0.1.1、v0.1.0、v0.0.9)都被标记为 \"no platform binaries\"。README 没有提到任何应用商店,也没有发布校验和或签名;唯一的安装说明是 ./gradlew installStandardDebug 和 installEnhancedRelease。","source":"ai","title":"没有应用商店,文件也没有校验和"},{"code":"crash_reporting_library_included","severity":"info","evidence":"Third-Party Libraries 列出了 ACRA(github.com/ACRA/acra),一个安卓崩溃报告库;README 没有说明报告是否会被发送到某处,或发送给谁。","source":"ai","title":"包含崩溃报告库"},{"code":"needs_your_own_jellyfin_server","severity":"info","evidence":"该仓库的描述是 \"Dune is a Jellyfin Android tv Client\",所有功能都围绕浏览和播放媒体库;没有自己的 Jellyfin 服务器就没有可看的内容。","source":"ai","title":"需要你自己的 Jellyfin 服务器"},{"code":"early_version_numbers_and_open_issues","severity":"info","evidence":"最新版本是 v0.1.1(2025-12-23),该仓库有 60 个未解决的 issue,并在那个版本发布约一个月后被归档。","source":"ai","title":"版本号较早,且有未解决的 issue"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["video","hardware"],"alternativeTo":[{"slug":"plex","name":"Plex","confirmed":true},{"slug":"emby","name":"Emby","confirmed":true}],"declaredAlternatives":[],"stars":968,"downloads":30786,"downloadsWeek":40,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":970},{"date":"2026-11-07","stars":970},{"date":"2026-11-08","stars":970},{"date":"2026-11-09","stars":970},{"date":"2026-11-10","stars":968}],"firstPublishedAt":"2026-11-09T23:02:44.444Z","commitSha":"bab9d3c8a29ea92cdfcbd51ffa2969d0018fc71f","verifiedAt":"2026-11-09T22:49:47.999Z","lastCommitAt":"2026-01-19T14:33:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android TV app, redesigned from the official Jellyfin client, for watching the films and series on your own server.","summary":"DUNE is an unofficial, restyled copy of the Jellyfin app for Android TV: a new home screen, several themes including one for OLED screens, voice search, subtitle controls and the option to keep it installed next to the official app. It only shows what is already on a Jellyfin media server you run yourself. The repository is archived, so it gets no more changes, and the README documents only building the file yourself with Android Studio, Android SDK 35 and OpenJDK 21. It does not run on a computer.","easeRationale":"Android TV is the only system it runs on, and there it earns a D: the README's only documented path is \"Building from Source\" with Android Studio Giraffe, Android SDK API 35 and OpenJDK 21 and commands like ./gradlew assembleEnhancedRelease, plus you must already have a Jellyfin server. Computers get no grade; there is no desktop version.","costNote":"Free and GPL-2.0; the README has a \"Support Me - Buy a Coffee\" badge pointing to coff.ee/sam42.","notVerified":["whether any ready-made file matches the source in this repository, since the README offers no checksum or signature","where the built-in automatic update looks for new versions now that the project is archived","whether crash reports leave the TV box, and to whom","whether it still works with current Jellyfin server versions","which TV boxes and Android TV versions it runs on, since the README names none"]},"zh-Hans":{"summaryShort":"一款基于官方 Jellyfin 客户端重新设计的安卓电视应用,用来观看你自己服务器上的电影和剧集。","summary":"DUNE 是 Jellyfin 安卓电视应用的非官方改版副本:全新的主屏幕、多种主题(包括一个专为 OLED 屏幕设计的主题)、语音搜索、字幕控制,还可以选择与官方应用同时安装。它只会显示你自己搭建的 Jellyfin 媒体服务器上已有的内容。该仓库已被归档,不再有新的改动,README 只记录了如何用 Android Studio、Android SDK 35 和 OpenJDK 21 自行编译文件。它不能在电脑上运行。","easeRationale":"安卓电视是它唯一能运行的系统,在那里它得到 D 级:README 中唯一记录的途径是 \"Building from Source\",需要使用 Android Studio Giraffe、Android SDK API 35 和 OpenJDK 21,以及像 ./gradlew assembleEnhancedRelease 这样的命令,而且你必须已经拥有一台 Jellyfin 服务器。电脑没有评级;不存在桌面版本。","costNote":"免费,采用 GPL-2.0 许可;README 中有一个指向 coff.ee/sam42 的 \"Support Me - Buy a Coffee\" 徽章。","notVerified":["该仓库中是否有任何现成文件与源代码一致,因为 README 没有提供校验和或签名","既然项目已被归档,内置的自动更新现在从哪里查找新版本","崩溃报告是否会离开电视盒子,发送给谁","它是否仍能在当前版本的 Jellyfin 服务器上使用","它支持哪些电视盒子和安卓电视版本,因为 README 没有提到任何一个"]}}},{"id":"R_kgDOQxQRhQ","slug":"samuel-vitorino__sopro","name":"Sopro","owner":"samuel-vitorino","fullName":"samuel-vitorino/sopro","type":"needs_setup","status":"active","grade":"A","easeByPlatform":{"windows":"C","mac":"C","linux":"C","web":"A"},"risk":"attention","signals":[{"code":"copies_a_voice_and_no_watermark","severity":"attention","evidence":"Main features:\"Zero-shot voice cloning from 5-20 seconds of reference audio\"。Disclaimers:\"We did not add watermarking: with an open-source inference pipeline it would be trivial to remove... Please use the model for good: do not impersonate people.\"因此,生成的音频没有任何标记能表明它是由机器制作的。","source":"ai","title":"复刻声音且没有水印"},{"code":"model_downloads_on_first_use","severity":"info","evidence":"\"The model downloads on first use and stays in the Hugging Face cache\",来自 huggingface.co/samuel-vitorino/sopro-v2-turbo;仓库本身不包含任何模型文件。","source":"ai","title":"模型在首次使用时下载"},{"code":"no_github_releases_only_package_registries","severity":"info","evidence":"项目资料显示 \"GitHub releases: 0\";版本只通过 `pip install -U sopro`(pyproject.toml 声明版本为 2.2.0)和 `npm install @soprotts/onnx-web@latest` 发布,仓库中没有校验和。","source":"ai","title":"没有 github release,只有包注册源"},{"code":"author_documents_quality_limits","severity":"info","evidence":"Disclaimers 说文本前端是 \"deliberately minimal\",所以 \"some abbreviations, numbers, and symbols may not be pronounced correctly\"(要写成 `one plus two`),混合语言的句子 \"can be mispronounced\",而流式路径 \"not bit-exact with the offline path\"。","source":"ai","title":"作者记录了质量限制"},{"code":"training_code_not_released","severity":"info","evidence":"Disclaimers:\"We are not planning to release the training code in the near future due to its complexity.\"该仓库只包含运行模型的代码,列出的训练数据有 Emilia YODAS、LibriTTS-R 和 FalAR。","source":"ai","title":"训练代码未公开"}],"platforms":["windows","mac","linux","web"],"tools":["python"],"buildTools":["python","gpu_any"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":983,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":5,"growthDays":6,"history":[{"date":"2026-11-08","stars":982},{"date":"2026-11-09","stars":982},{"date":"2026-11-10","stars":983}],"firstPublishedAt":"2026-11-09T23:02:44.453Z","commitSha":"7bfcf9a0539d274f6593959a21da948f506ceee1","verifiedAt":"2026-11-09T22:50:41.228Z","lastCommitAt":"2026-09-01T03:34:07Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Reads your text aloud in a voice copied from a 5 to 20 second clip, on a normal computer or straight in the browser.","summary":"Sopro is a small text-to-speech model (120 million parameters) that speaks English, European Portuguese, French and German and copies a voice from 5 to 20 seconds of reference audio. The fastest way to try it is the in-browser demo page, where nothing is installed. On your own machine, `uvx --from sopro soprotts serve` opens a local page at http://localhost:7860; it also works from the command line or from Python. It runs on a laptop processor, uses a graphics card when there is one, and downloads the model on first use.","easeRationale":"The browser demo (samuel-vitorino.github.io/sopro) is an A: you open the page and the model loads by itself, though the README warns that on phones the quantized version is weaker and low-memory devices \"may crash\". On Windows, macOS and Linux it is a C: install Python or uv, run `uvx --from sopro soprotts serve` or `pip install -U sopro`, and wait for the model download.","costNote":"Free software under Apache-2.0 with free model files; the cost is your own machine plus roughly 0.9 to 4.3 GB of disk for the downloaded model.","notVerified":["how close the cloned voice sounds to the original recording","whether the speed figures (about 300 ms to first audio on a laptop processor) hold on your machine","how much disk the model really takes after the first download","whether the browser demo works on your phone or runs out of memory","how well the three languages other than English are pronounced"]},"zh-Hans":{"summaryShort":"用从 5 到 20 秒片段中复制的声音朗读你的文本,可以在普通电脑上使用,也可以直接在浏览器中使用。","summary":"Sopro 是一个小型文本转语音模型(1.2 亿参数),支持英语、欧洲葡萄牙语、法语和德语,并能根据 5 到 20 秒的参考音频复刻一个声音。最快的体验方式是浏览器内的演示页面,不需要安装任何东西。在你自己的电脑上,`uvx --from sopro soprotts serve` 会在 http://localhost:7860 打开一个本地页面;它也可以通过命令行或 Python 使用。它可以在笔记本电脑的处理器上运行,有显卡时会使用显卡,并在首次使用时下载模型。","easeRationale":"浏览器演示版(samuel-vitorino.github.io/sopro)评级为 A:打开页面后模型会自动加载,不过 README 提醒,在手机上量化版本效果较弱,内存较小的设备 \"may crash\"。在 Windows、macOS 和 Linux 上评级为 C:安装 Python 或 uv,运行 `uvx --from sopro soprotts serve` 或 `pip install -U sopro`,然后等待模型下载。","costNote":"免费软件,采用 Apache-2.0 许可,模型文件也是免费的;成本是你自己的设备,再加上下载模型所需的大约 0.9 到 4.3 GB 磁盘空间。","notVerified":["克隆出的声音与原始录音有多相似","速度数据(在笔记本电脑处理器上大约 300 毫秒生成第一段音频)在你的设备上是否成立","首次下载之后,模型实际占用多少磁盘空间","浏览器演示版在你的手机上是否能用,还是会出现内存不足","除英语之外的三种语言发音效果如何"]}}},{"id":"R_kgDOOdFkDA","slug":"sandai-org__magi-1","name":"MAGI-1","owner":"SandAI-org","fullName":"SandAI-org/MAGI-1","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; flash-attn: GHSA-7g5w-pq96-8c5w, PYSEC-2026-2477; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8qvm-5x2c-j2w7, PYSEC-2026-1805; sentencepiece: GHSA-38vq-g6vr-w8wf, PYSEC-2026-1909; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74","source":"check","title":"已知漏洞"},{"code":"needs_expensive_graphics_hardware","severity":"attention","evidence":"Model Zoo 建议 24B 模型使用 \"H100/H800 x 8\",4.5B 模型使用 \"RTX 4090 x 1\";说明中写道,4.5B 模型 \"any machine with at least 24GB of GPU memory is sufficient\",量化版本并将 window_size 设为 1 时只需要 12 GB。","source":"ai","title":"需要昂贵的显卡硬件"},{"code":"documented_docker_run_is_privileged","severity":"attention","evidence":"推荐的命令是 `docker run -it --gpus all --privileged --shm-size=32g --net=host --ipc=host --ulimit memlock=-1 ... sandai/magi:latest`:该容器以特权模式运行,使用主机网络和主机的共享内存。","source":"ai","title":"文档中的 docker 运行命令使用特权模式"},{"code":"installs_from_an_extra_package_index","severity":"info","evidence":"requirements.txt 中的一行:`flashinfer-python==0.2.0.post2 --extra-index-url https://flashinfer.ai/whl/cu124/torch2.4/`,因此 pip 还会从 flashinfer.ai 拉取一个包,而不只是从 PyPI;MagiAttention 的源代码路径是 `git clone git@github.com:SandAI-org/MagiAttention.git`,这需要在 GitHub 上配置 SSH 密钥。","source":"ai","title":"从额外的包索引安装"},{"code":"model_files_come_from_hugging_face","severity":"info","evidence":"Model Zoo 把每个检查点(T5、VAE、4.5B 和 24B、distill 和 fp8_quant)都链接到了 huggingface.co/sand-ai/MAGI-1;该仓库本身只包含代码,因此权重及其许可条款需要单独下载。","source":"ai","title":"模型文件来自 hugging face"},{"code":"no_releases_and_no_generation_mark","severity":"info","evidence":"项目资料显示 \"GitHub releases: 0\"(变更只在 README 的新闻列表中公布,最近一次是 2026-06-17),README 中没有提到水印或生成视频上的任何标记。","source":"ai","title":"没有 release,生成内容也没有标记"}],"platforms":["linux"],"tools":["python","docker","gpu_nvidia","ffmpeg"],"buildTools":["python","ffmpeg"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"ltx-studio","name":"LTX Studio","confirmed":true},{"slug":"kling","name":"Kling","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":3805,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":7,"growthDays":6,"history":[{"date":"2026-11-06","stars":3803},{"date":"2026-11-07","stars":3804},{"date":"2026-11-08","stars":3804},{"date":"2026-11-09","stars":3805},{"date":"2026-11-10","stars":3805}],"firstPublishedAt":"2026-11-09T23:08:44.868Z","commitSha":"0fcefdef8ce2df37a3b8890979433c06eb003328","verifiedAt":"2026-11-09T23:05:52.960Z","lastCommitAt":"2026-06-17T07:06:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes videos from a text prompt, a picture or the start of a clip, using open model files on strong NVIDIA cards.","summary":"MAGI-1 is Sand AI's video generation model with the code to run it and the model files published on Hugging Face. From the command line you pick a mode (text to video, image to video, or continuing an existing clip), write a prompt and get a video file; custom nodes also let it run inside ComfyUI. It is for people with serious hardware: the small 4.5B model asks for an NVIDIA card with 24 GB (12 GB in the quantized version) and the 24B model for eight H100 cards. The authors also offer MAGI-1 to use online at sand.ai.","easeRationale":"Linux is the only documented system and it is a D: the recommended path is `docker pull sandai/magi:latest`, otherwise conda with Python 3.10.12, PyTorch 2.4 with CUDA 12.4, ffmpeg and a build of MagiAttention from source, then editing a JSON config and running example/4.5B/run.sh. There are no instructions for Windows or macOS.","costNote":"Code and model files are free under Apache-2.0; the real cost is the graphics hardware (or rented machines) needed to run them.","notVerified":["what the online service at sand.ai costs, since the README only says \"directly create magic with MAGI-1 [here](http://sand.ai)\" and mentions no price or plan","how much disk the model files really take, since they are downloaded from Hugging Face and are not part of the repository","how long one video takes on a single RTX 4090 and what resolution and length are usable there","whether the published videos and evaluation scores can be reproduced at home","what license the model files on Hugging Face carry, which can differ from the Apache-2.0 code","whether the ComfyUI nodes still work with current ComfyUI versions"]},"zh-Hans":{"summaryShort":"可以根据文字提示、一张图片或一段视频的开头生成视频,使用开放的模型文件,需要强力的 NVIDIA 显卡。","summary":"MAGI-1 是 Sand AI 的视频生成模型,提供运行代码,模型文件发布在 Hugging Face 上。在命令行中选择一种模式(文字生成视频、图片生成视频,或续写已有的片段),写一段提示词即可得到一个视频文件;自定义节点也能让它在 ComfyUI 中运行。它面向拥有强力硬件的用户:较小的 4.5B 模型需要一张显存 24 GB 的 NVIDIA 显卡(量化版本需要 12 GB),24B 模型则需要八张 H100 显卡。作者还提供了在线使用 MAGI-1 的方式,网址是 sand.ai。","easeRationale":"Linux 是唯一有文档说明的系统,评级为 D:推荐的方式是 `docker pull sandai/magi:latest`,否则就要用 conda 配合 Python 3.10.12、搭配 CUDA 12.4 的 PyTorch 2.4、ffmpeg,以及从源代码编译 MagiAttention,然后编辑一个 JSON 配置文件并运行 example/4.5B/run.sh。没有针对 Windows 或 macOS 的说明。","costNote":"代码和模型文件在 Apache-2.0 许可下免费提供;真正的成本是运行它们所需的显卡硬件(或租用的机器)。","notVerified":["sand.ai 上的在线服务收费多少,因为 README 只说了 \"directly create magic with MAGI-1 [here](http://sand.ai)\",没有提到任何价格或套餐","模型文件实际占用多少磁盘空间,因为它们是从 Hugging Face 下载的,并不属于该仓库","在单张 RTX 4090 上生成一个视频需要多长时间,以及在那种配置下可用的分辨率和时长是多少","发布的视频和评测分数能否在自己家里复现","Hugging Face 上的模型文件采用什么许可,这可能和 Apache-2.0 的代码许可不同","ComfyUI 节点是否仍能在当前版本的 ComfyUI 中使用"]}}},{"id":"R_kgDONaGk6Q","slug":"sashenjayathilaka__saas-instagram-dm-automations","name":"Instagram DM Automations","owner":"SashenJayathilaka","fullName":"SashenJayathilaka/SAAS-Instagram-DM-Automations","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget manifest 都没有检测到任何许可证","source":"check","title":"没有许可证"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):next: GHSA-2464-fhr4-w9xw, GHSA-2xp9-vwfh-vxw4, GHSA-36qx-fr4f-26g5","source":"check","title":"已知漏洞"},{"code":"install_instructions_belong_to_another_project","severity":"attention","evidence":"\"Run Locally\" 部分克隆了 `https://github.com/SashenJayathilaka/Miro-Clone.git` 并执行 `cd Miro-Clone`;Installation 部分从 `npx create-next-app@latest my-project` 开始;文中写道 \"This project was bootstrapped with Create React App\";Prerequisites 中 \"Sign up for a Stripe\" 的链接指向了 liveblocks.io。按照 README 原样操作并不能把这个项目配置起来。","source":"ai","title":"安装说明属于另一个项目"},{"code":"automated_dm_outreach_against_platform_rules","severity":"attention","evidence":"项目描述中宣传的是 \"Instagram comment automations to automate dm outreach\" 和 \"Instagram DM automations for keywords\"。大规模发送未经请求的自动消息受到 Meta 平台条款的限制,这样做的应用可能会被封锁访问权限,或连接的账号被封锁。","source":"ai","title":"自动私信外联违反平台规则"},{"code":"handles_other_peoples_instagram_access","severity":"attention","evidence":".env.example 中包含 INSTAGRAM_CLIENT_ID、INSTAGRAM_CLIENT_SECRET、INSTAGRAM_EMBEDDED_OAUTH_URL 和 INSTAGRAM_TOKEN_URL,描述开头是 \"Instagram API integration with OAuth\":运营这个网站的人会把其他人的 Instagram 访问令牌存放在自己的数据库里,仓库中既没有许可证,也没有任何隐私说明文本。","source":"ai","title":"处理他人的 instagram 访问权限"},{"code":"demo_link_does_not_match_the_project","severity":"info","evidence":"主页、\"View Demo\" 链接和 \"LIVE DEMO\" 标题都指向 https://slideproject.vercel.app,这个名字和 Instagram 自动化毫无关系,而 package.json 中的包名是 \"ig-automations\",版本号为 0.1.0。","source":"ai","title":"演示链接与项目不匹配"},{"code":"maintained_badge_but_no_recent_work","severity":"info","evidence":"README 显示了一个 \"Maintained-Yes\" 徽章,而最后一次 push 是在 2025-03-19,距此次检查已有 600 天,且没有任何 release。","source":"ai","title":"有维护中徽章,但没有最近的工作"},{"code":"instagram_app_review_not_documented","severity":"info","evidence":"Topics 中列出了 \"ngrok\",.env.example 把 INSTAGRAM_BASE_URL 指向 https://graph.instagram.com,但 README 从未提到创建 Meta 应用、所需的权限,或者 Meta 的审核——而这项审核是此类应用能够向真实账号发送消息之前必须完成的。","source":"ai","title":"没有记录 instagram 应用审核流程"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Instagram app (Meta) client ID and secret","Clerk (user logins)","Neon or another PostgreSQL database","OpenAI","Stripe"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[],"stars":54,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":1,"growthDays":6,"history":[{"date":"2026-11-05","stars":53},{"date":"2026-11-06","stars":53},{"date":"2026-11-07","stars":53},{"date":"2026-11-08","stars":53},{"date":"2026-11-09","stars":53},{"date":"2026-11-10","stars":54}],"firstPublishedAt":"2026-11-09T23:02:44.471Z","commitSha":"c0f27c20870a5c602ecab751bfdcedd97cbc6876","verifiedAt":"2026-11-09T22:52:32.622Z","lastCommitAt":"2025-03-19T02:12:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Source code for a website that answers Instagram comments and messages automatically and charges subscribers.","summary":"This is the code of a subscription website for Instagram automation: people connect their Instagram account, build rules that reply to comments and send direct messages when a keyword appears or someone answers a story, and see numbers on a dashboard. It is a developer project, not a finished product: it has no license, and to run it you need Node.js plus paid accounts at Clerk (logins), Neon (database), Stripe (charging) and OpenAI, with an Instagram app of your own. The README's own install steps point to a different repository.","easeRationale":"E on Windows, macOS and Linux alike: there is no installer and no working walkthrough. The \"Run Locally\" block tells you to clone `https://github.com/SashenJayathilaka/Miro-Clone.git` and `cd Miro-Clone`, fifteen environment variables must be filled in by hand, and nothing explains how to create the Instagram app or prepare the database.","costNote":"The code carries no license and no price. Running it means paying for OpenAI use and for the Clerk, Neon and Stripe accounts it is built on; STRIPE_SUBSCRIPTION_PRICE_ID shows it is meant to charge your own subscribers.","notVerified":["whether the code runs at all, since the README's steps are from another project","whether Meta approves an app with these permissions for messaging","what the Clerk, Neon, Stripe and OpenAI accounts cost once there is real traffic","whether the site at slideproject.vercel.app is built from this code","what the project does with the Instagram tokens and messages it stores, as there is no documentation about it"]},"zh-Hans":{"summaryShort":"一个网站的源代码,能自动回复 Instagram 评论和消息,并向订阅用户收费。","summary":"这是一个 Instagram 自动化订阅网站的代码:用户连接自己的 Instagram 账号,设置规则,在出现某个关键词或有人回复故事时自动回复评论、发送私信,并在仪表盘上查看数据。这是一个开发者项目,不是成品:它没有许可证,运行它需要 Node.js,还需要 Clerk(登录)、Neon(数据库)、Stripe(收费)和 OpenAI 的付费账户,以及你自己的 Instagram 应用。README 自己的安装步骤指向了另一个仓库。","easeRationale":"在 Windows、macOS 和 Linux 上都是 E 级:没有安装程序,也没有能跑通的操作说明。\"Run Locally\" 部分让你克隆 `https://github.com/SashenJayathilaka/Miro-Clone.git` 并执行 `cd Miro-Clone`,还要手动填写十五个环境变量,而且没有任何说明讲解如何创建 Instagram 应用或准备数据库。","costNote":"该代码没有许可证,也没有标价。运行它意味着要为 OpenAI 的使用付费,还要支付它所依赖的 Clerk、Neon 和 Stripe 账户的费用;STRIPE_SUBSCRIPTION_PRICE_ID 表明它是用来向你自己的订阅用户收费的。","notVerified":["这份代码是否真的能运行,因为 README 的步骤来自另一个项目","Meta 是否会批准一个拥有这些消息权限的应用","一旦有真实流量,Clerk、Neon、Stripe 和 OpenAI 的账户要花多少钱","slideproject.vercel.app 这个网站是否就是用这份代码搭建的","该项目如何处理它存储的 Instagram 令牌和消息,因为没有相关文档说明"]}}},{"id":"R_kgDOH9df7A","slug":"satnaing__astro-paper","name":"AstroPaper","owner":"satnaing","fullName":"satnaing/astro-paper","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"windows":"E","mac":"E","linux":"E"},"risk":"low","signals":[{"code":"needs_a_very_recent_node","severity":"info","evidence":"package.json 声明了 `\"engines\": { \"node\": \">=22.23.4\" }`,compose.yaml 设置了 `pnpm runtime set node 24 -g`,所以较旧的 Node 安装无法构建这个网站。","source":"ai","title":"需要非常新的 node 版本"},{"code":"no_admin_screen_and_hosting_is_yours","severity":"info","evidence":"文章以文件形式存放在 `src/content/posts/` 中,设置在 `astro-paper.config.ts` 里;Tech Stack 中把 Cloudflare Pages 列为 Deployment。没有编辑器、登录或评论系统:构建产生的是静态文件,需要你自己把它们发布上线。","source":"ai","title":"没有管理界面,托管由你自己负责"},{"code":"accessibility_tested_on_two_screen_readers","severity":"info","evidence":"README 中的说明:\"I've tested screen-reader accessibility of AstroPaper using VoiceOver on Mac and TalkBack on Android. I couldn't test all other screen-readers out there.\"","source":"ai","title":"在两款屏幕阅读器上测试过无障碍功能"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node","docker"],"disk":{"minMB":150,"maxMB":650,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","productivity"],"alternativeTo":[{"slug":"squarespace","name":"Squarespace","confirmed":true},{"slug":"wix","name":"Wix","confirmed":true}],"declaredAlternatives":[],"stars":5198,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":21,"growthDays":6,"history":[{"date":"2026-11-06","stars":5183},{"date":"2026-11-07","stars":5187},{"date":"2026-11-08","stars":5194},{"date":"2026-11-09","stars":5196},{"date":"2026-11-10","stars":5198}],"firstPublishedAt":"2026-11-09T23:02:44.481Z","commitSha":"6f61e2dc0e45bbcefc05f4055f293cd7f81bcb03","verifiedAt":"2026-11-09T22:53:15.544Z","lastCommitAt":"2026-11-06T12:48:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A ready-made design for a personal blog: you write posts as text files and publish a fast, light-and-dark website.","summary":"AstroPaper is a blog theme for the Astro website builder. You start a project from the template, change a configuration file, and write each post as a markdown file in src/content/posts; the site you get has light and dark mode, a search that works without a server (Pagefind), pagination, draft posts, RSS, sitemap, automatic preview images and translations. It is work for someone comfortable with a terminal: Node 22 or newer, pnpm commands, and your own hosting such as Cloudflare Pages. The project is active, MIT licensed and at version 7.1.0.","easeRationale":"This is a website template, not a program to open, so it is E on Windows, macOS and Linux alike: `pnpm create astro@latest --template satnaing/astro-paper`, then `pnpm install` and `pnpm dev` at localhost:4321, editing TypeScript config files and markdown by hand, and arranging hosting yourself.","costNote":"Free under the MIT License; you pay only for hosting and a domain if you want one (the author's own site runs on Cloudflare Pages, which has a free tier).","notVerified":["whether the Lighthouse score in the README holds once you add your own images and posts","how long the build and the Pagefind search indexing take on a large blog","how hard it is to bring later versions of the theme into a site you have already customized","whether the dynamic preview images work on every hosting service, since they are generated at build time"]},"zh-Hans":{"summaryShort":"一套现成的个人博客设计:把文章写成文本文件,就能发布一个支持浅色和深色模式的快速网站。","summary":"AstroPaper 是面向 Astro 网站构建器的一个博客主题。你从模板开始一个项目,修改一个配置文件,再把每篇文章写成 src/content/posts 目录下的一个 markdown 文件;得到的网站支持浅色和深色模式、不需要服务器的搜索(Pagefind)、分页、草稿文章、RSS、站点地图、自动生成的预览图,以及多语言翻译。这项工作适合熟悉终端的人:需要 Node 22 或更新版本、pnpm 命令,以及你自己的托管服务,例如 Cloudflare Pages。该项目正在积极维护,采用 MIT 许可,目前版本为 7.1.0。","easeRationale":"这是一个网站模板,不是一个可以打开的程序,所以在 Windows、macOS 和 Linux 上都是 E 级:运行 `pnpm create astro@latest --template satnaing/astro-paper`,然后执行 `pnpm install` 和 `pnpm dev`,在 localhost:4321 查看效果,还要手动编辑 TypeScript 配置文件和 markdown 文件,并自己安排托管。","costNote":"采用 MIT 许可,免费使用;只有在你想要域名和托管服务时才需要付费(作者自己的网站运行在 Cloudflare Pages 上,该服务有免费套餐)。","notVerified":["一旦加入自己的图片和文章,README 中的 Lighthouse 分数是否还能保持","在一个很大的博客中,构建和 Pagefind 搜索索引需要多长时间","把主题的后续版本合并到一个你已经自定义过的网站中有多难","动态预览图是否能在所有托管服务上正常工作,因为它们是在构建时生成的"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzNTMzMzczOA==","slug":"scribusproject__scribus","name":"Scribus","owner":"scribusproject","fullName":"scribusproject/scribus","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"unofficial_manual_mirror","severity":"info","evidence":"README.md：\"This mirror is updated manually and is not supported by the Scribus team\"；官方代码托管在 svn://scribus.net 的 Subversion 上。GitHub 的数据（0 个发布版本，未检测到许可证）并不能说明真实项目的情况。","source":"ai","title":"非官方的手动镜像"},{"code":"development_version_in_repo","severity":"info","evidence":"README：\"the 1.7.x series is not an officially stable version. Instead, all 1.7.x versions are development candidates of the future stable Scribus 1.8.0! We strongly recommend using version 1.6.0\"。","source":"ai","title":"仓库中是开发版本"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0-or-later","category":"copyleft","detectedFrom":"file"},"categories":["design_cad","office"],"alternativeTo":[{"slug":"indesign","name":"InDesign","confirmed":true},{"slug":"quarkxpress","name":"QuarkXPress","confirmed":true},{"slug":"microsoft-publisher","name":"Publisher","confirmed":true}],"declaredAlternatives":[],"stars":657,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":5,"growthDays":7,"history":[{"date":"2026-10-25","stars":643},{"date":"2026-10-26","stars":643},{"date":"2026-10-27","stars":644},{"date":"2026-10-28","stars":644},{"date":"2026-10-29","stars":646},{"date":"2026-10-30","stars":647},{"date":"2026-10-31","stars":648},{"date":"2026-11-01","stars":648},{"date":"2026-11-02","stars":650},{"date":"2026-11-03","stars":652},{"date":"2026-11-04","stars":653},{"date":"2026-11-05","stars":654},{"date":"2026-11-06","stars":656},{"date":"2026-11-07","stars":656},{"date":"2026-11-08","stars":656},{"date":"2026-11-09","stars":657},{"date":"2026-11-10","stars":657}],"firstPublishedAt":"2026-10-25T23:19:27.407Z","commitSha":"52dd809058b7c17a2df11ef6df9eef23bb0d18cd","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-08T20:38:36Z","isMirror":true,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Lay out magazines, newsletters, brochures, posters and books for printing or PDF.","summary":"Scribus is a desktop publishing program for page layout: magazines, newsletters, brochures, posters and books, ready for print or PDF. It handles right-to-left and Indian scripts, OpenType font features and many languages. This GitHub page is an unofficial, manually updated copy of the source code; the program itself is distributed through scribus.net, and its license file mentions a Windows version. The README says the 1.7.x versions are development versions and recommends the stable 1.6 series.","easeRationale":"This mirror has no release files, but the program is distributed from scribus.net and COPYING refers to \"the Windows version of Scribus\", so a normal user downloads it there instead of building it.","costNote":"Free software under the GNU GPL version 2 or later, according to the COPYING file.","notVerified":["Download size and disk space of the Windows installer, which the checks could not measure","Which version scribus.net currently offers for Windows","How well it opens InDesign or Publisher files"]},"zh-Hans":{"summaryShort":"排版杂志、简报、宣传册、海报和书籍，可用于印刷或输出 PDF。","summary":"Scribus 是一款用于页面排版的桌面出版程序：杂志、简报、宣传册、海报和书籍，可直接用于印刷或生成 PDF。它支持从右向左书写的文字和印度系文字、OpenType 字体功能以及多种语言。这个 GitHub 页面是源代码的非官方副本，由人工手动更新；程序本身通过 scribus.net 分发，其许可证文件中提到了 Windows 版本。README 说 1.7.x 系列是开发版，并推荐稳定的 1.6 系列。","easeRationale":"这个镜像没有发布文件，但程序是从 scribus.net 分发的，COPYING 文件提到了“Windows 版 Scribus”，所以普通用户应在那里下载，而不是自己编译。","costNote":"根据 COPYING 文件，这是自由软件，采用 GNU GPL 第 2 版或更高版本。","notVerified":["Windows 安装程序的下载大小和磁盘占用，检查时无法测量","scribus.net 目前为 Windows 提供哪个版本","它打开 InDesign 或 Publisher 文件的效果如何"]}}},{"id":"R_kgDOVG4bQg","slug":"seedds__nxnative","name":"NXNative","owner":"seedds","fullName":"seedds/nxnative","type":"needs_setup","status":"alert","grade":"E","easeByPlatform":{"mac":"E"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"14天内获得7849个star","source":"check","title":"新仓库却有大量star"},{"code":"console_keys_and_game_decryption","severity":"high","evidence":"主要用途依赖于任天堂主机的解密密钥：README\"Bring your game (its NSP files) and your `prod.keys`\"；构建命令\"build_ipa_from_nsp.py ~/Games/MarioWonder -k ~/keys/prod.keys\"；LibHac\"reads NSPs\"。prod.keys是Switch用来解密游戏的密钥，只能从改装过的主机中提取。许多国家的法律限制提取此类密钥和解密游戏。","source":"ai","title":"主机密钥与游戏解密"},{"code":"output_contains_full_game","severity":"attention","evidence":"每个IPA都包含完整的游戏（README列出的大小例如\"Mario Kart 8 Deluxe ~11.9 GB\"和\"Zelda: Tears of the Kingdom ~17 GB\"）。分享这些文件等同于传播受版权保护的游戏。","source":"ai","title":"生成结果包含完整游戏"},{"code":"sideloading_required","severity":"attention","evidence":"README第4步：\"Install the IPA with your sideloading tool (AltStore, Sideloadly, LiveContainer, …)\"，并使用\"Increased Memory Limit\"权限对其签名。应用是在App Store之外安装的。","source":"ai","title":"需要使用越狱/旁加载方式安装"},{"code":"strict_device_requirements","severity":"info","evidence":"README：\"Needs iOS / iPadOS 26 or later on an iPhone 15 Pro or newer (A17 Pro chip or later) or an iPad with an M1 chip or later\"；对于不在列表中的游戏，脚本\"still tries its best to pack an IPA, guessing the game's settings; such an IPA may not start\"。","source":"ai","title":"设备要求非常严格"},{"code":"unaffiliated_with_the_console_maker","severity":"info","evidence":"README：\"Nothing from any game is in this repository, and the project is not affiliated with Nintendo or any game publisher.\"该仓库只提供转换代码；游戏内容由你自己提供。","source":"ai","title":"与主机厂商无关"}],"platforms":["mac"],"tools":["python","dotnet","other"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":14498,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":2624,"growthDays":7,"history":[{"date":"2026-10-27","stars":7822},{"date":"2026-10-28","stars":9052},{"date":"2026-10-29","stars":9734},{"date":"2026-10-30","stars":10238},{"date":"2026-10-31","stars":10639},{"date":"2026-11-01","stars":11000},{"date":"2026-11-02","stars":11464},{"date":"2026-11-03","stars":11874},{"date":"2026-11-04","stars":12129},{"date":"2026-11-05","stars":12464},{"date":"2026-11-06","stars":12905},{"date":"2026-11-07","stars":13323},{"date":"2026-11-08","stars":13694},{"date":"2026-11-09","stars":14098},{"date":"2026-11-10","stars":14498}],"firstPublishedAt":"2026-10-27T02:46:39.239Z","commitSha":"712eb8167230a02023cfa57d10bb07b748b8933f","verifiedAt":"2026-10-28T01:42:59.433Z","lastCommitAt":"2026-11-10T23:12:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns Nintendo Switch game files and console keys into apps you sideload onto an iPhone, iPad or Apple silicon Mac.","summary":"A set of scripts that converts a Nintendo Switch game dump (NSP or XCI files) into an iPhone/iPad app (IPA) or a Mac app, so the game's own code runs on the Apple chip with graphics rebuilt on Metal. You must supply the game files and your Switch's prod.keys. It needs a Mac with Xcode, Homebrew, the .NET SDK and Python, an iPhone 15 Pro or M1 iPad on iOS 26 or later, and a sideloading tool such as AltStore. The first build takes 30–60 minutes and needs about twice the game's size plus 10 GB free. Windows is not supported.","easeRationale":"The Mac is the only system that can do this and it is still an E: there are no releases, so you clone the repository, install Xcode 27.1 with the Metal toolchain, Homebrew packages, the .NET SDK and Python modules, run a build script with your own game dump and prod.keys, then sideload the result onto the phone. Windows and Linux are not supported at all.","notVerified":["Whether each listed game actually runs well on a given iPhone, iPad or Mac","Whether the build works on a Mac other than the authors' setup and with game versions they did not test","How long sideloaded apps keep working before they must be signed again","Real battery use and heat on the phone while playing","The legal status of extracting prod.keys and converting your own games in your country"]},"zh-Hans":{"summaryShort":"把Nintendo Switch的游戏文件和主机密钥转换成App,侧载安装到iPhone、iPad或搭载Apple芯片的Mac上。","summary":"这是一套脚本,可以把Nintendo Switch的游戏导出文件(NSP或XCI文件)转换成iPhone/iPad的App(IPA)或Mac应用,让游戏自身的代码运行在Apple芯片上,图形部分用Metal重建。你需要自己提供游戏文件和你Switch的prod.keys。需要一台装有Xcode、Homebrew、.NET SDK和Python的Mac,一台运行iOS 26或更高版本的iPhone 15 Pro或M1 iPad,以及像AltStore这样的侧载工具。首次编译需要30到60分钟,所需空间大约是游戏大小的两倍再加10 GB。不支持Windows。","easeRationale":"Mac是唯一能做到这件事的系统,但评级仍然是E:没有发布版本,你需要克隆代码仓库,安装带有Metal工具链的Xcode 27.1、Homebrew软件包、.NET SDK和Python模块,用你自己的游戏导出文件和prod.keys运行一个编译脚本,然后把结果侧载到手机上。完全不支持Windows和Linux。","notVerified":["列表中的每款游戏是否真的能在具体的iPhone、iPad或Mac上良好运行","在作者使用的配置之外的Mac上,以及作者没有测试过的游戏版本上,编译是否也能成功","侧载安装的App在需要重新签名之前能持续使用多久","玩游戏时手机的实际电量消耗和发热情况","在你所在的国家,提取prod.keys并转换自己拥有的游戏,法律上处于什么状态"]}}},{"id":"R_kgDOJ4QHUg","slug":"serpapi__clauneck","name":"ClauNeck","owner":"serpapi","fullName":"serpapi/clauneck","type":"dev_tool","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布于 2024-03-19（965 天前）","source":"check","title":"版本更新停滞"},{"code":"collects_personal_contacts_for_cold_email","severity":"attention","evidence":"该仓库的描述是 \"A tool for scraping emails, social media accounts, and much more information from websites\"，标签包括 email-scraper、email-marketing 和 email-extract-with-proxy，README 还链接了自己的博客文章 \"Cold Email Marketing with Open-Source Email Extractor\" 作为使用场景。以这种方式收集地址并在未经同意的情况下发送邮件，在欧盟、巴西及其他国家都受数据保护和反垃圾邮件规则的限制。","source":"ai","title":"为冷邮件收集个人联系方式"},{"code":"recommends_public_proxy_servers","severity":"attention","evidence":"先决条件中写道：\"List of Proxies in a text document (You may use public proxies. Only HTTP proxies are accepted.)\"，进阶部分展示了 `http://username:password@ip:port` 这样的行。所有抓取流量，以及这些行中可能包含的凭据，都会经过陌生人运行的服务器，且走的是未加密的 HTTP。","source":"ai","title":"建议使用公共代理服务器"},{"code":"built_around_google_cache_addresses","severity":"info","evidence":"README 要求提供的 URL 列表包含 \"Google web cache links that start with https://webcache.googleusercontent.com\"，并说明如何手动找到缓存链接；该工具读取的是每个页面的缓存副本，而不是网站本身。","source":"ai","title":"依赖 Google 缓存地址构建"},{"code":"tool_funnels_into_its_own_paid_service","severity":"info","evidence":"该 gem 由 SerpApi 发布（LICENSE：\"Copyright (c) 2023 SerpApi\"），README 反复链接到 serpapi.com/search-api 以及 \"Register to Claim Free Credits\"；没有这个密钥，唯一的选择就是使用你自己提供的 URL 列表。","source":"ai","title":"工具把用户导向自己的付费服务"},{"code":"no_activity_and_very_early_versions","severity":"info","evidence":"仅有三次发布（v0.1.0 于 2023-07-06，随后 v0.0.2 和 v0.0.3 于 2024-03-19），最后一次提交在 2024-03-19，距本次检查 965 天，README 也没有测试或更新日志部分。","source":"ai","title":"没有活跃动态且版本非常早期"},{"code":"collects_other_people_s_contact_data_in_bulk","severity":"high","evidence":"描述：\"A tool for scraping emails, social media accounts, and much more information from websites using Google Search Results\"；标签：\"email-scraper\"、\"email-marketing\"、\"email-extract-with-proxy\"。README 把 \"Cold Email Marketing with Open-Source Email Extractor\" 链接为 \"A Blog Post about the usecase of the tool\"，示例搜索词是 \"site:*.ai AND inurl:/contact OR inurl:/contact-us\"，结果是一个 CSV 文件，包含每个网站的电子邮件地址，以及 Instagram、Facebook、Twitter、Tiktok、Youtube、Github 和 Medium 的账号。欧盟的 GDPR、巴西的 LGPD 等数据保护规则，以及其他国家的反垃圾邮件规则，都限制未经同意大量收集他人联系方式并发送信息的行为。README 还提到页面是通过 \"public proxies\" 和 \"Google web cache links that start with https://webcache.googleusercontent.com\" 读取的，而不是直接访问网站本身。","source":"ai","title":"批量收集他人的联系方式"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["SerpApi (Google Search API)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"hunter-io","name":"Hunter.io","confirmed":true},{"slug":"phantombuster","name":"PhantomBuster","confirmed":true},{"slug":"bright-data","name":"Bright Data","confirmed":true}],"declaredAlternatives":[],"stars":206,"downloads":7,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":206},{"date":"2026-11-06","stars":206},{"date":"2026-11-07","stars":206},{"date":"2026-11-08","stars":206},{"date":"2026-11-09","stars":206},{"date":"2026-11-10","stars":206}],"firstPublishedAt":"2026-11-09T23:02:44.491Z","commitSha":"c32f4121eaa89348c1e316050982ebebc65ff464","verifiedAt":"2026-11-09T22:54:06.129Z","lastCommitAt":"2024-03-19T09:31:33Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A command-line tool that collects e-mail addresses and social media handles from websites found by a Google search.","summary":"Clauneck is a Ruby command-line tool from SerpApi. You give it a Google search (for example every .ai site with a /contact page) and it goes through the results, pulls out e-mail addresses and Instagram, Facebook, X, TikTok, YouTube, GitHub and Medium handles, and writes everything into a CSV file. Its stated purpose is cold e-mail marketing. To use it you need Ruby, a SerpApi key for the searches and, as the README recommends, a list of proxy servers. The last release is from March 2024.","easeRationale":"Same on Windows, macOS and Linux: install Ruby, run `gem install clauneck`, create a SerpApi account for the key and then one command such as `clauneck --api_key YOUR_SERPAPI_KEY --output results.csv --q \"...\"`. Preparing a proxy list, which the README recommends, is extra work.","costNote":"The gem is MIT. The searches go through SerpApi, a paid service: the README says \"SerpApi offers free credits that renew every month\" and estimates about 10,000 pages on a free account, after which the searches are billed.","notVerified":["whether the Google cache addresses the tool depends on still answer","how many pages the free SerpApi credits really cover","what the public proxy servers do with the traffic that passes through them","how accurate the extracted addresses and handles are, since the README itself shows fields filled with null and error"]},"zh-Hans":{"summaryShort":"一个命令行工具，从 Google 搜索找到的网站中收集电子邮件地址和社交媒体账号。","summary":"Clauneck 是 SerpApi 推出的一个 Ruby 命令行工具。你给它一个 Google 搜索词（例如所有带 /contact 页面的 .ai 网站），它会遍历搜索结果，提取电子邮件地址以及 Instagram、Facebook、X、TikTok、YouTube、GitHub 和 Medium 的账号，并把全部内容写入一个 CSV 文件。它声明的用途是冷邮件营销。要使用它，你需要 Ruby、一个用于搜索的 SerpApi 密钥，以及（正如 README 建议的）一份代理服务器列表。最新版本发布于 2024 年 3 月。","easeRationale":"在 Windows、macOS 和 Linux 上流程相同：安装 Ruby，运行 `gem install clauneck`，注册一个 SerpApi 账号获取密钥，然后运行一条命令，例如 `clauneck --api_key YOUR_SERPAPI_KEY --output results.csv --q \"...\"`。准备 README 建议的代理列表则是额外的工作。","costNote":"该 gem 采用 MIT 许可。搜索请求经过 SerpApi，这是一项付费服务：README 写道 \"SerpApi offers free credits that renew every month\"，并估计免费账号大约能用于 10,000 页，之后的搜索将按量计费。","notVerified":["该工具依赖的 Google 缓存地址是否仍能访问","SerpApi 的免费额度实际能覆盖多少页面","公共代理服务器会如何处理经过它们的流量","提取出的地址和账号有多准确，因为 README 自身展示的字段中就有填着 null 和 error 的情况"]}}},{"id":"R_kgDOIuCgEw","slug":"shadcn-ui__ui","name":"shadcn/ui","owner":"shadcn-ui","fullName":"shadcn-ui/ui","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"useless_without_an_existing_react_project","severity":"info","evidence":"项目描述是 \"Build your own component library with code you can customize, extend, and make your own\"，主题标签包括 react、nextjs、vite、tailwindcss、radix-ui、react-aria、tanstack 和 laravel。这些组件是网页项目的源代码文件，本身没有任何可独立运行的东西。","source":"ai","title":"没有现成 react 项目就没用"},{"code":"components_arrive_over_the_network_into_your_files","severity":"info","evidence":"发布的版本是 npm 包（shadcn@4.23.0、@shadcn/registry@0.2.1），仓库的 package.json 中有 \"registry:build\"、\"registry:health\" 和 \"validate:registries\" 这类注册表脚本。命令行工具从某个注册表获取组件代码并写入你的项目，所以最终写进你文件里的内容取决于你指向的是哪个注册表。","source":"ai","title":"组件通过网络进入你的文件"},{"code":"large_issue_backlog","severity":"info","evidence":"在 127062 个星标的情况下有 1771 个未关闭的 issue，该项目每天都有提交（最近一次提交在 2026-10-31）。项目足够受欢迎，以至于报告积压的速度比关闭的速度更快。","source":"ai","title":"大量未处理的 issue 积压"},{"code":"ai_assistant_files_included_in_the_repository","severity":"info","evidence":"仓库顶层有 .claude/、.cursor-plugin/ 和 skills/ 文件夹，所以该项目除了组件之外，还附带了给 AI 编程助手使用的指令包。","source":"ai","title":"仓库中包含 ai 助手文件"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"tailwind-plus","name":"Tailwind Plus","confirmed":true}],"declaredAlternatives":[],"stars":127839,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":555,"growthDays":7,"history":[{"date":"2026-10-26","stars":126622},{"date":"2026-10-27","stars":126723},{"date":"2026-10-28","stars":126836},{"date":"2026-10-29","stars":126903},{"date":"2026-10-30","stars":126996},{"date":"2026-10-31","stars":127062},{"date":"2026-11-01","stars":127124},{"date":"2026-11-02","stars":127206},{"date":"2026-11-03","stars":127284},{"date":"2026-11-04","stars":127360},{"date":"2026-11-05","stars":127436},{"date":"2026-11-06","stars":127517},{"date":"2026-11-07","stars":127576},{"date":"2026-11-08","stars":127642},{"date":"2026-11-09","stars":127724},{"date":"2026-11-10","stars":127839}],"firstPublishedAt":"2026-10-27T01:10:12.681Z","commitSha":"c0d63c339d6cd2fc26135590a54cf1f362269451","verifiedAt":"2026-11-01T01:42:56.902Z","lastCommitAt":"2026-11-08T08:14:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ready-made React interface parts (buttons, forms, menus) that a developer copies into a project and is free to edit.","summary":"shadcn/ui is a collection of interface components for web projects built with React and Tailwind CSS, plus a command-line tool (`shadcn`) that copies the code of each component straight into your project instead of installing a package you cannot change. It is made for developers who want to build their own component library and keep control of the code. You need Node and an existing React project (Next.js, Vite, TanStack and Laravel are among the project's topics). The repository README carries no instructions: everything is on ui.shadcn.com/docs.","easeRationale":"The same on Windows, macOS and Linux: install Node, then run the `shadcn` command inside a React project you already have to add each component. There is nothing to download and open, and the repository itself explains none of it — the README only links to ui.shadcn.com/docs.","costNote":"Free and MIT-licensed, with no paid tier in the repository.","notVerified":["Whether the components work with the versions of React, Tailwind and your framework that you already use","What exactly the command writes into your project, which only running it shows","Whether later versions change components you already copied, since the code becomes yours after it is added","Disk use, estimated at 100 to 500 MB by a rule of thumb and not from a real download","Whether the accessibility claims in the description hold for each component"]},"zh-Hans":{"summaryShort":"现成的 React 界面组件（按钮、表单、菜单），开发者可以复制到项目中并自由修改。","summary":"shadcn/ui 是一套用 React 和 Tailwind CSS 构建的网页界面组件集合，还附带一个命令行工具（`shadcn`），它会把每个组件的代码直接复制到你的项目里，而不是安装一个你无法修改的包。它是为想要搭建自己的组件库并掌控代码的开发者而做的。你需要 Node 和一个已有的 React 项目（Next.js、Vite、TanStack 和 Laravel 都在项目的主题列表中）。仓库的 README 没有任何说明：所有内容都在 ui.shadcn.com/docs 上。","easeRationale":"在 Windows、macOS 和 Linux 上都一样：先安装 Node，然后在你已有的 React 项目里运行 `shadcn` 命令来添加每个组件。没有东西需要下载和打开，仓库本身也没有任何说明——README 只链接到 ui.shadcn.com/docs。","costNote":"免费，采用 MIT 许可证，仓库中没有任何付费层级。","notVerified":["这些组件是否能配合你已经在用的 React、Tailwind 及框架版本工作","这个命令到底会在你的项目里写入什么内容，只有运行后才能看到","后续版本是否会改变你已经复制过的组件，因为代码添加后就归你所有","磁盘占用，按经验估算为 100 到 500 MB，并非来自真实下载测得","描述中关于无障碍功能的说法是否对每个组件都成立"]}}},{"id":"MDEwOlJlcG9zaXRvcnk3NjAwNDA5","slug":"shadowsocks__shadowsocks-windows","name":"Shadowsocks for Windows","owner":"shadowsocks","fullName":"shadowsocks/shadowsocks-windows","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2022-02-08（1723 天前）","source":"check","title":"发布版本已久未更新"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 1.8/10","source":"check","title":"Scorecard 评分偏低"},{"code":"legal_status_varies_by_country","severity":"attention","evidence":"该应用的用途是让流量经由 Shadowsocks 代理转发（topics 'proxy, shadowsocks'），默认的 PAC 规则会让非中国域名走代理（README：'geositeProxiedGroups is initialized with geolocation-!cn'）。使用绕过网络封锁的工具在许多国家是合法的，但在另一些国家受到限制或被禁止。","source":"ai","title":"各国的法律地位不同"},{"code":"no_recent_commits","severity":"info","evidence":"最近一次提交在 2025-01-01（665 天前），最近一次发布是 2022-02-08 的 4.4.1.0。","source":"ai","title":"近期没有提交"},{"code":"no_files_in_latest_releases","severity":"info","evidence":"最近三个发布条目（4.4.1.0、4.4.0.0、4.3.3.0）在收集到的数据中没有列出任何平台文件，所以可能需要到发布页面的其他位置去找下载。","source":"ai","title":"最新发布中没有文件"}],"platforms":["windows"],"tools":["dotnet","other"],"buildTools":[],"disk":{"minMB":0,"maxMB":1,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"file"},"categories":["network_system"],"alternativeTo":[{"slug":"proxifier","name":"Proxifier","confirmed":true}],"declaredAlternatives":[],"stars":59487,"downloads":144807348,"downloadsWeek":26834,"downloadsDays":7,"growth":0,"growthDays":7,"history":[{"date":"2026-10-26","stars":59522},{"date":"2026-10-27","stars":59518},{"date":"2026-10-28","stars":59509},{"date":"2026-10-29","stars":59509},{"date":"2026-10-30","stars":59512},{"date":"2026-10-31","stars":59506},{"date":"2026-11-01","stars":59505},{"date":"2026-11-02","stars":59500},{"date":"2026-11-03","stars":59500},{"date":"2026-11-04","stars":59499},{"date":"2026-11-05","stars":59496},{"date":"2026-11-06","stars":59492},{"date":"2026-11-07","stars":59490},{"date":"2026-11-08","stars":59491},{"date":"2026-11-09","stars":59489},{"date":"2026-11-10","stars":59487}],"firstPublishedAt":"2026-10-27T01:10:12.682Z","commitSha":"891d971682eefcaa2e640258d3b352a3ad3b2233","verifiedAt":"2026-10-29T01:39:54.993Z","lastCommitAt":"2025-01-01T08:08:23Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A small Windows tray app that sends your internet traffic through a Shadowsocks proxy server to get around network blocks.","summary":"Shadowsocks for Windows is a tray app that connects your computer to a Shadowsocks proxy server. It can set the Windows system proxy, send only some websites through the proxy using rules (PAC mode), switch between several servers and use plugins. It is for people who already have a Shadowsocks server or provider, often to get around censorship. It needs .NET Framework 4.8 and the Visual C++ 2015 runtime. The last release is from February 2022, so the app is no longer actively updated.","easeRationale":"Windows is the only platform: you download the release package and run it from the tray, then type in your server details; it also needs .NET Framework 4.8 (built into recent Windows) and the Visual C++ 2015 x86 runtime.","costNote":"Free (GPL-3.0); you still need a Shadowsocks server or a paid provider.","notVerified":["Whether the 2022 release still works on current Windows 11 builds.","Whether its encryption options are still up to date against current blocking methods.","How the server you connect to handles your traffic; that depends on who runs it.","Whether it is legal to use where you live.","Which file you actually download from the release page."]},"zh-Hans":{"summaryShort":"一个小巧的 Windows 系统托盘应用,通过 Shadowsocks 代理服务器转发你的网络流量,绕过网络封锁。","summary":"Shadowsocks for Windows 是一个系统托盘应用,可以把你的电脑连接到 Shadowsocks 代理服务器。它可以设置 Windows 系统代理,用规则(PAC 模式)只让部分网站走代理,在多个服务器之间切换,并支持插件。它适合已经拥有 Shadowsocks 服务器或提供商的人,通常用于绕过网络审查。需要 .NET Framework 4.8 和 Visual C++ 2015 运行库。最新版本发布于 2022 年 2 月,此后该应用不再持续更新。","easeRationale":"仅支持 Windows:下载发布包并从托盘运行,然后输入你的服务器信息;还需要 .NET Framework 4.8(近期 Windows 已内置)和 Visual C++ 2015 x86 运行库。","costNote":"免费(GPL-3.0);你仍然需要一个 Shadowsocks 服务器或付费提供商。","notVerified":["2022 年的版本是否仍能在当前的 Windows 11 版本上运行。","它的加密方式面对当前的封锁手段是否仍然有效。","你连接的服务器如何处理你的流量,这取决于运营者是谁。","在你居住的地方使用它是否合法。","你从发布页面实际下载的是哪个文件。"]}}},{"id":"R_kgDOIS4yWA","slug":"shadps4-emu__shadps4","name":"shadPS4","owner":"shadps4-emu","fullName":"shadps4-emu/shadPS4","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"legal_status_varies_by_country","severity":"attention","evidence":"README 的 Firmware files：\"The above modules are required to run the games properly and must be dumped from your legally owned PlayStation 4 console.\" 把这些系统文件和自己的游戏从主机里复制出来，意味着越过主机自身的保护机制，即便光盘是你自己的，这在不少国家也受到限制或被禁止。","source":"ai","title":"各国法律情况不同"},{"code":"early_development","severity":"info","evidence":"README 的 Status：\"shadPS4 is early in development, don't expect a flawless experience\"，目前有 337 个未关闭的问题，另有一份单独的游戏兼容性列表，可用来查询某款作品能否运行。","source":"ai","title":"仍处于开发早期"},{"code":"this_build_has_no_interface","severity":"info","evidence":"README 的 General information：\"This is the emulator core, which does not include a GUI. If you just want to use the emulator as an end user, download the QtLauncher instead\"，而 QtLauncher 位于另一个仓库（shadps4-emu/shadps4-qtlauncher）。","source":"ai","title":"这个版本没有图形界面"},{"code":"most_downloads_are_test_builds","severity":"info","evidence":"最新的下载标记为 \"Pre-release-shadPS4-2026-11-10-4788f0b1\"，是当天由单个提交生成的构建，旁边才是 v.0.21.0 这样的编号版本；每日构建受到的测试远少于正式发布。","source":"ai","title":"多数下载都是测试构建"},{"code":"mac_support_limited","severity":"info","evidence":"README 的 Building/macOS：\"macOS users need at least macOS 26.0 to run shadPS4. Intel Macs are not supported.\"","source":"ai","title":"Mac 支持有限"},{"code":"needs_console_firmware_dump","severity":"high","evidence":"README 的 Firmware files：\"The above modules are required to run the games properly and must be dumped from your legally owned PlayStation 4 console.\" 同一份 README 还说程序会启动从主机复制来的游戏目录或 ELF 文件（\"shadPS4 CUSA00001\"、\"shadPS4 /path/to/game.elf\"）。把这些系统文件和游戏从 PlayStation 4 中取出，意味着绕过主机自身的保护机制，而许多国家都有法律限制绕过复制保护，即使游戏是你自己的。","source":"ai","title":"需要导出主机固件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":30,"maxMB":90,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[{"slug":"playstation-plus-premium","name":"PlayStation Plus Premium","confirmed":true}],"declaredAlternatives":[],"stars":34113,"downloads":4751675,"downloadsWeek":26803,"downloadsDays":6,"growth":64,"growthDays":6,"history":[{"date":"2026-11-04","stars":34049},{"date":"2026-11-05","stars":34066},{"date":"2026-11-06","stars":34069},{"date":"2026-11-07","stars":34084},{"date":"2026-11-08","stars":34089},{"date":"2026-11-09","stars":34100},{"date":"2026-11-10","stars":34113}],"firstPublishedAt":"2026-11-10T23:52:44.020Z","commitSha":"4788f0b108a6faaae5105f8ac88ef8e1935e2b22","verifiedAt":"2026-11-10T23:37:18.199Z","lastCommitAt":"2026-11-10T10:41:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Free PlayStation 4 emulator for Windows, Mac and Linux that runs games and system files copied from your own console.","summary":"shadPS4 lets a PC run PlayStation 4 games. You unzip the program, put PS4 system modules dumped from your own console into its sys_modules folder, and point it at a game folder. It already runs titles such as Bloodborne, Dark Souls Remastered and Red Dead Redemption, and Xbox and DualShock controllers work without setup. This repository is the emulator core and has no window of its own: you start it with commands, or download the separate QtLauncher for a normal interface. The team says it is early in development.","easeRationale":"On Windows, the simplest case, you download a ready 30 MB zip and unzip it, but there is no installer and this build has no interface: you type commands such as `shadPS4 CUSA00001`, or fetch the separate QtLauncher, and first you have to copy about 34 system files and the game itself off a PlayStation 4. Mac needs macOS 26.0 or newer and an Apple Silicon machine; Intel Macs are not supported.","costNote":"Free and open source under GPL-2.0; the team takes donations on Ko-fi. The games and system files you need are not provided.","notVerified":["whether a particular game runs, and how well, since that depends on the title and changes with each build","what speed you get, which depends on your processor and graphics card; the project publishes no minimum requirements","how much disk space you need once PS4 games and shader caches are on your drive, beyond the 30 to 35 MB of the program itself","whether the system modules you copied from your console match the versions the emulator expects"]},"zh-Hans":{"summaryShort":"面向 Windows、Mac 和 Linux 的免费 PlayStation 4 模拟器，运行从你自己主机中导出的游戏和系统文件。","summary":"shadPS4 让电脑能运行 PlayStation 4 游戏。你解压程序，把从自己主机导出的 PS4 系统模块放进它的 sys_modules 目录，再指向某个游戏目录即可。它已经能运行《血源诅咒》《黑暗之魂 重制版》和《荒野大镖客：救赎》等作品，Xbox 和 DualShock 手柄无需设置即可使用。本仓库是模拟器内核，没有自己的窗口界面：你要用命令启动它，或者另外下载 QtLauncher 来获得常规界面。团队表示项目仍处于开发早期。","easeRationale":"在最简单的 Windows 上，你下载一个 30 MB 的现成压缩包解压即可，但没有安装程序，而且这个版本没有界面：你需要输入 `shadPS4 CUSA00001` 这样的命令，或另外去取 QtLauncher；在此之前还必须从 PlayStation 4 上复制约 34 个系统文件和游戏本体。Mac 需要 macOS 26.0 或更高版本以及 Apple Silicon 机型，不支持 Intel 版 Mac。","costNote":"免费开源，采用 GPL-2.0；团队在 Ko-fi 上接受捐助。所需的游戏和系统文件不随程序提供。","notVerified":["某款具体游戏能否运行、运行效果如何，这取决于作品本身，并且每个构建都会变化","你能获得多少帧率，这取决于你的处理器和显卡；项目没有公布最低配置要求","当 PS4 游戏和着色器缓存都放进硬盘后需要多少磁盘空间，程序本身只有 30 到 35 MB","你从主机复制来的系统模块，版本是否与模拟器期望的一致"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMzQyODI2NA==","slug":"sharex__sharex","name":"ShareX","owner":"ShareX","fullName":"ShareX/ShareX","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"uploads_to_third_party_services","severity":"info","evidence":"README.md：ShareX \"can upload images, text, files, folders, clipboard content and URLs to many different destinations\"，然后 \"copy the URL to the clipboard\" 或 \"shorten the URL\"。你上传的任何内容都会离开你的电脑，落到别人的服务上，可能因此可以通过链接访问。","source":"ai","title":"上传到第三方服务"},{"code":"cross_platform_version_not_released","severity":"info","evidence":"仓库里有一个 \"ShareX.Avalonia/\" 文件夹，主题标签里列有 \"avalonia\"（一种用于开发 Windows、macOS 和 Linux 应用的工具包），但全部 91 个版本发布都只提供 Windows 安装程序（ShareX-22.0.0-setup-x64.exe）。","source":"ai","title":"跨平台版本尚未发布"},{"code":"copies_distributed_elsewhere","severity":"info","evidence":"README.md 说 ShareX 也上架了 Microsoft Store 和 Steam，并补充 \"For the safest download options, use the official links below\"，这暗示在官方渠道之外流传着重新打包的副本。","source":"ai","title":"其他地方流传的副本"},{"code":"runs_external_commands","severity":"info","evidence":"README.md：截图之后你可以 \"create actions that run external tools or scripts\"，所以一份共享的配置文件可能让 ShareX 在你的电脑上运行其他程序。","source":"ai","title":"运行外部命令"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":126,"maxMB":377,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image","video","productivity"],"alternativeTo":[{"slug":"snagit","name":"Snagit","confirmed":true},{"slug":"camtasia","name":"Camtasia","confirmed":true},{"slug":"dropbox","name":"Dropbox","confirmed":true}],"declaredAlternatives":[],"stars":40927,"downloads":34483371,"downloadsWeek":208021,"downloadsDays":7,"growth":319,"growthDays":7,"history":[{"date":"2026-10-26","stars":40435},{"date":"2026-10-27","stars":40463},{"date":"2026-10-28","stars":40477},{"date":"2026-10-29","stars":40510},{"date":"2026-10-30","stars":40531},{"date":"2026-10-31","stars":40553},{"date":"2026-11-01","stars":40573},{"date":"2026-11-02","stars":40592},{"date":"2026-11-03","stars":40608},{"date":"2026-11-04","stars":40632},{"date":"2026-11-05","stars":40647},{"date":"2026-11-06","stars":40704},{"date":"2026-11-07","stars":40764},{"date":"2026-11-08","stars":40836},{"date":"2026-11-09","stars":40886},{"date":"2026-11-10","stars":40927}],"firstPublishedAt":"2026-10-27T01:10:12.683Z","commitSha":"0f2450be0ebf8efeefb6ac2c59826e9b1f164d2b","verifiedAt":"2026-11-02T02:54:31.717Z","lastCommitAt":"2026-11-10T06:51:09Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"ShareX-22.4.0-setup-x64.exe","sizeBytes":145391368,"url":"https://github.com/ShareX/ShareX/releases/download/v22.4.0/ShareX-22.4.0-setup-x64.exe","releaseUrl":"https://github.com/ShareX/ShareX/releases/tag/v22.4.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Takes screenshots or records your Windows screen with one key, marks them up and can upload and copy the link for you.","summary":"ShareX is a Windows program for capturing the screen: a region, a window, a scrolling page, a video or an animated GIF, started from a hotkey. It then does what you set up for it, such as opening the built-in editor to draw arrows, blur private details, saving the file, copying it, reading text out of the image with OCR or uploading it and copying the link. It also bundles dozens of small tools (colour picker, image resizer, video trimmer, QR codes). It is free, needs no account, and the installer is a normal Windows setup file.","easeRationale":"On Windows you download ShareX-22.0.0-setup-x64.exe (126 MB) from the releases page and open it; everything works without an account, so it is an A. There is no build for macOS or Linux.","notVerified":["Which upload service each destination is set to by default, and what that service does with your images","Real disk use after installing (the installer is 126 MB; the checks estimate 126-377 MB)","Whether the screen recorder and OCR work on your version of Windows and your graphics card","Whether the Avalonia rewrite in the repository already runs on macOS or Linux"]},"zh-Hans":{"summaryShort":"一键截图或录制 Windows 屏幕，可以做标注，还能上传并替你复制链接。","summary":"ShareX 是一款用于捕获屏幕的 Windows 程序：可以截取一个区域、一个窗口、一整页滚动内容、一段视频或一张 GIF 动图，用快捷键即可启动。随后它会按你的设置执行，比如打开内置编辑器画箭头、模糊隐私内容、保存文件、复制文件、用 OCR 识别图片中的文字，或者上传并复制链接。它还自带几十个小工具（取色器、图片缩放、视频裁剪、二维码）。它免费、无需账号，安装程序就是普通的 Windows 安装文件。","easeRationale":"在 Windows 上，从发布页下载 ShareX-22.0.0-setup-x64.exe（126 MB）并打开即可；一切无需账号就能使用，所以评级为 A。没有 macOS 或 Linux 版本。","notVerified":["每个目的地默认使用哪个上传服务，以及该服务会如何处理你的图片","安装后的实际磁盘占用（安装程序 126 MB；检查估算为 126-377 MB）","屏幕录制和 OCR 在你的 Windows 版本和显卡上能否正常工作","仓库里基于 Avalonia 的重写版本是否已经能在 macOS 或 Linux 上运行"]}}},{"id":"R_kgDOUpnk2g","slug":"shihabal3amri__diplay","name":"DiPlay","owner":"shihabal3amri","fullName":"shihabal3amri/DiPlay","type":"needs_setup","status":"alert","grade":"E","easeByPlatform":{},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"33 天内获得 12567 个星标","source":"check","title":"新仓库，星标很多"},{"code":"reuses_accessory_identity_from_other_firmware","severity":"high","evidence":"README.md：\"This is not an Apple-certified product. The APK bundles an experimental accessory identity recovered from public Carlinkit firmware, not a newly provisioned MFi identity for DiPlay. A bundled private key is extractable.\" 该应用通过复用另一家公司的配件凭据来让 CarPlay 接受它，而这既未获得 Apple 的授权，也未获得那家公司的授权，并且在一些国家提取此类密钥是违法的。","source":"ai","title":"复用其他固件的配件身份"},{"code":"public_preview_unverified_compatibility","severity":"attention","evidence":"README.md：\"This release invites community testing; it is not a guarantee of universal compatibility\"、\"Android 7.0-8.1 support is not yet confirmed on a vehicle\"、\"wheel-speed dead reckoning in tunnels remains unverified\" 以及 \"General stutter, calls/Siri, decoder and model-specific reports still need current-device evidence\"。","source":"ai","title":"公开预览版，兼容性未经验证"},{"code":"may_stop_working_after_updates","severity":"attention","evidence":"README.md：\"Acceptance after future iOS updates, reliability across head units and suitability of that identity for general distribution are unresolved.\" 手机或汽车的更新可能让它无法使用，而且共用的身份可能被封锁。","source":"ai","title":"更新后可能停止工作"},{"code":"brand_scope_limited","severity":"info","evidence":"README.md：\"These projects focus on BYD cars. They may work on other brands, but other brands are unsupported and there are no plans to add support or fix brand-specific incompatibilities.\"","source":"ai","title":"仅限特定品牌"},{"code":"mixed_copyleft_licenses","severity":"info","evidence":"README.md：\"Based on xcertplay, GPL-3.0. The home/settings UI and website adapt DiAuto, AGPL-3.0 ... Preserve those notices when distributing modifications\"，而 GitHub 将该仓库报告为 GPL-3.0。","source":"ai","title":"混合的 copyleft 许可证"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware"],"alternativeTo":[{"slug":"carlinkit-wireless-carplay-adapter","name":"wireless CarPlay adapter","confirmed":true}],"declaredAlternatives":[],"stars":13590,"downloads":925522,"downloadsWeek":62045,"downloadsDays":7,"growth":414,"growthDays":7,"history":[{"date":"2026-10-26","stars":12482},{"date":"2026-10-27","stars":12561},{"date":"2026-10-28","stars":12685},{"date":"2026-10-29","stars":12785},{"date":"2026-10-30","stars":12893},{"date":"2026-10-31","stars":12964},{"date":"2026-11-01","stars":13034},{"date":"2026-11-02","stars":13112},{"date":"2026-11-03","stars":13176},{"date":"2026-11-04","stars":13258},{"date":"2026-11-05","stars":13332},{"date":"2026-11-06","stars":13380},{"date":"2026-11-07","stars":13430},{"date":"2026-11-08","stars":13480},{"date":"2026-11-09","stars":13524},{"date":"2026-11-10","stars":13590}],"firstPublishedAt":"2026-10-27T01:14:22.104Z","commitSha":"a2dc2e70b3ff24e9e18e56db71cde776b30a4eec","verifiedAt":"2026-10-28T01:41:23.896Z","lastCommitAt":"2026-10-28T09:01:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Adds wired and wireless CarPlay to the Android screen of some BYD cars, without buying an adapter.","summary":"DiPlay is an app you install on the Android screen built into certain BYD cars, not on the iPhone, so the car can show CarPlay over USB or wirelessly, including BYD's windshield navigation. It is a public preview by one developer: the head unit must allow installing app files, some extras need developer access (ADB) to the car, and only BYD is supported. It is not an Apple-approved product: it works by reusing an accessory identity taken from another company's adapter firmware, and that may stop being accepted at any time.","easeRationale":"There is nothing to run on Windows, Mac, Linux or the browser, so it is grade E here: the app is installed on a car's Android screen (\"Install on the car, not the iPhone\"), the head unit must permit installing app files, and some features need developer access to the car over ADB. Car head units are not a system Radar grades.","costNote":"Free and open source (GPL-3.0). No account, dongle or paid server is needed, according to the README.","notVerified":["Whether it works on your exact car, head unit firmware and iPhone version","Whether a car or phone update makes it stop working, and whether the shared identity gets blocked","What the saved diagnostic report contains before you send it to the project","Whether installing it can disturb the car's own software or the factory projection app","Whether using it this way breaks the terms of your car or phone warranty where you live"]},"zh-Hans":{"summaryShort":"为部分比亚迪(BYD)汽车的Android车机添加有线和无线CarPlay功能,不需要购买适配器。","summary":"DiPlay 是一款安装在部分比亚迪汽车自带Android车机上的应用,不是装在iPhone上,它让车机能够通过USB有线或无线方式显示CarPlay,包括比亚迪的前挡风玻璃导航。这是由一位开发者发布的公开预览版:车机必须允许安装应用文件,部分附加功能需要对车机进行开发者(ADB)访问,目前只支持比亚迪。这不是苹果官方认可的产品:它的原理是复用另一家公司适配器固件中的配件身份标识,这种做法随时可能不再被接受。","easeRationale":"在Windows、Mac、Linux或浏览器上都没有可运行的内容,所以这里评级为E:这款应用安装在汽车的Android车机屏幕上(“请装在车机上,不要装在iPhone上”),车机必须允许安装应用文件,部分功能需要通过ADB获得对车机的开发者访问权限。Radar不对汽车车机这类系统进行评级。","costNote":"免费且开源(GPL-3.0许可)。根据README的说明,不需要账号、硬件适配器(dongle)或付费服务器。","notVerified":["是否适用于你具体的汽车、车机固件版本和iPhone版本","汽车或手机的更新是否会导致它失效,以及共用的身份标识是否会被屏蔽","在你把保存的诊断报告发给项目之前,报告里到底包含哪些内容","安装它是否会干扰汽车自身的软件或原厂的投屏应用","在你所在地区,这样使用是否违反了汽车或手机质保条款"]}}},{"id":"R_kgDOP42VuA","slug":"showlab__code2video","name":"Code2Video","owner":"showlab","fullName":"showlab/Code2Video","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"cost_per_video_unknown","severity":"attention","evidence":"README 的 Configure LLM API Keys：流程需要一个付费文本模型（\"Best Manim code quality achieved with Claude-4-Opus\"），外加一个供 Critic 使用的付费视觉模型，run_agent.sh 里还有 PARALLEL_GROUP_NUM 设置可同时跑多个主题。仓库没有给出每个视频的价格，只把 \"Token usage\" 列为需要衡量的指标。","source":"ai","title":"每个视频的花费不明"},{"code":"your_topics_go_to_outside_ai_services","severity":"info","evidence":"README 的 Run Agents：每个视频都通过调用 api_config.json 里配置的 LLM API 和 VLM API 生成，因此主题、生成的代码和渲染出的画面都会发送到这些公司的服务器。","source":"ai","title":"你的主题会发送给外部 AI 服务"},{"code":"icons_come_from_stock_libraries","severity":"info","evidence":"README 的 Acknowledgements：\"High-quality visual assets (icons) are provided by IconFinder and Icons8\"，更新说明里还提到 IconFinder 账号出了问题，于是把图标复制到一个 HuggingFace 数据集 \"as a temporary alternative\"。这些图标的许可条款同样适用于你发布的任何视频。","source":"ai","title":"图标来自素材图库"},{"code":"research_code_not_a_packaged_program","severity":"info","evidence":"该仓库是 ICML 2026 论文的配套代码：没有发布版本、没有安装程序、没有版本号，使用方式是修改 run_agent.sh 和 run_agent_single.sh 里的变量。最后一次提交是在本次检查的 78 天前。","source":"ai","title":"这是研究代码，不是打包好的软件"},{"code":"quality_compared_only_with_its_own_benchmark","severity":"info","evidence":"README 的 Evaluation：质量结论建立在 MMMC 之上，而这个基准集由同一批作者构建，并用他们自己的脚本（eval_TQ.py、eval_AES.py）打分，以 3Blue1Brown 的课程作为优质视频的参照。","source":"ai","title":"质量只与自家基准做过比较"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["LLM API (Claude or another model, for the Planner and Coder)","Gemini API (vision model, for the Critic that fixes layout)","IconFinder"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","ai_agents"],"alternativeTo":[{"slug":"google-veo-3","name":"Veo 3","confirmed":true},{"slug":"iconfinder","name":"Iconfinder","confirmed":false}],"declaredAlternatives":[{"target":"iconfinder","stance":"claims","evidence":"we’ve updated Code2Video auto-collected icons at [MMMC](https://huggingface.co/datasets/YanzheChen/MMMC/tree/main/assets) as a temporary alternative"}],"stars":2355,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":114,"growthDays":6,"history":[{"date":"2026-11-08","stars":2347},{"date":"2026-11-09","stars":2353},{"date":"2026-11-10","stars":2355}],"firstPublishedAt":"2026-11-10T23:52:44.030Z","commitSha":"1142d8e14cdc2806df85aedb0fbb5dca474caa0f","verifiedAt":"2026-11-10T23:39:09.311Z","lastCommitAt":"2026-08-24T11:00:47Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a topic you type into an animated teaching video: AI agents write the animation code, run it and fix the layout.","summary":"Code2Video is research code from Show Lab (NUS) that makes 3Blue1Brown-style teaching videos. You type a topic such as \"Linear transformations and matrices\" and three AI agents work together: one plans the storyboard, one writes Manim animation code and debugs it, and one looks at the result and fixes the layout. It comes with MMMC, a benchmark of 117 learning topics, and scripts to score the videos. You run it from a terminal with Python and Manim installed, and you pay for the AI services it calls.","easeRationale":"The same on all three systems, Linux being the usual one for this kind of code: there is no download and no window. You clone the repository, install Python packages, set up Manim Community 0.19 with the FFmpeg and LaTeX pieces its own installation guide asks for, write your keys into api_config.json, then edit and run a shell script.","costNote":"The code is MIT. Each video is made by calling paid AI services; the README recommends Claude-4-Opus for the code and gemini-2.5-pro-preview for the layout checks, and names token usage as one of its efficiency metrics.","notVerified":["how much the AI services charge for one finished video, and how often a run fails halfway and has to be paid for again","whether the generated animation code runs without errors for topics outside the 117 in the benchmark","how long one video takes on a normal computer, and how much disk the rendered frames and cached icons use","whether Manim, LaTeX and FFmpeg install cleanly on Windows and macOS for this exact version","what the licences of the downloaded icons allow if you publish or sell the videos"]},"zh-Hans":{"summaryShort":"把你输入的主题变成动画教学视频：AI 智能体负责写动画代码、运行并修正画面排版。","summary":"Code2Video 是 Show Lab（新加坡国立大学）的研究代码，用来制作 3Blue1Brown 风格的教学视频。你输入一个主题，例如 \"Linear transformations and matrices\"，三个 AI 智能体协同工作：一个规划分镜，一个编写 Manim 动画代码并调试，一个查看结果并修正排版。它附带 MMMC——一个包含 117 个学习主题的基准集，以及给视频打分的脚本。你需要在装好 Python 和 Manim 的终端里运行它，并自行支付它调用的 AI 服务费用。","easeRationale":"三个系统上的做法相同，这类代码通常跑在 Linux 上：没有下载包，也没有图形界面。你要克隆仓库、安装 Python 包、按照 Manim Community 0.19 官方安装指南配好 FFmpeg 和 LaTeX 组件，把密钥写进 api_config.json，然后编辑并运行一个 shell 脚本。","costNote":"代码采用 MIT 许可。每个视频都要调用付费 AI 服务；README 推荐用 Claude-4-Opus 生成代码、用 gemini-2.5-pro-preview 检查排版，并把 token 用量列为效率指标之一。","notVerified":["做成一个视频，AI 服务要收多少钱，以及运行中途失败、需要重新付费的概率有多高","对于基准集 117 个主题之外的题目，生成的动画代码能否无错运行","在普通电脑上做一个视频要多久，渲染出的帧和缓存的图标又会占多少磁盘空间","在 Windows 和 macOS 上，这个指定版本的 Manim、LaTeX 和 FFmpeg 能否顺利装好","如果你发布或出售这些视频，所下载图标的许可条款是否允许"]}}},{"id":"R_kgDOP7ezIg","slug":"showlab__paper2video","name":"Paper2Video","owner":"showlab","fullName":"showlab/Paper2Video","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"clones_a_face_and_a_voice","severity":"attention","evidence":"README 的 Inference：流程接受 `--ref_img`（\"must be square portrait\"）和 `--ref_audio`（\"recommended: ~10s\"），并用 Hallo2 模型生成该声音的语音以及该人脸的数字主讲人视频。同样的步骤，用在未经本人同意的照片和语音样本上也一样有效。","source":"ai","title":"会克隆人脸和声音"},{"code":"needs_a_data_center_graphics_card","severity":"attention","evidence":"README 的 Inference：\"The minimum recommended GPU for running this pipeline is NVIDIA A6000 with 48G\"，而两条示例命令都传入 `--gpu_list [0,1,2,3,4,5,6,7]`，也就是八块显卡，用于生成光标和渲染数字主讲人。","source":"ai","title":"需要数据中心级显卡"},{"code":"cost_per_video_unknown","severity":"attention","evidence":"README 的 Configure LLMs：整篇论文都要经过 GPT-4.1 或 Gemini 2.5 Pro 处理文本和视觉，幻灯片排版还使用树搜索（`--if_tree_search` 默认为 True），会反复调用模型。仓库没有给出做一个视频大约要花多少钱。","source":"ai","title":"每个视频的花费不明"},{"code":"pulls_in_another_project_and_model_weights","severity":"info","evidence":"README 的 Requirements：要生成真人主讲画面，你得执行 \"git clone https://github.com/fudan-generative-vision/hallo2.git\"，为它单独建一个 conda 环境并下载模型权重，原因是 \"potential package conflicts\"。这些代码都不在本仓库里。","source":"ai","title":"需要额外拉取另一个项目和模型权重"},{"code":"not_updated_for_months","severity":"info","evidence":"最后一次提交是 2026-03-05，距本次检查 250 天，没有任何发布版本，README 里最新的更新说明来自 2025-10-15。","source":"ai","title":"已有数月未更新"},{"code":"benchmark_videos_taken_from_public_sites","severity":"info","evidence":"README 的 Acknowledgements：\"The souces of the presentation videos are SlideLive and YouTuBe\"，也就是说，它用来比较的数据集是别人录制的演讲。","source":"ai","title":"基准视频取自公开网站"}],"platforms":["linux"],"tools":["python","gpu_nvidia","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["OpenAI","Google Gemini"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","audio","ai_agents"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true}],"declaredAlternatives":[],"stars":2422,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":7,"growthDays":6,"history":[{"date":"2026-11-08","stars":2418},{"date":"2026-11-09","stars":2421},{"date":"2026-11-10","stars":2422}],"firstPublishedAt":"2026-11-10T23:52:44.047Z","commitSha":"47beb503e24218f8f8b3075f5d2bed7844d199b4","verifiedAt":"2026-11-10T23:40:04.697Z","lastCommitAt":"2026-03-05T09:05:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a research paper, a photo and a 10-second voice clip into a narrated talk video with slides and a talking head.","summary":"Paper2Video is research code from Show Lab (NUS) that makes the presentation video researchers record for a conference. You give it the LaTeX source of your paper, a square photo of the presenter and about ten seconds of their voice; an agent writes the slides, the subtitles and the narration, moves a cursor over the slides, speaks the text in the cloned voice and can add a talking head. There is a lighter mode without the talking head. It runs from the terminal on a machine with a large NVIDIA card and paid AI service keys.","easeRationale":"Linux is the only system with instructions, and it is a long setup: a conda environment with Python 3.10, the tectonic LaTeX engine, API keys exported in the shell, and for the talking head a second conda environment plus cloning the Hallo2 project and downloading its model weights. The README says the minimum recommended card is an NVIDIA A6000 with 48 GB, and its examples pass eight GPUs.","costNote":"The code is MIT, but the pipeline calls paid services: the README recommends GPT-4.1 or Gemini 2.5 Pro for both the text and the vision parts. It says locally run open models such as Qwen are also supported, pointing to another project for the details.","notVerified":["how much the OpenAI and Gemini calls cost for one paper, and how often a run has to be repeated","whether the lighter pipeline without the talking head works on a smaller graphics card","how much disk the Hallo2 weights and the rendered video need","whether the slides and narration describe the paper correctly, or invent content a reader would have to check","whether it runs on Windows or macOS at all, since only Linux-style instructions are given"]},"zh-Hans":{"summaryShort":"把一篇论文、一张照片和 10 秒语音，变成带幻灯片和数字主讲人的讲解视频。","summary":"Paper2Video 是 Show Lab（新加坡国立大学）的研究代码，用来生成研究者为学术会议录制的那种演讲视频。你提供论文的 LaTeX 源码、一张主讲人的方形照片和大约十秒的语音；智能体会撰写幻灯片、字幕和旁白，在幻灯片上移动光标，用克隆出的声音朗读文本，还可以加上数字主讲人画面。它还有一个不带主讲人的轻量模式。整个流程在终端运行，需要一块大显存的 NVIDIA 显卡和付费 AI 服务的密钥。","easeRationale":"只有 Linux 有安装说明，而且准备工作很长：一个 Python 3.10 的 conda 环境、tectonic 这个 LaTeX 引擎、在 shell 里导出的 API 密钥；要生成数字主讲人，还需要第二个 conda 环境，并克隆 Hallo2 项目、下载它的模型权重。README 说最低推荐显卡是 48 GB 的 NVIDIA A6000，示例命令里一次用了八块 GPU。","costNote":"代码采用 MIT 许可，但流程会调用付费服务：README 推荐文本和视觉部分都用 GPT-4.1 或 Gemini 2.5 Pro。它也说支持在本地运行 Qwen 等开源模型，并把细节指向另一个项目。","notVerified":["处理一篇论文时 OpenAI 和 Gemini 的调用要花多少钱，以及需要重跑的概率有多高","不带数字主讲人的轻量流程，能否在较小的显卡上跑起来","Hallo2 的权重和渲染出的视频需要多少磁盘空间","幻灯片和旁白是否准确描述了论文，还是会编造需要读者自行核实的内容","它在 Windows 或 macOS 上是否能运行，因为文档只给了 Linux 风格的说明"]}}},{"id":"R_kgDOIrCGYA","slug":"showlab__tune-a-video","name":"Tune-A-Video","owner":"showlab","fullName":"showlab/Tune-A-Video","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：torch: GHSA-3749-ghw9-m3mg, GHSA-47fc-vmwq-366v, GHSA-53q9-r3pm-6pq6; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446","source":"check","title":"存在已知漏洞"},{"code":"pinned_versions_from_2022","severity":"attention","evidence":"requirements.txt 把 torch==1.12.1、torchvision==0.13.1、diffusers[torch]==0.11.1 和 bitsandbytes==0.35.4 全部锁死，都是 2022 年的版本。这些版本在当前的 Python 和显卡驱动上很难装上，也正是检查中被标记存在已知漏洞的那些依赖。","source":"ai","title":"依赖锁定在 2022 年的版本"},{"code":"not_maintained","severity":"attention","evidence":"最后一次提交是 2023-10-25，距本次检查 1112 天，没有任何发布版本，有 37 个未关闭的问题，README 的 News 列表里最新一条是 2023 年 2 月。","source":"ai","title":"已无人维护"},{"code":"no_interface_and_no_finished_program","severity":"info","evidence":"仓库里有 train_tuneavideo.py、配置文件和一个 Python 模块；\"Inference\" 一节只是一段代码片段，需要你自己粘进脚本里。唯一现成的界面是别人在 Hugging Face 上搭的 Gradio 演示。","source":"ai","title":"没有界面，也没有成品程序"},{"code":"example_prompts_use_protected_characters","severity":"info","evidence":"README 的结果表格使用了 \"Spider Man is skiing on the beach\"、\"Wonder Woman, wearing a cowboy hat\"、\"James Bond is dribbling a basketball\"、\"Mr Potato Head\" 和 \"modern disney style\" 等提示词，还指向 Modern Disney、Anything V4.0 等微调模型。这类视频用到的角色和风格都归别人所有。","source":"ai","title":"示例提示词用了受保护的角色"},{"code":"edits_real_footage_of_people","severity":"info","evidence":"该方法拿一段真实视频，换一个主体重新生成，就像 \"A man is skiing\" 的示例那样；同样的步骤也适用于未经本人同意、可被辨认出身份的影像。","source":"ai","title":"会改写真人的实拍画面"}],"platforms":["linux"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_any"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":4359,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":4359},{"date":"2026-11-06","stars":4359},{"date":"2026-11-07","stars":4359},{"date":"2026-11-08","stars":4359},{"date":"2026-11-09","stars":4359},{"date":"2026-11-10","stars":4359}],"firstPublishedAt":"2026-11-10T23:52:44.064Z","commitSha":"7d4a89a2b90067e8d805ef1287beb4e4adf37e18","verifiedAt":"2026-11-10T23:41:03.966Z","lastCommitAt":"2023-10-25T02:00:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that teaches Stable Diffusion a single short video, then remakes that clip from a new text prompt.","summary":"Tune-A-Video is the code behind a 2023 research paper (ICCV). You give it one short video and a sentence describing it, and it fine-tunes a Stable Diffusion image model for about 10 to 15 minutes on a strong NVIDIA card. After that you can ask for the same motion with a different subject or style, such as turning \"a man is skiing\" into \"Spider Man is skiing on the beach\". Everything happens in a terminal and in Python; there is no interface, and the project has not been touched since October 2023.","easeRationale":"Linux is the only setup described: install the Python packages, download Stable Diffusion weights from Hugging Face yourself, edit a YAML config and run `accelerate launch train_tuneavideo.py` on a card like an A100. Getting a video out afterwards means writing your own Python script from the snippet in the README; a Colab notebook is linked for trying it without installing anything.","costNote":"Apache-2.0 code. The Stable Diffusion models it builds on have their own licences, and the training needs a rented or owned GPU.","notVerified":["whether the 2022 package versions still install and run on a current machine and driver","how much graphics memory is really needed, since the only figure given is 10 to 15 minutes on one A100","how much disk the Stable Diffusion weights and the per-video fine-tuned copies take once you keep several of them","whether the linked Colab notebook and Hugging Face demo still work three years after their last update","what the Stable Diffusion and DreamBooth model licences allow you to do with the videos you make"]},"zh-Hans":{"summaryShort":"研究代码：用一段短视频微调 Stable Diffusion，再按新的文字提示重做这段画面。","summary":"Tune-A-Video 是 2023 年一篇研究论文（ICCV）的配套代码。你给它一段短视频和一句描述，它会在一块性能较强的 NVIDIA 显卡上微调 Stable Diffusion 图像模型约 10 到 15 分钟。之后你就可以让同样的动作换个主体或风格，例如把 \"a man is skiing\" 变成 \"Spider Man is skiing on the beach\"。整个过程都在终端和 Python 里完成；没有图形界面，项目自 2023 年 10 月起就没有再改动。","easeRationale":"文档只描述了 Linux 的配置方式：安装 Python 包，自己从 Hugging Face 下载 Stable Diffusion 权重，修改 YAML 配置，然后在 A100 这类显卡上运行 `accelerate launch train_tuneavideo.py`。之后要产出视频，还得照 README 里的片段自己写 Python 脚本；文档也给了一个 Colab 笔记本链接，可以不装任何东西先试试。","costNote":"代码采用 Apache-2.0。它所依赖的 Stable Diffusion 模型有各自的许可，训练还需要租用或自备 GPU。","notVerified":["2022 年的这些固定版本依赖，在当前的系统和显卡驱动上还能否安装并运行","实际需要多少显存，文中只给出了在一块 A100 上耗时 10 到 15 分钟","当你保留多份按视频微调的副本后，Stable Diffusion 权重和这些副本会占多少磁盘空间","文中链接的 Colab 笔记本和 Hugging Face 演示，在最后一次更新三年后是否还能用","Stable Diffusion 和 DreamBooth 的模型许可，允许你拿生成的视频做什么"]}}},{"id":"R_kgDOR9KaBA","slug":"shudesu__ig-harness-oss","name":"IG Harness","owner":"Shudesu","fullName":"Shudesu/ig-harness-oss","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"depends_on_meta_approval_and_your_account","severity":"attention","evidence":"README 的 \"Standard Accessの壁について\"：Meta 的 Standard Access 和 Advanced Access 取决于账号归属和 App Review，作者也写明不能保证所有功能在未过审的情况下都可用。健康检查接口甚至会按账号报告 \"凍結検知\"（封禁检测）和令牌过期时间，因此一旦 Meta 限制了账号或应用，服务就可能停摆。","source":"ai","title":"依赖 Meta 的审核和你的账号状态"},{"code":"secrets_stored_in_plain_text_by_default","severity":"attention","evidence":".env.example 的 TOKEN_ENCRYPTION_KEY：\"Without it, app secrets / LINE keys stay plaintext, warn once per isolate, and their scheduled encryption migrations are skipped.\" 在示例文件里这个密钥是可选项，因此跳过它的部署会把 Instagram 和 LINE 的凭据以未加密的形式留在数据库里。","source":"ai","title":"密钥默认以明文保存"},{"code":"stores_your_followers_personal_data","severity":"attention","evidence":"README 的 CRM 部分：粉丝会通过 webhook 自动登记，包含资料、自定义字段、标签、表单回答、点击统计和完整的对话记录，全部存放在你的 Cloudflare D1 数据库里。.env.example 把 MESSAGES_RETENTION_DAYS 设为 0，即 \"preserves existing history\"——除非你改动，否则消息会一直保留。","source":"ai","title":"会保存你粉丝的个人数据"},{"code":"ai_assistant_gets_full_control","severity":"attention","evidence":"README 的 \"運用上の権限と更新\"：\"MCPへowner/adminキーを渡すと、DM送信・一斉配信を含む管理権限を渡します\"——把 owner 或 admin 密钥交给 MCP 服务器，就等于把发送私信和群发的管理权限一并交出；只有设置 IG_HARNESS_READ_ONLY=true 才能屏蔽写入类工具。","source":"ai","title":"AI 助手会获得完全控制权"},{"code":"published_releases_lag_the_code","severity":"info","evidence":"package.json 写的是 0.12.0，但 GitHub 上最新的发布版是 2026-04-25 的 v0.4.6，而最后一次提交在 2026-11-03。只看发布页的人，比仓库里的内容落后半年。","source":"ai","title":"发布版本落后于代码"},{"code":"supporting_material_written_by_the_author","severity":"info","evidence":"README 的 \"公式情報・検証資料\"：它引用的研究站点、目录和 wiki 都由开发者自己运营，文中也直说：\"各Researchサイトは開発元が運営する一次情報であり、独立した第三者レビューではありません\"（这些不是独立第三方评测）。对比表格里，付费竞品只以 \"A社\" 和 \"B社\" 代称。","source":"ai","title":"佐证材料由作者本人撰写"},{"code":"main_documentation_in_japanese","severity":"info","evidence":"主 README、安装视频和 wiki 都是日文；仓库里虽然有英文、西班牙文、韩文和中文的 README，但 docs/ 下的详细指南和所链接的 wiki 文章都是日文。","source":"ai","title":"主要文档只有日文"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["Meta / Instagram Graph API (access token and app secret)","Cloudflare account (Workers, D1, R2, Pages)","LINE (only to link Instagram followers with LINE friends)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["social_marketing","automation","ai_skills"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[],"stars":63,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":63},{"date":"2026-11-06","stars":64},{"date":"2026-11-07","stars":63},{"date":"2026-11-08","stars":63},{"date":"2026-11-09","stars":63},{"date":"2026-11-10","stars":63}],"firstPublishedAt":"2026-11-10T23:52:44.077Z","commitSha":"cecad5861cb306978c3aa8fa37e0e9f4471ab79c","verifiedAt":"2026-11-10T23:42:13.748Z","lastCommitAt":"2026-11-03T18:19:05Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Instagram DM automation you host yourself on Cloudflare: comments trigger private messages, follow checks and follow-ups.","summary":"IG Harness automates Instagram direct messages for a business or creator account. Someone comments on your post, the tool sends them a DM, checks whether they follow you, hands over the promised file or link, then sends follow-up messages on a delay. It also keeps a contact list with tags and notes, replies from a dashboard, tracks link clicks and can post content. One command deploys it to your own free Cloudflare account; you need Node 22, an Instagram professional account and a Meta developer app. A bundled MCP server lets Claude Code drive it.","easeRationale":"Same on all three systems, with Linux no harder than the others: there is nothing to download and open. You need Node 22 and pnpm, then `npx create-ig-harness` logs into Cloudflare, creates the database and storage and deploys the service and dashboard; before that you have to set up an Instagram professional account, a Meta developer app with webhooks, and several secrets by hand. The program itself then runs on Cloudflare, not on your computer.","costNote":"The software costs nothing (MIT). The README says the Cloudflare account and its usage, a domain, and any AI service you connect are paid separately, and that Cloudflare's free D1 plan starts returning errors once the daily read and write limits are reached.","notVerified":["whether Meta grants the permissions your use needs, and how long approval takes","whether a real account stays within Cloudflare's free limits once message volume grows","whether the automated messages follow Instagram's current rules for your country and account type","what the npm packages (create-ig-harness, the SDK and the MCP server) do when they run, compared with the code in this repository","how well the one-command setup copes when a step fails halfway through"]},"zh-Hans":{"summaryShort":"部署在你自己 Cloudflare 账号上的 Instagram 私信自动化：评论触发私信、检查是否关注并自动跟进。","summary":"IG Harness 为企业号或创作者账号自动处理 Instagram 私信。有人在你的帖子下留言，它就给对方发私信，检查对方是否关注你，把承诺的文件或链接发过去，之后再按设定的间隔发送跟进消息。它还会维护带标签和备注的联系人列表，支持在后台面板里回复、统计链接点击，并能发布内容。一条命令就能部署到你自己的免费 Cloudflare 账号；你需要 Node 22、一个 Instagram 专业账号和一个 Meta 开发者应用。它还自带 MCP 服务器，可以让 Claude Code 来操作。","easeRationale":"三个系统上的做法相同，Linux 也不会更难：没有可下载直接打开的程序。你需要 Node 22 和 pnpm，然后用 `npx create-ig-harness` 登录 Cloudflare，创建数据库和存储，并部署服务与面板；在此之前，还得手动准备 Instagram 专业账号、带 webhook 的 Meta 开发者应用以及若干密钥。之后程序运行在 Cloudflare 上，而不是你的电脑上。","costNote":"软件本身免费（MIT）。README 说 Cloudflare 账号及其用量、域名，以及你接入的任何 AI 服务都要另外付费，并提醒 Cloudflare 免费版 D1 一旦达到每日读写上限就会开始返回错误。","notVerified":["Meta 是否会批准你这种用法所需的权限，以及审核要多久","当消息量增大后，一个真实账号能否仍停留在 Cloudflare 的免费额度内","这些自动消息是否符合 Instagram 针对你所在国家和账号类型的现行规则","npm 上的包（create-ig-harness、SDK 和 MCP 服务器）运行时做了什么，与本仓库的代码是否一致","一条命令的安装流程在中途某一步失败时处理得如何"]}}},{"id":"R_kgDOJKSTjA","slug":"significant-gravitas__autogpt","name":"AutoGPT","owner":"Significant-Gravitas","fullName":"Significant-Gravitas/AutoGPT","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"B"},"risk":"attention","signals":[{"code":"platform_folder_not_open_source_license","severity":"attention","evidence":"LICENSE：\"Everything inside the autogpt_platform folder is under the Polyform Shield License. Everything outside the autogpt_platform folder is under the MIT License.\" README.md：Polyform Shield 是 \"Free for personal and internal business use; cannot be sold as a competing hosted service\"。Polyform 文本还增加了一条禁止竞争条款：\"Any purpose is a permitted purpose, except for providing any product that competes with the software\"。","source":"ai","title":"平台文件夹不是开源许可证"},{"code":"readme_promotes_paid_hosted_service","severity":"info","evidence":"README.md 多次出现带有跟踪参数（utm_source=github）的 \"Sign up for a trial\" 链接，并声明 \"The hosted Platform is a paid service.\"","source":"ai","title":"README 推广付费托管服务"},{"code":"agents_act_on_connected_accounts","severity":"info","evidence":"README.md：\"Each Expert has a role you can shape with your instructions, context, skills, and connected apps\" 以及 \"You choose which accounts each Expert can access, set spending controls\"。","source":"ai","title":"智能体可在已连接的账户上执行操作"},{"code":"platform_releases_marked_beta","severity":"info","evidence":"版本标签本身就写着 beta：autogpt-platform-beta-v0.10.0（2026-10-27）和 autogpt-platform-beta-v0.9.0（2026-10-23）。LICENSE 也称它为 \"our in-developlemt platform\"。","source":"ai","title":"平台版本标记为测试版"},{"code":"marketplace_workflows_from_strangers","severity":"info","evidence":"README.md：\"Marketplace: discover workflows you can adapt to your work.\" 你导入的工作流是别人写的，可能会使用你已连接的应用和密钥。","source":"ai","title":"市场中的工作流来自陌生人"}],"platforms":["windows","mac","linux","web"],"tools":["docker"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["AI model provider credentials for self-hosting (or a local model)"],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["ai_agents","automation","productivity"],"alternativeTo":[{"slug":"zapier","name":"Zapier","confirmed":true},{"slug":"make","name":"Make","confirmed":true},{"slug":"n8n-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":187985,"downloads":3733,"downloadsWeek":0,"downloadsDays":0,"growth":95,"growthDays":7,"history":[{"date":"2026-10-26","stars":187779},{"date":"2026-10-27","stars":187791},{"date":"2026-10-28","stars":187804},{"date":"2026-10-29","stars":187819},{"date":"2026-10-30","stars":187840},{"date":"2026-10-31","stars":187850},{"date":"2026-11-01","stars":187860},{"date":"2026-11-02","stars":187882},{"date":"2026-11-03","stars":187890},{"date":"2026-11-04","stars":187902},{"date":"2026-11-05","stars":187912},{"date":"2026-11-06","stars":187928},{"date":"2026-11-07","stars":187933},{"date":"2026-11-08","stars":187950},{"date":"2026-11-09","stars":187972},{"date":"2026-11-10","stars":187985}],"firstPublishedAt":"2026-10-27T01:10:12.685Z","commitSha":"07212df377c4ea0ac5ed6a7c9c8ff72203564650","verifiedAt":"2026-10-30T01:44:22.950Z","lastCommitAt":"2026-11-05T21:18:08Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Set up a team of AI assistants that write content, research prospects and handle recurring work, hosted or on your own computer.","summary":"AutoGPT is a platform where AI \"Experts\" (for content, sales research, email drafts and more) take a plain-English brief, work together under a coordinator called Otto, and can run routines and visual workflows connected to your apps. You can pay for the hosted version in a browser, or run it yourself with Docker and your own AI model credentials or a local model. It is for small teams and power users who are comfortable with some setup. The README mostly promotes the paid hosted service.","easeRationale":"The easiest route is the maker's hosted Platform in a browser: nothing to install, but you sign up for a paid service with a trial. Running it yourself on Windows, Mac or Linux is grade D: the README's quickest path is a long `docker run` command followed by `docker exec autogpt autogpt-admin promote you@example.com`, and even the one-command installer it mentions sets up the same Docker image.","costNote":"Self-hosting is free, but you pay for your own servers and AI model usage (or use a local model). The hosted Platform is a paid service with a trial; prices are on agpt.co/pricing, not in the package.","notVerified":["Whether the one-command installer works smoothly on Windows and what it installs besides Docker","Real disk and memory use of the self-hosted container (the check could not estimate disk use)","Real AI model cost when the Experts run routines on their own","What the agents can do in the accounts you connect and how well spending controls work","How the self-hosted version differs from the hosted one in features"]},"zh-Hans":{"summaryShort":"组建一支AI助手团队来撰写内容、调研潜在客户并处理重复性工作,可以用托管版,也可以在自己电脑上运行。","summary":"AutoGPT 是一个平台,AI\"专家\"(负责内容、销售调研、邮件草稿等)接收一份用简单语言写的任务说明,在一个叫Otto的协调者下协同工作,还可以运行连接到你各类应用的例行任务和可视化工作流。你可以在浏览器里付费使用托管版,也可以用Docker和自己的AI模型凭证或本地模型自己运行。它面向需要一些配置经验的小团队和高级用户。README主要是在推广付费托管服务。","easeRationale":"最简单的方式是使用开发者的浏览器托管平台:不需要安装任何东西,但你要注册一个带试用期的付费服务。自己在Windows、Mac或Linux上运行则是D级:README里最快的方式是一条很长的`docker run`命令,后面还要跟上`docker exec autogpt autogpt-admin promote you@example.com`,就连文档提到的一键安装程序,最终配置的也是同一个Docker镜像。","costNote":"自己运行是免费的,但你要承担自己服务器和AI模型使用的费用(或者使用本地模型)。托管平台是带试用期的付费服务;价格在agpt.co/pricing上,而不在这个代码包里。","notVerified":["一键安装程序在Windows上是否能顺利运行,以及除了Docker之外它还安装了什么","自行托管容器的真实磁盘和内存占用(检查工具无法估算磁盘占用)","专家自行运行例行任务时真实的AI模型费用","代理在你连接的账户里能做什么,以及支出控制机制效果如何","自行托管版本和托管版本在功能上有什么区别"]}}},{"id":"R_kgDOVQBDQw","slug":"simbasjungles__free-my-tv","name":"Free My TV","owner":"simbasjungles","fullName":"simbasjungles/free-my-tv","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"9 天内获得 5238 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"tv_adb_stays_enabled","severity":"attention","evidence":"README.md，How it works：\"The app connects to 127.0.0.1:5555 with its own key ... and runs the same commands you would: cmd package set-home-activity, ... pm disable-user\"。切换回原状时，看门狗（watchdog）和开机检查都依赖 ADB，因此电视的调试端口 5555 会一直开启，并且在你的家庭网络中可以访问。","source":"ai","title":"电视的 ADB 保持开启"},{"code":"ai_assistant_gets_shell_access","severity":"info","evidence":"README.md，The easy way：让 AI 编程助手来安装它，助手会 \"downloads the code, installs the tools it needs, builds the app and puts it on your TV\"，并在你的电脑上运行命令。","source":"ai","title":"AI 助手获得命令行权限"},{"code":"self_updating_apk","severity":"info","evidence":"README.md：\"Once a week the APK from the release looks for a newer release and downloads it\"；它通过 Android 的安装程序进行安装，安装前会先询问，前提是用户已允许 Free My TV 安装应用。","source":"ai","title":"APK 会自动更新"},{"code":"play_protect_warning_bypass","severity":"info","evidence":"README.md：\"On Android 14, Google Play Protect may hold the install with Risky app blocked: choose More details → Install anyway.\"","source":"ai","title":"绕过 Play Protect 警告"},{"code":"accessibility_and_notification_access","severity":"info","evidence":"README.md，What it changes：在 TCL 和其他品牌的电视上，一个 \"small accessibility service that only reads those buttons\" 负责处理遥控器按键，可选的 Now Playing 小组件则使用通过电视 ADB 授予权限的通知监听器。","source":"ai","title":"无障碍和通知权限"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["TMDB","Watchmode","Anthropic (only for the translation helper tools/translate.py)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","productivity"],"alternativeTo":[{"slug":"projectivy-launcher","name":"Launcher","confirmed":true}],"declaredAlternatives":[],"stars":6245,"downloads":35269,"downloadsWeek":7083,"downloadsDays":7,"growth":424,"growthDays":7,"history":[{"date":"2026-10-26","stars":4863},{"date":"2026-10-27","stars":5056},{"date":"2026-10-28","stars":5235},{"date":"2026-10-29","stars":5393},{"date":"2026-10-30","stars":5488},{"date":"2026-10-31","stars":5575},{"date":"2026-11-01","stars":5684},{"date":"2026-11-02","stars":5767},{"date":"2026-11-03","stars":5821},{"date":"2026-11-04","stars":5876},{"date":"2026-11-05","stars":5921},{"date":"2026-11-06","stars":5965},{"date":"2026-11-07","stars":6027},{"date":"2026-11-08","stars":6119},{"date":"2026-11-09","stars":6199},{"date":"2026-11-10","stars":6245}],"firstPublishedAt":"2026-10-27T01:10:12.685Z","commitSha":"aeffdb4c4faa781069ab5713890fa0b9efebb4ee","verifiedAt":"2026-10-29T01:56:29.679Z","lastCommitAt":"2026-11-09T23:57:56Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Replace your TV's ad-filled home screen with a simple one showing your apps, the weather and trending picks, and undo it anytime.","summary":"Free My TV is a replacement home screen for Google TV and Android TV, with app versions for LG and Samsung TVs. It shows your streaming apps, trending titles from those apps, the weather and optional crypto prices, with no ads. On Google TV it turns off (does not uninstall) the stock home screen and some preinstalled promo apps, and one setting brings them back. Installing needs a computer on the same Wi-Fi, the TV's developer mode and the adb tool, or an AI coding assistant to do it. Trending picks need a free TMDB key, which you add on the TV itself.","easeRationale":"There is a ready FreeMyTV.apk, but on Windows you must install adb, turn on developer mode and USB debugging on the TV, and type adb commands (or let an AI assistant do it); Mac and Linux are the same, and LG and Samsung need Python or a community installer.","costNote":"MIT license. The TMDB key for trending picks and the optional Watchmode key have free plans. The suggested 'easy way' uses an AI coding assistant such as Claude Code, which may need a paid plan.","notVerified":["Whether it works on your exact TV model and firmware","Whether 'Original' really brings back every app that was turned off","Whether a TV firmware update undoes the changes or breaks the home screen","Which commands an AI coding assistant actually runs when it follows the instructions","What the accessibility service and the in-app ADB connection do beyond what the README describes"]},"zh-Hans":{"summaryShort":"用一个干净的主屏幕替换电视上满是广告的主屏幕,展示你的应用、天气和热门推荐,随时可以撤销。","summary":"Free My TV 是 Google TV 和 Android TV 的替代主屏幕,也有适用于 LG 和三星电视的应用版本。它显示你的流媒体应用、这些应用中的热门内容、天气以及可选的加密货币价格,没有广告。在 Google TV 上,它会关闭(而不是卸载)原生主屏幕和部分预装的推广应用,一个设置选项可以把它们恢复回来。安装需要一台连接同一 Wi-Fi 的电脑、电视的开发者模式和 adb 工具,或者让 AI 编程助手代为完成。热门推荐需要一个免费的 TMDB 密钥,可以直接在电视上添加。","easeRationale":"有现成的 FreeMyTV.apk,但在 Windows 上你必须安装 adb,在电视上打开开发者模式和 USB 调试,并输入 adb 命令(或者让 AI 助手代劳);Mac 和 Linux 操作相同,LG 和三星则需要 Python 或社区提供的安装工具。","costNote":"MIT 许可证。用于热门推荐的 TMDB 密钥和可选的 Watchmode 密钥都有免费额度。推荐的'简便方式'会用到 AI 编程助手,比如 Claude Code,这可能需要付费计划。","notVerified":["是否适用于你具体的电视型号和固件版本","'Original'是否真的能恢复所有被关闭的应用","电视固件更新是否会撤销这些更改或破坏主屏幕","AI 编程助手在执行安装说明时实际运行了哪些命令","应用内的无障碍服务和 ADB 连接除了 README 所述之外还做了什么"]}}},{"id":"R_kgDOVS6VyQ","slug":"snapintosh__snapintosh","name":"Snapintosh","owner":"Snapintosh","fullName":"Snapintosh/Snapintosh","type":"framework_infra","status":"alert","grade":"E","easeByPlatform":{"mac":"E"},"risk":"high","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 4 天前","source":"check","title":"所有者账号较新"},{"code":"runs_macos_outside_apples_licence","severity":"high","evidence":"README 中写明了整个目的：\"Snapintosh boots Apple's macOS 27 (build 26A428) on an ASUS Zenbook A16 laptop with a Qualcomm Snapdragon X2 Elite\"，其自带的 Legal 部分也说 \"Running macOS on non-Apple hardware may conflict with Apple's macOS software licence agreement.\" 该监控程序提供 \"the SPTM and TXM monitor calls\"，让苹果的内核接受非苹果硬件，用户还必须用 tools/fetch-ipsw.py 获取苹果的恢复镜像。","source":"ai","title":"在苹果许可协议之外运行 macOS"},{"code":"can_destroy_data_on_the_laptop","severity":"attention","evidence":"README Warning：\"It writes the internal NVMe SSD. Installing macOS changes the partition table... A bug can corrupt or destroy data on the SSD, including the Windows installation\"，并且 \"programs clocks, voltages, the GPU and memory-bandwidth votes directly\"。","source":"ai","title":"可能毁掉笔记本上的数据"},{"code":"big_claims_almost_nobody_can_check","severity":"attention","evidence":"仓库创建于 2026-10-22，已有 1012 个星标，没有 GitHub 发布版本，README 说它只能在一台机器上运行：\"An ASUS Zenbook A16 with the Snapdragon X2 Elite X2E-96-100 (UX3607OA). Nothing else is supported.\" 再加上对创建仅 4 天的所有者账号的检查，几乎没有人能复现这些说法。","source":"ai","title":"宏大的说法几乎无人能验证"},{"code":"mixed_licences_in_one_project","severity":"info","evidence":"LICENSE 文件：\"Snapintosh is MIT with GPL-2.0 components\"，而使用 Adreno 表构建的二进制文件 \"is a combined work and is distributable under GPL-2.0\"，因此任何重新分发构建版本的人都要遵守 GPL-2.0，而不是 MIT。GitHub 本身把该许可证显示为 NOASSERTION。","source":"ai","title":"同一项目混用多种许可证"},{"code":"pulls_firmware_from_other_sources","severity":"info","evidence":"README \"Bring your own Apple software\"：GPU 微码（gen80100_gmu.bin、gen80100_sqe.fw）和 Wi-Fi 固件来自 linux-firmware，Wi-Fi 板文件来自华硕的 Windows 驱动包；这些工具会按固定的 SHA-256 检查每个文件。","source":"ai","title":"从其他来源获取固件"}],"platforms":["mac"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["network_system","hardware"],"alternativeTo":[{"slug":"parallels-desktop","name":"Parallels Desktop","confirmed":true}],"declaredAlternatives":[],"stars":1246,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":96,"growthDays":7,"history":[{"date":"2026-10-27","stars":1012},{"date":"2026-10-28","stars":1033},{"date":"2026-10-29","stars":1051},{"date":"2026-10-30","stars":1063},{"date":"2026-10-31","stars":1086},{"date":"2026-11-01","stars":1115},{"date":"2026-11-02","stars":1138},{"date":"2026-11-03","stars":1150},{"date":"2026-11-04","stars":1165},{"date":"2026-11-05","stars":1174},{"date":"2026-11-06","stars":1190},{"date":"2026-11-07","stars":1205},{"date":"2026-11-08","stars":1218},{"date":"2026-11-09","stars":1240},{"date":"2026-11-10","stars":1246}],"firstPublishedAt":"2026-10-27T18:40:28.046Z","commitSha":"940eeb95ba637528dc45ba098e6b8091163dd489","verifiedAt":"2026-10-28T01:38:38.686Z","lastCommitAt":"2026-10-25T04:18:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An experimental hypervisor that boots Apple's macOS on one specific ASUS Snapdragon laptop, driven from a Mac over USB-C.","summary":"A low-level monitor (hypervisor) plus build tools that make Apple's macOS 27 kernel start on an ASUS Zenbook A16 with a Snapdragon X2 Elite chip, including a hand-written graphics driver for the laptop's Adreno GPU, keyboard, touchpad and NVMe storage. Nothing else is supported. You need that exact laptop, a second Mac running macOS 27 connected by USB-C, Xcode, Apple's Kernel Debug Kit, Python 3 and several build tools, and you must supply Apple's macOS restore image yourself. The authors call it experimental and say to expect crashes.","easeRationale":"On macOS, the only system the tools run on, this is an E: there are no releases or ready files, and the Building section asks for Xcode with the macOS 27 SDK, Apple's Kernel Debug Kit for build 26A428, Python 3, Homebrew LLVM, Meson and Ninja, a second repository checked out next to this one, and a boot cycle driven over USB-C from the Mac.","costNote":"The project's own code is free (MIT, with GPL-2.0 parts). You supply the ASUS laptop, a Mac host, Apple's macOS restore image and the Kernel Debug Kit from Apple's developer downloads.","notVerified":["Whether it really boots to a working macOS desktop as the status table says; there are no releases and only one laptop model is supported.","Whether the pinned file hashes still match the firmware Apple and ASUS ship today.","Whether the battery and thermal guards prevent damage to the laptop; the README says they are \"not a guarantee\".","How much disk space the build, the restore image and the Kernel Debug Kit take together.","Whether Apple's kernel really runs unmodified in practice, since the docs also describe opt-in kernel patches."]},"zh-Hans":{"summaryShort":"一个实验性的虚拟机监控程序,可在某款特定的华硕骁龙笔记本上启动苹果 macOS,由一台 Mac 通过 USB-C 驱动控制。","summary":"这是一个底层监控程序(虚拟机监控程序)加上一套编译工具,能让苹果 macOS 27 的内核在搭载骁龙 X2 Elite 芯片的华硕 Zenbook A16 上启动,其中包括为该笔记本的 Adreno GPU、键盘、触控板和 NVMe 存储手写的显卡驱动。除此之外不支持任何其他设备。你需要这台特定型号的笔记本、一台通过 USB-C 连接并运行 macOS 27 的第二台 Mac、Xcode、苹果的 Kernel Debug Kit、Python 3 以及若干编译工具,并且必须自行提供苹果的 macOS 还原镜像。作者称这是实验性项目,并提醒会出现崩溃。","easeRationale":"在 macOS 上(这些工具唯一能运行的系统),评级是 E:没有发布版本,也没有现成文件,编译部分要求安装带 macOS 27 SDK 的 Xcode、适用于 26A428 版本的苹果 Kernel Debug Kit、Python 3、Homebrew LLVM、Meson 和 Ninja,还需要在旁边克隆第二个仓库,并通过 USB-C 从 Mac 驱动整个启动流程。","costNote":"该项目自身的代码是免费的(MIT 许可,部分为 GPL-2.0)。你需要自行提供华硕笔记本、作为主机的 Mac、苹果的 macOS 还原镜像,以及来自苹果开发者下载页面的 Kernel Debug Kit。","notVerified":["它是否真的能像状态表所说的那样启动到可用的 macOS 桌面;目前没有发布版本,且只支持一种笔记本型号。","固定的文件哈希是否仍与苹果和华硕目前发布的固件相符。","电池和温度保护措施是否真的能防止笔记本受损;README 说这些措施\"并非保证\"。","编译过程、还原镜像和 Kernel Debug Kit 加在一起需要多少磁盘空间。","苹果内核在实际中是否真的未经修改运行,因为文档中也提到了可选的内核补丁。"]}}},{"id":"R_kgDOVlsLZA","slug":"sostenesfreitas__ocarina-recomp","name":"Ocarina Recomp","owner":"sostenesfreitas","fullName":"sostenesfreitas/ocarina-recomp","type":"needs_setup","status":"alert","grade":"D","easeByPlatform":{"windows":"D"},"risk":"high","signals":[{"code":"needs_copy_of_commercial_game","severity":"attention","evidence":"README.md写道:\"Este repositório não contém o jogo, keys nem executável, e não explica como obtê-los. Você precisa da sua própria cópia\",以及\"Sua cópia do jogo na versão 1.1.1\"。这个程序是从受保护的商业主机游戏的dump文件构建出来的,README里还提到了\"keys\";从主机里提取这些文件在不少国家受限,项目本身也说明它产生的内容不能分享。","source":"ai","title":"需要一份商业游戏的拷贝"},{"code":"derived_from_nintendo_game","severity":"attention","evidence":"README.md写道:\"Projeto de fãs, sem nenhuma relação com a Nintendo. The Legend of Zelda e Ocarina of Time são marcas da Nintendo.\" 编译链使用了\"um Ryujinx com patches (tools/xlate/patches)\"来转换着色器。任天堂曾对粉丝移植项目和Ryujinx模拟器采取过行动,所以这个项目也可能被下架或面临法律压力。","source":"ai","title":"衍生自任天堂的游戏"},{"code":"unsigned_installer_antivirus_warnings","severity":"info","evidence":"README.md写道:\"O executável não tem assinatura digital, então o Windows pode avisar. Clique em Mais informações e depois em Executar assim mesmo\",以及\"Alguns antivírus reclamam de programas novos sem assinatura e de scripts do PowerShell\"。安装程序运行PowerShell时带有-ExecutionPolicy Bypass参数。","source":"ai","title":"安装程序未签名,杀毒软件会报警"},{"code":"downloads_code_during_install","severity":"info","evidence":"README.md写道:\"Baixa os pacotes Python (tools\\requirements-build.txt) e .NET do tradutor de shaders e, se o executável do jogo estiver comprimido, o zbic\"。安装期间需要联网,所以在你电脑上运行的一部分内容是安装时才下载下来的。","source":"ai","title":"安装过程中会下载代码"},{"code":"very_new_and_barely_tested","severity":"attention","evidence":"该仓库创建于2026-11-06,已经有13个发布版本。README.md写道:\"Experimental. Só foi testado em um PC (Ryzen 9 9950X3D, RTX 5090), na região do início do jogo\",并把\"Hyrule Field e o resto do jogo\"以及\"Placas AMD e Intel\"标记为未测试。","source":"ai","title":"非常新,几乎没怎么测试过"},{"code":"rebuilds_nintendo_game_from_console_files","severity":"high","evidence":"README.md写道:\"O código do jogo é traduzido para C e compilado como um programa do Windows. Os gráficos do Switch 2 são convertidos para Vulkan. Tudo isso acontece no seu PC, a partir do seu próprio arquivo do jogo.\" 它需要\"Sua cópia do jogo na versão 1.1.1\",可以选择\"o `.nup` do seu jogo (ou uma pasta com `exefs` e `romfs`)\",README还声明\"Este repositório não contém o jogo, keys nem executável, e não explica como obtê-los\"以及\"não compartilhe\"。编译链使用\"um Ryujinx com patches (tools/xlate/patches)\"来转换主机的着色器,README最后写道\"Projeto de fãs, sem nenhuma relação com a Nintendo. The Legend of Zelda e Ocarina of Time são marcas da Nintendo.\" 从主机中取出游戏和密钥,意味着绕过其版权保护机制,而许多国家的法律都限制这种绕过行为。","source":"ai","title":"用主机文件重建任天堂的游戏"}],"platforms":["windows"],"tools":[],"buildTools":[],"disk":{"minMB":110,"maxMB":331,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":1078,"downloads":47835,"downloadsWeek":12563,"downloadsDays":2,"growth":1078,"growthDays":4,"history":[{"date":"2026-11-08","stars":947},{"date":"2026-11-09","stars":1060},{"date":"2026-11-10","stars":1078}],"firstPublishedAt":"2026-11-07T22:56:16.927Z","commitSha":"c4896f8ed5eb122510f36001ae34607d9c7af568","verifiedAt":"2026-11-07T22:37:58.073Z","lastCommitAt":"2026-11-09T22:22:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your own copy of Ocarina of Time into a Windows program built on your PC, with no emulator.","summary":"A fan project that translates the game's code into C and compiles it as a native Windows program, converting the console graphics to Vulkan. It is for people who already own the game file (version 1.1.1 only) and want to play it without an emulator. You download a ZIP installer, point it at your game file and wait 30 minutes to 2 hours while it builds on your PC. It needs Windows 10 or 11 64-bit, a graphics card with Vulkan 1.3 and mesh shaders, 16 GB of RAM or more and about 40 GB free. The repository includes no game files.","easeRationale":"On Windows the ZIP installer needs no administrator rights, but you must supply your own game file, have a graphics card with Vulkan 1.3 and mesh shaders (the README recommends NVIDIA and says AMD and Intel were never tested), raise the Windows paging file on 16 GB machines and leave the PC building for 30 minutes to 2 hours. Linux has scripts (compilar.sh), but the README points to docs/building.md as \"linux-in-progress\", so no grade is given there.","notVerified":["Whether the build finishes and the game runs on anything other than the one PC the author tested (RTX 5090); AMD and Intel cards were never tried.","How much of the game is playable past the opening area.","Real disk and memory use: the README claims about 40 GB free and up to 24 GB of memory on the first run.","What the Python, .NET and zbic downloads bring to your PC during installation.","Whether the installer and scripts only write inside the project folder, as the README states."]},"zh-Hans":{"summaryShort":"把你自己的《塞尔达传说:时之笛》游戏文件,在你的电脑上编译成Windows程序,不需要模拟器。","summary":"这是一个粉丝项目,把游戏代码翻译成C语言并编译成原生Windows程序,同时把主机图形转换成Vulkan。它面向已经拥有游戏文件(仅限1.1.1版本)、想不用模拟器就玩的人。你下载一个ZIP安装程序,指向自己的游戏文件,然后等30分钟到2小时,让它在你的电脑上完成编译。需要64位Windows 10或11、支持Vulkan 1.3和mesh shaders的显卡、16 GB以上内存,以及大约40 GB空闲空间。仓库里不包含任何游戏文件。","easeRationale":"在Windows上,这个ZIP安装程序不需要管理员权限,但你得提供自己的游戏文件,显卡要支持Vulkan 1.3和mesh shaders(README推荐NVIDIA,并说从未测试过AMD和Intel),16 GB内存的电脑需要调大Windows分页文件,还要让电脑编译30分钟到2小时。Linux有脚本(compilar.sh),但README把docs/building.md标注为\"linux-in-progress\",所以这里不给出评级。","notVerified":["编译能否完成,游戏能否在作者测试用机(RTX 5090)之外的电脑上运行;从没测试过AMD和Intel显卡。","开场区域之外,还有多少内容是可玩的。","真实的磁盘和内存占用:README声称大约需要40 GB空闲空间,第一次运行时内存最多用到24 GB。","安装过程中下载的Python、.NET和zbic会给你的电脑带来什么。","安装程序和脚本是否只写入项目文件夹内部,正如README所说。"]}}},{"id":"R_kgDORpCz1g","slug":"stablyai__orca","name":"Orca","owner":"stablyai","fullName":"stablyai/orca","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"233 天内获得 116609 颗星(约每天 500 颗)","source":"check","title":"星标增长速度异常"},{"code":"very_new_project_with_huge_star_count","severity":"info","evidence":"GitHub 数据显示该仓库创建于 2026-03-17,已经拥有 116609 颗星和 1001 个发行版;README 写道\"we ship daily, so this list is perpetually behind\"。","source":"ai","title":"项目非常新却星标数巨大"},{"code":"anonymous_usage_data_collected","severity":"info","evidence":"README,Community & Support 部分:\"See the privacy & telemetry docs for what anonymous usage data Orca collects and how to opt out.\"","source":"ai","title":"会收集匿名使用数据"},{"code":"broad_control_of_your_computer","severity":"attention","evidence":"README 中列出了\"Computer Use — Let agents operate desktop apps and visible UI\"、在远程机器上运行代理的\"SSH Worktrees\",以及带有\"click\"和\"fill\"命令的\"Orca CLI\",因此这些助手可以操作你的电脑,也可以操作你连接的服务器。","source":"ai","title":"对你的电脑有广泛的控制权"},{"code":"android_app_outside_the_play_store","severity":"attention","evidence":"README,Mobile Companion 部分:Android 版本是从 www.onorca.dev 提供的\"[Download APK]\",并附有单独的\"Install guide\";只有 iOS 版本来自应用商店。","source":"ai","title":"android 应用来自应用商店之外"},{"code":"many_open_issues","severity":"info","evidence":"GitHub 数据显示有 8301 个未解决的 issue,同时还有 116609 颗星和 7374 次 fork。","source":"ai","title":"未解决的 issue 很多"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":200,"maxMB":599,"basis":"installer"},"apiKeys":["An account for at least one coding assistant (for example Claude Code, Codex or Grok)","GitHub","Linear"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents","productivity"],"alternativeTo":[{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"conductor","name":"Conductor","confirmed":true},{"slug":"warp","name":"Warp","confirmed":true}],"declaredAlternatives":[],"stars":121340,"downloads":168788358,"downloadsWeek":21501148,"downloadsDays":7,"growth":6830,"growthDays":7,"history":[{"date":"2026-10-25","stars":105924},{"date":"2026-10-26","stars":107076},{"date":"2026-10-27","stars":108077},{"date":"2026-10-28","stars":109057},{"date":"2026-10-29","stars":110199},{"date":"2026-10-30","stars":111129},{"date":"2026-10-31","stars":111742},{"date":"2026-11-01","stars":112629},{"date":"2026-11-02","stars":113621},{"date":"2026-11-03","stars":114510},{"date":"2026-11-04","stars":115364},{"date":"2026-11-05","stars":116578},{"date":"2026-11-06","stars":117737},{"date":"2026-11-07","stars":118511},{"date":"2026-11-08","stars":119216},{"date":"2026-11-09","stars":120301},{"date":"2026-11-10","stars":121340}],"firstPublishedAt":"2026-10-25T23:19:28.267Z","commitSha":"c75538441d1db6cb4821e945dfb55f313ddfd2a4","verifiedAt":"2026-11-05T22:47:12.700Z","lastCommitAt":"2026-11-10T23:13:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Desktop app to run several AI coding assistants at the same time, each on its own copy of your project.","summary":"Orca is a desktop program that puts many AI coding assistants (Claude Code, Codex, Grok, OpenCode and others) in one window. Each one works in its own separate copy of your project, so you can compare the results and keep the best. It also has a built-in editor, terminals, GitHub and Linear screens, and a phone app to follow the work from away. It is for people who already use these coding assistants: you install the assistant and its account yourself, Orca only drives them.","easeRationale":"On Windows there is a signed installer (orca-windows-setup.exe, 200 MB in v1.4.250), and macOS and Linux have .dmg, .zip, .deb, .rpm and AppImage builds, so the app itself is a download-and-open. The grade is C because Orca is only an orchestrator: you still have to install Node and at least one coding assistant CLI from the terminal and sign in to it before anything runs.","costNote":"Orca itself is free and MIT-licensed, but the repository description says it runs the coding agents \"with your own subscription\" (Claude, Codex and similar paid plans).","notVerified":["Whether the app really works with each of the 30-plus listed assistants, since most are third-party programs.","Whether the Windows, macOS and Linux builds install and run as described.","How much disk space the worktrees take after you copy a project several times; the v1.4.250 download files run from 149 to 276 MB and the checks estimate 200 to 599 MB of disk.","What the mobile relay under cloud/ sends through the internet when you pair the phone app.","What the paid assistant plans cost in real use, since Orca only drives accounts you pay for."]},"zh-Hans":{"summaryShort":"一个桌面应用,可以同时运行多个 AI 编程助手,每个助手都在你项目的独立副本上工作。","summary":"Orca 是一款桌面程序,把多个 AI 编程助手(Claude Code、Codex、Grok、OpenCode 等)集中在一个窗口里。每个助手都在你项目的独立副本上工作,这样你就可以比较结果,留下最好的那个。它还内置了编辑器、终端、GitHub 和 Linear 界面,以及一个手机应用,方便你在外面查看工作进度。它适合已经在用这些编程助手的人:助手本身和对应账号由你自己安装、登录,Orca 只是负责驱动它们。","easeRationale":"在 Windows 上有一个签名的安装程序(orca-windows-setup.exe,v1.4.250 版本 200 MB),macOS 和 Linux 则提供 .dmg、.zip、.deb、.rpm 和 AppImage 构建版本,所以应用本身下载后就能打开使用。评分是 C,因为 Orca 只是一个调度器:你仍然需要在终端里安装 Node 和至少一个编程助手的 CLI,并登录之后才能让一切运转起来。","costNote":"Orca 本身是免费的,采用 MIT 许可,但仓库描述中说它运行编程代理时\"with your own subscription\"(即 Claude、Codex 等类似的付费套餐)。","notVerified":["这款应用是否真的能配合列出的 30 多个助手逐一正常工作,因为其中大多数是第三方程序","Windows、macOS 和 Linux 的构建版本是否真的能按描述安装并运行","多次复制项目之后,worktree 会占用多少磁盘空间;v1.4.250 的下载文件大小在 149 到 276 MB 之间,检查估算磁盘占用为 200 到 599 MB","当你配对手机应用时,cloud/ 目录下的移动中继会通过互联网发送什么内容","这些付费助手套餐在实际使用中要花多少钱,因为 Orca 只是驱动你自己付费的账号"]}}},{"id":"R_kgDOVQJmWw","slug":"stackedlol__ascii-sharks","name":"ASCII Sharks","owner":"stackedlol","fullName":"stackedlol/ascii-sharks","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"very_new_and_very_small","severity":"info","evidence":"仓库创建于 2026-10-19，最后一次提交在 2026-10-24，顶层只有 .gitignore、LICENSE、README.md、demo.ts、index.html、package.json、src/ 和 tsconfig.json，却已有 1141 个星标。","source":"ai","title":"非常新且非常小"},{"code":"not_published_as_a_package","severity":"info","evidence":"package.json 中有 \"private\": true、版本 0.1.0 和 \"main\": \"src/index.ts\"，并且没有 GitHub 发布版本，所以你需要把源代码复制到自己的项目里，而不是直接安装。","source":"ai","title":"没有作为软件包发布"},{"code":"almost_no_documentation","severity":"info","evidence":"整个 README 只有 ASCII 图案、一行说明、一条安装命令和三行代码示例；提到的唯一生物是 \"hammerhead\"，也没有给出选项列表。","source":"ai","title":"几乎没有文档"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","image"],"alternativeTo":[],"declaredAlternatives":[],"stars":1497,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":108,"growthDays":7,"history":[{"date":"2026-10-27","stars":1140},{"date":"2026-10-28","stars":1165},{"date":"2026-10-29","stars":1188},{"date":"2026-10-30","stars":1207},{"date":"2026-10-31","stars":1219},{"date":"2026-11-01","stars":1264},{"date":"2026-11-02","stars":1345},{"date":"2026-11-03","stars":1389},{"date":"2026-11-04","stars":1410},{"date":"2026-11-05","stars":1436},{"date":"2026-11-06","stars":1456},{"date":"2026-11-07","stars":1468},{"date":"2026-11-08","stars":1477},{"date":"2026-11-09","stars":1485},{"date":"2026-11-10","stars":1497}],"firstPublishedAt":"2026-10-27T18:40:28.055Z","commitSha":"a0074c16a106f0abc665fe9712ca445f8f98ad8e","verifiedAt":"2026-10-28T01:39:09.860Z","lastCommitAt":"2026-10-24T16:07:31Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A small TypeScript code library that draws animated ASCII fish and sharks swimming on a web page.","summary":"Code you drop into your own web project to draw fish and sharks made of text characters, animated on an HTML canvas. You create an ocean engine, tell it which creatures and sizes you want, and it animates them; the README says it has no outside dependencies. It is not a program you open: to see the included demo you install Node, run two commands and open the local page. Written in TypeScript and meant for people building a web page or a toy project. Free, MIT-licensed, and only a few days old.","easeRationale":"Windows, Mac and Linux all land on C for the same reason: the only instruction is `npm install && npm run dev`, so you need Node installed and a terminal, and what you get is a developer preview page rather than an app. There are no releases and nothing ready to download.","notVerified":["Which creatures, sizes and settings the engine actually accepts, beyond the \"hammerhead\" shown in the README.","Whether the animation stays smooth with many creatures on screen, and how much CPU it uses.","Which browsers it works in.","Whether the demo runs as written, since there is no test or build output in the repository."]},"zh-Hans":{"summaryShort":"一个小型 TypeScript 代码库,可在网页上绘制游动的 ASCII 字符鱼和鲨鱼动画。","summary":"这是一段可以嵌入你自己网页项目的代码,用文本字符绘制鱼和鲨鱼,并在 HTML canvas 上制作动画。你创建一个海洋引擎,告诉它你想要哪些生物和尺寸,它就会让它们动起来;README 说它没有外部依赖。它不是一个可以打开的程序:要查看自带的演示,你需要安装 Node,运行两条命令,再打开本地页面。该项目用 TypeScript 编写,面向正在搭建网页或小项目的人。免费,采用 MIT 许可,只有几天历史。","easeRationale":"Windows、Mac 和 Linux 都被评为 C,原因相同:唯一的操作说明是 `npm install && npm run dev`,所以你需要安装好 Node 并打开终端,得到的只是一个开发者预览页面,而不是一个应用程序。目前没有发布版本,也没有现成的可下载内容。","notVerified":["这个引擎实际支持哪些生物、尺寸和设置,而不只是 README 中展示的“hammerhead”。","屏幕上生物很多时动画是否依然流畅,以及会占用多少 CPU。","它在哪些浏览器中可用。","演示是否真的能按原样运行,因为仓库中没有测试或构建产物。"]}}},{"id":"R_kgDOO4dzqA","slug":"startrail-org__leann","name":"LEANN","owner":"StarTrail-org","fullName":"StarTrail-org/LEANN","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"D","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把远程脚本直接灌进 shell：curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"安装时直接下载并执行脚本"},{"code":"reads_your_most_private_files","severity":"attention","evidence":"README：自带的应用会索引 Apple Mail、`~/Library/Messages/chat.db`（iMessage）、Chrome 浏览历史、微信消息以及 ChatGPT 和 Claude 的导出记录，其中两个还要求你 \"grant full disk access to your terminal/VS Code in System Preferences → Privacy & Security → Full Disk Access\"。这项权限会让你在那个终端里运行的任何程序都能读取 Mac 上的所有文件，而不只是这一个程序。","source":"ai","title":"会读取你最私密的文件"},{"code":"wechat_export_patches_the_app","severity":"attention","evidence":"README 的微信部分：导出消息需要 WeChatTweak-CLI，用 `brew install sunnyyoung/repo/wechattweak-cli` 或 `sudo packages/wechat-exporter/wechattweak-cli install` 安装。这是一个以管理员权限运行的第三方工具，会修改微信应用，以便读取聊天数据库。","source":"ai","title":"导出微信记录需要改动微信本体"},{"code":"cloud_mode_sends_your_content_away","severity":"info","evidence":"README 的云服务表格及其自带提醒：\"Before choosing a cloud provider, carefully review their privacy and data retention policies. Depending on their terms, your data may be used for their own purposes, including but not limited to human reviews and model training.\" 命令行默认后端是 `--llm openai`，使用 gpt-4o。","source":"ai","title":"云端模式会把你的内容发送出去"},{"code":"numbers_measured_by_the_authors","severity":"info","evidence":"\"97% less storage without accuracy loss\" 这一宣传语，以及在 30 个 SWE-Bench Pro 任务上与 BM25 的比较，都是项目自己在论文和测试框架中得出的结果；README 自己也补充说 \"Better context access does not guarantee issue resolution.\"","source":"ai","title":"这些数字由作者自己测得"},{"code":"installing_from_source_is_heavy","severity":"info","evidence":"README 的 Build from Source：macOS 需要一批 brew 包（libomp、boost、protobuf、zeromq），Linux 需要一长串 apt、pacman 或 dnf 包外加 Intel oneAPI MKL，Windows 则需要 Visual Studio 2022 Build Tools 和 vcpkg。README 还记录了安装失败的情况，例如 Linux 上 nltk 的 \"Security Violation [pathsec.open]\" 错误。","source":"ai","title":"从源码安装十分繁重"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["OpenAI (or any OpenAI-compatible provider, only if you want a cloud model)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["local_ai","ai_skills","development"],"alternativeTo":[{"slug":"notebooklm","name":"NotebookLM","confirmed":true},{"slug":"supermemory","name":"Supermemory","confirmed":true}],"declaredAlternatives":[],"stars":13789,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":17,"growthDays":6,"history":[{"date":"2026-11-07","stars":13778},{"date":"2026-11-08","stars":13780},{"date":"2026-11-09","stars":13786},{"date":"2026-11-10","stars":13789}],"firstPublishedAt":"2026-11-10T23:52:44.087Z","commitSha":"07796bd7e40f2d5f3a4be9e45adf1f62a68481fd","verifiedAt":"2026-11-10T23:43:10.032Z","lastCommitAt":"2026-10-28T21:26:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Search your own files, emails, chats and code by meaning on your computer, with an index that uses far less disk.","summary":"LEANN is a vector index and search engine that runs on your own machine. Instead of storing an embedding for every piece of text, it keeps a pruned graph and recomputes embeddings when it searches, which the authors say cuts index size by about 97 percent. Ready-made examples index documents and PDFs, Apple Mail, Chrome history, WeChat, iMessage, ChatGPT and Claude exports, Slack and a code base, and you can then ask questions about them with a local model through Ollama or with a cloud model. It also plugs into Claude Code as an MCP search tool.","easeRationale":"On macOS and Linux, the easiest case, you install the uv tool, create a virtual environment, run `uv pip install leann` and then type commands such as `python -m apps.document_rag --query \"...\"`. Windows is harder: the only Windows instructions in the README are the build-from-source path, which needs Visual Studio 2022 Build Tools, vcpkg, chocolatey and several environment variables.","costNote":"MIT code, and it can run end to end with local models through Ollama, LM Studio or similar. If you point it at OpenAI, Anthropic or another cloud provider instead, you pay that provider.","notVerified":["whether the \"zero telemetry\" claim holds, and what the bundled apps and MCP servers send out while they run","whether the 97 percent storage saving and the search quality hold on your own messy data, not the published benchmarks","how long building an index of a large mailbox or chat history takes, and how much slower searching is when embeddings are recomputed each time","how much disk and memory you need once model weights, PDF parsers and indexes are all installed","whether the Windows path works without building from source"]},"zh-Hans":{"summaryShort":"在自己的电脑上按语义搜索你的文件、邮件、聊天记录和代码，索引占用的磁盘空间要少得多。","summary":"LEANN 是一个在本机运行的向量索引和搜索引擎。它不为每段文本都存一份向量，而是保留一张经过剪枝的图，在搜索时重新计算向量，作者称这样可以把索引体积减少约 97%。自带的示例可以索引文档和 PDF、Apple Mail、Chrome 浏览历史、微信、iMessage、ChatGPT 和 Claude 的导出记录、Slack 以及代码库，之后你可以用 Ollama 上的本地模型或云端模型对这些内容提问。它还能作为 MCP 搜索工具接入 Claude Code。","easeRationale":"在最省事的 macOS 和 Linux 上，你安装 uv 工具，建一个虚拟环境，运行 `uv pip install leann`，然后输入 `python -m apps.document_rag --query \"...\"` 这样的命令。Windows 要麻烦得多：README 里针对 Windows 的说明只有从源码编译这一条路，需要 Visual Studio 2022 Build Tools、vcpkg、chocolatey 和若干环境变量。","costNote":"代码采用 MIT 许可，并且可以全程使用 Ollama、LM Studio 等本地模型。如果你改用 OpenAI、Anthropic 或其他云端服务，就要向那家服务付费。","notVerified":["\"zero telemetry\" 的说法是否成立，自带的应用和 MCP 服务器运行时又会向外发送什么","在你自己杂乱的数据上，而不是公布的基准测试上，97% 的空间节省和搜索质量是否还能保持","为一个大邮箱或长聊天记录建索引要多久，以及每次重新计算向量会让搜索慢多少","装好模型权重、PDF 解析器和索引之后，总共需要多少磁盘和内存","Windows 上不走源码编译这条路是否行得通"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMDcyNDUwOTI=","slug":"stenzek__duckstation","name":"DuckStation","owner":"stenzek","fullName":"stenzek/duckstation","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"high","signals":[{"code":"bios_file_not_included","severity":"attention","evidence":"README.md：\"A PS1 or PS2 'BIOS' ROM image is required to to start the emulator and to play games... A ROM image is not provided with the emulator for legal reasons, you should dump this from your own console using Caetla or other means.\" BIOS 的著作权属于索尼，因此从网上下载的副本在多数国家都属于受版权保护的内容。","source":"ai","title":"不附带 BIOS 文件"},{"code":"copy_protection_workaround","severity":"attention","evidence":"README.md 的 \"LibCrypt protection and SBI files\" 一节：\"A number of PAL region games use LibCrypt protection, requiring additional CD subchannel information to run properly... make sure that the CD image and its corresponding SBI (.sbi) file have the same name and are placed in the same directory.\" SBI 文件提供的正是光盘防拷贝机制要检查的数据，而模拟器还能读取 bin/cue、CHD 和 PBP 等镜像格式。","source":"ai","title":"绕开光盘防拷贝机制"},{"code":"license_not_open_source","severity":"attention","evidence":"LICENSE 文件是 \"Attribution-NonCommercial-NoDerivatives 4.0 International\"，README.md 写道：\"As per the terms of CC-BY-NC-ND, you may redistribute unmodified versions of this repository, and the compiled binaries... You are not permitted to publish modified versions of this repository.\" 也就是不允许商业使用，也不允许发布修改版。","source":"ai","title":"许可证并非开源"},{"code":"auto_update_built_in","severity":"info","evidence":"README.md：\"Automatic updates with preview and latest channels\"——程序会自行从 GitHub 更新，而 \"Preview\" 通道每次提交都会构建，用作者的话说只经过 \"minimal testing\"。","source":"ai","title":"内置自动更新"},{"code":"mac_app_not_signed","severity":"info","evidence":"README.md：\"If you receive a message about the app being from an unidentified developer... Click 'Open Anyway'\"，因为作者没有苹果的代码签名证书。","source":"ai","title":"Mac 版应用没有签名"},{"code":"needs_console_bios_dump","severity":"high","evidence":"README.md：\"A PS1 or PS2 'BIOS' ROM image is required to to start the emulator and to play games... A ROM image is not provided with the emulator for legal reasons, you should dump this from your own console using Caetla or other means.\" 同一份 README 的 \"LibCrypt protection and SBI files\" 一节又说：\"A number of PAL region games use LibCrypt protection, requiring additional CD subchannel information to run properly\"，DuckStation 会读取光盘镜像旁的 .sbi 文件。把 BIOS 从主机里导出，并补上光盘防拷贝机制要检查的子通道数据，意味着绕过主机和光盘自身的保护，而许多国家都有法律限制绕过复制保护，即使光盘是你自己的。","source":"ai","title":"需要导出主机 BIOS"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["ffmpeg"],"disk":{"minMB":44,"maxMB":133,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["games"],"alternativeTo":[{"slug":"playstation-plus-premium","name":"PlayStation Plus Premium","confirmed":true}],"declaredAlternatives":[],"stars":10976,"downloads":3800205,"downloadsWeek":28345,"downloadsDays":2,"growth":34,"growthDays":6,"history":[{"date":"2026-11-08","stars":10971},{"date":"2026-11-09","stars":10974},{"date":"2026-11-10","stars":10976}],"firstPublishedAt":"2026-11-10T23:52:44.100Z","commitSha":"73c4055ad047c0c93f8143a767205330c8dfaf1a","verifiedAt":"2026-11-10T23:32:15.117Z","lastCommitAt":"2026-11-10T07:05:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Play PlayStation 1 games on your computer, with sharper graphics, save states and game controller support.","summary":"DuckStation is a program that imitates the original Sony PlayStation 1 console, so old PS1 games run on a Windows, Mac or Linux computer. It can show the games at higher resolution, save your progress at any moment, rewind, and use a game controller. It is for people who want to play their own PS1 discs or disc images. To start it you also need a PlayStation BIOS file, which is not included: the project says you should copy it from a console you own.","easeRationale":"On Windows there is a ready installer (44 MB), so it is download and run, but the emulator still will not start until you add a PlayStation BIOS file yourself. On Mac you unzip the app and allow it in Privacy & Security, and on Linux you mark the AppImage as executable first.","notVerified":["Whether a given game actually runs well on your computer, since speed depends on the processor and graphics card","How much disk space you end up using, because game images and save states are not included","Whether the automatic updater downloads anything beyond the official GitHub releases","Whether the Windows, Mac and Linux builds behave the same, since each is built by the project's own automation"]},"zh-Hans":{"summaryShort":"在电脑上玩 PlayStation 1 游戏，画面更清晰，可随时存档，并支持手柄。","summary":"DuckStation 是一款模拟索尼初代 PlayStation 主机的程序，让老的 PS1 游戏能在 Windows、Mac 或 Linux 电脑上运行。它可以用更高分辨率显示画面、随时保存进度、倒带回放，并支持游戏手柄。它面向想玩自己手上 PS1 光盘或光盘镜像的人。启动时你还需要一个 PlayStation 的 BIOS 文件，程序不附带：项目说应当从你自己拥有的主机里导出。","easeRationale":"Windows 上有现成的安装程序（44 MB），下载运行即可，但在你自己放入 PlayStation 的 BIOS 文件之前，模拟器仍然启动不了。Mac 上要解压应用，并在系统设置里允许它运行；Linux 上则要先把 AppImage 设为可执行。","notVerified":["某款具体游戏在你的电脑上是否流畅，因为速度取决于处理器和显卡","最终会占用多少磁盘空间，因为游戏镜像和存档都不随程序提供","自动更新功能除了 GitHub 官方发布之外，是否还会下载别的东西","Windows、Mac 和 Linux 版本的表现是否一致，因为它们都由项目自己的自动化流程构建"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMDU0MTI5NzE=","slug":"stereobooster__react-snap","name":"react-snap","owner":"stereobooster","fullName":"stereobooster/react-snap","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：express: GHSA-qw6h-vgh9-j6wx, GHSA-rv95-896h-c2vc; html-minifier: GHSA-pfq8-rq6v-vf5m; serve-static: GHSA-cm22-4g7w-348p","source":"check","title":"存在已知漏洞"},{"code":"unmaintained_for_years","severity":"attention","evidence":"被分析仓库的最后一次提交是 2022-11-14，GitHub 上没有任何发布版本。README 仍然链接着 Travis CI 和 loadable-components 2.2.3 这类老旧库，而未关闭的问题有 192 个。","source":"ai","title":"多年无人维护"},{"code":"very_old_browser_dependency","severity":"attention","evidence":"package.json 锁定了 \"puppeteer\": \"^1.8.0\"，这是 2018 年版本的无界面 Chrome 库。安装它会把一个旧版 Chromium 下载到你的电脑上。","source":"ai","title":"依赖非常旧的浏览器组件"},{"code":"runs_your_site_in_a_real_browser","severity":"info","evidence":"Readme.md：\"Uses Headless Chrome to crawl all available links starting from the root\"，而 \"cacheAjaxRequests\" 之类的选项会把你网站 AJAX 请求的响应存进生成的页面里，因此构建时页面加载的任何内容都可能出现在发布的 HTML 中。","source":"ai","title":"会用真实浏览器运行你的网站"},{"code":"sandbox_disabled_in_docs","severity":"info","evidence":"Readme.md 针对容器环境建议使用 \"puppeteerArgs\": [\"--no-sandbox\", \"--disable-setuid-sandbox\"]，这会在生成页面期间关闭浏览器的一项保护机制。","source":"ai","title":"文档里建议关闭沙盒"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"prerender-io","name":"Prerender.io","confirmed":true}],"declaredAlternatives":[{"target":"make","stance":"claims","evidence":"Make sure to use [`replace: false`](https://v1.vuejs.org/api/#replace) for root components"}],"stars":5122,"downloads":0,"downloadsWeek":0,"downloadsDays":4,"growth":0,"growthDays":6,"history":[{"date":"2026-11-06","stars":5123},{"date":"2026-11-07","stars":5123},{"date":"2026-11-08","stars":5123},{"date":"2026-11-09","stars":5123},{"date":"2026-11-10","stars":5122}],"firstPublishedAt":"2026-11-10T23:52:44.118Z","commitSha":"e2746ebe63cc75a4943d076585fff5aa5b50751d","verifiedAt":"2026-11-10T23:32:57.977Z","lastCommitAt":"2022-11-14T10:30:54Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a JavaScript web app into ready-made HTML pages, so search engines and social networks can read the content.","summary":"react-snap is a command-line tool for people who build web apps with React, Vue, Preact and similar tools. Before you publish the site, it opens every page in an invisible Chrome browser and saves the finished HTML, which helps search engines and link previews on social networks. It also tries to speed up page loading. You add it to your project with npm or yarn and usually change a few lines in your app's code. The repository has had no new commit since November 2022.","easeRationale":"On Windows, Mac or Linux you need Node.js 8.6+, an existing web app project, a line added to package.json and usually a change in the app's source code (switching render to hydrate). Installing it also pulls a full Chrome browser through Puppeteer, and inside Docker or Alpine extra flags are needed.","notVerified":["Whether it still works with current versions of React, Vue and Node.js, given the age of the code","How much disk space the bundled Chromium download really takes","Whether the pinned dependencies with known vulnerabilities affect a normal build run","Whether your particular site renders correctly, since the README lists many special cases (async components, Redux, JSS, service workers)"]},"zh-Hans":{"summaryShort":"把 JavaScript 网页应用预先生成为现成的 HTML 页面，让搜索引擎和社交网络能读到内容。","summary":"react-snap 是一个命令行工具，面向使用 React、Vue、Preact 等框架开发网页应用的人。在你发布网站之前，它会用一个看不见的 Chrome 浏览器打开每个页面，把渲染完成的 HTML 保存下来，有助于搜索引擎收录和社交网络的链接预览。它还会尝试加快页面加载。你用 npm 或 yarn 把它加进项目，通常还要改动应用代码里的几行。该仓库自 2022 年 11 月起就没有新的提交。","easeRationale":"在 Windows、Mac 或 Linux 上，你需要 Node.js 8.6 以上版本、一个已有的网页应用项目，在 package.json 里加一行配置，通常还要改动应用源码（把 render 换成 hydrate）。安装时还会通过 Puppeteer 下载一整个 Chrome 浏览器，在 Docker 或 Alpine 环境里还需要额外参数。","notVerified":["考虑到代码年代久远，它是否还能配合当前版本的 React、Vue 和 Node.js 工作","随附下载的 Chromium 实际占用多少磁盘空间","那些被锁定、存在已知漏洞的依赖，是否会影响一次普通的构建","你自己的网站能否被正确生成，因为 README 列出了许多特殊情况（异步组件、Redux、JSS、service worker）"]}}},{"id":"R_kgDORq3pFQ","slug":"stevenflanagan1__social-ai-team","name":"Social AI Team","owner":"stevenflanagan1","fullName":"stevenflanagan1/social-ai-team","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 目录和 winget 清单中都没有发现任何许可证","source":"check","title":"没有许可证"},{"code":"publishes_to_your_social_accounts","severity":"attention","evidence":"README.md：/publisher 技能 \"Takes approved content from any content skill and schedules it via Blotato MCP\"，并会执行 \"Full schedule confirmation before submitting anything\"。一旦接通，AI 就能通过那个外部服务把内容发到你真实的社交账号上。","source":"ai","title":"会发布到你的社交账号"},{"code":"reads_client_website_and_instagram","severity":"info","evidence":"README.md：/brand-onboarding \"Uses Playwright to capture evidence from the client's website and Instagram\"。它会打开浏览器抓取外部网站的页面，这些内容随后交给 AI。","source":"ai","title":"会读取客户的网站和 Instagram"},{"code":"installer_copies_into_your_claude_folder","severity":"info","evidence":"install.bat 和 install.sh 用 XCOPY /E /Y 和 \"cp -r\" 把 10 个目录复制到 \"%USERPROFILE%\\.claude\\skills\" / \"$HOME/.claude/skills\"，并覆盖同名文件。这些技能会成为 Claude Code 在你机器上遵循的指令。","source":"ai","title":"安装脚本会写入你的 Claude 目录"},{"code":"license_text_is_informal","severity":"info","evidence":"README 的 \"License\" 一节只写了 \"Created by the AI Impact use it, adapt it, build on it. Want a custom build? https://theaiimpact.co/\"，仓库里没有许可证文件。作者同时也在出售定制开发服务。","source":"ai","title":"许可说明很随意"},{"code":"mostly_written_instructions","severity":"info","evidence":"仓库里只有 README.md、SETUP.md、install.bat、install.sh 和 skills/ 目录：这里没有程序，只有告诉 AI 该做什么的文本文件，因此效果取决于你使用的 AI 模型。","source":"ai","title":"主要就是一堆书面指令"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":["ffmpeg"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Nano Banana image generation (MCP)","Blotato, for scheduling posts (MCP)","Firecrawl, for competitor research (MCP)","SerpApi, for trend research (MCP)","X API, for X trend research"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["ai_skills","social_marketing","image"],"alternativeTo":[{"slug":"jasper","name":"Jasper","confirmed":true},{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true}],"declaredAlternatives":[],"stars":274,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":4,"growthDays":6,"history":[{"date":"2026-11-05","stars":270},{"date":"2026-11-06","stars":270},{"date":"2026-11-07","stars":270},{"date":"2026-11-08","stars":271},{"date":"2026-11-09","stars":273},{"date":"2026-11-10","stars":274}],"firstPublishedAt":"2026-11-10T23:52:44.133Z","commitSha":"01255e2df8a3316c066745a745c0bcc2bab4f1a6","verifiedAt":"2026-11-10T23:34:11.863Z","lastCommitAt":"2026-08-24T09:08:39Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ten Claude Code skills that plan a month of social posts, write captions, make visuals and schedule them.","summary":"Social AI Team is a pack of 10 skills you install inside Claude Code. Each one plays a role of a social media team for a small business: brand setup, monthly content calendar, captions for Instagram and Facebook, posts written for LinkedIn, Threads and X, branded images, scheduling and an end-of-month review. Everything is saved as files in a folder per client. You need Claude Code, and some skills only work with extra connections (MCP) such as Playwright, Nano Banana for images and Blotato for scheduling.","easeRationale":"On Windows you download the repository as a ZIP and double-click install.bat, which copies the 10 skills into your .claude/skills folder; on Mac and Linux you run \"bash install.sh\" in a terminal. On every system you must already have Claude Code, and the images, research and scheduling parts need extra MCP connections set up by hand.","costNote":"The files are free, but they only run inside Claude Code, which needs a paid plan, and the image, research and scheduling parts depend on outside services (Nano Banana, Blotato, Firecrawl, SerpApi, X API) that have their own prices.","notVerified":["The quality of the posts, calendars and reviews it produces for a real business","How much of your Claude plan a full month of content uses","What each of the 10 skill files actually instructs the AI to do, since their text was not read here","The real cost of the outside services (Nano Banana, Blotato, Firecrawl, SerpApi, X API)","Whether the optional stop-motion Reel export works, since it needs Python with imageio and ffmpeg"]},"zh-Hans":{"summaryShort":"十个 Claude Code 技能：规划一个月的社交内容、撰写文案、生成配图并安排发布。","summary":"Social AI Team 是一套装进 Claude Code 的 10 个技能。每个技能扮演小企业社媒团队里的一个角色：品牌设定、月度内容日历、Instagram 和 Facebook 文案、为 LinkedIn、Threads 和 X 量身写的帖子、带品牌风格的配图、排期发布，以及月末复盘。所有产出都按客户分文件夹保存为文件。你需要有 Claude Code，而且部分技能还要额外的连接（MCP），例如 Playwright、生成图片的 Nano Banana 和负责排期的 Blotato。","easeRationale":"在 Windows 上，你把仓库下载为 ZIP，双击 install.bat，它会把 10 个技能复制到你的 .claude/skills 目录；在 Mac 和 Linux 上则在终端里运行 \"bash install.sh\"。无论哪个系统，都必须先有 Claude Code，而图片、调研和排期部分还需要手动配置额外的 MCP 连接。","costNote":"这些文件本身免费，但只能在需要付费订阅的 Claude Code 里运行，图片、调研和排期部分还依赖外部服务（Nano Banana、Blotato、Firecrawl、SerpApi、X API），这些服务各有各的价格。","notVerified":["它为真实企业产出的帖子、日历和复盘质量如何","做满一个月的内容会消耗你 Claude 订阅额度的多少","这 10 个技能文件实际上让 AI 做了什么，因为本次没有逐一阅读其内容","外部服务（Nano Banana、Blotato、Firecrawl、SerpApi、X API）的实际花费","可选的定格动画 Reel 导出功能是否可用，因为它需要带 imageio 和 ffmpeg 的 Python"]}}},{"id":"R_kgDOI2oX4A","slug":"stirling-tools__stirling-pdf","name":"Stirling-PDF","owner":"Stirling-Tools","fullName":"Stirling-Tools/Stirling-PDF","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[{"code":"open_core_license","severity":"info","evidence":"LICENSE：\"app/proprietary/\"、\"app/saas/\"、\"engine/\" 和 \"frontend/src/editor/desktop/\"（等）下的内容 \"is licensed under the license defined in\" 单独的 LICENSE 文件，而不是 MIT；README：\"Stirling PDF is open-core.\"","source":"ai","title":"开放核心（open core）许可"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":257,"maxMB":770,"basis":"installer"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["office","productivity"],"alternativeTo":[{"slug":"acrobat","name":"Acrobat","confirmed":true}],"declaredAlternatives":[],"stars":96214,"downloads":2966300,"downloadsWeek":68797,"downloadsDays":7,"growth":523,"growthDays":7,"history":[{"date":"2026-10-25","stars":95007},{"date":"2026-10-26","stars":95084},{"date":"2026-10-27","stars":95181},{"date":"2026-10-28","stars":95266},{"date":"2026-10-29","stars":95363},{"date":"2026-10-30","stars":95425},{"date":"2026-10-31","stars":95481},{"date":"2026-11-01","stars":95525},{"date":"2026-11-02","stars":95610},{"date":"2026-11-03","stars":95691},{"date":"2026-11-04","stars":95746},{"date":"2026-11-05","stars":95801},{"date":"2026-11-06","stars":95874},{"date":"2026-11-07","stars":95955},{"date":"2026-11-08","stars":96045},{"date":"2026-11-09","stars":96149},{"date":"2026-11-10","stars":96214}],"firstPublishedAt":"2026-10-25T23:19:29.142Z","commitSha":"684608d143779bf3a8ffc67639f077a1202c62ff","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-10T14:22:58Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Stirling-PDF-windows-arm64-setup.exe","sizeBytes":209854472,"url":"https://github.com/Stirling-Tools/Stirling-PDF/releases/download/v3.3.0/Stirling-PDF-windows-arm64-setup.exe","releaseUrl":"https://github.com/Stirling-Tools/Stirling-PDF/releases/tag/v3.3.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Merge, split, sign, compress, convert and edit PDF files on your own computer or server.","summary":"Stirling PDF is a PDF toolbox with more than 50 tools: merge, split, sign, redact, convert, compress, read text from scans (OCR) and edit PDFs. You can install it as a desktop app, or run it on your own server and use it in a web browser, so documents do not have to go to an outside website. It also offers no-code automation and an API for businesses. The core is open source; some parts and the Server and Enterprise plans are paid.","easeRationale":"The latest release includes Windows installers (Stirling-PDF-windows-x86_64.msi and an arm64 setup.exe); the Docker command in the README is only needed for the server version.","costNote":"Open-core: most code is MIT, but some folders have their own licenses, and the README links to paid Server and Enterprise plans.","notVerified":["Which features in the desktop app are free and which need a paid plan","Whether the desktop app sends any data or usage statistics online","How well PDF editing keeps the original layout on complex documents","Real disk use after installing, which Radar estimates from the installer size"]},"zh-Hans":{"summaryShort":"在你自己的电脑或服务器上合并、拆分、签名、压缩、转换和编辑 PDF 文件。","summary":"Stirling PDF 是一个拥有 50 多种工具的 PDF 工具箱：合并、拆分、签名、涂黑敏感内容、转换、压缩、识别扫描件中的文字（OCR）以及编辑 PDF。你可以把它安装为桌面应用，也可以在自己的服务器上运行并通过网页浏览器使用，这样文档就不必上传到外部网站。它还提供无代码自动化功能和面向企业的 API。核心部分是开源的；部分功能以及 Server 和 Enterprise 套餐需要付费。","easeRationale":"最新版本包含 Windows 安装程序（Stirling-PDF-windows-x86_64.msi 和一个 arm64 的 setup.exe）；README 中的 Docker 命令只在使用服务器版本时才需要。","costNote":"开放核心（open-core）模式：大部分代码采用 MIT 许可，但部分文件夹有各自的许可，README 还链接到付费的 Server 和 Enterprise 套餐。","notVerified":["桌面应用中哪些功能免费，哪些需要付费套餐","桌面应用是否会在网上发送任何数据或使用统计","在复杂文档上，PDF 编辑能否很好地保留原有版式","安装后实际占用的磁盘空间，Radar 是根据安装程序大小估算的"]}}},{"id":"R_kgDOU_DMSg","slug":"storytold__cadcraft","name":"CADCraft","owner":"storytold","fullName":"storytold/cadcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"19 天内获得 6905 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"unsupported_entities_lost_on_save","severity":"attention","evidence":"README 的 'What works today' 表格：'Entities CADCraft doesn't model yet are not kept on save: see file-format parity'。保存在其他程序中制作的 DWG 或 DXF 文件时，可能会丢掉其中的一部分内容。","source":"ai","title":"不支持的图元在保存时丢失"},{"code":"early_development","severity":"info","evidence":"README 徽章 'Status: early development' 以及 'CADCraft is in early, fast development.'","source":"ai","title":"处于早期开发阶段"},{"code":"unusual_fork_count","severity":"info","evidence":"GitHub 数据：一个创建于 19 天前的仓库有 3008 个 fork、6905 个星标，fork 占比异常地高。","source":"ai","title":"fork 数量异常"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":50,"maxMB":149,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["design_cad"],"alternativeTo":[{"slug":"autocad","name":"AutoCAD","confirmed":true}],"declaredAlternatives":[],"stars":7905,"downloads":1879570,"downloadsWeek":379967,"downloadsDays":7,"growth":340,"growthDays":7,"history":[{"date":"2026-10-26","stars":6909},{"date":"2026-10-27","stars":7015},{"date":"2026-10-28","stars":7101},{"date":"2026-10-29","stars":7194},{"date":"2026-10-30","stars":7253},{"date":"2026-10-31","stars":7325},{"date":"2026-11-01","stars":7401},{"date":"2026-11-02","stars":7484},{"date":"2026-11-03","stars":7565},{"date":"2026-11-04","stars":7631},{"date":"2026-11-05","stars":7698},{"date":"2026-11-06","stars":7748},{"date":"2026-11-07","stars":7783},{"date":"2026-11-08","stars":7821},{"date":"2026-11-09","stars":7866},{"date":"2026-11-10","stars":7905}],"firstPublishedAt":"2026-10-27T01:10:12.687Z","commitSha":"c3b8e7b7fe3f8cec4f1353d680e6cdd27fe10979","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T18:20:25Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"cadcraft-0.19.0-windows-x64.msi","sizeBytes":59527168,"url":"https://github.com/storytold/cadcraft/releases/download/v0.19.0/cadcraft-0.19.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/cadcraft/releases/tag/v0.19.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"A free AutoCAD-style drawing program for 2D technical drawings and floor plans that opens DXF and DWG files.","summary":"CADCraft is a computer-aided design app that works like AutoCAD: you type commands, snap to points and use layers, dimensions, hatches and blocks. It opens and saves DXF and DWG files and exports PDF, SVG and PNG. It is for drafters, architects, engineers and hobbyists who want a free CAD tool, and AI agents can drive it too. Windows has signed installers; macOS and Linux builds also exist. The project is in early development, and parts of a drawing it does not understand yet are dropped when you save.","easeRationale":"Each release ships code-signed Windows .msi installers and portable .zip files for x64, arm64 and x86 (about 50–60 MB), so you download and open it.","costNote":"MIT or Apache-2.0, no account and no subscription.","notVerified":["How faithfully real-world DWG files from AutoCAD open and save without losing content.","Whether the app starts on every Windows graphics driver; the README describes crashes with some AMD drivers and a workaround.","How close it is to AutoCAD for daily professional work, beyond the feature list in the README.","Whether the downloaded installers match the published source code."]},"zh-Hans":{"summaryShort":"一款免费的类 AutoCAD 绘图程序，用于 2D 技术图纸和平面图，可打开 DXF 和 DWG 文件。","summary":"CADCraft 是一款计算机辅助设计应用，用法与 AutoCAD 相似：输入命令、捕捉点位，并使用图层、标注、填充图案和图块。它可以打开和保存 DXF、DWG 文件，并导出 PDF、SVG 和 PNG。它面向希望使用免费 CAD 工具的绘图员、建筑师、工程师和爱好者，AI 智能体也可以操作它。Windows 提供带签名的安装程序，也有 macOS 和 Linux 版本。项目处于早期开发阶段，保存时会丢弃图纸中它暂时还不能识别的部分。","easeRationale":"每个版本都提供带代码签名的 Windows .msi 安装程序，以及适用于 x64、arm64 和 x86 的便携 .zip 文件（约 50–60 MB），下载后即可打开使用。","costNote":"MIT 或 Apache-2.0，无需账户，也无需订阅。","notVerified":["来自 AutoCAD 的真实 DWG 文件在打开和保存时能多忠实地保留全部内容。","应用能否在所有 Windows 显卡驱动上启动；README 提到部分 AMD 驱动会崩溃，并给出了变通办法。","除了 README 中的功能列表之外，它在日常专业工作中与 AutoCAD 的接近程度。","下载的安装程序是否与公开的源代码一致。"]}}},{"id":"R_kgDOVeM_1A","slug":"storytold__craft-anything","name":"Craft Anything","owner":"storytold","fullName":"storytold/craft-anything","type":"needs_setup","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"3天内获得8421个star","source":"check","title":"年轻的仓库却有很多star"},{"code":"reverse_engineering_use_limits","severity":"attention","evidence":"README.md：\"Use Craft Anything on software you own or are allowed to examine.\" 拆解软件并重新创建它，在一些国家可能违反原版的使用条款或当地法律，尽管README说这个工具\"never bypasses DRM, logins or anti-cheat\"。Deep模式会读取程序的代码和文件：\"A program's code and files are read with the built-in engines (and Ghidra when it's installed)\"。","source":"ai","title":"逆向工程的使用限制"},{"code":"copying_someone_elses_work","severity":"attention","evidence":"这个工具的目的就是重建他人的网站、应用和游戏（README.md：\"Recreate this website\"，\"Recreate this game's gameplay in Rust\"）。README说助手\"is told not to copy logos, photos or text unless you confirm you own them or have permission\"，这只是给AI的一条指令，不是技术上的拦截。","source":"ai","title":"复制他人的作品"},{"code":"early_alpha_self_declared","severity":"attention","evidence":"README.md的Status部分：\"Craft Anything is in early alpha\"，并提到\"Partly there: building with API keys and other coding agents (built, not yet run live), behaviour checks for apps and games, a few engine tools tested only against stand-ins.\"","source":"ai","title":"作者自称是早期alpha版本"},{"code":"very_new_repository","severity":"info","evidence":"这个仓库创建于2026-10-31，已经有8421个star和3个发行版（v0.1.0到v0.3.0在三天内完成），几乎还没有任何历史记录。","source":"ai","title":"非常新的仓库"},{"code":"ai_agent_runs_commands_on_your_computer","severity":"info","evidence":"README.md描述了助手在`Documents/Craft Anything`里构建项目，\"an approval card asks before running a command\"，所以AI会在你的机器上写文件、运行命令，但需要你的批准。","source":"ai","title":"AI助手会在你的电脑上运行命令"},{"code":"recreates_other_peoples_software","severity":"high","evidence":"README.md：\"See how any website, app or game works, then have AI build your own version\"，举例\"Recreate this website\"和\"Recreate this game's gameplay in Rust\"。在Deep模式下，\"A program's code and files are read with the built-in engines (and Ghidra when it's installed)\"。README自己也说\"Use Craft Anything on software you own or are allowed to examine\"：很多付费程序的条款限制拆解程序，很多国家的版权规则也限制未经许可复制他人的作品。","source":"ai","title":"重建他人的软件"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":["rust"],"disk":{"minMB":53,"maxMB":158,"basis":"installer"},"apiKeys":["Anthropic (Claude Code sign-in, or an API key for questions and analysis)","OpenAI or OpenRouter (optional, for questions and analysis)","fal.ai (optional, for generated pictures and 3D models)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"ida-pro","name":"IDA Pro","confirmed":true},{"slug":"hopper","name":"Hopper","confirmed":true},{"slug":"v0-by-vercel","name":"v0 by Vercel","confirmed":true}],"declaredAlternatives":[],"stars":26053,"downloads":565012,"downloadsWeek":416026,"downloadsDays":7,"growth":17730,"growthDays":7,"history":[{"date":"2026-11-02","stars":4250},{"date":"2026-11-03","stars":8323},{"date":"2026-11-04","stars":11246},{"date":"2026-11-05","stars":14720},{"date":"2026-11-06","stars":17749},{"date":"2026-11-07","stars":20390},{"date":"2026-11-08","stars":22423},{"date":"2026-11-09","stars":24294},{"date":"2026-11-10","stars":26053}],"firstPublishedAt":"2026-11-02T03:04:39.400Z","commitSha":"ec93bc20e6d3a13de3e071e4e0e7a8a579978f20","verifiedAt":"2026-11-04T01:45:03.885Z","lastCommitAt":"2026-11-10T22:34:19Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Looks inside a website, app or game to see how it is made, then has an AI coding agent build your own version.","summary":"Craft Anything is a desktop program that inspects software you point it at (a website, a desktop or phone app, a game or a file) with built-in reverse-engineering tools, then asks an AI coding agent on your computer to build a new project like it and compares the result with the original. There are ready installers for Windows, macOS and Linux, plus a command-line tool and an MCP server. Looking inside works with no AI; building needs Claude Code or another agent already signed in. The authors call it early alpha. For developers and technically curious makers.","easeRationale":"On Windows there is a signed .msi installer (plus a portable zip), and macOS has a notarized .dmg, so opening the app and inspecting a target is one download away. The extra step on all three is installing and signing in to a coding agent such as Claude Code before the building part works; on Linux the ready builds are AppImage, Flatpak, .deb and .rpm, and building from source there also needs several system libraries.","costNote":"The program is free (MIT or Apache-2.0). The rebuilding part runs your own coding agent, and the README names Claude Code first, which needs a paid Anthropic plan or API credit. Generated pictures and 3D models need your own fal.ai key.","notVerified":["Whether the app and its 320 tools really work on each platform, since the authors say some tools were only tested against stand-ins","How good the rebuilt websites, apps and games actually are compared with the originals","What the coding agent will cost you per rebuild, since it uses your own account","What the agent sends to outside services while studying a target and building","Whether the Codex and other agent adapters work at all, as the README says they have not been run on a real install"]},"zh-Hans":{"summaryShort":"查看一个网站、应用或游戏内部的构建方式，再让AI编程助手构建出你自己的版本。","summary":"Craft Anything是一款桌面程序，用内置的逆向工程工具检查你指定的软件（一个网站、一个桌面或手机应用、一个游戏或一个文件），然后让你电脑上的AI编程助手构建一个类似的新项目，并把结果和原版做对比。它有Windows、macOS和Linux的现成安装程序，还有一个命令行工具和一个MCP服务器。查看内部不需要任何AI；构建则需要Claude Code或其他已登录的助手。作者把它称为早期alpha版本。适合开发者和有技术兴趣的创作者。","easeRationale":"在Windows上有经过签名的.msi安装程序（还有一个便携版zip），macOS上有经过公证的.dmg，所以打开应用并检查一个目标只差一次下载。三个系统都有的额外步骤是在构建功能生效之前，先安装并登录像Claude Code这样的编程助手；在Linux上，现成的构建版本有AppImage、Flatpak、.deb和.rpm，从源码编译还需要几个系统库。","costNote":"程序本身是免费的（MIT或Apache-2.0）。重建部分运行的是你自己的编程助手，README里第一个提到的是Claude Code，需要一个付费的Anthropic套餐或API额度。生成的图片和3D模型需要你自己的fal.ai密钥。","notVerified":["这个应用和它的320个工具是否真的能在每个平台上正常工作，因为作者说有些工具只针对替代品测试过","重建出来的网站、应用和游戏，和原版相比到底有多好","每次重建编程助手会花你多少钱，因为用的是你自己的账号","助手在研究目标和构建过程中向外部服务发送了什么内容","Codex和其他助手适配器是否真的能用，因为README说它们还没有在真实安装环境里运行过"]}}},{"id":"R_kgDOVDHGvA","slug":"storytold__craft-launcher","name":"Craft Launcher","owner":"storytold","fullName":"storytold/craft-launcher","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"27 天内获得 5029 个 star","source":"check","title":"年轻仓库却有大量 star"},{"code":"early_alpha_young_project","severity":"info","evidence":"README.md：\"Status: ArtCraft Launcher is new and in early alpha. Expect rough edges\"，徽章上写着\"status-early alpha\"。项目信息显示仓库创建于 2026-10-09，已有 21 个 release 和 756 个未关闭的 issue，最新的 release（v0.15.0）在 2026-11-06。","source":"ai","title":"早期 alpha 阶段的年轻项目"},{"code":"installs_and_updates_other_programs","severity":"info","evidence":"README.md（Where apps are installed）：在 Windows 上，每个应用都会通过\"The release MSI, through Windows Installer, for everyone on the computer\"安装到\"C:\\Program Files\\ArtCraft\\<App>\"，并且\"Windows asks for administrator permission for each new install\"；在 macOS 上会复制到\"/Applications/<App>.app\"，在 Linux 上则使用 release 的 .deb 或 .rpm，\"with your system's password prompt\"。这个启动器还会替换自己：\"Update and restart downloads it, verifies it, puts it where this copy is installed\"。README 说每次下载都\"is checked against the release's published SHA-256 before it is installed\"。","source":"ai","title":"会安装和更新其他程序"},{"code":"depends_on_forks_of_its_own_libraries","severity":"info","evidence":"Cargo.toml 把整个 egui 家族和 winit 都替换成了项目自己按 commit 锁定的 fork（[patch.crates-io] 条目指向 github.com/storytold/egui.git 和 storytold/winit.git），而且 craft-update 和 craft-ui 是按 revision 从 github.com/storytold/craft-libs 拉取的。因此构建过程并不使用这些库的官方发布版本。","source":"ai","title":"依赖项目自己 fork 的库"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":10,"maxMB":29,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["productivity"],"alternativeTo":[{"slug":"adobe-creative-cloud","name":"Creative Cloud","confirmed":true}],"declaredAlternatives":[],"stars":5206,"downloads":6489034,"downloadsWeek":1223830,"downloadsDays":7,"growth":333,"growthDays":7,"history":[{"date":"2026-10-26","stars":4128},{"date":"2026-10-27","stars":4194},{"date":"2026-10-28","stars":4252},{"date":"2026-10-29","stars":4319},{"date":"2026-10-30","stars":4366},{"date":"2026-10-31","stars":4433},{"date":"2026-11-01","stars":4698},{"date":"2026-11-02","stars":4795},{"date":"2026-11-03","stars":4873},{"date":"2026-11-04","stars":4921},{"date":"2026-11-05","stars":4977},{"date":"2026-11-06","stars":5026},{"date":"2026-11-07","stars":5072},{"date":"2026-11-08","stars":5119},{"date":"2026-11-09","stars":5161},{"date":"2026-11-10","stars":5206}],"firstPublishedAt":"2026-10-27T01:10:12.688Z","commitSha":"ea25ae5795bcd88231dbc664cde1cdf05da2d9c7","verifiedAt":"2026-11-06T23:05:20.925Z","lastCommitAt":"2026-11-10T19:48:18Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"artcraft-launcher-0.18.0-windows-x64.msi","sizeBytes":10153984,"url":"https://github.com/storytold/craft-launcher/releases/download/v0.18.0/artcraft-launcher-0.18.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/craft-launcher/releases/tag/v0.18.0"},{"os":"mac","file":"artcraft-launcher-0.18.0-macos-universal.dmg","sizeBytes":21613746,"url":"https://github.com/storytold/craft-launcher/releases/download/v0.18.0/artcraft-launcher-0.18.0-macos-universal.dmg","releaseUrl":"https://github.com/storytold/craft-launcher/releases/tag/v0.18.0"},{"os":"linux","file":"artcraft-launcher-0.18.0-linux-x86_64.AppImage","sizeBytes":13457912,"url":"https://github.com/storytold/craft-launcher/releases/download/v0.18.0/artcraft-launcher-0.18.0-linux-x86_64.AppImage","releaseUrl":"https://github.com/storytold/craft-launcher/releases/tag/v0.18.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"One app to install, update and open the free Craft creative apps, such as PhotoCraft and WordCraft, on your computer.","summary":"ArtCraft Launcher is a small desktop app that lists the free, open-source Craft App Suite (image, video, vector, PDF, office, CAD and audio tools), installs the right version for your computer, checks each download against its published checksum, and tells you when updates are out. It can sit in the system tray or menu bar. It is for anyone who uses the Craft apps and does not want to visit a dozen download pages. Windows, Mac and Linux get ready installers. The project says it is in early alpha.","easeRationale":"Windows is the simplest at grade A: the latest release carries artcraft-launcher-0.15.0-windows-x64.msi (10 MB), which the README describes as \"Installer for most PCs (Intel and AMD, 64-bit). Installs for the current user, no admin rights needed\", plus a portable zip you only unzip. macOS gets artcraft-launcher-0.15.0-macos-universal.dmg (20 MB), \"signed and notarized. Drag ArtCraft Launcher to Applications\", and Linux gets an AppImage, .deb, .rpm or Flatpak of 8 to 13 MB, so all three stay at A.","costNote":"Free (MIT or Apache-2.0). The README says the ArtCraft name, logo and \"A\" mark \"are not licensed under MIT or Apache-2.0\" and that \"Forks and modified versions must remove them\".","notVerified":["How much disk space the Craft apps you install take; the 10 to 29 MB figure covers only the launcher","How stable the launcher is, given its early alpha status","Whether update checks keep working when GitHub limits how often it can be asked","Whether an installed app really matches the app's own official release, which the README states but the package cannot show","The Craft apps themselves, which this card does not cover"]},"zh-Hans":{"summaryShort":"一个应用程序即可安装、更新和打开免费的 Craft 创意应用，例如 PhotoCraft 和 WordCraft，在你的电脑上使用。","summary":"ArtCraft Launcher 是一个小型桌面应用，列出免费、开源的 Craft 应用套件（图像、视频、矢量、PDF、办公、CAD 和音频工具），为你的电脑安装合适的版本，用发布的校验和核对每次下载，并在有更新时提醒你。它可以常驻在系统托盘或菜单栏。它适合使用 Craft 应用、又不想一个个跑十几个下载页面的人。Windows、Mac 和 Linux 都有现成的安装程序。项目方说它还处于早期 alpha 阶段。","easeRationale":"Windows 最简单，评级为 A：最新版本带有 artcraft-launcher-0.15.0-windows-x64.msi（10 MB），README 把它描述为\"Installer for most PCs (Intel and AMD, 64-bit). Installs for the current user, no admin rights needed\"，再加上一个只需解压的便携版 zip。macOS 有 artcraft-launcher-0.15.0-macos-universal.dmg（20 MB），\"signed and notarized. Drag ArtCraft Launcher to Applications\"；Linux 则有 8 到 13 MB 的 AppImage、.deb、.rpm 或 Flatpak，所以三个平台都是 A 级。","costNote":"免费（MIT 或 Apache-2.0）。README 说 ArtCraft 这个名字、logo 和 \"A\" 标志\"are not licensed under MIT or Apache-2.0\"，而且\"Forks and modified versions must remove them\"。","notVerified":["你安装的 Craft 应用占用多少磁盘空间；10 到 29 MB 这个数字只是启动器本身的大小","这个启动器有多稳定，毕竟它还处于早期 alpha 阶段","当 GitHub 限制查询频率时，更新检查是否还能正常工作","安装的应用是否真的和官方发布版一致，这一点 README 有说明，但这个包本身无法证明","Craft 应用本身，这张卡片不涉及"]}}},{"id":"R_kgDOU3RjOg","slug":"storytold__effectcraft","name":"EffectCraft","owner":"storytold","fullName":"storytold/effectcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"25 天内获得 7968 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"early_alpha_windows_bugs","severity":"attention","evidence":"README \"Where it stands\"：\"its first commit was on 1 October 2026\"、\"It isn't yet recommended for production client work\" 以及 \"macOS is the most tested platform. Linux and Windows users have hit basic interaction bugs\"。","source":"ai","title":"早期 alpha 版本，Windows 上有错误"},{"code":"unusual_fork_and_issue_volume","severity":"attention","evidence":"GitHub 数据：一个创建于 2026-10-01 的仓库（距检查时 25 天），有 4262 个 fork 和 1018 个未解决的 issue，星标为 7968 个。","source":"ai","title":"fork 和 issue 数量异常"},{"code":"optional_model_downloads","severity":"info","evidence":"README：\"Roto Brush can use MobileSAM (an optional 40 MB download in Settings ▸ Roto Brush) and face tracking can use Google's MediaPipe Face Landmarker (an optional 3.8 MB download)\"。","source":"ai","title":"可选的模型下载"},{"code":"local_control_port","severity":"info","evidence":"README：\"effectcraft --control 9877 accepts JSON lines to run commands, inspect and click any widget by its automation id, and take screenshots\"（仅在使用该选项启动时）。","source":"ai","title":"本地控制端口"},{"code":"third_party_effect_names","severity":"info","evidence":"README 的效果列表使用了另一款产品自带的商业插件效果的名称，例如 \"CC Particle World\"、\"CC Power Pin\" 和 \"CC Rainfall\"，同时又声称这项工作是净室（clean room）开发。","source":"ai","title":"使用第三方效果名称"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust","ffmpeg"],"disk":{"minMB":49,"maxMB":147,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","image","design_cad"],"alternativeTo":[{"slug":"after-effects","name":"After Effects","confirmed":true},{"slug":"blackmagic-fusion-studio","name":"Blackmagic Fusion Studio","confirmed":true}],"declaredAlternatives":[],"stars":8628,"downloads":2101564,"downloadsWeek":382590,"downloadsDays":7,"growth":224,"growthDays":7,"history":[{"date":"2026-10-26","stars":7968},{"date":"2026-10-27","stars":8044},{"date":"2026-10-28","stars":8097},{"date":"2026-10-29","stars":8163},{"date":"2026-10-30","stars":8211},{"date":"2026-10-31","stars":8276},{"date":"2026-11-01","stars":8306},{"date":"2026-11-02","stars":8352},{"date":"2026-11-03","stars":8404},{"date":"2026-11-04","stars":8439},{"date":"2026-11-05","stars":8476},{"date":"2026-11-06","stars":8508},{"date":"2026-11-07","stars":8537},{"date":"2026-11-08","stars":8568},{"date":"2026-11-09","stars":8597},{"date":"2026-11-10","stars":8628}],"firstPublishedAt":"2026-10-27T01:10:12.689Z","commitSha":"cca8ce06f3945ea9ec950c9837f73550279a7698","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T23:00:52Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"effectcraft-0.23.1-windows-x64.msi","sizeBytes":63840256,"url":"https://github.com/storytold/effectcraft/releases/download/v0.23.1/effectcraft-0.23.1-windows-x64.msi","releaseUrl":"https://github.com/storytold/effectcraft/releases/tag/v0.23.1"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Make animated titles, motion graphics and visual effects with layers, keyframes and hundreds of effects, then export video.","summary":"EffectCraft is a desktop studio for motion graphics and visual effects. You build a scene from layers (text, shapes, video, images), animate them over time, add effects, 3D cameras and lights, and export video, image sequences, GIFs or Lottie web animations. Its panels and shortcuts copy the style of well-known motion design tools. It is for motion designers and video creators. Windows, macOS and Linux installers are ready to download. The project is about a month old and says it is in alpha and not yet recommended for client work.","easeRationale":"Code-signed Windows MSI installers and portable zips (x64, x86, ARM64) are attached to every release; the README says video encoding is built in, with no FFmpeg inside.","costNote":"Free and open source (MIT or Apache-2.0).","notVerified":["Whether the Windows version is stable enough for everyday work, given the reported interaction bugs.","How closely effects and renders match the professional tools it imitates.","Speed and memory use on long or complex projects, and which graphics cards it needs.","Whether the app really makes no network connections unless asked, as the README states."]},"zh-Hans":{"summaryShort":"用图层、关键帧和数百种效果制作动画标题、动态图形和视觉特效，并导出视频。","summary":"EffectCraft 是一款用于动态图形和视觉特效的桌面工作室。你可以用图层（文字、形状、视频、图片）搭建场景，随时间为它们设置动画，添加效果、3D 摄像机和灯光，并导出视频、图像序列、GIF 或 Lottie 网页动画。它的面板和快捷键模仿了知名动态设计工具的风格。它面向动态设计师和视频创作者。提供可直接下载的 Windows、macOS 和 Linux 安装程序。该项目大约只有一个月历史，自称处于 alpha 阶段，暂不建议用于客户项目。","easeRationale":"每个版本都附带带代码签名的 Windows MSI 安装程序和便携 zip 包（x64、x86、ARM64）；README 称视频编码已内置，内部不含 FFmpeg。","costNote":"免费开源（MIT 或 Apache-2.0）。","notVerified":["鉴于已报告的交互问题，Windows 版本是否稳定到足以日常使用。","效果和渲染结果与它所模仿的专业工具有多接近。","处理长时间或复杂项目时的速度和内存占用，以及需要什么样的显卡。","应用是否真如 README 所述，在未被要求时不会建立任何网络连接。"]}}},{"id":"R_kgDOU1UwWQ","slug":"storytold__filmcraft","name":"FilmCraft","owner":"storytold","fullName":"storytold/filmcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"26 天内获得 16415 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"unusual_fork_and_issue_counts","severity":"info","evidence":"GitHub API：16415 个星标对应 6858 个 fork（约 42%）和 1838 个未解决的 issue，这一切都发生在仓库于 2026-09-30 创建后的 26 天之内。","source":"ai","title":"fork 和 issue 数量异常"},{"code":"readme_lists_files_not_in_releases","severity":"info","evidence":"README.md 写道 \"every release ships it as filmcraft-web-<version>.zip\"，并列出了 filmcraft-<ver>-freebsd-x86_64.tar.gz 构建，但这两个文件都没有出现在 v0.11.0、v0.12.0 或 v0.13.0 的发布文件列表中。","source":"ai","title":"README 列出的文件不在发布版本中"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust","ffmpeg"],"disk":{"minMB":80,"maxMB":239,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"premiere-pro","name":"Premiere Pro","confirmed":true},{"slug":"final-cut-pro","name":"Final Cut Pro","confirmed":true}],"declaredAlternatives":[],"stars":17992,"downloads":2836024,"downloadsWeek":497713,"downloadsDays":7,"growth":562,"growthDays":7,"history":[{"date":"2026-10-26","stars":16419},{"date":"2026-10-27","stars":16581},{"date":"2026-10-28","stars":16731},{"date":"2026-10-29","stars":16865},{"date":"2026-10-30","stars":16958},{"date":"2026-10-31","stars":17088},{"date":"2026-11-01","stars":17223},{"date":"2026-11-02","stars":17315},{"date":"2026-11-03","stars":17430},{"date":"2026-11-04","stars":17520},{"date":"2026-11-05","stars":17599},{"date":"2026-11-06","stars":17687},{"date":"2026-11-07","stars":17756},{"date":"2026-11-08","stars":17840},{"date":"2026-11-09","stars":17916},{"date":"2026-11-10","stars":17992}],"firstPublishedAt":"2026-10-27T01:10:12.690Z","commitSha":"4764b1a481be6aed25e2e64aa09e18c04c5f6baf","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T23:03:13Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"filmcraft-0.19.0-windows-x64.msi","sizeBytes":90718208,"url":"https://github.com/storytold/filmcraft/releases/download/v0.19.0/filmcraft-0.19.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/filmcraft/releases/tag/v0.19.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Edit videos on a timeline: cut clips, grade color, add effects, titles and captions, mix sound and export.","summary":"FilmCraft is a desktop video editor with a multi-track timeline, color grading, over 90 effects, titles, captions and an audio mixer. Its panels and shortcuts follow professional editors, and it can exchange timelines with other editors through XML, AAF and EDL files. AI agents can also control it through a command-line tool. The project is less than a month old, and its README calls it alpha and estimates it is about 56% ready for real production work. Windows has ready .msi installers and portable .zip files.","easeRationale":"Each release has Windows .msi installers and portable .zip files for x64, ARM64 and 32-bit PCs (about 80–96 MB), so you download, install and open it.","costNote":"Open source under MIT or Apache-2.0.","notVerified":["How stable it is for real editing work, given the README's alpha status and how new the project is","How well it handles real camera and phone footage on Windows, which the README says gets far less testing than macOS","Whether the Windows installers are really code-signed, as the README states","What the built-in update check (craft-update library) connects to","Whether the very fast growth in stars and forks reflects real use"]},"zh-Hans":{"summaryShort":"在时间线上剪辑视频：剪切片段、调色、添加特效、标题和字幕、混音并导出。","summary":"FilmCraft 是一款桌面视频编辑器，具有多轨时间线、调色、90 多种特效、标题、字幕和音频混音器。它的面板和快捷键参照专业剪辑软件，并能通过 XML、AAF 和 EDL 文件与其他剪辑软件交换时间线。AI 智能体也可以通过命令行工具控制它。该项目诞生不到一个月，其 README 称它处于 alpha 阶段，并估计距离真正用于生产大约完成了 56%。Windows 提供现成的 .msi 安装程序和便携 .zip 文件。","easeRationale":"每个版本都提供适用于 x64、ARM64 和 32 位电脑的 Windows .msi 安装程序和便携 .zip 文件（约 80–96 MB），下载、安装后即可打开使用。","costNote":"开源，采用 MIT 或 Apache-2.0 许可证。","notVerified":["鉴于 README 所说的 alpha 状态以及项目很新，它用于真实剪辑工作时稳定性如何","在 Windows 上处理真实相机和手机素材的表现如何；README 称 Windows 的测试远少于 macOS","Windows 安装程序是否真如 README 所述带有代码签名","内置的更新检查（craft-update 库）会连接到哪里","星标和复刻数的飞速增长是否反映了真实使用"]}}},{"id":"R_kgDOU_C7jg","slug":"storytold__gridcraft","name":"GridCraft","owner":"storytold","fullName":"storytold/gridcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"19 天内获得 6602 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"unusual_fork_and_issue_counts","severity":"info","evidence":"GitHub API：6602 个星标对应 3060 个 fork（约 46%）和 1313 个未解决的 issue，这一切都发生在仓库于 2026-10-07 创建后的 19 天之内。","source":"ai","title":"fork 和 issue 数量异常"},{"code":"readme_lists_files_not_in_releases","severity":"info","evidence":"README.md 列出了 gridcraft-<ver>-freebsd-x86_64.tar.gz 和 gridcraft-web-<ver>.zip 两个构建，但这两个文件都没有出现在 v0.10.0、v0.11.1 或 v0.12.1 的发布文件列表中。","source":"ai","title":"README 列出的文件不在发布版本中"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":14,"maxMB":43,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[{"slug":"excel","name":"Excel","confirmed":true}],"declaredAlternatives":[],"stars":7404,"downloads":2225456,"downloadsWeek":479427,"downloadsDays":7,"growth":300,"growthDays":7,"history":[{"date":"2026-10-26","stars":6601},{"date":"2026-10-27","stars":6680},{"date":"2026-10-28","stars":6741},{"date":"2026-10-29","stars":6820},{"date":"2026-10-30","stars":6881},{"date":"2026-10-31","stars":6938},{"date":"2026-11-01","stars":6993},{"date":"2026-11-02","stars":7037},{"date":"2026-11-03","stars":7104},{"date":"2026-11-04","stars":7148},{"date":"2026-11-05","stars":7198},{"date":"2026-11-06","stars":7252},{"date":"2026-11-07","stars":7291},{"date":"2026-11-08","stars":7328},{"date":"2026-11-09","stars":7370},{"date":"2026-11-10","stars":7404}],"firstPublishedAt":"2026-10-27T01:10:12.691Z","commitSha":"8117b66635e79d0fa6bb61661768e8a82fd62454","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T23:05:44Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"gridcraft-0.18.0-windows-x64.msi","sizeBytes":41652224,"url":"https://github.com/storytold/gridcraft/releases/download/v0.18.0/gridcraft-0.18.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/gridcraft/releases/tag/v0.18.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Make spreadsheets with formulas, tables, charts and pivot tables, and open and save Excel .xlsx files.","summary":"GridCraft is a desktop spreadsheet program laid out like Excel, with a ribbon, formula bar, over 500 functions, tables, conditional formatting, 17 chart types, pivot tables and PDF export. Its main file format is .xlsx, and it can also read CSV and older .xls files. AI agents can control it through a command-line tool. The project is less than three weeks old, and its README calls it alpha and estimates it is about 50% ready for real work. Windows has ready .msi installers and portable .zip files.","easeRationale":"Each release has Windows .msi installers and portable .zip files for x64, ARM64 and 32-bit PCs (about 13–19 MB), so you download, install and open it.","costNote":"Open source under MIT or Apache-2.0; the README says no subscription and no licence server.","notVerified":["How faithfully real-world Excel files with complex formatting, macros or charts open and save, given the README's alpha status","How it copes with very large workbooks, which the README lists as future work","Whether the Windows installers are really code-signed, as the README states","What the built-in update check (craft-update library) connects to, given the README's claim of no telemetry","Whether the very fast growth in stars and forks reflects real use"]},"zh-Hans":{"summaryShort":"制作带公式、表格、图表和数据透视表的电子表格，可打开和保存 Excel .xlsx 文件。","summary":"GridCraft 是一款桌面电子表格程序，界面布局与 Excel 相似，有功能区、公式栏、500 多个函数、表格、条件格式、17 种图表类型、数据透视表和 PDF 导出。它的主要文件格式是 .xlsx，也能读取 CSV 和较旧的 .xls 文件。AI 智能体可以通过命令行工具控制它。该项目诞生不到三周，其 README 称它处于 alpha 阶段，并估计距离真正可用大约完成了 50%。Windows 提供现成的 .msi 安装程序和便携 .zip 文件。","easeRationale":"每个版本都提供适用于 x64、ARM64 和 32 位电脑的 Windows .msi 安装程序和便携 .zip 文件（约 13–19 MB），下载、安装后即可打开使用。","costNote":"开源，采用 MIT 或 Apache-2.0 许可证；README 称无需订阅，也没有许可证服务器。","notVerified":["鉴于 README 所说的 alpha 状态，带有复杂格式、宏或图表的真实 Excel 文件在打开和保存时能多忠实地保留内容","它如何应对非常大的工作簿（README 将其列为未来工作）","Windows 安装程序是否真如 README 所述带有代码签名","鉴于 README 声称没有遥测，内置的更新检查（craft-update 库）会连接到哪里","星标和复刻数的飞速增长是否反映了真实使用"]}}},{"id":"R_kgDOU1ac_A","slug":"storytold__lightcraft","name":"LightCraft","owner":"storytold","fullName":"storytold/lightcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"26 天内获得 14756 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"alpha_software","severity":"info","evidence":"README.md：\"LightCraft is young and moving fast: **alpha**.\" 以及 \"As a day-to-day tool we're nearer 61% for the whole product\"。","source":"ai","title":"alpha 阶段的软件"},{"code":"optional_models_installed_separately","severity":"info","evidence":"README.md：Object 和 Describe 蒙版 \"use SAM 3 ... with a separately installed model\"；人脸检测使用 \"optional models installed through Settings → Faces; no weights are bundled\"。","source":"ai","title":"可选模型需单独安装"},{"code":"built_in_updater","severity":"info","evidence":"Cargo.toml 依赖来自 github.com/storytold/craft-libs 的 craft-update（features = [\"egui\"]），这表明该应用可以检查并下载更新。","source":"ai","title":"内置更新程序"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":87,"maxMB":261,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"lightroom","name":"Lightroom","confirmed":true},{"slug":"adobe-camera-raw","name":"Camera Raw","confirmed":true},{"slug":"capture-one","name":"Capture One","confirmed":true}],"declaredAlternatives":[],"stars":15811,"downloads":2409892,"downloadsWeek":432981,"downloadsDays":7,"growth":387,"growthDays":7,"history":[{"date":"2026-10-26","stars":14758},{"date":"2026-10-27","stars":14869},{"date":"2026-10-28","stars":14982},{"date":"2026-10-29","stars":15054},{"date":"2026-10-30","stars":15108},{"date":"2026-10-31","stars":15188},{"date":"2026-11-01","stars":15265},{"date":"2026-11-02","stars":15338},{"date":"2026-11-03","stars":15424},{"date":"2026-11-04","stars":15483},{"date":"2026-11-05","stars":15538},{"date":"2026-11-06","stars":15593},{"date":"2026-11-07","stars":15648},{"date":"2026-11-08","stars":15714},{"date":"2026-11-09","stars":15763},{"date":"2026-11-10","stars":15811}],"firstPublishedAt":"2026-10-27T01:10:12.692Z","commitSha":"bb08a697a41b2561a54f1b740245119503b24758","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T23:07:21Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"lightcraft-0.20.0-windows-x64.msi","sizeBytes":99655680,"url":"https://github.com/storytold/lightcraft/releases/download/v0.20.0/lightcraft-0.20.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/lightcraft/releases/tag/v0.20.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Organize your photo library and edit photos, including camera RAW files, without ever changing the original files.","summary":"LightCraft is a photo library and RAW photo editor for photographers. You can rate and sort photos, adjust light and color, paint masks, crop, merge panoramas and export. Edits never touch the original files, and it works without an account or cloud. There are ready installers for Windows, Mac and Linux. The project calls itself alpha and says support varies by camera. Some AI features (object masks, face recognition, AI noise removal) need models you install separately.","easeRationale":"The latest release has a code-signed Windows .msi installer and a portable .zip for x64, arm64 and 32-bit PCs; Rust is only needed to build it from source.","costNote":"No account or subscription, according to the README.","notVerified":["Whether your specific camera's RAW files open with correct colors (the README says support differs by camera).","Real disk use once the optional AI models for masks, faces and denoise are installed.","Where the optional AI models are downloaded from and under what license terms.","Whether the built-in updater contacts the internet without being asked.","Speed on PCs without a strong graphics card."]},"zh-Hans":{"summaryShort":"整理你的照片库并编辑照片（包括相机 RAW 文件），而且永远不会改动原始文件。","summary":"LightCraft 是面向摄影师的照片库和 RAW 照片编辑器。你可以给照片评级和排序、调整光线和色彩、绘制蒙版、裁剪、合成全景图并导出。编辑从不改动原始文件，无需账户或云端即可使用。提供适用于 Windows、Mac 和 Linux 的现成安装程序。项目自称处于 alpha 阶段，并表示对不同相机的支持程度各不相同。部分 AI 功能（对象蒙版、人脸识别、AI 降噪）需要另行安装模型。","easeRationale":"最新版本提供带代码签名的 Windows .msi 安装程序，以及适用于 x64、arm64 和 32 位电脑的便携 .zip；只有从源代码构建时才需要 Rust。","costNote":"据 README 称，无需账户，也无需订阅。","notVerified":["你的具体相机拍的 RAW 文件打开后颜色是否正确（README 称对不同相机的支持各不相同）。","安装可选的蒙版、人脸和降噪 AI 模型后实际占用多少磁盘空间。","可选的 AI 模型从哪里下载，适用什么许可条款。","内置更新程序是否会在未被要求时联网。","在没有高性能显卡的电脑上的速度。"]}}},{"id":"R_kgDOU1UaEg","slug":"storytold__pdfcraft","name":"PdfCraft","owner":"storytold","fullName":"storytold/pdfcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"26 天内获得 18048 个星标","source":"check","title":"新仓库获得大量星标"},{"code":"self_described_alpha","severity":"info","evidence":"README.md：\"PdfCraft is **alpha**\" 以及 \"fuzzing still turns up crashes and hangs on hostile files\"。","source":"ai","title":"自称处于 alpha 阶段"},{"code":"installer_needs_admin_and_adds_menus","severity":"info","evidence":"README.md：\"The MSI installs for all users and requires administrator privileges. It adds \\\"Edit with PdfCraft\\\" to the right-click menu of PDF files\"；便携版 .zip 可以避免这一点。","source":"ai","title":"安装程序需要管理员权限并添加菜单"},{"code":"homepage_name_mismatch","severity":"info","evidence":"仓库主页是 https://getartcraft.com/apps/printcraft，而 README 指向的是 https://getartcraft.com/apps/pdfcraft。","source":"ai","title":"主页名称与项目不一致"},{"code":"update_component_included","severity":"info","evidence":"Cargo.toml 列出了 `craft-update = { git = \"https://github.com/storytold/craft-libs\", ... features = [\"egui\"] }`，这表明该应用可以在线检查更新，尽管 README 写着 \"no telemetry\"。","source":"ai","title":"包含更新组件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":70,"maxMB":211,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[{"slug":"acrobat","name":"Acrobat","confirmed":true}],"declaredAlternatives":[],"stars":20290,"downloads":3851083,"downloadsWeek":804680,"downloadsDays":7,"growth":851,"growthDays":7,"history":[{"date":"2026-10-26","stars":18051},{"date":"2026-10-27","stars":18247},{"date":"2026-10-28","stars":18455},{"date":"2026-10-29","stars":18637},{"date":"2026-10-30","stars":18792},{"date":"2026-10-31","stars":18945},{"date":"2026-11-01","stars":19095},{"date":"2026-11-02","stars":19249},{"date":"2026-11-03","stars":19439},{"date":"2026-11-04","stars":19546},{"date":"2026-11-05","stars":19686},{"date":"2026-11-06","stars":19809},{"date":"2026-11-07","stars":19912},{"date":"2026-11-08","stars":20025},{"date":"2026-11-09","stars":20175},{"date":"2026-11-10","stars":20290}],"firstPublishedAt":"2026-10-27T01:10:12.693Z","commitSha":"9a9813bf7d97b6786471a4158dcef8fc709d6cbb","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T22:53:58Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"pdfcraft-0.20.0-windows-x64.msi","sizeBytes":98263040,"url":"https://github.com/storytold/pdfcraft/releases/download/v0.20.0/pdfcraft-0.20.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/pdfcraft/releases/tag/v0.20.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"A free desktop app to read PDFs, reorder, combine and split pages, add comments, fill forms and set passwords.","summary":"PdfCraft is a free PDF reader and editor for Windows, Mac and Linux. You can read documents, search text, rotate, delete and reorder pages, combine or split files, add comments, fill in forms and protect files with a password. It works offline and needs no account. There are ready installers and a portable version for Windows. The project calls itself alpha: editing existing text, office file conversion and some advanced features are still missing or weak. A command-line tool is also offered for scripts and AI agents.","easeRationale":"Each release has a signed Windows .msi installer and a portable .zip for x64, x86 and ARM64, so you download and open it.","costNote":"Open source under MIT or Apache-2.0.","notVerified":["Whether the app checks the internet for updates and what it sends when it does.","How well it handles large or complex PDFs compared with paid PDF editors.","Whether the Windows installer's code signature matches the publisher named in the README.","How reliable saving is for documents with forms, signatures or encryption in real use."]},"zh-Hans":{"summaryShort":"一款免费桌面应用，可阅读 PDF、重排、合并和拆分页面、添加批注、填写表单并设置密码。","summary":"PdfCraft 是适用于 Windows、Mac 和 Linux 的免费 PDF 阅读器和编辑器。你可以阅读文档、搜索文字、旋转、删除和重排页面、合并或拆分文件、添加批注、填写表单，并用密码保护文件。它可离线使用，无需账户。Windows 有现成的安装程序和便携版。项目自称处于 alpha 阶段：编辑已有文字、办公文件转换以及一些高级功能仍然缺失或较弱。它还提供命令行工具，供脚本和 AI 智能体使用。","easeRationale":"每个版本都提供带签名的 Windows .msi 安装程序，以及适用于 x64、x86 和 ARM64 的便携 .zip，下载后即可打开使用。","costNote":"开源，采用 MIT 或 Apache-2.0 许可证。","notVerified":["应用是否会联网检查更新，以及检查时会发送什么。","与付费 PDF 编辑器相比，它处理大型或复杂 PDF 的能力如何。","Windows 安装程序的代码签名是否与 README 中所述的发布者一致。","在真实使用中，带有表单、签名或加密的文档保存起来有多可靠。"]}}},{"id":"R_kgDOU1UDDw","slug":"storytold__photocraft","name":"PhotoCraft","owner":"storytold","fullName":"storytold/photocraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"31 天内获得 75575 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"early_alpha_self_declared","severity":"attention","evidence":"README 的状态栏写道：\"PhotoCraft is in early alpha ... it is not yet a Photoshop replacement for daily professional work. The biggest gaps are AI/generative features, 15 missing tools, depth in typography and pro workflows, and plug-in compatibility ... Expect rough edges\"。当前版本为 0.18.0，仓库有 3628 个未关闭的 issue。","source":"ai","title":"项目自称仍处于早期 alpha"},{"code":"clean_room_claim_unverified","severity":"info","evidence":"README 称 PhotoCraft 是 \"an open-source, clean-room reimplementation of Adobe Photoshop\"，说它是 \"implemented from public specs and observed behaviour only. No proprietary code, shaders or assets\"，并声明它 \"is not affiliated with, sponsored by or endorsed by Adobe Inc.\"。这些说法无法从项目外部核实。","source":"ai","title":"“洁净室”重制说法无法核实"},{"code":"local_control_channel_and_mcp_server","severity":"info","evidence":"README 写道，该应用提供 \"an authenticated, loopback-only control channel (`photocraft --control`) for inspecting UI state, driving tools with pointer events, and taking offscreen screenshots\"，而 `photocraft-cli mcp` 会启动一个 MCP 服务器，这样其他程序或 AI 代理就能操作这个编辑器。","source":"ai","title":"本地控制通道和 mcp 服务器"},{"code":"builds_depend_on_another_repository","severity":"info","evidence":"Cargo.toml 从 git 仓库 `https://github.com/storytold/craft-libs` 的固定版本拉取 `craft-ui`、`craft-print`、`craft-jpeg`、`craft-docpath` 和 `craft-update`，其中 `craft-update` 意味着应用启动时会检查新版本。","source":"ai","title":"构建依赖另一个仓库"},{"code":"known_limitations_on_linux_wayland","severity":"info","evidence":"README 写道：\"On a Wayland session, files dropped on the window don't open yet\"（winit 在 Wayland 上没有拖放功能，issue #386），而 XWayland 的变通方案会在分数缩放的显示器上导致界面变模糊（#1072）。","source":"ai","title":"linux wayland 下的已知限制"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","rust"],"disk":{"minMB":102,"maxMB":305,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["image","design_cad"],"alternativeTo":[{"slug":"photoshop","name":"Photoshop","confirmed":true}],"declaredAlternatives":[{"target":"eraser","stance":"claims","evidence":"· Brush · Pencil · Mixer Brush · Color Replacement · Clone Stamp · Pattern Stamp · History Brush · Art History Brush · Eraser · Background Eraser · Magic Eraser · Gradient · Paint Bucket · Blur · Shar"}],"stars":78475,"downloads":4672734,"downloadsWeek":788337,"downloadsDays":7,"growth":1864,"growthDays":7,"history":[{"date":"2026-10-25","stars":72865},{"date":"2026-10-26","stars":73378},{"date":"2026-10-27","stars":73950},{"date":"2026-10-28","stars":74418},{"date":"2026-10-29","stars":74839},{"date":"2026-10-30","stars":75230},{"date":"2026-10-31","stars":75573},{"date":"2026-11-01","stars":75913},{"date":"2026-11-02","stars":76226},{"date":"2026-11-03","stars":76611},{"date":"2026-11-04","stars":76879},{"date":"2026-11-05","stars":77177},{"date":"2026-11-06","stars":77450},{"date":"2026-11-07","stars":77711},{"date":"2026-11-08","stars":77952},{"date":"2026-11-09","stars":78216},{"date":"2026-11-10","stars":78475}],"firstPublishedAt":"2026-10-25T23:19:30.018Z","commitSha":"1f984bffb3793a284f861e4b753fc5da1ac5214d","verifiedAt":"2026-11-01T01:43:51.264Z","lastCommitAt":"2026-11-10T23:00:07Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"photocraft-0.22.0-windows-x64.msi","sizeBytes":117329920,"url":"https://github.com/storytold/photocraft/releases/download/v0.22.0/photocraft-0.22.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/photocraft/releases/tag/v0.22.0"},{"os":"mac","file":"photocraft-0.22.0-macos-universal.dmg","sizeBytes":269729474,"url":"https://github.com/storytold/photocraft/releases/download/v0.22.0/photocraft-0.22.0-macos-universal.dmg","releaseUrl":"https://github.com/storytold/photocraft/releases/tag/v0.22.0"},{"os":"linux","file":"photocraft-0.22.0-linux-x86_64.AppImage","sizeBytes":130148856,"url":"https://github.com/storytold/photocraft/releases/download/v0.22.0/photocraft-0.22.0-linux-x86_64.AppImage","releaseUrl":"https://github.com/storytold/photocraft/releases/tag/v0.22.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Photoshop-style image editor with layers, masks, type and real PSD files, installed like a normal program.","summary":"PhotoCraft is a desktop image editor written in Rust: layers, masks, adjustment layers, layer styles, text, vectors, brushes, filters and real PSD files, and it also opens Paint.NET, GIMP, Krita, OpenRaster and Affinity documents. Signed installers are published for Windows, macOS and Linux (roughly 100-230 MB), plus a command-line version that scripts or an AI agent can drive. The team says it is in early alpha and, in their words, \"not yet a Photoshop replacement for daily professional work\".","easeRationale":"A on all three: each release has a ready installer — a signed .msi (or a portable .zip) for Windows, a signed and notarized .dmg for macOS, and an AppImage, .deb, .rpm, Flatpak or tarball for Linux, with SHA256SUMS.txt for checking the download. Nothing else has to be installed.","costNote":"Dual-licensed MIT or Apache-2.0, no paid version mentioned.","notVerified":["Whether it is stable enough for real work, which is what \"early alpha\" leaves open","Whether your own PSD files survive a round trip with all layers and effects intact","How it performs on large documents and on your graphics card","Whether the installers were built from this source code and what the update check sends"]},"zh-Hans":{"summaryShort":"一款 Photoshop 风格的图像编辑器，支持图层、蒙版、文字和真正的 PSD 文件，像普通程序一样安装。","summary":"PhotoCraft 是一款用 Rust 编写的桌面图像编辑器：图层、蒙版、调整图层、图层样式、文字、矢量、笔刷、滤镜和真正的 PSD 文件，还能打开 Paint.NET、GIMP、Krita、OpenRaster 和 Affinity 的文档。Windows、macOS 和 Linux 都发布了经过签名的安装程序（大约 100 到 230 MB），另外还有一个脚本或 AI 代理可以驱动的命令行版本。团队表示它还在早期 alpha 阶段，用他们自己的话说，\"not yet a Photoshop replacement for daily professional work\"。","easeRationale":"三个系统都是 A：每个版本都带有现成的安装程序——Windows 上是经过签名的 .msi（或便携版 .zip），macOS 上是经过签名并通知公证的 .dmg，Linux 上是 AppImage、.deb、.rpm、Flatpak 或 tarball，并附有 SHA256SUMS.txt 用于校验下载内容。不需要再安装别的东西。","costNote":"采用 MIT 或 Apache-2.0 双重许可证，没有提到任何付费版本。","notVerified":["它是否足够稳定可用于实际工作，这正是 \"early alpha\" 留下的疑问","你自己的 PSD 文件经过一次往返后，图层和效果是否完整保留","在大型文档和你的显卡上表现如何","安装程序是否由这份源代码构建而成，更新检查又会发送什么内容"]}}},{"id":"R_kgDOU1VGZg","slug":"storytold__vectorcraft","name":"VectorCraft","owner":"storytold","fullName":"storytold/vectorcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"26 天内获得 10760 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"unusual_fork_ratio","severity":"attention","evidence":"GitHub API：创建于 2026-09-30，星标：10760，fork：5255。不到一个月内几乎每两个星标就对应一个 fork，远高于面向普通用户的应用的常见水平。","source":"ai","title":"fork 比例异常"},{"code":"alpha_software","severity":"info","evidence":"README.md：\"Where we are (2026-10-10): alpha.\" 以及 \"we estimate ~65% ready for real work for a mainstream illustrator\"；这些评分被描述为 \"self-assessed\"。","source":"ai","title":"处于 alpha 阶段的软件"},{"code":"readme_files_missing_from_release","severity":"info","evidence":"README.md 列出了 `SHA256SUMS.txt`、`vectorcraft-<ver>-freebsd-x86_64.tar.gz` 和 `vectorcraft-web-<ver>.zip`，但 v0.17.0 版本的文件中一个也没有。","source":"ai","title":"README 提到的文件未出现在发布版本中"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":102,"maxMB":305,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["design_cad","image"],"alternativeTo":[{"slug":"illustrator","name":"Illustrator","confirmed":true},{"slug":"affinity-designer","name":"Affinity Designer","confirmed":true},{"slug":"coreldraw","name":"CorelDRAW","confirmed":true}],"declaredAlternatives":[],"stars":11595,"downloads":2698931,"downloadsWeek":490525,"downloadsDays":7,"growth":287,"growthDays":7,"history":[{"date":"2026-10-26","stars":10761},{"date":"2026-10-27","stars":10855},{"date":"2026-10-28","stars":10940},{"date":"2026-10-29","stars":11006},{"date":"2026-10-30","stars":11074},{"date":"2026-10-31","stars":11133},{"date":"2026-11-01","stars":11190},{"date":"2026-11-02","stars":11243},{"date":"2026-11-03","stars":11308},{"date":"2026-11-04","stars":11364},{"date":"2026-11-05","stars":11400},{"date":"2026-11-06","stars":11440},{"date":"2026-11-07","stars":11475},{"date":"2026-11-08","stars":11512},{"date":"2026-11-09","stars":11559},{"date":"2026-11-10","stars":11595}],"firstPublishedAt":"2026-10-27T01:10:12.694Z","commitSha":"5815205324eaf7873265348e33a3513db181e5cf","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T22:55:33Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"vectorcraft-0.23.0-windows-x64.msi","sizeBytes":126746624,"url":"https://github.com/storytold/vectorcraft/releases/download/v0.23.0/vectorcraft-0.23.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/vectorcraft/releases/tag/v0.23.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Draw and edit vector illustrations, logos and posters in a desktop app with tools like those in pro design software.","summary":"VectorCraft is a free vector drawing app for illustrations, logos, posters and simple technical drawings. Its layout, tools and shortcuts copy the style of mainstream illustration software. It opens and saves SVG and PDF, and can open .ai, EPS and Affinity files. It is still alpha: the team itself estimates it is about 65% ready for real work, and reliable .ai exchange is listed as missing. There are installers for Windows, Mac and Linux. It also has an MCP server so AI agents such as Claude can draw and export with it.","easeRationale":"The latest release has a Windows .msi installer and a portable .zip for x64, ARM and 32-bit PCs, so you download it and open it.","notVerified":["How stable the alpha version is in daily use on Windows.","How well real .ai and Affinity files from other people open.","Whether the built-in update component (craft-update in Cargo.toml) contacts a server and what it downloads.","Whether the Windows installers are really code-signed, as the README says.","Why the repository gained so many stars and forks so quickly."]},"zh-Hans":{"summaryShort":"在桌面应用中绘制和编辑矢量插画、标志和海报，工具与专业设计软件相似。","summary":"VectorCraft 是一款免费的矢量绘图应用，可用于插画、标志、海报和简单的技术图纸。它的布局、工具和快捷键模仿了主流插画软件的风格。它可以打开和保存 SVG 与 PDF，也能打开 .ai、EPS 和 Affinity 文件。它仍处于 alpha 阶段：团队自己估计，离可用于实际工作还差一些，目前约完成了 65%，并且列出了尚缺少可靠的 .ai 文件互通。提供适用于 Windows、Mac 和 Linux 的安装程序。它还带有 MCP 服务器，让 Claude 等 AI 智能体可以用它来绘图和导出。","easeRationale":"最新版本提供 Windows .msi 安装程序，以及适用于 x64、ARM 和 32 位电脑的便携 .zip，下载后即可打开使用。","notVerified":["alpha 版本在 Windows 上日常使用的稳定程度。","他人制作的真实 .ai 和 Affinity 文件能打开到什么程度。","内置更新组件（Cargo.toml 中的 craft-update）是否会连接服务器，以及会下载什么。","Windows 安装程序是否真的像 README 所说那样经过代码签名。","该仓库为什么能在这么短的时间内获得这么多 star 和 fork。"]}}},{"id":"R_kgDOU_PZLA","slug":"storytold__wordcraft","name":"WordCraft","owner":"storytold","fullName":"storytold/wordcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"19 天内获得 16224 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"alpha_software","severity":"info","evidence":"README.md，Roadmap：\"WordCraft is in alpha.\" 以及 \"we estimate it is about 60% ready for real work\"；beta 版本仍需要 \"testing against real-world .docx files\" 和 \"native printing\"。","source":"ai","title":"处于 alpha 阶段的软件"},{"code":"many_open_issues_young_repo","severity":"info","evidence":"GitHub API：一个创建于 2026-10-07 的仓库（检查前 19 天）有 1871 个未关闭的 issue 和 4429 个 fork。","source":"ai","title":"新仓库有大量未关闭的 issue"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":81,"maxMB":242,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[{"slug":"word","name":"Word","confirmed":true}],"declaredAlternatives":[{"target":"word","stance":"claims","evidence":"`), dashes; word, sentence, paragraph selection; drag-select; clipboard with formatting; undo/redo; find and replace with regex |"}],"stars":17754,"downloads":2778366,"downloadsWeek":614081,"downloadsDays":7,"growth":551,"growthDays":7,"history":[{"date":"2026-10-26","stars":16227},{"date":"2026-10-27","stars":16370},{"date":"2026-10-28","stars":16481},{"date":"2026-10-29","stars":16613},{"date":"2026-10-30","stars":16746},{"date":"2026-10-31","stars":16864},{"date":"2026-11-01","stars":16965},{"date":"2026-11-02","stars":17073},{"date":"2026-11-03","stars":17203},{"date":"2026-11-04","stars":17306},{"date":"2026-11-05","stars":17391},{"date":"2026-11-06","stars":17466},{"date":"2026-11-07","stars":17536},{"date":"2026-11-08","stars":17612},{"date":"2026-11-09","stars":17689},{"date":"2026-11-10","stars":17754}],"firstPublishedAt":"2026-10-27T01:10:12.695Z","commitSha":"f54ea731cc584002786498b603383c85f2e7ee89","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T17:43:18Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"wordcraft-0.17.0-windows-x64.msi","sizeBytes":94769152,"url":"https://github.com/storytold/wordcraft/releases/download/v0.17.0/wordcraft-0.17.0-windows-x64.msi","releaseUrl":"https://github.com/storytold/wordcraft/releases/tag/v0.17.0"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"A free word processor that looks and works like Microsoft Word and opens and saves .docx files.","summary":"WordCraft is a free desktop word processor with a Word-style ribbon, styles, tables, track changes, comments, footnotes, citations and mail merge. It opens and saves .docx and also handles .odt, .rtf, Markdown and PDF export. It is for anyone who writes documents and wants an open-source option. On Windows you download an installer or a portable zip. The project calls itself alpha and estimates it is about 60% ready for real work, so complex Word files may not look the same.","easeRationale":"The latest release has Windows .msi installers and portable .zip files for x64, ARM64 and 32-bit PCs, so you download and open it.","costNote":"Free and open source (MIT or Apache-2.0).","notVerified":["How faithfully real-world .docx files from Microsoft Word keep their layout after opening and saving.","Whether the app starts on your graphics card; the README documents DirectX 12 and driver crash workarounds on Windows.","Whether the built-in update component (craft-update in Cargo.toml) contacts a server, and what it sends.","Stability with long documents, since the project says a stability sweep is still pending.","Whether the code-signed release files match the source code."]},"zh-Hans":{"summaryShort":"一款免费文字处理软件，外观和用法都像 Microsoft Word，可打开和保存 .docx 文件。","summary":"WordCraft 是一款免费的桌面文字处理软件，具有 Word 风格的功能区、样式、表格、修订、批注、脚注、引文和邮件合并。它可以打开和保存 .docx，也支持 .odt、.rtf、Markdown 和导出 PDF。适合需要写文档、又想用开源软件的人。在 Windows 上可以下载安装程序或便携 zip。项目自称处于 alpha 阶段，估计大约完成了 60%，可用于实际工作，所以复杂的 Word 文件显示效果可能不一样。","easeRationale":"最新版本提供 Windows .msi 安装程序，以及适用于 x64、ARM64 和 32 位电脑的便携 .zip 文件，下载后即可打开。","costNote":"免费开源（MIT 或 Apache-2.0）。","notVerified":["来自 Microsoft Word 的真实 .docx 文件在打开并保存后，版式能保持到什么程度。","应用能否在你的显卡上启动；README 记录了 Windows 上 DirectX 12 和驱动崩溃的应对办法。","内置更新组件（Cargo.toml 中的 craft-update）是否会连接服务器，以及会发送什么。","处理长文档时的稳定性，因为项目称稳定性排查仍未完成。","经过代码签名的发布文件是否与源代码一致。"]}}},{"id":"R_kgDONh_JYw","slug":"sudershhh__saas-dm-automations","name":"SaaS DM Automations","owner":"Sudershhh","fullName":"Sudershhh/saas-dm-automations","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"web":"E"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 目录和 winget 清单中都没有发现任何许可证","source":"check","title":"没有许可证"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：next: GHSA-2464-fhr4-w9xw, GHSA-2xp9-vwfh-vxw4, GHSA-36qx-fr4f-26g5","source":"check","title":"存在已知漏洞"},{"code":"ai_replies_in_your_name","severity":"attention","evidence":"README.md 的 \"AI Response Flow\"：\"AI context is generated from conversation history\"、\"OpenAI generates appropriate response\"、\"Response is sent back to user via Instagram API\"。粉丝发来的消息会被发送给 OpenAI，回复则直接以你的账号发出，中间没有人过目。","source":"ai","title":"AI 会以你的名义回复"},{"code":"depends_on_meta_app_approval","severity":"attention","evidence":"README.md 要求提供 INSTAGRAM_APP_ID 和 INSTAGRAM_APP_SECRET，并描述了 \"User connects Instagram account via OAuth\" 以及一个 webhook 接口。发送自动消息需要经过 Meta 审核的应用，而 Meta 关于自动化消息的规则同样适用于你的账号。","source":"ai","title":"取决于 Meta 是否批准应用"},{"code":"small_unfinished_project","severity":"info","evidence":"仓库有 19 个星标，没有任何发布版本，也没有测试目录，最后一次提交是 2026-03-24。package.json 标注为 \"private\": true，版本号 0.1.0，可见这是作为源码公开的个人项目。","source":"ai","title":"规模很小且尚未完成"},{"code":"setup_instructions_incomplete","severity":"info","evidence":"package.json 里列出了 \"stripe\": \"^17.5.0\"，但 README 的环境变量清单里没有任何 Stripe 密钥，也没有说明如何注册 Instagram 的 webhook。缺的部分得你自己补上。","source":"ai","title":"安装说明不完整"}],"platforms":["web"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":["OpenAI","Instagram Graph API (app ID and secret from Meta)","Clerk, for login","Stripe, present in the dependencies for payments"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"manychat","name":"ManyChat","confirmed":true},{"slug":"chatfuel","name":"Chatfuel","confirmed":true},{"slug":"sendpulse","name":"SendPulse","confirmed":true}],"declaredAlternatives":[],"stars":19,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":19},{"date":"2026-11-08","stars":19},{"date":"2026-11-09","stars":19},{"date":"2026-11-10","stars":19}],"firstPublishedAt":"2026-11-10T23:52:44.150Z","commitSha":"575a213d3fd33c561f4e6557ad644183f97ab46a","verifiedAt":"2026-11-10T23:34:57.704Z","lastCommitAt":"2026-03-24T23:49:39Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A website you host yourself that answers your Instagram direct messages and comments with AI or fixed replies.","summary":"DM Automations is the source code of an Instagram automation site. You connect an Instagram account, choose keywords, and the system answers direct messages and comments by itself, either with a ready template or with a reply written by OpenAI. It also shows engagement charts. There is no ready program to download: you must install Node.js, create a Postgres database, register an app with Instagram, Clerk and OpenAI, and publish the site (the author uses Vercel). The last change is from March 2026.","easeRationale":"There is nothing to install and open: you clone the code, run npm install and prisma migrate, fill six environment variables (database, Instagram app, OpenAI, Clerk) and deploy the site, plus register an Instagram app with webhooks at Meta. That is work for someone who builds websites.","costNote":"The code is free, but to work it needs a paid OpenAI key, a Postgres database and hosting; Clerk and Stripe (in the dependencies) have free tiers with paid plans.","notVerified":["Whether the site works end to end today, since the Instagram API changes often","What the demo at saas-dm-automations.vercel.app really does","The monthly cost of OpenAI, the database and hosting for a busy account","Whether automatic replies keep your Instagram account within Meta's rules","Whether the payment part with Stripe is finished"]},"zh-Hans":{"summaryShort":"一个需要自己部署的网站，用 AI 或固定话术自动回复你的 Instagram 私信和评论。","summary":"DM Automations 是一个 Instagram 自动化网站的源码。你连接一个 Instagram 账号、设定关键词，系统就会自动回复私信和评论，既可以用现成模板，也可以用 OpenAI 生成的回复。它还会展示互动数据图表。没有现成的程序可下载：你必须安装 Node.js、建一个 Postgres 数据库，在 Instagram、Clerk 和 OpenAI 上注册应用，再把网站部署上线（作者用的是 Vercel）。最后一次改动是 2026 年 3 月。","easeRationale":"没有可以装好就打开的程序：你要克隆代码，运行 npm install 和 prisma migrate，填好六个环境变量（数据库、Instagram 应用、OpenAI、Clerk），再部署网站，还要在 Meta 那边注册一个带 webhook 的 Instagram 应用。这是做网站的人才干得来的活。","costNote":"代码免费，但要跑起来需要付费的 OpenAI 密钥、一个 Postgres 数据库和主机托管；依赖里的 Clerk 和 Stripe 有免费额度，也有付费套餐。","notVerified":["这个网站今天是否还能完整跑通，因为 Instagram 的 API 经常变动","saas-dm-automations.vercel.app 上的演示实际能做什么","对一个消息量大的账号来说，OpenAI、数据库和主机每月要花多少钱","自动回复是否会让你的 Instagram 账号仍符合 Meta 的规则","Stripe 支付部分是否已经做完"]}}},{"id":"R_kgDOVKM4cg","slug":"supermedo__uncharted_pc","name":"Uncharted PC","owner":"Supermedo","fullName":"Supermedo/uncharted_pc","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B"},"risk":"high","signals":[{"code":"needs_a_decrypted_copy_of_a_commercial_game","severity":"high","evidence":"README：\"No game files are included. You need your own decrypted dump of the collection (CUSA02343, CUSA02320, ...: the folder with eboot.bin, big2-ps4_Shipping.elf, ... u1data, u2data and u3data).\" PS4 的光盘和下载内容都是加密的，所以获取这些文件意味着绕过主机的保护措施，这在许多国家是违法的。此外，该项目会运行游戏本身的代码，并声明它 \"is not affiliated with Sony Interactive Entertainment or Naughty Dog\"。","source":"ai","title":"需要一份已解密的商业游戏副本"},{"code":"very_new_and_crash_prone","severity":"attention","evidence":"该仓库创建于 2026-10-15，当前版本为 0.3.9，有 408 个未关闭的 issue。README 的 \"Known problems\" 列表包括图形设备崩溃、加载时崩溃、Intel 显卡上黑屏以及旧处理器上崩溃，每一项都已在最近的 0.3.x 版本中修复。","source":"ai","title":"非常新，容易崩溃"},{"code":"heavy_hardware_requirements","severity":"info","evidence":"README 的要求：支持 Vulkan 1.3 的显卡，并且 \"6 GB of video memory or more\"；文中说 GTX 1650 或 RX 570 这类 4 GB 显卡 \"run out of video memory and crash after a while\"。另外还需要约 8 GB 的可用内存。","source":"ai","title":"硬件要求较高"},{"code":"bundles_closed_source_libraries","severity":"info","evidence":"致谢部分：下载内容包含 NVIDIA 的 nvngx_dlss.dll（\"under NVIDIA's license\"）和 Intel 的 libxess.dll（\"under Intel's license\"），同时还有项目自己的 GPL v2 代码。","source":"ai","title":"附带闭源库"}],"platforms":["windows"],"tools":["gpu_any","ffmpeg"],"buildTools":["ffmpeg"],"disk":{"minMB":117,"maxMB":352,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":2259,"downloads":188073,"downloadsWeek":29272,"downloadsDays":7,"growth":92,"growthDays":7,"history":[{"date":"2026-10-27","stars":2046},{"date":"2026-10-28","stars":2070},{"date":"2026-10-29","stars":2086},{"date":"2026-10-30","stars":2097},{"date":"2026-10-31","stars":2110},{"date":"2026-11-01","stars":2137},{"date":"2026-11-02","stars":2153},{"date":"2026-11-03","stars":2167},{"date":"2026-11-04","stars":2185},{"date":"2026-11-05","stars":2203},{"date":"2026-11-06","stars":2214},{"date":"2026-11-07","stars":2231},{"date":"2026-11-08","stars":2239},{"date":"2026-11-09","stars":2245},{"date":"2026-11-10","stars":2259}],"firstPublishedAt":"2026-10-27T18:40:28.063Z","commitSha":"754ed2ab4f1fdd21250c8785d4d4f49517d31292","verifiedAt":"2026-10-27T18:24:33.987Z","lastCommitAt":"2026-11-07T17:39:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Runs the three PS4 Uncharted games on Windows, using game files you have to supply yourself.","summary":"A Windows launcher that runs the PS4 code of Uncharted 1, 2 and 3 natively and translates their graphics to Vulkan. It adds higher resolutions and frame rates, ultrawide screens, DLSS and FSR, mouse and keyboard, 5.1 and 7.1 sound, trophies and cheats. You unzip it, run Uncharted.exe and point it at your game folder. No game files are included: the README says you need your own decrypted dump of the PS4 collection. It needs a Vulkan 1.3 graphics card with 6 GB of video memory. GPL v2.","easeRationale":"The download is a ready zip: unpack it and run Uncharted.exe. The extra step is the hard one — you must already have your own decrypted dump of the PS4 collection, which the project does not provide and which a normal PS4 does not produce.","costNote":"The launcher is free. You still need a copy of the PS4 game, which is a paid product.","notVerified":["Whether it runs the games well on your own graphics card and processor.","What the zip actually contains and whether the published build matches the source in this repository.","How much disk space you need once your own game files are added.","Whether the automatic update feature in the launcher only downloads from this project's releases."]},"zh-Hans":{"summaryShort":"在Windows上运行PS4版的三款Uncharted游戏,游戏文件需要你自己提供。","summary":"一个Windows启动器,原生运行PS4版《神秘海域》1、2和3的代码,并将画面转换为Vulkan渲染。它加入了更高的分辨率和帧率、超宽屏支持、DLSS和FSR、鼠标键盘操作、5.1和7.1音效、奖杯和作弊功能。你只需解压、运行Uncharted.exe,然后指向你的游戏文件夹。不附带任何游戏文件:README说你需要自己准备一份已解密的PS4游戏合集转存文件。需要一块支持Vulkan 1.3且显存6GB的显卡。采用GPL v2许可。","easeRationale":"下载的是一个现成的压缩包:解压后运行Uncharted.exe即可。真正困难的额外步骤是——你必须已经拥有自己解密好的PS4游戏合集转存文件,这个项目不提供,普通PS4也不会生成这种文件。","costNote":"启动器本身是免费的。但你仍需要一份PS4游戏的拷贝,那是付费产品。","notVerified":["它在你自己的显卡和处理器上运行效果如何。","压缩包里实际包含什么内容,发布的版本是否与本仓库的源码一致。","加入你自己的游戏文件后,实际需要多少磁盘空间。","启动器的自动更新功能是否只从本项目的发布版本下载。"]}}},{"id":"R_kgDOQX8qIw","slug":"supertone-oss-archive__supertonic","name":"Supertonic","owner":"supertone-oss-archive","fullName":"supertone-oss-archive/supertonic","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"young_owner_account","severity":"attention","evidence":"作者账号创建于 62 天前","source":"check","title":"作者账号刚注册不久"},{"code":"external_download_links","severity":"attention","evidence":"README 链接到 https://drive.google.com/file/d/1eancUOhiSXCVoTu9ddh4S-OcVQaWrPV-/view?usp=sharing","source":"check","title":"指向站外的下载链接"},{"code":"archived","severity":"info","evidence":"该仓库已归档（只读，不再维护）","source":"check","title":"仓库已归档"},{"code":"model_license_restricts_use","severity":"attention","evidence":"README.md 的 \"License\"：\"This project's sample code is released under the MIT License\"，但 \"The accompanying model is released under the OpenRAIL-M License\"。真正有用的语音模型带有 MIT 代码所没有的使用限制。","source":"ai","title":"模型许可证限制使用方式"},{"code":"project_discontinued_by_company","severity":"attention","evidence":"README.md 开头写着 \"This repository is archived. Development and support have ended... No updates, bug fixes, security patches, or support will be provided\"，保留下来的公告还写道 \"Voice Builder will no longer be accessible after August 31, 2026\"。在线演示和定制声音服务都已下线。","source":"ai","title":"公司已终止该项目"},{"code":"model_files_downloaded_separately","severity":"info","evidence":"Quick Start 用 \"hf download supertone-oss-archive/supertonic-3 --revision aafc6e32... --local-dir assets\" 下载权重。仓库本身不含模型，因此音质取决于从 Hugging Face 取回的文件。","source":"ai","title":"模型文件要另行下载"},{"code":"no_voice_cloning","severity":"info","evidence":"README.md：\"This open-weight repository focuses on fixed-voice, local TTS and does not include an official voice-cloning pipeline.\" 你能用的是预设声音，而不是你自己的声音。","source":"ai","title":"不支持声音克隆"},{"code":"comparisons_made_by_the_authors","severity":"info","evidence":"那些给 Supertonic 打勾、给 ElevenLabs、OpenAI 和 Gemini 打叉的表格，以及准确率图表，都是项目作者自己做的；示例音频则放在 Google Drive 链接上。","source":"ai","title":"对比由作者自己完成"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai","development"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"speechify","name":"Speechify","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":13752,"downloads":0,"downloadsWeek":0,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":13755},{"date":"2026-11-09","stars":13754},{"date":"2026-11-10","stars":13752}],"firstPublishedAt":"2026-11-10T23:52:44.165Z","commitSha":"1e9799e964ea4c0dad7cde993b65c3c813a7b373","verifiedAt":"2026-11-10T23:35:36.319Z","lastCommitAt":"2026-09-09T06:53:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns written text into spoken audio on your own computer, in 31 languages, with no internet and no paid service.","summary":"Supertonic is a small text-to-speech model with example code for Python, Node.js, browser, Java, C++, C#, Go, Swift, iOS, Rust and Flutter. It reads text aloud in 31 languages and saves a 44.1 kHz WAV file, running on the processor alone, with no graphics card and no internet once the files are downloaded. It is not a finished program: you clone the repository, download the model from Hugging Face and run an example. The company ended the project and the repository is archived.","easeRationale":"The documented path is the same on Windows, Mac and Linux: install Python 3.11, create a virtual environment, download the model files with the Hugging Face tool and run \"python example_onnx.py\" from the py/ folder. The examples for Go, Java, C# and C++ need extra runtimes on top of that.","notVerified":["How natural the voices sound in each of the 31 languages","The real speed and memory use on an ordinary computer or a Raspberry Pi","The total disk space after downloading the model files","Whether the examples for Node.js, browser, Java, C++, C#, Go, Swift, Rust and Flutter still build today","What the OpenRAIL-M license allows for commercial work"]},"zh-Hans":{"summaryShort":"在自己的电脑上把文字转成语音，支持 31 种语言，不联网、不用付费服务。","summary":"Supertonic 是一个小体积的文字转语音模型，附带 Python、Node.js、浏览器、Java、C++、C#、Go、Swift、iOS、Rust 和 Flutter 的示例代码。它能用 31 种语言朗读文本并保存为 44.1 kHz 的 WAV 文件，只靠处理器运行，不需要显卡，文件下载完成后也不需要联网。它不是成品软件：你要克隆仓库、从 Hugging Face 下载模型，再运行示例。公司已经终止该项目，仓库也已归档。","easeRationale":"文档给出的流程在 Windows、Mac 和 Linux 上相同：安装 Python 3.11，建一个虚拟环境，用 Hugging Face 的工具下载模型文件，然后在 py/ 目录下运行 \"python example_onnx.py\"。Go、Java、C# 和 C++ 的示例还需要各自额外的运行环境。","notVerified":["31 种语言下的语音听起来有多自然","在普通电脑或树莓派上的真实速度和内存占用","下载完模型文件后总共占用多少磁盘空间","Node.js、浏览器、Java、C++、C#、Go、Swift、Rust 和 Flutter 的示例如今是否还能编译","OpenRAIL-M 许可证在商业用途上允许做什么"]}}},{"id":"R_kgDOM9RF8g","slug":"swivid__f5-tts","name":"F5-TTS","owner":"SWivid","fullName":"SWivid/F5-TTS","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"models_not_free_for_business","severity":"attention","evidence":"README.md 的 License：\"Our code is released under MIT License. The pre-trained models are licensed under the CC-BY-NC license due to the training data Emilia, which is an in-the-wild dataset.\" 你生成的语音来自禁止商业使用的模型。","source":"ai","title":"模型不可免费商用"},{"code":"clones_anyone_s_voice","severity":"attention","evidence":"命令行接受 \"--ref_audio 'provide_prompt_wav_path_here.wav'\" 以及对应文字稿，然后用那个声音读出任意文本，项目标题就是 \"A Fairytaler that Fakes Fluent and Faithful Speech\"。仓库里没有任何检查授权同意的机制，README 中的水印图片也被注释掉了。","source":"ai","title":"可以克隆任何人的声音"},{"code":"downloads_models_on_first_use","severity":"info","evidence":"依赖项 \"cached_path\" 和 Docker 卷 \"/root/.cache/huggingface/hub/\" 表明，模型文件是在第一次运行时从 Hugging Face 取回的；Gradio 应用还提供 \"Voice Chat powered by Qwen2.5-3B-Instruct\"，这是另一个大体积下载。","source":"ai","title":"首次使用时才下载模型"},{"code":"public_link_option","severity":"info","evidence":"README.md 给出了 \"f5-tts_infer-gradio --share\" 用于 \"Launch a share link\"，这会把运行在你机器上的界面通过一个公开地址暴露出去。","source":"ai","title":"带有公开链接选项"},{"code":"research_code","severity":"info","evidence":"该仓库是 arXiv 论文 2410.06885 的 \"Official code\"，包含训练、评测和微调目录。它面向研究者发布，并没有打包成成品应用。","source":"ai","title":"这是研究代码"}],"platforms":["windows","mac","linux"],"tools":["python","ffmpeg","gpu_any"],"buildTools":["python","docker","gpu_any","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"kits-ai","name":"Kits AI","confirmed":true},{"slug":"fish-audio","name":"Fish Audio","confirmed":true}],"declaredAlternatives":[],"stars":15560,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":48,"growthDays":6,"history":[{"date":"2026-11-07","stars":15530},{"date":"2026-11-08","stars":15535},{"date":"2026-11-09","stars":15546},{"date":"2026-11-10","stars":15560}],"firstPublishedAt":"2026-11-10T23:52:44.179Z","commitSha":"600e5fdfbf319f84735557aca80a7aaf60d487e6","verifiedAt":"2026-11-10T23:36:13.806Z","lastCommitAt":"2026-11-03T15:05:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Copies a voice from a short recording and makes it read any text, running on your own computer.","summary":"F5-TTS is the research code of a text-to-speech model that copies a voice. You give it a short reference recording with its transcript, type the text you want, and it produces audio in that voice, in English or Chinese and other languages added by the community. It opens a web page on your own machine (Gradio) or runs from the terminal, and it can also be trained on new voices. You install it with Python and PyTorch, and a graphics card is strongly recommended.","easeRationale":"On all three systems you create a Python 3.10+ environment, install FFmpeg, pick the PyTorch build that matches your hardware (CUDA, ROCm, Intel XPU or Apple Silicon) and then \"pip install f5-tts\" before typing \"f5-tts_infer-gradio\" to open the local web page. There is also a Docker image, which still needs an NVIDIA card with --gpus=all.","notVerified":["How close the cloned voice sounds to the original recording","Whether it runs at a usable speed without an NVIDIA card, for example on a Mac","The total disk space after the model files are downloaded","Which languages beyond English and Chinese work well, since they come from community models","How much memory is needed when the reference text is transcribed automatically"]},"zh-Hans":{"summaryShort":"用一小段录音复制出某个声音，让它朗读任意文字，全程在你自己的电脑上运行。","summary":"F5-TTS 是一个会复制声音的文字转语音模型的研究代码。你提供一小段参考录音及其文字稿，再输入想要朗读的文本，它就会用那个声音生成音频，支持中英文以及社区补充的其他语言。它可以在本机打开一个网页界面（Gradio），也可以在终端里运行，还能用新的声音继续训练。安装需要 Python 和 PyTorch，并强烈建议配备显卡。","easeRationale":"三个系统上的步骤相同：建一个 Python 3.10 以上的环境，安装 FFmpeg，按自己的硬件（CUDA、ROCm、Intel XPU 或 Apple Silicon）选择对应的 PyTorch 版本，然后运行 \"pip install f5-tts\"，再输入 \"f5-tts_infer-gradio\" 打开本地网页。项目也提供 Docker 镜像，但同样需要 NVIDIA 显卡并加上 --gpus=all。","notVerified":["克隆出的声音与原始录音有多接近","在没有 NVIDIA 显卡的设备上（例如 Mac）速度是否仍可接受","下载完模型文件后总共占用多少磁盘空间","除中英文之外，还有哪些语言效果不错，因为它们来自社区模型","当参考文本由系统自动转写时，需要多少内存"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNDYxNDM0NA==","slug":"swordfish90__cool-retro-term","name":"cool-retro-term","owner":"Swordfish90","fullName":"Swordfish90/cool-retro-term","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 目录和 winget 清单中都没有发现任何许可证","source":"check","title":"没有许可证"},{"code":"low_scorecard","severity":"attention","evidence":"OpenSSF Scorecard 2.6/10","source":"check","title":"Scorecard 评分偏低"},{"code":"latest_versions_are_beta","severity":"info","evidence":"最新发布的标签是 \"2.0.0-beta2\"（2026-05-31），另有一个 \"rolling\" 标签存放以单个提交命名的文件（cool-retro-term-0269a33.AppImage 等）。目前可下载的都是测试构建，而不是正式版。","source":"ai","title":"最新版本仍是 Beta"},{"code":"license_files_without_a_declared_license","severity":"info","evidence":"根目录里有 gpl-2.0.txt 和 gpl-3.0.txt，但没有 LICENSE 文件，GitHub 也识别不出任何许可证，因此这款程序到底适用哪些条款并不明确。","source":"ai","title":"有许可证文本却没有声明许可证"},{"code":"slow_pace_and_many_open_issues","severity":"info","evidence":"该项目有 578 个未关闭的问题，最后一次提交是 2026-05-31，而仓库建于 2013 年。","source":"ai","title":"进展缓慢且积压大量问题"},{"code":"it_is_a_real_terminal","severity":"info","evidence":"README.md：\"It uses the QML port of qtermwidget (Konsole)\"。在这个窗口里输入的命令，会以你当前用户的权限在电脑上真实执行，和其他终端完全一样。","source":"ai","title":"它是一个真正的终端"}],"platforms":["linux","mac"],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["network_system"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":26567,"downloads":176367,"downloadsWeek":501,"downloadsDays":5,"growth":16,"growthDays":6,"history":[{"date":"2026-11-05","stars":26555},{"date":"2026-11-06","stars":26556},{"date":"2026-11-07","stars":26559},{"date":"2026-11-08","stars":26563},{"date":"2026-11-09","stars":26564},{"date":"2026-11-10","stars":26567}],"firstPublishedAt":"2026-11-10T23:52:44.198Z","commitSha":"1394ce82fa53d2a87d5adc4d99f9eeba598ea53d","verifiedAt":"2026-11-10T23:36:47.030Z","lastCommitAt":"2026-05-31T10:04:28Z","isMirror":false,"collectFailing":false,"installers":[{"os":"mac","file":"cool-retro-term-2.0.0-beta2.dmg","sizeBytes":93111461,"url":"https://github.com/Swordfish90/cool-retro-term/releases/download/2.0.0-beta2/cool-retro-term-2.0.0-beta2.dmg","releaseUrl":"https://github.com/Swordfish90/cool-retro-term/releases/tag/2.0.0-beta2"},{"os":"linux","file":"cool-retro-term-2.0.0-beta2.AppImage","sizeBytes":76294648,"url":"https://github.com/Swordfish90/cool-retro-term/releases/download/2.0.0-beta2/cool-retro-term-2.0.0-beta2.AppImage","releaseUrl":"https://github.com/Swordfish90/cool-retro-term/releases/tag/2.0.0-beta2"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A terminal window that looks like an old green or amber cathode screen, with flicker and curved glass effects.","summary":"cool-retro-term is a terminal for Linux and macOS made to look like the old cathode tube monitors: green or amber letters, scan lines, glow and a curved screen. Underneath it behaves like a normal terminal, and colours, fonts and effects are changed in the right-click menu. You download a ready file from the releases page (AppImage for Linux, dmg for Mac), or install it from your Linux distribution's own store. There is no Windows version.","easeRationale":"On Linux it is already packaged by Ubuntu, Fedora and Arch, so it installs like any other program, and a 73 MB AppImage is offered as well; on Mac you download the 89 MB dmg from the releases page and open it. Nothing else has to be installed.","notVerified":["Whether the Mac dmg opens without warnings about an unidentified developer","How much the visual effects load the graphics card and the battery","How much disk space it uses after installation","Whether the beta is stable for daily use"]},"zh-Hans":{"summaryShort":"一个看起来像老式显像管屏幕的终端窗口，绿色或琥珀色字符，带闪烁和弧面玻璃效果。","summary":"cool-retro-term 是面向 Linux 和 macOS 的终端，外观刻意模仿老式显像管显示器：绿色或琥珀色字符、扫描线、荧光和弧形屏幕。底层它就是个普通终端，颜色、字体和特效都可以在右键菜单里调整。你可以从发布页下载现成文件（Linux 用 AppImage，Mac 用 dmg），也可以通过 Linux 发行版自带的软件源安装。它没有 Windows 版本。","easeRationale":"在 Linux 上，Ubuntu、Fedora 和 Arch 都已打包，可以像装普通软件一样安装，另外还提供 73 MB 的 AppImage；在 Mac 上，从发布页下载 89 MB 的 dmg 打开即可。不需要再装别的东西。","notVerified":["Mac 的 dmg 打开时是否会提示来自身份不明的开发者","这些视觉特效会给显卡和电池带来多大负担","安装之后会占用多少磁盘空间","Beta 版本用于日常工作是否足够稳定"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxNDcxMjg1MA==","slug":"syncthing__syncthing","name":"Syncthing","owner":"syncthing","fullName":"syncthing/syncthing","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"low","signals":[{"code":"listens_on_network_ports","severity":"info","evidence":"Dockerfile：\"EXPOSE 8384 22000/tcp 22000/udp 21027/udp\" — 该程序会打开一个用于设置的网页，并使用网络端口与你的其他设备通信，这可能需要路由器或防火墙的许可。","source":"ai","title":"监听网络端口"},{"code":"built_in_auto_upgrade","severity":"info","evidence":"README.md：\"There is also a built-in automatic upgrade mechanism (disabled in some distribution channels) which uses a compiled in ECDSA signature.\" 该程序可以自行下载并安装新版本。","source":"ai","title":"内置自动升级"},{"code":"signed_releases","severity":"info","evidence":"README.md：\"Release binaries are GPG signed with the key available from https://syncthing.net/security/ ... macOS and Windows binaries are also code-signed.\"","source":"ai","title":"发布版本带签名"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","go"],"disk":{"minMB":14,"maxMB":41,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["productivity","network_system"],"alternativeTo":[{"slug":"dropbox","name":"Dropbox","confirmed":true},{"slug":"resilio-sync","name":"Sync","confirmed":true}],"declaredAlternatives":[],"stars":90322,"downloads":112032844,"downloadsWeek":59272,"downloadsDays":7,"growth":275,"growthDays":7,"history":[{"date":"2026-10-25","stars":89746},{"date":"2026-10-26","stars":89781},{"date":"2026-10-27","stars":89818},{"date":"2026-10-28","stars":89845},{"date":"2026-10-29","stars":89879},{"date":"2026-10-30","stars":89916},{"date":"2026-10-31","stars":89951},{"date":"2026-11-01","stars":89995},{"date":"2026-11-02","stars":90020},{"date":"2026-11-03","stars":90047},{"date":"2026-11-04","stars":90077},{"date":"2026-11-05","stars":90131},{"date":"2026-11-06","stars":90163},{"date":"2026-11-07","stars":90195},{"date":"2026-11-08","stars":90238},{"date":"2026-11-09","stars":90283},{"date":"2026-11-10","stars":90322}],"firstPublishedAt":"2026-10-25T23:19:30.883Z","commitSha":"ed38528d7fba3dd6a42c9c9fc6780af432dc15d2","verifiedAt":"2026-10-30T01:42:04.246Z","lastCommitAt":"2026-11-09T03:48:13Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Keep folders automatically in sync between your own devices, directly and without a cloud service.","summary":"Syncthing copies changes in the folders you choose between two or more of your devices as they happen, so the same files are on each one. Files travel straight from device to device, encrypted, without being stored on a company's server. It is for anyone who wants Dropbox-style syncing without cloud storage or a subscription; it is not a backup, because there is no online copy. You run a small program on each device and manage it in your web browser. Ready downloads exist for Windows, macOS and Linux (the Windows one is a 14 MB zip).","easeRationale":"On Windows the release is a ready program in a zip (syncthing-windows-amd64-v2.1.7.zip, about 14 MB) that you unzip and run, and then you pair devices by exchanging device IDs in the browser page; macOS is a zip with the same program and Linux also has .deb packages, so all three need that same pairing step and some extra setup to start with the system.","notVerified":["How it behaves on Windows when files are open or locked by other programs during sync.","Whether conflicts are handled the way you expect when the same file changes on two devices.","Whether your router or firewall lets the devices find each other without manual changes.","How the separate community helper apps that add a Windows or Mac interface work; they are not part of this repository."]},"zh-Hans":{"summaryShort":"在你自己的设备之间自动保持文件夹同步，直接传输，不需要云服务。","summary":"Syncthing 会在你选择的文件夹发生变化时，立即把变化复制到你两台或更多设备之间，让每台设备上的文件保持一致。文件直接从一个设备传到另一个设备，并经过加密，不会存放在任何公司的服务器上。它适合那些想要类似 Dropbox 的同步效果、但不想用云存储或订阅的人；它不是备份工具，因为没有在线保存的副本。你需要在每台设备上运行一个小程序，并通过网页浏览器来管理它。Windows、macOS 和 Linux 都已提供可直接使用的下载（Windows 版是一个 14 MB 的 zip 文件）。","easeRationale":"在 Windows 上，发行版是一个压缩包里的现成程序（syncthing-windows-amd64-v2.1.7.zip，约 14 MB），你解压并运行它，然后在浏览器页面上通过交换设备 ID 来配对设备；macOS 版本是包含同一程序的压缩包，Linux 还提供 .deb 安装包，所以这三个系统都需要同样的配对步骤，以及一些让程序随系统启动的额外设置。","notVerified":["在 Windows 上，当文件在同步过程中被其他程序打开或锁定时会怎样处理。","当同一个文件在两台设备上都发生变化时，冲突是否会按你预期的方式处理。","你的路由器或防火墙是否能让设备之间互相找到对方，而不需要手动调整设置。","社区提供的、用于添加 Windows 或 Mac 界面的独立辅助应用是如何工作的；它们并不属于这个代码仓库。"]}}},{"id":"R_kgDOVAlDig","slug":"t-auto__dsh-ops","name":"DSH Ops","owner":"T-Auto","fullName":"T-Auto/dsh-ops","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"24 天内获得 5008 个星标","source":"check","title":"新仓库，星标很多"},{"code":"hides_host_tools","severity":"info","evidence":"README.md：\"提供新增的ops系列工具并隐藏部分官方工具\" 以及 \"（仅windows平台）启用与dsh原生bash行为一致的ops_bash，并可选禁用powershell\"；package.json 导出了一个 \"./hide-official\" 组件和一个 \"cordis.patch.yml\" 打包补丁。","source":"ai","title":"隐藏宿主程序的工具"},{"code":"third_party_npm_mirror","severity":"info","evidence":"README.md 让用户从自定义镜像源安装：\"选择自定义地址，输入 https://mirrors.huaweicloud.com/repository/npm/\"，而不是官方 npm 仓库。","source":"ai","title":"第三方 npm 镜像"},{"code":"installs_latest_without_pinning","severity":"info","evidence":"README.md 的安装和更新命令：\"npx --yes dsh-ops@latest install --profile desktop\"。每次运行都会获取 npm 上最新的版本。","source":"ai","title":"安装最新版而不固定版本"},{"code":"bundled_bash_payload","severity":"info","evidence":"发布文件 \"dsh-ops-bash-win32-x64-0.2.1.tgz (windows, 154 MB)\"，以及 README.md：可选包 @dsh-ops/bash-win32-x64 包含 Git for Windows 的组件，采用 GPL-2.0 和其他许可证，与插件的 MIT 许可证相互独立。","source":"ai","title":"捆绑的 bash 组件包"},{"code":"unpublished_benchmark","severity":"info","evidence":"README.md 图片说明：\"ops在自用benchmark上取得的的成绩。完整benchmark将随后续更新发布\"。也就是说，这是内部基准测试的结果，完整的基准测试之后才会发布。","source":"ai","title":"未公开的基准测试"}],"platforms":["windows","mac","linux"],"tools":["other"],"buildTools":["node"],"disk":{"minMB":154,"maxMB":461,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":5669,"downloads":9227,"downloadsWeek":2845,"downloadsDays":7,"growth":528,"growthDays":7,"history":[{"date":"2026-10-26","stars":4441},{"date":"2026-10-27","stars":4576},{"date":"2026-10-28","stars":4689},{"date":"2026-10-29","stars":4779},{"date":"2026-10-30","stars":4866},{"date":"2026-10-31","stars":4941},{"date":"2026-11-01","stars":5007},{"date":"2026-11-02","stars":5064},{"date":"2026-11-03","stars":5141},{"date":"2026-11-04","stars":5214},{"date":"2026-11-05","stars":5300},{"date":"2026-11-06","stars":5435},{"date":"2026-11-07","stars":5517},{"date":"2026-11-08","stars":5586},{"date":"2026-11-09","stars":5635},{"date":"2026-11-10","stars":5669}],"firstPublishedAt":"2026-10-27T01:10:12.696Z","commitSha":"065cd7fba266a24ffded20c78a404860c033ba72","verifiedAt":"2026-11-02T02:48:40.683Z","lastCommitAt":"2026-11-10T15:29:20Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A plugin for the DeepSeek Harness (DSH) AI coding app that adds leaner file tools to cut token use and task time.","summary":"dsh-ops is a plugin for DSH (DeepSeek Harness), an AI coding assistant app. It adds its own file and document tools (including PDF and Word reading and bulk edits), hides some of DSH's built-in tools, and returns shorter, line-numbered results so the AI spends fewer tokens. On Windows it can add a Bash shell based on Git for Windows and optionally turn off PowerShell. It is for people who already use DSH 0.2.0-rc.2 or 0.2.1-rc.2. You install it from DSH's plugin marketplace or with an npx command.","easeRationale":"If DSH is already installed, you add it in DSH's plugin marketplace by typing the package name and a custom npm mirror address, then enable it; the npx route needs Node.js.","costNote":"The plugin is free under the MIT license. DSH itself runs on AI model tokens, which may cost money.","notVerified":["Whether it really reduces token use and task time as claimed.","Whether hiding DSH's built-in tools or turning off PowerShell breaks other plugins or workflows.","Whether it keeps working with DSH versions other than 0.2.0-rc.2 and 0.2.1-rc.2.","What exactly is inside the 154 MB Windows Bash package.","Whether the Huawei Cloud npm mirror serves the same files as the official npm registry."]},"zh-Hans":{"summaryShort":"DeepSeek Harness（DSH）AI 编程应用的插件，提供更精简的文件工具，以减少 token 用量和任务耗时。","summary":"dsh-ops 是 DSH（DeepSeek Harness）的插件，DSH 是一款 AI 编程助手应用。它添加了自己的文件和文档工具（包括读取 PDF 和 Word 以及批量编辑），隐藏 DSH 的部分内置工具，并返回更短、带行号的结果，让 AI 少消耗 token。在 Windows 上，它可以添加基于 Git for Windows 的 Bash shell，并可选择关闭 PowerShell。适合已经在使用 DSH 0.2.0-rc.2 或 0.2.1-rc.2 的人。可通过 DSH 的插件市场或 npx 命令安装。","easeRationale":"如果已经装好 DSH，就在 DSH 的插件市场里输入包名和自定义 npm 镜像地址来添加，然后启用；npx 方式需要 Node.js。","costNote":"插件本身在 MIT 许可证下免费。DSH 本身要消耗 AI 模型 token，这可能需要花钱。","notVerified":["它是否真的像宣称的那样减少了 token 用量和任务耗时。","隐藏 DSH 的内置工具或关闭 PowerShell 是否会破坏其他插件或工作流程。","它在 0.2.0-rc.2 和 0.2.1-rc.2 之外的 DSH 版本上是否还能正常使用。","154 MB 的 Windows Bash 包里究竟有什么。","华为云 npm 镜像提供的文件是否与官方 npm 仓库一致。"]}}},{"id":"R_kgDOVVf_hg","slug":"t3dotgg__tstack","name":"tstack","owner":"t3dotgg","fullName":"t3dotgg/tstack","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"agent_can_merge_pull_requests_unattended","severity":"attention","evidence":"README 的技能表格：`full-send` \"Takes a change to a merged PR with nobody watching: proves it, reviews it, files it, babysits it, merges it.\" `full-stop` 技能被列为在合并前停下来的变体。","source":"ai","title":"代理可以无人值守地合并拉取请求"},{"code":"acts_on_your_github_account","severity":"attention","evidence":"README 的要求：\"An authenticated GitHub CLI (gh)\"。这些技能会开拉取请求、以你的身份发表评论（`leaving-pr-comments`），还会接管别人的分支，全部都在你的账号下进行。","source":"ai","title":"以你的 github 账号行事"},{"code":"some_features_need_another_tool","severity":"info","evidence":"README：一些技能 \"use T3 Code features when they are available and fall back to native tools when they are not\"，而 `council` 和 `sol-loop` \"need subagents on other providers' models\"，T3 Code 通过 `delegate_task` 来实现这一点。","source":"ai","title":"部分功能需要另一个工具"},{"code":"very_new_repository_fast_growth","severity":"info","evidence":"GitHub 数据：仓库创建于 2026-10-24，六天后就有 2471 个星标，没有发行版，只有一个 README 和技能文件夹。","source":"ai","title":"仓库非常新，增长很快"},{"code":"instructions_run_as_your_assistant","severity":"info","evidence":"整个仓库都是文本，由你的 AI 助手读取并执行（顶层文件：LICENSE、README.md、skills/）。更新技能意味着再复制一遍新文本，README 还让你去编辑它们：\"These are starting points. Edit them.\"","source":"ai","title":"指令以你的助手身份运行"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["GitHub (through an authenticated gh CLI)","Models from other providers, for the council and sol-loop reviews"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","automation"],"alternativeTo":[{"slug":"coderabbit","name":"CodeRabbit","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true},{"slug":"qodo","name":"Qodo","confirmed":true}],"declaredAlternatives":[],"stars":3445,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":735,"growthDays":7,"history":[{"date":"2026-10-27","stars":2008},{"date":"2026-10-28","stars":2265},{"date":"2026-10-29","stars":2396},{"date":"2026-10-30","stars":2469},{"date":"2026-10-31","stars":2536},{"date":"2026-11-01","stars":2592},{"date":"2026-11-02","stars":2638},{"date":"2026-11-03","stars":2710},{"date":"2026-11-04","stars":2941},{"date":"2026-11-05","stars":3079},{"date":"2026-11-06","stars":3177},{"date":"2026-11-07","stars":3233},{"date":"2026-11-08","stars":3281},{"date":"2026-11-09","stars":3330},{"date":"2026-11-10","stars":3445}],"firstPublishedAt":"2026-10-27T18:40:28.072Z","commitSha":"1085050f3baed1ac724caa4a8166145fb2ac88af","verifiedAt":"2026-10-31T01:40:59.979Z","lastCommitAt":"2026-10-28T02:38:02Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Nine agent skills for Claude Code and Codex that carry a code change through review, pull request, CI and merge.","summary":"tstack is a set of nine instruction packs (\"skills\") that you drop into Claude Code or Codex. Once installed you say things like \"full send\" or \"babysit\", and the assistant takes a code change through review by other AI models, opens a pull request on GitHub, watches the CI checks and review bots, fixes what they find and can merge it. It is for developers who already use an AI coding assistant. You need the GitHub CLI logged in and Python 3; the files are copied in by hand, there is no installer.","easeRationale":"The same on every platform: no installer. You clone the repository and copy the skill folders into ~/.claude/skills or ~/.codex/skills, then install and log in to the GitHub CLI and have Python 3 available. On Windows the README's cp commands need a shell like Git Bash.","costNote":"The skills are MIT. The AI assistants they run in, and the extra models the review skills call, are paid separately.","notVerified":["Whether an unattended run really stops at a bad change, since the only guard described is other AI models reviewing it.","How much the multi-model review loops cost in tokens for one pull request.","Whether the skills behave the same outside T3 Code, where the fallback path is used.","Whether the automatic merge respects your repository's branch protection and required reviews."]},"zh-Hans":{"summaryShort":"面向 Claude Code 和 Codex 的九个代理技能，带着代码改动走完审查、拉取请求、CI 和合并全流程。","summary":"tstack 是一套九个指令包（“技能”），你把它们放进 Claude Code 或 Codex。安装后，你只需说“full send”或“babysit”之类的话，助手就会让其他 AI 模型审查这次代码改动，在 GitHub 上开一个拉取请求，盯着 CI 检查和审查机器人，修复它们发现的问题，还能完成合并。它面向已经在用 AI 编程助手的开发者。你需要已登录的 GitHub CLI 和 Python 3；文件是手动复制进去的，没有安装程序。","easeRationale":"在任何平台上都一样：没有安装程序。你克隆仓库，把技能文件夹复制到 ~/.claude/skills 或 ~/.codex/skills，然后安装并登录 GitHub CLI，确保有 Python 3 可用。在 Windows 上，README 里的 cp 命令需要像 Git Bash 这样的 shell。","costNote":"这些技能是 MIT 许可的。它们所运行的 AI 助手，以及审查技能调用的额外模型，需要单独付费。","notVerified":["无人值守运行时是否真的会在改动出问题时停下来，因为文中描述的唯一保障就是其他 AI 模型的审查。","一次拉取请求的多模型审查循环在 token 上要花多少钱。","在 T3 Code 之外使用回退路径时，这些技能是否表现一致。","自动合并是否遵守你仓库的分支保护和必需的审查要求。"]}}},{"id":"R_kgDOSPdiBQ","slug":"t8y2__dbx","name":"DBX","owner":"t8y2","fullName":"t8y2/dbx","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A","web":"D"},"risk":"attention","signals":[{"code":"every_release_is_a_beta","severity":"attention","evidence":"最新的发布版标签形如 v0.6.69-beta.20261110.1，仓库自 2026-04-29 建立以来已发布 431 次，未关闭的问题有 1729 个。程序版本号还不到 1.0，几乎每天都在变，却要用来操作真实的数据库。","source":"ai","title":"每个发布版都是 Beta"},{"code":"stored_passwords_can_be_decrypted_by_whoever_copies_the_folder","severity":"attention","evidence":"README.md 在介绍 Windows 便携版时写道：\"Keeping the key beside the database means anyone who obtains or copies the whole drive can also decrypt its credentials.\" 关于 Docker 它也说，密钥 \"cannot protect the whole data volume if the volume itself is copied or exposed\"。","source":"ai","title":"谁复制了整个目录，谁就能解开保存的密码"},{"code":"ai_agents_can_write_to_your_databases","severity":"attention","evidence":"README.md 介绍 MCP 服务器有 \"Read only, Data read/write, and Full access modes\"（\"read_only\"、\"safe_write\"、\"high_risk_write\"），让 Claude Code、Cursor 或 Windsurf \"can query your databases through connections you already set up\"。在权限更大的模式下，AI 助手可以修改数据。","source":"ai","title":"AI 智能体可以写入你的数据库"},{"code":"ai_assistant_sends_your_data_outside","severity":"info","evidence":"README.md：\"Highlight a table, describe what you want, get SQL back... Works with Claude, OpenAI, or local models via Ollama. Built-in safety checks review AI-generated SQL before it runs.\" 如果用的是在线模型，表名和你选中的文本就会离开你的电脑。","source":"ai","title":"AI 助手会把你的数据发到外部"},{"code":"readme_full_of_sponsor_and_referral_links","severity":"info","evidence":"Sponsors 一节放了多家 AI API 转售商的带折扣联盟链接（AICodeMirror \"register through this link to enjoy 20% off\"、PackyCode \"aff=wyDA\"、Atlas Cloud \"?ref=6YYXWA\"、HiAPI \"invite/dbx\"）。页面把产品介绍和付费推广混在了一起。","source":"ai","title":"README 里满是赞助和推广链接"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["node","rust"],"disk":{"minMB":241,"maxMB":722,"basis":"installer"},"apiKeys":["Claude, OpenAI or another OpenAI-compatible model, for the AI assistant"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_skills"],"alternativeTo":[{"slug":"datagrip","name":"DataGrip","confirmed":true},{"slug":"navicat","name":"Navicat","confirmed":true},{"slug":"dbeaver-pro","name":"PRO","confirmed":true},{"slug":"tableplus","name":"TablePlus","confirmed":true}],"declaredAlternatives":[],"stars":29912,"downloads":4284573,"downloadsWeek":329626,"downloadsDays":6,"growth":843,"growthDays":6,"history":[{"date":"2026-11-04","stars":29069},{"date":"2026-11-05","stars":29214},{"date":"2026-11-06","stars":29373},{"date":"2026-11-07","stars":29470},{"date":"2026-11-08","stars":29567},{"date":"2026-11-09","stars":29724},{"date":"2026-11-10","stars":29912}],"firstPublishedAt":"2026-11-10T23:52:44.209Z","commitSha":"bef20fdc2214c4af8fb0aff1ae08ce8a1a309c53","verifiedAt":"2026-11-10T23:37:38.677Z","lastCommitAt":"2026-11-10T14:24:35Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Opens MySQL, PostgreSQL, SQLite, Redis, MongoDB and 100+ other databases in one small desktop app.","summary":"DBX is a desktop program to look at and change data in more than 100 kinds of database, plus message queues such as Kafka and RabbitMQ. It has an SQL editor with autocomplete, a table grid with inline editing, diagrams of table relations, import and export in CSV, Excel and JSON, and SSH tunnels. An AI assistant writes SQL from plain language using your own Claude, OpenAI or local model key, and an MCP server lets AI assistants query the same connections. Installers exist for Windows, Mac and Linux.","easeRationale":"On Windows there is a 33 MB installer (also winget and scoop), on Mac a dmg or \"brew install --cask dbx\", and on Linux deb, rpm, AppImage and Flatpak: download and open. The browser version is different: it is a self-hosted Docker container on port 4224, which means Docker and a server.","costNote":"Apache-2.0 and free. The AI assistant uses a key you provide (Claude, OpenAI, any OpenAI-compatible service or a local model through Ollama).","notVerified":["Whether all the listed databases really work, since many are reached through agent or JDBC drivers","How stable the beta is with large databases and long sessions","Whether the installers are signed on Windows and Mac","How the built-in automatic update behaves and where it downloads from","Whether the plugin store signature check works as described"]},"zh-Hans":{"summaryShort":"用一个小巧的桌面程序打开 MySQL、PostgreSQL、SQLite、Redis、MongoDB 等 100 多种数据库。","summary":"DBX 是一款桌面程序，可以查看和修改 100 多种数据库里的数据，还支持 Kafka、RabbitMQ 等消息队列。它带有自动补全的 SQL 编辑器、可直接改数据的表格视图、表关系图、CSV/Excel/JSON 的导入导出，以及 SSH 隧道。AI 助手可以用你自己的 Claude、OpenAI 或本地模型密钥，把自然语言变成 SQL；它还提供 MCP 服务器，让 AI 助手查询同样的连接。Windows、Mac 和 Linux 都有安装包。","easeRationale":"Windows 上有 33 MB 的安装程序（也支持 winget 和 scoop），Mac 上有 dmg 或 \"brew install --cask dbx\"，Linux 上有 deb、rpm、AppImage 和 Flatpak：下载打开即可。浏览器版则不同：那是你自己部署、监听 4224 端口的 Docker 容器，需要 Docker 和一台服务器。","costNote":"采用 Apache-2.0，免费。AI 助手使用你自己提供的密钥（Claude、OpenAI、任何兼容 OpenAI 的服务，或通过 Ollama 使用本地模型）。","notVerified":["列出的所有数据库是否都真的可用，因为不少是通过代理或 JDBC 驱动连接的","面对大型数据库和长时间使用，这个 Beta 版本有多稳定","Windows 和 Mac 上的安装包是否带有签名","内置的自动更新具体如何工作、从哪里下载","插件市场的签名校验是否真如文档所述那样生效"]}}},{"id":"R_kgDOU_9jPg","slug":"tangency-labs__solidcraft","name":"SolidCraft","owner":"tangency-labs","fullName":"tangency-labs/solidcraft","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"19 天内获得 10824 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 10 天前","source":"check","title":"所有者账号很新"},{"code":"unsigned_downloads","severity":"attention","evidence":"README.md：\"The downloads aren't signed yet: on Windows choose **More info**, then **Run anyway** in SmartScreen\"。","source":"ai","title":"下载文件未签名"},{"code":"early_stage_software","severity":"info","evidence":"README.md：\"Early but usable for simple parts\"。","source":"ai","title":"早期阶段的软件"},{"code":"many_issues_in_short_time","severity":"info","evidence":"一个创建于 2026-10-07 的仓库（检查前 19 天）有 3547 个未关闭的 issue，在此期间发布了 58 个版本。","source":"ai","title":"短时间内出现大量 issue"},{"code":"repository_owner_mismatch","severity":"info","evidence":"Cargo.toml 设置了 repository = \"https://github.com/bherbruck/solidcraft\"，而该项目发布在 github.com/tangency-labs/solidcraft；README 还链接到 github.com/bherbruck/solidcore。","source":"ai","title":"仓库所有者不一致"}],"platforms":["windows","mac","linux","web"],"tools":[],"buildTools":["rust"],"disk":{"minMB":136,"maxMB":407,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["design_cad"],"alternativeTo":[{"slug":"autodesk-fusion","name":"Fusion","confirmed":true},{"slug":"solidworks","name":"SolidWorks","confirmed":true},{"slug":null,"name":"Fusion 360","confirmed":false}],"declaredAlternatives":[],"stars":15474,"downloads":1275463,"downloadsWeek":279318,"downloadsDays":7,"growth":2073,"growthDays":7,"history":[{"date":"2026-10-26","stars":10830},{"date":"2026-10-27","stars":11257},{"date":"2026-10-28","stars":11663},{"date":"2026-10-29","stars":12062},{"date":"2026-10-30","stars":12380},{"date":"2026-10-31","stars":12639},{"date":"2026-11-01","stars":12922},{"date":"2026-11-02","stars":13186},{"date":"2026-11-03","stars":13401},{"date":"2026-11-04","stars":13612},{"date":"2026-11-05","stars":14120},{"date":"2026-11-06","stars":14494},{"date":"2026-11-07","stars":14775},{"date":"2026-11-08","stars":14997},{"date":"2026-11-09","stars":15263},{"date":"2026-11-10","stars":15474}],"firstPublishedAt":"2026-10-27T01:10:12.697Z","commitSha":"56bcd872cb57aabcfdedb0767f1fbc198f923991","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-10T23:02:50Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"SolidCraft-0.30.1-windows-x64.exe","sizeBytes":172718280,"url":"https://github.com/tangency-labs/solidcraft/releases/download/v0.30.1/SolidCraft-0.30.1-windows-x64.exe","releaseUrl":"https://github.com/tangency-labs/solidcraft/releases/tag/v0.30.1"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Design 3D parts from measured sketches, change a size and watch the whole model update, then export them for 3D printing.","summary":"SolidCraft is a 3D design (CAD) app for making mechanical parts. You draw a sketch with exact sizes, turn it into a solid, and add holes, rounded edges, threads and more. Change a number later and the whole part rebuilds. It also handles assemblies and sheet metal, and opens and saves common formats such as STEP, STL and 3MF. The project calls itself early but usable for simple parts. There are installers for Windows, Mac and Linux and a version that runs in the browser.","easeRationale":"Windows has a ready .msi installer, a .exe and a portable .zip, and there is a browser version with no install; Rust is only needed to build it from source.","notVerified":["Whether complex parts and assemblies rebuild correctly without errors.","Whether STEP files from other CAD programs open accurately.","Whether the app sends crash reports, and what they contain.","Whether the built-in update process downloads code without asking.","Speed on PCs without a dedicated graphics card."]},"zh-Hans":{"summaryShort":"根据带尺寸的草图设计 3D 零件，修改一个尺寸，整个模型随之更新，再导出用于 3D 打印。","summary":"SolidCraft 是一款用于制作机械零件的 3D 设计（CAD）应用。你先画出带精确尺寸的草图，把它变成实体，再添加孔、圆角、螺纹等特征。之后只要修改一个数字，整个零件就会重新生成。它还支持装配体和钣金，并能打开和保存 STEP、STL、3MF 等常见格式。项目自称仍处于早期阶段，但简单零件已经可以使用。提供适用于 Windows、Mac 和 Linux 的安装程序，还有可在浏览器中运行的版本。","easeRationale":"Windows 上有现成的 .msi 安装程序、.exe 和便携 .zip，另有无需安装的浏览器版；只有从源代码构建时才需要 Rust。","notVerified":["复杂零件和装配体能否无误地正确重新生成。","来自其他 CAD 程序的 STEP 文件能否被准确打开。","应用是否会发送崩溃报告，以及报告里包含什么。","内置更新流程是否会在不询问的情况下下载代码。","在没有独立显卡的电脑上的运行速度。"]}}},{"id":"R_kgDOVmGDFA","slug":"tangerineturtle1__nx2navi","name":"NX2navi","owner":"tangerineturtle1","fullName":"tangerineturtle1/NX2navi","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B"},"risk":"high","signals":[{"code":"requires_dumped_commercial_game","severity":"high","evidence":"README.md:\"No game files are included. You need your own legally obtained copy of the game, dumped from your own console. Instructions for obtaining those files will not be provided.\" 该项目只能配合任天堂付费游戏的转储文件使用,而转储 Switch 2 卡带意味着绕过主机的保护机制,这在多个国家是被禁止的。","source":"ai","title":"需要转储的商业游戏"},{"code":"recompiles_copyrighted_game_code","severity":"high","evidence":"README.md 将该项目描述为 \"a static recompilation of The Legend of Zelda: Ocarina of Time for Nintendo Switch 2 (game version 1.1.1) into a native x86-64 application\",并警告 \"the result contains the game. It is for your own use. Do not share it.\" 任天堂的代码和资源都保留在你所构建的内容中。","source":"ai","title":"重新编译受版权保护的游戏代码"},{"code":"source_not_freely_reusable","severity":"attention","evidence":"GitHub 上标注的许可证是 MIT,但 README.md 写道 \"The source is public to read. Its additions and changes are all rights reserved (LICENSE-ADDITIONS): no part of them may be copied or published without written permission.\" 因此项目的大部分内容不能被重用或再分发。","source":"ai","title":"源代码不能自由重用"},{"code":"downloads_build_tools_at_run_time","severity":"attention","evidence":"README.md:安装好的启动器会 \"finds the dump, downloads the build tools, builds the game on your PC (10 to 25 minutes, once) and starts it\"。它获取哪些工具、从哪里获取,都是在运行时决定的。","source":"ai","title":"运行时下载构建工具"},{"code":"unsigned_installer","severity":"attention","evidence":"README.md 的 Requirements 部分写道 \"The setup is not signed, so SmartScreen warns on its first start.\" Windows 无法确认文件的创建者身份。","source":"ai","title":"安装程序未签名"},{"code":"very_new_repository","severity":"info","evidence":"该仓库创建于 2026-11-07,已经有 1237 个星标、2 个版本和 155 个未解决的 issue;README 称其为 \"Early development, playable\",并提示仍会出现 \"Visual bugs and occasional crashes\"。","source":"ai","title":"仓库非常新"}],"platforms":["windows"],"tools":["gpu_any"],"buildTools":[],"disk":{"minMB":36,"maxMB":107,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":2812,"downloads":238463,"downloadsWeek":90408,"downloadsDays":1,"growth":2812,"growthDays":3,"history":[{"date":"2026-11-09","stars":2271},{"date":"2026-11-10","stars":2812}],"firstPublishedAt":"2026-11-08T23:01:18.951Z","commitSha":"a3246088771ac68006cca4739e8e226552fd0827","verifiedAt":"2026-11-08T22:42:40.750Z","lastCommitAt":"2026-11-10T18:49:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your own dumped copy of Zelda: Ocarina of Time for Switch 2 into a native Windows game with 4K, HDR and 60+ fps.","summary":"NX2navi rebuilds The Legend of Zelda: Ocarina of Time (Switch 2 version 1.1.1) as a native Windows program, so it runs as PC code instead of inside an emulator. It adds widescreen and ultrawide, resolutions up to 5K, HDR, DLSS or FSR upscaling, frame generation, remappable keyboard and gamepad controls and an in-game settings menu. It is for people who already own and have dumped the game: the installer finds your .nup dump, downloads build tools and builds the game on your PC once (10 to 25 minutes). Windows 10 or later, 16 GB of RAM, about 30 GB free and a recent Vulkan 1.3 graphics card.","easeRationale":"On Windows there is a ready installer (NX2navi-Setup-0.5.0.exe) and no commands to type, but you must already have your own .nup game dump sitting next to it and wait 10 to 25 minutes for the one-time build, plus a graphics card with Vulkan 1.3 and mesh shaders. No release for Mac or Linux.","costNote":"No price in the repository; the README calls it a non-commercial fan project.","notVerified":["Whether the launcher really finds a dump and finishes the build on a normal PC, and how often it fails","Which build tools the launcher downloads at run time and from which servers","Real disk use after the build (the README says about 30 GB for the game plus 1.5 GB of tools)","Whether the frame rate, 4K, HDR and frame generation claims hold on a given graphics card","Whether the installer behaves the same way on hardware without mesh shader support"]},"zh-Hans":{"summaryShort":"将你自己转储的 Switch 2 版《塞尔达传说:时光之笛》变成原生 Windows 游戏,支持 4K、HDR 和 60 以上帧率。","summary":"NX2navi 把《塞尔达传说:时光之笛》(Switch 2 1.1.1 版)重建为原生 Windows 程序,让它以 PC 代码的方式运行,而不是在模拟器里运行。它加入了宽屏和超宽屏支持、最高 5K 分辨率、HDR、DLSS 或 FSR 超分辨率、帧生成、可重新映射的键盘和手柄按键,以及游戏内设置菜单。它面向已经拥有并自行转储了游戏的人:安装程序会找到你的 .nup 转储文件,下载构建工具,并在你的电脑上一次性构建游戏(10 到 25 分钟)。需要 Windows 10 或更高版本、16 GB 内存、大约 30 GB 可用空间,以及一块支持 Vulkan 1.3 的较新显卡。","easeRationale":"在 Windows 上有现成的安装程序(NX2navi-Setup-0.5.0.exe),不需要输入任何命令,但你必须已经有自己的 .nup 游戏转储文件放在旁边,并等待 10 到 25 分钟完成一次性构建,还需要支持 Vulkan 1.3 和网格着色器(mesh shaders)的显卡。没有 Mac 或 Linux 版本。","costNote":"仓库里没有标价;README 称这是一个非商业的爱好者项目。","notVerified":["安装程序是否真的能在普通电脑上找到转储文件并完成构建,以及失败的频率如何","安装程序在运行时会下载哪些构建工具,来自哪些服务器","构建完成后的实际磁盘占用(README 说游戏约 30 GB,加上 1.5 GB 的工具)","帧率、4K、HDR 和帧生成方面的说法在具体显卡上是否成立","在不支持网格着色器的硬件上,安装程序是否表现一致"]}}},{"id":"R_kgDOSIqPOQ","slug":"tashfeenahmed__freellmapi","name":"FreeLLMAPI","owner":"tashfeenahmed","fullName":"tashfeenahmed/freellmapi","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"provider_free_tier_terms","severity":"attention","evidence":"该项目的存在就是为了把多家提供方的免费额度叠加在一个端点后面（仓库描述：\"7.4 billion tokens per month. 34 free LLM providers... Personal experimentation only.\"）。是否允许这样汇聚免费额度取决于每家提供方自己的条款，而作者自己的措辞把它限定在个人实验范围内。","source":"ai","title":"提供方免费档位的条款"},{"code":"install_script_piped_to_shell","severity":"attention","evidence":"README.md 快速开始：`curl -fsSL https://freellmapi.co/install.sh | bash`。这会从作者的网站下载一个脚本，并在你没有先读过的情况下直接在你的机器上运行；README 把脚本链接作为替代方式提供。","source":"ai","title":"安装脚本被管道传入 shell"},{"code":"downloads_catalog_at_run_time","severity":"info","evidence":"README.md：\"the router syncs a signed catalog from freellmapi.co twice a day\" 以及 \"every download is verified against a pinned Ed25519 key before it is applied\"，也就是说运行中的安装会持续从作者的服务器获取数据。","source":"ai","title":"运行时下载目录"},{"code":"docker_socket_access_if_autoupdate_enabled","severity":"info","evidence":"docker-compose.yml 为可选的 Watchtower 更新器挂载了 `/var/run/docker.sock`，其自带注释写道：\"It needs the Docker socket, which is root-equivalent on the host — leave the profile off if that is not acceptable.\"","source":"ai","title":"启用自动更新时会访问 docker 套接字"},{"code":"must_not_be_exposed_to_the_internet","severity":"info","evidence":"docker-compose.yml：\"Bound to localhost by default — FreeLLMAPI is single-user and must not be exposed to the internet\"；.env.example 补充说，把它开放到局域网会让代理 \"guarded only by the unified API key\"。","source":"ai","title":"不得暴露给互联网"},{"code":"all_provider_keys_in_one_place","severity":"info","evidence":"README.md：提供方密钥会 \"AES-256-GCM encrypted in SQLite and decrypted in-memory per request\"，.env.example 提到如果你不自己设置加密密钥，它会自动生成到数据库旁的 `.encryption-key` 文件中。每一条提示词和每一个密钥都会经过这一个程序。","source":"ai","title":"所有提供方密钥集中在一处"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node","docker"],"disk":{"minMB":104,"maxMB":311,"basis":"installer"},"apiKeys":["Free AI provider keys (Google, Groq, Cerebras, Mistral, OpenRouter and others — at least one)","FreeLLMAPI premium key (fla_, optional, for the same-day catalog)"],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","network_system"],"alternativeTo":[{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"portkey","name":"Portkey","confirmed":true}],"declaredAlternatives":[],"stars":50214,"downloads":579812,"downloadsWeek":127862,"downloadsDays":7,"growth":3175,"growthDays":7,"history":[{"date":"2026-10-29","stars":44036},{"date":"2026-10-30","stars":44685},{"date":"2026-10-31","stars":45200},{"date":"2026-11-01","stars":45786},{"date":"2026-11-02","stars":46405},{"date":"2026-11-03","stars":47039},{"date":"2026-11-04","stars":47506},{"date":"2026-11-05","stars":47963},{"date":"2026-11-06","stars":48383},{"date":"2026-11-07","stars":48802},{"date":"2026-11-08","stars":49226},{"date":"2026-11-09","stars":49679},{"date":"2026-11-10","stars":50214}],"firstPublishedAt":"2026-10-29T01:54:21.478Z","commitSha":"741d03094398facceb1b6019a838f806e4d9f7f9","verifiedAt":"2026-11-03T02:54:54.565Z","lastCommitAt":"2026-11-10T15:51:18Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Pools the free plans of many AI providers into one local address, so your apps and coding tools can use free models.","summary":"FreeLLMAPI runs on your own computer or server and puts the free plans of dozens of AI providers (Google, Groq, Mistral, Cerebras, OpenRouter and more) behind a single OpenAI-style address. You add each provider's key in a dashboard, and it picks a model for every request and switches provider when one is out of quota. There are installers for Windows, macOS and Linux, plus Docker. It is for people who already connect apps or coding tools to an AI endpoint; the owner says it is for personal experimentation only.","easeRationale":"On Windows there is a ready .exe installer with a tray app and, the README says, no password to set up, so the extra step is getting provider keys and pasting them in the dashboard. macOS has .dmg downloads (choose arm64 for Apple Silicon or x64 for Intel) and Linux has AppImage, .deb and .rpm, with the same single extra step; the Docker route is harder but optional.","costNote":"The router is MIT and free. The paid tier ($19/year or $49 once, on freellmapi.co) only gives the same-day model catalog; the free install uses a monthly snapshot that the README says sits about 303 models behind.","notVerified":["Whether the headline numbers (34 providers, 635 free endpoints, 7.4 billion tokens a month) match what the providers actually give you","Whether each provider allows this kind of pooled use, and what happens to your account if it does not","Whether the Windows and macOS installers are signed, since the repository does not say","How well the encryption of stored provider keys holds up in practice","The real speed and answer quality, which the README itself says drops late in the day as daily caps are hit","What the install script does step by step, and what the twice-daily catalog sync sends to the author's server"]},"zh-Hans":{"summaryShort":"把多家 AI 提供方的免费额度汇聚到一个本地地址，让你的应用和编程工具都能用上免费模型。","summary":"FreeLLMAPI 运行在你自己的电脑或服务器上，把几十家 AI 提供方（Google、Groq、Mistral、Cerebras、OpenRouter 等）的免费额度统一放在一个 OpenAI 风格的地址后面。你在控制面板里添加每家提供方的密钥，它会为每个请求挑选一个模型，并在某家额度用完时自动切换提供方。提供 Windows、macOS 和 Linux 的安装程序，还有 Docker 版本。它面向已经把应用或编程工具连接到 AI 接口的人；作者说这只用于个人实验。","easeRationale":"在 Windows 上有一个现成的 .exe 安装程序，带有托盘应用，README 说不需要设置密码，所以额外的步骤只是获取提供方的密钥并粘贴到控制面板里。macOS 提供 .dmg 下载（Apple Silicon 选 arm64，Intel 选 x64），Linux 提供 AppImage、.deb 和 .rpm，同样只有这一个额外步骤；Docker 路线更麻烦，但是可选的。","costNote":"路由器本身是 MIT 许可且免费。付费档（freellmapi.co 上每年 19 美元或一次性 49 美元）只是提供当天更新的模型目录；免费安装使用的是按月快照，README 说它大约落后 303 个模型。","notVerified":["主要数字（34 家提供方、635 个免费端点、每月 74 亿 token）是否和提供方实际给出的一致","每家提供方是否允许这种汇聚使用方式，以及如果不允许你的账户会怎样","Windows 和 macOS 的安装程序是否经过签名，因为仓库里没有说明","存储的提供方密钥的加密在实际使用中能撑住多少","真实的速度和回答质量，README 自己说随着每日限额用尽，当天晚些时候会下降","安装脚本具体做了哪些步骤，以及每天两次的目录同步会向作者的服务器发送什么内容"]}}},{"id":"R_kgDONjF_4A","slug":"tauricresearch__tradingagents","name":"TradingAgents","owner":"TauricResearch","fullName":"TauricResearch/TradingAgents","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"not_financial_advice","severity":"info","evidence":"README：\"TradingAgents framework is designed for research purposes... It is not intended as financial, investment, or trading advice.\"","source":"ai","title":"不构成投资建议"},{"code":"results_change_from_run_to_run","severity":"info","evidence":"README 的 \"Reproducibility\" 部分：\"two runs of the same ticker and date can differ\"，以及 \"Backtest results are not guaranteed to match any published figure... Treat the framework as a research scaffold... not as a strategy with a fixed, replicable return.\"","source":"ai","title":"结果每次运行都会变化"},{"code":"ratings_rest_partly_on_social_media_posts","severity":"info","evidence":"README：Sentiment Analyst \"Aggregates news headlines, StockTwits, and Reddit chatter into a single sentiment read\"；如果没有可选的 TypeSafe 密钥，\"posts pass through unscreened\"，所以任何人发关于某只股票的帖子都可能被带入分析。","source":"ai","title":"评级部分依赖社交媒体帖子"},{"code":"startup_update_check","severity":"info","evidence":"README：\"At start-up the interactive CLI mentions a newer release and notices from Tauric Research. It checks at most once a day... set TRADINGAGENTS_NO_UPDATE_CHECK=1 to turn it off.\"","source":"ai","title":"启动时检查更新"},{"code":"stores_account_sign_in","severity":"info","evidence":"README：ChatGPT 套餐的登录信息 \"is kept in ~/.tradingagents/chatgpt-auth.json, and tradingagents logout chatgpt revokes it.\"","source":"ai","title":"保存账号登录信息"},{"code":"sends_a_default_contact_address_to_the_sec","severity":"info","evidence":"README 的 \"Fundamentals as filed\" 部分：\"SEC asks callers to identify themselves and refuses requests that carry no contact address, so a default one is sent\"，并要求你改用 `SEC_EDGAR_USER_AGENT` 设置你自己的姓名和邮箱。","source":"ai","title":"向 sec 发送默认联系地址"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["AI model provider of your choice (OpenAI, Anthropic, Google, DeepSeek, OpenRouter, Groq and others)","Alpha Vantage (market data)","FRED (US economic data, free key)","TypeSafe Jev (social post screening)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","development"],"alternativeTo":[{"slug":"tradingview","name":"TradingView","confirmed":true},{"slug":"benzinga-pro","name":"Pro","confirmed":true}],"declaredAlternatives":[],"stars":121869,"downloads":3580,"downloadsWeek":969,"downloadsDays":7,"growth":1522,"growthDays":7,"history":[{"date":"2026-10-26","stars":116104},{"date":"2026-10-27","stars":117331},{"date":"2026-10-28","stars":117897},{"date":"2026-10-29","stars":118332},{"date":"2026-10-30","stars":118720},{"date":"2026-10-31","stars":119062},{"date":"2026-11-01","stars":119566},{"date":"2026-11-02","stars":119973},{"date":"2026-11-03","stars":120347},{"date":"2026-11-04","stars":120614},{"date":"2026-11-05","stars":120802},{"date":"2026-11-06","stars":120992},{"date":"2026-11-07","stars":121167},{"date":"2026-11-08","stars":121364},{"date":"2026-11-09","stars":121571},{"date":"2026-11-10","stars":121869}],"firstPublishedAt":"2026-10-27T01:10:12.698Z","commitSha":"c4544913a14ed3df5f1ed7b9e8483270ce20682a","verifiedAt":"2026-10-31T01:45:32.278Z","lastCommitAt":"2026-11-04T10:13:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ask a team of AI agents to study a stock or crypto coin and write a report with a buy, hold or sell opinion.","summary":"TradingAgents is a research tool where several AI agents play the roles of a trading firm (analysts, researchers, a trader, risk managers) and debate a stock or coin using prices, news, company filings and social posts. It ends with a written report and a rating, and trades only go to a simulated exchange. It can also score past decisions and run backtests over a grid of dates. The README says it is for research and is not financial advice. You need Python 3.11+, a few commands, and an AI model: a paid API key, a ChatGPT plan sign-in or a local Ollama model.","easeRationale":"C on Windows, Mac and Linux alike: none of the 15 releases carries an installer, so on each one you install Python 3.11 or newer, download the code, run `pip install .`, put an AI key in a `.env` file and start the `tradingagents` command in a terminal. Docker is offered as an alternative route.","costNote":"The software is free (Apache-2.0). Each analysis makes many calls to an AI model, which usually means a paid API key or a ChatGPT Plus/Pro plan; a local Ollama model avoids that cost. Docker is an optional way to run it.","notVerified":["How much one analysis costs in AI API fees, since many agents call the model several times per run.","Whether the install steps work on Windows as written; the README shows Linux/macOS commands such as \"source .venv/bin/activate\" and \"export\".","How good or repeatable the ratings are; the README itself says two runs of the same stock can differ.","Whether free data sources such as Yahoo Finance keep working; the README mentions Yahoo's news feed being down.","What the memory log and the per-ticker checkpoint databases under ~/.tradingagents keep about your own portfolio once you pass one in."]},"zh-Hans":{"summaryShort":"让一队 AI 代理研究一只股票或一种加密货币，写出一份带买入、持有或卖出意见的报告。","summary":"TradingAgents 是一个研究工具，多个 AI 代理扮演一家交易公司里的各种角色（分析师、研究员、交易员、风险经理），用价格、新闻、公司文件和社交媒体帖子来讨论一只股票或一种代币。最后输出一份书面报告和一个评级，交易只会发给模拟交易所。它还可以给过去的决策打分，并在一组日期上运行回测。README 说这是用于研究的，不构成投资建议。你需要 Python 3.11 以上版本、一些命令，以及一个 AI 模型：一个付费 API 密钥、登录一个 ChatGPT 套餐，或者一个本地的 Ollama 模型。","easeRationale":"Windows、Mac 和 Linux 上都是 C 级：15 个发行版里没有一个带安装程序，所以在每个系统上你都要安装 Python 3.11 或更新版本，下载代码，运行 `pip install .`，把 AI 密钥放进 `.env` 文件，然后在终端里启动 `tradingagents` 命令。Docker 是另一种可选方式。","costNote":"软件本身是免费的（Apache-2.0）。每次分析都会多次调用 AI 模型，通常意味着需要付费 API 密钥或 ChatGPT Plus/Pro 套餐；使用本地 Ollama 模型可以省下这笔钱。Docker 是运行它的一种可选方式。","notVerified":["一次分析在 AI API 费用上要花多少钱，因为很多代理每次运行都会多次调用模型。","安装步骤是否能在 Windows 上照着文档原样运行；README 里给的是 Linux/macOS 的命令，比如 \"source .venv/bin/activate\" 和 \"export\"。","评级有多靠谱、能否重现；README 自己也说同一只股票跑两次结果可能不同。","像 Yahoo Finance 这样的免费数据源能否持续可用；README 提到过 Yahoo 的新闻源出现过中断。","一旦你传入自己的投资组合，~/.tradingagents 下的记忆日志和各个 ticker 的检查点数据库到底保存了关于它的什么内容。"]}}},{"id":"R_kgDOL-fjTg","slug":"tencent-ailab__v-express","name":"V-Express","owner":"tencent-ailab","fullName":"tencent-ailab/V-Express","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 目录和 winget 清单中都没有发现任何许可证","source":"check","title":"没有许可证"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-5pcm-hx3q-hm94; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74","source":"check","title":"存在已知漏洞"},{"code":"model_weights_non_commercial_only","severity":"attention","evidence":"README：\"both manual-downloading and auto-downloading models from V-Express are for non-commercial research purposes. Our released checkpoints are also for research purposes only.\" 文中称代码本身可商用，但仓库没有许可证文件，条款就只有这一段话。","source":"ai","title":"模型权重仅限非商业用途"},{"code":"makes_video_of_a_real_persons_face","severity":"attention","evidence":"该工具把一张真人照片加上任意音频变成说话视频（\"Scenario 2 (A's picture and any talking audio)\"），可以让某个人看起来说了他从未说过的话。README 把责任推给使用者：\"they are obligated to comply with local laws and utilize it responsibly. The developers will not assume any responsibility for potential misuse by users.\"","source":"ai","title":"会生成真人面孔的视频"},{"code":"pinned_old_dependencies","severity":"info","evidence":"requirements.txt 锁定了 torch==2.0.1、diffusers==0.29.2、transformers==4.41.1 和 xformers==0.0.22，都是 2023 至 2024 年的版本，在较新的 Python 或较新的显卡驱动上可能装不上。","source":"ai","title":"锁定了过时的依赖"}],"platforms":["windows","linux"],"tools":["python","gpu_any","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":1050,"maxMB":4750,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["video"],"alternativeTo":[{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"synthesia","name":"Synthesia","confirmed":true}],"declaredAlternatives":[],"stars":2354,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":0,"growthDays":6,"history":[{"date":"2026-11-05","stars":2354},{"date":"2026-11-06","stars":2354},{"date":"2026-11-07","stars":2354},{"date":"2026-11-08","stars":2354},{"date":"2026-11-09","stars":2354},{"date":"2026-11-10","stars":2354}],"firstPublishedAt":"2026-11-10T23:52:44.232Z","commitSha":"6814f7db362dc9e26644162034b4d28c5277651a","verifiedAt":"2026-11-10T23:33:05.842Z","lastCommitAt":"2025-01-24T14:40:41Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Make a photo of a face talk: you give one picture and an audio file, and it builds a video with matching mouth movement.","summary":"V-Express is research code from Tencent AI Lab that animates a portrait. You give it a square photo of a face (512x512 or larger), an audio clip and, if you want, a reference video that supplies the head movement; it writes out a video where the face speaks the audio. Everything runs from the command line: install Python packages, download several AI models from Hugging Face, then run inference.py with the file paths. A graphics card is needed, and the authors say results are better with English audio.","easeRationale":"On Linux and Windows the only documented path is the terminal: pip install -r requirements.txt, git lfs to download the model files, move folders by hand, then run inference.py with several arguments. There is no interface and no installer, and a graphics card is needed (the README reports 7956 MiB of video memory for a 31-second clip).","notVerified":["How many gigabytes the Hugging Face model files really take after the download","How long one video takes on a normal home graphics card (the README's only number is 2617 seconds for 31 seconds of audio on a data-centre V100)","Whether the installation works on Windows, since every command in the README is written for a Linux shell","How good the lip movement is with languages other than English, which the authors say has not been tested in detail"]},"zh-Hans":{"summaryShort":"让照片里的人脸开口说话：提供一张图片和一段音频，就能生成口型对上的视频。","summary":"V-Express 是腾讯 AI Lab 的研究代码，用来让一张人像动起来。你提供一张正方形的人脸照片（512x512 或更大）、一段音频，如果愿意还可以再给一段提供头部动作的参考视频；程序会输出一段人脸说出这段音频的视频。全部流程都在命令行里完成：安装 Python 包、从 Hugging Face 下载若干 AI 模型，然后带上文件路径运行 inference.py。需要显卡，作者还说英文音频的效果更好。","easeRationale":"在 Linux 和 Windows 上，文档给出的唯一方式就是终端：pip install -r requirements.txt，用 git lfs 下载模型文件，手动移动目录，再带上一串参数运行 inference.py。没有界面，也没有安装程序，而且必须有显卡（README 提到 31 秒的片段用了 7956 MiB 显存）。","notVerified":["从 Hugging Face 下载完模型文件后，实际占用多少 GB","在普通家用显卡上生成一段视频要多久（README 里唯一的数字是数据中心 V100 上 31 秒音频耗时 2617 秒）","在 Windows 上能否顺利安装，因为 README 里的命令都是按 Linux shell 写的","英语以外的语言口型效果如何，作者表示这一点没有做过细致测试"]}}},{"id":"R_kgDONV-LBQ","slug":"tencent-hunyuan__hunyuanvideo","name":"HunyuanVideo","owner":"Tencent-Hunyuan","fullName":"Tencent-Hunyuan/HunyuanVideo","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：torch: GHSA-63cw-57p8-fm3p, GHSA-887c-mr87-cxwp, GHSA-qfhq-4f3w-5fph; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; tqdm: GHSA-g7vv-2v7x-gj9p, PYSEC-2026-1976; gradio: GHSA-39mp-8hj3-5c49, GHSA-5cpq-9538-jm2j, GHSA-6655-8ph2-63j3","source":"check","title":"存在已知漏洞"},{"code":"license_excludes_eu_uk_south_korea","severity":"attention","evidence":"LICENSE.txt 第一行：\"THIS LICENSE AGREEMENT DOES NOT APPLY IN THE EUROPEAN UNION, UNITED KINGDOM AND SOUTH KOREA AND IS EXPRESSLY LIMITED TO THE TERRITORY\"，第 5(c) 条又写明，在该地域之外使用或展示这些作品及其产出 \"is unlicensed and unauthorized\"。","source":"ai","title":"许可证不适用于欧盟、英国和韩国"},{"code":"license_extra_commercial_conditions","severity":"info","evidence":"Tencent Hunyuan Community License 并非标准开源许可：第 4 条要求月活跃用户超过一亿时须另行向腾讯取得授权，第 5(b) 条则禁止用其输出去改进任何其他 AI 模型。","source":"ai","title":"许可证附带额外的商业条件"},{"code":"needs_datacenter_class_gpu","severity":"info","evidence":"README 的配置要求表：720px1280px129f 需 60 GB 显存峰值，544px960px129f 需 45 GB；\"The model is tested on a single 80G GPU\"。文中称使用 FP8 权重可节省约 10 GB。","source":"ai","title":"需要数据中心级 GPU"},{"code":"docker_command_runs_privileged","severity":"info","evidence":"README 里的 Docker 命令使用了 --privileged 和 --security-opt=seccomp=unconfined，这会让容器获得对宿主机相当广泛的访问权限。","source":"ai","title":"Docker 命令以特权模式运行"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":1010,"maxMB":4570,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"kling","name":"Kling","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"ltx-studio","name":"LTX Studio","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":12754,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":30,"growthDays":6,"history":[{"date":"2026-11-04","stars":12724},{"date":"2026-11-05","stars":12733},{"date":"2026-11-06","stars":12736},{"date":"2026-11-07","stars":12738},{"date":"2026-11-08","stars":12742},{"date":"2026-11-09","stars":12749},{"date":"2026-11-10","stars":12754}],"firstPublishedAt":"2026-11-10T23:52:44.250Z","commitSha":"4eda9cb44c3be82a6b9c4261897cb6bd3cfb084e","verifiedAt":"2026-11-10T23:33:49.151Z","lastCommitAt":"2026-10-08T08:57:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Write a sentence and get a short video out of it, using Tencent's 13-billion-parameter video model on your own machine.","summary":"HunyuanVideo is Tencent's open-weights text-to-video model with the code to run it. You type a description and it produces a video of about five seconds, up to 720p. It runs from the command line or from a simple web page it starts locally (Gradio). The hardware bar is high: the authors list 45 GB of video memory for 544x960 and 60 GB for 720p, test only on Linux with an NVIDIA card, and the model files are downloaded separately from Hugging Face.","easeRationale":"Linux is the only system the authors test: you create a conda environment, install a specific PyTorch build for your CUDA version, install the requirements, optionally build flash-attention, download the model files and then run a Python script (a Docker image is offered as an alternative). An NVIDIA card is required and the stated peak memory is 45-60 GB.","notVerified":["How many gigabytes the model files take once downloaded from Hugging Face, since the download instructions live in a separate file","Whether it can be made to run on a normal home graphics card with FP8 weights and CPU offload, and how slow that is","Whether the install works on Windows or macOS, which the authors do not test","How long one 5-second video takes on a single GPU outside the authors' multi-GPU measurements"]},"zh-Hans":{"summaryShort":"写一句话就能生成一小段视频，在自己的机器上运行腾讯 130 亿参数的视频模型。","summary":"HunyuanVideo 是腾讯开放权重的文生视频模型，并附带运行代码。你输入一段描述，它就生成约五秒、最高 720p 的视频。可以在命令行里运行，也可以用它在本地启动的简易网页界面（Gradio）。硬件门槛很高：作者给出 544x960 需 45 GB 显存、720p 需 60 GB，只在装有 NVIDIA 显卡的 Linux 上测试，模型文件还要另外从 Hugging Face 下载。","easeRationale":"作者只在 Linux 上做过测试：你需要建 conda 环境，按自己的 CUDA 版本安装对应的 PyTorch，安装依赖，可选地编译 flash-attention，下载模型文件，然后运行一个 Python 脚本（另有 Docker 镜像可选）。必须有 NVIDIA 显卡，文中给出的显存峰值是 45 到 60 GB。","notVerified":["从 Hugging Face 下载完模型文件后共占多少 GB，因为下载说明放在另一个文件里","用 FP8 权重和 CPU 卸载，能否在普通家用显卡上跑起来，以及会慢到什么程度","在 Windows 或 macOS 上能否安装成功，作者并未测试这两个系统","在单张 GPU 上生成 5 秒视频要多久，作者只给了多卡的测量结果"]}}},{"id":"R_kgDOQZWsmQ","slug":"tencent-hunyuan__hunyuanvideo-1.5","name":"HunyuanVideo-1.5","owner":"Tencent-Hunyuan","fullName":"Tencent-Hunyuan/HunyuanVideo-1.5","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h; diffusers: GHSA-7wx4-6vff-v64p, GHSA-98h9-4798-4q5v, PYSEC-2026-2446; transformers: GHSA-29pf-2h5f-8g72, GHSA-69w3-r845-3855, GHSA-fgcw-684q-jj6r","source":"check","title":"存在已知漏洞"},{"code":"license_excludes_eu_uk_south_korea","severity":"attention","evidence":"LICENSE 第一行：\"THIS LICENSE AGREEMENT DOES NOT APPLY IN THE EUROPEAN UNION, UNITED KINGDOM AND SOUTH KOREA AND IS EXPRESSLY LIMITED TO THE TERRITORY\"；第 2 条的授权也只 \"for the Territory only\"，把这些地区排除在外。","source":"ai","title":"许可证不适用于欧盟、英国和韩国"},{"code":"install_step_builds_code_from_a_personal_fork","severity":"attention","evidence":"README 里 SageAttention 这一步不是从原项目安装，而是从某个个人的分支安装：\"git clone https://github.com/cooper1637/SageAttention.git ... python3 setup.py install\"。","source":"ai","title":"安装步骤会编译个人分支的代码"},{"code":"install_uses_tencent_package_mirror","severity":"info","evidence":"安装第 2 步执行 \"pip install -i https://mirrors.tencent.com/pypi/simple/ --upgrade tencentcloud-sdk-python\"，从 PyPI 之外的软件源拉取腾讯云 SDK，却没有说明它的用途。","source":"ai","title":"安装过程使用腾讯的软件源"},{"code":"license_extra_commercial_conditions","severity":"info","evidence":"Tencent Hunyuan Community License 并非标准开源许可：第 4 条要求月活跃用户超过一亿时须另行向腾讯取得授权，第 3(e) 条要求任何基于它提供服务的人都必须显著声明腾讯并不为其背书。","source":"ai","title":"许可证附带额外的商业条件"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":950,"maxMB":4450,"basis":"heuristic"},"apiKeys":["Prompt-rewriting model (Gemini or a self-hosted vLLM server)"],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"kling","name":"Kling","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"ltx-studio","name":"LTX Studio","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":4598,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":7,"growthDays":6,"history":[{"date":"2026-11-05","stars":4591},{"date":"2026-11-06","stars":4591},{"date":"2026-11-07","stars":4592},{"date":"2026-11-08","stars":4593},{"date":"2026-11-09","stars":4596},{"date":"2026-11-10","stars":4598}],"firstPublishedAt":"2026-11-10T23:52:44.261Z","commitSha":"60783e704160023913bee78f0b47036d393d4dfa","verifiedAt":"2026-11-10T23:34:22.004Z","lastCommitAt":"2026-04-10T06:14:58Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Make short videos from a sentence or from one picture, with Tencent's smaller video model that fits a 14 GB graphics card.","summary":"HunyuanVideo-1.5 is the lighter version of Tencent's video model: 8.3 billion parameters instead of 13. You give it a text description, or a photo plus a description, and it writes out a video at 480p or 720p, with an optional step that upscales to 1080p. It runs from the command line on Linux with an NVIDIA card; the authors state a minimum of 14 GB of video memory with offloading on. Model files are downloaded separately, and the project also documents use through ComfyUI and Diffusers.","easeRationale":"Linux is the only system listed: you clone the repository, pip install the requirements, optionally build extra attention libraries from source, download the chosen checkpoints from Hugging Face, and run generate.py with a long list of arguments. An NVIDIA card with at least 14 GB of video memory is required.","notVerified":["How many gigabytes the checkpoints take, since you pick among a dozen model variants listed in the README","Whether the stated 14 GB minimum really holds in practice, and how long one video takes on a single consumer card","Whether the install works on Windows or macOS, which the authors do not list","How much quality is lost without the prompt-rewriting step, which is on by default and needs a separate large model"]},"zh-Hans":{"summaryShort":"用一句话或一张图生成短视频，腾讯这款更小的视频模型在 14 GB 显存的显卡上就能跑。","summary":"HunyuanVideo-1.5 是腾讯视频模型的轻量版：参数量 83 亿，而不是 130 亿。你给它一段文字描述，或者一张照片加一段描述，它就输出 480p 或 720p 的视频，还有一个可选步骤能放大到 1080p。它在装有 NVIDIA 显卡的 Linux 上以命令行运行；作者称开启卸载后最低需要 14 GB 显存。模型文件要另行下载，项目还提供了通过 ComfyUI 和 Diffusers 使用的说明。","easeRationale":"文档只列出了 Linux：你要克隆仓库，用 pip 安装依赖，可选地从源码编译额外的注意力库，从 Hugging Face 下载选定的权重，然后带上一长串参数运行 generate.py。必须有至少 14 GB 显存的 NVIDIA 显卡。","notVerified":["这些权重文件共占多少 GB，因为 README 里列了十来个模型变体供你挑选","文中说的 14 GB 下限在实际使用中是否成立，以及在单张消费级显卡上生成一段视频要多久","在 Windows 或 macOS 上能否安装成功，作者并未列出这两个系统","关闭默认开启的提示词改写步骤后画质会损失多少，而该步骤还需要另一个大模型"]}}},{"id":"R_kgDOMNVlng","slug":"tencent__mimicmotion","name":"MimicMotion","owner":"Tencent","fullName":"Tencent/MimicMotion","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"can_animate_a_real_person","severity":"attention","evidence":"该工具用一张照片加一段动作视频实现 \"High-Quality Human Motion Video Generation\"，因此可以让一个真实的人做出他从未做过的动作。仓库里既没有水印机制，也没有征得同意的环节。","source":"ai","title":"可以让真人动起来"},{"code":"depends_on_third_party_model_terms","severity":"attention","evidence":"LICENSE：\"MimicMotion is licensed under the Apache License Version 2.0 except for the third-party components listed below.\" README 写道 \"The SVD model stabilityai/stable-video-diffusion-img2vid-xt-1-1 will be automatically downloaded\"，而这个 Stability AI 的模型自带使用条款。","source":"ai","title":"受第三方模型条款约束"},{"code":"no_updates_for_about_a_year","severity":"info","evidence":"最后一次提交是 2025-11-18，没有任何发布版本；News 一栏最新的条目是 2025-05-03 被 ICML 接收，而目前仍有 93 个未关闭的问题。锁定的依赖也很旧（pytorch 2.0.1、diffusers 0.27.0、transformers 4.32.1）。","source":"ai","title":"约一年没有更新"},{"code":"heavy_hardware","severity":"info","evidence":"README.md：生成 35 秒的视频时，\"the 72-frame model requires 16GB VRAM (4060ti) and finishes in 20 minutes on a 4090 GPU\"，并且 \"The minimum VRAM requirement for the 16-frame U-Net model is 8GB; however, the VAE decoder demands 16GB\"。","source":"ai","title":"硬件要求很高"},{"code":"weights_downloaded_from_outside","severity":"info","evidence":"配置过程需要手动从 Hugging Face 取文件（yolox_l.onnx、dw-ll_ucoco_384.onnx、MimicMotion_1-1.pth），并建议使用镜像 \"export HF_ENDPOINT=https://hf-mirror.com\"。这些模型都不在仓库里。","source":"ai","title":"权重要从站外下载"}],"platforms":["windows","linux"],"tools":["python","gpu_nvidia"],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":150,"maxMB":450,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"deepmotion","name":"DeepMotion","confirmed":true},{"slug":"viggle","name":"Viggle","confirmed":true}],"declaredAlternatives":[],"stars":2690,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":2,"growthDays":6,"history":[{"date":"2026-11-07","stars":2687},{"date":"2026-11-08","stars":2690},{"date":"2026-11-09","stars":2690},{"date":"2026-11-10","stars":2690}],"firstPublishedAt":"2026-11-10T23:52:44.273Z","commitSha":"6907bdcc259a6a048d41a365e840d22274f9256c","verifiedAt":"2026-11-10T23:39:13.045Z","lastCommitAt":"2025-11-18T06:29:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Takes a photo of a person and a dance video, and creates a new video of that person doing the same moves.","summary":"MimicMotion is the research code from Tencent for a model that animates a person. You give it one photo and a video of someone moving, and it produces a video of the person in the photo repeating those movements, up to 72 frames at 576x1024 per run. There is no interface: you build a conda environment, download several model files by hand and run inference.py with a configuration file. An NVIDIA card with 16 GB of memory is needed for the full quality.","easeRationale":"On Linux and Windows the path is the same and it is all terminal: install conda, create the environment from environment.yaml with CUDA PyTorch, download the DWPose and MimicMotion weights with wget, then run \"python inference.py --inference_config configs/test.yaml\". It also needs an NVIDIA card, 16 GB of video memory for the 72-frame model.","notVerified":["The quality of the result with your own photo and motion video","The total disk space after all the model files are downloaded","Whether the old pinned versions still install on a current graphics driver","Whether it runs on a card with less memory using the CPU for the decoder","What the third-party component list in the truncated part of the LICENSE allows"]},"zh-Hans":{"summaryShort":"给一张人物照片和一段舞蹈视频，生成这个人做同样动作的新视频。","summary":"MimicMotion 是腾讯的研究代码，对应一个让人物动起来的模型。你提供一张照片和一段有人在动的视频，它就生成照片中的人重复这些动作的视频，单次最多 72 帧、分辨率 576x1024。没有界面：你需要搭 conda 环境，手动下载若干模型文件，再带上配置文件运行 inference.py。要达到完整画质需要一块 16 GB 显存的 NVIDIA 显卡。","easeRationale":"Linux 和 Windows 的流程一样，全程都在终端里：安装 conda，用 environment.yaml 创建带 CUDA 版 PyTorch 的环境，用 wget 下载 DWPose 和 MimicMotion 的权重，然后运行 \"python inference.py --inference_config configs/test.yaml\"。同样需要 NVIDIA 显卡，72 帧模型要 16 GB 显存。","notVerified":["换成你自己的照片和动作视频后，效果如何","下载完全部模型文件后共占用多少磁盘空间","锁定的这些旧版本依赖，在当前显卡驱动上是否还能装上","用 CPU 跑解码器时，显存更小的显卡能否运行","LICENSE 被截断部分里列出的第三方组件允许哪些用途"]}}},{"id":"R_kgDOPQrONg","slug":"tencent__weknora","name":"WeKnora","owner":"Tencent","fullName":"Tencent/WeKnora","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"license_text_disagrees_with_detection","severity":"info","evidence":"LICENSE 开头写着 \"WeKnora is licensed under the MIT except for the third-party components listed below\"，README 也说是 MIT，但同一个文件里还嵌入了某个内置组件完整的 Apache-2.0 文本，而自动检查读到的正是后者。","source":"ai","title":"许可证文本与自动识别结果不一致"},{"code":"owner_warns_not_to_expose_publicly","severity":"attention","evidence":"README 的警示段落：\"deploy it in an internal / private network rather than on the public internet; avoid exposing the service directly to public networks, to prevent information leakage\"。","source":"ai","title":"项目方提醒不要暴露到公网"},{"code":"optional_docker_socket_equals_host_root","severity":"attention","evidence":"docker-compose.yml 中那行被注释掉的 /var/run/docker.sock 挂载旁的说明：\"Docker 沙箱默认关闭... 这等同宿主机 root\"。","source":"ai","title":"可选的 Docker socket 等同于宿主机 root"},{"code":"runs_third_party_skills_from_the_internet","severity":"attention","evidence":"README：\"Skills installed from ClawHub / SkillHub / Git / ZIP run in session-persistent Docker / E2B / Cube sandboxes\"，而且智能体还能通过 BrowserSkill 扩展操控你自己的 Chrome 或 Edge。","source":"ai","title":"会运行来自网上的第三方技能"},{"code":"example_file_ships_default_passwords","severity":"info","evidence":".env.example 里提供了可直接使用的默认值，例如 DB_PASSWORD=postgres123!@#、REDIS_PASSWORD=redis123!@#，以及一个标注为 \"rotate-before-exposing-publicly\" 的 SearXNG 默认密钥。","source":"ai","title":"示例配置里带着默认密码"}],"platforms":["windows","mac","linux"],"tools":["docker"],"buildTools":["docker","go"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["LLM provider (OpenAI, DeepSeek, Qwen, Gemini and 25 others)","Web search provider (Tavily, Bing, Google and others)"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["ai_agents","productivity","local_ai"],"alternativeTo":[{"slug":"notebooklm","name":"NotebookLM","confirmed":true},{"slug":"chatbase","name":"Chatbase","confirmed":true},{"slug":"notion-ai","name":"Notion AI","confirmed":true}],"declaredAlternatives":[],"stars":37106,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":721,"growthDays":7,"history":[{"date":"2026-11-05","stars":36584},{"date":"2026-11-06","stars":36682},{"date":"2026-11-07","stars":36744},{"date":"2026-11-08","stars":36828},{"date":"2026-11-09","stars":36995},{"date":"2026-11-10","stars":37106}],"firstPublishedAt":"2026-11-10T23:52:44.292Z","commitSha":"cbdcebabf48c5ee2d6868b370326aa8341703f62","verifiedAt":"2026-11-10T23:32:16.031Z","lastCommitAt":"2026-11-09T15:19:49Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turn your team's documents into a searchable knowledge base you can ask questions to, with sources shown in the answer.","summary":"WeKnora, from Tencent, is a knowledge platform you run on your own machine or server. You upload documents (PDF, Word, Excel, images and more), and it answers questions about them with links to the passages it used. It can also act as an agent that does multi-step tasks and build an automatic wiki from your files. You start it with Docker Compose and use it in the browser. The AI models come from outside: an online provider with a key, or Ollama running on your own computer.","easeRationale":"On all three systems the published way to install is the same: clone the repository, copy and edit the .env file, then run docker compose pull and up, which starts several services (app, web interface, PostgreSQL, Redis and a document reader). There is no ready installer; the README says the desktop app has no installer yet and must be built from source.","costNote":"The software is free. The AI models are not included: you either pay an online provider per use or run a local model with Ollama.","notVerified":["How much memory and disk the full Docker stack really needs once PostgreSQL, Redis and the document reader are running","Whether answers are correct for your own documents, and how good the automatic wiki and knowledge graph are","The real monthly cost if you point it at a paid model provider instead of Ollama","Whether the Lite single binary and the unpublished desktop app work as described","How much of the documentation site is usable in English, since the full docs are described as being in Chinese"]},"zh-Hans":{"summaryShort":"把团队的文档变成可提问的知识库，回答里会标明依据的出处。","summary":"WeKnora 是腾讯推出的知识平台，运行在你自己的电脑或服务器上。你上传文档（PDF、Word、Excel、图片等），它就能回答相关问题，并给出所引用段落的链接。它还能作为智能体执行多步骤任务，并根据你的文件自动生成 wiki。用 Docker Compose 启动，在浏览器里使用。AI 模型需要外部提供：要么用带密钥的在线服务，要么在自己电脑上跑 Ollama。","easeRationale":"三个系统上公布的安装方式相同：克隆仓库，复制并修改 .env 文件，然后执行 docker compose pull 和 up，启动多个服务（应用、网页界面、PostgreSQL、Redis 和文档解析器）。没有现成的安装程序；README 说桌面版目前还没有安装包，必须自行从源码构建。","costNote":"软件本身免费。AI 模型不包含在内：你要么按用量付费给在线服务，要么用 Ollama 跑本地模型。","notVerified":["当 PostgreSQL、Redis 和文档解析器都在运行时，整套 Docker 组件实际需要多少内存和磁盘","针对你自己的文档，回答是否准确，自动生成的 wiki 和知识图谱质量如何","如果接入付费模型服务而不是 Ollama，每月实际要花多少钱","单文件的 Lite 版本和尚未发布的桌面应用是否如文档所述那样可用","文档站有多少内容有英文版本可用，因为完整文档据称是中文的"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0NDgwNDIxNg==","slug":"termux__termux-app","name":"Termux","owner":"termux","fullName":"termux/termux-app","type":"ready_app","status":"active","grade":"A","easeByPlatform":{},"risk":"attention","signals":[{"code":"github_builds_signed_with_public_test_key","severity":"attention","evidence":"README 的警示：GitHub 上的 APK \"are signed with a test key that has been shared with community... Everyone is able to use it to forge a malicious Termux update installable over the GitHub build. Think twice about installing Termux builds distributed via Telegram or other social media.\"","source":"ai","title":"GitHub 上的构建用公开测试密钥签名"},{"code":"sources_cannot_be_mixed","severity":"info","evidence":"README：主程序和各个插件共用同一个 sharedUserId，必须全部来自同一来源；要换来源就得先把所有组件卸载干净，否则安卓会以 INSTALL_FAILED_UPDATE_INCOMPATIBLE 等错误拒绝安装。","source":"ai","title":"不同来源的安装包不能混用"},{"code":"unstable_on_recent_android","severity":"attention","evidence":"README 的提示：\"Termux may be unstable on Android 12+. Android OS will kill any (phantom) processes greater than 32... You may get [Process completed (signal 9) - press Enter] message in the terminal\"。","source":"ai","title":"在较新的安卓版本上不稳定"},{"code":"maintainers_wanted","severity":"info","evidence":"README 用粗体写道：\"We are looking for Termux Android application maintainers.\" 列出的最新版本是 2025 年 5 月的 v0.118.3，而 Google Play 版放在另一个仓库里，被描述为实验性且功能不全。","source":"ai","title":"项目正在招募维护者"},{"code":"license_is_gpl3_with_exceptions","severity":"info","evidence":"LICENSE.md：\"The termux/termux-app repository is released under GPLv3 only license\"，其中部分终端代码采用 Apache 2.0；GitHub 没有为该仓库识别出单一的标准许可证。","source":"ai","title":"许可证是带例外的 GPL3"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":64559,"downloads":11185007,"downloadsWeek":106309,"downloadsDays":6,"growth":520,"growthDays":7,"history":[{"date":"2026-11-04","stars":64098},{"date":"2026-11-05","stars":64171},{"date":"2026-11-06","stars":64257},{"date":"2026-11-07","stars":64345},{"date":"2026-11-08","stars":64421},{"date":"2026-11-09","stars":64497},{"date":"2026-11-10","stars":64559}],"firstPublishedAt":"2026-11-10T23:52:44.305Z","commitSha":"8629e632fcb95da272221be327db653fb24befe9","verifiedAt":"2026-11-10T23:35:46.499Z","lastCommitAt":"2026-09-26T05:11:43Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Linux terminal on your Android phone: install packages, run scripts and connect to other machines over SSH.","summary":"Termux is an Android app that gives your phone a terminal and a small Linux environment. After installing it you can add packages with its own package manager (Python, git, ffmpeg, SSH and many more) and run them on the phone, no root needed. There are optional companion apps for buttons, widgets, boot scripts and phone functions. It is Android only: nothing runs on Windows, Mac or Linux desktops. The install is about 180 MB with the bundled base packages.","easeRationale":"Android only, so none of the desktop platforms gets a grade. On Android you download a ready APK from F-Droid or from the GitHub releases page and open it; the terminal works right away and extra packages are installed from inside the app.","notVerified":["Whether it runs well on your own phone and Android version, given the warning about Android 12 and later","How much space it really takes after you install extra packages, beyond the stated ~180 MB base","Whether the Google Play version has the features you need, since it is a separate, reduced build","Whether the package repositories are reachable in your country, which the project itself links troubleshooting for"]},"zh-Hans":{"summaryShort":"在安卓手机上的 Linux 终端：安装软件包、运行脚本，还能用 SSH 连接其他机器。","summary":"Termux 是一款安卓应用，为手机提供一个终端和一个小型 Linux 环境。装好之后，你可以用它自带的包管理器安装 Python、git、ffmpeg、SSH 等众多软件包，并在手机上直接运行，不需要 root。它还有可选的配套应用，用于附加按键、桌面小组件、开机脚本和手机功能。它只支持安卓：Windows、Mac 和 Linux 桌面都无法运行。连同自带的基础包，安装后约占 180 MB。","easeRationale":"仅限安卓，因此桌面平台都不给等级。在安卓上，你从 F-Droid 或 GitHub 发布页下载现成的 APK 打开即可；终端马上就能用，额外的软件包在应用内部安装。","notVerified":["考虑到关于 Android 12 及更高版本的提醒，它在你自己的手机和系统版本上是否运行良好","在你装了额外软件包之后实际占用多少空间，官方只给出约 180 MB 的基础体积","Google Play 版本是否具备你需要的功能，因为那是一个单独的精简构建","在你所在的国家能否访问软件源，项目自己也为此给出了排错指引"]}}},{"id":"R_kgDOOVj18A","slug":"test-time-training__ttt-video-dit","name":"TTT Video DiT","owner":"test-time-training","fullName":"test-time-training/ttt-video-dit","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"low","signals":[{"code":"final_model_weights_not_released","severity":"info","evidence":"README 在下载说明下方注明：\"These are the pretrained model weights, not the final model weights used to generate our demonstration videos.\" 文中只给出了 CogVideoX-5b 基础权重的链接。","source":"ai","title":"最终模型权重没有公开"},{"code":"needs_datacenter_gpu_and_a_compiled_kernel","severity":"info","evidence":"README：\"You must have cuda toolkit (12.3+) and gcc11+ installed to build the TTT-MLP kernel. We only support training on H100s for TTT-MLP.\"","source":"ai","title":"需要数据中心级 GPU 并自行编译内核"},{"code":"builds_on_a_model_with_its_own_terms","severity":"info","evidence":"代码采用 MIT 许可，但它运行在从 Hugging Face 取得的 THUDM CogVideoX-5b 权重之上，而那些权重另有自己的许可条款，本仓库并未复述。","source":"ai","title":"依赖一个带有自身条款的模型"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":1050,"maxMB":4750,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","development"],"alternativeTo":[{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":2453,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":0,"growthDays":6,"history":[{"date":"2026-11-07","stars":2454},{"date":"2026-11-08","stars":2453},{"date":"2026-11-09","stars":2453},{"date":"2026-11-10","stars":2453}],"firstPublishedAt":"2026-11-10T23:52:44.326Z","commitSha":"e43707e9b8594ace5a73c4a4438f486d5d17729c","verifiedAt":"2026-11-10T23:36:19.638Z","lastCommitAt":"2026-02-25T04:11:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that teaches a video AI model to make one-minute clips instead of three-second ones.","summary":"This is the official code from the paper \"One-Minute Video Generation with Test-Time Training\". It adds new layers to an existing text-to-video model (CogVideoX 5B) and trains it in stages so the videos get longer, up to 63 seconds, keeping the same style from start to end. It is a research project for people with machine-learning experience: you build a CUDA kernel from source, fetch the base model yourself and train on H100 cards. There is no app, no interface and no finished model to download.","easeRationale":"Developers only, on Linux: you create a conda environment, pull a git submodule and compile the TTT-MLP kernel, which needs the CUDA toolkit 12.3+ and gcc 11+, and the authors say training is only supported on H100 cards. The base model weights have to be collected by hand from two other projects.","notVerified":["Whether anyone can reproduce the one-minute demo videos, since the finished weights are not published","How much video memory and disk the training and sampling steps really need","What renting H100 time would cost to train the staged schedule described in the README","Whether the kernel still builds with current CUDA and PyTorch versions, as the last commit is from February 2026"]},"zh-Hans":{"summaryShort":"研究代码：让视频 AI 模型能生成一分钟的片段，而不只是三秒。","summary":"这是论文《One-Minute Video Generation with Test-Time Training》的官方代码。它在已有的文生视频模型（CogVideoX 5B）上加入新层，并分阶段训练，使生成的视频更长，最长可达 63 秒，且首尾风格保持一致。这是面向有机器学习经验者的研究项目：你要从源码编译 CUDA 内核、自行获取基础模型，并在 H100 显卡上训练。它没有应用程序、没有界面，也没有可下载的成品模型。","easeRationale":"只适合开发者，且限于 Linux：你要创建 conda 环境，拉取一个 git 子模块并编译 TTT-MLP 内核，这需要 CUDA toolkit 12.3 以上和 gcc 11 以上，作者还说训练只支持 H100 显卡。基础模型的权重则要从另外两个项目手动收集。","notVerified":["是否有人能复现那些一分钟的演示视频，因为最终权重并未公开","训练和采样环节实际需要多少显存和磁盘空间","按 README 里描述的分阶段方案训练，租用 H100 大约要花多少钱","该内核在当前版本的 CUDA 和 PyTorch 上是否还能编译，因为最后一次提交是 2026 年 2 月"]}}},{"id":"R_kgDOVmXm5g","slug":"th0rgal__satmux","name":"satmux","owner":"Th0rgal","fullName":"Th0rgal/satmux","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"routes_around_onboard_network_controls","severity":"attention","evidence":"README.md:它与地面站附近的 Linux 虚拟机保持 \"12 pre-warmed TCP connections open to a Linux VM near the satellite ground station, bypassing the cabin :443 proxy and onboard DNS\",端口为 8444,因为 \"Non-443 TCP ports pass through untouched\"。它使用你已付费的套餐,README 也说 \"it does not bypass captive portals\",但绕开运营商自己的设备可能与航空公司或服务商的使用条款相冲突。","source":"ai","title":"绕过机上网络管控"},{"code":"hand_rolled_encryption","severity":"attention","evidence":"README.md 将与地面服务器的连接描述为 \"an encrypted (BLAKE2b-512 stream + HMAC-SHA256) request frame\",使用你自己生成的共享密钥(`openssl rand -hex 32 > mux.psk`)。这是建立在哈希函数之上的自制方案,而不是像 TLS 那样经过审查的协议,所有 ground-fetch 流量在服务器上都以明文形式经过它。","source":"ai","title":"自制加密方案"},{"code":"changes_system_and_tool_settings","severity":"attention","evidence":"README.md:`bin/satmux-macos on` 会设置系统的 Wi-Fi 代理,关闭时则 \"Restores macOS Wi-Fi proxy settings, ~/.gitconfig, gh http_unix_socket, and launchctl env vars\"。如果你落地后忘了运行 `bin/satmux-macos off`,你的 Mac、git 和 gh 会一直指向一个已经不存在的服务器。","source":"ai","title":"更改系统和工具设置"},{"code":"server_started_as_root","severity":"info","evidence":"README.md 安装步骤 2 把脚本复制到虚拟机上并以 root 身份运行:`ssh root@<SERVER_IP> 'nohup env SATMUX_PSK=$(cat /root/mux.psk) python3 /root/satmux.py server --port 8444 ...'`,TCP 端口 8444 对互联网开放。","source":"ai","title":"服务器以 root 身份启动"},{"code":"ground_server_sees_decrypted_traffic","severity":"attention","evidence":"README.md:在 \"1-RTT ground fetch (127.0.0.1:8081)\" 模式下,\"The ground VM performs DNS, verified TLS 1.3, and HTTP redirects on fiber\",并且 \"compresses plaintext responses with zlib\",因此在该模式下,是租用的服务器而不是你的笔记本电脑持有 TLS 连接和页面内容。SOCKS5 模式则保持端到端的 TLS。","source":"ai","title":"地面服务器能看到解密后的流量"},{"code":"very_new_repository","severity":"info","evidence":"创建于 2026-11-07,最后一次提交也在同一天,已有 922 个星标,没有发行版本,没有测试目录,0 个未解决的 issue;整个程序只有一个文件,satmux.py。","source":"ai","title":"仓库非常新"}],"platforms":["mac","linux"],"tools":["python"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":1030,"downloads":0,"downloadsWeek":0,"downloadsDays":1,"growth":1030,"growthDays":3,"history":[{"date":"2026-11-09","stars":1004},{"date":"2026-11-10","stars":1030}],"firstPublishedAt":"2026-11-08T23:01:18.959Z","commitSha":"2922ebb835d0eb939ddbdfa3b5e7e0f247a0dc0c","verifiedAt":"2026-11-08T22:43:30.783Z","lastCommitAt":"2026-11-07T16:24:29Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Makes slow satellite inflight Wi-Fi usable by sending your traffic through a rented server near the ground station.","summary":"satmux is a single Python file that speeds up airplane Wi-Fi on satellite connections (Inmarsat GX, Viasat, Panasonic, SITA). It keeps twelve connections open to a small Linux server you rent near the satellite's ground station, so web pages, APIs, git and the gh command stop timing out on the plane's own proxy. It is for technical users: you rent a server, copy the file over by SSH, start it on both ends and run a script that changes your macOS proxy settings. Pure Python 3 with no extra packages. It does not get you past the paid Wi-Fi login.","easeRationale":"There is no installer or release: on macOS, the best-documented side, you clone the repository, generate a key with openssl, rent a Linux VM, copy the file over with scp, start it over SSH as root and run the client plus a proxy script. Linux works the same way but has no helper script to switch the system proxy on and off, so you would wire that up yourself.","costNote":"The program is free (MIT). The README says you also need to rent a Linux VM, \"$0.01–$0.03/h\", and an already paid inflight Wi-Fi session.","notVerified":["Whether the speed numbers in the README (3.6x faster, fixed timeouts) happen on other airlines, satellites or seats","Whether the home-made key-based encryption of the carrier link holds up, since no review or test suite is in the repository","What satmux.py actually sends to the ground server, and whether it logs requests there","Whether `bin/satmux-macos off` restores every macOS, git and gh setting it changed","The real cost of the rented VM and of the inflight Wi-Fi session on a given flight"]},"zh-Hans":{"summaryShort":"让缓慢的卫星机上 Wi-Fi 变得可用,方法是把你的流量发送到地面站附近租用的服务器。","summary":"satmux 是一个单文件的 Python 程序,用于加速卫星连接(Inmarsat GX、Viasat、Panasonic、SITA)上的飞机 Wi-Fi。它与你在卫星地面站附近租用的一台小型 Linux 服务器保持十二条连接,这样网页、API、git 和 gh 命令就不会在飞机自身的代理上超时。它面向技术用户:你需要租一台服务器,通过 SSH 把文件复制过去,在两端都启动它,并运行一个脚本来修改你 macOS 的代理设置。纯 Python 3 实现,不依赖额外的包。它不能帮你绕过付费 Wi-Fi 的登录。","easeRationale":"没有安装程序,也没有发行版本:在文档最完善的 macOS 上,你需要克隆仓库、用 openssl 生成密钥、租用一台 Linux 虚拟机、用 scp 复制文件、以 SSH 方式在两端都启动它,再运行客户端和一个代理脚本。Linux 的做法类似,但没有辅助脚本来开关系统代理,需要你自己处理。","costNote":"该程序是免费的(MIT 许可)。README 说你还需要租一台 Linux 虚拟机,\"$0.01–$0.03/h\",以及一个已经付费的机上 Wi-Fi 会话。","notVerified":["README 中的速度数字(快 3.6 倍、不再超时)在其他航空公司、卫星或座位上是否同样成立","由于仓库里没有审查或测试套件,运营商线路上这种自制的基于密钥的加密是否可靠","satmux.py 实际向地面服务器发送了什么内容,以及它是否在那里记录请求","`bin/satmux-macos off` 是否恢复了它更改过的所有 macOS、git 和 gh 设置","在具体某次航班上,租用虚拟机和机上 Wi-Fi 会话的实际费用"]}}},{"id":"R_kgDOJ4EDkg","slug":"the-pr-agent__pr-agent","name":"PR Agent","owner":"The-PR-Agent","fullName":"The-PR-Agent/pr-agent","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"your_code_is_sent_to_the_model_provider","severity":"attention","evidence":"工具会把 pull request 的差异内容发送给所配置的 AI 服务商；README 的 Data Privacy 一节只说 \"If you host PR-Agent with your OpenAI API key, it is between you and OpenAI\"，并链接到 OpenAI 的政策。","source":"ai","title":"你的代码会发送给模型服务商"},{"code":"feature_disabled_for_credential_exposure","severity":"attention","evidence":"README 的 Features 一节：\"/help_docs is temporarily disabled since v0.36.1 pending a fix for a credential-exposure issue (#2445)\"。","source":"ai","title":"因凭据泄露问题而停用的功能"},{"code":"ownership_changed_and_not_the_vendor_product","severity":"info","evidence":"仓库描述：\"This project is not the Qodo free tier.\" README 说 Qodo 已把项目捐给社区，它是 \"a community-maintained legacy project\"，目前有了第一位外部维护者，并且正 \"in the process of being donated to an open-source foundation\"。","source":"ai","title":"项目已易主，并非厂商自家产品"},{"code":"docker_images_moved_namespace","severity":"info","evidence":"README 的说明：0.34.2 及之后的版本发布在 pragent/pr-agent 下，而旧的 codiumai/pr-agent 已冻结存档、不再接收新镜像，因此旧的固定引用会一直停留在过时版本上。","source":"ai","title":"Docker 镜像换了命名空间"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["AI model provider (OpenAI, Anthropic, Gemini and others via LiteLLM)","Git host token (GitHub, GitLab, Bitbucket, Azure DevOps or Gitea)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"coderabbit","name":"CodeRabbit","confirmed":true},{"slug":"qodo","name":"Qodo","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true}],"declaredAlternatives":[],"stars":15860,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1180,"growthDays":6,"history":[{"date":"2026-11-07","stars":15431},{"date":"2026-11-08","stars":15586},{"date":"2026-11-09","stars":15711},{"date":"2026-11-10","stars":15860}],"firstPublishedAt":"2026-11-10T23:52:44.347Z","commitSha":"f4257a28b55fe5850a1d3c3740cdb66ad97f9d7c","verifiedAt":"2026-11-10T23:36:55.329Z","lastCommitAt":"2026-11-10T20:22:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An AI assistant that reads a pull request and writes the review, the description and improvement suggestions for you.","summary":"PR-Agent is a tool for software teams. It reads a pull request on GitHub, GitLab, Bitbucket, Azure DevOps or Gitea and answers commands such as /review, /describe, /improve and /ask with AI-written feedback posted back as a comment. You can run it as a GitHub Action, as a command on your own computer, or as a webhook service. The program is free, but it needs an AI model: a paid key (OpenAI, Anthropic, Gemini and others through LiteLLM) or a local model through Ollama.","easeRationale":"The same three steps work on Windows, Mac and Linux: install Python, run pip install \"pr-agent[github]\", set your model key in an environment variable and run the pr-agent command with a pull-request link. The simplest route for a repository owner is instead a ten-line GitHub Actions file with two secrets.","costNote":"The code is MIT and free. Each command is one call to an AI model, which is normally a paid provider; a local model through Ollama is also supported.","notVerified":["What a month of reviews actually costs with a paid model, since it depends on the provider and the size of your pull requests","How useful the review comments are on your own codebase","Whether every listed git host and model provider works as well as GitHub with OpenAI","Whether a local Ollama model gives usable reviews, which the README mentions but does not demonstrate"]},"zh-Hans":{"summaryShort":"一个 AI 助手，读完 pull request 后替你写好评审意见、描述和改进建议。","summary":"PR-Agent 是面向软件团队的工具。它会读取 GitHub、GitLab、Bitbucket、Azure DevOps 或 Gitea 上的 pull request，并响应 /review、/describe、/improve、/ask 等命令，把 AI 写好的反馈以评论形式发回去。你可以把它作为 GitHub Action 运行，也可以在自己电脑上作为命令运行，或部署成 webhook 服务。程序本身免费，但需要一个 AI 模型：付费密钥（通过 LiteLLM 支持 OpenAI、Anthropic、Gemini 等），或通过 Ollama 使用本地模型。","easeRationale":"Windows、Mac 和 Linux 上的三步相同：装好 Python，运行 pip install \"pr-agent[github]\"，把模型密钥写进环境变量，再带上 pull request 链接运行 pr-agent 命令。对仓库所有者来说，更简单的做法是写一个十行的 GitHub Actions 配置文件，配两个密钥。","costNote":"代码采用 MIT 许可且免费。每执行一次命令就调用一次 AI 模型，通常来自付费服务；也支持通过 Ollama 使用本地模型。","notVerified":["用付费模型做一个月的评审实际要花多少钱，这取决于服务商和你 pull request 的规模","在你自己的代码库上，这些评审意见到底有多大帮助","文中列出的所有 git 平台和模型服务，是否都能像 GitHub 配 OpenAI 那样好用","用 Ollama 的本地模型能否给出可用的评审，README 提到了这一点但没有演示"]}}},{"id":"R_kgDOQLtlug","slug":"thecraighewitt__seomachine","name":"SEO Machine","owner":"TheCraigHewitt","fullName":"TheCraigHewitt/seomachine","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"command_hides_signs_of_ai_writing","severity":"attention","evidence":"README 的命令列表：\"/scrub [file] Remove AI watermarks and patterns from content (em-dashes, filler phrases, robotic patterns)\"，Editor 智能体还会给出 \"Humanity score (0-100)\" 和 \"specific rewrites to inject personality\"。其目的就是让 AI 写的文字读起来像真人写的。","source":"ai","title":"有命令专门抹去 AI 写作的痕迹"},{"code":"handles_credentials_for_your_accounts","severity":"attention","evidence":"配置过程要求在 ./credentials 放入 Google 服务账号的 JSON 文件，在 .env 里填写 DataForSEO 的账号密码、Exa 密钥以及 WordPress 的 WP_USERNAME / WP_APP_PASSWORD；docker-compose.yml 还会把 ./credentials 挂载进容器。在该目录下运行的 AI 助手可以读取全部内容。","source":"ai","title":"会经手你各个账号的凭据"},{"code":"publishing_needs_a_plugin_on_your_site","severity":"info","evidence":"WordPress 集成说明：\"Install wordpress/seo-machine-yoast-rest.php as an MU-plugin on your WordPress site\"，并在主题的 functions.php 里加一段代码，这样工具就能在正式站点上创建文章。","source":"ai","title":"发布功能需要在你的网站上装插件"},{"code":"paid_accounts_required_for_the_main_use","severity":"info","evidence":"前置条件：Claude Code 并拥有 \"Claude Pro, Max, Team, or Enterprise subscription\" 或 \"Anthropic Console account with API billing enabled\"，另加一个 \"DataForSEO account\"；README 指出这些都是单独计费的。","source":"ai","title":"主要用途需要付费账号"}],"platforms":["windows","mac","linux"],"tools":["python","other"],"buildTools":["docker"],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Anthropic Claude (subscription or API billing, through Claude Code)","DataForSEO","Google Analytics 4 and Search Console (service account)","Exa","WordPress application password"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","social_marketing"],"alternativeTo":[{"slug":"semrush","name":"Semrush","confirmed":true},{"slug":"jasper","name":"Jasper","confirmed":true},{"slug":"surfer-seo","name":"SEO","confirmed":true},{"slug":"ahrefs","name":"Ahrefs","confirmed":true},{"slug":"undetectable-ai","name":"Undetectable AI","confirmed":true}],"declaredAlternatives":[],"stars":7593,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":18,"growthDays":6,"history":[{"date":"2026-11-05","stars":7576},{"date":"2026-11-06","stars":7583},{"date":"2026-11-07","stars":7586},{"date":"2026-11-08","stars":7588},{"date":"2026-11-09","stars":7589},{"date":"2026-11-10","stars":7593}],"firstPublishedAt":"2026-11-10T23:52:44.510Z","commitSha":"120e7b46dc07c98382cd4f68ac5b6515ab9f2604","verifiedAt":"2026-11-10T23:37:55.121Z","lastCommitAt":"2026-10-23T20:55:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A pack of commands for Claude Code that researches keywords, writes long blog articles for your business and scores them for SEO.","summary":"SEO Machine is a ready-made workspace for Claude Code aimed at people who publish blog content. It adds commands like /research, /write, /optimize and /publish-draft, ten review agents and 26 marketing skills, plus Python scripts that pull numbers from Google Analytics, Search Console and DataForSEO. You fill in files describing your brand voice, keywords and internal links, and it drafts 2,000-3,000 word articles and can post them to WordPress. You need Claude Code with a paid plan and a DataForSEO account.","easeRationale":"The same steps on Windows, Mac and Linux: install Claude Code, clone the repository, run pip install for the analysis scripts, then open the folder with the claude command (a Docker Compose file is offered as an alternative). The long part is not technical: you must fill in eight context files and sign up for DataForSEO and, optionally, Google service accounts.","costNote":"The repository is MIT and free, but the README requires Claude Code with a Claude Pro/Max/Team/Enterprise subscription or Anthropic API billing, plus a paid DataForSEO account; Exa, Google Analytics and Search Console are extra.","notVerified":["Whether the articles it writes actually rank, which no repository content can show","What a month costs in practice between the Claude plan, DataForSEO queries and optional Exa calls","Whether the Python scripts still match the current Google Analytics, Search Console and DataForSEO APIs","How much of the result depends on how well you fill in the brand-voice and keyword files"]},"zh-Hans":{"summaryShort":"一套 Claude Code 命令：做关键词调研、为你的业务写长篇博客文章，并给出 SEO 评分。","summary":"SEO Machine 是为 Claude Code 准备好的工作区，面向经营博客内容的人。它添加了 /research、/write、/optimize、/publish-draft 等命令、十个评审智能体和 26 项营销技能，还有从 Google Analytics、Search Console 和 DataForSEO 拉取数据的 Python 脚本。你需要填写描述品牌口吻、关键词和内链的文件，它就能写出 2000 到 3000 字的文章，并可以直接发布到 WordPress。你需要付费订阅的 Claude Code 和一个 DataForSEO 账号。","easeRationale":"Windows、Mac 和 Linux 上的步骤相同：安装 Claude Code，克隆仓库，为分析脚本运行 pip install，然后用 claude 命令打开该目录（另有 Docker Compose 文件可选）。真正费时的部分不是技术活：你得填好八个背景资料文件，并注册 DataForSEO，必要时还要配置 Google 服务账号。","costNote":"仓库采用 MIT 许可且免费，但 README 要求使用 Claude Code，并拥有 Claude Pro/Max/Team/Enterprise 订阅或开通了 Anthropic API 计费，另外还要一个付费的 DataForSEO 账号；Exa、Google Analytics 和 Search Console 另算。","notVerified":["它写出的文章是否真能获得搜索排名，这一点仓库里的任何内容都无法证明","算上 Claude 订阅、DataForSEO 查询和可选的 Exa 调用，一个月实际要花多少钱","这些 Python 脚本是否仍与当前的 Google Analytics、Search Console 和 DataForSEO 接口相符","最终效果有多大程度取决于你把品牌口吻和关键词文件填得好不好"]}}},{"id":"R_kgDOPng1Jw","slug":"thedotmack__claude-mem","name":"claude-mem","owner":"thedotmack","fullName":"thedotmack/claude-mem","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本通过管道传给 shell：curl -fsSL https://install.cmem.ai/openclaw.sh | bash","source":"check","title":"curl 管道传输到 shell"},{"code":"records_everything_your_assistant_does","severity":"attention","evidence":"仓库描述说它 \"Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions\"。功能列表提供了 `<private>` 标签 \"to exclude sensitive content from storage\"，这意味着除此之外的一切，包括助手在你文件里读到的内容，默认都会被存储。","source":"ai","title":"记录你的助手做的一切"},{"code":"default_memory_provider_is_the_vendors_cloud","severity":"attention","evidence":"README Quick Start：\"Default is CMEM Pro, the hosted memory. Local observer is opt-in: `--provider host`\"，安装程序 \"asks you to sign in to claude-mem in your browser (email magic link)\"。除非你传入 `--provider` 或设置 `CLAUDE_MEM_ONLINE_OPTIN=false`，否则你捕获的工作内容会由该公司的服务来总结。","source":"ai","title":"默认的记忆提供方是厂商的云服务"},{"code":"installs_extra_runtimes_without_asking","severity":"attention","evidence":"README 系统要求：\"Bun: JavaScript runtime and process manager (auto-installed if missing)\" 以及 \"uv: Python package manager for vector search (auto-installed if missing)\"。安装过程还会在 `%LOCALAPPDATA%\\claude-mem\\bin` 放置一个启动器，并在 Windows 上修改你的用户 Path。","source":"ai","title":"未经询问就安装额外的运行环境"},{"code":"runs_hooks_and_a_local_web_service","severity":"attention","evidence":"README「它是如何工作的」部分：它注册了生命周期 hooks（SessionStart、UserPromptSubmit、PreToolUse、PostToolUse、Stop、SessionEnd），在每个事件上运行，另外还有一个 \"Worker Service - Local HTTP API with web viewer UI and search endpoints, managed by Bun\"，端口在启动时打印出来。自托管服务器的 `docker-compose.yml` 警告说 \"MUST NOT be deployed unmodified to any environment that is reachable from the public internet\"。","source":"ai","title":"运行 hooks 和一个本地网页服务"},{"code":"promotes_a_crypto_token","severity":"attention","evidence":"README.md 的最后一节「What About CMEM?」说 \"CMEM is a token created by a 3rd party but officially embraced by the creator of Claude-Mem\"，并打印了一个合约地址：\"Official BASE CA: 0x76b1967eec0ccaeb001bbbb2b40dc4badba31ba3\"。一个开发者工具的页面在推广一种代币，说明它对关注度有超出软件本身的兴趣。","source":"ai","title":"推广一种加密代币"},{"code":"usage_analytics_are_built_into_the_worker","severity":"info","evidence":"`package.json` 说明 `posthog-node` 等库 \"esbuild-inlined into the worker/server/npx bundles\"，也就是说分析代码被打包进了运行在你机器上的服务里。","source":"ai","title":"使用情况分析内置在工作进程里"},{"code":"very_fast_releases_and_star_growth","severity":"info","evidence":"项目数据：仓库创建于 2025-08-31，已有 110810 个星标、386 个发行版，仅 2026-11-02 当天就发布了三个 15.9.0 系列版本（v15.9.0、v15.8.3、v15.8.2）。功能可能一天之内就发生变化。","source":"ai","title":"发行版和星标增长速度都非常快"}],"platforms":["windows","mac","linux"],"tools":["node","docker"],"buildTools":["node","docker"],"disk":{"minMB":150,"maxMB":650,"basis":"heuristic"},"apiKeys":["claude-mem / CMEM Pro account (email sign-in, default provider)","OpenRouter, Gemini or your Anthropic plan, if you supply the model yourself"],"cost":{"model":"freemium"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","productivity"],"alternativeTo":[{"slug":"mem0","name":"Mem0","confirmed":true},{"slug":"zep","name":"Zep","confirmed":true},{"slug":"supermemory","name":"Supermemory","confirmed":true}],"declaredAlternatives":[],"stars":115770,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":3895,"growthDays":7,"history":[{"date":"2026-10-31","stars":107881},{"date":"2026-11-01","stars":109394},{"date":"2026-11-02","stars":110748},{"date":"2026-11-03","stars":111875},{"date":"2026-11-04","stars":112717},{"date":"2026-11-05","stars":113488},{"date":"2026-11-06","stars":113958},{"date":"2026-11-07","stars":114404},{"date":"2026-11-08","stars":114844},{"date":"2026-11-09","stars":115355},{"date":"2026-11-10","stars":115770}],"firstPublishedAt":"2026-10-31T01:54:33.221Z","commitSha":"02584c4ac9d92fea45adf5c61e22c546b5738621","verifiedAt":"2026-11-03T02:45:48.058Z","lastCommitAt":"2026-11-10T21:22:24Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Remembers what your AI coding assistant did in earlier sessions and feeds the useful parts back into new ones.","summary":"A plugin that gives an AI coding assistant a memory. While you work, it records what the assistant did, has an AI model turn that into short notes, stores them in a database on your machine, and puts the relevant ones back at the start of your next session. It also adds search tools so you can ask about your project's history. It works with Claude Code, OpenClaw, Codex, OpenCode, Antigravity and others, is installed with `npx claude-mem install`, and runs a small background service with a local web page for browsing what it saved.","easeRationale":"Windows, macOS and Linux follow the same path: have Node.js 20+ and a supported AI assistant, run `npx claude-mem install` (or two `/plugin` commands inside Claude Code), sign in by email or pass a `--provider` flag, then restart the assistant. Windows needs one extra check, since the README has a section for when `npm` is not on your PATH and the launcher must be added to it.","costNote":"The code is free (Apache-2.0), but the default memory provider is the paid hosted service CMEM Pro with a \"30 Day Free Trial\"; after that the README says memory \"falls back to your Anthropic plan unless you subscribe\". You can instead point it at your own OpenRouter or Gemini key, which you pay for.","notVerified":["Exactly which parts of your sessions and files are stored, and which of them leave your machine","What it costs after the 30-day trial, and what happens to your saved memories if you stop paying","How many extra tokens the injected context adds to every session of your AI assistant","Whether the `<private>` tags and the opt-in screenshot setting really keep that content out of storage and out of the cloud","Real disk use as the database, the vector index and the converted screenshots grow under `~/.claude-mem`","What the background worker and the installed hooks do on your computer between sessions"]},"zh-Hans":{"summaryShort":"记住你的 AI 编程助手在之前会话中做过的事，并把有用的部分带回新会话里。","summary":"这是一个给 AI 编程助手添加记忆的插件。你工作时，它会记录助手做过的事，让一个 AI 模型把这些内容整理成简短的笔记，存到你机器上的一个数据库里，然后把相关的部分在你下次会话开始时带回来。它还增加了搜索工具，方便你查询项目的历史记录。它适用于 Claude Code、OpenClaw、Codex、OpenCode、Antigravity 等，用 `npx claude-mem install` 安装，并运行一个带本地网页的小型后台服务，供你浏览保存的内容。","easeRationale":"Windows、macOS 和 Linux 的步骤相同：安装 Node.js 20+ 和一个受支持的 AI 助手，运行 `npx claude-mem install`（或者在 Claude Code 里执行两个 `/plugin` 命令），用邮箱登录或传入 `--provider` 参数，然后重启助手。Windows 需要多一步检查，因为 README 里有一节专门讲当 `npm` 不在你的 PATH 中时，要把启动器手动加进去。","costNote":"代码是免费的（Apache-2.0 许可），但默认的记忆提供方是付费托管服务 CMEM Pro，带有「30 Day Free Trial」；之后 README 说记忆功能 \"falls back to your Anthropic plan unless you subscribe\"。你也可以改用自己的 OpenRouter 或 Gemini 密钥，费用由你自己承担。","notVerified":["你的会话和文件中具体哪些部分被存储，哪些会离开你的机器","30 天试用之后要花多少钱，如果你停止付费，保存的记忆会怎样","注入的上下文给你 AI 助手的每次会话增加了多少额外 token","`<private>` 标签和可选的截图设置是否真的能让那部分内容不进入存储和云端","随着数据库、向量索引和转换后的截图在 `~/.claude-mem` 下不断增长，实际占用多少磁盘","后台工作进程和安装的 hooks 在会话之间在你电脑上做了什么"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyNDc3NzAxNTg=","slug":"theothernt__aerialviews","name":"Aerial Views","owner":"theothernt","fullName":"theothernt/AerialViews","type":"ready_app","status":"active","grade":"C","easeByPlatform":{},"risk":"attention","signals":[{"code":"optional_http_server_without_authentication","severity":"attention","evidence":"Message API \"allows you to display custom messages on your TV from any device on your network\"，监听 8081 端口；文档列出的接口（GET /status、POST /message/1）没有任何密码、令牌或其他身份验证。","source":"ai","title":"可选的 HTTP 服务没有身份验证"},{"code":"setup_asks_you_to_enable_usb_debugging","severity":"info","evidence":"把它设为默认屏保的说明要求在电视上开启开发者模式和 USB/ADB 调试，再通过网络用 adb 连接；调试一直开着，就意味着同一网络里的任何设备都能连到这台电视。","source":"ai","title":"设置过程要求开启 USB 调试"},{"code":"videos_come_from_third_parties","severity":"info","evidence":"README：\"Over 250 videos from Apple, Amazon, Jetson Creative and Robin Fourcade\"，这些内容从上述来源流式播放或下载，项目本身既不拥有也不托管它们。","source":"ai","title":"视频来自第三方"},{"code":"integrations_hold_your_media_credentials","severity":"info","evidence":"Immich 需要一个具备 album.read、asset.read、asset.view 和 asset.download 权限的 API 密钥；对于 Nextcloud Memories，README 写道 \"You can use either user password or app password\" 并建议用应用密码——也就是说，你照片服务器的凭据会保存在电视上。","source":"ai","title":"这些集成会保存你的媒体服务器凭据"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["Immich server API key","Nextcloud app password"],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"file"},"categories":["hardware","video"],"alternativeTo":[],"declaredAlternatives":[],"stars":1154,"downloads":50233,"downloadsWeek":1422,"downloadsDays":4,"growth":6,"growthDays":6,"history":[{"date":"2026-11-06","stars":1149},{"date":"2026-11-07","stars":1149},{"date":"2026-11-08","stars":1151},{"date":"2026-11-09","stars":1154},{"date":"2026-11-10","stars":1154}],"firstPublishedAt":"2026-11-10T23:52:44.556Z","commitSha":"de93d572c78dfd45119804a8c6735294b3618dea","verifiedAt":"2026-11-10T23:38:55.078Z","lastCommitAt":"2026-11-09T22:39:22Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns an Android TV, Google TV or Fire TV into an Apple TV-style screensaver with aerial videos or your own photos.","summary":"Aerial Views is a screensaver app for Android TV, Google TV, Fire TV, Nvidia Shield and Android phones. When the device is idle it plays more than 250 aerial videos, including 4K Dolby Vision ones, or your own photos and videos from a USB stick, an Immich or Nextcloud Memories server, Samba or WebDAV. Clock, weather, date and custom messages can be shown in the corners. It installs from Google Play, the Amazon Appstore or an APK, but on most recent Google TV devices making it the default screensaver needs ADB commands from a computer or phone.","easeRationale":"Android TV only, so none of the desktop platforms gets a grade. Installing is a one-tap store download, but since 2023 nearly all Google TV devices have no menu to pick a third-party screensaver: the README's route is to enable developer mode and USB debugging and run two ADB commands from a PC, Mac or phone. On an Nvidia Shield or older Android TV it is just a settings menu, and a paid third-party app (about $2) can do it instead.","costNote":"Free on Google Play, the Amazon Appstore and as an APK; the author accepts donations through Ko-fi.","notVerified":["Whether the screensaver can be set as the default on your exact TV model and system version, which the README says varies a lot","How much internet data the streamed 4K videos use per hour","Whether the Fire TV steps still work after recent Fire OS updates, which the README flags as changing","How well it behaves on phones and tablets, which are mentioned but not the main target"]},"zh-Hans":{"summaryShort":"把 Android TV、Google TV 或 Fire TV 变成 Apple TV 风格的屏保，播放航拍视频或你自己的照片。","summary":"Aerial Views 是面向 Android TV、Google TV、Fire TV、Nvidia Shield 和安卓手机的屏保应用。设备闲置时，它会播放 250 多段航拍视频（部分为 4K 杜比视界），也可以播放来自 U 盘、Immich 或 Nextcloud Memories 服务器、Samba 或 WebDAV 的自有照片和视频。屏幕角落可以显示时钟、天气、日期和自定义文字。它可以通过 Google Play、亚马逊应用商店或 APK 安装，但在多数较新的 Google TV 设备上，要把它设为默认屏保还得从电脑或手机执行 ADB 命令。","easeRationale":"仅限 Android TV，因此桌面平台都不给等级。安装只需在商店点一下，但自 2023 年起几乎所有 Google TV 设备都不再提供选择第三方屏保的菜单：README 给出的办法是开启开发者模式和 USB 调试，再从 PC、Mac 或手机执行两条 ADB 命令。在 Nvidia Shield 或较老的 Android TV 上，只要在设置菜单里选一下即可，也可以用一款约 2 美元的第三方付费应用代劳。","costNote":"在 Google Play、亚马逊应用商店和 APK 形式下均免费；作者通过 Ko-fi 接受捐助。","notVerified":["在你具体的电视型号和系统版本上能否设为默认屏保，README 说这一点差异很大","流式播放 4K 视频每小时消耗多少流量","在 Fire OS 最近更新之后，针对 Fire TV 的步骤是否仍然有效，README 提到这部分经常变动","在手机和平板上表现如何，文中有提及但那并非主要目标设备"]}}},{"id":"R_kgDOIzPnqw","slug":"ther1d__shell_gpt","name":"ShellGPT","owner":"TheR1D","fullName":"TheR1D/shell_gpt","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"ai_can_run_commands_on_your_computer","severity":"attention","evidence":"函数调用功能自带一个执行 shell 命令的函数：\"we are allowing LLM to execute any Shell commands in our system\"，示例中模型自己装了软件（brew install jq）。README 提醒：\"LLM might execute destructive commands, so please use it at your own risk\"，而配置文件里写的是 OPENAI_USE_FUNCTIONS=true。","source":"ai","title":"AI 可以在你的电脑上执行命令"},{"code":"whatever_you_pipe_in_is_sent_to_the_provider","severity":"info","evidence":"文档中的用法会把本地内容传给模型，例如 \"git diff | sgpt\"、\"docker logs -n 20 my_app | sgpt\" 和 \"sgpt \\\"summarise\\\" < document.txt\"；除非你改用本地模型，这些文本都会离开你的电脑发往 OpenAI。","source":"ai","title":"你用管道传入的一切都会发给服务商"},{"code":"edits_your_shell_startup_file","severity":"info","evidence":"README：执行 sgpt --install-integration 会 \"will add few lines to your .bashrc or .zshrc file\"，这样按 Ctrl+L 就能把你输入的内容替换成建议的命令。","source":"ai","title":"会修改你的 shell 启动文件"},{"code":"api_key_stored_in_plain_text","severity":"info","evidence":"你粘贴的密钥 \"will then be stored in ~/.config/shell_gpt/.sgptrc\"，这是一个纯文本配置文件，其他设置也都存在里面。","source":"ai","title":"API 密钥以明文保存"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","docker","ffmpeg"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":["OpenAI"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","productivity","ai_agents"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true},{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true}],"declaredAlternatives":[{"target":"buffer","stance":"claims","evidence":"When you press `Ctrl+l` it will replace you current input line (buffer) with suggested command"}],"stars":12309,"downloads":2838,"downloadsWeek":3,"downloadsDays":2,"growth":5,"growthDays":6,"history":[{"date":"2026-11-08","stars":12307},{"date":"2026-11-09","stars":12306},{"date":"2026-11-10","stars":12309}],"firstPublishedAt":"2026-11-10T23:52:44.652Z","commitSha":"a082bd5327ce0c4ef5a0284d9060e833be9444a6","verifiedAt":"2026-11-10T23:39:35.014Z","lastCommitAt":"2026-07-02T06:01:36Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ask questions in your terminal and get answers, ready-to-run commands or code, without leaving the command line.","summary":"ShellGPT (sgpt) puts an AI assistant inside your terminal. You type a question and it answers; with -s it writes the exact command for your system and offers to run it, and with -c it returns only code. You can pipe files, logs or a git diff into it, keep named conversations, and press Ctrl+L to turn what you typed into a command. It works on Windows, Mac and Linux after a pip install. By default it uses OpenAI's paid API; a local model through Ollama is possible but the author says it is not optimized for that.","easeRationale":"Same on Windows, Mac and Linux: install Python, run pip install shell-gpt, paste an OpenAI key when asked, then type sgpt with your question. A ready Docker image is offered as an alternative for people who prefer containers.","costNote":"The tool is MIT and free; it calls OpenAI's API, which is paid per use. A free local model through Ollama is documented in the wiki but the README warns it \"is not optimized for local models and may not work as expected\".","notVerified":["What the OpenAI usage costs in a normal month of use","How well the suggested commands match your own system and shell","Whether a local Ollama model gives usable results, which the author says is not optimized","Whether it still works with the newest OpenAI models, as the last commit is from July 2026"]},"zh-Hans":{"summaryShort":"在终端里提问，直接得到答案、可执行的命令或代码，不用离开命令行。","summary":"ShellGPT（sgpt）把 AI 助手搬进你的终端。你输入问题，它就回答；加上 -s 时，它会写出适配你系统的确切命令并询问是否执行，加上 -c 时只返回代码。你可以用管道把文件、日志或 git diff 传给它，保存带名字的会话，还能按 Ctrl+L 把刚输入的内容变成命令。安装只需 pip install，Windows、Mac 和 Linux 都能用。它默认调用 OpenAI 的付费 API；也可以通过 Ollama 使用本地模型，但作者说针对本地模型并未做过优化。","easeRationale":"Windows、Mac 和 Linux 上一样：装好 Python，运行 pip install shell-gpt，按提示粘贴 OpenAI 密钥，然后输入 sgpt 加上你的问题。对偏好容器的人，项目还提供了现成的 Docker 镜像。","costNote":"工具本身采用 MIT 许可且免费；它调用 OpenAI 的 API，按用量付费。wiki 里有通过 Ollama 使用免费本地模型的说明，但 README 提醒它 \"is not optimized for local models and may not work as expected\"。","notVerified":["按正常使用强度，一个月的 OpenAI 费用大概是多少","它建议的命令与你自己的系统和 shell 匹配得如何","用 Ollama 的本地模型能否得到可用结果，作者说这方面没有优化","它是否还能配合 OpenAI 最新的模型，因为最后一次提交是 2026 年 7 月"]}}},{"id":"R_kgDOLlJhbw","slug":"tmelyralab__musev","name":"MuseV","owner":"TMElyralab","fullName":"TMElyralab/MuseV","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{"windows":"D","linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-37q5-v5qm-c9v8; accelerate: GHSA-4j2p-28q2-5m79, PYSEC-2026-3804; onnx: GHSA-3r9x-f23j-gc73, GHSA-48m4-24cq-37qf, GHSA-538c-55jv-c5g9; protobuf: GHSA-7gcm-g887-7qv7, GHSA-8qvm-5x2c-j2w7, PYSEC-2026-1805; opencv-contrib-python: GHSA-cxjf-x6jp-p7mc, PYSEC-2023-181; pytorch-lightning: GHSA-4cv3-v7pv-rfhf, GHSA-75m9-98v2-hjpm, GHSA-98fp-7v67-4v3q; gdown: GHSA-76hw-p97h-883f, PYSEC-2026-2158; redis: GHSA-24wv-mv5m-xv4h, GHSA-8fww-64cx-x8p5, PYSEC-2023-45","source":"check","title":"存在已知漏洞"},{"code":"model_license_non_commercial","severity":"attention","evidence":"README.md 的 Disclaimer/License：\"model: The trained model are available for non-commercial research purposes only.\" 以及 \"The testdata are collected from internet, which are available for non-commercial research purposes only.\"","source":"ai","title":"模型许可仅限非商业用途"},{"code":"setup_runs_installs","severity":"info","evidence":"setup.py 在安装项目的过程中还会执行包安装：subprocess.run([\"pip\", \"install\", \"basicsr\"])，以及用 mim install 安装 mmengine、mmcv、mmdet 和 mmpose。","source":"ai","title":"安装脚本会再去装别的东西"},{"code":"dependencies_from_git_branches","severity":"info","evidence":"requirements.txt 直接从会变动的分支拉取代码：\"diffusers @ git+https://github.com/TMElyralab/diffusers.git@tme\"、\"ip_adapter @ git+https://github.com/tencent-ailab/IP-Adapter.git@main\"、\"clip @ git+https://github.com/openai/CLIP.git@main\"，因此装到的东西会随时间变化。","source":"ai","title":"依赖直接取自 git 分支"},{"code":"third_party_download_link","severity":"attention","evidence":"README.md 指向一个放在网盘上的非官方 Windows 整合包（\"One click integration package in windows\"，提取码 \"glut\"），同时写明：\"we have not verified, maintained, or updated third-party\"。","source":"ai","title":"指向第三方的下载链接"},{"code":"license_file_mismatch","severity":"info","evidence":"LICENSE 文件开头是 \"MIT License Copyright (c) 2024 Tencent Music Entertainment Group\"，后面又抄了内置依赖的 Apache 2.0 文本；README 则说 \"The code of MuseV is released under the MIT License\"。","source":"ai","title":"许可证文件自相矛盾"},{"code":"generated_people_videos","severity":"info","evidence":"README.md 展示了用真实人物和名人面孔生成的视频（例如蒙娜丽莎，以及一位指名的网红：\"Sun Xinying is a supermodel KOL\"），免责声明则称用户 \"are expected to comply with local laws\"，开发者对滥用不承担责任。","source":"ai","title":"会生成真人视频"},{"code":"project_quality_warnings_by_authors","severity":"info","evidence":"README.md 的 Limitation 一节：\"Lack of generalization ability\"、\"Watermarks may appear because of webvid\"、\"Understructured code... complex and unrefacted codes\"。","source":"ai","title":"作者自己也提醒了质量问题"}],"platforms":["linux","windows"],"tools":["python","docker","gpu_nvidia","ffmpeg"],"buildTools":["python","docker","gpu_nvidia","ffmpeg"],"disk":{"minMB":1610,"maxMB":6370,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"file"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true},{"slug":"viggle","name":"Viggle","confirmed":true}],"declaredAlternatives":[],"stars":2851,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":2849},{"date":"2026-11-08","stars":2849},{"date":"2026-11-09","stars":2850},{"date":"2026-11-10","stars":2851}],"firstPublishedAt":"2026-11-10T23:52:44.757Z","commitSha":"43370a6215afdfcd6d6af404350c132cd3b6eef8","verifiedAt":"2026-11-10T23:32:22.413Z","lastCommitAt":"2024-05-29T08:22:27Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a photo or a text description into a video of a person moving, made on your own computer with an NVIDIA card.","summary":"MuseV, from Tencent Music's Lyra Lab, makes videos of virtual people from a photo, a text prompt or another video, and can keep going for long clips. It is a research release: you run it by cloning the repository and typing commands, ideally inside the project's Docker image, on a machine with an NVIDIA card (the README mentions about 8 to 12 GB of video memory). Model files are downloaded separately from Hugging Face. The code is under the MIT license, but the trained models are stated to be for non-commercial research only.","easeRationale":"On Linux, the recommended path is pulling the Docker image, cloning the repository with submodules, setting PYTHONPATH by hand, downloading model files and running Python commands with many flags, so it is a D. Windows is the same route through Docker with an NVIDIA card; the README also links a third-party one-click Windows package but says third-party integrations are not verified or maintained by the authors.","costNote":"No payment to use the code; the README states the trained models are for non-commercial research purposes only.","notVerified":["Whether it still installs today: the last commit was in May 2024 and the pinned versions (torch 2.0.1, transformers 4.33.1) are old.","Real disk space and video memory needed after downloading the model files from Hugging Face.","Whether the Hugging Face demo space linked in the README is still online.","Whether the third-party Windows one-click package and ComfyUI node work or match this code.","Video quality for your own photos: the authors say results vary a lot with the image and the base model used."]},"zh-Hans":{"summaryShort":"把一张照片或一段文字描述变成人物动起来的视频，在自己装有 NVIDIA 显卡的电脑上生成。","summary":"MuseV 出自腾讯音乐天琴实验室，可以根据照片、文字提示或另一段视频生成虚拟人物视频，并能持续延长片段。这是一个研究性质的发布：使用方式是克隆仓库并敲命令，最好在项目提供的 Docker 镜像里运行，机器需配 NVIDIA 显卡（README 提到约 8 到 12 GB 显存）。模型文件要另外从 Hugging Face 下载。代码采用 MIT 许可，但训练好的模型声明仅供非商业研究使用。","easeRationale":"在 Linux 上，推荐流程是拉取 Docker 镜像、带子模块克隆仓库、手动设置 PYTHONPATH、下载模型文件，再运行带一堆参数的 Python 命令，所以给 D。Windows 同样走 Docker 加 NVIDIA 显卡这条路；README 还链接了一个第三方的 Windows 一键整合包，但声明第三方集成未经作者验证，也不由作者维护。","costNote":"使用代码无需付费；README 声明训练好的模型仅供非商业研究使用。","notVerified":["今天是否还能装上：最后一次提交是 2024 年 5 月，锁定的版本（torch 2.0.1、transformers 4.33.1）已经很旧。","从 Hugging Face 下载完模型文件后，实际需要的磁盘空间和显存。","README 里链接的 Hugging Face 演示空间是否还在线。","第三方的 Windows 一键整合包和 ComfyUI 节点是否可用、是否与这份代码对应。","用你自己的照片时的画质：作者说结果会因图片和所用基础模型差异很大。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMDA4OTIxNzY=","slug":"tomiinek__multilingual_text_to_speech","name":"Multilingual Text to Speech","owner":"Tomiinek","fullName":"Tomiinek/Multilingual_Text_to_Speech","type":"dev_tool","status":"active","grade":"E","easeByPlatform":{"linux":"E"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布是 2020-05-26（2359 天前）","source":"check","title":"发布早已停止"},{"code":"archived","severity":"info","evidence":"该仓库已归档（只读，不再维护）","source":"check","title":"仓库已归档"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：pillow: GHSA-22hg-r758-rrw9, GHSA-24rc-8hj2-v98c, GHSA-254f-3936-288h; torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-5pcm-hx3q-hm94","source":"check","title":"存在已知漏洞"},{"code":"voice_cloning_without_misuse_notice","severity":"info","evidence":"仓库描述和标签都在宣传 \"voice cloning\"（标签：voice-cloning、code-switching），README 还提供了一个 \"supporting code-switching or voice-cloning\" 的检查点，却没有任何关于本人同意或滥用他人声音的提示。","source":"ai","title":"可克隆声音却无滥用提醒"},{"code":"research_code_not_an_app","severity":"info","evidence":"README.md 围绕复现论文展开（\"source code, and parameters for the paper One Model, Many Languages\"）；使用章节一上来就是下载 CSS10 和 Common Voice 数据集并训练模型。","source":"ai","title":"这是研究代码，不是应用程序"},{"code":"old_pinned_dependencies","severity":"attention","evidence":"requirements.txt 锁定了 2019 至 2020 年的版本（torch==1.13.1、librosa==0.7.0、numba==0.48、tensorboard==2.0.2、phonemizer==2.1），在当前 Python 版本上通常装不上。","source":"ai","title":"锁定了过时的依赖"},{"code":"dataset_licenses_vary","severity":"info","evidence":"README.md：训练需要 CSS10 数据集（Apache License 2.0）和一份清洗过的 Common Voice 数据（Creative Commons CC0），都要在代码之外单独下载。","source":"ai","title":"数据集许可各不相同"}],"platforms":["linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":900,"maxMB":4300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true}],"declaredAlternatives":[],"stars":843,"downloads":14700,"downloadsWeek":19,"downloadsDays":2,"growth":0,"growthDays":6,"history":[{"date":"2026-11-08","stars":843},{"date":"2026-11-09","stars":843},{"date":"2026-11-10","stars":843}],"firstPublishedAt":"2026-11-10T23:52:44.774Z","commitSha":"2c75185337d26cc48750d669fd12f5f80777f4f1","verifiedAt":"2026-11-10T23:33:10.669Z","lastCommitAt":"2023-10-10T12:35:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that reads text aloud in several languages with one model, and can switch languages inside a sentence.","summary":"This is the code behind a 2020 research paper on multilingual text to speech (a Tacotron 2 model). It can train and run a single voice model for many languages, mix languages in one sentence and copy a speaker's voice. It is meant for researchers and developers: you clone the repository, install Python packages, download datasets and model files, and run scripts from the terminal. The repository is archived, so it gets no more fixes. The authors also link Google Colab notebooks and a page with audio samples.","easeRationale":"On Linux you clone the repository, install old pinned Python packages (torch 1.13.1, librosa 0.7.0, numba 0.48), download datasets and checkpoint files by hand and run training or synthesis scripts; there is no installer and no end-user app, so it is an E. The linked Colab notebooks are the only click-and-try option, and they are from 2020.","notVerified":["Whether the pinned 2020 packages still install on a current Python, and whether the code runs at all today.","Whether the Google Colab demo notebooks linked in the README still work.","How the generated voices actually sound compared with current text-to-speech services.","Whether synthesis works at a usable speed without a graphics card.","Real disk space needed once the datasets and the WaveRNN vocoder weights are downloaded."]},"zh-Hans":{"summaryShort":"研究代码：用同一个模型朗读多种语言的文本，还能在一句话里切换语言。","summary":"这是 2020 年一篇多语种文字转语音研究论文的配套代码（基于 Tacotron 2 模型）。它可以训练和运行一个覆盖多种语言的语音模型，在同一句话里混用不同语言，并复制某个人的声音。它面向研究者和开发者：克隆仓库、安装 Python 包、下载数据集和模型文件，再在终端里运行脚本。该仓库已归档，不会再有修复。作者还提供了 Google Colab 笔记本和一个音频示例页面的链接。","easeRationale":"在 Linux 上，你要克隆仓库，安装一批锁死的旧版 Python 包（torch 1.13.1、librosa 0.7.0、numba 0.48），手动下载数据集和检查点文件，再运行训练或合成脚本；既没有安装程序，也没有面向普通用户的应用，所以给 E。文中链接的 Colab 笔记本是唯一点开就能试的方式，而它们来自 2020 年。","notVerified":["2020 年锁定的这些包在当前 Python 上是否还能安装，代码今天是否还跑得起来。","README 中链接的 Google Colab 演示笔记本是否还能用。","与当下的文字转语音服务相比，生成的声音实际听感如何。","没有显卡时，合成速度是否还可接受。","下载完数据集和 WaveRNN 声码器权重后，实际需要多少磁盘空间。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2NzcwMjE4NA==","slug":"topjohnwu__magisk","name":"Magisk","owner":"topjohnwu","fullName":"topjohnwu/Magisk","type":"needs_setup","status":"active","grade":"D","easeByPlatform":{},"risk":"attention","signals":[{"code":"deep_system_access","severity":"attention","evidence":"README.MD：\"MagiskSU: Provide root access for applications\" 以及 \"Zygisk: Run code in every Android applications' processes\"；任何被授予 root 权限的应用都可以更改整部手机。","source":"ai","title":"可深入访问系统"},{"code":"unlocking_the_phone_has_consequences","severity":"attention","evidence":"文档中的安装方式（\"Installation Instruction\"）需要修补并刷入启动镜像，这要求手机的 bootloader 已解锁。在大多数手机上，解锁会清除所有数据，可能使保修失效，有些厂商甚至根本不允许解锁；刷入错误的启动镜像可能导致手机无法开机。","source":"ai","title":"解锁手机会带来后果"},{"code":"official_source_only","severity":"info","evidence":"README.MD：\"Github is the only source where you can get official Magisk information and downloads.\" 来自其他网站的副本可能已被改动。","source":"ai","title":"只认官方来源"},{"code":"android_only_no_desktop_app","severity":"info","evidence":"描述：\"The Magic Mask for Android\"；对发布版本的检查没有发现 Windows、Mac 或 Linux 的程序文件。","source":"ai","title":"仅限 Android，没有桌面应用"}],"platforms":[],"tools":["other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":63921,"downloads":173316260,"downloadsWeek":306671,"downloadsDays":7,"growth":182,"growthDays":7,"history":[{"date":"2026-10-26","stars":63516},{"date":"2026-10-27","stars":63539},{"date":"2026-10-28","stars":63567},{"date":"2026-10-29","stars":63598},{"date":"2026-10-30","stars":63631},{"date":"2026-10-31","stars":63655},{"date":"2026-11-01","stars":63689},{"date":"2026-11-02","stars":63714},{"date":"2026-11-03","stars":63739},{"date":"2026-11-04","stars":63765},{"date":"2026-11-05","stars":63798},{"date":"2026-11-06","stars":63825},{"date":"2026-11-07","stars":63850},{"date":"2026-11-08","stars":63875},{"date":"2026-11-09","stars":63896},{"date":"2026-11-10","stars":63921}],"firstPublishedAt":"2026-10-27T01:10:12.699Z","commitSha":"33bfff7ce750c27f157ddb4580e36bf842d6d95b","verifiedAt":"2026-10-28T01:42:25.225Z","lastCommitAt":"2026-11-10T13:58:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Gives you full control (root access) of an Android phone and lets you install add-ons that change the system.","summary":"Magisk is a long-running tool for customizing Android phones and tablets running Android 7.0 or newer. It gives chosen apps full administrator access to the phone (called root), lets you install modules that change system files without overwriting them, and includes tools for advanced users and developers. It is for experienced Android users. It runs on the phone, not on Windows, Mac or Linux. Installing it usually means unlocking the phone's bootloader, which can erase the phone, then patching and flashing a system file with a computer and command-line tools.","easeRationale":"Android is the only system it runs on, and there it is a D: there is no desktop program to install, you need a phone whose bootloader can be unlocked, the Magisk app on the phone, and Android command-line tools on a computer to flash a patched boot image.","costNote":"Open source under GPL-3.0.","notVerified":["Whether your specific phone model can be unlocked and patched","Whether banking, payment or streaming apps keep working on a rooted phone","Whether unlocking the bootloader voids your warranty or erases your data on your model","Risk of the phone failing to start if the patching step goes wrong","What a third-party Magisk module does once it is installed, since modules come from outside this repository"]},"zh-Hans":{"summaryShort":"让你获得 Android 手机的完整控制权(root 权限),并可以安装修改系统的插件。","summary":"Magisk 是一款历史悠久的工具,用于定制运行 Android 7.0 或更新版本的手机和平板。它让你选定的应用获得对手机的完整管理员权限(称为 root),允许安装在不覆盖系统文件的情况下修改系统的模块,并包含面向高级用户和开发者的工具。它面向有经验的 Android 用户。它运行在手机上,而不是 Windows、Mac 或 Linux 电脑上。安装它通常意味着先解锁手机的 bootloader(这可能会清空手机数据),然后用电脑和命令行工具修补并刷入系统文件。","easeRationale":"Android 是它唯一支持的系统,在该系统上评级为 D:没有可安装的桌面程序,你需要一台 bootloader 可解锁的手机、手机上的 Magisk 应用,以及电脑上的 Android 命令行工具来刷入修补后的 boot 镜像。","costNote":"基于 GPL-3.0 许可的开源项目。","notVerified":["你的具体手机型号能否解锁并修补","银行、支付或流媒体应用在已 root 的手机上能否继续正常使用","解锁 bootloader 是否会使你型号的保修失效或清空数据","修补过程出错导致手机无法开机的风险","第三方 Magisk 模块安装后实际会做什么,因为模块来自本仓库之外"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMzA2MjA2Mg==","slug":"torakiki__pdfsam","name":"PDFsam","owner":"torakiki","fullName":"torakiki/pdfsam","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A"},"risk":"low","signals":[],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":93,"maxMB":278,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["office","productivity"],"alternativeTo":[{"slug":"acrobat","name":"Acrobat","confirmed":true}],"declaredAlternatives":[],"stars":4665,"downloads":24624417,"downloadsWeek":201953,"downloadsDays":7,"growth":8,"growthDays":7,"history":[{"date":"2026-10-25","stars":4641},{"date":"2026-10-26","stars":4642},{"date":"2026-10-27","stars":4645},{"date":"2026-10-28","stars":4649},{"date":"2026-10-29","stars":4649},{"date":"2026-10-30","stars":4652},{"date":"2026-10-31","stars":4653},{"date":"2026-11-01","stars":4653},{"date":"2026-11-02","stars":4656},{"date":"2026-11-03","stars":4657},{"date":"2026-11-04","stars":4657},{"date":"2026-11-05","stars":4657},{"date":"2026-11-06","stars":4657},{"date":"2026-11-07","stars":4658},{"date":"2026-11-08","stars":4662},{"date":"2026-11-09","stars":4663},{"date":"2026-11-10","stars":4665}],"firstPublishedAt":"2026-10-25T23:19:31.763Z","commitSha":"4e4dd4dc5b14a1d328837c2b549fbe0b43cec83f","verifiedAt":"2026-10-27T00:49:36.422Z","lastCommitAt":"2026-11-04T12:09:05Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"pdfsam-basic-6.0.10-windows-x64.msi","sizeBytes":96768000,"url":"https://github.com/torakiki/pdfsam/releases/download/v6.0.10/pdfsam-basic-6.0.10-windows-x64.msi","releaseUrl":"https://github.com/torakiki/pdfsam/releases/tag/v6.0.10"}],"reviewedBy":"founder","text":{"en":{"summaryShort":"Split, merge, mix and rotate PDF files and pull out pages, all on your own computer.","summary":"PDFsam Basic is a desktop program for everyday PDF page tasks: joining several PDFs into one, splitting a PDF into parts, mixing pages from two files, rotating pages and extracting selected pages. It does not edit the text inside a PDF. It runs on Windows, macOS and Linux, and since version 4 it comes with its own Java built in, so nothing else needs to be installed.","easeRationale":"The releases include a Windows installer (pdfsam-basic-6.0.6-windows-x64.msi, about 93 MB) that bundles its own Java, so you download it and open it.","costNote":"PDFsam Basic is free and open source under the AGPL-3.0 license.","notVerified":["Whether it handles very large or password-protected PDFs well","Whether the app connects to the internet, for example to check for updates or show news","Real disk use after installation"]},"zh-Hans":{"summaryShort":"拆分、合并、混合和旋转 PDF 文件并提取页面，全部在你自己的电脑上完成。","summary":"PDFsam Basic 是一款桌面程序，用于日常的 PDF 页面操作：把多个 PDF 合并成一个，把一个 PDF 拆分成几部分，混合两个文件中的页面，旋转页面，以及提取选定的页面。它不能编辑 PDF 内部的文字。它可在 Windows、macOS 和 Linux 上运行，并且从第 4 版起自带 Java，因此无需再安装其他东西。","easeRationale":"发布版本包含 Windows 安装程序（pdfsam-basic-6.0.6-windows-x64.msi，约 93 MB），内置自带的 Java，下载后打开即可。","costNote":"PDFsam Basic 免费并开源，采用 AGPL-3.0 许可证。","notVerified":["它能否很好地处理超大或有密码保护的 PDF","该应用是否会连接互联网，例如检查更新或显示新闻","安装后的实际磁盘占用"]}}},{"id":"R_kgDOQ4pdCg","slug":"tracer-cloud__opensre","name":"OpenSRE","owner":"Tracer-Cloud","fullName":"Tracer-Cloud/opensre","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把远程脚本直接灌进 shell：curl -fsSL https://install.opensre.com | bash","source":"check","title":"安装时直接下载并执行脚本"},{"code":"account_required_for_self_hosted_tool","severity":"attention","evidence":"README.md：\"Run opensre to sign in and get started. The first launch activates the hosted model.\" 以及 \"the shell only opens for an active account\"——它自称自托管，但不登录厂商服务就打不开。","source":"ai","title":"号称自托管却必须注册账号"},{"code":"telemetry_on_by_default","severity":"attention","evidence":"README.md 的 Telemetry 一节：\"First-party product analytics and Sentry (errors) are opt-out\"，只有设置 OPENSRE_NO_TELEMETRY=1 才会关闭。","source":"ai","title":"默认开启数据上报"},{"code":"installs_unreleased_main_build_by_default","severity":"attention","evidence":"install.sh 里写着 INSTALL_CHANNEL=\"${OPENSRE_INSTALL_CHANNEL:-main}\"，install.ps1 也默认使用 \"main\" 通道，因此安装命令给你的是主分支上的最新构建，而不是经过审核的正式版本；Dockerfile 也说 \"On start the supervisor installs the current main-channel binary\"。","source":"ai","title":"默认安装未发布的 main 分支构建"},{"code":"broad_production_credentials","severity":"attention","evidence":".env.example 收集了 AWS、Kubernetes、Datadog、Grafana、Splunk、Sentry、GitHub、GitLab、Slack、PagerDuty、Twilio 等数十项服务的令牌，而 README 说智能体会 \"Fixes the cause and pushes it to the PR branch\"，还可以 \"optionally, execute remediation actions\"。","source":"ai","title":"需要大量生产环境凭据"},{"code":"installer_also_installs_other_software","severity":"info","evidence":"install.sh 在缺少 GitHub CLI 时会自行安装（\"Installing GitHub CLI (gh) for OpenSRE GitHub tools\"，通过 Homebrew 或下载发布包），除非设置了 OPENSRE_SKIP_GH_INSTALL。","source":"ai","title":"安装脚本还会装别的软件"},{"code":"coding_agent_without_sandbox_in_container","severity":"info","evidence":"Dockerfile 设置了 ENV CODING_AGENT_SANDBOX=host，并注释说 \"the coding agent's own sandbox cannot start; this task is the isolation boundary and the agent gets the whole process\"。","source":"ai","title":"容器里的编码智能体没有沙盒"},{"code":"public_alpha","severity":"info","evidence":"README.md：\"Public Alpha: Core workflows are usable for early exploration, though not yet fully stable... APIs and integrations may evolve.\"","source":"ai","title":"公开 Alpha 阶段"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","docker"],"disk":{"minMB":117,"maxMB":350,"basis":"installer"},"apiKeys":["OpenSRE account sign-in (hosted model activated on first launch)","One model provider: Anthropic, OpenAI, Google Gemini, OpenRouter, DeepSeek, Groq, NVIDIA, MiniMax, Azure OpenAI, Bedrock or local Ollama","GitHub or GitLab token (to read runs and push fixes)","Observability services: Grafana, Datadog, Honeycomb, Sentry, Splunk, New Relic and others","Chat and incident services: Slack, Telegram, Discord, PagerDuty, Jira, incident.io","Cloud and database services: AWS, Azure, Vercel, Railway, Neon"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"resolve-ai","name":"Resolve.ai","confirmed":true},{"slug":null,"name":"Datadog","confirmed":false},{"slug":"github-copilot","name":"Copilot","confirmed":false},{"slug":"claude-code","name":"Claude Code","confirmed":false}],"declaredAlternatives":[],"stars":12805,"downloads":20190,"downloadsWeek":178,"downloadsDays":2,"growth":97,"growthDays":6,"history":[{"date":"2026-11-08","stars":12764},{"date":"2026-11-09","stars":12793},{"date":"2026-11-10","stars":12805}],"firstPublishedAt":"2026-11-10T23:53:36.492Z","commitSha":"3bd3ac9100b93718f444303bd2a22ef1cf6300ae","verifiedAt":"2026-11-10T23:34:57.378Z","lastCommitAt":"2026-11-10T23:16:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A terminal assistant that looks into why your build or service broke, explains it and can push a fix for review.","summary":"OpenSRE is a command-line AI agent for teams that run software in production. When a build fails or an alert fires, it reads the logs, checks recent commits, tests ideas and answers with evidence; it can also push a suggested fix to the pull request branch and post a summary to Slack or PagerDuty. It connects to 60+ services (Grafana, Datadog, Kubernetes, AWS, GitHub). You install it from the terminal, sign in to an account and pick a model: the hosted one, your own key or local Ollama. It calls itself public alpha.","easeRationale":"On macOS and Linux you paste one install command in the terminal (curl ... | bash), then run opensre, sign in and connect at least one service with opensre integrations setup; Windows uses the same route with a PowerShell one-liner and has its own extra instructions page, so all three are a C. Ready archives for Windows, macOS and Linux are published on every release, so nothing has to be compiled.","costNote":"The code is Apache 2.0, but the shell only opens for an active account and the agent needs a model: either the hosted one activated at first launch or your own provider key (a local Ollama model is listed as an option). No prices are stated in the repository.","notVerified":["What the hosted model costs after the first launch, and what an account includes.","How much you will spend on model calls when the agent investigates a real failure.","What the opt-out analytics and error reports actually send.","Whether the suggested fixes it pushes are correct, and how often a person has to undo them.","How well the Windows build works compared with macOS and Linux.","Whether the many listed integrations all work, since the project is in public alpha."]},"zh-Hans":{"summaryShort":"一个终端助手，帮你查清构建或服务为什么出故障，解释原因，还能把修复推上去待审。","summary":"OpenSRE 是一个命令行 AI 智能体，面向在生产环境里运维软件的团队。当构建失败或告警触发时，它会读日志、查看近期提交、验证假设，并给出有依据的回答；它还能把建议的修复推到 pull request 分支，并把摘要发到 Slack 或 PagerDuty。它可以对接 60 多个服务（Grafana、Datadog、Kubernetes、AWS、GitHub）。你在终端里安装，登录账号，再选一个模型：官方托管的模型、你自己的密钥，或本地 Ollama。项目自称处于公开 Alpha 阶段。","easeRationale":"在 macOS 和 Linux 上，你在终端里粘贴一条安装命令（curl ... | bash），然后运行 opensre 登录，并用 opensre integrations setup 接入至少一个服务；Windows 走同样的路线，只是换成一行 PowerShell，并另有一页额外说明，因此三者都是 C。每次发布都会提供 Windows、macOS 和 Linux 的现成压缩包，不需要自己编译。","costNote":"代码采用 Apache 2.0，但只有账号处于有效状态才能打开控制台，而且智能体需要一个模型：首次启动时激活的托管模型，或你自己服务商的密钥（本地 Ollama 模型也列为可选）。仓库里没有给出任何价格。","notVerified":["首次启动之后，托管模型要收多少钱，一个账号又包含哪些内容。","当智能体去排查一次真实故障时，模型调用会花掉多少钱。","默认开启的使用统计和错误上报实际发送了哪些内容。","它推送的修复是否正确，以及需要人工撤销的频率有多高。","Windows 版本相比 macOS 和 Linux 运行得如何。","列出的众多集成是否都可用，因为项目还处在公开 Alpha 阶段。"]}}},{"id":"R_kgDORz9IfA","slug":"tradesdontlie__tradingview-mcp","name":"TradingView MCP","owner":"tradesdontlie","fullName":"tradesdontlie/tradingview-mcp","type":"dev_tool","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"may_conflict_with_tradingview_terms","severity":"attention","evidence":"README.md 的 Disclaimer：\"TradingView's Terms of Use restrict automated data collection, scraping, and non-display usage of their platform and data. This tool uses Chrome DevTools Protocol to programmatically interact with the TradingView Desktop app, which may conflict with those terms\"，以及 \"The authors are not responsible for any account bans, suspensions, legal actions\"。","source":"ai","title":"可能与 TradingView 的条款相冲突"},{"code":"opens_local_debug_port","severity":"attention","evidence":"要使用它，必须带 \"--remote-debugging-port=9222\" 启动 TradingView；只要这个端口开着，你机器上运行的任何程序都能通过它操控该应用，而不只是这个工具。","source":"ai","title":"会开启本地调试端口"},{"code":"uses_undocumented_internal_interfaces","severity":"info","evidence":"README.md 的 CAUTION：\"This tool accesses undocumented internal TradingView APIs via the Electron debug interface. These can change or break without notice in any TradingView update.\"","source":"ai","title":"调用未公开的内部接口"},{"code":"assistant_can_run_code_in_the_app","severity":"info","evidence":"工具列表里有 \"ui_open_panel / ui_click / ui_evaluate\" 这类界面自动化命令，命令行也提供 \"tv ui click/keyboard/hover/scroll/find/eval\"，也就是说助手可以在你的 TradingView 窗口里点击并执行代码。","source":"ai","title":"助手可以在应用内执行代码"},{"code":"install_by_pasting_a_prompt","severity":"info","evidence":"README.md 的 \"Install with Claude Code\" 让你粘贴一段提示词，指示助手去克隆仓库、运行 npm install，并替你修改 ~/.claude/.mcp.json。","source":"ai","title":"安装方式是粘贴一段提示词"},{"code":"automated_use_restricted_by_platform_terms","severity":"high","evidence":"README.md 的 Disclaimer：\"TradingView's Terms of Use restrict automated data collection, scraping, and non-display usage of their platform and data. This tool uses Chrome DevTools Protocol to programmatically interact with the TradingView Desktop app, which may conflict with those terms\"，以及 \"You assume all risk associated with using this tool. The authors are not responsible for any account bans, suspensions, legal actions, or other consequences resulting from its use.\" 同一份 README 也声明该工具 \"does not bypass or circumvent any TradingView paywall or access control\"，并且 \"Requires a valid TradingView subscription.\"","source":"ai","title":"平台条款限制自动化使用"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["ai_skills","ai_agents","development"],"alternativeTo":[{"slug":"tradingview","name":"TradingView","confirmed":false}],"declaredAlternatives":[{"target":"tradingview","stance":"claims","evidence":"Replace `/path/to/tradingview-mcp` with your actual path"}],"stars":7150,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":68,"growthDays":6,"history":[{"date":"2026-11-05","stars":7088},{"date":"2026-11-06","stars":7095},{"date":"2026-11-07","stars":7110},{"date":"2026-11-08","stars":7125},{"date":"2026-11-09","stars":7134},{"date":"2026-11-10","stars":7150}],"firstPublishedAt":"2026-11-10T23:52:44.800Z","commitSha":"c05b8f5755ed8e64ea242de88ddbf46aa24d56a4","verifiedAt":"2026-11-10T23:35:46.668Z","lastCommitAt":"2026-07-28T17:28:37Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets Claude Code read and drive the TradingView Desktop app on your computer: charts, indicators and Pine Script.","summary":"This is an MCP server that plugs the TradingView Desktop app into Claude Code, so you can ask an AI assistant to read your chart, change symbol or timeframe, add indicators, draw lines, manage alerts, take screenshots and write Pine Script. It talks only to the app running on your own machine, through the Chrome debugging port you have to switch on yourself. It does not replace TradingView: you still need the Desktop app and, for real-time data, a paid subscription. You install it with Node.js, git clone and npm install, then point your MCP config at it.","easeRationale":"On all three systems you install Node.js 18+, clone the repository, run npm install, edit your Claude Code MCP config by hand and start TradingView with the --remote-debugging-port=9222 flag (there is a launch script for Mac, Windows and Linux), so it is a C everywhere. There are no releases or installers to download.","costNote":"The code is free (MIT). The README says it \"Requires a valid TradingView subscription\" and the TradingView Desktop app, plus an assistant that supports MCP such as Claude Code.","notVerified":["Whether it still works with the current TradingView Desktop version, since it depends on internal parts of the app.","Whether using it puts your TradingView account at risk; the repository says you take that risk.","Whether the replay trading commands really stay inside TradingView's practice mode and never place a real order.","The exact number of tools: the README says 78 in one place and 84 in another.","How much of your assistant's context (and cost) a chart analysis session uses in practice."]},"zh-Hans":{"summaryShort":"让 Claude Code 读取并操控你电脑上的 TradingView 桌面版：图表、指标和 Pine Script。","summary":"这是一个 MCP 服务器，把 TradingView 桌面应用接入 Claude Code，于是你可以让 AI 助手读你的图表、切换品种或周期、添加指标、画线、管理提醒、截图并编写 Pine Script。它只与运行在你自己电脑上的那个应用通信，通过你手动开启的 Chrome 调试端口。它并不能取代 TradingView：你仍然需要桌面版应用，要看实时行情还得有付费订阅。安装需要 Node.js，git clone 后执行 npm install，再把 MCP 配置指向它。","easeRationale":"三个系统都一样：安装 Node.js 18 以上版本，克隆仓库，运行 npm install，手动修改 Claude Code 的 MCP 配置，并带上 --remote-debugging-port=9222 参数启动 TradingView（Mac、Windows 和 Linux 都有启动脚本），所以统一给 C。项目没有可下载的发布版或安装程序。","costNote":"代码免费（MIT）。README 写明 \"Requires a valid TradingView subscription\"，还需要 TradingView 桌面应用，以及支持 MCP 的助手（例如 Claude Code）。","notVerified":["它在当前版本的 TradingView 桌面版上是否还能用，因为它依赖应用的内部机制。","使用它是否会让你的 TradingView 账号面临风险；仓库明说这个风险由你自己承担。","回放交易相关命令是否真的只停留在 TradingView 的模拟模式里，绝不会下真实订单。","工具的确切数量：README 一处写 78 个，另一处写 84 个。","一次图表分析会话在实际使用中会消耗助手多少上下文（以及多少费用）。"]}}},{"id":"R_kgDOSDCHbQ","slug":"tt-a1i__archify","name":"Archify","owner":"tt-a1i","fullName":"tt-a1i/archify","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"readme_sponsor_links_to_model_resellers_with_referral_codes","severity":"attention","evidence":"README 的 Sponsors 表格推广了几项第三方服务，它们以远低于标价的价格转售 Claude、OpenAI 和 Gemini 模型的访问权限，每一项都通过带有该项目自己推荐码的链接：OpenLux（`openlux.ai/register?channel=c_qdvanbpc`，'as low as 8.82%, 4% and 8% of official rates'）、PackyCode（`packyapi.ai/register?aff=bN97`）、AICodeMirror（`aicodemirror.ai/register?invitecode=WC26FL`，'Codex pricing starts at 7% of standard rates'）和 Infistar（`infistar.cc/register?aff=9N89LEMV`，'starts at 1% of official rates'）。这些是其他公司的付费转售服务，不属于 Archify。","source":"ai","title":"README 链接到带推荐码的模型转售服务"},{"code":"optional_update_check_contacts_the_authors_server","severity":"info","evidence":"README 快速开始：'Archify may GET the fixed stable manifest solely to show an optional reminder; it never downloads or installs updates. (...) The server sees normal HTTP metadata (IP and time), but receives no version, Agent, project data, prompts, account/device ID, or ETag. (...) Set ARCHIFY_UPDATE_CHECK_DISABLED=1 to disable networking and reminder-state writes.'","source":"ai","title":"可选的更新检查会联系作者的服务器"},{"code":"zip_archive_committed_in_the_repository","severity":"info","evidence":"顶层文件列表中包含 `archify.zip`，README 要求你把它上传到 Claude.ai Settings → Capabilities → Skills。在仓库的文件列表里，无法以纯文本方式查看这个打包文件的内容。","source":"ai","title":"仓库中直接提交了 zip 压缩包"},{"code":"skill_tells_the_assistant_to_read_your_repository","severity":"info","evidence":"README：'Analyze this repository, then use archify to create a high-level runtime architecture diagram' 以及 'Evidence-backed Architecture nodes mark themselves SRC n and open Git-verified files and line ranges pinned to one public commit'。有源码依据的图表意味着助手会读取你的代码路径和文件片段，这些内容可能最终出现在你分享的 HTML 里。","source":"ai","title":"该 skill 会让助手读取你的仓库"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","development","productivity"],"alternativeTo":[{"slug":"lucidchart","name":"Lucidchart","confirmed":true},{"slug":"miro","name":"Miro","confirmed":true}],"declaredAlternatives":[],"stars":95355,"downloads":41774,"downloadsWeek":3645,"downloadsDays":7,"growth":2225,"growthDays":7,"history":[{"date":"2026-10-29","stars":91440},{"date":"2026-10-30","stars":91837},{"date":"2026-10-31","stars":92154},{"date":"2026-11-01","stars":92409},{"date":"2026-11-02","stars":92736},{"date":"2026-11-03","stars":93130},{"date":"2026-11-04","stars":93427},{"date":"2026-11-05","stars":93825},{"date":"2026-11-06","stars":94177},{"date":"2026-11-07","stars":94437},{"date":"2026-11-08","stars":94707},{"date":"2026-11-09","stars":95028},{"date":"2026-11-10","stars":95355}],"firstPublishedAt":"2026-10-29T01:54:21.485Z","commitSha":"d6059331abbddb49bca380ff46dca8ce33158b04","verifiedAt":"2026-10-29T01:38:42.066Z","lastCommitAt":"2026-11-10T01:32:12Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Ask your AI coding assistant to turn an idea, a plan or a codebase into an interactive diagram saved as one HTML file.","summary":"Archify is a skill you add to an AI coding assistant such as Claude Code, Codex CLI, Cursor or OpenCode. You describe a system or a plan, or point it at a repository, and the assistant writes a diagram that comes out as one interactive HTML file you can open in a browser, explore, export as an image and send to others. Ten diagram shapes are covered, from architecture and sequence to timelines and database schemas. You need Node.js 18 or newer and one of those assistants. The README says it is not a general-purpose drawing editor, so it does not replace free-form tools like Lucidchart.","easeRationale":"Same work on Windows, macOS and Linux: install Node.js 18 or newer plus a supported AI assistant, then run one command in the terminal (`npx skills add tt-a1i/archify -g`) and ask the assistant for a diagram. There is no installer and no ready binary; the 23 releases carry no platform files.","costNote":"The project is free under Apache-2.0. It only works inside an AI coding assistant, which normally needs a paid plan or paid API use. The README also links a WeChat Pay and an Alipay QR code for donations.","notVerified":["Whether the generated diagrams are accurate for a real codebase, since the assistant writes them","How much AI usage, and therefore cost, one diagram or an explainer page consumes","Whether the Claude.ai upload path works, since the README says it 'depends on Node.js access in the sandbox'","What is inside the committed archify.zip compared with the source folders","Whether the optional update check really sends no project data, as the README states","The real disk use after installing the skill and its viewer assets"]},"zh-Hans":{"summaryShort":"让你的 AI 编程助手把一个想法、一份计划或一份代码库变成一个交互式图表,保存为单个 HTML 文件。","summary":"Archify 是一个可以添加到 AI 编程助手(如 Claude Code、Codex CLI、Cursor 或 OpenCode)中的技能。你描述一个系统或计划,或者指向一个代码仓库,助手就会生成一个图表,输出为一个可在浏览器中打开、浏览、导出为图片并发送给他人的交互式 HTML 文件。它支持十种图表形式,从架构图、序列图到时间线和数据库结构图。你需要 Node.js 18 或更新版本,以及上述助手之一。README 说明它不是通用的自由绘图编辑器,因此不能取代 Lucidchart 这类工具。","easeRationale":"在 Windows、macOS 和 Linux 上流程相同:安装 Node.js 18 或更新版本及一个受支持的 AI 助手,然后在终端运行一条命令(npx skills add tt-a1i/archify -g),再向助手请求生成图表。没有安装程序,也没有现成的二进制文件;23 个发布版本都不包含平台文件。","costNote":"该项目基于 Apache-2.0 协议免费提供。它只能在 AI 编程助手内使用,通常需要付费计划或付费的 API 调用。README 中还附有用于捐赠的微信支付和支付宝二维码。","notVerified":["生成的图表对真实代码库是否准确,因为图表是由助手编写的","生成一个图表或一页说明会消耗多少 AI 用量,从而产生多少费用","上传到 Claude.ai 的路径是否可用,因为 README 说这'取决于沙盒中 Node.js 的访问权限'","仓库中提交的 archify.zip 内容与源码文件夹相比有什么不同","可选的更新检查是否真的不发送任何项目数据,正如 README 所说","安装该技能及其查看器资源后,实际占用的磁盘空间"]}}},{"id":"R_kgDOVSMwTA","slug":"udhaykumarbala__cleanyourtv-app","name":"Clean Your TV","owner":"udhaykumarbala","fullName":"udhaykumarbala/cleanyourtv-app","type":"ready_app","status":"active","grade":"B","easeByPlatform":{},"risk":"attention","signals":[{"code":"opens_tv_network_debugging","severity":"attention","evidence":"README，Supported TVs：\"the TV needs ADB over the network (port 5555)\"。开启期间，同一 Wi-Fi 下的任何人都可以向电视发送命令。README 说明调试 \"is switched off again at the end unless you keep it on\"。","source":"ai","title":"开启电视的网络调试"},{"code":"very_new_and_in_flux","severity":"attention","evidence":"项目信息：创建于 2026-10-21，已发布 17 个版本，最近的三个版本是 v0.4.6-beta3、v0.4.6-beta2 和 v0.4.5，日期都是 2026-10-27。有 1274 个未关闭的 issue，而星标为 910 个。","source":"ai","title":"非常新，变动频繁"},{"code":"installed_outside_the_app_store","severity":"info","evidence":"README 唯一的安装方式是从 GitHub 发布页面 \"Download the APK\"，或使用 Obtainium。Android 会要求你允许安装来自未知来源的应用，而且没有任何商店的审核。每个版本都由同一个密钥签名，证书 SHA-256 为 eaf7af3867bb29ff4bb038854842c93f42ffe4198a087e611cb18f925201cd65。","source":"ai","title":"在应用商店之外安装"},{"code":"changes_tv_system_software","severity":"info","evidence":"该应用会为主用户停用电视上预装的应用，并可以替换主屏幕启动器；README：\"accessibility, casting, voice and updates go off only if you choose\"，在某些型号上它只做 \"clean-up only\"，因为更换主屏幕 \"would cost casting\"。","source":"ai","title":"更改电视的系统软件"},{"code":"written_mostly_by_ai_agents","severity":"info","evidence":"README，For developers：\"The app is developed with AI coding agents under human review: the maintainer set the design and reviewed and tested the app, and every change went through automated tests and review.\"","source":"ai","title":"主要由 AI 智能体编写"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":1930,"downloads":381759,"downloadsWeek":100368,"downloadsDays":7,"growth":382,"growthDays":7,"history":[{"date":"2026-10-28","stars":1011},{"date":"2026-10-29","stars":1105},{"date":"2026-10-30","stars":1184},{"date":"2026-10-31","stars":1278},{"date":"2026-11-01","stars":1371},{"date":"2026-11-02","stars":1463},{"date":"2026-11-03","stars":1548},{"date":"2026-11-04","stars":1605},{"date":"2026-11-05","stars":1670},{"date":"2026-11-06","stars":1744},{"date":"2026-11-07","stars":1786},{"date":"2026-11-08","stars":1830},{"date":"2026-11-09","stars":1886},{"date":"2026-11-10","stars":1930}],"firstPublishedAt":"2026-10-28T01:50:06.062Z","commitSha":"f58c8f7c5a6d8a85f0c4214a4beeb704142ab33e","verifiedAt":"2026-10-28T01:39:47.647Z","lastCommitAt":"2026-11-03T10:20:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android phone app that switches off ads and preinstalled bloat on an Android TV over Wi-Fi, with no computer.","summary":"You install this app on an Android phone (Android 9 or newer) on the same Wi-Fi as the TV, and it connects to the TV over network debugging. It reads the TV, shows a plan of what it would switch off — adverts, tracking, preinstalled apps, brand extras — and after you agree it disables them for the main user and can put the Clean Your TV home screen in place. Nothing is uninstalled, and a restore point can put things back. It uses community lists covering 173 TV models from 47 brands, and it is a community app built on Gathin23/cleanyourtv.","easeRationale":"Android is the only system it runs on, and there it is a B: the release page has a ready APK you install by hand on the phone, and the one extra step is switching on network debugging on the TV with the remote and allowing the phone. Some sets only offer wireless debugging with a pairing code, and those need a computer once for three commands. Computers are not a target of this app at all.","notVerified":["Whether it recognises and cleans your exact TV model, or falls back to the careful brand list","Whether undo really brings every app and the original home screen back","Whether network debugging on the TV is actually closed again when it finishes","Whether the published APK matches the public source, as the reproducible-build claim says","Whether the TV keeps working normally after its own next system update"]},"zh-Hans":{"summaryShort":"一款 Android 手机应用,通过 Wi-Fi 关闭 Android TV 上的广告和预装软件,无需电脑。","summary":"你在一台 Android 手机(Android 9 或更新版本)上安装这个应用,并让它与电视处于同一个 Wi-Fi,它通过网络调试连接到电视。它会读取电视的信息,显示一份将要关闭的内容清单——广告、跟踪、预装应用、品牌附加项——在你确认后,它会为主用户关闭这些内容,并可以替换成 Clean Your TV 的主屏幕。没有任何东西会被卸载,恢复点可以把一切还原。它使用社区整理的列表,覆盖 47 个品牌的 173 款电视型号,是基于 Gathin23/cleanyourtv 构建的社区应用。","easeRationale":"Android 是它唯一支持的系统,在该系统上评级为 B:发布页面提供了现成的 APK,可手动安装到手机上,唯一的额外步骤是用遥控器在电视上打开网络调试并授权手机。部分电视只提供带配对码的无线调试,这类情况需要用电脑执行一次性的三条命令。电脑完全不是这个应用的目标平台。","notVerified":["它是否能识别并清理你确切的电视型号,还是会退回到更谨慎的品牌通用列表","撤销操作是否真的能恢复所有应用和原来的主屏幕","电视上的网络调试在完成后是否真的会重新关闭","发布的 APK 是否与公开源代码一致,正如可复现构建所声称的那样","电视在自身下一次系统更新后是否仍能正常工作"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyNjQ4MTg2ODY=","slug":"ultralytics__yolov5","name":"YOLOv5","owner":"ultralytics","fullName":"ultralytics/yolov5","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布是在 2022-11-22（1437 天前）","source":"check","title":"发布版本已过时"},{"code":"code_downloaded_at_run_time","severity":"info","evidence":"README 快速开始使用 torch.hub.load(\"ultralytics/yolov5\", \"yolov5s\") 加载模型，运行时会从 GitHub 获取仓库代码；README 还写道 \"Models are automatically downloaded from the latest YOLOv5 release\"。","source":"ai","title":"运行时下载代码"},{"code":"superseded_by_newer_package","severity":"info","evidence":"README：\"the actively maintained `ultralytics` package brings the latest Ultralytics YOLO models together in one place\"；requirements.txt 也依赖 \"ultralytics>=8.4.164\"。","source":"ai","title":"已被更新的软件包取代"},{"code":"commercial_license_upsell","severity":"info","evidence":"README：\"To request an Enterprise License, please complete the form at Ultralytics Licensing\"（代码本身采用 AGPL-3.0，该许可证要求你公开自己所做的修改，即使程序只是通过网络被访问也一样）。","source":"ai","title":"推销商业许可证"},{"code":"datasets_downloaded_automatically","severity":"info","evidence":"README 的训练部分：\"Both models and datasets are downloaded automatically from the latest YOLOv5 release\"，验证示例中也提到了类似 \"Download ImageNet validation split (6.3GB, 50,000 images)\" 的下载。这些命令可能不再询问就下载数 GB 的数据。","source":"ai","title":"自动下载数据集"}],"platforms":["windows","mac","linux"],"tools":["python","gpu_any"],"buildTools":["python","gpu_any"],"disk":{"minMB":1460,"maxMB":5920,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image","video","development"],"alternativeTo":[{"slug":"roboflow","name":"roboflow","confirmed":true}],"declaredAlternatives":[],"stars":58260,"downloads":135223295,"downloadsWeek":82388,"downloadsDays":7,"growth":10,"growthDays":7,"history":[{"date":"2026-10-26","stars":58227},{"date":"2026-10-27","stars":58231},{"date":"2026-10-28","stars":58233},{"date":"2026-10-29","stars":58238},{"date":"2026-10-30","stars":58247},{"date":"2026-10-31","stars":58247},{"date":"2026-11-01","stars":58249},{"date":"2026-11-02","stars":58246},{"date":"2026-11-03","stars":58250},{"date":"2026-11-04","stars":58252},{"date":"2026-11-05","stars":58255},{"date":"2026-11-06","stars":58254},{"date":"2026-11-07","stars":58255},{"date":"2026-11-08","stars":58254},{"date":"2026-11-09","stars":58257},{"date":"2026-11-10","stars":58260}],"firstPublishedAt":"2026-10-27T01:10:12.699Z","commitSha":"af08ba156fd58ed5b0fe7e4e58a22c72c3cdd0ce","verifiedAt":"2026-10-30T01:43:47.005Z","lastCommitAt":"2026-11-07T16:59:42Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Teach a computer to find and label objects (people, cars, animals) in photos, videos or a webcam feed.","summary":"YOLOv5 is an image recognition model and set of scripts from the company Ultralytics. It can spot and outline objects in pictures, videos, screen captures and live streams, sort images into categories, and be trained on your own photos. It is aimed at developers, students and researchers. You need Python 3.8 or newer and PyTorch, and you run it by typing commands. Training your own model realistically needs a graphics card. The README itself points new projects to the newer \"ultralytics\" package.","easeRationale":"The three systems are equally hard and all sit at C, with Linux the best documented (pyproject.toml lists Linux, macOS and Windows, and the only release file is a Linux one): there is no installer, so you install Python and PyTorch, clone the code, run pip install -r requirements.txt and then start detect.py or train.py from a terminal. Training on your own photos also needs a graphics card in practice.","costNote":"Free under the AGPL-3.0 license. Ultralytics sells an Enterprise License for companies that cannot follow the AGPL terms.","notVerified":["Whether the install and scripts work on Windows without errors.","Real disk use after installing PyTorch (with or without graphics card support) and downloading model files.","How fast detection runs on a normal computer without a graphics card.","Which files the scripts download from the internet at run time besides the model weights.","Whether the code still matches the last release, which dates from 2022, since commits continue."]},"zh-Hans":{"summaryShort":"教电脑在照片、视频或摄像头画面中找到并标注物体（人、汽车、动物）。","summary":"YOLOv5 是 Ultralytics 公司开发的一个图像识别模型及配套脚本。它可以在图片、视频、屏幕截图和实时画面中找出并框出物体，把图片分类，还可以用你自己的照片来训练。它主要面向开发者、学生和研究人员。你需要 Python 3.8 或更高版本以及 PyTorch，并通过输入命令来使用它。训练自己的模型在实际使用中通常需要一块显卡。README 本身也建议新项目改用更新的“ultralytics”包。","easeRationale":"三个系统的难度相同，评级都是 C，其中 Linux 的文档最完整（pyproject.toml 中列出了 Linux、macOS 和 Windows，而唯一发布的版本文件是 Linux 版）：没有安装程序，你需要先安装 Python 和 PyTorch，克隆代码，运行 pip install -r requirements.txt，然后在终端里启动 detect.py 或 train.py。实际使用中，用自己的照片训练模型也需要一块显卡。","costNote":"基于 AGPL-3.0 许可证免费提供。对于无法遵守 AGPL 条款的公司，Ultralytics 另外出售企业许可证。","notVerified":["安装过程和脚本在 Windows 上是否能无错误运行。","安装 PyTorch（无论是否带显卡支持）并下载模型文件之后，实际占用的磁盘空间。","在没有显卡的普通电脑上，检测速度有多快。","脚本在运行时除了模型权重之外，还会从网上下载哪些文件。","由于代码仍在不断提交更新，目前的代码是否还和 2022 年发布的最后一个版本一致。"]}}},{"id":"R_kgDOR1kXog","slug":"ultraworkers__claw-code","name":"Claw Code","owner":"ultraworkers","fullName":"ultraworkers/claw-code","type":"dev_tool","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"unusual_star_rate","severity":"attention","evidence":"215 天内获得 194992 个星标（约每天 907 个）","source":"check","title":"星标增长速度异常"},{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把一个远程脚本直接交给 shell 执行：curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh","source":"check","title":"把远程脚本直接交给 shell 执行"},{"code":"self_described_not_production","severity":"attention","evidence":"README.md：\"Claw Code is not the serious production project here. This repository is closer to a museum exhibit than a product pitch\"；仓库描述：\"developed and maintained with no human intervention\"。","source":"ai","title":"自称不是生产级项目"},{"code":"possible_reproduction_of_proprietary_product","severity":"attention","evidence":"README.md：\"This repository does not claim ownership of the original Claude Code source material\" 和 \"not affiliated with, endorsed by, or maintained by Anthropic\"；PARITY.md 被描述为 \"Rust-port parity status\"。","source":"ai","title":"可能复刻了一款专有产品"},{"code":"unusual_fork_count","severity":"info","evidence":"一个创建于 2026-03-31 的仓库，194992 个星标对应 107597 个 fork（约 55%），却只有 1 个未解决的问题（issue）。","source":"ai","title":"fork 数量异常"},{"code":"readme_redirects_to_other_projects","severity":"info","evidence":"README.md：\"If you want to actually run work, start with LazyCodex or Gajae-Code\"，并带有链接到这些仓库的大幅横幅。","source":"ai","title":"README 把读者引向其他项目"},{"code":"package_name_confusion","severity":"info","evidence":"README.md：\"`cargo install claw-code` installs the wrong thing. The `claw-code` crate on crates.io is a deprecated stub\"，运行后会打印 \"claw-code has been renamed to agent-code\"。","source":"ai","title":"软件包名称容易混淆"},{"code":"release_files_contradict_readme","severity":"info","evidence":"README 说该仓库 \"build-from-source only\"，但 v0.1.10 版本（2026-11-02）却发布了现成的二进制文件（claw-windows-x64.exe、claw-macos-arm64、claw-linux-x64），而且没有为它们提供任何安装说明。","source":"ai","title":"发布文件与 README 相矛盾"},{"code":"announced_features_not_there_yet","severity":"info","evidence":"README.md：\"`claw-code` does not ship an ACP/Zed daemon or JSON-RPC entrypoint yet\" 和 \"`claw acp serve` is currently a discoverability alias only\"。docker-compose.yml 中可选的搜索套件也以 `CLAW_RAG_MOCK_PROVIDERS: \"1\"`（模拟的嵌入向量）运行，直到你设置真正的提供方。","source":"ai","title":"宣传的功能尚未实现"},{"code":"rust_port_of_claude_code_source_material","severity":"high","evidence":"README.md：\"Claw Code is the public Rust implementation of the `claw` CLI agent harness\"，列出 \"**`PARITY.md`** — Rust-port parity status and migration notes\"，其 \"Ownership / affiliation disclaimer\" 写道 \"This repository does **not** claim ownership of the original Claude Code source material\" 和 \"This repository is **not affiliated with, endorsed by, or maintained by Anthropic**\"。Claude Code 是 Anthropic 自己的产品，按其商业条款分发，没有公开的开源许可证，而这个仓库则以 MIT 许可证发布自己的代码（LICENSE）。","source":"ai","title":"Claude Code 源材料的 Rust 移植版"}],"platforms":["windows","mac","linux"],"tools":["rust"],"buildTools":["docker"],"disk":{"minMB":15,"maxMB":45,"basis":"installer"},"apiKeys":["AI model provider key (Anthropic, OpenAI or a local OpenAI-compatible server)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true}],"declaredAlternatives":[],"stars":194797,"downloads":3110,"downloadsWeek":659,"downloadsDays":7,"growth":0,"growthDays":7,"history":[{"date":"2026-10-26","stars":194988},{"date":"2026-10-27","stars":194969},{"date":"2026-10-28","stars":194970},{"date":"2026-10-29","stars":194968},{"date":"2026-10-30","stars":194995},{"date":"2026-10-31","stars":194997},{"date":"2026-11-01","stars":194994},{"date":"2026-11-02","stars":194973},{"date":"2026-11-03","stars":194970},{"date":"2026-11-04","stars":194962},{"date":"2026-11-05","stars":194949},{"date":"2026-11-06","stars":194948},{"date":"2026-11-07","stars":194943},{"date":"2026-11-08","stars":194948},{"date":"2026-11-09","stars":194804},{"date":"2026-11-10","stars":194797}],"firstPublishedAt":"2026-10-27T01:10:12.701Z","commitSha":"51cfca114cf506a5d7a53ab1bffa8e1bbd3c780d","verifiedAt":"2026-11-02T02:58:02.201Z","lastCommitAt":"2026-11-10T05:59:17Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A terminal-based AI coding assistant that its own README calls an experimental exhibit rather than a real product.","summary":"Claw Code is a command-line AI coding assistant written in Rust. You type requests in a terminal and it talks to an AI model to work on code. The README says it is \"not the serious production project here\", calls it a \"museum exhibit\" run by AI agents with no human intervention, and points users to two other projects instead. It is for programmers curious about AI agent tools. The README says to install Rust and build it from source, though releases also include ready files for Windows, Mac and Linux. You need your own paid AI API key, such as Anthropic or OpenAI.","easeRationale":"All three platforms are the same work: the README says this repo is \"build-from-source only\", so you install Rust and Git, run `cargo build --workspace`, set an API key and use it from a terminal; bare release files (claw-windows-x64.exe, claw-macos-arm64, claw-linux-x64) also exist but the project does not document them as the install path.","costNote":"The software is free (MIT); it requires an API key, and Claude subscription login is not supported.","notVerified":["Whether the release files match this source code and work without building.","Which files and project content are sent to the AI service while you use it.","Real API cost for a typical coding session.","Whether any of the code was copied from another company's product.","How stable it is, given that AI agents maintain it without human review."]},"zh-Hans":{"summaryShort":"一个终端里的 AI 编程助手,其 README 自称是实验性展品,而不是真正的产品。","summary":"Claw Code 是一个用 Rust 编写的命令行 AI 编程助手。你在终端里输入请求,它会与一个 AI 模型沟通来处理代码。README 说这'不是这里认真的生产项目',称它是由 AI 代理在没有人工干预的情况下运行的'博物馆展品',并建议用户转向另外两个项目。它适合对 AI 代理工具感兴趣的程序员。README 要求安装 Rust 并从源码构建,不过各版本也附带了 Windows、Mac 和 Linux 的现成文件。你需要自备付费的 AI API 密钥,例如 Anthropic 或 OpenAI 的。","easeRationale":"三个平台所需的操作相同:README 说这个仓库'只能从源码构建',所以你要安装 Rust 和 Git,运行 cargo build --workspace,设置一个 API 密钥,然后在终端中使用;也存在独立的发布文件(claw-windows-x64.exe、claw-macos-arm64、claw-linux-x64),但项目没有把它们记录为安装方式。","costNote":"软件本身免费(MIT);需要 API 密钥,且不支持用 Claude 订阅登录。","notVerified":["发布文件是否与这份源码一致,能否不经编译直接运行。","使用期间,哪些文件和项目内容会被发送给 AI 服务。","一次典型编程会话的真实 API 费用。","其中是否有代码抄自其他公司的产品。","在没有人工审查、仅由 AI 代理维护的情况下,它有多稳定。"]}}},{"id":"R_kgDOVPbxGw","slug":"unixnu__unixnu","name":"UniXNU","owner":"UniXNU","fullName":"UniXNU/UniXNU","type":"framework_infra","status":"alert","grade":"E","easeByPlatform":{},"risk":"high","signals":[{"code":"young_owner_account","severity":"attention","evidence":"所有者账号创建于 8 天前","source":"check","title":"所有者账号很新"},{"code":"runs_apple_systems_on_non_apple_hardware","severity":"high","evidence":"仓库描述和 README：\"An independent project to run iOS and macOS on non-Apple ARM64 hardware through UEFI.\" Apple 仅授权 iOS 和 macOS 在 Apple 的硬件上使用，而该项目没有说明系统镜像从何而来，因此这种用法违反了 Apple 的许可协议，在你所在的地区还可能违法。","source":"ai","title":"在非 Apple 硬件上运行 Apple 系统"},{"code":"replaces_device_firmware","severity":"attention","evidence":"仓库的表格列出了 \"uefi-sheng | UEFI firmware for the Xiaomi Pad 6S Pro\"，所以使用它就意味着要替换一台平板的启动固件。尝试失败可能导致设备无法开机。","source":"ai","title":"替换设备固件"},{"code":"very_early_and_one_device","severity":"attention","evidence":"仓库创建于 2026-10-19，没有发布过版本，有 1095 个星标，检查记录显示所有者账号只有 8 天；唯一的设备仓库是面向 Xiaomi Pad 6S Pro 的 UniXNU-sheng，README 开头有一则 \"Migration notice\"，要求移植者更新他们的仓库。","source":"ai","title":"非常早期，且只支持一款设备"},{"code":"instructions_outside_the_repository","severity":"info","evidence":"README.md 把所有实际操作都指向 unixnu.com：\"Start with Port a device or Build and boot\"。仓库本身没有记录任何步骤，所以无法根据这里的代码判断你实际需要做什么。","source":"ai","title":"操作说明在仓库之外"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"BSD-2-Clause","category":"permissive","detectedFrom":"file"},"categories":["hardware","network_system","development"],"alternativeTo":[],"declaredAlternatives":[],"stars":2391,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":438,"growthDays":7,"history":[{"date":"2026-10-27","stars":1095},{"date":"2026-10-28","stars":1146},{"date":"2026-10-29","stars":1225},{"date":"2026-10-30","stars":1320},{"date":"2026-10-31","stars":1490},{"date":"2026-11-01","stars":1703},{"date":"2026-11-02","stars":1837},{"date":"2026-11-03","stars":1953},{"date":"2026-11-04","stars":2008},{"date":"2026-11-05","stars":2047},{"date":"2026-11-06","stars":2078},{"date":"2026-11-07","stars":2101},{"date":"2026-11-08","stars":2134},{"date":"2026-11-09","stars":2151},{"date":"2026-11-10","stars":2391}],"firstPublishedAt":"2026-10-27T18:40:28.081Z","commitSha":"120067ccc43ebb5e76ec62c9219a48c897052838","verifiedAt":"2026-10-28T01:42:08.111Z","lastCommitAt":"2026-11-10T20:37:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A boot and driver kit for people trying to start Apple's iOS or macOS on non-Apple ARM tablets.","summary":"UniXNU is a kit for developers who want to boot Apple's iOS or macOS on hardware Apple never made, such as a Xiaomi Pad 6S Pro. Like OpenCore on PCs, it is a UEFI boot environment plus build tools, drivers and system profiles that a device port is built on. It is not a program with an installer: there are no releases, the device support lives in separate repositories, and the build and boot instructions are on the project's own website. Only one tablet is listed as a target. Apple's systems are licensed for Apple hardware only.","easeRationale":"Grade E: there are no releases at all, no install or build steps in the repository (they are on unixnu.com), and the work consists of cloning several repositories with submodules, building UEFI firmware and porting drivers for a specific tablet. Nothing here runs on Windows, Mac, Linux or the browser as a finished program.","costNote":"Free and open source; the LICENSE file is BSD-2-Clause-Patent.","notVerified":["Whether iOS or macOS actually boots and is usable on any of the listed hardware","Where the Apple system files would come from, which the repository does not say","Whether a tablet can be restored to its original software after the firmware is replaced","Which parts work and which do not, since there is no release and no changelog in the repository"]},"zh-Hans":{"summaryShort":"一个启动和驱动工具包,面向想在非苹果 ARM 平板上启动苹果 iOS 或 macOS 的人。","summary":"UniXNU 是一个面向开发者的工具包,用于在苹果从未制造过的硬件上启动苹果的 iOS 或 macOS,例如小米 Pad 6S Pro。就像 PC 上的 OpenCore 一样,它是一个 UEFI 启动环境,加上编译工具、驱动程序和系统配置文件,设备适配工作就建立在这些之上。它不是带安装程序的软件:没有发布版本,设备支持存放在单独的仓库中,编译和启动说明则放在项目自己的网站上。目前列出的目标设备只有一款平板。苹果的系统仅授权用于苹果自家硬件。","easeRationale":"评级为 E:完全没有发布版本,仓库里也没有安装或编译步骤(这些说明在 unixnu.com 上),相关工作包括克隆带子模块的多个仓库、编译 UEFI 固件,以及为特定平板适配驱动程序。这里没有任何内容能作为成品程序运行在 Windows、Mac、Linux 或浏览器上。","costNote":"免费且开源;LICENSE 文件为 BSD-2-Clause-Patent。","notVerified":["iOS 或 macOS 是否真的能在列出的硬件上启动并可用","苹果系统文件应从何处获取,仓库中并未说明","固件被替换后,平板是否可以恢复到原始软件","哪些部分能用、哪些不能,因为仓库中没有发布版本,也没有更新日志"]}}},{"id":"R_kgDOKznBOA","slug":"unslothai__unsloth","name":"Unsloth","owner":"unslothai","fullName":"unslothai/unsloth","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"curl_pipe_shell","severity":"attention","evidence":"安装命令把远程脚本直接灌进 shell：curl -LsSf https://astral.sh/uv/install.sh | sh","source":"check","title":"安装时直接下载并执行脚本"},{"code":"tools_enabled_when_exposed_to_network","severity":"attention","evidence":"README.md 的 Remote HTTPS & LAN Access：\"Server-side tools are on by default - so be careful! Keep your password safe, or use --disable-tools when exposing Unsloth.\" 同一节还提供了 Cloudflare 的公开 HTTPS 链接和监听 0.0.0.0 的做法。","source":"ai","title":"对外开放时工具仍默认开启"},{"code":"model_can_run_commands_and_browse","severity":"info","evidence":"README.md 的功能列表：\"tool calling and code execution\"、\"Browser use: Navigate sites, fill out forms, read PDFs, and edit files\"，以及 \"Sandboxing: Restrict which files your model can access when it runs commands in a sandbox\"。","source":"ai","title":"模型可以执行命令和上网"},{"code":"setup_downloads_extra_prebuilt_binaries","severity":"info","evidence":"安装程序会获取额外的预编译组件：README.md 写道 \"Setup also installs audio.cpp (prebuilt by unslothai/audio.cpp)\"，并允许选择 llama.cpp 的后端（UNSLOTH_LLAMA_CPP_BACKEND=vulkan/cpu/cuda/rocm）；pyproject.toml 里还带着 node_prebuilt_pins.json 和 lemonade_prebuilt_pins.json。","source":"ai","title":"安装过程还会下载其他预编译程序"},{"code":"beta_releases","severity":"info","evidence":"最近三个发布版都标为 beta（v0.1.916-beta、v0.1.915-beta、v0.1.914-beta，2026 年 11 月）。","source":"ai","title":"发布版本仍是 beta"},{"code":"local_server_port_open","severity":"info","evidence":"该应用会运行一个本地 HTTP 服务：README.md 介绍了 UNSLOTH_DESKTOP_PORT、默认端口 8888，以及 \"If the selected port is busy, Desktop tries the next 20 ports\"。","source":"ai","title":"会开启本地服务端口"},{"code":"dual_license_agpl_parts","severity":"info","evidence":"README.md 的 License：\"The core Unsloth package remains licensed under Apache 2.0, while certain optional components, such as the Unsloth Studio UI are licensed under... AGPL-3.0\"；仓库里同时放着 LICENSE（Apache）和 COPYING（AGPL-3.0）。","source":"ai","title":"双重许可，部分组件为 AGPL"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["python","gpu_nvidia"],"disk":{"minMB":21,"maxMB":62,"basis":"installer"},"apiKeys":["Optional: a ChatGPT/Codex subscription or cloud model providers can be connected; not needed for local models"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["local_ai","ai_agents","development"],"alternativeTo":[{"slug":"chatgpt-plus","name":"ChatGPT Plus","confirmed":true},{"slug":"midjourney","name":"Midjourney","confirmed":true}],"declaredAlternatives":[],"stars":79292,"downloads":5480586,"downloadsWeek":472004,"downloadsDays":6,"growth":271,"growthDays":7,"history":[{"date":"2026-11-04","stars":79049},{"date":"2026-11-05","stars":79080},{"date":"2026-11-06","stars":79124},{"date":"2026-11-07","stars":79155},{"date":"2026-11-08","stars":79198},{"date":"2026-11-09","stars":79251},{"date":"2026-11-10","stars":79292}],"firstPublishedAt":"2026-11-10T23:52:44.812Z","commitSha":"322c4c1a50541238632c2aeb95821151f575f29c","verifiedAt":"2026-11-10T23:36:44.924Z","lastCommitAt":"2026-11-10T20:23:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A desktop app to chat with AI models on your own computer, and to train or fine-tune them.","summary":"Unsloth is a desktop app (and a web interface and a Python package) that downloads and runs AI models on your own machine: chat models, image and video generators, speech models. The same app fine-tunes models, which the team says is about twice as fast and uses much less video memory. There are installers for Windows, macOS and Linux, and it works with NVIDIA, AMD and Intel cards or just the processor. Models are downloaded inside the app, so plan for several gigabytes of disk. The releases are marked beta.","easeRationale":"Every release ships ready installers (Windows .exe, macOS .dmg, Linux .deb and AppImage, around 20 to 170 MB), and there are Homebrew and Winget packages too, so on all three systems it is download, install and then pick and download a model inside the app: grade B. The command-line routes (install script, pip, Docker) exist but are not needed for the desktop app.","costNote":"Dual licence: the core package is Apache 2.0 and parts such as the Studio interface are AGPL-3.0, which matters if you want to redistribute a changed version or run it as a service.","notVerified":["Real disk space: the installer is small, but the models you download inside the app take gigabytes.","The speed and memory claims (\"2x faster\", \"70% less VRAM\", \"no accuracy loss\") were not measured here.","How well it runs and trains on AMD, Intel, Apple Silicon or processor-only machines compared with NVIDIA cards.","How stable the beta desktop app is day to day.","What the app sends out when you use web search, the Cloudflare link or connected cloud providers."]},"zh-Hans":{"summaryShort":"一款桌面应用，可以在自己电脑上与 AI 模型对话，也能训练或微调模型。","summary":"Unsloth 是一款桌面应用（同时提供网页界面和 Python 包），可以在你自己的机器上下载并运行 AI 模型：对话模型、图像和视频生成模型、语音模型。同一个程序还能微调模型，团队称速度约快一倍、显存占用大幅降低。它为 Windows、macOS 和 Linux 提供安装包，支持 NVIDIA、AMD、Intel 显卡，也可以只用处理器。模型在应用内下载，因此要预留好几个 GB 的磁盘空间。目前的发布版都标为 beta。","easeRationale":"每个发布版都提供现成安装包（Windows 的 .exe、macOS 的 .dmg、Linux 的 .deb 和 AppImage，约 20 到 170 MB），还有 Homebrew 和 Winget 包，因此三个系统都是下载、安装，然后在应用里挑选并下载模型：给 B。命令行方式（安装脚本、pip、Docker）也有，但使用桌面版并不需要。","costNote":"双重许可：核心包采用 Apache 2.0，Studio 界面等部分采用 AGPL-3.0；如果你想再分发修改版或把它当作服务来运营，这一点很重要。","notVerified":["真实磁盘占用：安装包很小，但你在应用里下载的模型会占好几个 GB。","关于速度和显存的说法（\"2x faster\"、\"70% less VRAM\"、\"no accuracy loss\"）本次没有实测。","在 AMD、Intel、Apple Silicon 或纯处理器的机器上，运行和训练效果与 NVIDIA 显卡相比如何。","处于 beta 的桌面应用在日常使用中有多稳定。","当你使用网页搜索、Cloudflare 链接或已连接的云服务时，应用会向外发送什么。"]}}},{"id":"R_kgDOHvghIw","slug":"upscayl__upscayl","name":"Upscayl","owner":"upscayl","fullName":"upscayl/upscayl","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"needs_vulkan_graphics_card","severity":"attention","evidence":"README 中标为 IMPORTANT：\"You'll need a Vulkan compatible GPU (Graphics Card) to upscale images. Many iGPUs (integrated graphics) do not work but, no harm in trying :)\"，常见问题（FAQ）也重复了这一点：\"Yes, unfortunately. NCNN Vulkan requires a Vulkan-compatible GPU. Upscayl won't work with most iGPUs or CPUs.\" 它可能顺利装上，却仍然无法放大图片。","source":"ai","title":"需要支持 Vulkan 的显卡"},{"code":"no_new_release_in_almost_two_years","severity":"attention","evidence":"最新版本是 2024-12-25 发布的 v2.15.0，所分析分支上的最后一次提交是 2026-03-27，距本次检查已过去 216 天。README 仍在宣布 \"v2.15 is out!\"。","source":"ai","title":"近两年没有新版本"},{"code":"installer_triggers_os_warning","severity":"attention","evidence":"Windows 的说明写道：\"If you get a SmartScreen warning - click 'More Info' and then 'Run Anyway' OR press 'YES' on the unverified publisher dialog\"，Mac 的步骤则要求你右键点击应用并按两次 'Open'。package.json 中设置了 `\"dmg\": { \"sign\": \"false\" }`。也就是说，你被要求绕过系统自带的检查。","source":"ai","title":"安装程序会触发系统警告"},{"code":"declared_alternative_comes_from_a_sponsor_banner","severity":"info","evidence":"检查发现了一个声称是 Postman 替代品的说法，出自这一行 \"Requestly - Free & Open-Source alternative to Postman\"。这一行是 README 顶部的付费赞助商横幅（旁边还有一个 Warp 的），并不是关于 Upscayl 的说法，Upscayl 只处理图片。","source":"ai","title":"所谓替代方案来自赞助商横幅"},{"code":"extra_models_from_elsewhere","severity":"info","evidence":"README 指向另一个仓库获取更多模型（\"More models can be taken from here: https://github.com/upscayl/custom-models\"），并注明社区模型托管在 openmodeldb.info。你自己添加的任何模型都来自这次下载之外。","source":"ai","title":"额外模型来自别处"}],"platforms":["windows","mac","linux"],"tools":["gpu_any"],"buildTools":["node"],"disk":{"minMB":249,"maxMB":746,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"AGPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["image"],"alternativeTo":[{"slug":"topaz-gigapixel-ai","name":"Topaz Gigapixel AI","confirmed":true},{"slug":"photoshop","name":"Photoshop","confirmed":true},{"slug":"lightroom","name":"Lightroom","confirmed":false}],"declaredAlternatives":[{"target":"postman","stance":"claims","evidence":"<b>Requestly - Free & Open-Source alternative to Postman</b>"}],"stars":52178,"downloads":84615048,"downloadsWeek":352720,"downloadsDays":7,"growth":480,"growthDays":7,"history":[{"date":"2026-10-26","stars":51105},{"date":"2026-10-27","stars":51179},{"date":"2026-10-28","stars":51229},{"date":"2026-10-29","stars":51301},{"date":"2026-10-30","stars":51358},{"date":"2026-10-31","stars":51524},{"date":"2026-11-01","stars":51589},{"date":"2026-11-02","stars":51644},{"date":"2026-11-03","stars":51698},{"date":"2026-11-04","stars":51752},{"date":"2026-11-05","stars":51793},{"date":"2026-11-06","stars":51851},{"date":"2026-11-07","stars":51917},{"date":"2026-11-08","stars":52011},{"date":"2026-11-09","stars":52103},{"date":"2026-11-10","stars":52178}],"firstPublishedAt":"2026-10-27T01:10:12.701Z","commitSha":"a00d55fee90e0f9435d5eaa86e76700df8199af8","verifiedAt":"2026-10-30T01:44:15.614Z","lastCommitAt":"2026-03-27T16:33:02Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"upscayl-2.15.0-win.exe","sizeBytes":260912662,"url":"https://github.com/upscayl/upscayl/releases/download/v2.15.0/upscayl-2.15.0-win.exe","releaseUrl":"https://github.com/upscayl/upscayl/releases/tag/v2.15.0"},{"os":"mac","file":"upscayl-2.15.0-mac.dmg","sizeBytes":385264901,"url":"https://github.com/upscayl/upscayl/releases/download/v2.15.0/upscayl-2.15.0-mac.dmg","releaseUrl":"https://github.com/upscayl/upscayl/releases/tag/v2.15.0"},{"os":"linux","file":"upscayl-2.15.0-linux.AppImage","sizeBytes":312749864,"url":"https://github.com/upscayl/upscayl/releases/download/v2.15.0/upscayl-2.15.0-linux.AppImage","releaseUrl":"https://github.com/upscayl/upscayl/releases/tag/v2.15.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Enlarges small or pixelated photos with AI on your own computer, with nothing uploaded and no account needed.","summary":"Upscayl is a desktop program that makes a low-resolution picture bigger while filling in detail, using AI models that run on your own graphics card. You pick an image or a whole folder, choose a model and a scale, and it saves the enlarged copy; nothing is uploaded and there is no account. It is for anyone with old photos, small scans or pixelated images. The AI models come inside the download, so there is nothing else to fetch, but you do need a graphics card that supports Vulkan, and the project warns that many built-in laptop graphics chips will not work.","easeRationale":"Every system has a ready download: a Windows installer (upscayl-2.15.0-win.exe, 249 MB), a Mac .dmg you drag to Applications, and for Linux an AppImage, .deb and .rpm plus Flathub, Snap and AUR packages. The models are bundled, so there is nothing to set up after installing — but the program only runs the AI if your graphics card supports Vulkan.","costNote":"Free and open source (AGPL-3.0). The download includes the AI models; the project asks only for donations.","notVerified":["Whether your particular graphics card is one of the supported ones, which decides if the program works at all","How good the result looks on your own pictures, and how long each image takes","Whether a build from December 2024 still runs on the newest Windows and macOS versions","Real disk use once you add extra community models","What the separate macOS App Store listing costs or contains, which Radar did not check"]},"zh-Hans":{"summaryShort":"在你自己的电脑上用 AI 放大模糊或像素化的照片，无需上传任何内容，也不需要账号。","summary":"Upscayl 是一款桌面程序，它使用在你自己显卡上运行的 AI 模型，把低分辨率图片放大并补充细节。你选择一张图片或整个文件夹，选择一个模型和放大倍数，程序就会保存放大后的副本；不会上传任何内容，也不需要账号。它适合任何有老照片、小尺寸扫描件或像素化图片的人。AI 模型已经包含在下载包里，所以不需要再下载别的东西，但你需要一块支持 Vulkan 的显卡，项目也提醒很多笔记本电脑自带的显卡芯片无法使用。","easeRationale":"每个系统都有现成的下载：Windows 安装程序（upscayl-2.15.0-win.exe，249 MB）、可拖到“应用程序”文件夹的 Mac .dmg 文件，以及 Linux 的 AppImage、.deb 和 .rpm，还有 Flathub、Snap 和 AUR 软件包。模型已经打包在内，安装后不需要再做任何设置——但程序只有在你的显卡支持 Vulkan 时才能运行 AI。","costNote":"免费且开源（AGPL-3.0）。下载包中已包含 AI 模型；项目只接受捐赠，不收取费用。","notVerified":["你的显卡具体型号是否在支持列表中，这决定了程序能不能运行","放大你自己的照片效果如何，以及每张图片需要多长时间","2024 年 12 月的版本是否仍能在最新的 Windows 和 macOS 系统上运行","添加社区提供的额外模型后，实际占用的磁盘空间","macOS App Store 中单独上架的版本收费多少或包含什么内容，Radar 未对此进行核实"]}}},{"id":"R_kgDOIFKaXA","slug":"usebruno__bruno","name":"Bruno","owner":"usebruno","fullName":"usebruno/bruno","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"collections_can_run_scripts","severity":"attention","evidence":"该项目自带用于请求的 JavaScript 引擎（package.json 的 workspaces 中有 `packages/bruno-js`），README 还链接了一个名为 \"Scriptmania\" 的社区讨论。因此，别人发给你的集合文件可能带有代码，当你运行该请求时，这些代码会在你的电脑上运行。","source":"ai","title":"集合文件可以运行脚本"},{"code":"secrets_live_in_plain_files","severity":"attention","evidence":"README：\"Bruno stores your collections directly in a folder on your filesystem. We use a plain text markup language, Bru, to save information about API requests\"，以及 \"You can use Git or any version control of your choice to collaborate over your API collections.\" 你在请求中填写的密钥和密码就存放在这些文件里，如果不小心，可能会被推送到共享仓库。","source":"ai","title":"密钥保存在明文文件里"},{"code":"owner_says_it_stays_offline","severity":"info","evidence":"README：\"Bruno is offline-only. There are no plans to add cloud-sync to Bruno, ever. We value your data privacy and believe it should stay on your device.\" 这是项目自己的说法；Radar 没有观察该程序实际发送了什么。","source":"ai","title":"作者称它只在本地运行"},{"code":"some_features_are_sold_separately","severity":"info","evidence":"README 的 \"Commercial Versions\" 一节：\"Majority of our features are free and open source. We strive to strike a harmonious balance between open-source principles and sustainability\"，并附有付费版本的链接。哪些功能需要付费由公司网站决定，而不是由这个仓库决定。","source":"ai","title":"部分功能单独收费"},{"code":"many_open_issues","severity":"info","evidence":"47931 个星标对应 1969 个未关闭的 issue（问题反馈），最后一次提交就在本次检查当天，4.3.0 版本发布于 2026-10-22。","source":"ai","title":"未解决的问题很多"},{"code":"name_is_a_trademark","severity":"info","evidence":"README：\"`Bruno` is a trademark held by Anoop M D\"。代码采用 MIT 许可，但如果你重新发布修改后的版本，名称和标志并不能随意复用。","source":"ai","title":"名称是注册商标"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":129,"maxMB":386,"basis":"installer"},"apiKeys":[],"cost":{"model":"freemium"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development"],"alternativeTo":[{"slug":"postman","name":"Postman","confirmed":true},{"slug":"insomnia","name":"Insomnia","confirmed":true}],"declaredAlternatives":[],"stars":48324,"downloads":76056249,"downloadsWeek":920176,"downloadsDays":7,"growth":291,"growthDays":7,"history":[{"date":"2026-10-26","stars":47856},{"date":"2026-10-27","stars":47879},{"date":"2026-10-28","stars":47909},{"date":"2026-10-29","stars":47932},{"date":"2026-10-30","stars":47956},{"date":"2026-10-31","stars":47965},{"date":"2026-11-01","stars":47980},{"date":"2026-11-02","stars":47994},{"date":"2026-11-03","stars":48033},{"date":"2026-11-04","stars":48135},{"date":"2026-11-05","stars":48207},{"date":"2026-11-06","stars":48245},{"date":"2026-11-07","stars":48267},{"date":"2026-11-08","stars":48281},{"date":"2026-11-09","stars":48306},{"date":"2026-11-10","stars":48324}],"firstPublishedAt":"2026-10-27T01:10:12.703Z","commitSha":"b569a7288a6ca38377f23878d9ab43da7dc3d511","verifiedAt":"2026-10-30T01:50:32.445Z","lastCommitAt":"2026-11-09T10:47:10Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"bruno_4.3.0_x64_win.exe","sizeBytes":123071368,"url":"https://github.com/usebruno/bruno/releases/download/v4.3.0/bruno_4.3.0_x64_win.exe","releaseUrl":"https://github.com/usebruno/bruno/releases/tag/v4.3.0"},{"os":"mac","file":"bruno_4.3.0_arm64_mac.dmg","sizeBytes":167922780,"url":"https://github.com/usebruno/bruno/releases/download/v4.3.0/bruno_4.3.0_arm64_mac.dmg","releaseUrl":"https://github.com/usebruno/bruno/releases/tag/v4.3.0"},{"os":"linux","file":"bruno_4.3.0_x86_64_linux.AppImage","sizeBytes":174598148,"url":"https://github.com/usebruno/bruno/releases/download/v4.3.0/bruno_4.3.0_x86_64_linux.AppImage","releaseUrl":"https://github.com/usebruno/bruno/releases/tag/v4.3.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Try out and test web APIs from a desktop app that keeps your requests as plain files you can put in Git.","summary":"Bruno is a desktop program for sending requests to web APIs and checking what comes back: you build a request, fill in the address, headers and body, send it and read the response. Each collection of requests is saved as plain text files in a folder on your computer, so you can keep them in Git next to the code and share them with your team without an account. There is also a command-line version, `bru`, for running a whole collection automatically. It is for developers and testers. Installers are ready for Windows, macOS and Linux.","easeRationale":"Release 4.3.0 has ready installers for all three systems (.exe and .msi for Windows, .dmg and .pkg for macOS on both chips, .deb, .rpm and AppImage for Linux), and the README also lists Homebrew, Chocolatey, Scoop, winget, Snap, Flatpak, AUR and an apt repository. You install it and start sending requests; nothing else is needed.","costNote":"The program in this repository is MIT-licensed and free. The README says \"Majority of our features are free and open source\" and points to paid versions on usebruno.com/pricing for extra team features; the repository does not list prices.","notVerified":["Whether the program really sends nothing anywhere, which Radar did not test","Which features need a paid version, since the repository only links to a pricing page","Whether importing an existing Postman or Insomnia collection keeps everything intact","What a collection from someone else does when you run it, since requests can contain scripts","Real disk use, which the checks estimate between 129 MB and 386 MB"]},"zh-Hans":{"summaryShort":"通过一个桌面应用来尝试和测试网络 API，它把你的请求保存为可以放进 Git 的普通文件。","summary":"Bruno 是一款桌面程序，用于向网络 API 发送请求并查看返回结果：你构建一个请求，填写地址、请求头和请求体，发送后查看响应。每一组请求集合都以纯文本文件的形式保存在你电脑上的一个文件夹中，这样你可以把它们和代码一起放进 Git，并与团队共享，不需要账号。它还提供一个命令行版本 `bru`，可以自动运行整组请求集合。它适合开发者和测试人员使用。Windows、macOS 和 Linux 都有现成的安装程序。","easeRationale":"4.3.0 版本为三大系统都提供了现成的安装程序（Windows 的 .exe 和 .msi，macOS 两种芯片的 .dmg 和 .pkg，Linux 的 .deb、.rpm 和 AppImage），README 中还列出了 Homebrew、Chocolatey、Scoop、winget、Snap、Flatpak、AUR 和一个 apt 仓库。安装之后就可以直接发送请求，不需要其他任何操作。","costNote":"这个代码仓库中的程序采用 MIT 许可证，可免费使用。README 中写道“我们大部分功能都是免费且开源的”，并指向 usebruno.com/pricing 上的付费版本以获取额外的团队功能；代码仓库本身没有列出价格。","notVerified":["该程序是否真的不会把任何数据发送到别处，Radar 未对此进行测试","哪些功能需要付费版本，因为代码仓库只给出了一个价格页面的链接","导入现有的 Postman 或 Insomnia 请求集合时，是否能完整保留所有内容","运行别人创建的请求集合时会发生什么，因为请求中可能包含脚本","实际占用的磁盘空间，检查结果估计在 129 MB 到 386 MB 之间"]}}},{"id":"R_kgDOVbwqRA","slug":"valentderah__totk-pc-recomp","name":"TOTK PC Recomp","owner":"valentderah","fullName":"valentderah/totk-pc-recomp","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","linux":"C"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"7 天内获得 5127 个星标","source":"check","title":"仓库年轻但星标很多"},{"code":"built_from_the_games_own_files","severity":"high","evidence":"README.md：该项目是 \"a static recompilation of The Legend of Zelda: Tears of the Kingdom (Nintendo Switch, 2023) into a native x86-64 application\"，并且 \"The game is built from your files on your machine, so the result contains the game. It is for your own use. Do not share it.\"。许可证部分写道 \"The license grants nothing for Nintendo's game\"，并说明这是 \"an independent, non-commercial fan project, not affiliated with or endorsed by Nintendo\"。游戏及其代码属于任天堂，许多国家都有关于复制作品以及规避技术保护措施的法律。","source":"ai","title":"基于游戏自身文件构建"},{"code":"requires_dumping_your_own_console","severity":"attention","evidence":"README.md：\"You need your own legally obtained copy of Tears of the Kingdom, dumped from your own console: the base game and the 1.1.0, 1.2.1 or 1.4.3 update. Instructions for obtaining those files will not be provided.\"。该项目需要从主机中提取的文件，但没有说明如何获取；许多国家的法律限制规避技术保护措施。","source":"ai","title":"需要从自己的主机提取文件"},{"code":"built_in_cheats_and_mods","severity":"info","evidence":"README.md 列出了 \"Mods made for emulators\" 和 \"Atmosphère cheats: put a cheat file into a mod and switch its cheats in the menu\"，其中包括会重建游戏的代码补丁（`.ips`、`.pchtxt`）。","source":"ai","title":"内置作弊和模组"},{"code":"early_development_self_declared","severity":"attention","evidence":"README.md 的 Status 部分：\"Early development. The game is playable on Windows and Linux; visual bugs and crashes are still expected.\"","source":"ai","title":"项目自称处于早期开发阶段"},{"code":"mostly_ai_written_and_very_new","severity":"info","evidence":"README.md：\"Most of the code and documentation is written by AI coding agents (Claude Code) under the maintainer's direction; the maintainer plays every build and decides what is merged.\"。该仓库创建于 2026-10-29，已有 5127 个星标和 10 个 release，几乎没有历史记录可参考。","source":"ai","title":"大部分由 ai 编写且非常新"},{"code":"unsigned_installer","severity":"info","evidence":"README.md 安装步骤 2：\"If SmartScreen warns about an unsigned app, choose More info, then Run anyway.\"。安装程序没有代码签名，因此 Windows 无法说明它的开发者是谁。","source":"ai","title":"安装程序未签名"},{"code":"downloads_a_closed_nvidia_library","severity":"info","evidence":"README.md 的 Features 部分：\"the setup downloads `nvngx_dlss.dll` from NVIDIA's DLSS SDK repository (github.com/NVIDIA/DLSS, NVIDIA RTX SDK license) when it finds an RTX GPU; untick it in the setup to skip it.\"。该文件来自此仓库之外，并保留 NVIDIA 自己的许可证。","source":"ai","title":"下载一个封闭的 nvidia 库"}],"platforms":["windows","linux"],"tools":["gpu_any"],"buildTools":[],"disk":{"minMB":157,"maxMB":471,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":6387,"downloads":235387,"downloadsWeek":112863,"downloadsDays":7,"growth":1885,"growthDays":7,"history":[{"date":"2026-10-30","stars":1576},{"date":"2026-10-31","stars":2613},{"date":"2026-11-01","stars":3241},{"date":"2026-11-02","stars":4008},{"date":"2026-11-03","stars":4502},{"date":"2026-11-04","stars":4807},{"date":"2026-11-05","stars":5123},{"date":"2026-11-06","stars":5366},{"date":"2026-11-07","stars":5661},{"date":"2026-11-08","stars":5895},{"date":"2026-11-09","stars":6165},{"date":"2026-11-10","stars":6387}],"firstPublishedAt":"2026-10-30T01:57:54.489Z","commitSha":"0ae47a578285cde080642c9e6c40316c99171ef1","verifiedAt":"2026-11-05T22:47:42.457Z","lastCommitAt":"2026-11-10T16:25:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns your own dumped copy of Zelda: Tears of the Kingdom into a native Windows or Linux game at 60 fps.","summary":"TotK Recomp rebuilds The Legend of Zelda: Tears of the Kingdom as a normal Windows or Linux program instead of running it in an emulator. No game files come with it: a setup program takes the game files you dumped from your own Switch and builds the game on your machine, with 60 fps, custom resolutions, HDR, ultrawide, gamepad support and emulator mods. You need a Vulkan 1.3 graphics card and about 36 GiB free while it installs. The authors call it early development, with crashes still expected.","easeRationale":"On Windows you unpack a zip and run `TotK-Setup.exe`, with no tools to install, but first you have to get the game files off your own Switch yourself, and the README says it will not tell you how. Linux is the same plus making the AppImage executable from a terminal (`chmod +x`), in Desktop Mode on a Steam Deck.","costNote":"The project is free and the README calls it \"an independent, non-commercial fan project\". You still need a paid copy of the game.","notVerified":["How well the game actually plays, and how often it crashes, on a given PC","How long the one-time build takes and how much disk it really uses beyond the stated 36 GiB during setup and 20 GiB after","Whether every listed feature (HDR, 144 fps, ultrawide, mods, cheats) works as described","Whether the three supported game updates (1.1.0, 1.2.1, 1.4.3) all behave the same","What the setup program does with your game files while it builds"]},"zh-Hans":{"summaryShort":"把你自己从主机转储出来的《塞尔达传说：王国之泪》变成一款原生运行在 Windows 或 Linux 上、60 帧的游戏。","summary":"TotK Recomp 把《塞尔达传说：王国之泪》重建为一个普通的 Windows 或 Linux 程序，而不是在模拟器里运行。它本身不附带任何游戏文件：一个安装程序会使用你从自己 Switch 上转储出来的游戏文件，在你的机器上构建游戏，支持 60 帧、自定义分辨率、HDR、超宽屏、手柄，以及模拟器的 mod。你需要一块支持 Vulkan 1.3 的显卡，安装过程中大约需要 36 GiB 的空闲空间。作者称其处于早期开发阶段，仍可能出现崩溃。","easeRationale":"在 Windows 上你解压一个 zip 并运行 `TotK-Setup.exe`，不需要安装任何工具，但你得先自己把游戏文件从自己的 Switch 上取出来，README 表示不会告诉你具体怎么做。Linux 的步骤相同，另外还要在终端里让 AppImage 可执行（`chmod +x`），在 Steam Deck 的桌面模式下操作。","costNote":"该项目免费，README 称其为 \"an independent, non-commercial fan project\"。你仍然需要一份付费购买的游戏。","notVerified":["在某台特定 PC 上，游戏实际的流畅度如何，崩溃有多频繁","一次性构建需要多长时间，实际占用的磁盘空间除了安装期间声明的 36 GiB 和安装后的 20 GiB 之外还有多少","列出的每项功能（HDR、144 帧、超宽屏、mod、作弊码）是否真的像描述那样工作","三个受支持的游戏版本（1.1.0、1.2.1、1.4.3）表现是否完全一致","安装程序在构建过程中如何处理你的游戏文件"]}}},{"id":"MDEwOlJlcG9zaXRvcnkyMDMxNTgwODQ=","slug":"vasani-arpit__social-media-automation","name":"Social Media Automation","owner":"vasani-arpit","fullName":"vasani-arpit/Social-Media-Automation","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 目录和 winget 清单中都没有发现任何许可证","source":"check","title":"没有许可证"},{"code":"pretends_to_be_the_user_on_social_networks","severity":"attention","evidence":"readme.md 的 Solution：\"The solution to this problem is a bot. It will act as a logged in user and do the same activities as the user. So that platform thinks that the user is active\"；它通过浏览器操控 Twitter、Facebook、Instagram 和 LinkedIn，而这些平台的规则通常禁止此类行为，可能导致账号被封。","source":"ai","title":"在社交网络上冒充你本人"},{"code":"license_terms_contradict_each_other","severity":"attention","evidence":"仓库没有许可证文件；package.json 写的是 \"license\": \"MIT\"，而 readme.md 却说 \"Commercial use of this code/repo is strictly prohibited\"，因此再使用的权利并不清楚。","source":"ai","title":"许可条款自相矛盾"},{"code":"unfinished_project","severity":"attention","evidence":"readme.md 里有一节 \"When it will be ready? That's where we need you\" 在招募贡献者；package.json 版本号是 0.1.0，也没有任何发布版本。","source":"ai","title":"项目尚未完成"},{"code":"old_unsupported_dependencies","severity":"attention","evidence":"package.json 锁定了 electron ^6.0.1 和 puppeteer-core ^1.19.0，都是 2019 年的版本，早已不再收到修复更新。","source":"ai","title":"依赖陈旧且已无人支持"},{"code":"logs_into_your_accounts_locally","severity":"info","evidence":"readme.md 声称 \"Complete Privacy (we don't even have servers to store your data)\"：应用在你自己机器上的浏览器窗口里登录你的账号，会话确实留在本地，但由这个程序来操控。","source":"ai","title":"在本地登录你的账号"},{"code":"no_affiliation_use_at_own_risk","severity":"info","evidence":"readme.md 的 Legal：\"This code is in no way affiliated with, authorized, maintained, sponsored or endorsed by any social media platforms... Use at your own risk.\"","source":"ai","title":"与平台无关联，风险自负"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["social_marketing","automation"],"alternativeTo":[{"slug":"buffer","name":"Buffer","confirmed":true},{"slug":"hootsuite","name":"Hootsuite","confirmed":true},{"slug":"phantombuster","name":"PhantomBuster","confirmed":true}],"declaredAlternatives":[],"stars":487,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":1,"growthDays":6,"history":[{"date":"2026-11-05","stars":486},{"date":"2026-11-06","stars":486},{"date":"2026-11-07","stars":486},{"date":"2026-11-08","stars":486},{"date":"2026-11-09","stars":486},{"date":"2026-11-10","stars":487}],"firstPublishedAt":"2026-11-10T23:52:44.833Z","commitSha":"4fc906ae759f64a745fc4d2990a47ae968788731","verifiedAt":"2026-11-10T23:37:38.682Z","lastCommitAt":"2025-02-02T18:51:57Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An unfinished desktop app that drives your social media accounts in a browser: reposting, scheduling, likes.","summary":"This project is a desktop app built with Electron that opens a browser in the background and acts as if it were you on social networks: reposting what you publish on one network to another, scheduling posts and liking or retweeting posts with a chosen hashtag. It is early and incomplete, the author's own README asks for help to finish it, and the last change was in February 2025. To try it you need Node.js, then clone the repository and run two commands. Automating accounts this way goes against the rules of most networks.","easeRationale":"There are no releases or installers: on Windows, macOS and Linux alike you install Node.js, clone the repository, run npm install and npm start, which is a C. It also uses puppeteer-core, the version of Puppeteer that does not bring its own browser, so a Chrome install on the machine is expected.","costNote":"The README says \"Totally Free for personal use\" and \"Commercial use of this code/repo is strictly prohibited\".","notVerified":["Whether it still runs and which networks actually work; the code is from 2019 with a small change in 2025.","What happens to your accounts if a network detects the automation.","Where the app keeps your logged-in sessions and how they are protected on your machine.","Whether scheduling and hashtag liking are really implemented or only described in the README."]},"zh-Hans":{"summaryShort":"一个尚未完成的桌面应用，用浏览器操控你的社交账号：转发、定时发布、点赞。","summary":"这个项目是用 Electron 做的桌面应用，会在后台打开浏览器，以你的身份在社交网络上行动：把你在一个平台发的内容转发到另一个平台、定时发帖、给带指定话题标签的帖子点赞或转推。它还处于早期、功能不全，作者自己的 README 也在请人帮忙完成，最后一次改动是 2025 年 2 月。要试用，你需要 Node.js，然后克隆仓库并运行两条命令。这样自动操作账号违反大多数平台的规则。","easeRationale":"没有发布版也没有安装包：Windows、macOS 和 Linux 上都要装 Node.js、克隆仓库，再运行 npm install 和 npm start，因此给 C。它还使用 puppeteer-core——不自带浏览器的那个版本，所以默认你机器上已经装有 Chrome。","costNote":"README 写着 \"Totally Free for personal use\" 和 \"Commercial use of this code/repo is strictly prohibited\"。","notVerified":["它今天是否还能运行、哪些平台真的可用；代码出自 2019 年，2025 年只有一处小改动。","如果平台检测到自动化行为，你的账号会怎样。","应用把你的登录会话存在哪里，在你机器上又是如何保护的。","定时发布和按话题标签点赞是真的实现了，还是只写在 README 里。"]}}},{"id":"R_kgDOQMFwdQ","slug":"vectorize-io__hindsight","name":"Hindsight","owner":"vectorize-io","fullName":"vectorize-io/hindsight","type":"framework_infra","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"everything_you_store_passes_through_a_model_provider","severity":"attention","evidence":"README：\"Behind the scenes, retain uses an LLM to extract key facts, temporal data, entities, and relationships\"，快速入门还把 `HINDSIGHT_API_LLM_API_KEY` 设为 OpenAI 的密钥。除非你改用本地模型，否则你的记忆文本会发送给该提供商。","source":"ai","title":"你存入的所有内容都会经过模型提供商"},{"code":"coding_agent_setup_ingests_your_repository_history_by_itself","severity":"info","evidence":"README 的 \"Coding Agents\" 部分：`npx @vectorize-io/hindsight-coding-agents install all` 会接入 \"every detected agent\"，建立 \"a per-repo bank built automatically from git history and past sessions\"，并且 \"Ingestion is automatic — there is no setup command.\"","source":"ai","title":"编程智能体的配置会自动读取你的仓库历史"},{"code":"secret_and_personal_data_scanning_is_off_by_default","severity":"info","evidence":"README，Memory Banks 部分：Memory Defense 是 \"An opt-in, per-bank policy that scans every retain for secrets and PII against 45 patterns\"。在你开启它之前，你存入内容中的令牌和个人数据会原样保存。","source":"ai","title":"密钥和个人数据扫描默认关闭"},{"code":"command_line_client_installed_by_piping_a_script_into_bash","severity":"info","evidence":"README 的 \"Connect a client\" 部分：`curl -fsSL https://hindsight.vectorize.io/get-cli | bash` 会运行该地址返回的任何内容，且没有给出校验和。","source":"ai","title":"命令行客户端通过把脚本直接交给 bash 来安装"},{"code":"benchmark_comparison_comes_from_the_vendor","severity":"info","evidence":"README：\"Hindsight is the most accurate agent memory system ever tested according to benchmark performance\"，配有项目自己制作的图表；同一部分也承认 \"Other scores are self-reported by software vendors.\" 它还声称，却没有在仓库中给出证据，称其 \"is being used in production at Fortune 500 enterprises\"。","source":"ai","title":"基准对比来自厂商自己"},{"code":"counting_pixel_in_the_readme","severity":"info","evidence":"README.md 的最后一行是一张 1x1 的图片：`<img src=\"https://umami-pixel.chris-latimer.workers.dev/?id=a8b043e6-6964-454d-80df-69b69d3f0d50&host=github.com&url=/vectorize-io/hindsight\" width=\"1\" height=\"1\">`，它会把页面的访问情况报告给作者自己的服务器。","source":"ai","title":"README 中有统计像素"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":["python","node"],"disk":{"minMB":200,"maxMB":800,"basis":"heuristic"},"apiKeys":["OpenAI, or any of the 25+ supported model providers (a local Ollama or LM Studio needs no key)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"mem0","name":"Mem0","confirmed":true},{"slug":"zep","name":"Zep","confirmed":true},{"slug":"supermemory","name":"Supermemory","confirmed":true}],"declaredAlternatives":[],"stars":60285,"downloads":68574,"downloadsWeek":8803,"downloadsDays":7,"growth":1783,"growthDays":7,"history":[{"date":"2026-10-29","stars":56204},{"date":"2026-10-30","stars":56643},{"date":"2026-10-31","stars":56983},{"date":"2026-11-01","stars":57704},{"date":"2026-11-02","stars":58084},{"date":"2026-11-03","stars":58502},{"date":"2026-11-04","stars":58822},{"date":"2026-11-05","stars":59107},{"date":"2026-11-06","stars":59423},{"date":"2026-11-07","stars":59707},{"date":"2026-11-08","stars":59893},{"date":"2026-11-09","stars":60090},{"date":"2026-11-10","stars":60285}],"firstPublishedAt":"2026-10-29T01:54:21.503Z","commitSha":"6decc291bc0b9f99568ad53cb0501f75128c6d7b","verifiedAt":"2026-10-29T01:43:05.766Z","lastCommitAt":"2026-11-10T17:14:29Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A memory server for AI agents: it stores what they are told and gives the useful parts back later.","summary":"A memory system for AI agents that you run yourself. Your program sends text in (retain), asks questions later (recall), or asks for a worked-out answer built from everything stored (reflect). Facts, experiences and background summaries live in a PostgreSQL database, and there are clients for Python, Node, Go and a CLI, plus ready-made hooks for coding agents, agent frameworks and MCP. For developers building agents or chatbots that should remember. Start with one Docker command or `pip install hindsight-api`, plus a key for one of 25+ model providers; local models also work.","easeRationale":"C on Windows, Mac and Linux: the shortest path is `pip install hindsight-api`, one environment variable with a model key and one command, with the database bundled (the README's platform table marks the embedded database as supported on all three). The Docker route is one long command instead. The release files are only a small command-line client for Linux and Mac, and Intel Macs need the separate `hindsight-all-slim` package.","costNote":"The code is MIT and self-hosting is free, but every retain, recall and reflect calls a language model, so you pay that provider. The README says local providers (`ollama`, `lmstudio`, `llamacpp`) and existing subscriptions (ChatGPT, Claude, Cursor, GitHub Copilot) work too. The company behind it also sells Hindsight Cloud, a usage-based hosted version.","notVerified":["What it really costs in model tokens to keep a busy agent's memory up to date.","Whether the benchmark numbers in the README reproduce on your own data.","How much disk the database uses as memories pile up.","Whether the pip and embedded-database routes behave on Windows as the platform table claims.","Whether bank isolation (\"no cross-bank leakage\") holds in practice when several users share one server.","What Hindsight Cloud costs once the free credits are used."]},"zh-Hans":{"summaryShort":"一个面向 AI 代理的记忆服务器:存储别人告诉它的内容,之后把有用的部分取出来。","summary":"一个你自己运行的 AI 代理记忆系统。你的程序可以存入文本(retain)、之后提出问题(recall),或者要求根据存储的一切给出一个综合答案(reflect)。事实、经历和背景摘要都保存在 PostgreSQL 数据库中,并提供 Python、Node、Go 的客户端和一个命令行工具,还有面向编程代理、代理框架和 MCP 的现成接入方式。适合正在构建需要记忆能力的代理或聊天机器人的开发者。只需一条 Docker 命令或 pip install hindsight-api,再加上 25 多个模型提供商中任意一个的密钥即可开始;本地模型也可以使用。","easeRationale":"在 Windows、Mac 和 Linux 上均为 C 级:最短路径是 pip install hindsight-api,加一个带模型密钥的环境变量和一条命令,数据库是内置的(README 的平台表把内置数据库标记为三个平台都支持)。Docker 方式则换成了一条很长的命令。发布文件只有适用于 Linux 和 Mac 的小型命令行客户端,Intel 版 Mac 还需要单独的 hindsight-all-slim 软件包。","costNote":"代码采用 MIT 协议,自行托管是免费的,但每次 retain、recall 和 reflect 都会调用一个语言模型,因此你需要向该服务商付费。README 说本地模型(ollama、lmstudio、llamacpp)以及你已有的订阅(ChatGPT、Claude、Cursor、GitHub Copilot)也可以使用。项目背后的公司还销售 Hindsight Cloud,一个按用量付费的托管版本。","notVerified":["维持一个繁忙代理的记忆更新,实际会消耗多少模型 token 费用。","README 中的基准测试数字在你自己的数据上能否重现。","随着记忆不断累积,数据库会占用多少磁盘空间。","pip 和内置数据库的方式在 Windows 上是否真的像平台表所说那样工作。","当多个用户共用一台服务器时,'bank'之间的隔离('无跨 bank 泄露')在实际中是否成立。","免费额度用完后,Hindsight Cloud 的实际费用是多少。"]}}},{"id":"R_kgDOQ3j5Tg","slug":"vercel-labs__agent-browser","name":"Agent Browser","owner":"vercel-labs","fullName":"vercel-labs/agent-browser","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"an_ai_agent_can_act_inside_your_logged_in_accounts","severity":"attention","evidence":"命令列表中包括 \"cookies\"、\"storage\"、\"auth\"、\"profiles\" 和 \"eval\"，README 还提到你可以 \"attach to a running Chrome with `--cdp` or `--auto-connect`\"。因此，驱动它的助手可以读取你的 cookie，在页面上运行自己的 JavaScript，并使用你已经登录的网站。项目在其 Security 页面中记录了相关防护措施：\"the auth vault, domain allowlists, action policies, and content boundaries\"。","source":"ai","title":"ai 代理可以在你已登录的账户内操作"},{"code":"the_agent_follows_instructions_it_reads_from_web_pages","severity":"attention","evidence":"整个流程就是代理读取一个页面并据此行动：\"`snapshot` returns the page as compact text with a ref such as `@e1` for each element. The agent acts on a ref, reads the page again to check the result, and repeats.\" 被访问页面上的文字有可能试图向代理下达指令。文档在 Security 页面中列出了 \"content boundaries\"，这说明项目方意识到了这一点。","source":"ai","title":"代理会执行它从网页上读到的指令"},{"code":"a_script_runs_when_the_package_is_installed","severity":"info","evidence":"package.json 中：\"postinstall\": \"node scripts/postinstall.js\"。通过 npm 安装会运行这个脚本，它会为你的系统挑选并配置相应的原生二进制文件。","source":"ai","title":"安装该软件包时会运行一个脚本"},{"code":"it_downloads_a_separate_browser","severity":"info","evidence":"README 的安装部分写道：\"agent-browser install           # download Chrome for Testing (first time only)\"。发布的二进制文件大小为 15 到 18 MB，但之后下载的浏览器要大得多，并不计入这个大小。","source":"ai","title":"它会下载一个独立的浏览器"},{"code":"still_before_version_1_0_and_changing_fast","severity":"info","evidence":"最新版本是 v0.41.2（2026-11-03），自该仓库在 2026-01-11 创建以来已经有 120 个版本，还有 542 个未关闭的 issue。命令和参数在不同版本之间仍可能发生变化。","source":"ai","title":"版本号还在 1.0 之前，变化很快"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":17,"maxMB":52,"basis":"installer"},"apiKeys":["Cloud browser providers such as Browserbase, Browserless, Browser Use, Kernel or AgentCore"],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","development","ai_skills"],"alternativeTo":[{"slug":"browserbase","name":"Browserbase","confirmed":true},{"slug":"browser-use-cloud","name":"Cloud","confirmed":true}],"declaredAlternatives":[],"stars":49121,"downloads":667133,"downloadsWeek":0,"downloadsDays":0,"growth":3716,"growthDays":7,"history":[{"date":"2026-11-10","stars":49121}],"firstPublishedAt":"2026-11-09T23:02:44.507Z","commitSha":"e7519957b9067c541825afa21a792fbe1f483047","verifiedAt":"2026-11-09T22:49:43.846Z","lastCommitAt":"2026-11-10T19:56:18Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Gives an AI agent a browser it can drive from the command line: open pages, read them as text, click and fill forms.","summary":"agent-browser is a command-line tool from Vercel Labs that lets an AI assistant use a real browser. The agent asks for a snapshot, gets the page as short text with a reference for each element, then clicks, types, uploads files, takes screenshots or records the screen. The browser stays open between commands. It is for developers and for people running coding agents such as Claude Code or Codex; it also works as an MCP server. You install it with npm (Node 22 or newer) on Windows, Mac or Linux, then run one command to download its browser.","easeRationale":"It is a terminal tool with no window of its own: on all three systems you install Node 22 or newer, run \"npm install -g agent-browser\" and then \"agent-browser install\" to download Chrome for Testing. On Mac there is also a Homebrew package, and on Linux \"--with-deps\" adds the browser's system libraries.","costNote":"The tool is free (Apache-2.0) and drives a browser on your own computer. The optional cloud browser providers it can connect to (Browserbase, Browserless, Browser Use, Kernel, AgentCore) are paid services.","notVerified":["How much disk space Chrome for Testing adds on each system, on top of the 15 to 18 MB binary.","Whether the guards described on the Security page (auth vault, domain allowlists, action policies) are on by default or have to be configured.","What the background daemon keeps on disk between sessions when session persistence is turned on, including cookies and signed-in state.","Whether the published binaries and the npm package match the source in this repository.","What the optional cloud browser providers cost and what they do with the pages your agent visits.","How well it works on Windows, since the extra setup notes in the README cover Mac and Linux."]},"zh-Hans":{"summaryShort":"让 AI 代理拥有一个可以从命令行驱动的浏览器：打开页面、把页面读成文本、点击并填写表单。","summary":"agent-browser 是 Vercel Labs 推出的一个命令行工具，让 AI 助手可以使用一个真实的浏览器。代理请求一次快照（snapshot），得到的页面是简短的文本，每个元素都带有一个引用标识，然后代理可以点击、输入文字、上传文件、截图或录制屏幕。浏览器在多次命令之间保持打开状态。它面向开发者，以及运行像 Claude Code 或 Codex 这类编程代理的人；它也可以作为 MCP 服务器使用。你可以在 Windows、Mac 或 Linux 上通过 npm 安装它（需要 Node 22 或更新版本），然后运行一条命令来下载它所使用的浏览器。","easeRationale":"这是一个没有自己窗口的终端工具：在这三个系统上，你都要安装 Node 22 或更新版本，运行 \"npm install -g agent-browser\"，然后运行 \"agent-browser install\" 来下载 Chrome for Testing。在 Mac 上还有一个 Homebrew 安装包，在 Linux 上加上 \"--with-deps\" 会附带安装浏览器所需的系统库。","costNote":"该工具是免费的（Apache-2.0 许可），并在你自己的电脑上驱动浏览器。它可以连接的那些可选的云端浏览器服务商（Browserbase、Browserless、Browser Use、Kernel、AgentCore）是付费服务。","notVerified":["除了 15 到 18 MB 的二进制文件本身之外，Chrome for Testing 在每个系统上还会额外占用多少磁盘空间。","Security 页面中描述的那些防护措施（身份验证保险库、域名白名单、操作策略）是默认开启的，还是需要手动配置。","当会话持久化开启时，后台守护进程在会话之间会在磁盘上保留哪些内容，包括 cookie 和登录状态。","发布的二进制文件和 npm 包是否与这个仓库中的源代码一致。","那些可选的云端浏览器服务商收费多少，以及它们会如何处理你的代理访问过的页面。","它在 Windows 上效果如何，因为 README 里额外的安装说明只涉及 Mac 和 Linux。"]}}},{"id":"MDEwOlJlcG9zaXRvcnk2MjM2NzU1OA==","slug":"vercel__hyper","name":"Hyper","owner":"vercel","fullName":"vercel/hyper","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2023-07-13(1210 天前)","source":"check","title":"版本更新停滞"},{"code":"archived","severity":"info","evidence":"该仓库已被归档(只读,不再维护)","source":"check","title":"已归档"},{"code":"known_vulnerabilities","severity":"attention","evidence":"固定版本的依赖存在已知漏洞(OSV):lodash: GHSA-r5fr-rjxr-66jc, GHSA-xxjr-mmjv-4gpg; uuid: GHSA-w5hq-g745-h8pq","source":"check","title":"已知漏洞"},{"code":"only_prerelease_builds_recently","severity":"info","evidence":"最新的三个发布版本是 'v4.0.0-canary.5'(2023-07-13)、'v4.0.0-canary.4' 和 'v4.0.0-canary.3'——都是测试构建,不是稳定版本。","source":"ai","title":"最近只有预发布版本"},{"code":"outdated_browser_engine_inside_the_app","severity":"attention","evidence":"package.json 构建这个应用时用的是 \"electron\": \"22.3.25\",这是 2022 年的 Electron 版本。Electron 只维护最近的三个主版本,所以 2023 年发布的这个版本里的 Chrome 引擎已经不再获得修复。","source":"ai","title":"应用内置的浏览器引擎已过时"},{"code":"third_party_plugins","severity":"info","evidence":"顶层的 'PLUGINS.md' 和 README 都链接到了一个 'Sample Extension' 和 'Awesome Hyper';插件和主题都是别人写的代码,会在终端里运行。","source":"ai","title":"第三方插件"},{"code":"install_from_elsewhere_may_be_even_older","severity":"info","evidence":"README:\"The version available on Homebrew Cask, Chocolatey, Snapcraft or the AUR may not be the latest\",而这个项目自己最新的构建本身也已经是 2023 年的了。","source":"ai","title":"从别处安装可能版本更旧"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["node"],"disk":{"minMB":146,"maxMB":439,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"warp","name":"Warp","confirmed":true},{"slug":"termius","name":"Termius","confirmed":true}],"declaredAlternatives":[],"stars":44753,"downloads":57179867,"downloadsWeek":37436,"downloadsDays":7,"growth":0,"growthDays":7,"history":[{"date":"2026-10-26","stars":44753},{"date":"2026-10-27","stars":44756},{"date":"2026-10-28","stars":44758},{"date":"2026-10-29","stars":44760},{"date":"2026-10-30","stars":44763},{"date":"2026-10-31","stars":44762},{"date":"2026-11-01","stars":44763},{"date":"2026-11-02","stars":44764},{"date":"2026-11-03","stars":44762},{"date":"2026-11-04","stars":44762},{"date":"2026-11-05","stars":44760},{"date":"2026-11-06","stars":44757},{"date":"2026-11-07","stars":44755},{"date":"2026-11-08","stars":44754},{"date":"2026-11-09","stars":44753},{"date":"2026-11-10","stars":44753}],"firstPublishedAt":"2026-10-27T01:10:12.704Z","commitSha":"890a332ed9c35ed87343647d3ec6611b8045b157","verifiedAt":"2026-11-04T22:42:23.640Z","lastCommitAt":"2026-11-04T19:17:12Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Hyper-Setup-3.4.1.exe","sizeBytes":73828760,"url":"https://github.com/vercel/hyper/releases/download/v3.4.1/Hyper-Setup-3.4.1.exe","releaseUrl":"https://github.com/vercel/hyper/releases/tag/v3.4.1"},{"os":"mac","file":"Hyper-3.4.1-mac-arm64.dmg","sizeBytes":92046768,"url":"https://github.com/vercel/hyper/releases/download/v3.4.1/Hyper-3.4.1-mac-arm64.dmg","releaseUrl":"https://github.com/vercel/hyper/releases/tag/v3.4.1"},{"os":"linux","file":"Hyper-3.4.1.AppImage","sizeBytes":97878022,"url":"https://github.com/vercel/hyper/releases/download/v3.4.1/Hyper-3.4.1.AppImage","releaseUrl":"https://github.com/vercel/hyper/releases/tag/v3.4.1"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A customizable terminal window app, built with web technology, where you type commands with themes and plugins.","summary":"Hyper is a terminal app: the window where you type text commands, like Command Prompt or PowerShell, but restyled with web technology so you can change its look with themes and add features with plugins. It is for people who use the command line, mostly programmers. Installers exist for Windows, macOS and Linux, and it can also be installed with Chocolatey, Homebrew, Snap or the AUR. The newest builds on GitHub are test versions (4.0.0 canary) from July 2023, and the repository is now archived: read-only and no longer maintained.","easeRationale":"A on Windows: the releases include ready .exe installers (x64 and arm64, about 73 MB) that you download and open. Mac gets .dmg files and Linux gets .deb, .rpm, .snap and AppImage builds, so each system has a ready download — but all of them are the July 2023 canary builds.","costNote":"Free under the MIT license.","notVerified":["Whether the 2023 canary build runs well on current Windows 11.","Whether the known vulnerable lodash and uuid versions are reachable in the shipped app.","Memory use compared with lighter terminals, since it is built on Electron."]},"zh-Hans":{"summaryShort":"一款可定制的终端窗口应用,用网页技术构建,配有主题和插件,供你在里面输入命令。","summary":"Hyper 是一款终端应用:就是你输入文字命令的那个窗口,类似命令提示符或 PowerShell,但用网页技术重新设计了外观,这样你可以用主题改变它的样子,用插件添加功能。它适合使用命令行的人,主要是程序员。Windows、macOS 和 Linux 都有安装程序,也可以通过 Chocolatey、Homebrew、Snap 或 AUR 安装。GitHub 上最新的构建版本是 2023 年 7 月的测试版(4.0.0 canary),该仓库现在已被归档:只能读取,不再维护。","easeRationale":"Windows 上是 A 级:发布版本里有现成的 .exe 安装程序(x64 和 arm64,大约 73 MB),下载打开即可。Mac 拿到的是 .dmg 文件,Linux 拿到的是 .deb、.rpm、.snap 和 AppImage 构建,所以每个系统都有现成的下载——但全部都是 2023 年 7 月的 canary 测试版。","costNote":"采用 MIT 许可证,免费。","notVerified":["2023 年的 canary 测试版在当前的 Windows 11 上运行是否顺畅。","已知存在漏洞的 lodash 和 uuid 版本,在发布的应用里是否真的会被用到。","和更轻量的终端相比,内存占用如何,因为它是基于 Electron 构建的。"]}}},{"id":"R_kgDOVegS6A","slug":"verlintas__noir","name":"NOIR","owner":"Verlintas","fullName":"Verlintas/NOIR","type":"needs_setup","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"attention","signals":[{"code":"downloads_tools_at_run_time","severity":"attention","evidence":"README，第一次运行部分：\"检查 `adb`，没有就问你要不要自动下载（`~/.noir/tools`）\" —— 安装之后，程序会从网络获取 `adb`，以及用于深度扫描的 `jadx`，所以在你机器上运行的不只是发行文件里包含的内容。","source":"ai","title":"运行时下载工具"},{"code":"changes_settings_on_your_phone","severity":"attention","evidence":"README 使用列表：`noir auto --apply` 会在手机上执行 app-ops 更改，`noir auto --apply --aggressive` \"额外卸载被标记的用户应用\"。README 说会写入一个带撤销命令的 `rollback.txt`，默认情况下会读取它。","source":"ai","title":"修改你手机上的设置"},{"code":"requires_usb_debugging_on_the_phone","severity":"info","evidence":"README 要求你打开「开发者选项 → USB 调试」，并为电脑勾选「一律允许」。USB 调试保持开启意味着任何插入的电脑都能获得对手机的广泛访问权限。","source":"ai","title":"需要在手机上开启 usb 调试"},{"code":"very_new_repository","severity":"info","evidence":"项目数据：创建于 2026-10-31，已有 1356 个星标，9 个发行版，最新标签 v0.2.5 发布于 2026-11-02。目前还没有任何历史记录，版本号也低于 1.0。","source":"ai","title":"仓库非常新"},{"code":"results_are_static_evidence_only","severity":"info","evidence":"README 免责声明：该工具给出的是静态证据（\"这个应用具备这个能力，并且代码里有这个端点\"），并且 \"不能证明某次具体传输发生过\" —— 它无法证明某次数据传输真的发生了，所以高分并不能证明存在窃密行为。","source":"ai","title":"结果只是静态证据"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":59,"maxMB":176,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["hardware","network_system"],"alternativeTo":[{"slug":"iverify","name":"iVerify","confirmed":true}],"declaredAlternatives":[],"stars":3227,"downloads":15154,"downloadsWeek":10633,"downloadsDays":7,"growth":1468,"growthDays":7,"history":[{"date":"2026-11-03","stars":1759},{"date":"2026-11-04","stars":2001},{"date":"2026-11-05","stars":2439},{"date":"2026-11-06","stars":2741},{"date":"2026-11-07","stars":2885},{"date":"2026-11-08","stars":3083},{"date":"2026-11-09","stars":3153},{"date":"2026-11-10","stars":3227}],"firstPublishedAt":"2026-11-03T02:51:37.708Z","commitSha":"c0611f2c91f6eda0236354b286d9ca4dfae75f8d","verifiedAt":"2026-11-03T02:36:04.802Z","lastCommitAt":"2026-11-07T16:07:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Checks an Android phone from your computer by USB cable: app permissions, which servers apps contact, and reversible fixes.","summary":"A desktop tool that inspects an Android phone connected by USB cable. It lists which apps have sensitive permissions, pulls the web addresses found inside each app, scores them by risk, and writes a report plus a list of tracker domains. It can also turn off single permissions and save a file with the commands to undo every change. You need a computer with Windows, macOS or Linux, a phone with USB debugging turned on, and the Android `adb` tool, which the program offers to download on first run. The README and screens are mostly in Chinese.","easeRationale":"On Windows, macOS and Linux you unzip the portable package and double-click a launcher (`Start NOIR.bat`, `Start NOIR.command`, `Start-NOIR.sh`); the README says the Java runtime is inside the package. The extra steps are the same everywhere: turn on developer mode and USB debugging on the phone, and answer yes when the program asks to download `adb`.","notVerified":["Whether the double-click launcher really works on each system, especially past the Windows SmartScreen and macOS quarantine warnings the README mentions","Whether the `adb` and `jadx` files the program downloads on first run are the genuine ones","How accurate the risk scores and vendor labels are on a real phone, and how many apps are flagged by mistake","Real disk use of the deep scan, which decompiles whole apps into a cache under `~/.noir`","Whether turning permissions off breaks apps on your phone, and whether the saved undo commands fully restore them"]},"zh-Hans":{"summaryShort":"通过 USB 线从你的电脑检查 Android 手机：应用权限、应用联系的服务器，以及可逆的修复操作。","summary":"一个通过 USB 线连接 Android 手机进行检查的桌面工具。它列出哪些应用拥有敏感权限，提取每个应用内部找到的网址，按风险打分，并写出一份报告和一份跟踪域名列表。它还可以关闭单个权限，并保存一个包含撤销每次更改所需命令的文件。你需要一台 Windows、macOS 或 Linux 电脑，一台开启了 USB 调试的手机，以及 Android 的 `adb` 工具，程序会在首次运行时提供下载。README 和界面大多是中文。","easeRationale":"在 Windows、macOS 和 Linux 上，你解压便携包并双击启动器（`Start NOIR.bat`、`Start NOIR.command`、`Start-NOIR.sh`）；README 说 Java 运行环境已经包含在包里。各平台的额外步骤相同：在手机上打开开发者模式和 USB 调试，并在程序询问是否下载 `adb` 时回答是。","notVerified":["双击启动器在每个系统上是否真的能工作，尤其是在越过 README 提到的 Windows SmartScreen 和 macOS 隔离警告之后","程序在首次运行时下载的 `adb` 和 `jadx` 文件是否是正版的","在真实手机上，风险评分和厂商标签有多准确，有多少应用被错误标记","深度扫描的实际磁盘占用，它会把整个应用解包到 `~/.noir` 下的缓存中","关闭权限是否会让手机上的应用出问题，以及保存的撤销命令能否完全恢复它们"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMjM0MjY5MzQ=","slug":"videolan__vlc-android","name":"VLC for Android","owner":"videolan","fullName":"videolan/vlc-android","type":"ready_app","status":"active","grade":"A","easeByPlatform":{},"risk":"low","signals":[{"code":"no_downloads_in_this_repository","severity":"info","evidence":"该仓库没有 GitHub 发布版，也没有 APK：里面只有源码和构建脚本（buildsystem/compile.sh）。README.md 指向 videolan.org 和 code.videolan.org，因此成品应用来自 VideoLAN 官网，而不是这里。","source":"ai","title":"本仓库里没有可下载的程序"},{"code":"development_happens_elsewhere","severity":"info","evidence":"README.md：\"Pull requests must be proposed on our gitlab server\"，缺陷追踪也在 code.videolan.org，因此这个 GitHub 仓库只是 VideoLAN 自建平台上项目的一个副本。","source":"ai","title":"开发其实在别处进行"},{"code":"protected_disc_playback_rules_vary","severity":"info","evidence":"README.md 在 LibVLC 的功能里列出了 \"Playback of Audio CD, DVD and Bluray with menu navigation\"；播放带防拷贝保护的商业光盘需要额外的解密组件，而各国对此的规定并不相同。","source":"ai","title":"受保护光盘的播放规定各地不同"},{"code":"copyleft_license","severity":"info","evidence":"README.md 的 License：\"VLC for Android is licensed under GPLv2 (or later). Android libraries make this, de facto, a GPLv3 application\"，而 LibVLC 引擎采用 LGPLv2——如果你想基于这份代码发布自己的应用，这一点很重要。","source":"ai","title":"copyleft 许可证"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"mx-player-pro","name":"MX Player Pro","confirmed":true},{"slug":"plex","name":"Plex","confirmed":true}],"declaredAlternatives":[],"stars":4135,"downloads":0,"downloadsWeek":0,"downloadsDays":6,"growth":6,"growthDays":6,"history":[{"date":"2026-11-04","stars":4129},{"date":"2026-11-05","stars":4129},{"date":"2026-11-06","stars":4130},{"date":"2026-11-07","stars":4130},{"date":"2026-11-08","stars":4131},{"date":"2026-11-09","stars":4132},{"date":"2026-11-10","stars":4135}],"firstPublishedAt":"2026-11-11T00:01:13.675Z","commitSha":"14a7422459ca809b29cad567c3a59702178cb423","verifiedAt":"2026-11-10T23:57:02.524Z","lastCommitAt":"2026-11-09T13:39:51Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"The official VLC media player for Android phones, tablets, Android TV and ChromeOS.","summary":"This is the source code of VLC for Android, Android TV and ChromeOS, made by the VideoLAN team. The app plays practically any video or audio file, codec and stream, keeps a library of your media, opens files on network shares and servers (SMB, FTP, SFTP, NFS, UPnP, DLNA), plays Audio CD, DVD and Blu-ray with menu navigation, handles HDR, 360 video and 3D audio, and can send playback to a Chromecast. It is for anyone with an Android phone, tablet or TV box. This repository holds only the code and the build scripts: the ready app comes from the VideoLAN site.","easeRationale":"It is an Android, Android TV and ChromeOS app, so none of the desktop platforms get a grade; for a user it is a finished app from the VideoLAN site, install and open, which is an A. Building it from this repository is another story: the README asks for a recent Linux distribution, the Android SDK and NDK and a list of compiler packages.","costNote":"Free software under GPLv2 or later; the README notes that the Android libraries make it in practice a GPLv3 application.","notVerified":["Whether the app published by VideoLAN was built from this exact code.","How much space the app and its media library use on a phone.","How well it performs on older phones or on a specific Android TV box.","Which network and disc features are available on each Android version."]},"zh-Hans":{"summaryShort":"VLC 官方播放器，适用于安卓手机、平板、Android TV 和 ChromeOS。","summary":"这是 VideoLAN 团队开发的 VLC for Android、Android TV 和 ChromeOS 的源代码。这款应用几乎能播放任何视频或音频文件、编码格式和流媒体，会整理你的媒体库，能打开网络共享和服务器上的文件（SMB、FTP、SFTP、NFS、UPnP、DLNA），支持带菜单导航的音频 CD、DVD 和蓝光播放，支持 HDR、360 度视频和 3D 音频，还能投屏到 Chromecast。它适合所有拥有安卓手机、平板或电视盒子的人。本仓库只存放源码和构建脚本：成品应用要从 VideoLAN 官网获取。","easeRationale":"这是面向 Android、Android TV 和 ChromeOS 的应用，所以桌面平台都不给等级；对用户而言，它就是从 VideoLAN 官网下载、安装即用的成品应用，因此给 A。但从这个仓库自行编译就是另一回事：README 要求较新的 Linux 发行版、Android SDK 和 NDK，以及一长串编译工具包。","costNote":"自由软件，采用 GPLv2 或更高版本；README 指出，由于用到安卓的库，它实际上是一款 GPLv3 应用。","notVerified":["VideoLAN 发布的应用是否正是用这份代码构建的。","应用和它的媒体库在手机上会占多少空间。","在较老的手机或某款具体的 Android TV 盒子上表现如何。","各个安卓版本上分别能使用哪些网络和光盘相关功能。"]}}},{"id":"R_kgDOSRUoTw","slug":"virgiliojr94__book-to-skill","name":"Book to Skill","owner":"virgiliojr94","fullName":"virgiliojr94/book-to-skill","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"low","signals":[{"code":"copyright_of_processed_books","severity":"info","evidence":"README.md 的 \"Copyright & fair use\"：工具本身不含任何书籍内容，并要求你 \"Bring a book you bought\"，同时提醒 \"Don't redistribute... Publishing or sharing a generated skill of a copyrighted work can infringe the rights holder.\"","source":"ai","title":"被处理书籍的版权问题"},{"code":"can_publish_to_your_github","severity":"info","evidence":"README.md 的 Usage：\"After a conversion, the converter can publish the skill to GitHub (private by default) so any host installs it with npx skills add\"——也就是说它能在你的账号下创建仓库。","source":"ai","title":"可以发布到你的 GitHub"},{"code":"writes_into_assistant_folders","severity":"info","evidence":"README.md：生成的技能会写到 ~/.agents/skills/<slug>/，在 Claude Code 下 \"the converter also attempts a verified symlink at ~/.claude/skills/<slug>/\"；其他助手也各有自己的目录。","source":"ai","title":"会写入助手的目录"},{"code":"separate_security_notice_file","severity":"info","evidence":"仓库在常规的 SECURITY.md 之外还放了一份 SECURITY-NOTICE.md；本次材料里没有包含它的内容，安装前值得先读一读。","source":"ai","title":"另有一份 SECURITY-NOTICE 文件"},{"code":"author_measured_benefit","severity":"info","evidence":"\"24x-51x fewer tokens than dumping the book into context\" 这一宣传语来自作者在 docs/performance.md 里的自测，本次没有核实。","source":"ai","title":"收益数字由作者自己测得"}],"platforms":["windows","mac","linux"],"tools":["python","node","other"],"buildTools":["python"],"disk":{"minMB":100,"maxMB":300,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","productivity","office"],"alternativeTo":[{"slug":"notebooklm","name":"NotebookLM","confirmed":true}],"declaredAlternatives":[],"stars":37544,"downloads":0,"downloadsWeek":0,"downloadsDays":5,"growth":671,"growthDays":7,"history":[{"date":"2026-11-05","stars":37057},{"date":"2026-11-06","stars":37140},{"date":"2026-11-07","stars":37210},{"date":"2026-11-08","stars":37338},{"date":"2026-11-09","stars":37476},{"date":"2026-11-10","stars":37544}],"firstPublishedAt":"2026-11-10T23:52:44.859Z","commitSha":"e7bb659d3261491c8b28793dcb9102fa64a46c22","verifiedAt":"2026-11-10T23:40:24.569Z","lastCommitAt":"2026-11-03T17:53:10Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Turns a book or a folder of documents you own into a skill your AI assistant can consult while you work.","summary":"book-to-skill is a skill for AI coding assistants (Claude Code, GitHub Copilot CLI, Codex, OpenCode and others). You point it at a PDF, EPUB, DOCX or a folder of documents and it builds a structured set of notes: a main file with the key ideas, one file per chapter, a glossary, patterns and a cheat sheet, saved in your assistant's skills folder. Later you ask about a topic and the assistant reads only the right chapter. It ships no book content: you bring files you own, and everything is extracted on your machine with Python.","easeRationale":"On all three systems it is a C: you need an AI assistant that reads skills, then one command (npx skills add virgiliojr94/book-to-skill, which needs Node) or a git clone into the skills folder, plus Python 3.9+ and a pip install for the extractor your file type needs (pypdf, docling, ebooklib and so on). On Windows the extra tools shown in the README (poppler's pdftotext via apt, Calibre for MOBI) take more work to set up.","costNote":"MIT code with no fees; the author accepts GitHub sponsors. The conversion runs through your AI assistant, so whatever that assistant's model costs you still applies.","notVerified":["Whether the generated notes are faithful to the book, and what gets lost in the distillation.","The token savings claimed by the author.","How long a conversion takes and how much it costs in model usage for a long book (docling is listed at about 1.5 seconds per page).","Whether it behaves the same in every assistant it lists (Copilot CLI, Amp, Codex, Hermes, OpenCode, OpenClaw).","What the SECURITY-NOTICE.md file in the repository says."]},"zh-Hans":{"summaryShort":"把你自己的一本书或一个文档目录，变成 AI 助手在你工作时可随时查阅的技能。","summary":"book-to-skill 是给 AI 编程助手（Claude Code、GitHub Copilot CLI、Codex、OpenCode 等）用的技能。你指向一个 PDF、EPUB、DOCX 或一个文档目录，它就会生成一套结构化笔记：一份含核心要点的主文件、每章一个文件、术语表、常见范式和速查表，保存在助手的技能目录里。之后你提问时，助手只会去读相关的那一章。它不附带任何书籍内容：你提供自己拥有的文件，全部提取都在你本机用 Python 完成。","easeRationale":"三个系统都是 C：你需要一个支持技能的 AI 助手，然后执行一条命令（npx skills add virgiliojr94/book-to-skill，需要 Node），或者把仓库 git clone 进技能目录，另外还要 Python 3.9 以上，并按文件类型 pip 安装相应的提取器（pypdf、docling、ebooklib 等）。在 Windows 上，README 里提到的额外工具（通过 apt 安装的 poppler pdftotext、处理 MOBI 的 Calibre）配置起来更费事。","costNote":"代码采用 MIT 许可，不收费；作者接受 GitHub 赞助。转换过程要通过你的 AI 助手完成，因此该助手模型的费用照样会产生。","notVerified":["生成的笔记是否忠实于原书，提炼过程中又丢失了什么。","作者宣称的 token 节省幅度。","处理一本长书要多久、模型调用要花多少钱（docling 的速度标注为每页约 1.5 秒）。","在它列出的各个助手里（Copilot CLI、Amp、Codex、Hermes、OpenCode、OpenClaw）表现是否一致。","仓库里的 SECURITY-NOTICE.md 文件写了什么。"]}}},{"id":"R_kgDOP_Lcdw","slug":"vita-epfl__stable-video-infinity","name":"Stable Video Infinity","owner":"vita-epfl","fullName":"vita-epfl/Stable-Video-Infinity","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"linux":"D"},"risk":"attention","signals":[{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：torch: GHSA-3749-ghw9-m3mg, GHSA-53q9-r3pm-6pq6, GHSA-63cw-57p8-fm3p; transformers: GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm, GHSA-4w7r-h757-3r74","source":"check","title":"存在已知漏洞"},{"code":"talking_head_from_photo_and_audio","severity":"attention","evidence":"SVI-Talk 模型接受 \"Image + Audio\"，输出 \"Talking video\"（标签：audio-driven-talking-face），也就是能生成某人说出其从未说过的话的视频；README 里没有任何关于本人同意或滥用的提示。","source":"ai","title":"用照片加音频生成数字人说话视频"},{"code":"cartoon_variant_of_known_characters","severity":"attention","evidence":"其中一个变体被列为 \"SVI-Tom&Jerry — Cartoon animation\"，README 还链接了一个 \"8-minute crazy Tom & Jerry video made with SVI-Tom\"。svi-tom.safetensors 并不在仓库里：README 用 huggingface-cli 从 Hugging Face（vita-video-gen/svi-model）下载。卡通角色通常受版权保护，而权利方的条款不在本项目的 MIT 许可范围内。","source":"ai","title":"带知名卡通角色的动画变体"},{"code":"heavy_hardware_and_downloads","severity":"info","evidence":"README.md：\"We have tested the environment with A100 80G, cuda 12.0\"，weights/ 目录结构里列出了分成七个 safetensors 分片的 Wan 2.1 I2V 14B 基础模型，以及说话和舞蹈变体所需的额外编码器。","source":"ai","title":"硬件要求高，下载量大"},{"code":"third_party_workflows_and_hosting","severity":"info","evidence":"实际使用中很大一部分要依靠他人的成果：社区做的 ComfyUI 工作流（并带有 \"Avoid using the wrong SVI 1.0 workflow in this repo\" 这样的提醒，以及量化测试版本），README 还写道 \"SVI-2.0 Pro is now available on the Poe platform! You can access it through the Poe chat interface or integrate it via their API\"，并感谢 @empiriolabsai 的支持。","source":"ai","title":"依赖第三方的工作流和托管"},{"code":"training_data_from_third_parties","severity":"info","evidence":"列出的训练数据集包括 MixKit、UltraVideo、Hallo 3 的一个 5k 子集，以及来自 Kaggle 的 TikTok 舞蹈数据集，各自都有自己的条款。","source":"ai","title":"训练数据来自第三方"},{"code":"author_claims_not_measured","severity":"info","evidence":"\"This will never drift or forget in our 20 min test\" 和 \"infinite length\" 之类的说法，都出自作者自己的测试。","source":"ai","title":"这些说法没有独立验证"}],"platforms":["linux"],"tools":["python","gpu_nvidia","ffmpeg"],"buildTools":["python","gpu_any","ffmpeg"],"disk":{"minMB":1110,"maxMB":4870,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["video","local_ai"],"alternativeTo":[{"slug":"runway","name":"Runway","confirmed":true},{"slug":"kling","name":"Kling","confirmed":true},{"slug":"pika","name":"Pika","confirmed":true},{"slug":"heygen","name":"HeyGen","confirmed":true},{"slug":"d-id","name":"D-ID","confirmed":true},{"slug":"google-veo-3","name":"Veo 3","confirmed":true}],"declaredAlternatives":[],"stars":2620,"downloads":0,"downloadsWeek":0,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":2620},{"date":"2026-11-08","stars":2620},{"date":"2026-11-09","stars":2620},{"date":"2026-11-10","stars":2620}],"firstPublishedAt":"2026-11-11T00:01:13.691Z","commitSha":"557afca50d5766514c67ae39000b03e0dd66d6e5","verifiedAt":"2026-11-10T23:57:48.402Z","lastCommitAt":"2026-09-17T06:02:06Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Research code that makes very long AI videos from a photo and text, without the usual drift in quality.","summary":"Stable Video Infinity, from the VITA lab at EPFL, is research code that extends an existing video model (Wan 2.1) so it can keep generating minutes-long clips from one picture plus a stream of text prompts, instead of a few seconds. Variants do cartoons, dancing from a skeleton and talking heads driven by audio. It is run from the terminal on Linux with a strong NVIDIA card (the team tested an A100 80 GB) after downloading tens of gigabytes of model files, or through community ComfyUI workflows. Training code and datasets are included.","easeRationale":"Linux only in practice, and a D: you create a conda environment, run pip install -e ., install flash_attn (the README links an issue for when that fails), download the Wan 2.1 14B base model plus the SVI adapters with huggingface-cli, and launch bash scripts or the Gradio demo. The environment the authors tested is an A100 with 80 GB of video memory.","costNote":"MIT code and freely downloadable model files; the base Wan 2.1 weights and the datasets come with their own terms. The README also says SVI-2.0 Pro is available on the Poe platform, through its chat interface or its API, hosted with support from @empiriolabsai.","notVerified":["How much video memory you really need on a consumer graphics card, and whether it runs at all below the tested A100 80 GB.","Real disk space: the base Wan 2.1 14B model and the extra encoders are far larger than the estimate in the checks.","How long one minute of video takes to generate.","Whether it works on Windows or macOS; the instructions are Linux shell commands with flash_attn and CUDA 12.","Whether the quality in the community videos is reproducible with your own images and prompts.","What material each variant was fine-tuned on: the README lists the training datasets for the family, but not for svi-tom."]},"zh-Hans":{"summaryShort":"研究代码：用一张照片加文字生成超长 AI 视频，而不会出现常见的画质漂移。","summary":"Stable Video Infinity 来自洛桑联邦理工学院（EPFL）的 VITA 实验室，是一份研究代码，它扩展了已有的视频模型 Wan 2.1，使其能够从一张图片加上一连串文字提示，持续生成数分钟长的片段，而不只是几秒。它还有动画、按骨骼驱动的舞蹈，以及由音频驱动的数字人等变体。使用方式是在装有高性能 NVIDIA 显卡的 Linux 上通过终端运行（团队测试用的是 80 GB 的 A100），并先下载数十 GB 的模型文件，也可以走社区做的 ComfyUI 工作流。训练代码和数据集都包含在内。","easeRationale":"实际上只支持 Linux，给 D：你要创建 conda 环境，运行 pip install -e .，安装 flash_attn（README 附了安装失败时可参考的 issue 链接），用 huggingface-cli 下载 Wan 2.1 14B 基础模型和 SVI 适配器，然后运行 shell 脚本或 Gradio 演示。作者测试的环境是一块 80 GB 显存的 A100。","costNote":"代码采用 MIT 许可，模型文件可免费下载；Wan 2.1 的基础权重和各数据集另有自己的条款。README 还提到 SVI-2.0 Pro 已上线 Poe 平台，可通过 Poe 的聊天界面或 API 使用，托管由 @empiriolabsai 提供支持。","notVerified":["在消费级显卡上究竟需要多少显存，低于测试用的 80 GB A100 是否还跑得起来。","真实磁盘占用：Wan 2.1 14B 基础模型和额外的编码器，都远大于检查给出的估算值。","生成一分钟视频需要多长时间。","在 Windows 或 macOS 上能否使用；文档给的是带 flash_attn 和 CUDA 12 的 Linux shell 命令。","社区视频里的画质，换成你自己的图片和提示词是否还能复现。","各个变体分别用什么素材微调：README 列出了该系列的训练数据集，却没有列出 svi-tom 的。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzNDY0MDgxNzc=","slug":"voice-cloning-app__voice-cloning-app","name":"Voice Cloning App","owner":"voice-cloning-app","fullName":"voice-cloning-app/Voice-Cloning-App","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","linux":"D"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最后一次发布是 2022-02-07（1737 天前）","source":"check","title":"发布早已停止"},{"code":"known_vulnerabilities","severity":"attention","evidence":"锁定的依赖存在已知漏洞（OSV）：flask: GHSA-68rp-wp8r-4726, GHSA-m2qf-hxjv-5gpq, PYSEC-2023-62; requests: GHSA-9hjg-9r4m-mvj7, GHSA-9wx4-h78v-vm56, GHSA-gc5v-m9x4-r6x2; torch: GHSA-3749-ghw9-m3mg, GHSA-47fc-vmwq-366v, GHSA-53q9-r3pm-6pq6; nltk: GHSA-2jhm-w3mp-jcwr, GHSA-2ww3-fxvq-293j, GHSA-3gq4-3j92-5w49; werkzeug: GHSA-29vq-49wr-vm6x, GHSA-2g68-c3qc-8985, GHSA-87hc-h4r5-73f7; joblib: GHSA-6hrg-qmvc-2xh8, PYSEC-2022-288; python-socketio: GHSA-5w7q-77mv-v69f, GHSA-g8c6-8fjj-2r4m, PYSEC-2026-1854; flask: GHSA-68rp-wp8r-4726, GHSA-m2qf-hxjv-5gpq, PYSEC-2023-62; requests: GHSA-9hjg-9r4m-mvj7, GHSA-9wx4-h78v-vm56, GHSA-gc5v-m9x4-r6x2; torch: GHSA-3749-ghw9-m3mg, GHSA-47fc-vmwq-366v, GHSA-53q9-r3pm-6pq6; nltk: GHSA-2jhm-w3mp-jcwr, GHSA-2ww3-fxvq-293j, GHSA-3gq4-3j92-5w49; werkzeug: GHSA-29vq-49wr-vm6x, GHSA-2g68-c3qc-8985, GHSA-87hc-h4r5-73f7; joblib: GHSA-6hrg-qmvc-2xh8, PYSEC-2022-288; python-socketio: GHSA-5w7q-77mv-v69f, GHSA-g8c6-8fjj-2r4m, PYSEC-2026-1854","source":"check","title":"存在已知漏洞"},{"code":"voice_cloning_without_consent_notice","severity":"attention","evidence":"它的全部用途就是复制某个人的声音（标签：voice-cloning、tts），README 还提供 \"Automatic dataset generation (with support for subtitles and audiobooks)\" 以及社区做的 \"Youtube data fetching\" 笔记本，却完全没有提到要取得声音主人的许可，也没有提到录音的权利归属。","source":"ai","title":"可克隆声音却无授权提示"},{"code":"end_of_life_python_required","severity":"attention","evidence":"install.md 和 install.sh 要求 Python 3.6（从 deadsnakes PPA \"Installing Python 3.6\"），这个版本 2021 年起就不再收到修复，当前的 Linux 发行版里也没有了。","source":"ai","title":"要求使用已停止支持的 Python"},{"code":"old_web_server_on_local_port","severity":"attention","evidence":"该应用是运行在 http://localhost:5000/ 上的 Flask 网页界面，依赖锁定在 2020 至 2021 年的包（Flask==1.1.2、werkzeug==2.0.3、python-socketio==5.5.0）；Docker 说明用 \"docker run -d -p 5000:5000\" 把该端口发布出去，等于暴露到网络上。","source":"ai","title":"本地端口上跑着老旧的网页服务"},{"code":"windows_build_cannot_train","severity":"info","evidence":"install.md：\"The CPU Only version supports all features except local training\"，而最近三个发布版只提供 cpuonly 的可执行文件（Voice-Cloning-App-1.1.1-cpuonly.exe）。","source":"ai","title":"Windows 版本无法训练"},{"code":"depends_on_outside_services","severity":"info","evidence":"README.md 指向部署在 heroku 上的 \"Voice Sharing Hub\"、用 Google Colab 笔记本做远程训练和合成（部分由社区成员制作），以及用于托管模型的 uberduck.ai 服务。","source":"ai","title":"依赖外部服务"}],"platforms":["windows","linux"],"tools":[],"buildTools":["python","docker","gpu_any"],"disk":{"minMB":450,"maxMB":1351,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"BSD-3-Clause","category":"permissive","detectedFrom":"github"},"categories":["audio","local_ai"],"alternativeTo":[{"slug":"elevenlabs","name":"ElevenLabs","confirmed":true},{"slug":"respeecher","name":"Respeecher","confirmed":true},{"slug":"murf","name":"Murf","confirmed":true},{"slug":"kits-ai","name":"Kits AI","confirmed":true}],"declaredAlternatives":[],"stars":1438,"downloads":19037,"downloadsWeek":4,"downloadsDays":3,"growth":1,"growthDays":6,"history":[{"date":"2026-11-07","stars":1437},{"date":"2026-11-08","stars":1438},{"date":"2026-11-09","stars":1438},{"date":"2026-11-10","stars":1438}],"firstPublishedAt":"2026-11-10T23:52:44.891Z","commitSha":"45d7bc0156f4549ad06df7f86de81d275c0f2a22","verifiedAt":"2026-11-10T23:42:20.194Z","lastCommitAt":"2024-12-02T03:42:25Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Windows app that copies a voice from recordings and then reads any text in that voice, on your own computer.","summary":"Voice Cloning App builds a text-to-speech voice from audio recordings: it cuts the audio into clips with their transcript, trains a model (Tacotron 2) and then reads any text you type in that voice. It runs as a small web page on your own machine at localhost:5000. On Windows there is an installer, but the published one is the CPU-only build, which does not train locally; for training you need an NVIDIA card with 4 GB or more, a source install, or the authors' Colab notebook. The project is old: the last release is from February 2022.","easeRationale":"On Windows you download one .exe from the releases page and the app opens in your browser, but it only becomes useful after you build a dataset and train or import a voice, so it is a B. Linux is a D: the install script wants Python 3.6 from an extra PPA, ffmpeg, an NVIDIA driver and CUDA, or you build the Docker image yourself.","notVerified":["Whether the February 2022 installer still runs on a current Windows.","How good the cloned voice sounds compared with today's services.","How many hours of recordings and how much time on a graphics card a decent voice needs.","Whether the linked Colab notebooks and the voice sharing hub are still online.","Whether a source install still works, given Python 3.6 and torch 1.9 pins."]},"zh-Hans":{"summaryShort":"一款 Windows 应用，从录音中复制出某个声音，再用这个声音在你自己的电脑上朗读任意文本。","summary":"Voice Cloning App 用录音构建一个文字转语音的音色：它把音频切成带文字稿的片段，训练一个模型（Tacotron 2），之后就能用那个声音朗读你输入的任何文字。它以一个小网页的形式运行在你本机的 localhost:5000。Windows 上有安装程序，但公开发布的是仅用处理器的版本，无法在本地训练；要训练就需要 4 GB 以上显存的 NVIDIA 显卡、从源码安装，或使用作者提供的 Colab 笔记本。项目已经很旧：最后一次发布是 2022 年 2 月。","easeRationale":"在 Windows 上，你从发布页下载一个 .exe，应用会在浏览器里打开，但必须先整理数据集、训练或导入一个音色后才真正有用，因此给 B。Linux 则是 D：安装脚本要求从额外 PPA 装 Python 3.6，还要 ffmpeg、NVIDIA 驱动和 CUDA，否则就得自己构建 Docker 镜像。","notVerified":["2022 年 2 月的安装程序在当前的 Windows 上是否还能运行。","与如今的服务相比，克隆出的声音听感如何。","要得到一个像样的音色，需要多少小时录音和多少显卡时间。","文中链接的 Colab 笔记本和声音共享站是否还在线。","鉴于锁定了 Python 3.6 和 torch 1.9，从源码安装是否还行得通。"]}}},{"id":"R_kgDOVb5hFQ","slug":"wangzan101__mika","name":"Mika","owner":"wangzan101","fullName":"wangzan101/Mika","type":"ready_app","status":"active","grade":"C","easeByPlatform":{},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"4 天内获得 5399 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"needs_deep_control_of_the_phone","severity":"attention","evidence":"仓库的主题标签以 \"accessibility\" 开头，README 的第 3 步是 \"pair wireless debugging on the phone once\"。无线调试是 Android 用于控制设备的开发者通道，README 说 Mika \"Opens apps, taps, types and scrolls\" 并且 \"Looks at the screen and finds buttons the way a person does\"，所以它能触及你已登录的应用。","source":"ai","title":"需要对手机进行深度控制"},{"code":"automating_other_apps_may_break_their_rules","severity":"attention","evidence":"README，Privacy & safety：\"Operating third-party apps and websites automatically may break their terms of service, so use your own judgment.\" 项目把这个判断留给了你。它自己的演示之一是 \"Watches ads to unlock music perks\"，还有一些演示代表你在招聘网站和购物网站上行动。","source":"ai","title":"自动操作其他应用可能违反其条款"},{"code":"your_screen_is_sent_to_the_model_provider","severity":"attention","evidence":"README，Privacy & safety：\"There is no Mika server: what you say and what's on your screen go only to the model provider you set up.\" 演示 \"What did the group talk about today?\" 会滑动浏览一个微信群聊，于是私密的聊天内容离开手机，发送给那个提供方。","source":"ai","title":"你的屏幕内容被发送给模型提供方"},{"code":"acts_without_asking_by_default","severity":"attention","evidence":"README：\"By default Mika runs in Ask mode: it checks with you only before spending money (placing orders, paying) and just does everything else. You can switch to Autonomous mode instead.\" 也就是说，所有不花钱的操作都不需要确认就会执行。","source":"ai","title":"默认情况下未经询问就行动"},{"code":"very_new_and_moving_fast","severity":"info","evidence":"该仓库创建于 2026-10-29，已有 36 个发行版，最新的 v0.1.34 发布于 2026-11-02——版本号还在 1.0 之前，而且每天发布多个版本。确定性检查记录了 \"5399 stars in 4 days\"，说明很多人正在试用一个只有四天历史的软件。","source":"ai","title":"非常新且进展很快"},{"code":"sideloaded_app_that_updates_itself","severity":"info","evidence":"README FAQ：\"Where do updates come from? Only from this repository's Releases; you can also upgrade in the app from About › Check for updates.\" 这个应用以 APK 形式在任何应用商店之外安装，并从这个仓库自我更新，所以每次更新的可信度都取决于这个仓库账户本身。","source":"ai","title":"旁加载的应用会自我更新"}],"platforms":[],"tools":[],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["An AI model provider of your choice (DeepSeek, Qwen, Kimi, OpenAI, Anthropic, Gemini, OpenRouter, Groq, Mistral, Ollama and others)"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","automation"],"alternativeTo":[{"slug":"manus","name":"Manus","confirmed":true}],"declaredAlternatives":[],"stars":8062,"downloads":59103,"downloadsWeek":28995,"downloadsDays":7,"growth":2019,"growthDays":7,"history":[{"date":"2026-10-31","stars":2747},{"date":"2026-11-01","stars":3963},{"date":"2026-11-02","stars":5302},{"date":"2026-11-03","stars":6043},{"date":"2026-11-04","stars":6594},{"date":"2026-11-05","stars":7060},{"date":"2026-11-06","stars":7382},{"date":"2026-11-07","stars":7589},{"date":"2026-11-08","stars":7732},{"date":"2026-11-09","stars":7900},{"date":"2026-11-10","stars":8062}],"firstPublishedAt":"2026-10-31T01:54:33.233Z","commitSha":"4ce3ca67e8cf8af8323a4d74505f51fd94178137","verifiedAt":"2026-11-03T02:55:58.033Z","lastCommitAt":"2026-11-10T16:52:30Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An Android app where an AI assistant taps through your phone apps and websites for you, using a model key you provide.","summary":"Mika is a phone assistant for Android: you ask in plain words and it opens apps, taps, types and reads pages on a hidden background screen while you keep using the phone. It also has a built-in browser, timed tasks and a Linux sandbox for files on arm64 phones. There is no Mika server: you connect your own AI provider key (DeepSeek, OpenAI, Anthropic, Gemini, OpenRouter, a local Ollama and others) or sign in with a ChatGPT, xAI or OpenRouter account. Chat and browsing need Android 11+; operating other apps needs Android 12+ and a one-time wireless debugging pairing.","easeRationale":"It is an Android-only app, so none of the desktop platforms get a grade and the deterministic check found no desktop binaries. The README's \"three steps\" are: install the APK from the releases page, then enter an API key or sign in with a model account, then — for the headline feature of operating other apps — run the \"Set up phone control\" wizard and pair wireless debugging once in the phone's developer options, which the FAQ says some phones can only do from a computer over USB.","costNote":"The app is free and Apache-2.0, but it only works with an AI model you pay for: a provider API key, or a ChatGPT Plus/Pro, xAI or OpenRouter account. The README also allows a self-hosted model (Ollama, LM Studio), which has no per-use cost.","notVerified":["Whether the APK installs and really operates apps on your particular phone and Android version (the project ships its own 12-point compatibility self-check)","Whether the wireless debugging pairing survives Wi-Fi changes and restarts without a computer","What the AI provider you choose does with the screen contents and chat text it receives","How much the model calls cost in a month of real use","Disk and memory use on the phone, which the check could not estimate","Whether the Linux sandbox feature works on your device, since it needs an arm64 phone"]},"zh-Hans":{"summaryShort":"一个 Android 应用，AI 助手帮你在手机应用和网站上点按操作，使用你自己提供的模型密钥。","summary":"Mika 是一个 Android 手机助手：你用日常语言提出要求，它就会在一个隐藏的后台屏幕上打开应用、点按、输入和阅读页面，而你可以继续正常使用手机。它还内置浏览器、定时任务，以及在 arm64 手机上用于文件操作的 Linux 沙盒。Mika 没有自己的服务器：你需要接入自己的 AI 提供方密钥（DeepSeek、OpenAI、Anthropic、Gemini、OpenRouter、本地 Ollama 等），或者用 ChatGPT、xAI 或 OpenRouter 账户登录。聊天和浏览需要 Android 11+；操作其他应用需要 Android 12+，并且要做一次性的无线调试配对。","easeRationale":"这是一款仅支持 Android 的应用，所以没有任何桌面平台获得评级，确定性检查也没有发现桌面端的可执行文件。README 的「三步」是：从发行页安装 APK，然后输入 API 密钥或用模型账户登录，最后——为了实现操作其他应用这项核心功能——运行「Set up phone control」向导，在手机的开发者选项里完成一次无线调试配对，FAQ 提到有些手机只能通过电脑用 USB 才能完成这一步。","costNote":"该应用免费，采用 Apache-2.0 许可，但它只能配合你自己付费的 AI 模型使用：提供方的 API 密钥，或者 ChatGPT Plus/Pro、xAI 或 OpenRouter 账户。README 也允许使用自托管模型（Ollama、LM Studio），这样就没有按次付费的成本。","notVerified":["APK 能否在你具体的手机和 Android 版本上安装并真正操作应用（项目自带一个 12 项的兼容性自检）","无线调试配对在没有电脑的情况下，经过 Wi-Fi 变化和重启后是否还能保持","你选择的 AI 提供方会如何处理它收到的屏幕内容和聊天文本","一个月的真实使用中，模型调用会花多少钱","手机上的磁盘和内存占用，检查环节无法估算这一点","Linux 沙盒功能是否能在你的设备上使用，因为它需要 arm64 手机"]}}},{"id":"R_kgDOVSqAyw","slug":"wesbos__3d-model-anything","name":"3D Model Anything","owner":"wesbos","fullName":"wesbos/3d-model-anything","type":"dev_tool","status":"active","grade":"A","easeByPlatform":{"windows":"D","mac":"D","linux":"D","web":"A"},"risk":"low","signals":[{"code":"assets_not_under_the_code_license","severity":"info","evidence":"LICENSE：\"Files under data/ and docs/images/thumb-1.png are excluded; see ASSETS-LICENSE.md.\" README.md 补充说：\"The supplied office images/textures have separate asset terms. Third-party marks, artwork and depicted products are not licensed by the code license.\" 所以 MIT 许可覆盖的是代码和技能，不包括演示用的素材。","source":"ai","title":"素材不在代码许可范围内"},{"code":"iphone_app_not_distributed","severity":"info","evidence":"README.md：\"The app is source-available here; an App Store release is a separate distribution step. Do not assume an App Store download is available yet.\" iPhone 应用需要你自己在 Xcode 里编译并侧载安装。","source":"ai","title":"iphone 应用尚未正式分发"},{"code":"captures_photograph_your_home","severity":"info","evidence":"README.md：\"Keep your captures private; inspect assets before publishing your own demo.\" 一次捕获包含 LiDAR 几何数据和你自己房间的照片，这些数据保存在你机器上的 data/captures-in/ 目录里，会被你指定的 AI 代理读取。","source":"ai","title":"捕获过程会拍下你的家"}],"platforms":["windows","mac","linux","web"],"tools":["node","python","other"],"buildTools":["node"],"disk":{"minMB":100,"maxMB":500,"basis":"heuristic"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"file"},"categories":["design_cad","ai_skills","hardware"],"alternativeTo":[{"slug":"polycam","name":"Polycam","confirmed":true},{"slug":"matterport","name":"Matterport","confirmed":true}],"declaredAlternatives":[],"stars":2932,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":243,"growthDays":7,"history":[{"date":"2026-10-27","stars":2217},{"date":"2026-10-28","stars":2313},{"date":"2026-10-29","stars":2365},{"date":"2026-10-30","stars":2480},{"date":"2026-10-31","stars":2534},{"date":"2026-11-01","stars":2613},{"date":"2026-11-02","stars":2654},{"date":"2026-11-03","stars":2689},{"date":"2026-11-04","stars":2737},{"date":"2026-11-05","stars":2794},{"date":"2026-11-06","stars":2818},{"date":"2026-11-07","stars":2839},{"date":"2026-11-08","stars":2884},{"date":"2026-11-09","stars":2914},{"date":"2026-11-10","stars":2932}],"firstPublishedAt":"2026-10-27T18:40:28.090Z","commitSha":"31b2dc9da3fb541e501f2f0936e55b32afb70ee8","verifiedAt":"2026-10-31T01:39:55.244Z","lastCommitAt":"2026-10-22T13:10:20Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Scan a room with an iPhone, then have an AI coding agent rebuild it as an editable 3D scene you walk through in a browser.","summary":"Three parts that work together: an iPhone app that records a room with LiDAR and photos, a Three.js viewer that shows the result in a browser (dollhouse, walking, overhead and wireframe views), and an agent skill called scan-to-threejs that guides an AI coding assistant through measuring, modeling and comparing its renders with your photos. It is for people who write code. The README calls it an iterative workflow, not a one-click scan-to-model converter, and the current workflow covers rooms and interiors only.","easeRationale":"The hosted page opens in any browser with nothing to install, but it only shows the author's sample office and the README says Scan and Overlay are disabled there. To rebuild your own room you need Node.js 22.12+ with npm commands, Python 3 with Pillow (plus NumPy and Open3D for depth), an AI coding assistant, and for the capture app a Mac with Xcode and an iPhone or iPad with LiDAR.","costNote":"The code and the skill are MIT. The reconstruction is done by an AI coding assistant, which usually needs a paid plan or API use; the office images and textures have separate terms in ASSETS-LICENSE.md.","notVerified":["How close the rebuilt scene gets to a real room, and how many agent rounds it takes.","Whether the iPhone app builds and runs on a current Xcode and iOS version.","How much AI usage a full reconstruction consumes.","Real disk use for scans, photos and textures of your own rooms.","What the agent-browser CLI used by the capture scripts does."]},"zh-Hans":{"summaryShort":"用 iPhone 扫描一个房间，让 AI 编程代理把它重建成可在浏览器里行走浏览的可编辑 3D 场景。","summary":"三个部分协同工作：一个用 LiDAR 和照片记录房间的 iPhone 应用，一个在浏览器里展示结果的 Three.js 查看器（娃娃屋视图、步行视图、俯视视图和线框视图），以及一个叫 scan-to-threejs 的代理技能，它会引导 AI 编程助手完成测量、建模，并把渲染结果和你的照片做对比。这是给写代码的人用的。README 把它称为一个迭代式的工作流程，不是一键把扫描转成模型的工具，当前的流程只覆盖房间和室内空间。","easeRationale":"托管页面在任何浏览器里都能直接打开，不用安装任何东西，但那里只展示作者自己的示例办公室，README 说那里的 Scan 和 Overlay 功能是关闭的。要重建你自己的房间，你需要 Node.js 22.12+ 以及 npm 命令、带 Pillow 的 Python 3（深度处理还需要 NumPy 和 Open3D）、一个 AI 编程助手，以及用于捕获的一台装有 Xcode 的 Mac 和一台带 LiDAR 的 iPhone 或 iPad。","costNote":"代码和技能都是 MIT 许可。重建工作由 AI 编程助手完成，通常需要付费套餐或 API 用量；办公室的图片和材质在 ASSETS-LICENSE.md 中有单独的条款。","notVerified":["重建出的场景和真实房间有多接近，以及要跑多少轮代理才能完成。","iPhone 应用能否在当前版本的 Xcode 和 iOS 上编译运行。","一次完整重建要消耗多少 AI 用量。","你自己房间的扫描、照片和材质实际占用多少磁盘空间。","捕获脚本用到的 agent-browser 命令行工具具体做了什么。"]}}},{"id":"R_kgDOVAzMMQ","slug":"wheresryan22__anatomy","name":"Anatomy","owner":"wheresryan22","fullName":"wheresryan22/anatomy","type":"dev_tool","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"23 天内获得 23369 个星标","source":"check","title":"仓库很新但星标很多"},{"code":"asks_you_to_loosen_your_agents_sandbox","severity":"attention","evidence":"README 的 \"Codex: if the browser checks are blocked\" 部分写道，该技能 \"checks its own figures by starting headless Chrome, which writes a temporary profile and talks to it over a local port\"，当 Codex 的沙盒阻止这一行为时，它建议在沙盒外批准命令，或者使用 `sandbox_mode = \"workspace-write\"`、`network_access = true` 和 `writable_roots = [\"/tmp\"]` 的配置，或者用 `--sandbox danger-full-access` 启动 Codex，它自己也承认这会 \"lifts every limit for that session\"（解除该会话的一切限制）。它也提到，没有这些检查图形依然能构建出来。","source":"ai","title":"要求你放宽代理的沙盒限制"},{"code":"the_install_copies_the_whole_repository_and_wants_newer_node","severity":"info","evidence":"README 的 \"With the skills CLI\" 部分写道：\"The CLI copies the whole repository, examples included (about 10 MB). It needs a recent Node.js: version 1.7.1 of the CLI asks for 22.20 or later\"——高于需求表和 README 徽章里标注的 Node 20.10 及以上。安装命令还会直接从 npm 运行另一个独立项目，vercel-labs 的 `skills` 命令行工具。","source":"ai","title":"安装会复制整个仓库，并要求更新的 node 版本"},{"code":"each_figure_is_a_long_paid_session","severity":"info","evidence":"README 的 Limits 部分写道：\"It is slow on purpose. A figure takes a long session: the parts list, the geometry, the audit, and close-ups of every joint in both themes. The Raptor's build takes a few minutes, and a full 3D orbit audit takes 15 to 20 minutes for 30 parts.\" 这个技能还会 \"tells Claude to read every PNG it makes\"（让 Claude 读取它生成的每一张 PNG），所以助手要处理大量图片，而这正是你要付费的部分。","source":"ai","title":"每个图形都是一次漫长的付费会话"},{"code":"sharing_a_figure_opens_a_pull_request_in_your_name","severity":"info","evidence":"README 的 \"Share a figure\" 部分：让代理发布内容时会 \"asks for a title and your GitHub name, shows what it will send, and after your yes opens a pull request on anatomy-community for you\"。这只会在你主动要求时发生，而且会先展示内容，但结果是一条带有你 GitHub 名字的公开发布内容。","source":"ai","title":"分享图形会以你的名义开一个 pull request"},{"code":"3d_mode_is_beta_and_heavy","severity":"info","evidence":"README 中写道：\"The 3D mode is beta. It is proved on its example and one test figure, not yet on a production figure. A small 3D figure runs at 60 fps; the 380-part stress hand measured about 20 fps at 1440 px wide and 13 to 15 fps on a throttled phone profile\"。kit 文件夹里还把 turn-fixed.mjs、canvas-painter.mjs 和 gl-shared.mjs 列为 \"experimental variants, not yet documented or used by the examples\"（实验性变体，尚未写入文档，也未被示例使用）。","source":"ai","title":"3d 模式仍是测试版且较为吃资源"}],"platforms":["windows","mac","linux"],"tools":["node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_skills","image","design_cad"],"alternativeTo":[{"slug":"illustrator","name":"Illustrator","confirmed":true}],"declaredAlternatives":[],"stars":26482,"downloads":3184,"downloadsWeek":534,"downloadsDays":7,"growth":2033,"growthDays":7,"history":[{"date":"2026-10-25","stars":20645},{"date":"2026-10-26","stars":21075},{"date":"2026-10-27","stars":21463},{"date":"2026-10-28","stars":22116},{"date":"2026-10-29","stars":22622},{"date":"2026-10-30","stars":23031},{"date":"2026-10-31","stars":23364},{"date":"2026-11-01","stars":23806},{"date":"2026-11-02","stars":24119},{"date":"2026-11-03","stars":24449},{"date":"2026-11-04","stars":24840},{"date":"2026-11-05","stars":25208},{"date":"2026-11-06","stars":25528},{"date":"2026-11-07","stars":25740},{"date":"2026-11-08","stars":25991},{"date":"2026-11-09","stars":26232},{"date":"2026-11-10","stars":26482}],"firstPublishedAt":"2026-10-25T23:19:32.648Z","commitSha":"1482e8469b0a9e0d8cc18ea2740a01423d29ae80","verifiedAt":"2026-11-01T01:48:37.805Z","lastCommitAt":"2026-10-18T20:20:24Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"An add-on for Claude Code and similar AI assistants that draws interactive isometric figures of machines as web pages.","summary":"Anatomy is a skill you install inside an AI coding assistant (Claude Code, and also Antigravity, OpenCode, Codex and MiniMax Code). You ask it to explain an idea and it invents a machine whose working is that idea, then builds it as an isometric drawing in SVG, optionally with WebGL effects such as flame or water, as a standalone HTML page, a single SVG or React components you keep. It needs Node.js 20.10+ and Chrome for its own checks, and no API keys of its own. The author is clear that it is not a chart or diagram tool and does not draw people or maps.","easeRationale":"The same on Windows, macOS and Linux: you must already have one of the AI coding assistants, then install Node.js and run one command, `npx skills add wheresryan22/anatomy -g`, or copy the folder in by hand with git or a ZIP; Chrome or Chromium also has to be present for the skill's own picture checks. After that it is a `/anatomy` command typed into the assistant, with no window of its own.","costNote":"The skill itself is free and MIT-licensed, with \"No API keys and nothing to npm install\". What costs money is the AI assistant that runs it: Claude Code or one of the other agents, on a paid plan or API credits, for sessions the README says are long on purpose.","notVerified":["Whether your own request produces a figure as good as the examples in the gallery","How much a single figure costs in AI assistant usage, since the sessions are long by design","Whether the headless Chrome checks work on your machine, especially on Windows or inside a sandboxed agent","Disk use: the README says the install copies about 10 MB, which the automatic check could not measure","How 23369 stars in 23 days were gathered, and how the project behaves once the first attention passes"]},"zh-Hans":{"summaryShort":"一款用于 Claude Code 等 AI 助手的插件，能把机器画成交互式等距网页图。","summary":"Anatomy 是一个安装在 AI 编程助手里的技能（Claude Code，以及 Antigravity、OpenCode、Codex 和 MiniMax Code）。你让它解释一个想法，它就会发明一台体现那个想法的机器，然后把它画成 SVG 等距图，可以选择加上火焰、水流等 WebGL 特效，输出为独立的 HTML 页面、单个 SVG 或你可以保留的 React 组件。它需要 Node.js 20.10 及以上版本和 Chrome 来做自身的检查，本身不需要任何 API 密钥。作者明确表示它不是图表或示意图工具，也不会画人物或地图。","easeRationale":"在 Windows、macOS 和 Linux 上一样：你必须已经安装了某个 AI 编程助手，然后安装 Node.js 并运行一条命令 `npx skills add wheresryan22/anatomy -g`，或者用 git 或 ZIP 手动把文件夹复制进去；还需要安装 Chrome 或 Chromium，供该技能自身的图片检查使用。之后只需在助手里输入 `/anatomy` 命令，没有自己的窗口。","costNote":"这个技能本身免费，采用 MIT 许可证，\"No API keys and nothing to npm install\"（无需 API 密钥，也无需 npm 安装任何东西）。花钱的是运行它的 AI 助手：Claude Code 或其他代理，需要付费计划或 API 额度，而 README 说这些会话是故意设计得很长的。","notVerified":["你自己提出的要求能否产生像画廊示例一样出色的图形","由于会话被设计得很长，制作一张图形在 AI 助手使用上会花多少钱","无头 Chrome 检查是否能在你的电脑上运行，尤其是在 Windows 上或在沙盒化的代理内部","磁盘占用：README 说安装会复制大约 10 MB，这一点自动检测无法测量","23 天内获得 23369 个星标是怎么来的，以及项目在最初的关注过去之后会如何表现"]}}},{"id":"R_kgDOVShyNw","slug":"xevrion__hatchzip","name":"HatchZip","owner":"xevrion","fullName":"xevrion/hatchzip","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"very_young_project_fast_releases","severity":"attention","evidence":"该仓库创建于 2026-10-21，已经有 9 个发行版，最新的是 2026-10-28 发布的 v0.1.8，Cargo.toml 写着 version = \"0.1.8\"。一个在读取压缩包的同时就删除它的工具，目前还处于早期的 0.1.x 阶段。","source":"ai","title":"项目非常新，发行很快"},{"code":"deletes_the_archive_while_extracting","severity":"attention","evidence":"README：\"HatchZip destroys the archive as it extracts\" 和 \"When the extraction is complete, the archive has been removed and the files are in the output folder.\" 如果中途出了问题，压缩包会被部分销毁：\"if the download itself is damaged, the files before the damaged spot are extracted and the archive is partly gone.\" 要保留压缩包需要用 --keep-archive 选项，或者在应用里开启对应设置。","source":"ai","title":"解压的同时会删除压缩包"},{"code":"progress_files_must_not_be_deleted","severity":"attention","evidence":"README：进度保存在压缩包旁边的 \"game.zip.hatchzip\" 和 \"game.zip.hatchzip-tail\" 文件里，\"Do not delete them while an extraction is unfinished: they hold the archive's index, which is no longer inside the archive itself.\"","source":"ai","title":"进度文件不能被删除"},{"code":"not_code_signed_antivirus_flags","severity":"attention","evidence":"README：\"The Windows installers and the macOS apps are not code-signed yet\"，以及 \"On VirusTotal, the v0.1.3 Windows setup is flagged by 4 of 70 engines, all heuristic or machine-learning scores; Microsoft Defender reports it clean.\" 项目提供了 SHA-256 文件和 \"gh attestation verify\" 来检查下载内容。","source":"ai","title":"没有代码签名，杀毒软件可能会报警"},{"code":"written_with_ai_assistants","severity":"info","evidence":"README 的 \"How it is built\" 部分：\"HatchZip is written with the help of AI coding assistants, and AGENTS.md is the rulebook they work under.\" 它还宣称在 Linux、macOS 和 Windows 上有超过 130 个测试，我们并未自行验证。","source":"ai","title":"借助 ai 助手编写"},{"code":"copycat_download_sites","severity":"info","evidence":"README：\"HatchZip is only distributed from this repository's releases and its site, hatchzip.dev. Other sites using the name, such as hatchzip.com, are not affiliated with this project.\"","source":"ai","title":"存在冒充的下载站点"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["rust"],"disk":{"minMB":9,"maxMB":28,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["network_system","productivity"],"alternativeTo":[{"slug":"winrar","name":"WinRAR","confirmed":true},{"slug":"winzip","name":"WinZip","confirmed":true}],"declaredAlternatives":[],"stars":2806,"downloads":28093,"downloadsWeek":6306,"downloadsDays":7,"growth":321,"growthDays":7,"history":[{"date":"2026-10-27","stars":2176},{"date":"2026-10-28","stars":2267},{"date":"2026-10-29","stars":2315},{"date":"2026-10-30","stars":2355},{"date":"2026-10-31","stars":2385},{"date":"2026-11-01","stars":2434},{"date":"2026-11-02","stars":2462},{"date":"2026-11-03","stars":2485},{"date":"2026-11-04","stars":2497},{"date":"2026-11-05","stars":2538},{"date":"2026-11-06","stars":2585},{"date":"2026-11-07","stars":2617},{"date":"2026-11-08","stars":2681},{"date":"2026-11-09","stars":2745},{"date":"2026-11-10","stars":2806}],"firstPublishedAt":"2026-10-27T18:40:28.100Z","commitSha":"50fa69c1e8b73f53d7b0cc592c303c66fd7ad8c5","verifiedAt":"2026-10-31T02:01:16.240Z","lastCommitAt":"2026-11-10T07:21:06Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"HatchZip-x64-setup.exe","sizeBytes":10886083,"url":"https://github.com/xevrion/hatchzip/releases/download/v0.1.10/HatchZip-x64-setup.exe","releaseUrl":"https://github.com/xevrion/hatchzip/releases/tag/v0.1.10"},{"os":"mac","file":"HatchZip-aarch64.dmg","sizeBytes":12940610,"url":"https://github.com/xevrion/hatchzip/releases/download/v0.1.10/HatchZip-aarch64.dmg","releaseUrl":"https://github.com/xevrion/hatchzip/releases/tag/v0.1.10"},{"os":"linux","file":"HatchZip-x86_64.AppImage","sizeBytes":12499448,"url":"https://github.com/xevrion/hatchzip/releases/download/v0.1.10/HatchZip-x86_64.AppImage","releaseUrl":"https://github.com/xevrion/hatchzip/releases/tag/v0.1.10"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"Opens ZIP, 7z, RAR and tar archives without needing room for the archive and the extracted files at the same time.","summary":"HatchZip is a free archive extractor with a desktop app and a command line tool. It opens ZIP, 7z, RAR, tar and split archives and extracts them in place: the archive file shrinks as the files come out, so you only need free space for the extracted files, not for both. It helps when a large download will not fit on an almost full disk. It checks that the plan fits before changing anything and can carry on after a crash. When it finishes, the archive is gone, unless you pick the option that keeps it. There are installers for Windows, macOS and Linux.","easeRationale":"All three platforms are A because release v0.1.8 has ready installers: HatchZip-x64-setup.exe or HatchZip-x64.msi for Windows, HatchZip-aarch64.dmg and HatchZip-x86_64.dmg for macOS, and hatchzip_amd64.deb, hatchzip-x86_64.rpm or HatchZip-x86_64.AppImage for Linux, with nothing else to install. The files are not code-signed yet, so on Windows you have to click \"More info\" then \"Run anyway\", and on macOS 15 and later you open the app once and then allow it in System Settings, Privacy & Security.","costNote":"Free and open source, under the MIT or the Apache-2.0 licence at your choice. RAR reading in the release builds uses the UnRAR source under its own freeware licence.","notVerified":["Whether extracting in place really frees space on your own disk: the README says hole punching on NTFS and APFS is \"not yet verified\".","Whether a crash, a power cut or Ctrl+C always resumes with every file intact, as described.","The speed and space figures in the README, which were measured only on the author's Fedora machine.","Whether your antivirus will block the unsigned installer.","What happens to a damaged archive you cannot download again if you do not use --verify-first.","The real disk space used: the checks estimate 9 to 28 MB for the installers."]},"zh-Hans":{"summaryShort":"打开 ZIP、7z、RAR 和 tar 压缩包，不需要同时为压缩包本身和解压出的文件预留空间。","summary":"HatchZip 是一款免费的压缩包解压工具，提供桌面应用和命令行工具。它可以打开 ZIP、7z、RAR、tar 和分卷压缩包，并就地解压：压缩包会随着文件被解压出来而缩小，所以你只需要给解压后的文件留出空间，不用同时给两者留空间。当一个大文件下载下来，磁盘几乎满了放不下时，这就很有用。它会先检查计划是否可行，再开始改动，崩溃后也能接着继续。解压完成后，压缩包就会消失，除非你选择保留压缩包的选项。Windows、macOS 和 Linux 都有安装包。","easeRationale":"三个平台都是 A 级，因为 v0.1.8 发行版带有现成的安装包：Windows 用 HatchZip-x64-setup.exe 或 HatchZip-x64.msi，macOS 用 HatchZip-aarch64.dmg 和 HatchZip-x86_64.dmg，Linux 用 hatchzip_amd64.deb、hatchzip-x86_64.rpm 或 HatchZip-x86_64.AppImage，不需要再装别的东西。这些文件还没有代码签名，所以在 Windows 上你需要点击“更多信息”再点击“仍要运行”，在 macOS 15 及更新版本上，你需要先打开一次应用，再到系统设置的隐私相关面板里允许它运行。","costNote":"免费且开源，你可以选择 MIT 或 Apache-2.0 许可。发行版里读取 RAR 功能使用的是 UnRAR 源码，遵循其自身的免费许可条款。","notVerified":["就地解压是否真的能在你自己的磁盘上腾出空间：README 说 NTFS 和 APFS 上的打孔（hole punching）功能“尚未验证”。","崩溃、断电或 Ctrl+C 是否总能像描述的那样，恢复时保留所有完整文件。","README 里的速度和空间数据，只是在作者的 Fedora 机器上测得的。","你的杀毒软件是否会拦截这个未签名的安装包。","如果不使用 --verify-first，一个你无法重新下载的损坏压缩包会发生什么。","实际占用的磁盘空间：检测估算安装包为 9 到 28 MB。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzMTEzMTU3MzE=","slug":"xtls__xray-core","name":"Xray","owner":"XTLS","fullName":"XTLS/Xray-core","type":"dev_tool","status":"active","grade":"D","easeByPlatform":{"windows":"D","mac":"D","linux":"D"},"risk":"attention","signals":[{"code":"proxy_tool_restricted_in_some_countries","severity":"attention","evidence":"该仓库的主题标签本身就是 \"anticensorship, dns, network, proxy, reality, shadowsocks, socks5, tls, trojan, tunnel, utls, vision, vless, vmess, vpn, wireguard, xhttp, xray, xtls, xudp\"，项目描述是 \"Xray, Penetrates Everything\"。README 的教程列表中包括 \"XTLS-Iran-Reality\" 和 \"Xray REALITY with 'steal oneself'\"，项目还运营着俄语和波斯语的 Telegram 频道。这些协议的设计目的就是让流量躲过网络过滤，而在一些国家，运行或传播这类工具是受限制甚至被禁止的，具体的法律状况取决于你所在的地区。","source":"ai","title":"这类代理工具在一些国家受到限制"},{"code":"needs_a_server_and_a_config_file_you_write","severity":"info","evidence":"README 没有自己的配置章节：\"Usage\" 部分只链接到诸如 https://github.com/XTLS/Xray-examples 这样的示例仓库和第三方教程。发布版压缩包里只包含 xray 这一个程序，没有别的，所以在你提供一份 JSON 配置和一台自己租用或搭建的服务器地址之前，它什么都不会做。","source":"ai","title":"需要一台服务器和一份你自己写的配置文件"},{"code":"traffic_passes_through_whoever_owns_the_server","severity":"attention","evidence":"这个程序的全部工作就是把你的连接传送到一台远程服务器。运营那台服务器的人能看到流量，而 README 只谈协议和面板，完全没有提到谁在运营人们实际使用的服务器。它的赞助商横幅中还包括一个商业 VPN 链接（BlancVPN，blanc.link）。","source":"ai","title":"流量会经过服务器所有者之手"},{"code":"windows_zip_bundles_a_closed_source_driver","severity":"info","evidence":"README 的 \"Bundled Third-Party Components Redistribution\" 部分写道：\"This distribution contains unmodified official precompiled and pre-signed Wintun binaries\"（wintun.dll），采用 \"Prebuilt Binaries License (PBL)\" 许可证，包含在 Windows 32 位、64 位和 arm64 的软件包中。这部分不是开源的，会在运行时为 TUN 功能加载。","source":"ai","title":"windows 压缩包内置了一个闭源驱动"},{"code":"crypto_donations_and_nfts_in_readme","severity":"info","evidence":"README 的 \"Donation & NFTs\" 部分列出了 TRX/USDT/USDC、TON、BTC、XMR、SOL 和 ETH 的钱包地址，并在 OpenSea 上出售 \"Project X NFT\" 系列。程序本身不需要付费，但项目的资金是以加密货币的形式募集的，这样就无法追踪到谁收到了这些钱。","source":"ai","title":"readme 中出现加密货币捐赠和 nft"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["go"],"disk":{"minMB":16,"maxMB":48,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"MPL-2.0","category":"copyleft","detectedFrom":"github"},"categories":["network_system","development"],"alternativeTo":[{"slug":"surge","name":"Surge","confirmed":true},{"slug":"proxifier","name":"Proxifier","confirmed":true}],"declaredAlternatives":[],"stars":42791,"downloads":48765451,"downloadsWeek":744350,"downloadsDays":7,"growth":190,"growthDays":7,"history":[{"date":"2026-10-26","stars":42347},{"date":"2026-10-27","stars":42386},{"date":"2026-10-28","stars":42411},{"date":"2026-10-29","stars":42446},{"date":"2026-10-30","stars":42492},{"date":"2026-10-31","stars":42525},{"date":"2026-11-01","stars":42546},{"date":"2026-11-02","stars":42574},{"date":"2026-11-03","stars":42601},{"date":"2026-11-04","stars":42623},{"date":"2026-11-05","stars":42646},{"date":"2026-11-06","stars":42659},{"date":"2026-11-07","stars":42691},{"date":"2026-11-08","stars":42730},{"date":"2026-11-09","stars":42758},{"date":"2026-11-10","stars":42791}],"firstPublishedAt":"2026-10-27T01:10:12.705Z","commitSha":"a4a3dff5b20c2faae567b742c172c7b95e75c083","verifiedAt":"2026-11-01T01:39:36.568Z","lastCommitAt":"2026-11-10T22:13:21Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"The engine inside many proxy and VPN apps: a command-line program that routes your traffic through a server you run.","summary":"Xray-core is the connection engine that dozens of proxy apps use. It speaks protocols such as VLESS, VMess, Trojan, Shadowsocks, REALITY and WireGuard, and it is meant to carry traffic past network filtering. The downloads are a single command-line file for Windows, macOS and Linux, with no window and no server included: you write a JSON configuration by hand and point it at a server you rent or run yourself. It is aimed at technical users and at the people who build the client apps; everyone else usually installs one of the graphical clients listed in the README instead.","easeRationale":"Every release has a ready 14 to 16 MB zip for Windows, macOS and Linux, but the file inside is a terminal program with no window, and the README gives no end-user setup: for configuration it links to other repositories (XTLS/Xray-examples) and to the official website, and for installation it points to third-party scripts, Docker images and web panels. Linux is the best served of the three because the project ships an official install script and Docker image, and macOS has `brew install xray`, yet on all three you still have to write a JSON configuration and have your own server before anything connects.","costNote":"The program costs nothing and the code is MPL-2.0. The server it connects to is a separate expense: you rent it or host it yourself. The README also lists sponsors and crypto donation addresses.","notVerified":["Whether a connection really works on each of Windows, macOS and Linux, which depends on your server and your network","Whether the downloaded zip matches the .dgst checksum file published beside it, unless you check it yourself","What the owner of the server you connect to does with your traffic or logs","Real disk use after unzipping, estimated at 16 to 48 MB","Whether the bundled wintun.dll in the Windows packages is the untouched file from wintun.net"]},"zh-Hans":{"summaryShort":"许多代理和 VPN 应用背后的引擎：一个命令行程序，把你的流量通过你自己运营的服务器转发出去。","summary":"Xray-core 是数十个代理应用背后使用的连接引擎。它支持 VLESS、VMess、Trojan、Shadowsocks、REALITY 和 WireGuard 等协议，目的是让流量绕过网络过滤。下载内容是一个适用于 Windows、macOS 和 Linux 的单一命令行文件，没有窗口，也不包含服务器：你需要手写一份 JSON 配置文件，并指向一台自己租用或搭建的服务器。它面向技术用户和构建客户端应用的开发者；其他人通常会改用 README 中列出的某个图形化客户端。","easeRationale":"每个发布版本都有现成的 14 到 16 MB 压缩包，分别对应 Windows、macOS 和 Linux，但里面的文件是一个没有窗口的终端程序，README 也没有给出面向最终用户的设置说明：配置方面链接到其他仓库（XTLS/Xray-examples）和官方网站，安装方面则指向第三方脚本、Docker 镜像和网页面板。三者中 Linux 的支持最完善，因为项目提供了官方安装脚本和 Docker 镜像，macOS 则有 `brew install xray`，但三个系统都需要你手写一份 JSON 配置，并准备好自己的服务器才能连接成功。","costNote":"该程序免费，代码采用 MPL-2.0 许可证。你连接的服务器是另一笔开销：需要自己租用或搭建。README 还列出了赞助商和加密货币捐赠地址。","notVerified":["连接是否真的能在 Windows、macOS 和 Linux 上分别正常工作，这取决于你的服务器和网络","下载的压缩包是否与旁边发布的 .dgst 校验文件相符，除非你自己去核对","你所连接服务器的所有者会怎样处理你的流量或日志","解压后的实际磁盘占用，估算为 16 到 48 MB","Windows 软件包中附带的 wintun.dll 是否是来自 wintun.net 未经修改的原始文件"]}}},{"id":"MDEwOlJlcG9zaXRvcnk0OTk3MDY0Mg==","slug":"yarnpkg__yarn","name":"Yarn Classic","owner":"yarnpkg","fullName":"yarnpkg/yarn","type":"framework_infra","status":"active","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"attention","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2024-03-09（963 天前）","source":"check","title":"版本长期未更新"},{"code":"frozen_line_the_authors_tell_you_to_migrate","severity":"attention","evidence":"仓库描述为 'The 1.x line is frozen - features and bugfixes now happen on https://github.com/yarnpkg/berry'，README 还补充说：'If you hit bugs or issues with Yarn 1.x, we strongly suggest you migrate to the latest release'。贡献部分写道 'The 1.x codebase is fairly old and will only accept security fixes.'","source":"ai","title":"已冻结的版本线，作者建议你迁移"},{"code":"very_old_dependencies","severity":"attention","evidence":"package.json 仍然依赖早已被弃用的包，包括 `request` ^2.87.0（request 项目已于 2020 年被弃用）、`babel-runtime` ^6.26.0、`micromatch` ^2.3.11、`mkdirp` ^0.5.1 和 `rimraf` ^2.5.0，并声明 `engines.node: >=4.0.0`。","source":"ai","title":"依赖项非常陈旧"},{"code":"a_ci_token_is_published_in_the_readme","severity":"info","evidence":"README 的构建徽章在其 URL 中嵌入了一个令牌：'https://circleci.com/gh/yarnpkg/yarn.svg?style=shield&circle-token=5f0a78473b0f440afb218bf2b82323cc6b3cb43f'。徽章令牌本来只用于读取，但一个凭证就这样放在公开文件里。","source":"ai","title":"README 中公开了 CI 令牌"},{"code":"unreleased_version_in_the_source","severity":"info","evidence":"package.json 中的版本是 '1.23.0-0'，而最新发布的版本是 2024-03-09 的 v1.22.22，所以仓库里的代码比任何已发布版本都要新。最后一次提交是 2026-02-12。","source":"ai","title":"源码中有尚未发布的版本"},{"code":"a_large_backlog_on_a_closed_line","severity":"info","evidence":"项目信息：有 2055 个未关闭的 issue，而维护者说这条版本线只接受漏洞修复。","source":"ai","title":"已停止更新的版本线上积压大量问题"},{"code":"sponsor_logos_with_tracking_links","severity":"info","evidence":"README 的 'Gold sponsors' 表格通过活动链接把读者引向 Doppler 和 WorkOS，例如 'https://www.doppler.com/?utm_campaign=github_repo&utm_medium=referral&utm_content=yarn&utm_source=github'；两者都是付费的商业服务。","source":"ai","title":"赞助商标志带有跟踪链接"}],"platforms":["windows","mac","linux"],"tools":["node"],"buildTools":["node"],"disk":{"minMB":2,"maxMB":5,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"BSD-2-Clause","category":"permissive","detectedFrom":"file"},"categories":["development"],"alternativeTo":[],"declaredAlternatives":[],"stars":41459,"downloads":273001837,"downloadsWeek":562877,"downloadsDays":7,"growth":0,"growthDays":7,"history":[{"date":"2026-10-26","stars":41470},{"date":"2026-10-27","stars":41467},{"date":"2026-10-28","stars":41465},{"date":"2026-10-29","stars":41468},{"date":"2026-10-30","stars":41466},{"date":"2026-10-31","stars":41466},{"date":"2026-11-01","stars":41466},{"date":"2026-11-02","stars":41464},{"date":"2026-11-03","stars":41462},{"date":"2026-11-04","stars":41460},{"date":"2026-11-05","stars":41459},{"date":"2026-11-06","stars":41458},{"date":"2026-11-07","stars":41459},{"date":"2026-11-08","stars":41459},{"date":"2026-11-09","stars":41457},{"date":"2026-11-10","stars":41459}],"firstPublishedAt":"2026-10-27T01:10:12.706Z","commitSha":"c2dda503f3759b5be5f0e24ecd9cf5c97a540147","verifiedAt":"2026-10-29T01:43:04.794Z","lastCommitAt":"2026-02-12T04:44:38Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"The classic 1.x version of Yarn, the tool programmers use to download and lock the JavaScript libraries a project needs.","summary":"Yarn is a package manager for JavaScript projects: it downloads the libraries a project depends on, writes a lock file so everyone installs exactly the same versions, caches what it fetched so it works offline, and checks each package against a checksum. This repository is the old 1.x line. Its own README says the line is frozen, that new work happens in the yarnpkg/berry repository, and that users who hit bugs should move to the newer releases. It is for programmers; you need Node.js and you use it from a terminal.","easeRationale":"On all three you first install Node.js and then install Yarn and use it by typing commands. Windows has a 2 MB .msi and Linux has .deb and .rpm packages in the last release; macOS has no file in the releases and is covered only by the install guide on the project website.","notVerified":["Whether the old dependencies still install cleanly on current Node.js versions","Whether security fixes are still being published, given the last release was in March 2024","What the unreleased 1.23.0-0 code in the repository changes compared with 1.22.22","How the macOS install path behaves, since no macOS file ships in the releases","Real disk use after the package cache grows on a working machine"]},"zh-Hans":{"summaryShort":"Yarn 的经典 1.x 版本,程序员用它来下载并锁定项目所需的 JavaScript 库。","summary":"Yarn 是面向 JavaScript 项目的包管理工具:它下载项目依赖的库,写入一个锁定文件让所有人安装完全相同的版本,把下载过的内容缓存下来以便离线使用,并用校验和检查每个包。这个仓库是旧的 1.x 分支。其 README 说明这条分支已经冻结,新的开发工作在 yarnpkg/berry 仓库中进行,遇到问题的用户应该迁移到更新的版本。它面向程序员;你需要 Node.js,并通过终端使用它。","easeRationale":"三个平台的流程相同:先安装 Node.js,再安装 Yarn,然后通过输入命令来使用。最新版本中,Windows 有一个 2 MB 的 .msi 安装包,Linux 有 .deb 和 .rpm 软件包;macOS 在发布文件中没有对应文件,只能参考项目官网的安装指南。","notVerified":["旧的依赖项是否仍能在当前的 Node.js 版本上顺利安装","由于最新版本发布于 2024 年 3 月,是否仍在发布相关的修复更新","仓库中未发布的 1.23.0-0 代码相比 1.22.22 有哪些改动","由于发布文件中没有 macOS 版本,macOS 的安装方式实际表现如何","在实际使用的机器上,随着包缓存增长,真实的磁盘占用情况"]}}},{"id":"R_kgDOUns_9A","slug":"yetone__magpie","name":"Magpie","owner":"yetone","fullName":"yetone/magpie","type":"ready_app","status":"active","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"C"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"40 天内获得 28474 个星标","source":"check","title":"年轻仓库却有大量星标"},{"code":"paid_plans_rerouted_to_other_programs","severity":"attention","evidence":"README，Subscriptions：\"Codex / ChatGPT: your ChatGPT plan's models, in every other agent\" 以及 \"Claude: drives the real local claude binary\"。在其他程序中使用个人计划，或在你的网络上共享它，可能违反你与供应商约定的条款。","source":"ai","title":"付费计划被转发给其他程序使用"},{"code":"reads_the_sign_ins_of_other_programs","severity":"attention","evidence":"README，Quick start：\"magpie finds the subscriptions your agents are signed in to by reading their sign-ins (Claude Code's and Codex's files, the Keychain)\"。可以在 Settings → Privacy → Subscription accounts 中关闭。","source":"ai","title":"读取其他程序的登录信息"},{"code":"all_prompts_pass_through_the_app","severity":"attention","evidence":"README，How it fits together：每个助手都把请求发送到本地网关 127.0.0.1:3425，插件可以作为 \"gateway middleware\"：\"JavaScript that sees what every agent sends and gets back\"。","source":"ai","title":"所有提示词都经过这个应用"},{"code":"install_by_piping_a_script_from_a_website","severity":"attention","evidence":"README，Quick start：\"curl -fsSL https://usemagpie.ai/install.sh | sh\"，在 Linux 上还会用 sudo 安装系统包。从 GitHub 下载发行文件可以避免这种方式。","source":"ai","title":"安装方式是从网站管道传输脚本"},{"code":"plugins_download_and_run_outside_code","severity":"attention","evidence":"README，Plugins：\"magpie plugin add opencode-gemini-auth # an OpenCode plugin from npm\" 以及 \"Plugins run on Bun, downloaded the first time one needs it\"。","source":"ai","title":"插件在代码之外下载和运行"},{"code":"usage_data_sent_once_a_day_by_default","severity":"info","evidence":"README，Privacy：\"Once a day a released magpie tells us it is in use: a random id, its version and system, and which agents, providers and models it is used with\"。可以在 Settings → Privacy 中关闭，或设置 DO_NOT_TRACK=1。","source":"ai","title":"默认每天发送一次使用数据"},{"code":"outside_contributions_not_accepted","severity":"info","evidence":"README，Community：\"This repository doesn't take pull requests. Only maintainers can open them.\"","source":"ai","title":"不接受外部贡献"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","go"],"disk":{"minMB":76,"maxMB":228,"basis":"installer"},"apiKeys":["At least one AI provider: an API key or a plan you are already signed in to","OpenAI","Anthropic","Google Gemini"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["development","productivity"],"alternativeTo":[{"slug":"openrouter","name":"OpenRouter","confirmed":true},{"slug":"portkey","name":"Portkey","confirmed":true},{"slug":"langsmith","name":"LangSmith","confirmed":false}],"declaredAlternatives":[],"stars":33847,"downloads":87507,"downloadsWeek":10006,"downloadsDays":7,"growth":4800,"growthDays":7,"history":[{"date":"2026-10-26","stars":22604},{"date":"2026-10-27","stars":23615},{"date":"2026-10-28","stars":24582},{"date":"2026-10-29","stars":25396},{"date":"2026-10-30","stars":26276},{"date":"2026-10-31","stars":26956},{"date":"2026-11-01","stars":27651},{"date":"2026-11-02","stars":28410},{"date":"2026-11-03","stars":29047},{"date":"2026-11-04","stars":29709},{"date":"2026-11-05","stars":30427},{"date":"2026-11-06","stars":31224},{"date":"2026-11-07","stars":31785},{"date":"2026-11-08","stars":32266},{"date":"2026-11-09","stars":33078},{"date":"2026-11-10","stars":33847}],"firstPublishedAt":"2026-10-27T01:10:12.707Z","commitSha":"eb779ce92a9a5a4c77dd09a35fa8dfbe8c140df5","verifiedAt":"2026-11-03T02:37:25.830Z","lastCommitAt":"2026-11-10T23:24:28Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Puts all your AI model accounts in one place so any coding assistant on your computer can use any of them.","summary":"magpie is a small desktop app that sits in the menu bar and keeps all your AI providers in one list: API keys, plans you are signed in to, and models running on your own machine. From there you choose which model each coding assistant (Claude Code, Codex, Gemini CLI and dozens more) will use, and it switches accounts when one runs out of quota. It also shows token use and cost. You need your own keys or plans; macOS, Windows and Linux builds are ready to download.","easeRationale":"On macOS you download a .dmg that the README says is signed and notarised, and Windows has a ready .exe; in both cases the only extra step is adding a provider and pasting a key, so B. Linux gets C because the ready file is a plain binary and the install script may have to add WebKitGTK 4.1 with your package manager and ask for your password.","costNote":"The app is free and MIT-licensed, but it only routes accounts you pay for yourself (API keys or plans such as Claude, ChatGPT, Copilot). Local servers like Ollama and LM Studio are also listed as providers. The README shows paid partners as sponsors.","notVerified":["Whether it really works with each of the 70-plus assistants listed, since almost all are other people's programs.","Whether the privacy masking catches every key or personal detail in a prompt before it leaves your computer.","What the vendors do if they notice a plan being used through the gateway by other programs.","How much your own keys and plans will cost in real use, since magpie only shows prices at list value.","Whether 1200 releases in 40 days means a stable version today, since each build updates itself.","What the optional masking model really downloads (the README says 1.6 GB) and how much disk it needs."]},"zh-Hans":{"summaryShort":"把你所有的 AI 模型账户集中在一个地方，这样你电脑上的任何编程助手都能使用其中任意一个。","summary":"magpie 是一个驻留在菜单栏的小型桌面应用，把你所有的 AI 提供方集中在一个列表里：API 密钥、已登录的计划，以及在你自己机器上运行的模型。你可以在那里为每个编程助手（Claude Code、Codex、Gemini CLI 以及其他几十种）选择要使用的模型，当某个账户额度用完时它会自动切换账户。它还会显示 token 用量和费用。你需要自己的密钥或计划；提供 macOS、Windows 和 Linux 的现成构建版本。","easeRationale":"在 macOS 上你下载一个 README 说已签名并经过公证的 .dmg，Windows 有现成的 .exe；两种情况下唯一的额外步骤都是添加一个提供方并粘贴密钥，所以评级是 B。Linux 评级是 C，因为现成文件只是一个普通的二进制文件，安装脚本可能还需要你用自己的包管理器安装 WebKitGTK 4.1 并输入密码。","costNote":"该应用免费，采用 MIT 许可，但它只负责转发你自己付费的账户（API 密钥或 Claude、ChatGPT、Copilot 等计划）。本地服务如 Ollama 和 LM Studio 也作为提供方列出。README 把付费合作方列为赞助商。","notVerified":["它是否真的能配合列出的 70 多个助手中的每一个工作，因为几乎所有助手都是别人的程序。","隐私掩码是否能在提示词离开你电脑之前拦住每一个密钥或个人信息。","如果供应商发现一个计划被网关转发给其他程序使用，他们会怎么做。","你自己的密钥和计划在真实使用中要花多少钱，因为 magpie 只按标价展示价格。","40 天内 1200 个发行版是否意味着今天的版本是稳定的，因为每个构建都会自我更新。","可选的掩码模型实际会下载什么内容（README 说是 1.6 GB），需要多少磁盘空间。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzMDcyNjAyMDU=","slug":"yt-dlp__yt-dlp","name":"yt-dlp","owner":"yt-dlp","fullName":"yt-dlp/yt-dlp","type":"needs_setup","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"downloading_may_be_restricted_where_you_live","severity":"attention","evidence":"这个程序的工作就是从没有提供下载按钮的网站上抓取媒体内容，README 中描述了如何绕过网站的防护措施：\"pycryptodomex - For decrypting AES-128 HLS streams and various other data\"，\"--impersonate CLIENT[:OS] Client to impersonate for requests\" 用于 \"sites that employ TLS fingerprinting\"，\"--xff VALUE How to fake X-Forwarded-For HTTP header to try bypassing geographic restriction\"，还需要一个 JavaScript 运行环境 \"Required for full YouTube support\"。保存一份拷贝可能违反网站的使用条款，在一些国家，绕过技术保护措施本身就是违法的。","source":"ai","title":"下载行为在你所在地区可能受到限制"},{"code":"unverified_updates_from_other_repositories","severity":"info","evidence":"README.md 的 UPDATE 部分写道：\"You may also use `--update-to <repository>` ... Be careful with what repository you are updating to though, there is no verification done for binaries from different repositories.\" 普通的 `-U` 更新仍停留在官方渠道。","source":"ai","title":"来自其他仓库的更新未经验证"},{"code":"can_read_browser_cookies","severity":"info","evidence":"README.md 的 DEPENDENCIES 部分：\"secretstorage - For `--cookies-from-browser` to access the Gnome keyring while decrypting cookies of Chromium-based browsers on Linux\"。使用该选项时，程序可以读取你浏览器中保存的登录信息。","source":"ai","title":"可以读取浏览器 cookie"},{"code":"binary_license_differs_from_repository","severity":"info","evidence":"README.md 的 Licensing 部分：\"the PyInstaller-bundled executables include GPLv3+ licensed code, and as such the combined work is licensed under GPLv3+\"，而仓库本身采用的是 Unlicense。","source":"ai","title":"二进制文件的许可证与仓库不同"},{"code":"can_fetch_code_at_run_time","severity":"info","evidence":"README 中的选项 `--remote-components COMPONENT`：\"Remote components to allow yt-dlp to fetch when required ... ejs:npm (external JavaScript components from npm), ejs:github\"。README 还补充说 \"By default, no remote components are allowed\"。","source":"ai","title":"运行时可以获取外部代码"},{"code":"possible_illegal_use","severity":"high","evidence":"主要用途是保存那些没有提供下载功能的网站上的媒体内容，README 记录了如何绕过它们的防护措施：\"--xff VALUE How to fake X-Forwarded-For HTTP header to try bypassing geographic restriction\"；\"--impersonate CLIENT[:OS] Client to impersonate for requests\"；\"pycryptodomex - For decrypting AES-128 HLS streams and various other data\"。许多网站的使用条款不允许保留副本，一些国家的法律也限制绕过技术保护措施。","source":"ai","title":"可能存在非法用途"}],"platforms":["windows","mac","linux"],"tools":["ffmpeg","other"],"buildTools":["python","ffmpeg"],"disk":{"minMB":22,"maxMB":67,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Unlicense","category":"permissive","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"4k-video-downloader","name":"4K Video Downloader","confirmed":true},{"slug":"internet-download-manager","name":"Internet Download Manager","confirmed":true}],"declaredAlternatives":[],"stars":206703,"downloads":240728211,"downloadsWeek":2218266,"downloadsDays":7,"growth":1692,"growthDays":7,"history":[{"date":"2026-10-25","stars":202124},{"date":"2026-10-26","stars":202593},{"date":"2026-10-27","stars":203013},{"date":"2026-10-28","stars":203404},{"date":"2026-10-29","stars":203712},{"date":"2026-10-30","stars":203991},{"date":"2026-10-31","stars":204232},{"date":"2026-11-01","stars":204521},{"date":"2026-11-02","stars":204782},{"date":"2026-11-03","stars":205011},{"date":"2026-11-04","stars":205233},{"date":"2026-11-05","stars":205518},{"date":"2026-11-06","stars":205762},{"date":"2026-11-07","stars":205999},{"date":"2026-11-08","stars":206216},{"date":"2026-11-09","stars":206474},{"date":"2026-11-10","stars":206703}],"firstPublishedAt":"2026-10-25T23:19:33.496Z","commitSha":"f264a6cbd1f630894da0f368df00f8aae50140e0","verifiedAt":"2026-10-29T01:39:08.513Z","lastCommitAt":"2026-11-04T07:16:26Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Save videos and audio from YouTube and thousands of other sites by typing commands in a terminal.","summary":"yt-dlp is a command-line program that saves videos and audio from YouTube and thousands of other websites to your computer. It can pick the quality, download whole playlists, subtitles and thumbnails, and turn videos into audio files such as MP3. There is no window with buttons: you type commands in a terminal. Windows, Mac and Linux each get a ready single-file program. For the best quality you also need ffmpeg, and full YouTube support needs a JavaScript runtime such as Deno.","easeRationale":"On Windows there is a ready single file (yt-dlp.exe), but it has no graphical window, you use it by typing commands, and the README strongly recommends also installing ffmpeg and a JavaScript runtime such as Deno. Mac gets the same kind of single file; on Linux the recommended download is a small file that needs Python already installed.","costNote":"Free. The code is public domain (Unlicense); the ready Windows, Mac and Linux programs include GPLv3+ code, so those files are under GPLv3+.","notVerified":["Whether the current stable release still works with YouTube and other sites on the day you use it; the README says stable is \"often stale and prone to external breakage\" and recommends the nightly channel","Whether downloading from a given site is allowed by that site's terms or by the copyright law where you live","Which sites need ffmpeg or a JavaScript runtime in practice, and how much disk space they add","Whether Windows or antivirus software warns about the single-file yt-dlp.exe"]},"zh-Hans":{"summaryShort":"通过在终端输入命令,从 YouTube 和数千个其他网站保存视频和音频。","summary":"yt-dlp 是一个命令行程序,可以把 YouTube 和数千个其他网站上的视频和音频保存到你的电脑。它可以选择画质,下载整个播放列表、字幕和缩略图,还能把视频转换成 MP3 等音频文件。它没有带按钮的窗口:你需要在终端里输入命令。Windows、Mac 和 Linux 各有一个现成的单文件程序。为了获得最佳画质,你还需要 ffmpeg,完整的 YouTube 支持则需要像 Deno 这样的 JavaScript 运行环境。","easeRationale":"在 Windows 上有一个现成的单文件程序(yt-dlp.exe),但它没有图形窗口,需要输入命令来使用,README 强烈建议同时安装 ffmpeg 和像 Deno 这样的 JavaScript 运行环境。Mac 提供同类型的单文件程序;在 Linux 上,推荐下载的是一个需要预先安装好 Python 的小文件。","costNote":"免费。代码属于公共领域(Unlicense);Windows、Mac 和 Linux 的现成程序包含 GPLv3+ 代码,因此这些文件遵循 GPLv3+ 协议。","notVerified":["当前的稳定版在你使用的那天是否仍能在 YouTube 和其他网站上正常工作;README 说稳定版'常常过时,容易受外部变化影响',并建议使用 nightly 频道","从某个网站下载内容是否符合该网站的条款,或者是否符合你所在地的版权法律","实际使用中哪些网站需要 ffmpeg 或 JavaScript 运行环境,以及它们会增加多少磁盘占用","Windows 或杀毒软件是否会对单文件 yt-dlp.exe 发出警告"]}}},{"id":"MDEwOlJlcG9zaXRvcnkxMDM5NTIw","slug":"ytdl-org__youtube-dl","name":"youtube-dl","owner":"ytdl-org","fullName":"ytdl-org/youtube-dl","type":"dev_tool","status":"alert","grade":"C","easeByPlatform":{"windows":"C","mac":"C","linux":"C"},"risk":"high","signals":[{"code":"stale_releases","severity":"attention","evidence":"最近一次发布在 2021-12-16（1778 天前）","source":"check","title":"发布版本已过时"},{"code":"bypasses_site_restrictions","severity":"attention","evidence":"README 的 OPTIONS 部分：\"--geo-bypass  Bypass geographic restriction via faking X-Forwarded-For HTTP header\" 和 \"--geo-bypass-country CODE\"。缓存部分写道 \"only YouTube player files (for videos with obfuscated signatures) are cached\"，说明该程序会绕过 YouTube 经过混淆的视频地址；Adobe Pass 相关选项（\"--ap-mso MSO\"、\"--ap-username\"、\"--ap-password\"）则用付费电视提供商的账号登录来获取其视频。这样保存视频违反了大多数视频网站的条款，在一些国家，绕过这类技术限制本身就是违法的。","source":"ai","title":"绕过网站的限制"},{"code":"no_release_since_2021","severity":"attention","evidence":"GitHub 上最新的发布版本是 2021.12.17（发布于 2021-12-16），最近一次提交是在 2025-11-26，距本次检测 337 天。视频网站经常变化，所以发布的 youtube-dl.exe 可能已经无法在这些网站上正常工作。","source":"ai","title":"自 2021 年以来没有新发布"},{"code":"downloads_installed_from_vendor_site_not_github","severity":"info","evidence":"README 的 INSTALLATION 部分告诉 UNIX 用户运行 \"sudo curl -L https://yt-dl.org/downloads/latest/youtube-dl -o /usr/local/bin/youtube-dl\"，告诉 Windows 用户获取 https://yt-dl.org/latest/youtube-dl.exe，也就是说这个文件来自 yt-dl.org，而不是来自本仓库的发布页面。程序本身也内置了 \"-U, --update  Update this program to latest version\"。","source":"ai","title":"下载来自厂商网站而非 github"},{"code":"can_run_commands_on_downloaded_files","severity":"info","evidence":"README 的后处理选项：\"--exec CMD  Execute a command on the file after downloading and post-processing\"，设置也可以来自一个配置文件（~/.config/youtube-dl/config 或 %APPDATA%\\youtube-dl\\config.txt），youtube-dl 会自动读取该文件，除非你传入 --ignore-config。","source":"ai","title":"可以对下载的文件执行命令"},{"code":"stores_account_passwords_in_a_file","severity":"info","evidence":"README 写道：各网站的登录凭据可以以明文形式保存在你主目录下的 .netrc 文件中，格式为 \"machine youtube login myaccount@gmail.com password my_youtube_password\"。","source":"ai","title":"把账号密码保存在一个文件中"},{"code":"possible_illegal_use","severity":"high","evidence":"该程序绕过视频网站自身的限制。README 的 OPTIONS、Geo Restriction 部分：\"--geo-bypass                         Bypass geographic restriction via faking X-Forwarded-For HTTP header\" 和 \"--geo-bypass-country CODE            Force bypass geographic restriction with explicitly provided two-letter ISO 3166-2 country code\"。README 的 --cache-dir 部分：\"At the moment, only YouTube player files (for videos with obfuscated signatures) are cached\"，说明它处理的是 YouTube 经过混淆的视频地址。README 的 Adobe Pass Options 部分：\"--ap-mso MSO                         Adobe Pass multiple-system operator (TV provider) identifier\"、\"--ap-username USERNAME\" 和 \"--ap-password PASSWORD\"，用于登录付费电视账号以获取其视频。视频网站的条款通常不允许保留副本，多个国家的法律也限制绕过技术保护措施。","source":"ai","title":"可能存在非法用途"}],"platforms":["windows","mac","linux"],"tools":["ffmpeg"],"buildTools":["python","ffmpeg"],"disk":{"minMB":8,"maxMB":23,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"Unlicense","category":"permissive","detectedFrom":"github"},"categories":["video","audio"],"alternativeTo":[{"slug":"4k-video-downloader","name":"4K Video Downloader","confirmed":true},{"slug":"internet-download-manager","name":"Internet Download Manager","confirmed":true}],"declaredAlternatives":[],"stars":141527,"downloads":125147774,"downloadsWeek":4066,"downloadsDays":7,"growth":81,"growthDays":7,"history":[{"date":"2026-10-25","stars":141398},{"date":"2026-10-26","stars":141400},{"date":"2026-10-27","stars":141405},{"date":"2026-10-28","stars":141412},{"date":"2026-10-29","stars":141421},{"date":"2026-10-30","stars":141434},{"date":"2026-10-31","stars":141442},{"date":"2026-11-01","stars":141450},{"date":"2026-11-02","stars":141447},{"date":"2026-11-03","stars":141446},{"date":"2026-11-04","stars":141454},{"date":"2026-11-05","stars":141462},{"date":"2026-11-06","stars":141470},{"date":"2026-11-07","stars":141476},{"date":"2026-11-08","stars":141496},{"date":"2026-11-09","stars":141512},{"date":"2026-11-10","stars":141527}],"firstPublishedAt":"2026-10-25T23:19:34.355Z","commitSha":"956b8c585591b401a543e409accb163eeaaa1193","verifiedAt":"2026-10-30T01:46:00.376Z","lastCommitAt":"2026-11-06T02:12:14Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Download videos or their audio from YouTube and other video sites by typing commands.","summary":"youtube-dl is a command-line program that downloads videos from YouTube and many other video sites. You type a command with the video link, and it saves the file; it can also download whole playlists, pick a quality or keep only the audio. It has no windows or buttons, so it suits people comfortable with a terminal. On Windows there is a single .exe file; converting or merging formats needs ffmpeg. The last GitHub release is from December 2021, so it may not work with sites that changed since then.","easeRationale":"Windows is the shortest path and sets the grade: the release is a single 8 MB youtube-dl.exe that you place in a folder on your PATH and then drive by typing commands. Mac (brew install youtube-dl) and Linux (a sudo curl command into /usr/local/bin) take one terminal step more. On every system you also need ffmpeg to pull out audio or merge video and sound.","costNote":"Released to the public domain (Unlicense).","notVerified":["Whether it still downloads from YouTube and other major sites today.","Whether the yt-dl.org download links in the README still work and are still run by the project.","Whether the 2021 Windows .exe runs on current Windows without extra system files.","Whether downloading a given video is allowed by that site's terms or by the law where you live.","Whether the built-in --update command still finds a newer version."]},"zh-Hans":{"summaryShort":"通过输入命令，从 YouTube 和其他视频网站下载视频或其音频。","summary":"youtube-dl 是一个命令行程序，可以从 YouTube 和许多其他视频网站下载视频。你输入一条带有视频链接的命令，它就会保存文件；它还可以下载整个播放列表、选择清晰度，或者只保留音频。它没有窗口或按钮，所以更适合习惯使用终端的人。在 Windows 上只有一个单独的 .exe 文件；要转换或合并格式需要 ffmpeg。GitHub 上最后一次发布是在 2021 年 12 月，所以对于之后有变化的网站，它可能无法正常工作。","easeRationale":"Windows 是最简单的方式，也决定了评级：发行版只是一个 8 MB 的 youtube-dl.exe 文件，你把它放进 PATH 中的一个文件夹，然后通过输入命令来使用。Mac（brew install youtube-dl）和 Linux（一条 sudo curl 命令把文件放到 /usr/local/bin）都需要多一个终端步骤。在所有系统上，你还需要 ffmpeg 来提取音频或合并视频和声音。","costNote":"以公共领域方式发布（Unlicense）。","notVerified":["它目前是否仍能从 YouTube 和其他主要网站下载内容。","README 中 yt-dl.org 的下载链接是否仍然有效，以及是否仍由该项目维护。","2021 年的 Windows .exe 文件在当前的 Windows 系统上运行是否还需要额外的系统文件。","下载某个特定视频是否符合该网站的条款，或者是否符合你所在地区的法律规定。","内置的 --update 命令是否仍能找到更新的版本。"]}}},{"id":"R_kgDOVN7flQ","slug":"zagitta__gpu-unlock","name":"GPU Unlock","owner":"Zagitta","fullName":"Zagitta/gpu-unlock","type":"needs_setup","status":"alert","grade":"E","easeByPlatform":{"linux":"E"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"10 天内获得 11390 个星标","source":"check","title":"新仓库却有很多星标"},{"code":"no_license","severity":"info","evidence":"GitHub、许可证文件、LICENSES 文件夹和 winget 清单中都没有检测到许可证","source":"check","title":"没有许可证"},{"code":"vendor_firmware_lock_bypass","severity":"high","evidence":"这正是该项目的目的。README：它 \"combines the cmpunlocker's V67/PLM technique with the pre-driver timing of the UEFI approach\"，并在 NVIDIA 驱动绑定之前写入速度选择值 \"SS0 0x88888888\" 和 \"SS1 0x8\"；常见问题说 NVIDIA 出于 \"Market segmentation\" 限制了这些显卡。去掉厂商内置在显卡里的限制，违反 NVIDIA 的条款，并可能被视为规避技术保护措施，这在一些国家受到限制。","source":"ai","title":"绕过厂商的固件锁定"},{"code":"fake_installers_and_coins_circulating","severity":"attention","evidence":"FAQ：\"Do you have a website/coin/installer? A: No this is the one and only OG source for gpu-unlock. Crypto, websites, installers and what else is not associated with this project or the developer in any way shape or form.\" 关于 Windows：\"There are windows installer's floating around but I would recommend being ++very++ careful with those.\"","source":"ai","title":"流传着假冒的安装程序和代币"},{"code":"hardware_instability_risk","severity":"attention","evidence":"README：\"If the firmware on a card never stops, the module logs it and keeps the DMA buffers that firmware may still use until the next power off. Power off before trying again.\" 对于负责显示的那块显卡，\"screen is blank from that reset until nvidia-drm takes over\"。FAQ：\"It may heat up your GPU slightly\"。","source":"ai","title":"硬件运行不稳定的风险"},{"code":"license_stated_only_in_readme","severity":"info","evidence":"没有检测到许可证文件，但 README 写道 \"The kernel module and local packaging are GPL-2.0-only\"，并指向 licenses/ 文件夹；\"NVIDIA firmware remains under its own terms\"。","source":"ai","title":"许可证只写在 README 里"},{"code":"build_downloads_nvidia_files","severity":"info","evidence":"README，Local build：\"Every build downloads the NVIDIA/CachyOS inputs over HTTPS and generates firmware only in the build directory and the binary package.\" 软件包是 \"built from the working tree, uncommitted changes included\"。","source":"ai","title":"构建时会下载 NVIDIA 的文件"},{"code":"crypto_donation_addresses","severity":"info","evidence":"README 的 \"Support me\" 部分列出了 ETH、BTC 和 SOL 钱包地址，同时还有 GitHub Sponsors 和 Buy Me a Coffee。","source":"ai","title":"加密货币捐赠地址"}],"platforms":["linux"],"tools":["gpu_nvidia","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":null,"category":"none","detectedFrom":null},"categories":["hardware","network_system"],"alternativeTo":[],"declaredAlternatives":[],"stars":18550,"downloads":1102,"downloadsWeek":388,"downloadsDays":7,"growth":2911,"growthDays":7,"history":[{"date":"2026-10-26","stars":9731},{"date":"2026-10-27","stars":11367},{"date":"2026-10-28","stars":12372},{"date":"2026-10-29","stars":13087},{"date":"2026-10-30","stars":13490},{"date":"2026-10-31","stars":13854},{"date":"2026-11-01","stars":14171},{"date":"2026-11-02","stars":14657},{"date":"2026-11-03","stars":15639},{"date":"2026-11-04","stars":16561},{"date":"2026-11-05","stars":17162},{"date":"2026-11-06","stars":17568},{"date":"2026-11-07","stars":17835},{"date":"2026-11-08","stars":18130},{"date":"2026-11-09","stars":18367},{"date":"2026-11-10","stars":18550}],"firstPublishedAt":"2026-10-27T01:14:23.049Z","commitSha":"1fc6d628c49c7fba34ef3a2e7f758700d35ce890","verifiedAt":"2026-10-28T01:43:42.599Z","lastCommitAt":"2026-10-28T22:31:16Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"A Linux add-on that removes the half-speed limit NVIDIA puts on GeForce RTX cards for certain AI calculations.","summary":"A Linux kernel module that loads at startup, before the NVIDIA driver, and removes the cap NVIDIA puts on GeForce tensor cores for FP16, BF16 and TF32 maths with FP32 accumulate. The README reports about double the matrix-multiply rate (RTX 3090: 75 to 125 TFLOPS) and, from user reports, 25–55% faster LLM prompt processing and image generation, with no change for gaming, quantized models or token generation. RTX 30 cards are called stable; RTX 20 and 40 are experimental and RTX 50 is not supported. You build it with make install, add a kernel boot option and restart.","easeRationale":"Linux is the only system it supports and there it is an E: you clone the repository, run make install (it builds a kernel module and installs distribution packages with your sudo password), then edit the kernel command line in GRUB, systemd-boot or your UKI setup by hand and restart. The author says he does not run Windows and only points at someone else's port.","notVerified":["Whether it works on your card, motherboard and Linux distribution; the figures come from users' own reports in the issues","Long-term stability, heat and power draw after the change","How much of the doubled rate reaches the AI programs you actually use","Whether the unlock holds on a laptop, since the README says powering the GPU down \"silently resets the unlock\"","What the generated firmware buffer, derived from \"the pinned CMP reference\", does beyond the README's description","Warranty and legal consequences of changing NVIDIA's locked settings in your country"]},"zh-Hans":{"summaryShort":"一个 Linux 附加组件,可解除 NVIDIA 在特定 AI 计算中对 GeForce RTX 显卡施加的半速限制。","summary":"这是一个在系统启动时、早于 NVIDIA 驱动加载的 Linux 内核模块,它解除了 NVIDIA 对 GeForce 张量核心在 FP16、BF16 和带 FP32 累加的 TF32 运算上施加的限制。README 报告矩阵乘法速率大约提升一倍(RTX 3090:从 75 提升到 125 TFLOPS),根据用户反馈,LLM 提示处理和图像生成速度提升 25%–55%,而游戏、量化模型或逐词生成则没有变化。RTX 30 系列被称为稳定,RTX 20 和 40 系列为实验性,RTX 50 不受支持。安装方式是运行 make install,添加一个内核启动选项,然后重启。","easeRationale":"Linux 是它唯一支持的系统,在该系统上评级为 E:你需要克隆仓库,运行 make install(它会编译一个内核模块,并用你的 sudo 密码安装发行版软件包),然后手动在 GRUB、systemd-boot 或你的 UKI 配置中编辑内核命令行,再重启。作者说他自己不使用 Windows,只是指出了别人做的移植版本。","notVerified":["它是否能在你的显卡、主板和 Linux 发行版上正常工作;这些数据来自用户在 issue 中自行反馈","更改之后的长期稳定性、发热和功耗情况","翻倍的速率在你实际使用的 AI 程序中究竟能体现多少","解锁效果在笔记本电脑上是否能保持,因为 README 说显卡断电会“悄悄重置解锁状态”","生成的固件缓冲区(源自“固定的 CMP 参考值”)除 README 所述之外还会做什么","在你所在国家,更改 NVIDIA 锁定设置可能带来的保修和法律后果"]}}},{"id":"R_kgDOUicD1g","slug":"zai-org__zcode","name":"ZCode","owner":"zai-org","fullName":"zai-org/ZCode","type":"dev_tool","status":"active","grade":"B","easeByPlatform":{"windows":"B"},"risk":"attention","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"36 天内获得 8275 个星标","source":"check","title":"新仓库却有大量星标"},{"code":"github_release_behind_source","severity":"info","evidence":"README：'2026-10-19：更新至 ZCode v3.16.2 版本'，package.json 的版本为 '3.16.2'，但 GitHub 上唯一的发布版是 v3.14.3（2026-09-24）；更新的构建版本没有发布到 GitHub。","source":"ai","title":"GitHub 发布版落后于源码"},{"code":"usage_monitoring_sdk","severity":"info","evidence":"package.json patchedDependencies：'@arms/rum-electron@0.0.3'，这是一个面向 Electron 的真实用户监测（RUM）库；它上报哪些内容未经检查。","source":"ai","title":"使用情况监测 SDK"},{"code":"remote_assets_downloaded","severity":"info","evidence":".env.example：'ZCODE_REMOTE_ASSET_CDN_BASE_URL=https://cdn-zcode.z.ai/zcode/electron/releases/3.14.0' 和 'ZCODE_BASE_URL=https://zcode.z.ai'：该应用会从 Z.ai 的服务器获取运行时资源和内置的服务商配置。","source":"ai","title":"会下载远程资源"},{"code":"agent_shipped_as_bytecode","severity":"info","evidence":"package.json scripts：'build:desktop-agent:bytecode' 和 'dev:desktop:bytecode'：打包后的智能体可能已编译为字节码，这使它更难与源码进行比对。","source":"ai","title":"智能体以字节码形式发布"}],"platforms":["windows","mac"],"tools":[],"buildTools":["node"],"disk":{"minMB":143,"maxMB":429,"basis":"installer"},"apiKeys":["Z.ai account (OAuth login) or another model provider"],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"Apache-2.0","category":"permissive","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"claude-code","name":"Claude Code","confirmed":true},{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"github-copilot","name":"Copilot","confirmed":true}],"declaredAlternatives":[],"stars":8595,"downloads":8691,"downloadsWeek":1833,"downloadsDays":7,"growth":139,"growthDays":7,"history":[{"date":"2026-10-26","stars":8275},{"date":"2026-10-27","stars":8314},{"date":"2026-10-28","stars":8339},{"date":"2026-10-29","stars":8358},{"date":"2026-10-30","stars":8386},{"date":"2026-10-31","stars":8401},{"date":"2026-11-01","stars":8412},{"date":"2026-11-02","stars":8433},{"date":"2026-11-03","stars":8456},{"date":"2026-11-04","stars":8483},{"date":"2026-11-05","stars":8506},{"date":"2026-11-06","stars":8524},{"date":"2026-11-07","stars":8544},{"date":"2026-11-08","stars":8558},{"date":"2026-11-09","stars":8584},{"date":"2026-11-10","stars":8595}],"firstPublishedAt":"2026-10-27T01:10:12.709Z","commitSha":"3c5b227649371507ddc4eb46ac5133afe6db1cab","verifiedAt":"2026-10-27T18:15:07.907Z","lastCommitAt":"2026-11-03T09:30:55Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"founder","text":{"en":{"summaryShort":"Z.ai's AI coding assistant: a desktop app, web page and terminal tool where an AI agent writes and edits code in your projects.","summary":"ZCode is an AI programming workbench from Z.ai, the company behind the GLM models. It comes as a desktop app, a browser interface and a terminal agent that reads your project, writes code and runs tools, and it can work on remote machines over SSH or WSL. It is for programmers. Windows and macOS installers are on GitHub; you sign in with a Z.ai account or configure a model provider. Building from source needs Node.js 24 and pnpm. Prices and plan limits are not in the README.","easeRationale":"There is a 143 MB Windows .exe installer (v3.14.3); after installing you still need to sign in to Z.ai or set up a model provider before the agent works.","costNote":"The app is open source; it relies on Z.ai/GLM models through an account login or another model provider. The README points to NOTICE.md for plan and discount details and states no prices.","notVerified":["How much use is free and what a paid Z.ai plan costs.","What usage data the monitoring library and Z.ai services receive from your projects.","Whether the installer on GitHub matches this source, given the release is older than the code.","Which commands the AI agent runs on your machine and how it asks for permission."]},"zh-Hans":{"summaryShort":"Z.ai 的 AI 编程助手：桌面应用、网页和终端工具，由 AI 智能体在你的项目中编写和修改代码。","summary":"ZCode 是 Z.ai 推出的 AI 编程工作台，Z.ai 是 GLM 系列模型背后的公司。它提供桌面应用、浏览器界面和终端智能体三种形态，能读取你的项目、编写代码并运行工具，还可以通过 SSH 或 WSL 在远程机器上工作。它面向程序员。GitHub 上提供 Windows 和 macOS 安装程序；你需要用 Z.ai 账号登录，或配置一个模型服务商。从源代码构建需要 Node.js 24 和 pnpm。README 中没有写明价格和套餐限制。","easeRationale":"有一个 143 MB 的 Windows .exe 安装程序（v3.14.3）；安装后仍需登录 Z.ai 或配置模型服务商，智能体才能工作。","costNote":"应用本身开源；它通过账号登录或其他模型服务商使用 Z.ai/GLM 模型。README 指向 NOTICE.md 查看套餐和折扣详情，但没有列出价格。","notVerified":["免费额度有多少，以及 Z.ai 付费套餐的价格。","监控库和 Z.ai 服务会从你的项目中收到哪些使用数据。","鉴于发布的版本比代码更旧，GitHub 上的安装程序是否与当前源代码一致。","AI 智能体会在你的机器上运行哪些命令，以及它如何请求许可。"]}}},{"id":"R_kgDOVLVgEA","slug":"zcbacxc__mcd-party-planner","name":"MCD Party Planner","owner":"zcbacxc","fullName":"zcbacxc/mcd-party-planner","type":"needs_setup","status":"active","grade":"C","easeByPlatform":{"windows":"C"},"risk":"attention","signals":[{"code":"works_only_in_china","severity":"info","evidence":"README.md 和 .env.example 指向 https://open.mcd.cn/mcp，并通过手机号登录，所有示例都使用中国城市；它只与麦当劳中国的服务通信。","source":"ai","title":"只能在中国使用"},{"code":"can_place_a_real_paid_booking","severity":"attention","evidence":"README.md：派对预订默认是 dry_run（试运行），而 \"python -m src.cli party --city 成都市 --headcount 6 --confirm\" 会下达真实订单（示例中显示派对名额为 ¥50.0）。","source":"ai","title":"可能下达真实的付费预订"},{"code":"sends_your_search_to_mcdonalds","severity":"info","evidence":"README.md 说明，城市和关键词会作为查询参数发送给麦当劳的 MCP，输入内容只保留在内存中，不会写入磁盘。","source":"ai","title":"把你的搜索内容发给麦当劳"},{"code":"not_an_official_product","severity":"info","evidence":"README.md 第七节：这是参加麦当劳程序员节比赛的参赛作品，由个人独立开发，\"非麦当劳官方产品\"（不是麦当劳的官方产品）。","source":"ai","title":"不是官方产品"},{"code":"very_new_project","severity":"info","evidence":"项目信息：仓库创建于 2026-10-15，到 2026-10-27 已发布 7 个版本，1600 个星标。","source":"ai","title":"非常新的项目"}],"platforms":["windows","mac","linux"],"tools":["python"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":["McDonald's China MCP (MCD_MCP_TOKEN)"],"cost":{"model":"free"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["productivity","ai_skills"],"alternativeTo":[],"declaredAlternatives":[],"stars":1382,"downloads":0,"downloadsWeek":0,"downloadsDays":7,"growth":0,"growthDays":7,"history":[{"date":"2026-10-27","stars":1600},{"date":"2026-10-28","stars":1600},{"date":"2026-10-29","stars":1600},{"date":"2026-10-30","stars":1600},{"date":"2026-10-31","stars":1596},{"date":"2026-11-01","stars":1596},{"date":"2026-11-02","stars":1596},{"date":"2026-11-03","stars":1595},{"date":"2026-11-04","stars":1595},{"date":"2026-11-05","stars":1593},{"date":"2026-11-06","stars":1387},{"date":"2026-11-07","stars":1382},{"date":"2026-11-08","stars":1382},{"date":"2026-11-09","stars":1382},{"date":"2026-11-10","stars":1382}],"firstPublishedAt":"2026-10-27T18:40:28.119Z","commitSha":"40b91ffa00e189b4d8ed8011aa97ff52c0d8a124","verifiedAt":"2026-10-27T18:31:16.058Z","lastCommitAt":"2026-10-27T06:04:09Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Plans a group meal or a birthday party at McDonald's in China, with real store menus, real prices and booking slots.","summary":"You type a city, a neighbourhood, how many people and your budget, and this tool asks McDonald's China's official MCP service for nearby open stores, the real menu and the real price, then shows a few meal combinations with the total and the price per person. It can also look up birthday party slots and the campaign calendar. It works only with McDonald's China: you need a free token from open.mcd.cn (sign-in by Chinese phone number) and Python 3.9. It also runs as a WorkBuddy skill.","easeRationale":"You install Python 3.9, download the repository, request a token on open.mcd.cn and put it in a .env file, then run commands in the terminal; no extra packages are needed.","costNote":"The program is free and the McDonald's MCP token is free to request; the food and the party booking are of course paid.","notVerified":["Whether you can get a token without a Chinese phone number","Whether the commands run as shown on Windows","Whether the prices and store list match what you see in the McDonald's app on the day you run it","What exactly happens when you add --confirm and a real party order is created","Whether the offline replay snapshots in examples/ still match the current service"]},"zh-Hans":{"summaryShort":"帮你在中国麦当劳规划聚餐或生日派对,使用真实门店菜单、真实价格和预订时段。","summary":"你输入城市、商圈、人数和预算,这个工具就会向麦当劳中国官方MCP服务查询附近营业的门店、真实菜单和真实价格,然后给出几种套餐组合,包括总价和人均价格。它还能查询生日派对的预订时段和活动日历。它只支持中国麦当劳:你需要在open.mcd.cn申请一个免费token(用中国大陆手机号登录)并安装Python 3.9。它也可以作为WorkBuddy的一个skill运行。","easeRationale":"你安装Python 3.9,下载仓库,在open.mcd.cn申请一个token并填入.env文件,然后在终端里运行命令;不需要额外安装其他软件包。","costNote":"程序本身免费,麦当劳的MCP token申请也是免费的;当然,点餐和派对预订本身是要付费的。","notVerified":["没有中国大陆手机号是否也能申请到token。","这些命令在Windows上是否能按说明运行。","运行当天的价格和门店列表是否与麦当劳App里看到的一致。","加上--confirm参数、真正创建一个派对订单时具体会发生什么。","examples/目录里的离线回放快照是否仍与当前服务保持一致。"]}}},{"id":"MDEwOlJlcG9zaXRvcnkzNDA1NDc1MjA=","slug":"zed-industries__zed","name":"Zed","owner":"zed-industries","fullName":"zed-industries/zed","type":"ready_app","status":"active","grade":"A","easeByPlatform":{"windows":"A","mac":"A","linux":"A"},"risk":"attention","signals":[{"code":"license_not_detected_by_github","severity":"info","evidence":"GitHub 显示没有单一许可证（NOASSERTION）。该仓库同时带有 LICENSE-GPL 和 LICENSE-APACHE，README.md 写道：\"Zed source code is licensed primarily under GPL-3.0-or-later, with Apache-2.0 components where marked.\" GPL-3.0 是一种 copyleft（著佐权）许可证，所以你分发的改动必须在同样的条款下共享。","source":"ai","title":"github 未能识别出单一许可证"},{"code":"telemetry_built_in","severity":"attention","evidence":"Cargo.toml 中的 workspace 包含 `telemetry`、`telemetry_events`、`hang_telemetry` 和 `crashes` 这些 crate，说明该程序会发送使用数据和崩溃数据；README 没有说明具体收集了什么，也没有说明如何关闭。","source":"ai","title":"内置遥测功能"},{"code":"automatic_updates","severity":"info","evidence":"Cargo.toml 列出了 `auto_update`、`auto_update_helper` 和 `auto_update_ui` 这些 crate，说明已安装的程序可以自行下载并安装新版本。","source":"ai","title":"自动更新"},{"code":"sends_your_code_to_ai_services","severity":"attention","evidence":"`anthropic`、`open_ai`、`google_ai`、`mistral`、`deepseek`、`copilot_chat`、`cloud_llm_client` 和 `edit_prediction` 等 crate 是编辑器的一部分：当你使用 AI 面板或编辑预测功能时，你正在处理的文件会被发送给所选的服务商。","source":"ai","title":"把你的代码发送给 ai 服务"},{"code":"pre_release_published_alongside_stable","severity":"info","evidence":"最新的发布版本是 v1.27.0-pre（2026-10-28），这是一个与同日期的稳定版 v1.26.2 一同发布的预览通道。","source":"ai","title":"预览版与稳定版同时发布"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":["docker","rust"],"disk":{"minMB":62,"maxMB":185,"basis":"installer"},"apiKeys":["Anthropic","OpenAI","Google AI","GitHub Copilot"],"cost":{"model":"freemium"},"license":{"spdx":"NOASSERTION","category":"other","detectedFrom":"github"},"categories":["development","ai_agents"],"alternativeTo":[{"slug":"sublime-text","name":"Sublime Text","confirmed":true},{"slug":"cursor","name":"Cursor","confirmed":true},{"slug":"jetbrains-ides","name":"IDEs","confirmed":true}],"declaredAlternatives":[],"stars":94667,"downloads":38249657,"downloadsWeek":585574,"downloadsDays":7,"growth":2031,"growthDays":7,"history":[{"date":"2026-10-26","stars":92246},{"date":"2026-10-27","stars":92299},{"date":"2026-10-28","stars":92346},{"date":"2026-10-29","stars":92389},{"date":"2026-10-30","stars":92438},{"date":"2026-10-31","stars":92466},{"date":"2026-11-01","stars":92517},{"date":"2026-11-02","stars":92579},{"date":"2026-11-03","stars":92636},{"date":"2026-11-04","stars":92687},{"date":"2026-11-05","stars":92857},{"date":"2026-11-06","stars":93484},{"date":"2026-11-07","stars":94223},{"date":"2026-11-08","stars":94468},{"date":"2026-11-09","stars":94582},{"date":"2026-11-10","stars":94667}],"firstPublishedAt":"2026-10-27T01:10:12.710Z","commitSha":"a4b39c38e0056a5631a6f49f7fb13d4d1c4b6a9c","verifiedAt":"2026-11-01T01:41:49.176Z","lastCommitAt":"2026-11-10T23:00:03Z","isMirror":false,"collectFailing":false,"installers":[{"os":"windows","file":"Zed-x86_64.exe","sizeBytes":64751256,"url":"https://github.com/zed-industries/zed/releases/download/v1.27.0/Zed-x86_64.exe","releaseUrl":"https://github.com/zed-industries/zed/releases/tag/v1.27.0"},{"os":"mac","file":"Zed-aarch64.dmg","sizeBytes":88169290,"url":"https://github.com/zed-industries/zed/releases/download/v1.27.0/Zed-aarch64.dmg","releaseUrl":"https://github.com/zed-industries/zed/releases/tag/v1.27.0"}],"reviewedBy":"ai","text":{"en":{"summaryShort":"A fast code editor for Windows, Mac and Linux with built-in AI help and live editing with other people.","summary":"Zed is a code editor you download and install like any other program. It is written in Rust for speed, supports many languages, a terminal, a debugger, Git, extensions and themes, lets several people edit the same project together, and has a built-in AI panel that can work with Anthropic, OpenAI, Google, Mistral, DeepSeek, GitHub Copilot or a local model through Ollama or LM Studio. Installers are published for Windows, macOS and Linux (about 55-92 MB). There is no web version yet.","easeRationale":"A on all three: every release has a ready file to download and open — Zed-x86_64.exe for Windows, Zed-x86_64.dmg for macOS and zed-linux-x86_64.tar.gz for Linux — and the README also points to package managers for each system. Nothing else has to be installed to start editing.","costNote":"The editor is open source and free; the AI features need either your own API key or a paid plan from Zed Industries, Inc., the for-profit company that develops it.","notVerified":["Whether it runs smoothly on your graphics card, especially on Linux","What the telemetry crates actually send and whether a setting turns them off","How much the AI features cost in practice with your own key or a Zed plan","Whether the published installers were built from this source code"]},"zh-Hans":{"summaryShort":"一款适用于 Windows、Mac 和 Linux 的快速代码编辑器，内置 AI 辅助，并支持与他人实时协同编辑。","summary":"Zed 是一款像其他普通程序一样下载安装的代码编辑器。它用 Rust 编写以追求速度，支持多种编程语言、终端、调试器、Git、扩展和主题，可以让多人同时编辑同一个项目，还内置了一个 AI 面板，可以配合 Anthropic、OpenAI、Google、Mistral、DeepSeek、GitHub Copilot，或者通过 Ollama、LM Studio 使用本地模型。Windows、macOS 和 Linux 都发布了安装程序（大约 55-92 MB）。目前还没有网页版本。","easeRationale":"三个系统都是 A：每个发布版本都有现成的文件可以下载并打开——Windows 的 Zed-x86_64.exe、macOS 的 Zed-x86_64.dmg 和 Linux 的 zed-linux-x86_64.tar.gz——README 还为每个系统指出了对应的包管理器。不需要再安装别的东西就能开始编辑。","costNote":"该编辑器是开源且免费的；AI 功能需要你自己的 API 密钥，或者来自开发该编辑器的营利性公司 Zed Industries, Inc. 的付费计划。","notVerified":["它在你的显卡上是否能顺畅运行，尤其是在 Linux 上","遥测相关的 crate 实际发送了什么，是否有设置可以关闭它们","用自己的密钥或 Zed 计划实际使用 AI 功能要花多少钱","发布的安装程序是否确实由这份源代码构建而成"]}}},{"id":"R_kgDOU-TOoA","slug":"zhongerxin__iphone-use","name":"iPhone-use","owner":"zhongerxin","fullName":"zhongerxin/iPhone-use","type":"needs_setup","status":"active","grade":"E","easeByPlatform":{"mac":"E"},"risk":"attention","signals":[{"code":"telemetry_on_by_default","severity":"attention","evidence":"README.md：该插件默认会向一个 PostHog 项目和一个 GA4 属性发送匿名使用统计，内容包括工具调用、连接状态、耗时、错误类别以及它所启动应用的标识符；可以用 IPHONE_USE_ANALYTICS=0 或 DO_NOT_TRACK=1 关闭（详见 ANALYTICS.md）。","source":"ai","title":"默认开启遥测"},{"code":"ai_controls_your_real_phone","severity":"attention","evidence":"README.md 列出了 pua_tap、pua_swipe、pua_type_text、pua_launch_app 和 pua_batch 等工具，它们会直接操作已连接的 iPhone，示例包括操作微信的联系人和聊天列表。README 本身也提醒：超时或中断的操作结果可能不确定，而密码、验证码和 Face ID 必须由本人完成。","source":"ai","title":"AI 会操控你的真实手机"},{"code":"phone_developer_mode_and_resigned_app","severity":"attention","evidence":"README.md 要求信任这台 Mac、在 iPhone 上开启开发者模式，并安装用你自己的 Apple 团队签名的 WebDriverAgentRunner（固定的 WDA 提交 16.14.0），这会让一个自动化服务一直运行在手机上。","source":"ai","title":"手机需开启开发者模式并重新签名应用"},{"code":"readme_says_repository_is_private","severity":"info","evidence":"README.md 称该仓库目前是私有的，需要 GitHub 访问权限或源码包，但该仓库实际上是公开的，并且有 34 个发布版本：说明文字与项目的实际状态不一致。","source":"ai","title":"README 称仓库是私有的"}],"platforms":["mac"],"tools":["python","node","other"],"buildTools":[],"disk":{"minMB":null,"maxMB":null,"basis":"unknown"},"apiKeys":[],"cost":{"model":"free_with_paid_api"},"license":{"spdx":"MIT","category":"permissive","detectedFrom":"github"},"categories":["ai_agents","ai_skills","hardware"],"alternativeTo":[],"declaredAlternatives":[],"stars":4971,"downloads":448,"downloadsWeek":57,"downloadsDays":7,"growth":93,"growthDays":7,"history":[{"date":"2026-10-26","stars":4677},{"date":"2026-10-27","stars":4710},{"date":"2026-10-28","stars":4739},{"date":"2026-10-29","stars":4764},{"date":"2026-10-30","stars":4790},{"date":"2026-10-31","stars":4811},{"date":"2026-11-01","stars":4821},{"date":"2026-11-02","stars":4849},{"date":"2026-11-03","stars":4878},{"date":"2026-11-04","stars":4900},{"date":"2026-11-05","stars":4909},{"date":"2026-11-06","stars":4928},{"date":"2026-11-07","stars":4938},{"date":"2026-11-08","stars":4950},{"date":"2026-11-09","stars":4957},{"date":"2026-11-10","stars":4971}],"firstPublishedAt":"2026-10-27T01:10:12.711Z","commitSha":"3ce9c080534e89fab91a237bf1c53a4a7e9d80a7","verifiedAt":"2026-10-28T01:35:24.079Z","lastCommitAt":"2026-10-22T13:43:00Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Lets an AI assistant on your Mac open apps, tap, type and read screens on your own iPhone through a USB cable.","summary":"An add-on for AI assistants (Codex, WorkBuddy or Doubao) that gives them control of a real iPhone connected by USB. You describe a task in plain words and the assistant opens apps, reads the page, taps, scrolls, types text and collects lists, while a side panel shows the phone screen live. It only runs on a Mac and you need the full Xcode, your own Apple developer account, Python 3.9+, Node.js 20+ and an unlocked iPhone with developer mode on. The first setup builds and signs Appium's WebDriverAgent onto the phone.","easeRationale":"Mac only, and the hardest part is on the phone: the dependency table requires the full Xcode (not just Command Line Tools), an Apple account with a signing team, and building, signing and launching the WebDriverAgent app on your iPhone, plus Python, Node.js and npm. The README offers a prompt so the assistant can walk you through it, but Apple login, device trust, developer mode and signing still have to be done by hand.","costNote":"The code is free (MIT), but it only works inside a paid AI assistant such as Codex, WorkBuddy or Doubao.","notVerified":["Whether the WebDriverAgent build and signing really completes with a free or personal Apple account, and on which iOS versions","Whether the live screen preview and the taps work as described on each iPhone model","What exactly is sent to PostHog and GA4 while you use it, and whether the off switch stops all of it","How much disk space the full Xcode plus the WDA build take on the Mac","The cost of the AI assistant tokens consumed by long phone tasks"]},"zh-Hans":{"summaryShort":"让你 Mac 上的 AI 助手通过 USB 数据线打开应用、点击、输入文字,并读取你自己 iPhone 的屏幕内容。","summary":"这是一个面向 AI 助手(Codex、WorkBuddy 或豆包)的扩展,让它们能够控制通过 USB 连接的真实 iPhone。你用简单的语言描述一项任务,助手就会打开应用、读取页面内容、点击、滚动、输入文字并收集列表信息,同时一个侧边面板会实时显示手机屏幕。它只能在 Mac 上运行,你需要完整的 Xcode、你自己的苹果开发者账号、Python 3.9+、Node.js 20+,以及一台已解锁且开启了开发者模式的 iPhone。首次配置会在手机上编译并签名 Appium 的 WebDriverAgent。","easeRationale":"仅支持 Mac,而最难的部分发生在手机端:依赖列表要求安装完整的 Xcode(而不只是 Command Line Tools)、拥有签名团队的苹果账号,并在你的 iPhone 上编译、签名并启动 WebDriverAgent 应用,此外还需要 Python、Node.js 和 npm。README 提供了一段提示词,让助手引导你完成操作,但苹果账号登录、设备信任、开发者模式和签名仍然需要手动完成。","costNote":"代码本身免费(MIT 许可),但它只能在付费 AI 助手(如 Codex、WorkBuddy 或豆包)内部使用。","notVerified":["WebDriverAgent 的编译和签名是否真的能用免费或个人苹果账号完成,以及在哪些 iOS 版本上可行","实时屏幕预览和点击操作是否真的在每款 iPhone 机型上都能如描述那样工作","使用过程中究竟向 PostHog 和 GA4 发送了什么数据,以及关闭开关是否真的能停止全部数据发送","完整的 Xcode 加上 WDA 构建在 Mac 上一共占用多少磁盘空间","长时间手机任务所消耗的 AI 助手 token 费用"]}}},{"id":"R_kgDOVeLPDw","slug":"zortos293__lichii","name":"Lichii","owner":"zortos293","fullName":"zortos293/Lichii","type":"ready_app","status":"alert","grade":"B","easeByPlatform":{"windows":"B","mac":"B","linux":"B"},"risk":"high","signals":[{"code":"young_repo_many_stars","severity":"attention","evidence":"7天内获得5444颗星","source":"check","title":"仓库很新,星标却很多"},{"code":"needs_console_keys_and_game_dumps","severity":"attention","evidence":"README.md写道:\"Only use games and system files you have legally obtained. Games, keys and firmware are not included.\" Switch模拟器只有靠从主机上复制出来的解密密钥和固件才能运行,而复制游戏和系统文件,或者绕开主机的保护机制,在不少国家是受限甚至违法的,哪怕是你自己买的光盘和卡带也一样。","source":"ai","title":"需要主机密钥和游戏dump文件"},{"code":"fork_of_emulators_taken_down_after_legal_action","severity":"attention","evidence":"README.md写道:\"Lichii inherits the work of the Eglantine, Eden, yuzu and Sudachi projects\"。yuzu在被任天堂起诉后就被关停了,后续的分支项目也一再消失,所以这个下载和它的发布页面随时可能变得无法访问。","source":"ai","title":"是曾因法律诉讼被下架的模拟器的分支"},{"code":"repository_renamed_releases_point_to_other_repo","severity":"info","evidence":"README.md写道:\"Published builds belong on this fork's releases page\",链接指向github.com/zortos293/Eglantine/releases,还写道\"The repository still uses its historical Eglantine name\"。下载和问题反馈的链接,指向的仓库名字和卡片上的并不一样。","source":"ai","title":"仓库改了名,发布页却指向另一个仓库"},{"code":"very_new_repository_with_fast_growth","severity":"info","evidence":"项目数据:仓库创建于2026-10-31,已经有5444颗星和14个发布版本,最新的是2026-11-07的build-116;所有者自己写的仓库简介是\"VERY experimental Nintendo Switch 2 emulator\"。几乎没有历史记录可以用来评判这个项目或它的构建版本。","source":"ai","title":"仓库非常新,增长很快"},{"code":"crypto_donation_addresses_in_readme","severity":"info","evidence":"README.md在\"Support Lichii\"下列出了Solana、Ethereum、Bitcoin、BNB和Polygon的钱包地址,并警告说\"Coins sent on the wrong network can't be recovered\"。加密货币转账无法撤回,而README里的地址,任何获得写入权限的人都能改掉。","source":"ai","title":"README里有加密货币捐赠地址"},{"code":"needs_console_keys_and_firmware","severity":"high","evidence":"仓库简介写道:\"VERY experimental Nintendo Switch 2 emulator\";README.md写道:\"Lichii is an experimental, open-source Nintendo Switch and Switch 2 emulator written in C++\",关于用户需要自备什么,又写道\"Only use games and system files you have legally obtained. Games, keys and firmware are not included.\" 所以这个程序只有靠从主机本身取出的密钥、固件和游戏文件才能运行。按惯例,主机厂商的条款禁止绕过主机及其游戏内置的保护机制,而许多国家的法律也限制绕过版权保护的行为。","source":"ai","title":"需要主机密钥和固件"}],"platforms":["windows","mac","linux"],"tools":[],"buildTools":[],"disk":{"minMB":83,"maxMB":250,"basis":"installer"},"apiKeys":[],"cost":{"model":"free"},"license":{"spdx":"GPL-3.0","category":"copyleft","detectedFrom":"github"},"categories":["games"],"alternativeTo":[],"declaredAlternatives":[],"stars":8121,"downloads":1320790,"downloadsWeek":688593,"downloadsDays":3,"growth":7510,"growthDays":7,"history":[{"date":"2026-11-07","stars":5335},{"date":"2026-11-08","stars":6694},{"date":"2026-11-09","stars":7501},{"date":"2026-11-10","stars":8121}],"firstPublishedAt":"2026-11-06T22:57:16.784Z","commitSha":"89440b29a90708c0fefc126b2bd8c7a64a996bdd","verifiedAt":"2026-11-07T23:05:13.497Z","lastCommitAt":"2026-11-10T20:43:59Z","isMirror":false,"collectFailing":false,"installers":[],"reviewedBy":"ai","text":{"en":{"summaryShort":"Plays Nintendo Switch and Switch 2 games on a Windows, Mac or Linux computer, using game files you supply yourself.","summary":"Lichii is an experimental emulator that tries to run Nintendo Switch and Switch 2 games on a desktop computer. It is a fork of the Eglantine, Eden, yuzu and Sudachi emulators and is not connected to Nintendo. Ready builds exist for Windows (zip you extract and run), Mac (Apple Silicon .dmg) and Linux (AppImage), so no programming tools are needed. The README says games, keys and firmware are not included and must come from hardware you own. It is for people who want to try Switch games on a PC and accept crashes, graphics errors and games that do not work.","easeRationale":"On Windows you extract the release zip and run lichii.exe, with no tools to install; Mac (.dmg) and Linux (AppImage) are the same kind of download. It is a B rather than an A because nothing runs until you add the console keys, firmware and game files yourself, which the project does not provide.","costNote":"Free. The README asks for voluntary donations in crypto (Solana, Ethereum, Bitcoin, BNB, Polygon).","notVerified":["Whether the published builds for Windows, Mac and Linux actually start and play games on your hardware","Which games work and how well, since the README says compatibility and performance vary by game, driver and GPU","Whether the downloadable binaries match the source code in the repository","What the optional compatibility report sends to the project's website, beyond what the README describes","Whether using the emulator with keys and game files is allowed where you live"]},"zh-Hans":{"summaryShort":"在 Windows、Mac 或 Linux 电脑上玩任天堂 Switch 和 Switch 2 游戏，使用你自己提供的游戏文件。","summary":"Lichii 是一款实验性模拟器，尝试在桌面电脑上运行任天堂 Switch 和 Switch 2 游戏。它是 Eglantine、Eden、yuzu 和 Sudachi 模拟器的 fork，和任天堂没有关系。Windows（解压即可运行的 zip）、Mac（Apple Silicon 的 .dmg）和 Linux（AppImage）都有现成的构建版本，不需要任何编程工具。README 说游戏、密钥和固件都不包含在内，必须来自你自己拥有的硬件。它适合想在 PC 上尝试 Switch 游戏、并能接受崩溃、画面错误和无法运行的游戏的人。","easeRationale":"在 Windows 上，你只需解压发布的 zip 并运行 lichii.exe，不需要安装任何工具；Mac（.dmg）和 Linux（AppImage）是同类型的下载。评级是 B 而不是 A，因为在你自己添加主机密钥、固件和游戏文件之前什么都跑不起来，而这些东西项目方并不提供。","costNote":"免费。README 请求以加密货币自愿捐赠（Solana、Ethereum、Bitcoin、BNB、Polygon）。","notVerified":["发布的 Windows、Mac 和 Linux 构建版本是否真的能在你的硬件上启动并运行游戏","哪些游戏能玩、玩得怎么样，因为 README 说兼容性和性能会因游戏、驱动和 GPU 而不同","可下载的二进制文件是否和仓库里的源代码一致","可选的兼容性报告除了 README 里说的之外，还向项目的网站发送了什么","在你所在的地方，用密钥和游戏文件使用这个模拟器是否合法"]}}}]}