Planning with Files
OthmanAdi/planning-with-filesReviewed by AI
Checked by RepoPicks Radar on
What is Planning with Files?
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.
How easy is it to install Planning with Files?
C · install 1 or 2 tools · Windows
- Windows C
- macOS C
- Linux C
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+".
Planning with Files on GitHub
- Stars
- 27,989on Nov 10, 2026
- Downloads
- —all releases
- Growth
- +105stars in 6 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Node.js, Other tools
- Disk space
- not determined
- Accounts or keys
- None
- Cost
- Free, API costs extra · 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.
- License
- MIT (Permissive)
- Can replace
- Manus
- Last change on GitHub
- Nov 10, 2026
Planning with Files risk signals (7)
- Attentioninstalls scripts that run by themselves in every session
The quick install is `npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g`, a global install straight from the GitHub repository, and the project describes itself as "Not a prompt it might follow. A hook that fires every turn". The repository carries a `hooks/` and `scripts/` folder plus plugin folders for 17 assistants, so after one command those scripts run automatically inside your assistant's sessions, with your files.
AI review
- Infothe benchmark numbers are the authors own
The badges "benchmark-96.7% pass (29/30)" and "blind A/B 3/3 wins", the claim that the plan "turned a 13.3-turn re-orientation into 5.0 turns" and "One hook fire measures 289ms wall-clock" all come from the project's own docs/evals.md. The README itself points to "numbers and limits", and no independent test is cited.
AI review
- Inforeadme promotes a paid sponsor to connect to your agent
A Sponsor block near the top advertises "Scrape.ly by evomi" with tracking links (scrape.ly/?utm_source=github&utm_medium=readme&utm_campaign=planning-with-files) and invites you to "Connect the Scrape.ly MCP server to Claude Code, Cursor or any MCP client". There is also an "OSS code review support" block for CodeRabbit. The advice sits in the same README as the install steps.
AI review
- Infothe plan files are thrown away by design
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." Notes you want to keep have to be moved somewhere else by hand, or archived with `set-active-plan.sh --archive <slug> --yes`.
AI review
- Infowhat you get depends on your assistant
The README says "Commands and model-callable tools depend on the host; Qoder CLI loads the skill and lifecycle hooks without the Claude-specific command prompts", lists 13 hosts with native plugins and reaches its "60+" and "70+ others" figures through the Agent Skills file standard, with "the `npx skills` installer alone targets 71 agents". Only some of those are tested integrations.
AI review
- Infovery fast release pace
Created on 2026-01-03, the repository already has 127 releases, the newest being v3.28.1 from 2026-11-07 (one day before this check), and the README documents behaviour that changed in v3.6.0, v3.7.0, v3.10.0, v3.11.0 and v2.36.0. With 27953 stars and 17 open issues the project is busy, but a skill that reinstalls itself globally changes often.
AI review
- Infoproject site on a generated address
The repository's homepage field is https://gilded-laurel-c7nj.here.now/, a random-looking hosting subdomain rather than a name belonging to the project, while install counts are shown through third-party badge services (skills.sh, skillsplayground.com, skill-history.com).
AI review
Evidence is shown in its original language.
What Radar could not check about Planning with Files
- 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