tstack
t3dotgg/tstackReviewed by AI
What it does
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.
How easy it is to install
C · install 1 or 2 tools · Windows
- Windows C
- macOS C
- Linux C
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.
Numbers on GitHub
- Stars
- 2,536on Oct 31, 2026
- Downloads
- —all releases
- Growth
- +2,536stars in 7 days
Stars over time
Facts
- Runs on
- Windows, macOS, Linux
- You need to install
- Python, Other tools
- Disk space
- not determined
- Accounts or keys
- GitHub (through an authenticated gh CLI), Models from other providers, for the council and sol-loop reviews
- Cost
- Free · The skills are MIT. The AI assistants they run in, and the extra models the review skills call, are paid separately.
- License
- MIT (Permissive)
- Can replace
- CodeRabbit
- Last update
- Oct 28, 2026
Risk signals (5)
- Attentionagent can merge pull requests unattended
README skill table: `full-send` "Takes a change to a merged PR with nobody watching: proves it, reviews it, files it, babysits it, merges it." The `full-stop` skill is listed as the variant that stops before the merge.
AI review
- Attentionacts on your github account
README Requirements: "An authenticated GitHub CLI (gh)". The skills file pull requests, post comments on your behalf (`leaving-pr-comments`) and take over other people's branches, all under your account.
AI review
- Infosome features need another tool
README: some skills "use T3 Code features when they are available and fall back to native tools when they are not", and `council` and `sol-loop` "need subagents on other providers' models", which T3 Code does with `delegate_task`.
AI review
- Infovery new repository fast growth
GitHub facts: repository created 2026-10-24 and at 2471 stars six days later, with no releases and only a README plus the skills folder.
AI review
- Infoinstructions run as your assistant
The whole repository is text that your AI assistant reads and acts on (top-level files: LICENSE, README.md, skills/). Updating the skills means copying the new text in again, and the README tells you to edit them: "These are starting points. Edit them."
AI review
Evidence is shown in its original language.
What Radar could not check
- 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.