t3dotgg/tstack

In one line

You give your AI coding agent a short phrase like "full send" or "council", and tstack turns that phrase into a multi-step routine for handling a pull request (PR) — from review to merge — inside tools like Claude Code or Codex.

What it is

A set of small instruction files (called "skills") that tell the agent how to perform specific GitHub workflows. They were built for the author's own daily use, then cleaned up so anyone can install them. Some skills use a helper tool called T3 Code when available, but fall back to the agent's built-in tools when T3 Code isn't present.

The skills (trigger phrase → what happens)

Trigger	Purpose
`full-send`	Takes a merged PR that nobody is reviewing and drives it to completion: proves the change works, reviews it, files it, watches the PR through checks, and merges it.
`full-stop`	Same as `full-send`, but stops before merging and hands it back to you to merge yourself.
`council`	Gets review from three different AI models (from three different providers) and keeps looping until all three agree.
`sol loop`	Runs a review loop with a specific model called GPT Sol, going back and forth until both the agent and GPT Sol agree on the change.
`file a PR`	Files a short pull request that leads with the problem it solves rather than a bulleted list of every file changed.
`take over`	Picks up someone else's unfinished PR, branch, or agent thread, finishes it, and keeps the original author's credit (does not steal attribution).
`babysit`	Watches a PR through CI (automated checks) and review bots, and fixes only the findings that are actually valid.
`modernize`	Rebases an old PR onto the latest code and then decides whether the PR should still exist at all.
(automatic)	Formats any comments the agent posts on your behalf, so they look consistent (used by other skills).

How skills depend on each other

Some skills are built from others, so you need to install the whole set. The README lists these chains:

full-send → uses sol-loop, file-pr, babysit-pr
full-stop → uses full-send
takeover → uses modernize, file-pr, babysit-pr, leaving-pr-comments
file-pr, babysit-pr, modernize → use leaving-pr-comments

The README advises installing all the skills together.

Install & setup

Clone the repo: git clone https://github.com/t3dotgg/tstack
Copy the skills folder into your agent's skills directory:
- Claude Code: copy tstack/skills/* to ~/.claude/skills/
- Codex: copy tstack/skills/* to ~/.codex/skills/
Requirements: an authenticated GitHub CLI (gh), plus python3 for the babysit-pr skill's wait script. The council and sol-loop skills need subagents from other providers' models — T3 Code can spawn those via delegate_task, otherwise you need a harness that can run those models.

Customization

These skills are meant as starting points. The README encourages editing them, and keeping personal settings (such as which models sit on your "council") in a CLAUDE.md or AGENTS.md file — so you can update the skills without losing those personal choices.

License

MIT.