Minimum path
The shortest way to see the loop work. It assumes Node ≥ 22, gh signed
in (gh auth status), Claude Code, and a GitHub repo you can push to and
label. Before you start, read Risks and responsibilities. The loop
spends your Anthropic credits and your GitHub Actions minutes.
In the repo, in a terminal:
npx @rtorcato/repo-ai setup
Then in Claude Code, in the same repo:
/ai-issue
/ai-loop
setup installs the skills, creates the loop's labels, and allows the loop's
shell calls in this checkout. It asks before each step, then runs doctor.
/ai-issue files a small issue labelled ai-ready. /ai-loop picks it up,
opens a PR, has two agents review it, and hands it to you labelled
merge-ready. You read the diff and merge it. The loop keeps itself going
until you run /ai-loop-stop.
That's all you need. Everything else is optional, and Getting started covers it: running the loop as a bot account, the repo settings it prefers, the sandbox, stopping, and releasing.