Found loops! (loops.elorm.xyz). The site has ready-made agent loop templates you copy straight into your coding agent.
If you use Claude Code or Cursor, you know the pattern: define a goal, pick a check command, set an exit condition, cap the iterations. Not hard, but you end up writing it from scratch every time. This site has 40 of them packaged by category. Copy the kickoff prompt, paste, let the agent self-pace.
Ship PR Until Green - Give it a feature spec. It implements, runs tests, pushes, opens a PR, checks CI via gh pr checks, fixes what's red. Up to 10 iterations. Comes back when the PR is merge-ready.
CI Failure Watcher - Polls CI every 5 minutes. When a run goes red, it reads the logs, finds root cause, fixes locally, pushes. You come back to a green branch.
Guardrails Learning Loop - If a check fails the same way twice, the agent logs the pattern to a guardrails file. Next iteration reads that file first. Ten iterations in, the agent knows what not to try.
Pre-Commit Guard - Hook-based. Install it once, tests run automatically before every commit. Red suite blocks the commit. You don't trigger it.
40 loops across 16 categories. Works with Cursor, Claude Code, Codex, Gemini CLI, OpenCode. Most are prompt-only (copy and paste). A few hook-based ones ship a zip you extract at project root.
