ci-sweeper

name	ci-sweeper
description	Watches main's CI. On a red run, classifies flaky vs real and drafts a suggested fix in a report. Read-only — does not touch a worktree or open a PR until promoted past L1.

ci-sweeper procedure

The prompt carries intent; this skill carries the procedure — keep it a checklist, not an essay.

1. Read — the triggering CI-completion event for main (or, on the reconciliation sweep, any failing run on main not yet in ci-sweeper-state.md's per-failure register).
2. Classify — flaky (timeout, runner error, transient network — no code change needed) or real (reproducible failure — needs a fix).
3. Write ci-sweeper-report.md: one line per failure — [classification] check name · why it matters · suggested (NOT applied) fix. For a real failure, the suggested fix should be concrete enough that a human could apply it directly — but do not apply it yourself.
4. Record — update ci-sweeper-state.md's per-failure status, append one line to loop-run-log.md, commit both.
5. Take NO other action — no worktree writes, no PRs, no pushes. This loop is L1 report-only until a human promotes it (see LOOP.md → Promotion to L2).