watch-prlisted

# /watch-pr — wait for the review, then triage. Bounded. After a push, poll the PR until automated review and CI checks have landed, then address what they found — instead of the user relaying review comments back manually. **Deliberately narrow: it only touches what the review + CI flagged. It does NOT audit scope, re-review architecture, or improve things nobody asked about.** (That open-endedness is why designs like this historically never terminated.) ## Hard bounds (the whole point) - **Scope = review findings + CI failures only.** Nothing else gets touched. If you notice an unrelated issue, capture it with `/lesson note` — do not fix it here. - **Max 2 fix→re-review rounds.** After the 2nd, stop and report remaining findings to the user even if not all are resolved. - **Terminate explicitly** when: no actionable findings remain, OR all remaining are deferred-to-user, OR the round cap is hit. Say "**watch-pr done**" and stop. ## Loop 1. **Wait.** Poll `gh pr checks` + `gh pr view --json reviews,comments` periodically until CI completes and reviews are posted. Use long-interval waits, not a tight spin. Give up with a clear message if nothing lands after a reasonable window (~30 minutes). 2. **Fetch ALL comment types** — inline review comments, review bodies, and issue comments (per `rules/agent-operational.md`, with the three `gh api` endpoints listed there). 3. **Triage each finding into one of four buckets:** - **Mechanical** (lint, obvious