← ClaudeAtlas

ci-watchlisted

Diagnose failing CI checks, isolate the first real blocker (vs. noise), and surface the smallest viable fix. Use when a PR has failing checks, the pipeline is broken, or you need to understand why CI is red before proceeding with merge or release.
LucasSantana-Dev/sharekit-profile · ★ 1 · Data & Documents · score 70
Install: claude install-skill LucasSantana-Dev/sharekit-profile
# ci-watch Use for broken checks, flaky tests, or merge-blocking pipeline noise. ## Steps 1. Identify the active PR or HEAD commit. 2. List non-passing checks. 3. Separate required failures from advisory noise. 4. Inspect the first failing job deeply. 5. Name the smallest fix or the exact reason the failure is unrelated. ## Output Return: - failing job - first bad signal - likely cause - likely owner surface - smallest viable fix - whether it blocks shipping now ## PR state machine Read `gh pr view N --json mergeable,mergeStateStatus,reviewDecision` first, then: | mergeable / state | Action | |---|---| | `MERGEABLE` + `CLEAN` | proceed to merge | | `MERGEABLE` + `UNSTABLE` | non-required check failing or pending; poll required-only checks | | `MERGEABLE` + `BEHIND` | `gh pr update-branch` or local rebase + force-push | | `MERGEABLE` + `BLOCKED` | check `reviewDecision` and branch protection; if self-PR → enforce_admins toggle pattern | | `CONFLICTING` + `DIRTY` | local rebase first; if `git merge-tree` reports clean but GH disagrees → webhook desync, close+recreate PR | | `UNKNOWN` + `UNKNOWN` | GitHub still computing — wait 15s, recheck once; if STILL UNKNOWN, try `gh pr merge` directly (may already be merged) | ## Polling with Monitor For checks that take >1 min to settle, use the `Monitor` tool with an until-loop instead of busy-waiting: ``` Monitor command: until s=$(gh pr view N --json statusCheckRollup); \ pend=$(echo "$s" | python3 -c "..."); [ "$pend" = "