What problem does it solve? When a CI check goes red, teams waste time re-running jobs, dismissing failures as flakes, or misreading GitHub check state. This Skill enforces a disciplined workflow: read the actual failing log, diagnose to root cause, and push a real fix instead of guessing or blindly re-running. ## Core Features & Use Cases - Root-cause doctrine for red checks: Treats every failure as a real bug until the log proves otherwise, forbidding opening assumptions like "flake", "pre-existing", or "infra" and requiring root-cause fixes even for proven flakes. - GitHub check-state reading: Covers how to read PR check status correctly with gh versus mcp__github__* tools, including pagination traps, mergeable_state interpretation, draft-PR deferred suites, and timeout-versus-cancellation detection. - Deduplication and claim coordination: Searches issues and PRs before fixing shared reds, files claim issues to prevent duplicate work across sessions, and batches one fix push per round to avoid cancelling in-flight runs. - Use Case: A PR shows a red check after a webhook notification. Instead of re-running the job, you read the failing step's log excerpt, identify a race condition in a test, dispatch a subagent to fix it at the root, and verify the whole round goes green. ## Quick Start Ask the assistant to diagnose why the CI check on my pull request is failing and fix the root cause instead of re-running it.
