What problem does it solve? GitHub merge queues sometimes evict PRs with no visible test failure, and CI jobs hang silently on self-hosted runners. This Skill provides a systematic diagnostic path to distinguish required-check timeouts from real test failures, identify Blacksmith sticky-disk read failures and WarpCache corruption, and execute the correct runner-provider mitigation. ## Core Features & Use Cases - Merge-Queue Forensics: Query the GitHub GraphQL API for RemovedFromMergeQueueEvent reasons and timestamps to determine whether a PR was evicted by failed checks or job timeouts. - Runner Fleet Identification: Determine which fleet (Blacksmith, WarpBuild, or GitHub-hosted) ran a job and compare sticky-disk installs against cache-based control runs to isolate infrastructure degradation. - Mitigation & Escalation: Flip USE_GITHUB_RUNNERS or CI_RUNNER_PROVIDER repository variables to switch fleets, and escalate to vendors with the exact runner names, disk keys, and log evidence they require. - Use Case: A PR is removed from the merge queue at 04:01 UTC with failed_checks but no failing test. Use this Skill to confirm the unit job hit its 7-minute timeout after a stalled bun install on a Blacksmith sticky disk, then fail over to WarpBuild and re-enqueue the PR. ## Quick Start Ask the assistant to investigate why PR 31592 was evicted from the merge queue and determine whether the CI failure was caused by a runner infrastructure problem.
