diagnose-run

name	diagnose-run
description	Diagnose a GitHub Actions workflow run. Use when: check run, diagnose run, analyze run, what failed, run status, debug workflow, why did it fail, check build.
argument-hint	<workflow> [run-id|run-url|branch]
user-invocable	true
disable-model-invocation	false

Analyze Workflow Run

Preflight

Complete the shared repository skill preflight before continuing.

Steps

1. Identify the workflow 1.0. Resolve the repository 1.1. List available workflows 1.2. Exclude files ending in -reusable.yml; they cannot be selected directly by users 1.3. If the user did not provide a workflow, show the available workflows and ask which one 1.4. Validate the provided workflow against the available workflow file names or their name: values
2. Identify the run within the selected workflow 2.1. User provides run ID or URL — extract the numeric run ID and verify the run belongs to the selected workflow using view run summary 2.2. User provides a branch name — find latest run for branch 2.3. If none provided, list recent runs for the selected workflow and ask which run to diagnose
3. Get run status and identity — view run summary, retaining both the database ID used by gh run view and the workflow run number used in artifact names 3.1. If all jobs succeeded → report success, show duration, done 3.2. If any jobs failed or cancelled → continue to step 4
4. Identify failed jobs — get job details 4.1. List each job with its status (success/failure/cancelled/skipped) 4.2. For failed jobs, note which step failed
5. Fetch failure logs — get failed logs 5.1. Scan for error patterns: stack traces, Error:, exit code, OOM, timeout 5.2. Summarize root cause per failed job 5.3. If a benchmark experiment failed (log shows Experiment <name> failed to complete): - Extract the experiment name and architecture from the error line
   • Download benchmark artifact (<run_number>-benchmarks-<arch>)
   • If the artifact exists, read <experiment_name>/stderr.txt for the actual error
   • If it is missing or expired, report that limitation and use the available run log
6. Report — present findings: 6.1. Overall status + duration 6.2. Per-job breakdown (pass/fail/skip) 6.3. For failures: failed step, error summary, relevant log excerpt