diagnose-ci

Diagnose CI

Investigate and diagnose CI pipeline failures.

Steps

1. Find the failure:

   • If a run_id is provided, use gh run view <run_id> --log-failed.
   • Otherwise, run gh run list --status=failure --limit 5 --json databaseId,name,headBranch,conclusion,createdAt to find recent failures.
   • Pick the most recent failure on the current branch (or the most recent overall if none match).

2. Pull logs: Run gh run view <id> --log-failed to get the failed step logs. If output is large, focus on the last 100 lines of the failing step.

3. Diagnose: Analyze the logs to identify:

   • The specific step that failed
   • The root cause (compile error, test failure, lint issue, missing secret, timeout, etc.)
   • The relevant file(s) and line(s) if applicable

4. Recommend fix: Based on the diagnosis (read-only: recommend, do not apply):

   • If it's a code issue: show the exact fix and point the user to fix-and-retry to apply + re-ship it.
   • If it's a config issue (missing secret, wrong action version): explain what to change.
   • If it's flaky (timeout, network): suggest re-running with gh run rerun <id>.

5. Report: Present a concise summary:

   ## CI Failure: <workflow name> (run #<id>)
   - Branch: <branch>
   - Failed step: <step name>
   - Root cause: <explanation>
   - Fix: <what to do>
   

Rules

• Always check the current repo's workflows first (gh run list).
• If gh is not authenticated or the repo has no remote, report that clearly.
• Don't blindly re-run failed pipelines; diagnose first.

Gotchas

• gh run view <id> --log-failed returns empty for a cancelled run (no step is marked "failed", so there is nothing to dump). Fall back to gh run view <id> --json jobs and inspect each job's conclusion/steps to find the cancelled or failed step, then read its log directly.