Fix CI Skill

This skill helps fix failing CI checks in GitHub repositories.

Workflow

1. •

   Understand the Changes First

   • •Detect the base branch: gh repo view --json defaultBranchRef -q '.defaultBranchRef.name'
   • •Run git diff <base>...HEAD or git log --oneline <base>..HEAD to see what changed
   • •Read the modified files to understand the intent and context of the changes
   • •This context is critical - CI failures often relate directly to the changes made
2. •

   Check CI Status

   • •Run gh pr checks or gh run list to see current CI status
   • •If no PR exists, check the latest workflow runs on the current branch
3. •

   Analyze Failures

   • •For each failing check, run gh run view <run-id> --log-failed to get failure logs
   • •Cross-reference failures with the changes from step 1
   • •Identify the root cause of each failure
4. •

   Fix Issues

   • •If the fix is straightforward (linting, formatting, type errors, test fixes):
     • •Make the necessary code changes
     • •Stage and commit with a descriptive message
     • •Push the changes
   • •If the fix is unclear or involves architectural decisions:
     • •Explain the issue to the user
     • •Present options if multiple approaches exist
     • •Ask for permission before proceeding
5. •

   Wait and Verify

   • •After pushing, use gh run watch --exit-status to monitor the CI run
   • •Exit status 0 means CI passed - report success
   • •Non-zero exit status means CI failed - repeat from step 3

Commands Reference

# Check PR status
gh pr checks

# List recent workflow runs
gh run list --branch $(git branch --show-current)

# View failed run logs
gh run view <run-id> --log-failed

# Watch a run in progress (exits non-zero if run fails)
gh run watch <run-id> --exit-status

# Re-run failed jobs
gh run rerun <run-id> --failed

Guidelines

• •Always understand the diff first - don't blindly fix errors without knowing what changed
• •Always read error logs carefully before making changes
• •Prefer minimal, targeted fixes over large refactors
• •If a test is flaky, mention it to the user rather than silently retrying
• •Never skip tests or disable CI checks without explicit user approval
• •If secrets or environment variables are missing, ask the user for guidance