checksSafety 95Repository ShareFavorite skill

Monitor CI/CD pipeline checks for the current branch and help diagnose any failures. Supports both GitHub and GitLab.

When to use

• After pushing commits or creating a PR/MR.
• When CI/CD pipeline is running and you want to wait for results.
• When a pipeline fails and you need to diagnose the failure.

When NOT to use

• When there is no remote branch or PR/MR and no recent pipelines.
• When you want to fix failures automatically. This skill only diagnoses.

Arguments

This skill accepts optional arguments after /checks:

• No arguments: check the current branch.
• A PR/MR number (e.g. 123): check that specific PR/MR instead of the current branch.

Steps

1. Detect the git platform and CLI tool:

   • Run git remote get-url origin to get the remote URL.
   • If the URL contains github.com, use gh (GitHub CLI).
   • If the URL contains gitlab (e.g. gitlab.com or a self-hosted GitLab), use glab (GitLab CLI).
   • If neither matches, ask the user which platform they use.
   • Verify the CLI tool is installed with which <tool>. If not installed, stop and tell the user.

2. Run git branch --show-current to get the current branch name (skip if a PR/MR number was provided).

3. Check if a PR/MR exists:

   • If a number was provided, use that directly.
   • Otherwise, check the current branch:
     • GitHub: run gh pr view --json number,url,statusCheckRollup.
     • GitLab: run glab mr view.
   • If no PR/MR exists, fall back to checking branch pipelines directly:
     • GitHub: run gh run list --branch <branch> --limit 5.
     • GitLab: run glab ci list --branch <branch>.
     • If no pipelines exist either, say so and stop.

4. Check the status of all pipeline checks:

   • With PR/MR:
     • GitHub: run gh pr checks.
     • GitLab: run glab ci status.
   • Without PR/MR (branch pipelines only):
     • GitHub: run gh run list --branch <branch> --limit 1 to find the latest run, then gh run view <id>.
     • GitLab: run glab ci status.

5. If all checks pass, report success and stop.

6. If checks are still running, wait with a timeout:

   • GitHub: run gh pr checks --watch (or gh run watch <id> for branch pipelines) with a 10-minute timeout.
   • GitLab: run glab ci status --wait with a 10-minute timeout.
   • If the timeout is reached, report that checks are still running and show the URL for the user to monitor manually.

7. If any checks failed:

   • GitHub:
     • Identify the failed check names from the output.
     • Run gh run list --branch <branch> --limit 5 to find the run IDs.
     • For each failed run, get the logs with gh run view <id> --log-failed.
   • GitLab:
     • Run glab ci view to see the pipeline overview.
     • Identify failed jobs and run glab ci trace <job-id> to get the logs.
   • Before suggesting a fix, search for an existing fix:
     • Check recent commits on the branch: git log --oneline -5.
     • Check open PRs/MRs that might address it:
       • GitHub: gh pr list --search "<failed check name>".
       • GitLab: glab mr list --search "<failed check name>".
     • If an existing fix is found, report it and stop.

8. Present the diagnosis using this format for each failure:

   ### <check name>
   **URL:** <direct link to the failed check>
   **Error:**
   <relevant error message, file/line if available>
   
   **Log excerpt:**
   <the most relevant 10-20 lines from the failure log>
   

9. After diagnosing, suggest next steps but do not automatically fix anything.

Rules

• Always detect the git platform from the remote URL. Never assume GitHub or GitLab.
• Always search for existing fixes before suggesting corrections.
• Always include the direct URL to each failed check in the output.
• Always use a timeout when waiting for checks. Never wait indefinitely.
• Never mark a task as complete if checks are still running or failing.
• Do not automatically fix failures. Present the diagnosis and let the user decide.
• If the required CLI tool (gh or glab) is not installed, stop and tell the user.

Related skills

• /pr - Create or update a PR/MR before checking pipelines.
• /review - Review the PR/MR code after checks pass.
• /commit - Fix issues locally and commit before pushing again.