What problem does it solve?
Manually reviewing dozens of daily GitHub Actions runs to distinguish real problems from expected noise is tedious and error-prone. This Skill automates that triage, classifying failures as expected or actionable and filing a GitHub issue only when genuine problems exist.
Core Features & Use Cases
- Failure Classification: Applies explicit rules to ignore expected failures (Nightly Runner Cleanup, cascading CI failures, CLA Assistant, cancelled runs) and flag actionable ones (permission errors, main-branch CI breakage, infrastructure issues).
- Severity Assessment: Investigates failed runs via
gh run view logs and categorizes findings as SEVERE, MODERATE, or LOW.
- Deduplicated Issue Filing: Checks for existing open
workflow-health issues before creating a new one, adding comments instead of duplicates.
- Use Case: Run a daily health check on the dyad repository to catch a broken CI pipeline on main or a misconfigured secret before it blocks all contributors.
Quick Start
Ask the assistant to check GitHub Actions workflow runs from the past 24 hours and file an issue if any actionable failures are found.