repo-guardian

What This Skill Does

Repo Guardian is an automated governance and maintenance layer for GitHub repositories. Designed to ensure high code quality and repository health, it acts as a gatekeeper for your pull requests and a concierge for your issue tracker. By employing a dual-model consensus mechanism—where two distinct AI models evaluate every pull request independently—the skill eliminates bias and captures edge cases that single-model reviews might miss. It orchestrates the entire PR lifecycle: if both models reach a consensus on approval, the skill auto-merges the PR; if changes are required or critical vulnerabilities are detected, it posts detailed feedback and appropriate labels to the thread.

Beyond PRs, Repo Guardian performs intelligent issue triage. It assesses the complexity of incoming issues to determine if they are actionable for the gh-issues pipeline or require manual human intervention. This creates a friction-less developer experience where minor bugs and feature requests are addressed or labeled without human oversight, while complex problems are correctly prioritized.

Installation

To integrate Repo Guardian into your OpenClaw environment, execute the following command in your terminal:

clawhub install openclaw/skills/skills/bloodandeath/repo-guardian

Once installed, you can configure recurring repository maintenance via cron by adding an entry to your ~/.openclaw/cron/jobs.json file. This allows the guardian to check your repositories every six hours or on your custom schedule.

Use Cases

• Automated PR Governance: Automatically validate PR quality without manual intervention by running the dual-model review process.
• Repository Maintenance: Keep your issue board clean by automatically triaging and labeling new tickets based on complexity and scope.
• CI/CD Quality Gates: Enforce a zero-trust policy where all code changes must pass automated architecture, security, and quality analysis before entering the main codebase.
• Consistency Enforcement: Ensure that code styles and best practices are applied across multiple repositories consistently by leveraging the same dual-model prompts.

Example Prompts

1. "Run the repo-guardian for my-org/web-frontend to clear out the current backlog and review pending PRs."
2. "Check the status of the repository governance and perform a full triage on all open issues in the backend-service repo."
3. "Configure a cron job to trigger the repo-guardian every 12 hours for the core-library repository."

Tips & Limitations

• Scope Boundaries: Always remember that Repo Guardian is for reviewing and triaging. If you need to actually write code or implement a fix, hand off the task to the gh-issues skill. Attempting to use this skill for code authorship will result in errors.
• Dual-Model Fallback: The skill uses Opus and GPT-5.4 by default. If you encounter errors, ensure your API tokens for these specific models are active and authenticated in your OpenClaw settings.
• Human-in-the-loop: Always review the labels applied by the agent. If an issue is labeled 'needs-human', it means the agent identified high complexity that potentially involves breaking changes or sensitive infrastructure updates.
• Configuration: Use the provided bash scripts for manual overrides if a specific PR requires an immediate, one-off review outside of the scheduled cron cycle.