pr-review-loop

What This Skill Does

The pr-review-loop skill automates the feedback and refinement cycle between an AI agent and the Greptile code review bot. It is designed to minimize the human effort required to move a pull request from submission to a production-ready state. When triggered, the agent autonomously pulls Greptile's review comments, parses quality scores, applies necessary code fixes, pushes the changes, and initiates re-review. This cycle repeats systematically until the code meets a quality threshold of 4/5, or until safety parameters (like max rounds or stagnant scores) trigger an escalation to a human supervisor.

Installation

To integrate this skill into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/cemoso/pr-review-loop

Ensure that your environment has GitHub CLI (gh) installed and authenticated, as the skill relies on the gh API to interact with pull requests, fetch comment history, and execute merge commands.

Use Cases

This skill is ideal for teams relying on CI/CD pipelines where Greptile is used as a primary code quality gate. Use it to:

• Automate routine bug fixes and linting corrections found during PR reviews.
• Ensure faster PR turnaround times without constant manual intervention.
• Standardize the process of addressing bot-generated feedback.
• Manage large-scale refactors where multiple rounds of review are expected.

Example Prompts

1. "pr review https://github.com/my-org/my-repo/pull/123"
2. "review my PR my-org/my-repo#456"
3. "pr status"

Tips & Limitations

• Safety First: The skill is designed with safety in mind. If the agent hits architectural feedback or repeatedly fails to improve the score, it will automatically escalate the issue to you via Telegram or console alert.
• Monitoring: Always monitor the review-state.json file in your workspace to track how many rounds the agent has spent on a specific PR.
• Scope: The skill handles code-level feedback best. Architectural or high-level design changes will trigger the escalation protocol, preventing the agent from making potentially breaking changes without human oversight.
• Consistency: Use clear descriptive commit messages when the agent applies fixes so that human collaborators can audit changes easily.