pr-reviewer

Multi-agent code review is automated by coordinating four specialized sub-agents to analyze PR diffs, metadata, and modified files. Security, Logic, UX, and Conventions reviewers score findings on a 0–100 rubric, returning only high-confidence issues that score 80 or above in a consolidated report.

The PR review scoring mechanism works by having each of the four reviewer agents independently score findings on a 0–100 rubric. The orchestrator filters these results and returns only high-confidence findings that score at least 80, grouped by reviewer in a consolidated report.

Yes, you can review pull requests from GitHub or GitLab by providing a PR URL or number. The skill maintains full context including PR_DIFF, PR_META, and modified file contents to execute the multi-agent review across both platforms.

The best way to surface critical security issues is using a dedicated Security reviewer agent that analyzes the PR diff alongside modified files. It scores security findings on a 0–100 scale, filtering to only high-confidence results so engineers can focus remediation efficiently.

Yes, the multi-agent PR reviewer checks UX and code conventions by spawning dedicated UX and Conventions sub-agents. These agents analyze the PR diff and metadata, scoring findings to ensure only high-confidence UX and convention issues are returned in the final consolidated report.

The automated PR review process maintains full context including PR_DIFF, LOCAL_DIFF, PR_META, PR_SUMMARY, CLAUDE_MD_CONTENTS, and the contents of modified files. This context is passed to all four reviewer agents to ensure accurate and high-confidence findings.