review-team

Adversarial PR reviews are automated by a five-member team—staff-engineer, qa-engineer, security-reviewer, architect, and devils-advocate—that cross-examines diffs and metadata to generate high-confidence findings. The team fetches GitHub PR context and aggregates results into a final report.

Project conventions and ticket context are applied during code review by enforcing CLAUDE.md standards and fetching ticket metadata alongside the PR diff. This ensures the automated review process maintains alignment with your established project standards.

The Devil's Advocate role works in an automated code review process by challenging the findings produced by the four specialist reviewers. This phase-based cross-examination step filters out weak observations, ensuring only high-confidence findings remain in the final aggregated report.

Yes, you can use this automated review process with your existing GitHub workflow without extra dependencies. It operates independently by fetching PR diffs, metadata, and ticket context, while persisting its own state locally under a .review-team directory.

To start a team-based PR review, provide a GitHub PR URL to initiate the workflow. The five-member review team will then fetch the diff, assign specialist tasks, execute the cross-examination phase, and compile the final review report.

A limitation of using an automated five-member team for risk assessment is that the review quality depends heavily on the fetched diff and ticket context. Complex architectural nuances or unwritten team conventions not explicitly documented in the metadata may be missed during the automated cross-examination.