review

Multi-agent PR review works by reading pull request metadata and diffs to assign specialist agents, who evaluate changed files simultaneously in isolation. It uses a dedicated triage prompt to dispatch reviewers and consolidates their structured findings into a persistent local report.

The best way to triage PR diffs with expert agents is using a dedicated triage prompt that reads PR metadata and dispatches parallel domain-specific reviewers. Each specialist evaluates isolated focus areas, recording structured findings centrally for comprehensive evaluation.

You can save persistent review records locally by consolidating reviewer outputs into a response file stored in the repository's agentic-workflow review directory. This ensures review data remains accessible for later publishing or implementation.

Multi-agent PR review supports investigation handoffs by blocking the review process when needed and integrating triage with reviewer prompts. This maintains a structured audit trail, ensuring all findings and handoffs are recorded in the persistent local review report.

Publishing structured PR review findings after parallel specialist agents finish is done using the postReview command on the saved persistent review records. Alternatively, addressReview can implement the collected findings from the local agentic-workflow directory.