multi-agent-codereview

Multi-agent debate improves code review by assigning independent expert roles like backend architect and security to analyze a codebase, then enforcing cross-agent evaluation to surface defects and reach consensus. This workflow ensures robust pull request reviews through structured conflicting viewpoints.

You should use a multi-agent code review workflow for large or safety-critical codebases, architectural reviews, and service-layer assessments. It is specifically designed to evaluate complex changes across multiple viewpoints like tech, operations, and business where standard reviews might miss deep structural defects.

To start a multi-agent code review, trigger the skill with your target codebase path. The workflow then executes a staged process: code analysis, role assignment, self-configuration, independent reviews, two-round debates, and a final synthesis report with issue prioritization.

Yes, you can use this workflow for architectural reviews and service-layer assessments. It assigns independent expert roles to analyze the codebase and enforces cross-agent debate to evaluate practical risks and structural concerns across technical, operational, and business viewpoints.

The Devil's Advocate role in automated code review acts as a cross-entity evaluator during the two-round debate phase. It collaboratively challenges the independent expert agents to surface hidden defects, architectural concerns, and practical risks before the final synthesis.

You should consider that this multi-agent code review requires a frontmatter-defined name and description, along with optional references or scripts. Because it enforces independent reviews and two-round debates, the workflow is resource-intensive and best suited for large or safety-critical codebases.