How does the skill handle different PR sizes?

The skill classifies changes into 'Risk Lanes' (L0 to L2). Routine changes get a lightweight review, while significant features or architectural shifts trigger full cross-review protocols.

What is a multi-agent code review?

It is a process where multiple specialized AI agents, such as security, performance, and architecture experts, review code in parallel and then challenge each other's findings to ensure high rigor and accuracy.

Does this skill provide a final merge verdict?

Yes, it provides a binary 'APPROVE' or 'REQUEST CHANGES' verdict based on the presence of critical blocker items identified during the multi-phase review.

Can I use this for GitHub Pull Requests?

Yes, the skill integrates with the GitHub CLI to ingest changes via PR numbers, as well as supporting git branches, commit SHAs, or local staged files.

What is Lead-Mediated Cross-Review?

In Phase 2, the team lead (Claude) routes critical findings between different specialist agents to verify exploitability, confirm logic errors, and resolve contradictions before finalizing the report.

Multi-Agent Code Review

Name: Multi-Agent Code Review
Author: abatilo

byabatilo

•

Collaboration Tools

Orchestrates a comprehensive, three-phase parallel code review using a specialized team of AI agents for rigorous quality assurance.

The Multi-Agent Code Review skill transforms Claude into a lead architect who manages a team of specialist agents—including experts in security, architecture, performance, and correctness—to perform deep-dive code reviews. By implementing a structured workflow consisting of specialist analysis, lead-mediated cross-review, and priority-based synthesis, it identifies structural risks, logic errors, and security vulnerabilities that standard linters miss. The skill dynamically scales its rigor based on the change's risk profile, ensuring that critical infrastructure receives maximum scrutiny while routine updates remain streamlined and efficient.

Key Features

01Multi-specialist agent team covering Security, Architecture, Performance, and Governance

02Structured self-critique protocol to harden findings and reduce false positives

033 GitHub stars

04Three-phase workflow including mandatory cross-agent challenges for high-risk changes

05Binary 'Approve/Request Changes' verdict with prioritized, actionable remediation steps

06Risk-lane classification (L0-L2) to automatically match review depth to code impact

Use Cases

01Supplementing human peer reviews with an AI 'red team' focused on edge cases and logic flaws

02Conducting high-stakes reviews on architectural changes, security-sensitive modules, or public APIs

03Automating complex PR reviews in large codebases to maintain consistent quality and naming standards

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add abatilo/vimrc code-review

For use in Claude.ai and ChatGPT

Download Skill