Does it post duplicate comments on GitHub?

No, the skill automatically fetches existing PR comments and deduplicates its findings based on file path, line number, and content to ensure feedback remains fresh and relevant.

How does the multi-agent system work?

The skill spawns three independent Claude agents—a Correctness Expert, a Code Health Expert, and a UX Wizard—to review your PR diff from different technical perspectives simultaneously.

Do I need a separate API key for the sub-agents?

No, sub-agents are spawned via the built-in Task tool in Claude Code and automatically use your existing environment and permissions.

Can it detect security vulnerabilities?

Yes, the 'Correctness Expert' persona specifically audits the diff for security flaws, SQL injection risks, data loss vulnerabilities, and broken control flows.

Why is file ordering randomized for each agent?

Randomizing the file order reduces 'ordering bias,' ensuring that every file in a large PR receives equal attention from at least one reviewer agent.

Multi-Agent PR Reviewer

Name: Multi-Agent PR Reviewer
Author: dyad-sh

bydyad-sh

•

19,778

•

コラボレーションツール

Conducts comprehensive, multi-perspective code reviews using specialized AI sub-agents to identify bugs, maintainability issues, and UX gaps.

The Multi-Agent PR Reviewer skill enhances your development workflow by deploying a trio of independent Claude sub-agents to audit Pull Requests. By assigning distinct personas—Correctness Expert, Code Health Expert, and UX Wizard—and randomizing file ordering to eliminate bias, it provides a deep-dive analysis that exceeds standard linting. It intelligently validates findings through reasoned analysis rather than simple voting, deduplicates issues against existing comments, and automatically posts a structured summary and inline feedback directly to GitHub.

主な機能

01Bias reduction via randomized file traversal for each agent

02Automatic deduplication against existing GitHub PR comments

03Reasoned issue validation to eliminate false positives

04Automated GitHub integration for summaries and inline reviews

05Triple-perspective review (Correctness, Code Health, and UX personas)

0619,778 GitHub stars

ユースケース

01Automating initial code review passes for complex pull requests

02Catching subtle UX and accessibility regressions during development

03Ensuring code maintainability and identifying dead infrastructure

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add dyad-sh/dyad multi-pr-review

For use in Claude.ai and ChatGPT

Download Skill