How does the 3-round protocol work?

The protocol starts with independent opinions from all models, followed by a cross-review where models critique each others' findings, and concludes with Claude synthesizing a final report as the council chair.

Is the Multi-Agent Council compatible with Windows?

Yes, it offers cross-platform support for both macOS and Windows, handling path normalization and CLI transport differences internally.

Does it send sensitive data to external models?

The skill includes security considerations that exclude credentials, tokens, and certificates from the context pack by default to ensure privacy during external LLM calls.

Which external LLMs are supported by the council?

The council currently supports Claude (as the chair), OpenAI, and Gemini, which can be connected via MCP (Model Context Protocol) or CLI adapters.

Can this skill help with security audits?

Yes, the Review Mode is specifically designed to analyze the project state for security risks, defects, and operational vulnerabilities from multiple perspectives.

Multi-Agent Council

Name: Multi-Agent Council
Author: hideki5123

byhideki5123

0•

コラボレーションツール

Orchestrates a multi-LLM consensus protocol using Claude, OpenAI, and Gemini to validate architectural decisions and perform deep code reviews.

The Multi-Agent Council skill empowers developers to bypass the biases of a single AI model by convening a panel of expert LLMs to analyze complex technical challenges. By utilizing a structured three-round protocol—comprising independent opinions, cross-model critique, and a final synthesis by Claude—this skill provides rigorous validation for system design, tech stack selection, and comprehensive project state audits. It is particularly valuable for high-stakes decision-making where diverse perspectives and cross-verification are essential for mitigating technical debt and identifying hidden risks.

主な機能

010 GitHub stars

02Multi-provider integration supporting Claude, OpenAI, and Gemini

03Structured 3-round consensus protocol (Independent, Cross-Review, Synthesis)

04Review Mode for auditing project state and identifying hidden risks

05Architecture Mode for validating design patterns and technology choices

06Automated context packing with transparency reporting and security filtering

ユースケース

01Conducting comprehensive security and performance audits of a codebase

02Evaluating trade-offs between different databases, frameworks, or cloud providers

03Validating complex refactoring plans through diverse AI perspectives

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add hideki5123/multi-agent-council multi-agent-council

For use in Claude.ai and ChatGPT

Download Skill