What is the reporting threshold for review findings?

To ensure relevance, only findings that achieve a weighted synthesis score of 80 or higher are included in the final report.

How does this skill handle code review noise?

It utilizes a strict false positive taxonomy to automatically reject findings that are pre-existing, already detectable by linters, or fall outside the project's specific CLAUDE.md guidelines.

Can I invoke this skill manually in the terminal?

No, this is a non-invocable reference skill designed to provide background context for other subagents and automated workflows within the Claude Code environment.

What is the primary purpose of the Council Reference skill?

It provides the background knowledge, structured data formats, and evaluation logic used by multi-agent 'Council' systems to conduct consistent, high-quality code reviews.

How does the expertise weight matrix work?

It assigns specific coefficients to different AI models (like Gemini or Codex) based on their domain-specific strengths in areas such as security, architecture, or debugging.

Council Review Reference

Name: Council Review Reference
Author: rube-de

byrube-de

•

コラボレーションツール

Establishes standardized evaluation frameworks and multi-model expertise weights for multi-agent code review systems.

The Council Reference skill provides the foundational logic and reference data required for high-precision, multi-agent code reviews within Claude Code. It defines a mandatory JSON response structure for subagents, a detailed expertise weight matrix for various LLMs (including Gemini, Codex, and Qwen), and a rigorous scoring threshold to ensure only high-quality findings are surfaced. By implementing a strict false positive taxonomy and a weighted synthesis formula, this skill enables Claude to act as a sophisticated conductor for automated quality assurance and security audits.

主な機能

01Comprehensive expertise weight matrix across specialized tasks like Security and Architecture

025 GitHub stars

03Rigorous false positive taxonomy to filter linter noise and pedantic nits

04Mandatory location-tagging requirements for precise code finding identification

05Weighted synthesis formula for aggregating opinions from multiple AI agents

06Standardized JSON response schema for consistent multi-model outputs

ユースケース

01Standardizing output from disparate LLMs during deep security audits

02Filtering low-confidence AI suggestions to minimize developer fatigue and noise

03Automating high-precision multi-agent PR reviews to reduce human overhead

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add rube-de/cc-skills council-reference

For use in Claude.ai and ChatGPT

Download Skill