three-minds

What This Skill Does

Three Minds (three-minds) is an advanced multi-agent collaboration system designed for software development workflows. By leveraging the Claude Code agent, this skill orchestrates three distinct AI agents with unique personas to perform tasks within a shared project directory. Unlike static conversational agents, these agents are capable of reading files, modifying code, and executing tests. The system operates on a consensus-based protocol, requiring all agents to agree before finalizing a task. This ensures high-quality outcomes through multi-perspective evaluation—typically involving a Architect, an Engineer, and an Auditor. It is the ultimate tool for complex code reviews, system refactoring, and comprehensive debugging, mimicking a professional software engineering team within your terminal.

Installation

To install the Three Minds skill, ensure you have Node.js 18+ and the Claude Code CLI (the claude command) properly configured on your machine. Run the following command in your terminal:

clawhub install openclaw/skills/skills/enderfga/three-minds

Once installed, verify that the three-minds command is globally accessible in your environment.

Use Cases

• Deep Code Review: Have an Architect, Engineer, and Auditor review your Pull Request or existing codebase to catch design flaws, performance bottlenecks, and security vulnerabilities simultaneously.
• Complex Refactoring: Use the system to plan and execute code migrations or dependency updates where multi-stage verification is necessary to prevent breaking changes.
• Bug Hunting: Deploy the trio to identify the root cause of elusive bugs, fix them, and write verifying tests.
• Documentation Generation: Task the agents with updating project documentation while ensuring the code remains consistent with the new specs.

Example Prompts

1. "three-minds 'Perform a deep security and performance review on the authentication module located in src/auth' --dir ./project"
2. "三个臭皮匠，协作完成这个复杂的重构任务，将全局变量迁移到 Context API，确保测试通过 --max-rounds 5"
3. "让三个 agent 一起审查并优化以下目录的代码逻辑，重点关注内存泄漏问题 --dir ./src/services"

Tips & Limitations

• Version Control: Because this skill actively modifies your project files, always run it within a clean git branch so you can revert changes if the consensus cycle produces unwanted outcomes.
• Consensus Loop: The agents can run up to 15 rounds by default. If they fail to reach a consensus, review your prompt for clarity or increase the --max-rounds parameter.
• Cost Awareness: Since each agent is a separate Claude Code instance performing operations, monitor your API usage during extended collaboration sessions.
• Customization: You can define your own expert team by creating a JSON configuration file, allowing for domain-specific personas like 'DevOps Specialist' or 'UX Designer' to be integrated into your workflow.