Three Minds

三个能干活的 AI 分身协作系统。用 Claude Code 启动三个不同人设的 coding agent，共享同一个工作目录，多轮协作直到达成共识。适用于代码审查、方案评审、重构任务等需要多视角协作的场景。

Security Scan

The skill's code and instructions match its stated multi-agent coding purpose, but it forwards the host environment to spawned Claude processes and uses a '--dangerously-skip-permissions' flag, which can expose secrets and allow automated actions beyond what the skill declares.

Purpose & Capability

The name/description promise (three collaborating agents that read/modify files and run tests via Claude Code CLI) matches the included code, configs, and SKILL.md. The skill requires the 'claude' CLI (documented) and operates on a project directory as described.

Instruction Scope

Instructions and code legitimately direct agents to read, modify, and execute within the project directory (expected). However, the runtime uses '--dangerously-skip-permissions' when invoking the Claude CLI and constructs system prompts that fully control child agents. That combination removes interactive permission checks and gives child agents automated ability to run tools and modify files — widening the action scope beyond a cautious/confirming workflow.

Install Mechanism

No external download/install spec is present (instruction-only at registry level). The repository contains a normal package.json with common npm deps (commander, chalk, uuid) and TypeScript sources. There are no remote URL downloads or archive extracts in the codebase.

Credentials

Although the skill declares no required environment variables, runClaude spawns the 'claude' process with env: { ...process.env }, passing the entire host environment to the child. This can leak any secrets present in the environment (API keys, cloud credentials, tokens) to the Claude process and any tools it calls — disproportionate to the declared requirements.

Persistence & Privilege

The skill does not request always:true nor modify other skills or global agent settings. However, it enables fully automated agent actions (no permission prompts) and writes transcripts/modified files into the specified projectDir. Combined with env forwarding and autonomous invocation, this increases the blast radius if the Claude process or its tool integrations are compromised or misused.

Scan Findings in Context

[system-prompt-override] expected: The code explicitly appends system prompts (buildSystemPrompt) and passes them to the Claude CLI. Appending system prompts is expected for shaping agent personas, but 'system-prompt-override' is a prompt-injection pattern worth noting because the skill programmatically controls system prompts for child agents.

What to consider before installing

This skill does what it says (runs three Claude-based agents that can read/modify files), but you should be cautious before running it on sensitive systems: 1) It spawns the 'claude' CLI with the full host environment, which can expose any environment secrets to the child process — do not run it where API keys or cloud credentials are present. 2) It uses a flag to skip permission confirmations, letting agents act automatically; run first in a sandboxed environment (container or disposable VM) and on a git branch. 3) Inspect and, if needed, remove or modify the '--dangerously-skip-permissions' usage and avoid passing process.env wholesale to the child process. 4) Verify the 'claude' CLI you have installed is from a trusted source. 5) If you must use it on real projects, restrict the working directory to a copy of the repo and ensure no secrets are present in files or env. If you want, I can point out the exact lines to change to avoid env forwarding and the skip-permissions flag.

Three Minds - 三个臭皮匠顶个诸葛亮

真正能干活的多 Agent 协作系统。不是纸上谈兵，而是真的能读文件、改代码、跑测试。

核心特性

• 🔧 真正执行 - 每个 agent 都能读取文件、编写代码、运行测试
• 👥 多视角协作 - 三个不同人设的 agent 互相审核
• ✅ 共识机制 - 全员 YES 才结束，确保质量
• 📁 共享目录 - 在同一个项目上协作
• 📝 完整记录 - 自动保存讨论和改动历史

使用方法

# 基础用法
three-minds "审查并改进这个项目的代码质量" --dir ./my-project

# 使用代码审查配置（安全+性能+质量三人组）
three-minds "审查 src/ 目录的所有代码" --config code-review --dir ./project

# 指定最大轮数
three-minds "重构这个模块" --dir ./module --max-rounds 5

# 保存结果到 JSON
three-minds "任务描述" --dir ./project --output result.json

触发词

• "三个臭皮匠"
• "three minds"
• "多 agent 协作"
• "让三个 agent 一起审查"
• "协作完成这个任务"

预设配置

默认配置 - 代码协作三人组

• 🏗️ 架构师 - 关注代码结构、设计模式、可扩展性
• ⚙️ 工程师 - 关注代码质量、错误处理、性能优化
• 🔍 审核员 - 关注代码规范、潜在 bug、文档完整性

code-review - 代码审查三人组

• 🛡️ 安全专家 - 关注安全漏洞、注入风险、权限问题
• ⚡ 性能工程师 - 关注算法复杂度、内存使用、查询优化
• ✅ 质量审查员 - 关注可读性、命名规范、测试覆盖

自定义配置

创建 JSON 配置文件：

{
  "name": "我的三人组",
  "agents": [
    {
      "name": "专家A",
      "emoji": "🎯",
      "persona": "你是一位...关注..."
    },
    {
      "name": "专家B",
      "emoji": "🔬",
      "persona": "你是一位...擅长..."
    },
    {
      "name": "专家C",
      "emoji": "📊",
      "persona": "你是一位...负责..."
    }
  ],
  "maxRounds": 10,
  "projectDir": "."
}

然后：three-minds "任务" --config ./my-config.json

工作流程

┌──────────────────────────────────────────┐
│              第 N 轮                      │
├──────────────────────────────────────────┤
│  🏗️ 架构师                               │
│  → 读取文件，审查结构                     │
│  → 执行必要的重构                         │
│  → 投票 [CONSENSUS: YES/NO]              │
├──────────────────────────────────────────┤
│  ⚙️ 工程师                               │
│  → 审查架构师的改动                       │
│  → 补充实现细节，修复问题                 │
│  → 投票 [CONSENSUS: YES/NO]              │
├──────────────────────────────────────────┤
│  🔍 审核员                               │
│  → 审查所有改动                          │
│  → 检查规范、bug、文档                    │
│  → 投票 [CONSENSUS: YES/NO]              │
└──────────────────────────────────────────┘
          ↓
    全员 YES？→ 结束
          ↓ NO
    继续下一轮...

输出

1. 终端实时输出 - 每个 agent 的工作进展和投票
2. Markdown 记录 - 自动保存在工作目录 three-minds-{timestamp}.md
3. JSON 结果 - 使用 --output 保存完整 session 数据

使用场景

• 代码审查 - 多角度审查 PR 或代码变更
• 重构任务 - 协作完成复杂的代码重构
• 新功能开发 - 从设计到实现的完整协作
• Bug 修复 - 定位问题并验证修复
• 文档完善 - 补充和改进项目文档

注意事项

• 需要安装 Claude Code CLI (claude 命令)
• 每个 agent 会真的修改文件，建议在 git 分支上使用
• 默认最大 15 轮，可通过 --max-rounds 调整
• 如果长时间无法达成共识，检查任务描述是否清晰

依赖

• Node.js 18+
• Claude Code CLI（claude 命令可用）