Description

基于 Claude Code 架构增强 OpenClaw，实现多 Agent 协作、工具权限控制、结构化记忆和任务工作流管理能力。

README (SKILL.md)

🚀 Claude Code Enhancement

Name: Claude Code Enhancement
Author: ntaffffff

\x3Cp align="center"> \x3Cimg src="https://img.shields.io/badge/OpenClaw-增强组件-blue?style=for-the-badge&logo=rocket" alt="OpenClaw"> \x3Cimg src="https://img.shields.io/badge/版本-1.0.0-green?style=for-the-badge" alt="Version"> \x3Cimg src="https://img.shields.io/badge/Python-3.12+-yellow?style=for-the-badge" alt="Python"> \x3Cimg src="https://img.shields.io/badge/许可证-MIT-orange?style=for-the-badge" alt="License"> \x3C/p>

[!TIP] 借鉴 Claude Code 顶级架构设计，为 OpenClaw 打造企业级多 Agent 协作系统

✨ 为什么选择它？

\x3Cp align="center"> \x3Cimg src="https://streak-stats.demolab.com?user=ntaffffff&theme=dark&hide_border=true" width="400"> \x3C/p>

传统的 AI Agent 助手只能单打独斗，而这个技能让你拥有一个指挥官 + N 个执行者的超级团队！

传统模式	Claude Code Enhancement
❌ 单线程执行	✅ 多 Agent 并行处理
❌ 工具权限混乱	✅ 完整权限控制系统
❌ 记忆碎片化	✅ 结构化记忆体系
❌ 任务难以追踪	✅ 全链路进度监控

🎯 核心功能

1. 🧠 Coordinator - 多 Agent 协调器

┌─────────────────────────────────────────────────────────────┐
│                        Coordinator                          │
│  🧠 主 Agent (指挥官)                                       │
│     ├── 分析任务                                            │
│     ├── 拆分任务                                            │
│     └── 汇总结果                                            │
│                                                              │
│   ┌──────────┐   ┌──────────┐   ┌──────────┐              │
│   │ Worker 1 │   │ Worker 2 │   │ Worker 3 │              │
│   │  🔨 编码  │   │  🔍 测试  │   │  📝 文档  │              │
│   └──────────┘   └──────────┘   └──────────┘              │
│        ↓              ↓              ↓                     │
│   ─────────────────────────────────────────────            │
│                    结果聚合返回                              │
└─────────────────────────────────────────────────────────────┘

一部手机同时让 5 个人帮你干活！

# 启动协调者模式
/coord start

# 派生出 3 个 Worker 同时工作
/coord spawn "后端 API" "实现用户注册接口，用 FastAPI"
/coord spawn "前端组件" "实现注册页面，用 React"
/coord spawn "单元测试" "为注册功能编写测试用例"

# 查看所有 Worker 状态
/coord status

2. 🛡️ Tool Permission - 工具权限系统

\x3Cp align="center"> \x3Cimg src="https://img.shields.io/badge/安全-企业级-red?style=for-the-badge" alt="Security"> \x3C/p>

再也不用担心 AI 误删重要文件了！

# 配置文件示例
permission:
  default_mode: "ask"  # 默认每次询问
  
  always_allow:        # 始终允许（安全工具）
    - GlobTool
    - GrepTool
    - FileReadTool
  
  always_deny:         # 始终拒绝（危险操作）
    - FormatDiskTool
    - DeleteProduction
  
  always_ask:          # 始终询问（修改类）
    - BashTool
    - FileWriteTool
    - FileEditTool

工具类型	示例	默认行为
🔍 只读	`GlobTool`, `GrepTool`	自动允许
⚠️ 修改	`FileWriteTool`, `BashTool`	询问确认
☠️ 危险	`FormatDiskTool`	始终拒绝

3. 💾 Memory System - 结构化记忆

\x3Cp align="center"> \x3Cimg src="https://img.shields.io/badge/记忆-永久保存-blue?style=for-the-badge" alt="Memory"> \x3C/p>

再也不用每次都重复告诉 AI 你的偏好！

memory/
├── MEMORY.md           ← 入口文件（每次对话自动加载）
├── user/
│   └── preferences.md  ← 你的个人偏好
├── feedback/
│   └── corrections.md  ← 你的每次纠正
├── project/
│   └── context.md      ← 项目上下文
└── reference/
    └── docs.md         ← 参考资料

# 记住我的偏好
/memory add user "我喜欢用 TypeScript + React"

/memory add feedback "上次那个函数写错了，应该是这样..."
/memory search "TypeScript"

4. 📊 Task Workflow - 任务工作流

\x3Cp align="center"> \x3Cimg src="https://img.shields.io/badge/进度-可视化-green?style=for-the-badge" alt="Progress"> \x3C/p>

像项目管理工具一样追踪 AI 任务！

任务: 用户头像上传功能
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📋 阶段 1: 分析 ████████████ 100%
🎯 阶段 2: 规划 ████████░░░░ 80%
▶️ 阶段 3: 执行 ████░░░░░░░░ 40%
⏳ 阶段 4: 验证 ░░░░░░░░░░░ 0%
⏳ 阶段 5: 总结 ░░░░░░░░░░░ 0%
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

/task create "实现用户头像上传"
/task status      # 查看当前状态
/task progress    # 查看详细进度
/task complete    # 标记完成

5. 🤖 Enhanced Agent - 增强版 Agent

Agent 类型	描述	超能力
🧑‍💻 `coder`	编码专家	写代码、debug、重构
🔍 `reviewer`	审查专家	代码审查、安全扫描
📚 `researcher`	研究专家	信息收集、分析整理
👷 `worker`	多面手	通用任务执行
🧠 `general`	全能型	日常对话任务

# 调用专门的 Coding Agent
agent = await spawn_agent(
    type="coder",
    description="修复登录 bug",
    prompt="用户反馈登录后 token 过期时间不对，请检查并修复",
    isolation="worktree"  # 隔离环境，安全！
)

🚀 快速开始

安装

# 方式一：ClawHub 安装（推荐）
clawhub install claude-code-enhancement

# 方式二：GitHub 手动安装
git clone https://github.com/ntaffffff/openclaw-claude-code-enhancement.git
cd openclaw-claude-code-enhancement

使用

# 启动多 Agent 协作
/coord start

# 添加记忆
/memory add user "我喜欢用 Python + FastAPI"

# 创建任务
/task create "实现支付功能"

📖 文档导航

文档	说明
功能详解	所有功能的详细介绍
配置指南	如何自定义配置
API 参考	开发者 API
常见问题	常见问题解答

🆚 对比其他方案

特性	本技能	传统单 Agent	其他多 Agent 方案
多 Worker 并行	✅	❌	✅
工具权限控制	✅ 企业级	❌	⚠️ 基础
结构化记忆	✅	❌	❌
任务进度追踪	✅	❌	⚠️ 基础
隔离执行模式	✅ Git Worktree	❌	⚠️ 有限
零侵入	✅	-	❌

🧪 示例展示

示例 1：同时实现一个完整功能

用户: 帮我实现一个用户认证系统

> /coord start

[Coordinator] 分析任务中...
[Coordinator] 任务已拆分为 4 个子任务

> /coord spawn "用户模型" "实现 User 模型，包含 id, email, password, createdAt"
> /coord spawn "注册 API" "实现 POST /register 接口"
> /coord spawn "登录 API" "实现 POST /login 接口，返回 JWT"
> /coord spawn "中间件" "实现 JWT 验证中间件"

[Worker 1] ✅ 完成
[Worker 2] ✅ 完成
[Worker 3] ✅ 完成  
[Worker 4] ✅ 完成

[Coordinator] 结果聚合中...
[Coordinator] ✅ 任务完成！

示例 2：记住你的偏好

用户: 记住我喜欢用 Vitest 做测试

> /memory add user "测试框架: Vitest"
> /memory add user "代码风格: 函数式编程"
> /memory add user "喜欢用 async/await"

[Memory] ✅ 已记住！

用户: 帮我写个工具函数

[Agent] 好的！我会用 Vitest + 函数式风格来写～

🤝 贡献者

\x3Cp align="center"> \x3Cimg src="https://contributors-img.web.app/image?repo=ntaffffff/openclaw-claude-code-enhancement" width="400"> \x3C/p>

📝 更新日志

v1.0.0 (2026-04-03)

🎉 初始版本发布
✅ Coordinator 多 Agent 协调
✅ Tool Permission 权限系统
✅ Memory System 记忆系统
✅ Task Workflow 任务工作流
✅ Enhanced Agent 增强代理

📄 许可证

\x3Cp align="center"> \x3Cimg src="https://img.shields.io/badge/许可证-MIT-blue?style=for-the-badge" alt="License"> \x3C/p>

MIT License - 自由使用，保留署名即可。

\x3Cp align="center"> \x3Cstrong>⭐ 如果这个项目对你有帮助，请点个 Star！\x3C/strong> \x3C/p>

\x3Cp align="center"> \x3Ca href="https://github.com/ntaffffff/openclaw-claude-code-enhancement"> \x3Cimg src="https://img.shields.io/github/stars/ntaffffff/openclaw-claude-code-enhancement?style=social" alt="Star"> \x3C/a> \x3Ca href="https://github.com/ntaffffff/openclaw-claude-code-enhancement/fork"> \x3Cimg src="https://img.shields.io/github/forks/ntaffffff/openclaw-claude-code-enhancement?style=social" alt="Fork"> \x3C/a> \x3C/p>

\x3Cp align="center"> \x3Csub>Made with ❤️ by dxx\x3C/sub> \x3C/p>

Usage Guidance

What to consider before installing/running: 1) Don't blindly run 'git clone' or any installer pointed at an unknown GitHub repo — prefer reviewing the bundled code first. 2) Inspect SKILL.md and README in a raw/text editor for invisible/unicode control characters (they can hide malicious prompts). 3) Run the code in an isolated/sandbox environment (container or VM) and with a throwaway OpenClaw workspace to observe behavior and file writes. 4) Review and lock down the permission configuration (avoid enabling BYPASS mode). 5) Back up any existing ~/.openclaw workspace first; the memory system will create and delete files there. 6) Be aware nlp_parser.py contains a probable bug/truncation which may cause exceptions or unexpected fallbacks — consider running static checks and tests before using with real data. 7) If you need network installs, prefer known release sources (official GitHub releases or vetted packages) and verify their integrity.

Capability Analysis

Type: OpenClaw Skill Name: claude-code-enhancement Version: 1.0.1 The skill bundle provides a comprehensive framework for multi-agent coordination, structured memory management, and a tool permission system inspired by Claude Code. The Python modules (e.g., coordinator.py, permission.py, memory.py) are well-documented and align with the stated purpose of enhancing OpenClaw's capabilities. There is no evidence of data exfiltration, obfuscation, or malicious execution; notably, the permission system is designed to increase security by requiring user confirmation for high-risk tools like BashTool and FileWriteTool.

Capability Assessment

✓ Purpose & Capability

Name/description map to the included modules (coordinator, agent, permission, memory, workflow). No unrelated environment variables or external credentials are requested; the code's filesystem usage (under ~/.openclaw/workspace) and CLI commands align with the stated multi‑agent and memory features.

⚠ Instruction Scope

SKILL.md contains full runtime instructions that tell the agent to create persistent memory files, spawn agents, and suggests cloning from a GitHub repo. A pre-scan detected unicode control characters in SKILL.md (prompt‑injection pattern). The instructions also reference external resources (GitHub, remote images) which could cause network fetches if followed. The included nlp_parser.py appears truncated/buggy (a call to get_enhancemen is incomplete), so the runtime behavior may be unpredictable.

ℹ Install Mechanism

No formal install spec is included (instruction-only install), which reduces automated install risk. However the README/SKILL.md recommend 'clawhub install' or 'git clone' from an external GitHub repo (unknown owner). If a user follows the git clone path they will fetch code from an external source — that is an installation decision the user must make and carries extra risk.

ℹ Credentials

The skill requests no credentials or env vars, which is appropriate. It does create and modify files under the user's home (~/.openclaw/workspace/memory and skill config paths) and can delete memory files via clear(). This filesystem access is consistent with a memory/agent system but means it will persist data locally — review what is saved and avoid storing secrets there.

✓ Persistence & Privilege

The skill is not marked always:true and uses the normal autonomous-invocation default. It creates persistent memory files under the user's workspace (expected for its purpose). Note the PermissionSystem supports a BYPASS mode that would allow executing without confirmations — that is a configurable, high‑privilege mode and should be avoided unless you trust the code.

Version History

v1.0.1

- Added a new README.md with user-facing documentation. - Completely rewrote SKILL.md with a more visual, user-friendly style and enhanced explanations. - No functional code changes; this update focuses on improved documentation, examples, and feature highlights. - Clearer guidance for installation, usage, configuration, and comparison with alternatives. - Added quickstart, visual diagrams, shields, and contributor sections to increase accessibility for all users.

v1.0.0

- Initial release of Claude Code Enhancement Skill for OpenClaw. - Adds support for advanced multi-Agent collaboration with a Coordinator mode. - Integrates enhanced Agent tools with new types, isolation, and async/sync execution. - Introduces a comprehensive tool permission system for safe and flexible tool usage. - Implements a structured memory system for user, project, and feedback data retention. - Provides a modular, non-intrusive design fully compatible with existing OpenClaw features.

Metadata

Slug claude-code-enhancement

Version 1.0.1

License MIT-0

All-time Installs 0

Active Installs 0

Total Versions 2

Frequently Asked Questions

What is Claude Code Enhancement?

基于 Claude Code 架构增强 OpenClaw，实现多 Agent 协作、工具权限控制、结构化记忆和任务工作流管理能力。 It is an AI Agent Skill for Claude Code / OpenClaw, with 82 downloads so far.

How do I install Claude Code Enhancement?

Run "/install claude-code-enhancement" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Claude Code Enhancement free?

Yes, Claude Code Enhancement is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Claude Code Enhancement support?

Claude Code Enhancement is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Claude Code Enhancement?

It is built and maintained by ntaffffff (@ntaffffff); the current version is v1.0.1.

More Skills

Claude Code Enhancement