agency-agents-zh-skill

agency-agents-zh Skill

Skill by ara.so — AI Agent Skills collection.

agency-agents-zh is a Chinese agent library containing 215 plug-and-play AI expert roles, covering 18 departments including engineering, design, marketing, product, games, security, and finance. It supports 17 AI programming tools such as Claude Code, Cursor, Copilot, and OpenClaw. In addition to fully translating the 165 upstream roles, it adds 50 original agents for the Chinese market (Xiaohongshu, Douyin, WeChat, Feishu, DingTalk, cross-border e-commerce, etc.).

Core Features

215 expert roles: each agent has a clear identity, workflow, and deliverables
17 supported tools: Claude Code, Cursor, Copilot, OpenClaw, Gemini CLI, Qwen, Codex, etc.
50 original for the Chinese market: Xiaohongshu operations, Douyin seeding, WeChat Mini Program, Feishu integration, DingTalk development, etc.
18 professional departments: engineering, design, marketing, product, games, security, finance, data, etc.
Automated orchestration: works with Agency Orchestrator to enable multi-agent collaboration

Installation Methods

Method 1: One-click installation script

# 克隆仓库
git clone https://github.com/jnMetaCode/agency-agents-zh.git
cd agency-agents-zh

# 自动检测已安装工具并安装
./scripts/install.sh

# 指定工具安装
./scripts/install.sh --tool openclaw       # OpenClaw（推荐）
./scripts/install.sh --tool claude-code    # Claude Code
./scripts/install.sh --tool cursor         # Cursor
./scripts/install.sh --tool copilot        # GitHub Copilot
./scripts/install.sh --tool gemini-cli     # Gemini CLI
./scripts/install.sh --tool qwen           # Qwen Code
./scripts/install.sh --tool codex          # Codex CLI
./scripts/install.sh --tool aider          # Aider
./scripts/install.sh --tool windsurf       # Windsurf
./scripts/install.sh --tool hermes         # Hermes Agent

Method 2: OpenClaw-specific installation (recommended)

OpenClaw uses a three-file structure:SOUL.md(identity persona) + AGENTS.md(business capabilities) + IDENTITY.md(profile)

# 第一步：转换为 OpenClaw 格式
./scripts/convert.sh --tool openclaw

# 第二步：安装到 OpenClaw 目录
./scripts/install.sh --tool openclaw

# 智能体文件会安装到：~/.openclaw/souls/
# 重启 OpenClaw 网关后生效

Method 3: Manual installation (Claude Code / Copilot)

# Claude Code
cp -r marketing/*.md ~/.claude/agents/
cp -r engineering/*.md ~/.claude/agents/
cp -r design/*.md ~/.claude/agents/

# GitHub Copilot
cp -r marketing/*.md ~/.github/copilot/agents/
cp -r engineering/*.md ~/.github/copilot/agents/

# Cursor
cp -r marketing/*.md ~/.cursor/agents/

Using Agents

Activate in an AI programming tool

After installation, activate via natural language in any AI programming tool:

激活前端开发者模式，帮我构建一个 React 组件

切换到小红书运营专家角色，帮我策划一个产品种草方案

使用安全工程师角色审查这段代码的安全漏洞

View the list of available agents

# 列出所有智能体
ls -R engineering/ design/ marketing/ product/ game/ security/ finance/ data/

# 查看工程类智能体
ls engineering/

Agency Orchestrator Multi-Agent Orchestration

Agency Orchestrator enables multiple agents to collaborate on complex tasks.

Install Agency Orchestrator

npm install -g agency-orchestrator

Basic Usage

# 一句话描述需求，自动选角+编排+执行
ao compose "帮我写一篇关于 AI Agent 的深度分析文章" --run

# 输出：
# 🎭 自动选角 → 叙事学家 + 心理学家 + 内容创作者
# 📊 自动编排 → DAG 工作流，并行执行
# ✅ 自动交付 → 完整成果

Use preset templates

# 查看可用模板
ao templates

# 使用开发类模板
ao run --template web-app-mvp --params "项目名称=电商平台,技术栈=Next.js"

# 使用营销类模板
ao run --template social-media-campaign --params "产品=智能手表,平台=小红书"

YAML workflow configuration

Create workflow.yaml：

name: 小红书产品种草全流程
description: 从调研到投放的完整流程

agents:
  - id: researcher
    role: marketing/marketing-xiaohongshu-operator.md
    task: 分析目标用户画像和竞品内容策略
    
  - id: content_creator
    role: marketing/marketing-content-creator.md
    task: 创作 5 篇种草笔记（3 图文 + 2 视频）
    depends_on: [researcher]
    
  - id: kol_matcher
    role: marketing/marketing-xiaohongshu-operator.md
    task: 推荐 10 位匹配 KOL 及报价预估
    depends_on: [researcher]
    
  - id: ad_planner
    role: marketing/marketing-ad-campaign-manager.md
    task: 制定信息流投放计划和预算分配
    depends_on: [content_creator, kol_matcher]

config:
  llm: deepseek  # 或 claude, gemini, openai
  parallel: true # 无依赖步骤并行执行
  output_dir: ./output

Run the workflow:

ao run workflow.yaml

Resume from checkpoint

# 工作流失败后，单独重跑某个步骤
ao resume workflow.yaml --step content_creator

# 从某个步骤开始继续
ao resume workflow.yaml --from ad_planner

Core Agent Examples

1. Xiaohongshu Operations Expert

File: marketing/marketing-xiaohongshu-operator.md

Use Cases：

切换到小红书运营专家，帮我策划一款智能手表的种草方案

Deliverables：

User persona analysis
Content topic matrix (5W1H)
Note examples (image-text + video scripts)
KOL collaboration list and quotes
Data monitoring dashboard

2. Host Computer Engineer

File: engineering/engineering-pc-host-engineer.md

Use Cases：

激活上位机工程师角色，帮我开发一个工业检测设备的 Qt 界面

Core Capabilities：

Qt/QML UI development
QSerialPort/QModbusTCP communication
QChart real-time curves
SQLite data storage
Multilingual (Chinese and English)

Code Example(auto-generated):

// ModbusManager.h
class ModbusManager : public QObject {
    Q_OBJECT
public:
    explicit ModbusManager(QObject *parent = nullptr);
    bool connectDevice(const QString &ip, int port);
    
signals:
    void dataReceived(quint16 registerAddr, quint16 value);
    
private:
    QModbusTcpClient *modbusClient;
};

3. WeChat Mini Program Developer

File: engineering/engineering-wechat-mini-program-developer.md

Use Cases：

使用微信小程序开发者角色，帮我构建一个电商小程序

Core Capabilities：

WXML/WXSS component-based development
WeChat Pay integration
Cloud development database
Subpackage loading optimization
User authorization flow

Code Example：

// pages/product/product.js
Page({
  data: {
    product: null,
    canvasId: 'shareCanvas'
  },
  
  onLoad(options) {
    this.loadProduct(options.id);
  },
  
  async loadProduct(id) {
    const db = wx.cloud.database();
    const res = await db.collection('products').doc(id).get();
    this.setData({ product: res.data });
  },
  
  async onBuyNow() {
    const res = await wx.cloud.callFunction({
      name: 'createOrder',
      data: { productId: this.data.product._id }
    });
    
    wx.requestPayment({
      ...res.result.payment,
      success: () => wx.navigateTo({ url: '/pages/order/order' })
    });
  }
});

4. Feishu Integration Development Engineer

File: engineering/engineering-feishu-integration-developer.md

Use Cases：

切换到飞书集成开发工程师，帮我开发一个审批流机器人

Core Capabilities：

Feishu Bot Development
Approval Flow API
Multi-dimensional Table Operations
Feishu Document API
Event Subscription

Code Example：

// server.js
const express = require('express');
const app = express();

app.post('/webhook/approval', async (req, res) => {
  const { event } = req.body;
  
  if (event.type === 'approval_instance.approved') {
    const instanceCode = event.instance_code;
    
    // 更新多维表格
    await fetch('https://open.feishu.cn/open-apis/bitable/v1/apps/{app_token}/tables/{table_id}/records', {
      method: 'POST',
      headers: {
        'Authorization': `Bearer ${process.env.FEISHU_ACCESS_TOKEN}`,
        'Content-Type': 'application/json'
      },
      body: JSON.stringify({
        fields: {
          '审批状态': '已通过',
          '审批时间': new Date().toISOString()
        }
      })
    });
    
    // 发送通知
    await sendMessage(event.user_id, '您的审批已通过！');
  }
  
  res.json({ code: 0 });
});

5. Frontend Developer

Files: engineering/engineering-frontend-developer.md

Use Cases：

激活前端开发者模式，帮我构建一个现代化的 React Dashboard

Core Capabilities：

React/Vue Component Development
Responsive Design
Performance Optimization
State Management (Redux/Zustand)
Type Safety (TypeScript)

Code Example：

// components/Dashboard.tsx
import { useEffect, useState } from 'react';
import { Card, Chart } from '@/components/ui';

interface DashboardData {
  users: number;
  revenue: number;
  trend: Array<{ date: string; value: number }>;
}

export function Dashboard() {
  const [data, setData] = useState<DashboardData | null>(null);
  
  useEffect(() => {
    fetch('/api/dashboard')
      .then(res => res.json())
      .then(setData);
  }, []);
  
  if (!data) return <div>加载中...</div>;
  
  return (
    <div className="grid grid-cols-3 gap-4">
      <Card title="总用户数">
        <span className="text-3xl font-bold">{data.users}</span>
      </Card>
      <Card title="总收入">
        <span className="text-3xl font-bold">¥{data.revenue}</span>
      </Card>
      <Card title="增长趋势" className="col-span-3">
        <Chart data={data.trend} type="line" />
      </Card>
    </div>
  );
}

Agent Department Categories

Engineering

33 agents, covering:

Frontend/Backend/Full-stack Development
AI/ML Engineering
DevOps/SRE
Security Engineering
Embedded/IoT
HMI/Industrial Control
WeChat/Feishu/DingTalk Integration

Design

18 agents, covering:

UI/UX Design
Brand Design
Visual Storytelling
Accessibility Design
AI Image Generation

Marketing

40 agents, covering:

Xiaohongshu/Douyin/Bilibili Operations
SEO/SEM
Content Creation
Social Media
Cross-border E-commerce
Ad Placement

Product

22 agents, covering:

Product Management
Requirements Analysis
User Research
Data-driven Decision Making
A/B Testing

Other Departments

Game: 15
Security: 12
Finance: 10
Data: 18
Operations: 14
Support: 8
Legal: 6
HR (Human Resources): 9

Configuration and Environment Variables

Agency Orchestrator Configuration

Create ~/.agency-orchestrator/config.yaml：

# LLM 配置
llm:
  provider: deepseek  # deepseek, claude, openai, gemini
  api_key_env: DEEPSEEK_API_KEY  # 环境变量名称
  model: deepseek-chat
  temperature: 0.7
  max_tokens: 4000

# 智能体路径
agents:
  base_dir: ~/.openclaw/souls  # 或 ~/.claude/agents
  auto_discover: true

# 输出配置
output:
  default_dir: ./agency-output
  format: markdown  # markdown, json, html
  include_metadata: true

# 并行执行配置
execution:
  max_parallel: 3
  timeout: 300  # 秒
  retry: 2

Environment Variable Settings

# .env 文件
DEEPSEEK_API_KEY=your_deepseek_api_key
CLAUDE_API_KEY=your_claude_api_key
OPENAI_API_KEY=your_openai_api_key
GEMINI_API_KEY=your_gemini_api_key

# 飞书/钉钉集成
FEISHU_APP_ID=your_feishu_app_id
FEISHU_APP_SECRET=your_feishu_app_secret
DINGTALK_APP_KEY=your_dingtalk_app_key
DINGTALK_APP_SECRET=your_dingtalk_app_secret

# 微信小程序
WECHAT_APP_ID=your_wechat_app_id
WECHAT_APP_SECRET=your_wechat_app_secret

Common Usage Patterns

Pattern 1: Single Agent Consultation

切换到安全工程师角色
请审查以下代码的安全问题：
[粘贴代码]

Pattern 2: Agent Switching

激活前端开发者
[完成 UI 开发后]
现在切换到后端架构师，设计对应的 API

Pattern 3: Multi-Agent Collaboration (via Agency Orchestrator)

# 一句话描述复杂需求
ao compose "开发一个小红书运营工具，包含内容创作、数据分析、KOL 推荐功能" --run

# 自动选角：
# - 小红书运营专家（需求分析）
# - 前端开发者（UI 实现）
# - 后端架构师（API 设计）
# - 数据工程师（数据管线）

Pattern 4: Using Preset Templates

# 电商小程序全栈开发
ao run --template wechat-mini-program-ecommerce

# 工业上位机开发
ao run --template industrial-hmi-qt

# 小红书营销全流程
ao run --template xiaohongshu-full-campaign

Troubleshooting

Issue 1: Agent Not Loaded

Symptoms: Unable to activate the agent in the AI tool

Solution：

# 检查安装路径
ls ~/.claude/agents/
ls ~/.openclaw/souls/

# 重新安装
./scripts/install.sh --tool claude-code --force

# 重启 AI 工具

Issue 2: OpenClaw Format Conversion Failed

Symptoms：convert.sh error

Solution：

# 确保有执行权限
chmod +x scripts/convert.sh

# 手动指定输出目录
./scripts/convert.sh --tool openclaw --output ./converted

# 检查依赖
which python3  # 需要 Python 3.7+

Issue 3: Agency Orchestrator Cannot Connect to LLM

Symptoms：ao compose reports an API error

Solution：

# 检查环境变量
echo $DEEPSEEK_API_KEY

# 测试连接
ao test-connection --provider deepseek

# 切换到免费 LLM
ao compose "任务描述" --llm gemini-cli --run

Issue 4: Chinese Characters Garbled

Symptoms: Agent output is garbled

Solution：

# 设置终端编码
export LANG=zh_CN.UTF-8
export LC_ALL=zh_CN.UTF-8

# 检查文件编码
file -i marketing/marketing-xiaohongshu-operator.md
# 应显示：charset=utf-8

Issue 5: Workflow Execution Timeout

Symptoms: Agency Orchestrator task timed out

Solution：

# 在 workflow.yaml 中增加超时时间
config:
  timeout: 600  # 从 300 秒增加到 600 秒
  
# 或减少并行度
config:
  max_parallel: 1  # 串行执行，避免 API 限流

Advanced Tips

Custom Agents

Create custom/my-agent.md：

---
name: 我的自定义专家
role: 特定领域专家
department: 自定义部门
---

# 角色定义

你是一位...

## 核心能力
1. ...
2. ...

## 工作流程
- 步骤 1：...
- 步骤 2：...

## 交付物
- [ ] ...

Install:

cp custom/my-agent.md ~/.claude/agents/

Combined Use Agents

# combo-workflow.yaml
name: 全栈开发+运营组合
agents:
  - role: engineering/engineering-frontend-developer.md
  - role: engineering/engineering-backend-architect.md
  - role: marketing/marketing-xiaohongshu-operator.md
  - role: design/design-ui-designer.md

Batch Conversion Agents

# 转换所有智能体为 Cursor 格式
for dept in engineering design marketing product; do
  ./scripts/convert.sh --tool cursor --input $dept --output ./cursor-agents/$dept
done

Resource Links

GitHub Repository: https://github.com/jnMetaCode/agency-agents-zh
Agency Orchestrator: https://github.com/jnMetaCode/agency-orchestrator
Upstream English Version: https://github.com/msitarzewski/agency-agents
Companion Book: https://book.aibuzhiyu.com/（AI Programming in Practice Methodology)

Complete List of Supported AI Tools

OpenClaw ⭐ Recommended (Multi-Agent Collaboration)
Claude Code(Anthropic Official)
Cursor(Intelligent IDE)
GitHub Copilot
Gemini CLI（Google）
Qwen Code(Alibaba Cloud)
Codex CLI（OpenAI）
Aider(Open-Source AI Programming Assistant)
Windsurf
Antigravity
Kiro（Amazon）
Trae
OpenCode
DeerFlow 2.0(ByteDance)
WorkBuddy(Tencent)
Hermes Agent（NousResearch）
Qoder

Quick Start: Clone the repository → Run ./scripts/install.sh → Activate the agents in your AI tool → Start using 215 expert roles!

About this skill

Core Features

Installation Methods

Method 1: One-click installation script

Method 2: OpenClaw-specific installation (recommended)

Method 3: Manual installation (Claude Code / Copilot)

Using Agents

Activate in an AI programming tool

View the list of available agents

Agency Orchestrator Multi-Agent Orchestration

Install Agency Orchestrator

Basic Usage

Use preset templates

YAML workflow configuration

Resume from checkpoint

Core Agent Examples

1. Xiaohongshu Operations Expert

2. Host Computer Engineer

3. WeChat Mini Program Developer

4. Feishu Integration Development Engineer

5. Frontend Developer

Agent Department Categories

Engineering

Design

Marketing

Product

Other Departments

Configuration and Environment Variables

Agency Orchestrator Configuration

Environment Variable Settings

Common Usage Patterns

Pattern 1: Single Agent Consultation

Pattern 2: Agent Switching

Pattern 3: Multi-Agent Collaboration (via Agency Orchestrator)

Pattern 4: Using Preset Templates

Troubleshooting

Issue 1: Agent Not Loaded

Issue 2: OpenClaw Format Conversion Failed

Issue 3: Agency Orchestrator Cannot Connect to LLM

Issue 4: Chinese Characters Garbled

Issue 5: Workflow Execution Timeout

Advanced Tips

Custom Agents

Combined Use Agents

Batch Conversion Agents

Resource Links

Complete List of Supported AI Tools