
agency-agents-zh-skill
使用 agency-agents-zh 中文 AI 智能体角色库,为 AI 编程工具提供 215 个即插即用的专家角色
- Installs
- 323
- Last 24h
- —
- Trending rank
- —
- Security
- Not audited
Install
Works with every agent the skills CLI supports.
npx skills add https://github.com/reason-machines/ai-agent-skills --skill agency-agents-zh-skill- 01
Install
Run the command in your project. The CLI asks which agents to install for.
- 02
Describe the task
Ask your agent for something this skill covers. It loads agency-agents-zh-skill automatically when relevant.
- 03
Review the result
The agent follows the SKILL.md steps, scripts and references bundled with the skill.
SKILL.md
The instructions your agent reads when the skill is triggered.
agency-agents-zh Skill
Skill by ara.so — AI Agent Skills collection.
agency-agents-zh 是一个包含 215 个即插即用 AI 专家角色的中文智能体库,覆盖工程、设计、营销、产品、游戏、安全、金融等 18 个部门。支持 Claude Code、Cursor、Copilot、OpenClaw 等 17 种 AI 编程工具。除了完整翻译上游 165 个角色外,新增了 50 个中国市场原创智能体(小红书、抖音、微信、飞书、钉钉、跨境电商等)。
核心特性
- 215 个专家角色:每个智能体有明确的身份、工作流程和交付物
- 17 种工具支持:Claude Code、Cursor、Copilot、OpenClaw、Gemini CLI、Qwen、Codex 等
- 50 个中国市场原创:小红书运营、抖音种草、微信小程序、飞书集成、钉钉开发等
- 18 个专业部门:工程、设计、营销、产品、游戏、安全、金融、数据等
- 自动化编排:配合 Agency Orchestrator 实现多智能体协作
安装方法
方式一:一键安装脚本
# 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
方式二:OpenClaw 专用安装(推荐)
OpenClaw 使用三文件结构:SOUL.md(身份人设)+ AGENTS.md(业务能力)+ IDENTITY.md(简介)
# 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
方式三:手动安装(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/
使用智能体
在 AI 编程工具中激活
安装后,在任意 AI 编程工具中通过自然语言激活:
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
查看可用智能体列表
# 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 多智能体编排
Agency Orchestrator 让多个智能体协作完成复杂任务。
安装 Agency Orchestrator
npm install -g agency-orchestrator
基本使用
# 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
使用预设模板
# 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 工作流配置
创建 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
运行工作流:
ao run workflow.yaml
断点续跑
# 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
核心智能体示例
1. 小红书运营专家
文件: marketing/marketing-xiaohongshu-operator.md
使用场景:
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
交付物:
- 用户画像分析
- 内容选题矩阵(5W1H)
- 笔记范例(图文+视频脚本)
- KOL 合作清单及报价
- 数据监控仪表盘
2. 上位机工程师
文件: engineering/engineering-pc-host-engineer.md
使用场景:
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
核心能力:
- Qt/QML UI 开发
- QSerialPort/QModbusTCP 通信
- QChart 实时曲线
- SQLite 数据存储
- 多语言(中英文)
代码示例(自动生成):
// 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. 微信小程序开发者
文件: engineering/engineering-wechat-mini-program-developer.md
使用场景:
使用微信小程序开发者角色,帮我构建一个电商小程序
核心能力:
- WXML/WXSS 组件化开发
- 微信支付集成
- 云开发数据库
- 分包加载优化
- 用户授权流程
代码示例:
// 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. 飞书集成开发工程师
文件: engineering/engineering-feishu-integration-developer.md
使用场景:
切换到飞书集成开发工程师,帮我开发一个审批流机器人
核心能力:
- 飞书机器人开发
- 审批流接口
- 多维表格操作
- 飞书文档 API
- 事件订阅
代码示例:
// 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. 前端开发者
文件: engineering/engineering-frontend-developer.md
使用场景:
激活前端开发者模式,帮我构建一个现代化的 React Dashboard
核心能力:
- React/Vue 组件开发
- 响应式设计
- 性能优化
- 状态管理(Redux/Zustand)
- 类型安全(TypeScript)
代码示例:
// 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>
);
}
智能体部门分类
工程部(Engineering)
33 个智能体,涵盖:
- 前端/后端/全栈开发
- AI/ML 工程
- DevOps/SRE
- 安全工程
- 嵌入式/物联网
- 上位机/工控
- 微信/飞书/钉钉集成
设计部(Design)
18 个智能体,涵盖:
- UI/UX 设计
- 品牌设计
- 视觉叙事
- 无障碍设计
- AI 图像生成
营销部(Marketing)
40 个智能体,涵盖:
- 小红书/抖音/B站运营
- SEO/SEM
- 内容创作
- 社交媒体
- 跨境电商
- 广告投放
产品部(Product)
22 个智能体,涵盖:
- 产品管理
- 需求分析
- 用户研究
- 数据驱动决策
- A/B 测试
其他部门
- 游戏(Game):15 个
- 安全(Security):12 个
- 金融(Finance):10 个
- 数据(Data):18 个
- 运营(Operations):14 个
- 客服(Support):8 个
- 法务(Legal):6 个
- HR(Human Resources):9 个
配置与环境变量
Agency Orchestrator 配置
创建 ~/.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
环境变量设置
# .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
常见使用模式
模式一:单智能体咨询
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]
模式二:智能体切换
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
模式三:多智能体协作(通过 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)
模式四:使用预设模板
# 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
故障排除
问题 1:智能体未加载
症状:在 AI 工具中无法激活智能体
解决方案:
# 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
问题 2:OpenClaw 格式转换失败
症状:convert.sh 报错
解决方案:
# 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+
问题 3:Agency Orchestrator 无法连接 LLM
症状:ao compose 报 API 错误
解决方案:
# 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
问题 4:中文乱码
症状:智能体输出乱码
解决方案:
# 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
问题 5:工作流执行超时
症状:Agency Orchestrator 任务超时
解决方案:
# 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
高级技巧
自定义智能体
创建 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
- [ ] ...
安装:
cp custom/my-agent.md ~/.claude/agents/
组合使用智能体
# 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
批量转换智能体
# 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
资源链接
- GitHub 仓库: https://github.com/jnMetaCode/agency-agents-zh
- Agency Orchestrator: https://github.com/jnMetaCode/agency-orchestrator
- 上游英文版: https://github.com/msitarzewski/agency-agents
- 配套书籍: https://book.aibuzhiyu.com/(AI 编程实战方法论)
支持的 AI 工具完整列表
- OpenClaw ⭐ 推荐(多智能体协作)
- Claude Code(Anthropic 官方)
- Cursor(智能 IDE)
- GitHub Copilot
- Gemini CLI(Google)
- Qwen Code(阿里云)
- Codex CLI(OpenAI)
- Aider(开源 AI 编程助手)
- Windsurf
- Antigravity
- Kiro(Amazon)
- Trae
- OpenCode
- DeerFlow 2.0(字节跳动)
- WorkBuddy(腾讯)
- Hermes Agent(NousResearch)
- Qoder
快速开始:克隆仓库 → 运行 ./scripts/install.sh → 在 AI 工具中激活智能体 → 开始使用 215 个专家角色!
FAQ
What does the agency-agents-zh-skill skill do?
使用 agency-agents-zh 中文 AI 智能体角色库,为 AI 编程工具提供 215 个即插即用的专家角色
How do I install agency-agents-zh-skill?
Run npx skills add https://github.com/reason-machines/ai-agent-skills --skill agency-agents-zh-skill in your project. The skills CLI detects the agents you use and copies the skill into the right folder. For Claude Code you can also place the skill folder at ~/.claude/skills/agency-agents-zh-skill/ (personal) or .claude/skills/agency-agents-zh-skill/ (project).
Which AI agents can use agency-agents-zh-skill?
Any agent that supports the open Agent Skills format (a folder with a SKILL.md file), including Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, Windsurf, Cline, OpenCode and more.
Is agency-agents-zh-skill safe to install?
No partner security audit is available yet. Skills can run commands and scripts, so read the SKILL.md and any bundled files before installing.
Who maintains agency-agents-zh-skill?
It is published by reason-machines in reason-machines/ai-agent-skills. It has 323 recorded installs.
More from reason-machines/ai-agent-skills
agent-reach-internet-access
reason-machines/ai-agent-skills
Give AI agents eyes to see the internet — scrape Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu with zero API fees
agent-sprite-forge-2d-game-assets
reason-machines/ai-agent-skills
Generate 2D sprite sheets, animated GIFs, transparent PNGs, layered maps, and engine-ready prototypes for Godot/Unity using Codex-driven asset pipelines.
agentmemory-persistent-memory
reason-machines/ai-agent-skills
Add persistent memory to AI coding agents using agentmemory - remembers context, preferences, and decisions across sessions
awesome-agent-skills
reason-machines/ai-agent-skills
Curated collection of 1000+ agent skills compatible with Claude Code, Codex, Gemini CLI, Cursor, and more
oh-my-openagent-orchestration
reason-machines/ai-agent-skills
Orchestrate multi-agent AI workflows with ultrawork, discipline agents, team mode, and hash-anchored editing for autonomous code development
tencentdb-agent-memory
reason-machines/ai-agent-skills
Local-first long-term memory for AI agents with symbolic short-term memory and layered long-term recall via a 4-tier progressive pipeline
Popular ai & agents skills
google-agents-cli-publish
google/agents-cli
This skill should be used when the user wants to "publish an agent", "publish my ADK agent", "register an agent with Gemini", "publish to Gemini" (formerly Gemini Enterprise), or needs guidance on the agents-cli publish gemini command. Also use when the user wants to "manage agents in Agent Registry", "list/update/delete registered agents", or "register an MCP server". Covers ADK vs A2A registration modes, programmatic and interactive usage, flag reference, auto-detection from deployment metadata, Agent Registry fleet management, and troubleshooting. Part of the agents-cli skills suite. Do NOT use for deployment (use google-agents-cli-deploy).
google-agents-cli-eval
google/agents-cli
This skill should be used when the user wants to "run an evaluation", "evaluate my agent", "evaluate my ADK agent", "write an eval dataset", "analyze eval failures", "compare eval results", "optimize agent", or needs guidance on the Gemini platform eval methodology and the Quality Flywheel. Covers eval metrics, dataset schema, LLM-as-judge scoring, and common failure causes. Applies to any agents-cli project, whatever framework the agent is written in. Do NOT use for agent API code patterns (ADK: use google-agents-cli-adk-code), deployment (use google-agents-cli-deploy), or project scaffolding (use google-agents-cli-scaffold).
skill-creator
anthropics/skills
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
codex-pet
prime-skills/runcomfy-agent-skills
Codex Pet generator on RunComfy. Build a Codex-compatible Codex Pet spritesheet.webp + pet.json from a single reference image, drop it into ${CODEXHOME:-$HOME/.codex}/pets/ / and Codex picks it up as a custom Codex Pet next to the 8 built-ins. This skill produces the exact Codex Pet atlas Codex expects (1536x1872 PNG/WebP, 8 cols x 9 rows, 192x208 cells, 9 animation states — idle, running-right, running-left, waving, jumping, failed, waiting, running, review). Calls OpenAI GPT Image 2 edit ONCE via the local RunComfy CLI as runcomfy run openai/gpt-image-2/edit to produce a canonical Codex Pet pose, then assembles all 9 animation rows programmatically with ImageMagick micro-transforms — no Codex Pro, no $imagegen, no OPENAIAPIKEY required, only RUNCOMFYTOKEN. Triggers on "codex pet", "create codex pet", "make codex pet", "hatch codex pet", "/hatch image", "desktop pet codex", "codex pets", "spritesheet.webp", or any explicit ask to build a custom pet for OpenAI Codex.
full-output-enforcement
leonxlnx/taste-skill
Overrides default LLM truncation behavior. Enforces complete code generation, bans placeholder patterns, and handles token-limit splits cleanly. Apply to any task requiring exhaustive, unabridged output.
executing-plans
obra/superpowers
Use when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available