🎭 Agency Agents — AI Specialist Personalities
🎭 Agency Agents — AI专家人格集合
Skill by
ara.so — Daily 2026 Skills collection.
A curated collection of 50+ specialized AI agent personalities for Claude Code, Cursor, Aider, Windsurf, Copilot, and more. Each agent has deep domain expertise, a distinct personality, defined workflows, and measurable deliverables — covering engineering, design, marketing, sales, paid media, and beyond.
由
ara.so开发的技能——属于Daily 2026技能合集。
这是一个精心整理的、包含50+专业AI Agent人格的集合,适用于Claude Code、Cursor、Aider、Windsurf、Copilot等工具。每个Agent都具备深厚的领域专业知识、独特的人格、明确的工作流程和可衡量的交付成果——覆盖工程、设计、营销、销售、付费媒体等多个领域。
bash
git clone https://github.com/msitarzewski/agency-agents.git
cd agency-agents
bash
git clone https://github.com/msitarzewski/agency-agents.git
cd agency-agents
Claude Code (Recommended)
Claude Code(推荐)
Copy all agents to Claude's agents directory
将所有Agent复制到Claude的Agent目录
cp -r agency-agents/* ~/.claude/agents/
cp -r agency-agents/* ~/.claude/agents/
Or symlink for auto-updates
或创建符号链接以实现自动更新
ln -s /path/to/agency-agents ~/.claude/agents/agency
Then in any Claude Code session:
Hey Claude, activate Frontend Developer mode and help me build a React component
ln -s /path/to/agency-agents ~/.claude/agents/agency
Hey Claude, activate Frontend Developer mode and help me build a React component
All Other Tools (Interactive Installer)
其他所有工具(交互式安装程序)
Step 1: Generate integration files for all supported tools
步骤1:为所有支持的工具生成集成文件
Step 2: Auto-detect installed tools and install interactively
步骤2:自动检测已安装的工具并进行交互式安装
Or target a specific tool
或指定特定工具
./scripts/install.sh --tool cursor
./scripts/install.sh --tool copilot
./scripts/install.sh --tool aider
./scripts/install.sh --tool windsurf
./scripts/install.sh --tool cursor
./scripts/install.sh --tool copilot
./scripts/install.sh --tool aider
./scripts/install.sh --tool windsurf
| Tool | Install path |
|---|
| Claude Code | |
| Cursor | in project root |
| Copilot | .github/copilot-instructions.md
|
| Aider | or pass via |
| Windsurf | in project root |
| 工具 | 安装路径 |
|---|
| Claude Code | |
| Cursor | 项目根目录下的 |
| Copilot | .github/copilot-instructions.md
|
| Aider | 或通过传入 |
| Windsurf | 项目根目录下的 |
engineering/engineering-frontend-developer.md React/Vue/Angular, UI, Core Web Vitals
engineering/engineering-backend-architect.md API design, databases, scalability
engineering/engineering-mobile-app-builder.md iOS/Android, React Native, Flutter
engineering/engineering-ai-engineer.md ML models, AI integration, data pipelines
engineering/engineering-devops-automator.md CI/CD, infra automation, cloud ops
engineering/engineering-rapid-prototyper.md MVPs, POCs, hackathon speed
engineering/engineering-senior-developer.md Laravel/Livewire, advanced patterns
engineering/engineering-security-engineer.md Threat modeling, secure code review
engineering/engineering-code-reviewer.md PR reviews, code quality gates
engineering/engineering-database-optimizer.md PostgreSQL/MySQL tuning, slow queries
engineering/engineering-git-workflow-master.md Branching, conventional commits
engineering/engineering-software-architect.md System design, DDD, trade-off analysis
engineering/engineering-sre.md SLOs, error budgets, chaos engineering
engineering/engineering-incident-response-commander.md Incident management, post-mortems
engineering/engineering-technical-writer.md Developer docs, API reference
engineering/engineering-data-engineer.md Data pipelines, lakehouse, ETL/ELT
engineering/engineering-frontend-developer.md React/Vue/Angular, UI, Core Web Vitals
engineering/engineering-backend-architect.md API设计, 数据库, 可扩展性
engineering/engineering-mobile-app-builder.md iOS/Android, React Native, Flutter
engineering/engineering-ai-engineer.md ML模型, AI集成, 数据管道
engineering/engineering-devops-automator.md CI/CD, 基础设施自动化, 云运维
engineering/engineering-rapid-prototyper.md MVP, POC, 黑客松快速开发
engineering/engineering-senior-developer.md Laravel/Livewire, 高级模式
engineering/engineering-security-engineer.md 威胁建模, 安全代码审查
engineering/engineering-code-reviewer.md PR审查, 代码质量门禁
engineering/engineering-database-optimizer.md PostgreSQL/MySQL调优, 慢查询
engineering/engineering-git-workflow-master.md 分支管理, 规范化提交
engineering/engineering-software-architect.md 系统设计, DDD, 权衡分析
engineering/engineering-sre.md SLO, 错误预算, 混沌工程
engineering/engineering-incident-response-commander.md 事件管理, 事后复盘
engineering/engineering-technical-writer.md 开发者文档, API参考
engineering/engineering-data-engineer.md 数据管道, 数据湖仓, ETL/ELT
design/design-ui-designer.md Visual design, component libraries
design/design-ux-researcher.md User testing, behavior analysis
design/design-ux-architect.md CSS systems, technical UX
design/design-brand-guardian.md Brand identity and consistency
design/design-whimsy-injector.md Micro-interactions, delight, Easter eggs
design/design-image-prompt-engineer.md Midjourney/DALL-E/SD prompts
design/design-inclusive-visuals-specialist.md Representation, bias mitigation
design/design-ui-designer.md 视觉设计, 组件库
design/design-ux-researcher.md 用户测试, 行为分析
design/design-ux-architect.md CSS系统, 技术UX
design/design-brand-guardian.md 品牌标识与一致性
design/design-whimsy-injector.md 微交互, 愉悦体验, 复活节彩蛋
design/design-image-prompt-engineer.md Midjourney/DALL-E/SD提示词
design/design-inclusive-visuals-specialist.md 多元呈现, 偏见缓解
Marketing, Sales & Paid Media
营销、销售与付费媒体
marketing/marketing-growth-hacker.md
marketing/marketing-content-creator.md
paid-media/paid-media-ppc-strategist.md
paid-media/paid-media-creative-strategist.md
sales/sales-outbound-strategist.md
sales/sales-deal-strategist.md
sales/sales-discovery-coach.md
marketing/marketing-growth-hacker.md
marketing/marketing-content-creator.md
paid-media/paid-media-ppc-strategist.md
paid-media/paid-media-creative-strategist.md
sales/sales-outbound-strategist.md
sales/sales-deal-strategist.md
sales/sales-discovery-coach.md
Using Agents in Claude Code
在Claude Code中使用Agent
Activating a Single Agent
激活单个Agent
In Claude Code chat:
在Claude Code聊天中:
Activate the Backend Architect agent and help me design a REST API for a multi-tenant SaaS app.
Activate the Backend Architect agent and help me design a REST API for a multi-tenant SaaS app.
Using Multiple Agents in Sequence
按顺序使用多个Agent
First, design the system
首先,设计系统
Activate the Software Architect agent. Design the domain model for an e-commerce platform.
Activate the Software Architect agent. Design the domain model for an e-commerce platform.
Now activate the Senior Developer agent and implement the Order aggregate in Laravel.
Now activate the Senior Developer agent and implement the Order aggregate in Laravel.
Activate the Code Reviewer agent and review the implementation above.
Activate the Code Reviewer agent and review the implementation above.
Referencing an Agent File Directly
直接引用Agent文件
Pass an agent as a system prompt in Claude CLI
在Claude CLI中传入Agent作为系统提示词
claude --system-prompt "$(cat ~/.claude/agents/engineering-frontend-developer.md)"
"Build a responsive product card component in React with Tailwind CSS"
claude --system-prompt "$(cat ~/.claude/agents/engineering-frontend-developer.md)"
"Build a responsive product card component in React with Tailwind CSS"
Using Agents in Cursor
在Cursor中使用Agent
After running
./scripts/install.sh --tool cursor
, agent rules land in
. Reference them in chat:
@engineering-frontend-developer Build a data table component with sorting and pagination.
Or set a default rule in
.cursor/rules/default.mdc
:
markdown
---
alwaysApply: true
---
You are operating as the Senior Developer agent from The Agency.
Refer to .cursor/rules/engineering-senior-developer.md for your full persona and workflows.
运行
./scripts/install.sh --tool cursor
后,Agent规则会被放入
目录。在聊天中引用它们:
@engineering-frontend-developer Build a data table component with sorting and pagination.
或在
.cursor/rules/default.mdc
中设置默认规则:
markdown
---
alwaysApply: true
---
You are operating as the Senior Developer agent from The Agency.
Refer to .cursor/rules/engineering-senior-developer.md for your full persona and workflows.
Using Agents with Aider
与Aider一起使用Agent
Use a single agent as the system prompt
使用单个Agent作为系统提示词
aider --system-prompt "$(cat agency-agents/engineering/engineering-security-engineer.md)"
aider --system-prompt "$(cat agency-agents/engineering/engineering-security-engineer.md)"
Or reference in .aider.conf.yml
或在.aider.conf.yml中引用
echo "system-prompt: agency-agents/engineering/engineering-devops-automator.md" >> .aider.conf.yml
echo "system-prompt: agency-agents/engineering/engineering-devops-automator.md" >> .aider.conf.yml
Using Agents in Windsurf
在Windsurf中使用Agent
bash
./scripts/install.sh --tool windsurf
bash
./scripts/install.sh --tool windsurf
Agents are written to .windsurf/rules/
Agent会被写入.windsurf/rules/
Use the UX Architect agent rules from .windsurf/rules/ to audit my CSS architecture.
Use the UX Architect agent rules from .windsurf/rules/ to audit my CSS architecture.
Real Workflow Examples
实际工作流示例
Full-Stack Feature with Multiple Agents
使用多个Agent完成全栈功能
1. Architecture phase
1. 架构阶段
cat > task.md << 'EOF'
I need to add real-time notifications to my Node.js + React app.
Users should see in-app alerts and optionally receive email digests.
EOF
cat > task.md << 'EOF'
I need to add real-time notifications to my Node.js + React app.
Users should see in-app alerts and optionally receive email digests.
EOF
Invoke Software Architect
调用Software Architect
claude --system-prompt "$(cat ~/.claude/agents/engineering-software-architect.md)" < task.md
claude --system-prompt "$(cat ~/.claude/agents/engineering-software-architect.md)" < task.md
2. Backend implementation
2. 后端实现
claude --system-prompt "$(cat ~/.claude/agents/engineering-backend-architect.md)"
"Implement the notification service based on the architecture above using PostgreSQL LISTEN/NOTIFY and Socket.io"
claude --system-prompt "$(cat ~/.claude/agents/engineering-backend-architect.md)"
"Implement the notification service based on the architecture above using PostgreSQL LISTEN/NOTIFY and Socket.io"
3. Frontend implementation
3. 前端实现
claude --system-prompt "$(cat ~/.claude/agents/engineering-frontend-developer.md)"
"Build the React notification bell component that connects to the Socket.io feed"
claude --system-prompt "$(cat ~/.claude/agents/engineering-frontend-developer.md)"
"Build the React notification bell component that connects to the Socket.io feed"
4. Security review
4. 安全审查
claude --system-prompt "$(cat ~/.claude/agents/engineering-security-engineer.md)"
"Review the notification system implementation for security issues"
claude --system-prompt "$(cat ~/.claude/agents/engineering-security-engineer.md)"
"Review the notification system implementation for security issues"
Code Review Workflow
代码审查工作流
Generate a diff and pipe to the Code Reviewer agent
生成diff并传递给Code Reviewer Agent
git diff main..feature/payment-integration |
claude --system-prompt "$(cat ~/.claude/agents/engineering-code-reviewer.md)"
"Review this PR diff. Focus on security, correctness, and maintainability."
git diff main..feature/payment-integration |
claude --system-prompt "$(cat ~/.claude/agents/engineering-code-reviewer.md)"
"Review this PR diff. Focus on security, correctness, and maintainability."
Database Optimization
数据库优化
Paste slow query log and activate Database Optimizer
粘贴慢查询日志并激活Database Optimizer
claude --system-prompt "$(cat ~/.claude/agents/engineering-database-optimizer.md)" << 'EOF'
Here is a slow query from our PostgreSQL logs (avg 4200ms):
SELECT u., p., o.*
FROM users u
LEFT JOIN profiles p ON p.user_id = u.id
LEFT JOIN orders o ON o.user_id = u.id
WHERE u.created_at > NOW() - INTERVAL '30 days'
ORDER BY o.created_at DESC;
Table sizes: users=2M rows, orders=18M rows. No indexes on created_at columns.
EOF
claude --system-prompt "$(cat ~/.claude/agents/engineering-database-optimizer.md)" << 'EOF'
Here is a slow query from our PostgreSQL logs (avg 4200ms):
SELECT u., p., o.*
FROM users u
LEFT JOIN profiles p ON p.user_id = u.id
LEFT JOIN orders o ON o.user_id = u.id
WHERE u.created_at > NOW() - INTERVAL '30 days'
ORDER BY o.created_at DESC;
Table sizes: users=2M rows, orders=18M rows. No indexes on created_at columns.
EOF
Structured incident kick-off
结构化事件启动
claude --system-prompt "$(cat ~/.claude/agents/engineering-incident-response-commander.md)" << 'EOF'
SEV-1 INCIDENT: Payment processing returning 503 errors since 14:32 UTC.
Error rate: 94%. Affected: checkout, subscription renewals.
Recent deploys: payment-service v2.4.1 at 14:15 UTC.
EOF
claude --system-prompt "$(cat ~/.claude/agents/engineering-incident-response-commander.md)" << 'EOF'
SEV-1 INCIDENT: Payment processing returning 503 errors since 14:32 UTC.
Error rate: 94%. Affected: checkout, subscription renewals.
Recent deploys: payment-service v2.4.1 at 14:15 UTC.
EOF
Creating Custom Agents
创建自定义Agent
Agent files follow a consistent markdown structure:
You are [Name], [role] at The Agency...
你是[姓名],The Agency的[角色]...
[What this agent optimizes for]
Personality & Communication Style
人格与沟通风格
- [Concrete output 1]
- [Concrete output 2]
Save custom agents to `agency-agents/custom/` and re-run `./scripts/convert.sh` to generate tool integrations.
---
将自定义Agent保存到`agency-agents/custom/`目录,然后重新运行`./scripts/convert.sh`以生成工具集成文件。
---
Contributing New Agents
贡献新Agent
Create your agent in the appropriate division
在对应部门下创建你的Agent
cp engineering/engineering-senior-developer.md
engineering/engineering-YOUR-SPECIALTY.md
cp engineering/engineering-senior-developer.md
engineering/engineering-YOUR-SPECIALTY.md
Edit the file, then test it
编辑文件,然后测试
claude --system-prompt "$(cat engineering/engineering-YOUR-SPECIALTY.md)"
"Give me a sample deliverable to demonstrate your capabilities"
claude --system-prompt "$(cat engineering/engineering-YOUR-SPECIALTY.md)"
"Give me a sample deliverable to demonstrate your capabilities"
git checkout -b agent/your-specialty
git add engineering/engineering-YOUR-SPECIALTY.md
git commit -m "feat: add Your Specialty agent"
git push origin agent/your-specialty
git checkout -b agent/your-specialty
git add engineering/engineering-YOUR-SPECIALTY.md
git commit -m "feat: add Your Specialty agent"
git push origin agent/your-specialty
Agents not found in Claude Code
If empty, re-run:
如果为空,重新运行:
cp -r /path/to/agency-agents/* ~/.claude/agents/
**`convert.sh` fails with permission error**
```bash
chmod +x scripts/convert.sh scripts/install.sh
./scripts/convert.sh
Cursor not picking up agent rules
cp -r /path/to/agency-agents/* ~/.claude/agents/
**`convert.sh`因权限错误失败**
```bash
chmod +x scripts/convert.sh scripts/install.sh
./scripts/convert.sh
Cursor未识别Agent规则
Rules must be in project root .cursor/rules/
规则必须位于项目根目录的.cursor/rules/下
Re-run installer targeting cursor
重新运行针对Cursor的安装程序
./scripts/install.sh --tool cursor
**Agent personality not activating**
- Be explicit: *"Activate the Frontend Developer agent"* rather than just referencing the topic
- Paste the agent file contents directly into the system prompt if tool integration isn't working
- For Claude Code, confirm agents directory: `claude config get agentsDir`
**Agent conflicts when using multiple**
- Activate one agent per conversation session
- For multi-agent workflows, use separate sessions or Claude Code's subagent feature
- Sequence agents explicitly: architect → implement → review
---
./scripts/install.sh --tool cursor
**Agent人格未激活**
- 明确指令:使用“Activate the Frontend Developer agent”而非仅提及主题
- 如果工具集成无效,直接将Agent文件内容粘贴到系统提示词中
- 对于Claude Code,确认Agent目录:`claude config get agentsDir`
**使用多个Agent时发生冲突**
- 每个对话会话激活一个Agent
- 对于多Agent工作流,使用单独会话或Claude Code的子Agent功能
- 明确按顺序调用Agent:架构师→实现者→审查者
---
agency-agents/
├── engineering/ # 23 engineering specialist agents
├── design/ # 8 design specialist agents
├── marketing/ # Marketing and growth agents
├── sales/ # 8 sales specialist agents
├── paid-media/ # 7 paid media specialist agents
├── scripts/
│ ├── convert.sh # Generate tool-specific integration files
│ └── install.sh # Interactive installer (auto-detects tools)
└── README.md
agency-agents/
├── engineering/ # 23个工程领域专家Agent
├── design/ # 8个设计领域专家Agent
├── marketing/ # 营销与增长类Agent
├── sales/ # 8个销售领域专家Agent
├── paid-media/ # 7个付费媒体领域专家Agent
├── scripts/
│ ├── convert.sh # 生成工具特定的集成文件
│ └── install.sh # 交互式安装程序(自动检测工具)
└── README.md
- License: MIT
- 51,000+ stars — battle-tested by a large community
- No API keys required — agents are prompt files, not services
- Tool-agnostic — works with any LLM tool that accepts system prompts
- Extensible — add custom agents following the same markdown pattern
- PRs welcome — the roster grows through community contributions
- 许可证:MIT
- 51000+星标 — 经过大型社区的实战检验
- 无需API密钥 — Agent是提示词文件,而非服务
- 工具无关 — 适用于任何接受系统提示词的LLM工具
- 可扩展 — 遵循相同Markdown模式添加自定义Agent
- 欢迎PR — Agent清单通过社区贡献不断增长