lab-agent-skills-design

内容来源：README.md（说明文档） · 原始地址 · 查看安装指南

原始内容

# Claude AI Agent with Skills Architecture

A demonstration of modern AI agent design patterns using Claude API with a modular skills system, persistent memory, and security-first principles.

## 🎯 Purpose

This project showcases:
- **Agentic AI Design**: Modular skill system with progressive disclosure
- **Production Patterns**: Structured prompts, memory management, and error handling
- **Security Best Practices**: Least privilege, context minimization, and input validation
- **Clean Architecture**: Separation of concerns between agent core, skills, and memory

## 🏗️ Architecture

```
agent.py              # Main agent loop with Claude integration
├── prompts/
│   └── system.txt    # System prompt with skill discovery
├── skills/
│   ├── powershell.py # PowerShell execution skill
│   └── examples/
│       ├── SKILL.md      # Skill metadata & instructions
│       ├── reference.md  # Technical reference (loaded on-demand)
│       └── security.md   # Security boundaries
└── memory/
    └── short_term.json   # Conversation history & context
```

## 🚀 Quick Start

1. **Install dependencies:**
   ```bash
   pip install -r requirements.txt
   ```

2. **Configure API key:**
   ```bash
   cp .env.example .env
   # Add your ANTHROPIC_API_KEY to .env
   ```

3. **Run the agent:**
   ```bash
   python agent.py
   ```

## 💡 Key Features

### Modular Skills System
- Skills are self-contained Python modules
- Progressive disclosure: Load detailed context only when needed
- Clear metadata (name, description, parameters)

### Memory Management
- Short-term memory for conversation context
- Persistent storage between sessions
- Automatic memory pruning to stay within token limits

### Security-First Design
- Explicit security boundaries per skill
- Input validation and sanitization
- Audit logging for sensitive operations
- Principle of least privilege

## 📚 Skills Overview

### PowerShell Skill
Execute PowerShell commands with safety guardrails:
- Whitelist of allowed commands
- Output sanitization
- Execution timeout
- Error handling and logging

## 🎓 Learning Outcomes

This project demonstrates:
- ✅ Working with Claude API (Anthropic SDK)
- ✅ Prompt engineering with tool use
- ✅ Modular software architecture
- ✅ Security considerations in AI systems
- ✅ Production-ready error handling
- ✅ Documentation best practices

## 🔒 Security Considerations

- Skills operate with minimal permissions
- No external network access unless explicitly authorized
- All user inputs are validated
- Sensitive data is never logged
- Skills include security.md defining boundaries

## 📝 License

MIT License - Free for learning and demonstration purposes

## 👤 Author

Created as a portfolio demonstration of AI agent architecture and Python best practices.

---

*This is a demonstration project designed to showcase technical skills for potential employers.*
=======
## Security Focus
- Least privilege
- Context minimization
- Trusted content only


## Resources 
- https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview
>>>>>>> f52fbcd677b74b19ed210fb411f6b1469fba8da5

Claude AI Agent with Skills Architecture

A demonstration of modern AI agent design patterns using Claude API with a modular skills system, persistent memory, and security-first principles.

🎯 Purpose

This project showcases:

Agentic AI Design: Modular skill system with progressive disclosure
Production Patterns: Structured prompts, memory management, and error handling
Security Best Practices: Least privilege, context minimization, and input validation
Clean Architecture: Separation of concerns between agent core, skills, and memory

🏗️ Architecture

agent.py              # Main agent loop with Claude integration
├── prompts/
│   └── system.txt    # System prompt with skill discovery
├── skills/
│   ├── powershell.py # PowerShell execution skill
│   └── examples/
│       ├── SKILL.md      # Skill metadata & instructions
│       ├── reference.md  # Technical reference (loaded on-demand)
│       └── security.md   # Security boundaries
└── memory/
    └── short_term.json   # Conversation history & context

🚀 Quick Start

Install dependencies:
```
pip install -r requirements.txt
```

Configure API key:

cp .env.example .env
# Add your ANTHROPIC_API_KEY to .env

Run the agent:
```
python agent.py
```

💡 Key Features

Modular Skills System

Skills are self-contained Python modules
Progressive disclosure: Load detailed context only when needed
Clear metadata (name, description, parameters)

Memory Management

Short-term memory for conversation context
Persistent storage between sessions
Automatic memory pruning to stay within token limits

Security-First Design

Explicit security boundaries per skill
Input validation and sanitization
Audit logging for sensitive operations
Principle of least privilege

📚 Skills Overview

PowerShell Skill

Execute PowerShell commands with safety guardrails:

Whitelist of allowed commands
Output sanitization
Execution timeout
Error handling and logging

🎓 Learning Outcomes

This project demonstrates:

✅ Working with Claude API (Anthropic SDK)
✅ Prompt engineering with tool use
✅ Modular software architecture
✅ Security considerations in AI systems
✅ Production-ready error handling
✅ Documentation best practices

🔒 Security Considerations

Skills operate with minimal permissions
No external network access unless explicitly authorized
All user inputs are validated
Sensitive data is never logged
Skills include security.md defining boundaries

📝 License

MIT License - Free for learning and demonstration purposes

👤 Author

Created as a portfolio demonstration of AI agent architecture and Python best practices.

This is a demonstration project designed to showcase technical skills for potential employers.

Security Focus

Least privilege
Context minimization
Trusted content only

Resources

https://platform.claude.com/docs/en/agents-and-tools/agent-skills/overview

f52fbcd677b74b19ed210fb411f6b1469fba8da5