Claude Agent - ReAct Pattern Implementation

Learning project: Building AI agents from scratch without frameworks

What is this?

A Python implementation of the ReAct (Reasoning + Acting) pattern for AI agents using Claude API. Built to understand agent fundamentals before using frameworks like LangChain.

Features

• ✅ ReAct loop with tool calling
• ✅ Modular tool system (auto-discovery)
• ✅ Unit tests (no token usage)
• ✅ Integration tests (optional)
• ✅ Loop protection (max iterations)

Quick Start

# Setup
uv venv
source .venv/bin/activate
uv sync

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

# Run
python main.py

# Test (no tokens)
pytest test_tools.py -v

# Integration tests (uses tokens)
pytest -m integration -v

Adding a New Tool

Create tools/your_tool.py:

"""Your tool description"""

TOOL_DEFINITION = {
    "name": "your_tool",
    "description": "What it does",
    "input_schema": {
        "type": "object",
        "properties": {
            "param": {"type": "string", "description": "..."}
        },
        "required": ["param"]
    }
}

def execute(tool_input: dict) -> str:
    """Execute the tool"""
    return "result"

That's it! Auto-discovered on next run.

Tool Schema Format

Anthropic (Claude):

{
    "name": "calculator",
    "input_schema": { ... }  # JSON Schema
}

OpenAI:

{
    "type": "function",
    "function": {
        "name": "calculator",
        "parameters": { ... }  # JSON Schema
    }
}

Note: Internal schema (properties, required) is standard JSON Schema for both.

Learning Goals

• Understand ReAct pattern internals
• Build agents without framework magic
• Create production-ready code

Next Steps

• System prompts
• Streaming responses
• Persistent memory

License

MIT