🧠 LangGraph ReAct Agent with Claude "Think" Tool

This template showcases a ReAct agent implemented using LangGraph, extended with a Claude-style "think" tool that enables structured internal reasoning before taking any action—ideal for complex multi-step workflows and policy-compliant environments.

Inspired by Anthropic’s "think" tool for Claude 3.7 and τ-Bench benchmark: Read the blog post

---

🚀 Features

• 🧠 think Tool: Forces the agent to reflect before acting, logging structured reasoning steps.
• 🔁 ReAct Agent Loop: Reason → Act → Observe → Repeat.
• 🔌 Pluggable Tooling: Easily integrate new tools (e.g., search, APIs).
• 📚 Optimized Prompting: Includes domain-style guidance for better decision-making.
• 🧩 LangGraph Studio Ready: Visualize and edit the graph in-browser.

---

🔍 What’s New

This fork adds a Claude-inspired think tool with the following logic:

def think(thought: str, config: Annotated[RunnableConfig, InjectedToolArg]) -> str:
    """Think tool for structured intermediate reasoning before actions."""

Used as a scratchpad to:

• List applicable rules
• Check required information
• Validate planned actions
• Reflect on tool outputs

📌 Prompt template included in THINK_PROMPT, used dynamically with Configuration.

---

📈 Use Cases

• Complex workflows (e.g., booking, support, compliance)
• Long tool-call chains needing internal verification
• Reasoning transparency and debuggability
• Agents operating in policy-heavy domains (e.g., airline, finance, legal)

---

🛠️ Getting Started

1. Clone the repo
2. Create a .env file from .env.example
3. Install all requirements with pip install -r requirements.txt
4. Add your API keys (OpenAI, Tavily)
5. Run in LangGraph Studio

---

🧩 File Overview

File	Purpose
src/agent/tools.py	Includes think and search tools
src/agent/prompts.py	System and think prompt configuration
src/agent/configuration.py	Central configuration (e.g. model setup)

---

📚 Resources

• Anthropic "Think" Tool Post
• LangGraph Documentation
• LangChain Tools Guide

---

🧰 Tools Configuration

You can modify the tool list in src/agent/tools.py:

TOOLS: List[Callable[..., Any]] = [search, think]

And the system prompt logic in src/agent/prompts.py.

---

✅ Development Tips

• Use hot reload in LangGraph Studio
• Add custom nodes for internal logging or external validation
• Extend think() to support domain-specific policies or memory