LangGraph_Researcher: Instructions file for Claude Code

CLAUDE.md — LangGraph Researcher

This file provides guidance for AI assistants working in this repository.

Project Overview

LangGraph Researcher is a ReAct (Reasoning + Acting) agent built with LangGraph and Google Gemini. It combines web search with four specialized developer tools — called ADK Skills — that help with software tasks like formatting commits, adding license headers, generating Pydantic models, and validating database schemas.

The agent is exposed via LangGraph's API server and communicates entirely through the LangGraph chat interface.

---

Repository Structure

LangGraph_Researcher/
├── langgraph_101.py          # Agent entry point: LLM, tools, graph assembly
├── langgraph.json            # LangGraph configuration (graph name, version)
├── requirements.txt          # Python dependencies
├── verify_adk.py             # Test suite for all ADK skills
├── .env.example              # Required environment variables template
├── adk/
│   ├── __init__.py
│   ├── core.py               # @skill decorator (wraps functions as LangChain tools)
│   └── skills/
│       ├── __init__.py       # Re-exports all four skills
│       ├── git.py            # format_commit_message
│       ├── compliance.py     # add_license_header
│       ├── codegen.py        # generate_pydantic_model
│       └── data.py           # validate_schema
├── README.md                 # Documentation (Portuguese)
├── README.en.md              # Documentation (English)
└── CHANGELOG.md              # Version history (Keep a Changelog format)

---

Architecture

Agent (langgraph_101.py)

The agent is built with create_react_agent from langgraph.prebuilt. It:

• Uses gemini-2.0-flash via ChatGoogleGenerativeAI with temperature=0.2
• Has a Portuguese-language system prompt instructing it to search before answering, cite sources, and use ADK tools when appropriate
• Exposes five tools to the LLM:
  1. search_web — Tavily search (max 5 results, advanced depth)
  2. format_commit_message — Conventional Commits formatter
  3. add_license_header — MIT license header injector
  4. generate_pydantic_model — JSON → Pydantic converter
  5. validate_schema — DB schema governance checker