Add CLAUDE.md developer guide for AI assistants

[jpanat]

Summary

Add comprehensive developer documentation (CLAUDE.md) to guide AI assistants working with the ai-competitor-agents codebase. This file serves as a reference for understanding the project architecture, conventions, and development workflows.

Key Changes

• Project overview: High-level description of the multi-agent competitor intelligence system
• Repository structure: Clear layout of all source files and their purposes
• Tech stack reference: Pinned versions of all dependencies (LangChain, LangGraph, FastAPI, Streamlit, etc.)
• Environment setup: Required API keys and optional configuration variables
• Architecture documentation: Detailed explanation of the blackboard pattern with LangGraph, including:
  • Sequential agent workflow diagram
  • AgentState TypedDict field reference
  • Agent implementation pattern and conventions
• Running the system: Commands for all execution modes (direct, API, Streamlit, CLI, examples)
• Key conventions:
  • LLM configuration and model selection
  • Tavily web search limits and rationale
  • Pydantic model contracts for the API
  • CORS configuration notes
• Output structure: Complete schema of the final AgentState response
• Development tasks: Guides for common operations (adding agents, changing models, adding endpoints, modifying CLI)
• Dependency notes: Warnings about version pinning and API stability
• Git workflow: Development and main branch information

Notable Details

This documentation is specifically written for AI assistants (Claude) to understand the codebase structure, design patterns, and conventions before making modifications. It includes implementation details like JSON extraction fallback patterns and status tracking mechanisms that are important for maintaining consistency when extending the system.

https://claude.ai/code/session_01MR1rjF18oNGDSJyV1BFPsu