Add LangGraph framework starter

[justinmadison]

Summary

Add a LangGraph framework starter that teaches developers how to use LangGraph by building an agent that plays Agent Arena scenarios. This is Phase 2 of the framework adapter system.

Key framing: "Want to learn LangGraph? Build an AI agent that plays a game." This starter is a tutorial for LangGraph, not just a template.

Prerequisites

• Add framework adapter system for LangGraph, Claude Agent SDK, and other agent frameworks #74 (framework adapter foundation + Anthropic starter) — provides the adapter ABC, observation formatter, and canonical tool definitions that this starter builds on

Deliverables

LangGraph Starter (starters/langgraph/)

File	Purpose
agent.py	LangGraph adapter + agent graph with tutorial comments
run.py	One-command startup
requirements.txt	agent-arena-sdk, langgraph, langchain-anthropic (or langchain-openai)
README.md	Learning goals, setup, how to modify, debugging

What It Teaches

• Graph construction — StateGraph, nodes, edges, conditional routing
• Tool nodes — StructuredTool.from_function(), ToolNode
• State management — LangGraph state schema, message passing
• ReAct pattern — tool calling loop with action detection
• Checkpointing — LangGraph memory (optional section)

Architecture

The LangGraph starter uses the adapter base from #74:

class LangGraphAdapter(FrameworkAdapter):
    def decide(self, observation: Observation) -> Decision:
        # 1. Format observation as context (inherited from base)
        context = self.format_observation(observation)
        
        # 2. Convert action tools to LangChain StructuredTools
        tools = [StructuredTool.from_function(...) for t in self.get_action_tools()]
        
        # 3. Build and invoke LangGraph agent
        result = self.graph.invoke({"messages": [HumanMessage(context)]})
        
        # 4. Extract action tool call from result, return as Decision
        return self._extract_decision(result)

Tests

• Unit tests for LangGraph adapter (mock LLM responses)
• Verify tool schema conversion to LangChain format
• Integration test with hand-built Observations

Success Criteria

• Developer can have a LangGraph agent playing foraging in under 15 minutes
• Starter teaches LangGraph core concepts through tutorial-quality comments
• Uses the shared adapter base from Add framework adapter system for LangGraph, Claude Agent SDK, and other agent frameworks #74 (observation formatting, tool schemas)
• Works with at least one LLM provider (Anthropic or OpenAI)
• README includes learning goals, prerequisites, and modification guide

Context

See docs/framework_integration_strategy.md for the full strategic discussion. This is Phase 2 of the framework integration roadmap.