LangGraph Orchestration

Use this skill when the task is about building or fixing an orchestrated agent system on the LangGraph stack.

This skill is written for the agent, not as user-facing docs. Favor shipped patterns and official APIs over improvised orchestration.

First Move

Classify the request before touching code:

1. High-level agent harness: prefer deepagents.create_deep_agent(...)
2. General-purpose tool-calling agent: prefer langchain.agents.create_agent(...)
3. Custom workflow / graph topology: prefer langgraph.graph.StateGraph
4. MCP tool integration: prefer langchain_mcp_adapters
5. Human approval / resumability: prefer interrupt() + checkpointer
6. Specialist delegation: prefer deep-agent subagents, or explicit subgraphs