LangChain LangGraph
Use this skill as the project-local dispatcher for LangChain and LangGraph guidance derived from the curated reference in .github/LangChain-LangGraph_organized_reference.md.
Workflow
- Open
references/index.mdfirst. - Pick the smallest matching reference file instead of loading everything.
- Prefer the LangChain-side files for agent construction, prompts, tools, ToolRuntime, middleware, model routing, and structured output.
- Prefer the LangGraph-side files for graph design, reducers, checkpoints, interrupts, subgraphs, streaming, and durable execution.
- Prefer the memory and retrieval files for long-term memory, state/store separation, vector stores, loaders, splitters, embeddings, and RAG architecture.
- Preserve the original source doc as the canonical reference for cross-checking; this skill is a bounded access layer, not a replacement source.
High-Value Repo Rules
- Prefer explicit LangGraph orchestration for complex workflows instead of burying a full agent loop inside a graph node.
- Reuse model and agent instances instead of rebuilding them inside each call path.
- Use structured output consistently for LLM and tool-facing boundaries when typed data matters.
- Keep async usage end-to-end across LangChain and LangGraph integrations.
- Preserve provider-specific message metadata when the model requires it across turns.
- Normalize persisted agent state when loading from checkpointers if schema or version drift is possible.
- Keep code examples close to the topic-specific references instead of collapsing them into abstract summaries.
Reference Files
references/index.mdreferences/01-langchain-overview.mdreferences/02-prompts-and-messages.mdreferences/03-tools-and-toolruntime.mdreferences/04-model-selection-and-structured-output.mdreferences/05-middleware-and-guardrails.mdreferences/06-runtime-state-store-context.mdreferences/07-langgraph-state-nodes-edges.mdreferences/08-checkpointing-persistence-durability.mdreferences/09-interrupts-hitl-resume.mdreferences/10-subgraphs-and-streaming.mdreferences/11-multi-agent-patterns.mdreferences/12-memory.mdreferences/13-retrieval-rag.md
Source Preservation
The source reference remains in .github/LangChain-LangGraph_organized_reference.md. This skill is an opencode-native access layer and does not replace the preserved source document.
Source: Harmeet10000/AgentNexus-LangChain-FastAPI — distributed by TomeVault.