langchain-knowledge-patch
Installation
SKILL.md
LangChain Knowledge Patch
Use this guide when choosing current LangChain, LangGraph, provider-integration, MCP-adapter, or Deep Agents APIs. Read the topic reference before changing an agent loop, middleware policy, graph state, streaming contract, or provider tool.
Reference index
| Reference | Topics |
|---|---|
| Agents and tools | create_agent, middleware hooks and policies, structured output, model profiles, tool runtime injection, dynamic tools, custom state, and ToolNode errors |
| Anthropic integration | Native schemas, tool metadata and streaming, effort, citations, caching, context management, provider tools, code execution, remote MCP, and tool search |
| Deep Agents | Harness tools, filesystem and binary backends, protocol v2, background subagents, sandboxes, profiles, context storage, summarization, memory, and CLI use |
| LangGraph workflows | Typed stream and invoke results, event projections, delta checkpoints, timeouts and recovery, draining, JavaScript state schemas, node caching, fan-in, and reconnectable streams |
| MCP adapters | Session lifetime, structured and multimodal results, resources, prompts, runtime-aware interceptors, callbacks, progress, logging, and elicitation |
| OpenAI integration | Official and Azure endpoints, custom and built-in tools, Responses routing and continuation, approvals, compaction, reasoning, PDF input, and prompt caching |
| v1 migration | Package split, runtime floor, classic imports, agent migration, provider-neutral content blocks, rollout limits, and Google GenAI v4 |