LangChain
Your agent now builds LangChain apps using the modern agent runtime — create_agent(), @tool decorator, middleware for custom flows, and standard patterns for human-in-the-loop and error handling. Sourced from LangChain AI.
What it installs
| Artifact | Path (in your project) | Purpose |
|---|---|---|
| Skill | .agents/skills/langchain-fundamentals/ | Protocol the agent follows on LangChain work — sourced from langchain-ai/langchain-skills |
| Rule | .claude/rules/langchain.md | Always-loaded pointer: mandates create_agent() + middleware, triggers on any LangChain import |
How it works
LangChain has churned through many agent abstractions (AgentExecutor, LCEL, OpenAIAgent, ...). Most tutorials and older training data still reference deprecated patterns. The skill locks the agent onto the current, supported path: create_agent() for the runtime, @tool / tool() for tools, middleware for anything custom (retries, approval gates, logging), and documented streaming + error handling.
The rule loads a pointer into every session, so the agent consults the skill before writing imports from langchain / @langchain/core — no explicit invocation needed.
Setup
No env vars or external accounts required for the skill. Your provider API keys (OpenAI, Anthropic, etc.) are still needed to actually run agents. Skill fetched from GitHub during harness-kit add via npx skills add.
Pairs well with
langgraph— graph-based orchestration on top of LangChain agentsredis— LangCache for LLM response cachingcode-review-gates— enforce middleware + error-handling patterns on every commit