1---2name: langchain-skills3description: Use when building agents with LangChain, LangGraph, or Deep Agents. 21 skills covering ecosystem primer, quickstarts, Deep Agents (memory, orchestration, managed), LangChain (fundamentals, middleware, RAG), LangGraph (fundamentals, persistence, CLI, human-in-the-loop), evaluation (Harbor), and utilities (swarm).4---56# LangChain Skills: Agent Engineering with LangChain, LangGraph & Deep Agents78Official agent skills from LangChain for building production-grade LLM applications. 21 skills covering the full stack from fundamentals to managed deployment.910---1112## Anti-Rationalization Table1314| Excuse | Reality | Rule |15|--------|---------|------|16| "I'll just use raw LLM calls" | Raw calls lack observability, memory, streaming, and eval infrastructure | Use LangGraph for stateful agents, LangSmith for tracing |17| "LangChain is too heavy" | Core is modular; use only what you need (create_agent, StateGraph, checkpointers) | Import specific modules, not the whole framework |18| "I don't need evals" | Evals catch regressions before production; Harbor makes it systematic | Run eval-engineering skill to set up Harbor evals |19| "Deep Agents are overkill" | Deep Agents add planning, subagents, filesystem - essential for complex tasks | Use Deep Agents when task needs multi-step planning + tools |20| "I'll write my own RAG" | RAG has 50+ failure modes; langchain-rag skill covers loaders, embeddings, vector stores | Use the skill; customize only the retrieval strategy |2122---2324## When to Use2526**Use when you need to:**2728- **Start a new agent project** → `ecosystem-primer` (framework selection: LangChain vs LangGraph vs Deep Agents)29- **Quick prototype** → Quickstart skills (weather bot, math agent, research agent)30- **Build Deep Agents** → `deep-agents-core`, `deep-agents-memory`, `deep-agents-orchestration`, `managed-deep-agents`31- **Build LangChain agents** → `langchain-fundamentals`, `langchain-middleware`, `langchain-rag`32- **Build LangGraph workflows** → `langgraph-fundamentals`, `langgraph-persistence`, `langgraph-cli`, `langgraph-human-in-the-loop`33- **Evaluate agents** → `eval-engineering` (Harbor evals with user approval)34- **Parallelize independent work** → `swarm` (dispatch + aggregate)3536---3738## Workflow39401. **Select Framework** — Run `ecosystem-primer` to choose LangChain / LangGraph / Deep Agents412. **Verify Setup** — Run appropriate quickstart (weather, math, or research agent)423. **Configure API Keys** — Set `OPENAI_API_KEY` and/or `ANTHROPIC_API_KEY`434. **Load Framework Skill** — Load skill matching your framework choice445. **Build & Test** — Develop agent with framework-specific patterns456. **Evaluate (Production)** — Use `eval-engineering` to set up Harbor evals467. **Deploy** — Use `managed-deep-agents` or `langgraph-cli` for deployment4748---4950## Quick Start5152### Install All Skills (via npx)53```bash54# Local (current project)55npx skills add langchain-ai/langchain-skills --skill '*' --yes5657# Global (all projects)58npx skills add langchain-ai/langchain-skills --skill '*' --yes --global5960# Link to specific agent (e.g., Claude Code)61npx skills add langchain-ai/langchain-skills --agent claude-code --skill '*' --yes --global62```6364### Install via Script (Claude Code & Deep Agents CLI)65```bash66# Clone and install67git clone https://github.com/langchain-ai/langchain-skills68cd langchain-skills6970# Install for Claude Code in current directory71./install.sh7273# Install for specific project74./install.sh ~/my-project7576# Install globally77./install.sh --global7879# Install for Deep Agents CLI80./install.sh --deepagents ~/my-project8182# Install globally for Deep Agents (includes agent persona)83./install.sh --deepagents --global84```8586### Required API Keys87```bash88export OPENAI_API_KEY=<your-key> # For OpenAI models89export ANTHROPIC_API_KEY=<your-key> # For Anthropic models90```9192---9394## Available Skills (21)9596### Getting Started97| Skill | Description |98|-------|-------------|99| `ecosystem-primer` | **Start here.** Framework selection (LangChain vs LangGraph vs Deep Agents), env setup, which skill to load next |100| `langchain-dependencies` | Full package version and dependency management reference (Python + TypeScript) |101102### Quickstarts (Local)103| Skill | Description |104|-------|-------------|105| `langchain-python-quickstart` | Python weather agent (default: anthropic:claude-sonnet-5) |106| `langchain-typescript-quickstart` | TypeScript weather agent |107| `langgraph-python-quickstart` | Python math agent |108| `langgraph-typescript-quickstart` | TypeScript math agent |109| `deepagents-python-quickstart` | Python research agent (web search) |110| `deepagents-typescript-quickstart` | TypeScript research agent |111112### Deep Agents113| Skill | Description |114|-------|-------------|115| `deep-agents-core` | Agent architecture, harness setup, SKILL.md format |116| `deep-agents-memory` | Memory, persistence, filesystem middleware |117| `deep-agents-orchestration` | Subagents, task planning, human-in-the-loop |118| `managed-deep-agents` | Deploy with CLI, use SDKs, stream runs, connect MCP tools, build React useStream UIs |119120### LangChain121| Skill | Description |122|-------|-------------|123| `langchain-fundamentals` | Agents with create_agent, tools, structured output, middleware basics |124| `langchain-middleware` | Human-in-the-loop approval, custom middleware, Command resume patterns |125| `langchain-rag` | RAG pipeline: document loaders, embeddings, vector stores |126127### LangGraph128| Skill | Description |129|-------|-------------|130| `langgraph-fundamentals` | StateGraph, nodes, edges, state reducers |131| `langgraph-persistence` | Checkpointers, thread_id, cross-thread memory |132| `langgraph-cli` | CLI lifecycle: scaffold, dev, build, deploy, langgraph.json config |133| `langgraph-human-in-the-loop` | Interrupts, human review, approval workflows |134135### Evaluation136| Skill | Description |137|-------|-------------|138| `eval-engineering` | Build, run, and audit Harbor evals for existing agent with user approval |139140### Utilities141| Skill | Description |142|-------|-------------|143| `swarm` | Dispatch independent work items in parallel and aggregate results |144145---146147## Framework Selection Guide148149| Use Case | Recommended Framework | Start With Skill |150|----------|----------------------|------------------|151| Simple tool-calling agent | LangChain | `langchain-fundamentals` |152| Stateful multi-turn conversation | LangGraph | `langgraph-fundamentals` |153| Complex planning + subagents + filesystem | Deep Agents | `deep-agents-core` |154| Need eval/monitoring in prod | LangSmith + Harbor | `eval-engineering` |155| Quick prototype (weather, math, research) | Any | Quickstart skills |156157---158159## Verification Checklist160161- [ ] Choose framework via `ecosystem-primer`162- [ ] Run appropriate quickstart to verify setup163- [ ] Install API keys (OPENAI_API_KEY, ANTHROPIC_API_KEY)164- [ ] Load framework-specific skill for your use case165- [ ] For production: set up `eval-engineering` with Harbor166- [ ] For complex tasks: use `deep-agents-orchestration` for subagents167- [ ] For stateful workflows: use `langgraph-persistence` with checkpointers168169---170171## References172173- **GitHub Repository**: https://github.com/langchain-ai/langchain-skills174- **Documentation**: https://langchain-ai.github.io/langgraph/175- **LangSmith Platform**: https://smith.langchain.com/176- **Harbor Evaluation**: https://github.com/langchain-ai/harbor177- **Deep Agents**: https://github.com/langchain-ai/deepagents178- **LangChain Academy**: https://academy.langchain.com/179- **License**: MIT180181---182183## Related 1ai-skills184185- `langchain-patterns` — LangChain/LangGraph patterns: chains, agents, tools, memory, retrieval, graph workflows186- `langgraph-fundamentals` — LangGraph state machines, nodes, edges, reducers (if available)187- `rag-builder` — RAG pipeline design: chunking, embedding, retrieval, answer generation188- `agent-harness-optimizer` — Agent harness optimization for token efficiency, memory persistence189- `model-router` — Route AI model requests to optimal provider
oyi77