langchain-dev

A Claude Code (and compatible agent) skill for developing LangChain 1.x / LangGraph applications. This npm package is the installer: it drops the skill into a project's .claude/skills/ folder so any Claude-powered agent in that project can use it.

The skill is an API quick-reference + selection guide — not a tutorial. For each API it answers four questions: what it is / what problem it solves / when to use it / how to call it — so the agent can quickly decide which API to pick.

What the skill covers

Triggers whenever the user mentions LangChain, LangGraph, LangSmith, create_agent, init_chat_model, structured output, RAG, vector retrieval, Agents, tool/function calling, prompt templates, context memory (Checkpointer / thread_id), middleware, embeddings, text splitting, or Ollama local models.

Reference modules included under references/:

| File | Topic | |---|---| | models.md | init_chat_model, provider adapters, model selection | | messages-prompts.md | Message types, prompt templates | | agent.md | create_agent, LangGraph orchestration | | tools.md | Tools / function calling, args_schema | | structured-output.md | Structured output / with_structured_output | | memory.md | Checkpointer, thread_id, persistence | | middleware.md | LangGraph middleware | | rag.md | RAG: loaders, splitters, embeddings, vector stores, retrieval | | langsmith.md | Tracing / observability / eval |

Bundled assets:

• assets/requirements.txt — reference dependency list for a LangChain 1.x project.
• assets/.env.example — .env template (model keys, base URLs, LangSmith, vector DB).

⚠️ All model names / thresholds / endpoints in the skill are placeholders. The only fixed things are API names, parameter names, and parameter semantics — fill real values per project.

Quick start

In the root of your project (or any folder where you want the skill enabled):

# one-off, no global install needed
npx langchain-dev init

This creates:

.claude/skills/langchain-dev/
├── SKILL.md
├── assets/
│   ├── .env.example
│   └── requirements.txt
└── references/

Restart Claude Code so it discovers the new skill, then just describe what you want — e.g. "build a LangGraph agent that calls these tools with persistent memory".

Requirements

• Node.js ≥ 14 (only to run the installer; zero runtime dependencies)
• macOS, Linux, and Windows

CLI

Usage:
  npx langchain-dev              Install into ./.claude/skills/langchain-dev
  npx langchain-dev init         Same as above (explicit form)
  npx langchain-dev init <dir>   Install into <dir>/.claude/skills/langchain-dev
  npx langchain-dev init --force Overwrite an existing install
  npx langchain-dev --version
  npx langchain-dev --help

Behavior notes:

• Idempotent & safe — if the skill is already installed, it prints a hint instead of clobbering. Use --force to refresh files in place.
• No postinstall hooks — nothing happens automatically on npm install; the skill is only copied when you explicitly run init.

Global install (optional)

npm install -g langchain-dev
langchain-dev init

Programmatic usage

The skill files are plain markdown + templates. If you want to copy them without the CLI, the files live under the skill/ directory of this package:

node -e "console.log(require('path').join(require.resolve('langchain-dev'), '..', 'skill'))"

Troubleshooting

• Agent doesn't see the skill → restart Claude Code after running init; skills are loaded at startup.
• npx ran the wrong version → pin with npx langchain-dev@latest init.
• Already installed warning → re-run with --force to refresh.
• Wrong directory → pass a target explicitly: npx langchain-dev init path/to/project.

License

MIT © smartyu1993