langchain-skills / config

# LangChain + Deep Agents Development Guide

This project uses skills that contain up-to-date patterns and working reference scripts.

## CRITICAL: Invoke Skills BEFORE Writing Code

**ALWAYS** invoke the relevant skill first - skills have the correct imports, patterns, and scripts that prevent common mistakes.

### Getting Started
- **ecosystem-primer** - Invoke FIRST for any LangChain/LangGraph/Deep Agents project: framework selection, env setup, and which skill to load next
- **langchain-dependencies** - Invoke before installing packages or when resolving version issues (Python + TypeScript)

### Local quickstarts
Invoke when the user wants a minimal working local agent. Skills point at the official Mintlify quickstarts (fetch live docs); add provider/model prompt + provider-key-only setup:
- **langchain-python-quickstart** / **langchain-typescript-quickstart** (weather)
- **langgraph-python-quickstart** / **langgraph-typescript-quickstart** (math)
- **deepagents-python-quickstart** / **deepagents-typescript-quickstart** (research; provider web search, not Tavily)

### LangChain Skills
- **langchain-fundamentals** - Invoke for create_agent, @tool decorator, middleware patterns
- **langchain-rag** - Invoke for RAG pipelines, vector stores, embeddings
- **langchain-middleware** - Invoke for structured output with Pydantic

### LangGraph Skills
- **langgraph-cli** - Invoke for langgraph CLI commands: new, dev, build, up, deploy, and langgraph.json configuration
- **langgraph-fundamentals** - Invoke for StateGraph, state schemas, edges, Command, Send, invoke, streaming, error handling
- **langgraph-persistence** - Invoke for checkpointers, thread_id, time travel, memory, subgraph scoping
- **langgraph-human-in-the-loop** - Invoke for interrupts, human review, error handling, approval workflows
- **langgraph-decision-models** - Invoke for routing on a decision model (TypeSafe Jev, SemIf) instead of an LLM, or auditing an agent for LLM calls that only produce a decision

### Deep Agents Skills
- **deep-agents-core** - Invoke for Deep Agents harness architecture
- **deep-agents-memory** - Invoke for long-term memory with StoreBackend
- **deep-agents-orchestration** - Invoke for multi-agent coordination
- **managed-deep-agents** - Invoke for Managed Deep Agents: build a code-first agent, add tools, middleware, MCP connectors, schedules, and sandboxes, and deploy with the `mda` CLI

## Environment Setup

Required environment variables:

```bash
OPENAI_API_KEY=<your-key>  # For OpenAI models
ANTHROPIC_API_KEY=<your-key>  # For Anthropic models
```