AI-optimized skills for building multi-agent systems with Deep Agents and LangGraph (>=0.2.0).
| Path | When to use | Key API |
|---|---|---|
| Deep Agents | Medium complexity, 15 lines to a working agent | create_deep_agent() — built-in planning, filesystem, sub-agents |
| LangGraph Graph API | Full topology control | StateGraph + nodes + edges + reducers |
| LangGraph Functional API | Minimal boilerplate | @entrypoint + @task decorators |
deepagents-skill/
├── 00_quickstart/ # 4 template projects — copy and run
├── 01_atomic/ # 11 single-purpose operation guides
├── 02_patterns/ # 9 design patterns (reflection, parallel delegation, etc.)
├── 03_examples/ # Full projects (deep research, hybrid multi-agent)
├── 04_langgraph_integration/ # LangGraph-specific guides
├── ROUTER.md # Decision tree — start here
└── deepagents-rules.md # Project rules for AI assistants
For AI assistants (Cursor / GitHub Copilot / Claude): point the assistant at ROUTER.md — it contains a decision tree to locate the exact file needed.
For humans: browse 00_quickstart/ to get started, or read ROUTER.md for the full map.
The atomic skills are structured tutorials, not executable units — they don't take parameters and output code. The agent reads, understands, and decides how to act. Opus and Sonnet handle this well.