Deep Agents Core

This is a comprehensive wrapper around LangChain that adds production-ready middleware for complex AI workflows. Instead of building task planning, file management, and delegation logic yourself, you get TodoListMiddleware for breaking down work, filesystem tools with pluggable backends, and SubAgent spawning out of the box. The skill system uses progressive disclosure, loading SKILL.md files only when relevant rather than cramming everything into context upfront. Use this when you need persistent memory across sessions, human approval workflows, or agents that can delegate specialized work to subagents. It's opinionated but handles the plumbing so you focus on configuration rather than middleware implementation.