langchain-ai/deepagents ? reverse-engineered prompt

Build me a powerful AI agent that can act as a personal assistant, kind of like a self-hosted version of Cursor or Claude Code that I can run from my terminal. It should be a complete, "batteries-included" system that can handle complex requests out of the box.

The agent needs to be able to plan its own work, like creating a to-do list for a task I give it. It should have tools for accessing my local filesystem to read, write, and edit files, and also be able to run shell commands safely in a sandbox. I'd also like it to be able to delegate work by spawning sub-agents. Make it easy to customize, so I can swap in different models like GPT-4o or add my own tools. The main way I'll interact with it is through a nice terminal interface, so please build that too.