ndqkhanh/lyra — reverse-engineered prompt

Build me a terminal based coding agent called Lyra. I want it to feel like a lightweight open source assistant that can run from the command line, take a user goal, break it into tasks, and coordinate a main agent with specialist agents.

It should have a simple agent registry, a skills system where skills can be discovered and matched to a request, short term and long term memory saved locally as JSON, and lifecycle hooks that can run before and after tool use. Add basic task coordination, dependencies, and status tracking so it can handle small software work instead of just chatting.

Keep it practical and honest. Don’t try to fake voice, desktop control, or huge future features. Just leave clean extension points for more model providers, tools, safety checks, and memory backends later.

Use Python for the core CLI and TypeScript only where it already makes sense. Include a clear README, install instructions, example commands, and tests for the core pieces. Look up current docs online if you need to.