bojieli/loopx ? reverse-engineered prompt

Build me a lightweight local tool for managing long running AI agent work on my own machine.

I want something that keeps a project’s goal, current gate, todos, evidence, quota, and handoff notes in one stable place, so I can run bounded turns with Codex, Claude Code, Cursor, or a shell agent without losing context. It should be easy to install, easy to connect to an existing repo, and safe about not overwriting any current state. I’d like a simple command line flow for checking health, seeing the current status, starting a new goal with guidance, and recording evidence and next steps after each turn.

Please make it work as a local first control plane, with durable state on disk, clear human approval points, and a clean way to resume work later or pass it to another agent. If you need current docs or API details, look them up online and follow the existing project style.