tuchg/Lucarne — reverse-engineered prompt
Reverse engineered prompt
Build me a lightweight desktop background tool that lets me walk away from my computer while local AI coding agents keep running, and lets me manage them from Telegram or WeChat.
I want it to watch existing agent sessions like Claude Code, Codex, Gemini, Copilot, and Pi without requiring project changes, hooks, MCP, or special plugins. It should send mobile notifications when an agent needs approval, asks a question, fails, or makes important progress. From Telegram I should be able to open a simple control panel, see agents and workspaces, start or resume sessions, approve or deny requests, interrupt work, check status, and fork a session when supported. From WeChat I should be able to scan a QR code, receive messages, and reply by quoting a notification so the reply goes back to the right session.
Make it run as a small cross platform daemon with install, init, doctor, paths, update, and autostart commands for macOS, Linux, and Windows. Store config locally, keep session history, release idle agents, and keep the setup simple and secure.
Want more depth? Deep Reverse