anvie/evonic — reverse-engineered prompt

Build me an open agent platform where I can create AI agents, choose their model, give them tools, add knowledge files, install skills, and run them from a simple browser dashboard plus a command line tool.

I want agents to work in local folders, on remote machines, or through a lightweight tunnel connector. They should be able to talk to each other, delegate tasks, and run as a small swarm with supervisor style control. Include a scheduler for recurring jobs, persistent artifacts so outputs are saved, and backup and restore for the whole setup.

Please add safety controls so agent actions are checked before they run, especially risky file operations, privilege changes, remote execution, and prompt injection attempts. Support OpenAI compatible model endpoints, local or cloud. Add Telegram and WhatsApp channels for chatting with agents, and make skills and plugins installable as zip packages.

Make it easy to install, configure, start, stop, diagnose, and update. Look up current docs online if needed.