omnigent-ai/omnigent ? reverse-engineered prompt

Build me an early but usable version of Omnigent, a Python based tool that gives me one place to run and supervise different AI agents like Claude Code, Codex, Cursor, Pi, and simple custom agents I define in YAML.

I want to start a session from the terminal, have that same live session available in a local web app, and be able to pick it up from another device like my phone. Let multiple agents work inside the same conversation, hand work off to sub agents, compare answers side by side, and have one agent review another agent’s output. I also want live collaboration so a teammate can open the same session, watch what is happening, chat, and continue from there.

Please include guardrails like policies and sandboxing so risky actions can pause for approval, spend can be capped, and tool access can be limited. Add a couple of example agents, one for multi agent coding orchestration and one for side by side model debates. Make the setup and local run experience really smooth. Look up current docs online if you need to.