paradigmxyz/centaur — reverse-engineered prompt

Build me a self hosted team AI agent platform that works from Slack. I want people to mention a bot in a Slack thread, ask it to investigate something like broken tests or summarize context, and then get progress updates and a final answer back in the same thread.

Each conversation should get its own safe sandbox where the agent can run shell commands, inspect a repo, use Python and JavaScript tools, and call approved internal tools without seeing raw secrets. Add a simple API too, so the same flow can be started without Slack. Store messages, events, executions, and results so work can resume after restarts.

Make it possible to plug in different CLI agents like Claude Code or Codex, add Python tool plugins, and define longer workflows that can sleep, resume, schedule checks, or start child agent tasks. Include local setup for a small Kubernetes or k3s environment, Postgres, Slack credentials, and clear docs so a team can run it themselves. Look up current docs online if you need to.