bogdaniel/zenchron-engineering ? reverse-engineered prompt

Build me a Go tool that acts like a local control plane for AI coding agents. I want it to take work from GitHub issues, assign each issue to an available agent like Claude Code or Codex CLI, run the work in a bounded way, collect proof that the work was actually done, and then open a pull request for review.

It should keep me in charge of approvals, so the agents can do the coding but they cannot authorize their own changes. I want one command to start the service, another to check which agents are available, and commands to run or plan work, view status, and follow logs.

Please make it support multiple tasks at once, use policy and evidence to decide what is allowed, and keep the runtime local and truthful about what it can prove. If you need to look up current docs for the agent CLIs or GitHub behavior, go ahead and do that.