hfadhlullah/kaizen ? reverse-engineered prompt

Build me a tool for working with AI coding agents that always shows the plan first, waits for my approval, then does the work, and finally runs a second fresh reviewer that did not see the build. I want it to manage the whole flow for each request, with a clean workspace, saved run state, tasks, review notes, and a fix loop that only retries serious issues a couple of times before stopping.

It should work from the terminal and a web dashboard, keep project specific state in a .kaizen folder, and let me pick up a run later from another session. Make the setup simple, with install and init commands, and support using it with Claude Code, Codex, Antigravity, and Gemini CLI. If you need current docs for those tools or Bun, look them up online.

I also want a small notes area the agent reads before it starts, plus a way to share requests from the web app and open it on my phone through a local network connection.