chekusu/wanman — reverse-engineered prompt
Reverse engineered prompt
Build me an open source local command line tool called wanman. I want to point it at an existing git repo and have a small team of Claude Code or Codex agents work on it while I mostly watch.
It should have a supervisor that starts agents, passes messages between them, keeps shared context, tracks tasks, initiatives, hypotheses, change capsules, and saves artifacts. Each agent needs its own safe workspace and home folder so it doesn’t mess up my current checkout or shell settings.
Give me simple commands to start a run from a goal, take over a repo, send and receive agent messages, list agents, manage tasks, and watch live activity. Add a config file where I can define agent names, roles, lifecycle, models, port, database path, and workspace location.
Please build it in TypeScript as a monorepo with tests, docs, and a source install flow using pnpm. Also include a small experimental finance review app for API credentials, provider costs, Stripe revenue, and ROI summaries. Look up current Claude Code and Codex CLI docs if needed.
Want more depth? Deep Reverse