popmechanic/ultrapowers ? reverse-engineered prompt

Build me a Python project called ultrapowers that plugs into Superpowers and helps Claude Code run big approved plans more autonomously.

I want it to take a detailed plan, figure out which tasks depend on each other, split the independent work into parallel waves, and run those pieces in separate agents or isolated workspaces. Then it should check each task with a separate reviewer before marking it done, and only bring me back in for the final approval at the end. If the plan is too small or too sequential for this approach, it should say so and suggest using the simpler workflow instead.

Please make the whole thing feel polished, with good docs, clear tests, and a clean command line or skill based flow that fits the rest of the repo. Use the repo’s existing Claude Code and Superpowers setup, and if you need to check current docs for any integration details, look them up online.