This is a five-agent pipeline that handles refactoring with actual safety gates. Phoenix analyzes your code for smells and risks, plan-agent breaks the work into small reversible steps, kraken implements the changes, plan-reviewer checks you didn't break behavior, and arbiter runs your full test suite. The step-by-step approach is smart for large refactors where you want to catch issues early rather than debug a massive diff later. It assumes you have decent test coverage to begin with, otherwise the validation phase won't catch much. The dry-run and step-by-step flags are nice if you want to preview the plan or approve each change individually before moving forward.
npx skills add https://github.com/parcadei/continuous-claude-v3 --skill refactor