This is a disciplined refactoring workflow that makes you write tests first, refactor in micro-steps, and prove nothing broke before shipping. It's built around a baseline-establish-prove loop with explicit risk tiers: low risk needs unit tests, high risk (auth, money, migrations) requires integration tests plus canary deploys. The decision trees and quick reference tables are genuinely useful when you're staring at a god object or legacy code with no tests. The "avoid refactor plus feature in one PR" and "fix determinism before trusting flaky tests" rules are things teams learn the hard way. If you're cleaning up technical debt or doing strangler migrations, this gives you guard rails so you don't accidentally break production while improving the code.
npx skills add https://github.com/vasilyu1983/ai-agents-public --skill qa-refactoring