What problem does it solve?
Refactors often get risky and uncontrolled because it’s unclear what to change, how to preserve behavior, and how to keep progress verifiable with tests.
Core Features & Use Cases
- Find mode (opportunity scanning): identifies where modules are too shallow, where seams/adapters could improve testability, and where refactoring would increase leverage and locality.
- Plan mode (known refactor): scopes a specific refactor target, decomposes it into behavior-preserving tiny commits, and documents interface/seam decisions.
- Safety-first refactor planning: emphasizes characterization tests when coverage is thin, working-tree cleanliness, and a commit-by-commit verification mindset.
- Claude-Code refactoring workflow: produces a refactor plan written to .claude/refactor-plan.md and uses refactoring best practices (scope control, tests between commits, and stop for destructive operations).
Quick Start
Ask the AI to plan a safe refactor by saying: "Use the refactor skill to plan a shallow-to-deepening refactor for my codebase, then write the plan to .claude/refactor-plan.md."