What problem does it solve? Small-scale code cleanup often drifts into unplanned refactors, hidden behavior changes, or scope creep. This Skill enforces a disciplined workflow for minor optimizations—deduplication, naming improvements, local performance tuning—so changes stay minimal, verifiable, and safe to commit. ## Core Features & Use Cases - Understand-before-change workflow: Reads project conventions (AGENTS.md) and related code first, then plans by optimization goal, approach, and affected files. - Minimal-change ladder: A stepwise checklist decides whether to write code at all—reuse existing implementations, standard library, or deletion before adding anything new. - Behavior-preservation verification: Compiles and runs unit tests (Maven examples included), with a checklist to confirm interfaces, UI interactions, and flows remain unchanged. - Boundary discipline: Real defects found during optimization are reported separately and handed to bug-fix rather than silently fixed; structural refactors route to the refactor skill. - Use Case: Ask to extract duplicated logic in a service class; the Skill plans the extraction, verifies no caller behavior changes, runs tests, and commits with a conventional Chinese commit message. ## Quick Start Ask the assistant to optimize this duplicated code and clean up the naming in the current module without changing any external behavior.
