safe-refactor

Safe refactoring prevents behavior changes by running baseline tests before edits, then retesting and self-correcting up to 3 times before committing. This loop ensures your refactor preserves the original module's logic across common scenarios.

The safest refactor uses a plan-first workflow requiring explicit approval before edits, followed by lint and targeted test runs on changed files. Commits are guarded so they only happen when all tests pass.

Automated refactoring prevents regressions by establishing a baseline test run, applying edits, and entering a self-correcting loop if tests fail. It commits changes only after verifying the targeted tests pass.

Yes, plan-first execution requires explicit user approval before any edits are made. This governance ensures you review the intended renames, extractions, or restructures before the agent runs lint and tests.

When tests fail during a refactor, the agent enters a self-correcting loop and attempts to fix the failures up to 3 times. It keeps changes targeted and will only commit once all targeted tests pass.

You can use this for both small function extractions and large modular restructures. The baseline-test, edit, and retest loop applies to any behavior-preserving code changes as long as you provide an explicit plan.