Refactor Safe Workflow

Multi-Agent Pipeline

Stage 1: Assessment (Architect Agent)

• Input: Target file/module.
• Action: Identify code smells (Long Method, God Class, Duplication).
• Output: architectural refactoring plan (refactor_plan.md).

Stage 2: Safety Check (Test Writer Agent)

• Action: Run existing tests.
• Gap Analysis: If coverage < 100% on target logic, create Characterization Tests.
• Goal: Lock in current behavior (even bugs) to ensure we don't accidentally "fix" or break things silently.

Stage 3: Execution (Refactorer Agent)

• Loop:
  1. Apply one refactoring move from plan.
  2. Run tests.
  3. If Pass: Commit (refactor: ...).
  4. If Fail: specific Undo changes, verify Assumption, Retry or Pivot.

Stage 4: Review (Code Reviewer Agent)

• Check: Is the code actually better? (Cyclomatic complexity reduced? Naming clearer?).

Usage Example

/refactor --target src/legacy/OrderManager.ts --goal "Extract Payment Logic"

Error Recovery

• Test Failure: Instant revert (git reset --hard HEAD). Do not try to "debug" a refactoring. Logic should be preserved by definition.
• Type Errors: Fix immediately. Do not use ts-ignore.

Outputs

• Refactor plan and validation notes.

Related Skills

• /verify-code - Post-refactor review