R.E.F.A.C.T. Methodology

The R.E.F.A.C.T. methodology provides guardrails for AI-assisted refactoring, ensuring that improvements don't come at the cost of stability.

The Six Steps

R — Review

Understand what you're refactoring before changing anything.

Activities:

• Read the current implementation
• Identify what the code does (not how)
• Document expected behavior
• Note any existing tests

AI prompt: "Explain what this code does. What are its inputs, outputs, and side effects?"

E — Extract

Break down large pieces into smaller, focused units.

Activities:

• Identify extractable functions
• Find repeated code to consolidate
• Separate concerns
• Create clear interfaces

AI prompt: "Identify functions that could be extracted from this code. What would each do?"

F — Format

Apply consistent formatting and style.

Activities:

• Consistent naming
• Proper indentation
• Organized imports
• Clear structure

AI prompt: "Reformat this code following [your style guide]. Don't change behavior."

A — Analyze

Look for deeper issues now that code is cleaner.

Activities:

• Identify remaining complexity
• Find potential bugs
• Spot performance issues
• Check for security concerns

AI prompt: "Analyze this refactored code. What issues remain? What could still be improved?"

C — Clean

Remove unnecessary code and simplify.

Activities:

• Delete dead code
• Remove unused variables
• Simplify overly complex logic
• Eliminate redundancy

AI prompt: "Identify any dead code, unused variables, or unnecessary complexity. What can be safely removed?"

T — Test

Verify the refactoring didn't break anything.

Activities:

• Run existing tests
• Add tests for uncovered paths
• Manual verification of behavior
• Performance comparison

AI prompt: "Generate tests that verify this code still behaves correctly after refactoring."

Why Order Matters

Each step prepares for the next:

• Review before changing prevents blind refactoring
• Extract before format makes formatting easier
• Analyze after cleanup reveals deeper issues
• Test last catches any regressions