Code Refactoring Assistant

This skill is a comprehensive tool for safe code refactoring. It analyzes codebase changes by impact scope, creates prioritized checklists, facilitates user review, executes approved refactors, performs quality assurance, and generates before/after reports with metrics.

Distinguished from priority optimizers: Focuses on impact scope (files/modules/deps/tests) rather than business priority, with richer interactive review (yes/no/back/feedback).

Capabilities

Impact Analysis: Assess refactoring scope (low: <5 files/local; medium: 5-20 files/module; high: >20 files/cross-module/deps)
Checklist Generation: Structured lists grouped by impact, with risks and benefits
Interactive Review: yes/no/back/other feedback per item
Refactor Execution: Apply changes to files with git diffs
QA Validation: Check style, tests, coverage, performance
Reporting: Before/after comparisons, metrics delta, visualizations

Input Requirements

JSON format:

codebase: {files: [{path, lines, deps}], tests: [...], goals: ["improve modularity", "reduce complexity"]}
metrics: Optional baseline (coverage %, perf ms)
Data from Claude tools (Glob/Grep/Bash git status)

Quality: Accurate file lists, recent git state.

Output Formats

checklist.json: Items with id/impact/desc/risks
review_decisions.json: User choices per item
changes.diff: Executed diffs
qa_report.json: Pass/fail metrics
final_report.md: Full comparison (tables, charts)

How to Use

@[code-refactoring-assistant] Analyze current repo for low-impact refactors to improve modularity, then review checklist.

Scripts

refactoring_assistant.py: Orchestrates 6 core classes (ImpactAnalyzer, RefactoringPlanner, ReviewInterface, RefactoringExecutor, QAValidator, RefactoringReporter)

Best Practices

Start with low-impact items
Backup repo (git stash/branch) before execution
Provide test paths for accurate QA
Review high-impact manually
Iterate: Refactor → QA → Report → Re-analyze

Limitations

Relies on Claude tools for real file access (no direct FS write)
Simulated execution in samples; real uses git/Bash tools
Complex deploys need human oversight
Python sandbox limits (no external libs beyond std)