TDD Refactor Guard
A safety gate that runs before any AI-assisted refactoring. Ensures tests exist to catch regressions. If coverage is insufficient, generates characterization tests first.
When This Skill Activates
Use this skill when the user:
- Says "refactor this" or "clean this up"
- Wants AI to "restructure" or "reorganize" code
- Asks to "extract", "inline", "rename", or "simplify" existing code
- Plans to migrate between patterns (e.g., MVC → MVVM, CoreData → SwiftData)
- Wants AI to "improve" or "modernize" existing code
Why This Guard Exists
Without guard: With guard:
"Claude, refactor this" "Claude, refactor this"
→ AI rewrites code → Guard checks for tests
→ Something breaks → Tests missing → generate them
→ You don't know what → Tests pass → proceed with refactor
→ Debugging nightmare → Tests fail → exact regression found
Process
Step 1: Identify Scope
What code is being refactored?
Ask user: "Which files/classes are you planning to refactor?"
Or detect from context:
Glob: [files mentioned in conversation]
Read: each file to understand scope
Determine:
- [ ] Files involved
- [ ] Public API surface (methods, properties)
- [ ] Dependencies (what calls this code, what it calls)
- [ ] Side effects (network, disk, notifications, UI updates)
Step 2: Check Existing Test Coverage
Grep: "class.*Tests.*XCTestCase|@Suite.*Tests|struct.*Tests" in test targets
Grep: "@Test|func test" that reference the classes being refactored
Coverage Assessment
| Level | Criteria | Action | |-------|----------|--------| | Good (> 80%) | Most public methods have tests for happy + error paths | Proceed with refactoring | | Partial (40-80%) | Some methods tested, missing edge cases | Add characterization tests for uncovered methods | | Minimal (< 40%) | Few or no tests | Generate full characterization test suite first | | None (0%) | No test file exists | STOP — create characterization tests before any refactoring |
Step 3: Evaluate Test Quality
Even if tests exist, check quality:
// ❌ Low-quality test — doesn't actually verify behavior
@Test("test items")
func testItems() {
let vm = ViewModel()
vm.loadItems()
#expect(true) // Always passes, tests nothing
}
// ❌ Tests implementation, not behavior
@Test("calls repository")
func testCallsRepo() {
let vm = ViewModel()
vm.loadItems()
#expect(mockRepo.fetchCallCount == 1) // Brittle
}
// ✅ Tests observable behavior
@Test("loads items into published array")
func testLoadsItems() async {
let vm = ViewModel(repo: MockRepo(items: [.sample]))
await vm.loadItems()
#expect(vm.items.count == 1)
#expect(vm.items.first?.title == "Sample")
}
Quality checklist:
- [ ] Tests verify outputs/state, not implementation details
- [ ] Both happy path and error paths covered
- [ ] Edge cases (empty, nil, boundary) tested
- [ ] Async behavior properly awaited
- [ ] No
#expect(true)or always-passing assertions
Step 4: Generate Missing Tests
If coverage is insufficient, use characterization-test-generator/:
Before refactoring [ClassName], generate characterization tests:
1. Read the source file
2. Identify all public methods
3. Generate tests that capture CURRENT behavior
4. Run and verify all pass
Step 5: Green Light / Red Light
Green Light — Proceed
## Refactor Guard: ✅ PROCEED
**Files**: ItemManager.swift, ItemRepository.swift
**Test coverage**: Good (12 tests covering 8/9 public methods)
**Test quality**: Adequate — tests verify behavior, not implementation
Safe to refactor. Run tests after each change:
`xcodebuild test -scheme YourApp`
Yellow Light — Needs Work
## Refactor Guard: ⚠️ NEEDS TESTS
**Files**: ItemManager.swift, ItemRepository.swift
**Test coverage**: Partial (5 tests covering 4/9 public methods)
**Missing coverage**:
- `ItemManager.sort()` — no test
- `ItemManager.filter(by:)` — no test
- `ItemRepository.sync()` — no test
- Error paths for `save()` and `delete()` — not tested
**Action**: Generate characterization tests for uncovered methods before refactoring.
Use `characterization-test-generator` skill.
Red Light — Stop
## Refactor Guard: 🔴 STOP
**Files**: ItemManager.swift, ItemRepository.swift
**Test coverage**: None (0 tests found)
**Action**: Do NOT refactor until characterization tests exist.
Refactoring without tests is like surgery without monitoring —
you won't know if you killed the patient.
Run `characterization-test-generator` on:
1. ItemManager (7 public methods)
2. ItemRepository (5 public methods)
Then re-run this guard.
Step 6: Post-Refactor Verification
After refactoring is complete:
# Run all tests
xcodebuild test -scheme YourApp
# Check specifically for regressions
xcodebuild test -scheme YourApp \
-only-testing "YourAppTests/ItemManagerCharacterizationTests"
- [ ] All pre-existing tests still pass
- [ ] Characterization tests still pass
- [ ] No new warnings or errors
- [ ] Public API surface unchanged (unless intentionally changed)
Refactoring Safeguards
Safe Refactorings (Low Risk)
These typically don't change behavior:
- Rename variable/method/type
- Extract method/function
- Inline temporary variable
- Move method to another type (keeping same behavior)
- Extract protocol from class
Risky Refactorings (Need Good Coverage)
These can change behavior:
- Change method signature
- Reorder operations
- Replace algorithm
- Merge/split classes
- Change data structure
- Introduce async/await to sync code
- Migration between frameworks (CoreData → SwiftData)
Migration Refactorings (Need Comprehensive Coverage)
Full contract tests recommended:
- Architecture change (MVC → MVVM)
- Framework migration (UIKit → SwiftUI)
- Storage migration (UserDefaults → SwiftData)
- Concurrency model change (GCD → async/await)
Output Format
## Refactor Guard Report
**Scope**: [Files/classes to be refactored]
**Verdict**: ✅ PROCEED / ⚠️ NEEDS TESTS / 🔴 STOP
### Coverage Summary
| Class | Public Methods | Tests | Coverage | Verdict |
|-------|---------------|-------|----------|---------|
| ItemManager | 9 | 12 | 89% | ✅ |
| ItemRepository | 5 | 2 | 40% | ⚠️ |
### Missing Coverage
- [ ] `ItemRepository.sync()` — needs characterization test
- [ ] Error path for `ItemRepository.delete()` — untested
### Recommendations
1. [Action item]
2. [Action item]
References
testing/characterization-test-generator/— generates tests this guard may requiretesting/tdd-feature/— for building new code during refactor- Martin Fowler, Refactoring: Improving the Design of Existing Code