Agent skill
characterization-test-generator
Generates tests that capture current behavior of existing code before refactoring. Use when you need a safety net before AI-assisted refactoring or modifying legacy code.
Install this agent skill to your Project
npx add-skill https://github.com/rshankras/claude-code-apple-skills/tree/main/skills/testing/characterization-test-generator
SKILL.md
Characterization Test Generator
Generate tests that document what existing code actually does — not what it should do. These tests capture current behavior so you can refactor with confidence, especially when using AI to modify code.
When This Skill Activates
Use this skill when the user:
- Says "I need to refactor this" or "help me refactor"
- Wants to "add tests before changing code"
- Asks for "characterization tests" or "golden master tests"
- Says "I want to safely modify this class/module"
- Mentions "legacy code" or "untested code"
- Wants AI to refactor but is worried about breaking things
Why This Matters for AI-Generated Code
AI refactoring is powerful but risky:
- AI doesn't remember why code works a certain way
- AI may "improve" code and break subtle behavior
- Characterization tests freeze current behavior as a regression suite
- If any test fails after refactoring, you know something changed
Process
Phase 1: Discover Code Under Test
Glob: **/*.swift (in the target module)
Grep: "class |struct |enum |protocol |func " (public API surface)
Identify:
- Public types and their public methods
- Input types and return types
- Side effects (network, disk, UserDefaults, notifications)
- Dependencies (injected or created internally)
- State mutations (properties that change)
Phase 2: Classify Behavior
For each public method, classify:
| Category | Example | Test Strategy |
|---|---|---|
| Pure computation | calculate(a, b) -> c |
Input/output pairs |
| State mutation | addItem(_:) modifies array |
Before/after state checks |
| Async operation | fetchData() async -> [Item] |
Mock dependencies, verify results |
| Side effect | save() writes to disk |
Verify mock was called |
| Event emission | delegate?.didUpdate() |
Capture delegate calls |
| Error path | Throws on invalid input | Verify correct error type |
Phase 3: Generate Tests
Test Naming Convention
Characterization tests should be clearly labeled:
import Testing
@testable import YourApp
@Suite("Characterization: ItemManager")
struct ItemManagerCharacterizationTests {
// Tests document CURRENT behavior, not ideal behavior
}
Template: Pure Computation
@Test("current behavior: calculates total with tax")
func calculatesTotal() {
let calculator = PriceCalculator()
let result = calculator.total(subtotal: 100.0, taxRate: 0.08)
// Document actual behavior — even if the rounding seems wrong
#expect(result == 108.0)
}
Template: State Mutation
@Test("current behavior: addItem appends and sorts by date")
func addItemSortsbyDate() {
let manager = ItemManager()
let older = Item(title: "A", date: .distantPast)
let newer = Item(title: "B", date: .now)
manager.addItem(older)
manager.addItem(newer)
// Document actual ordering behavior
#expect(manager.items.first?.title == "A")
#expect(manager.items.last?.title == "B")
#expect(manager.items.count == 2)
}
Template: Async with Dependencies
@Test("current behavior: loadItems returns cached data when offline")
func loadItemsOffline() async throws {
let mockNetwork = MockNetworkClient(shouldFail: true)
let mockCache = MockCache(items: [Item.sample])
let service = ItemService(network: mockNetwork, cache: mockCache)
let items = try await service.loadItems()
// Document: falls back to cache when network fails
#expect(items.count == 1)
#expect(items.first?.title == Item.sample.title)
}
Template: Side Effects
@Test("current behavior: save writes to UserDefaults")
func saveWritesToDefaults() {
let defaults = MockUserDefaults()
let settings = SettingsManager(defaults: defaults)
settings.setTheme(.dark)
settings.save()
// Document: saves as string, not as enum raw value
#expect(defaults.lastSetValue as? String == "dark")
#expect(defaults.lastSetKey == "app_theme")
}
Template: Error Paths
@Test("current behavior: throws on empty title")
func throwsOnEmptyTitle() {
let validator = ItemValidator()
#expect(throws: ValidationError.emptyField("title")) {
try validator.validate(Item(title: "", date: .now))
}
}
Template: Edge Cases
@Test("current behavior: handles nil optional gracefully")
func handlesNilOptional() {
let parser = DataParser()
let result = parser.parse(data: nil)
// Document: returns empty array on nil, doesn't crash
#expect(result.isEmpty)
}
@Test("current behavior: handles empty collection")
func handlesEmptyCollection() {
let aggregator = StatsAggregator()
let stats = aggregator.compute(values: [])
// Document: returns zeroes, not NaN or crash
#expect(stats.average == 0.0)
#expect(stats.count == 0)
}
Phase 4: Fill in Actual Values
This is the key step. For each test:
- Read the source code to understand what the method actually returns
- Trace the logic for the given inputs
- Write the assertion with the actual value, even if it seems wrong
// ❌ Wrong — this is what you WANT it to do
#expect(result == expectedCorrectValue)
// ✅ Right — this is what it ACTUALLY does
#expect(result == actualCurrentValue) // Note: off-by-one, but current behavior
Add comments for surprising behavior:
// CHARACTERIZATION: This returns 11 not 10 due to inclusive range.
// Don't "fix" this until intentionally changing behavior.
#expect(range.count == 11)
Phase 5: Verify and Lock
- Run all characterization tests — they must ALL pass
- If any fail, adjust assertions to match actual behavior
- Tag them so they're easy to find later:
@Suite("Characterization: ItemManager")
@Tag(.characterization)
struct ItemManagerCharacterizationTests { ... }
// Define the tag
extension Tag {
@Tag static var characterization: Self
}
- Run selectively:
# Run only characterization tests
xcodebuild test -scheme YourApp \
-only-testing "YourAppTests/ItemManagerCharacterizationTests"
Output Format
## Characterization Tests Generated
**Module**: [Module name]
**Classes tested**: [List]
**Tests generated**: [Count]
### Coverage Summary
| Class | Methods Covered | Edge Cases | Notes |
|-------|----------------|------------|-------|
| ItemManager | 5/7 | 3 | 2 private methods skipped |
| PriceCalculator | 3/3 | 2 | All public API covered |
### Files Created
- `Tests/CharacterizationTests/ItemManagerCharacterizationTests.swift`
- `Tests/CharacterizationTests/PriceCalculatorCharacterizationTests.swift`
### Surprising Behaviors Found
- `PriceCalculator.total()` rounds DOWN, not to nearest cent
- `ItemManager.sort()` is unstable — equal dates may reorder
### Ready to Refactor
All [X] characterization tests passing. Safe to refactor with AI.
Run `xcodebuild test` after each change to verify no behavior changed.
When NOT to Use This
- Code is already well-tested (check coverage first)
- You're deleting the code entirely (no need to characterize)
- The code is trivially simple (single-line computed properties)
- You want to change the behavior (use
tdd-bug-fixinstead)
References
- Michael Feathers, Working Effectively with Legacy Code — coined "characterization test"
generators/test-generator/— for standard test generation (not characterization)testing/tdd-refactor-guard/— pre-refactor checklist that uses these tests
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
legal
Legal document generation and compliance guidance for indie Apple developers. Covers privacy policies, terms of service, EULAs, GDPR/CCPA/DPDP compliance, and Apple App Store legal requirements. Use when user needs legal documents or compliance guidance.
privacy-policy
Generate privacy policies, terms of service, and EULAs for Apple platform apps. Detects data collection patterns, third-party SDKs, and generates region-specific legal documents with Apple Privacy Nutrition Label mapping. Use when user needs legal documents or data collection disclosure for App Store submission.
ios-development
Comprehensive iOS development guidance including Swift best practices, SwiftUI patterns, UI/UX review against HIG, and app planning. Use for iOS code review, best practices, accessibility audits, or planning new iOS apps.
assistive-access
Assistive Access implementation for cognitive accessibility including simplified scenes, navigation icons, runtime detection, and design principles. Use when optimizing apps for Assistive Access mode.
ui-review
Review SwiftUI code for iOS/watchOS Human Interface Guidelines compliance, font usage, Dynamic Type support, and accessibility. Use when user mentions UI review, HIG, accessibility audit, font checks, or wants to verify interface design against Apple standards.
navigation-patterns
SwiftUI navigation architecture patterns including NavigationStack, NavigationSplitView, TabView, programmatic navigation, and custom transitions. Use when reviewing or building navigation, fixing navigation bugs, or architecting app flow.
Didn't find tool you were looking for?