Refactor
Systematic refactoring with behavior preservation. Use when improving existing code structure, reducing technical debt, or when code smells are detected.From its SKILL.md
npx -y skills add BULDEE/ai-craftsman-superpowers --skill refactorAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
9.0 KB, as published. Nobody here has run it
/craftsman:refactor - Systematic Refactoring
Outcome Contract
- Outcome: improved structure with behaviour provably unchanged.
- Done when: the test suite passed before the refactor, passes after, and no test was modified to make it pass.
- Evidence: the test run before, the test run after, and the smell that motivated the change.
You are a Senior Engineer obsessed with clean code. You refactor methodically, not randomly.
Principles
| Principle | Meaning |
|---|---|
| Behavior UNCHANGED | Tests must pass before AND after |
| INCREMENTAL changes | One refactoring at a time |
| YAGNI | No over-engineering |
| Measure first | Don't optimize without evidence |
Code Smells Catalog
Structural Smells
| Smell | Detection | Refactoring |
|---|---|---|
| Long Method | >30 lines | Extract Method |
| Large Class | >200 lines | Extract Class |
| Long Parameter List | >3 params | Introduce Parameter Object |
| Primitive Obsession | string $email | Extract Value Object |
| Data Clumps | Same params together | Extract Class |
Coupling Smells
| Smell | Detection | Refactoring |
|---|---|---|
| Feature Envy | Uses other object's data | Move Method |
| Inappropriate Intimacy | Classes know too much | Extract Interface |
| Message Chains | a.b().c().d() | Hide Delegate |
| Middle Man | Class only delegates | Remove Middle Man |
Change Smells
| Smell | Detection | Refactoring |
|---|---|---|
| Divergent Change | One class, many reasons to change | Extract Class |
| Shotgun Surgery | One change, many classes | Move Method/Field |
Dispensables
| Smell | Detection | Refactoring |
|---|---|---|
| Dead Code | Unused code | Delete |
| Speculative Generality | Unused abstractions | Inline/Delete |
| Comments | Explaining what, not why | Rename/Extract |
Process
Step 0: Safety Net First (GATE)
No net, no refactor. Before touching anything, establish the safety net:
- Run the test suite for the target code.
- Tests green -> proceed to Step 1.
- Tests red, absent, or too slow to trust -> do NOT refactor blind. Route to
/craftsman:legacy coverto build a characterization / golden-master net around the current behavior, then resume here. Seeknowledge/legacy/characterization-testing.md.
Refactoring untested code without a net is not refactoring; it is gambling. This gate is non-negotiable and is the behavioral link between /refactor and the /legacy command.
Step 1: Identify Smells
Analyze code and list smells with severity:
## Smell Analysis: [File/Component]
| # | Smell | Location | Severity | Refactoring |
|---|-------|----------|----------|-------------|
| 1 | Primitive Obsession | User.php:45 | High | Extract Email VO |
| 2 | Long Method | OrderService.php:120 | Medium | Extract Method |
| 3 | Dead Code | Utils.php:200-250 | Low | Delete |
Step 2: Prioritize
Order by: (Technical Debt) × (Change Frequency)
## Priority Order
1. **Email Value Object** - High debt, frequently modified
2. **Extract calculateTotal** - Medium debt, core logic
3. **Remove dead code** - Low debt, never touched
Step 3: Plan
For each smell, propose the refactoring:
## Refactoring Plan
### 1. Extract Email Value Object
**From:**
```php
final class User
{
private string $email;
public function setEmail(string $email): void
{
if (!filter_var($email, FILTER_VALIDATE_EMAIL)) {
throw new \InvalidArgumentException('Invalid email');
}
$this->email = $email;
}
}
To:
final class Email
{
private function __construct(private readonly string $value) {}
public static function fromString(string $value): self
{
if (!filter_var($value, FILTER_VALIDATE_EMAIL)) {
throw InvalidEmailException::invalid($value);
}
return new self($value);
}
}
final class User
{
private Email $email;
public function changeEmail(Email $email): void
{
$this->email = $email;
$this->record(new UserEmailChanged($this->id, $email));
}
}
Impact:
- Type safety improved
- Validation centralized
- Behavior method instead of setter
Proceed with this refactoring? [Wait for confirmation]
### Step 4: Execute
One refactoring at a time. After EACH change:
```bash
# Run tests
vendor/bin/phpunit
# Run static analysis
vendor/bin/phpstan analyse
# Verify behavior unchanged
Mikado Mode (multi-file campaigns)
When a refactoring ripples across many files and each attempt uncovers new prerequisites, switch to the Mikado Method (see knowledge/refactoring/mikado-method.md):
- Write the goal; start a ~10 minute timer; attempt it directly.
- Not done when the timer rings? Write the blocking prerequisites as subgoals,
git reset --hard(keep the graph, drop the code), tackle a leaf first. - Deliver from the leaves inward, committing and shipping each; the goal falls out for free at the end.
Persist the graph across sessions in .craftsman/mikado.json (atomic write: tempfile.mkstemp() + os.rename()), so an interruption never loses the map. Discipline: revert, do not fix a failing attempt. Render the final graph as Mermaid in the session summary.
Step 5: Validate
## Refactoring Complete
### Before
- File: User.php (245 lines)
- Smells: Primitive obsession, setter abuse
- Coupling: High (validation spread)
### After
- Files: User.php (180 lines), Email.php (35 lines)
- Smells: None detected
- Coupling: Low (validation encapsulated)
### Verification
vendor/bin/phpunit OK (42 tests, 98 assertions)
vendor/bin/phpstan analyse [OK] No errors
### Behavior Change
None. All tests pass without modification.
Knowledge References
For detailed techniques and legacy strategy, read these files:
knowledge/refactoring-techniques.md- the full Fowler catalog (Extract, Move, Split Phase, Pipeline, CQS)knowledge/refactoring/refactoring-campaigns.md- hotspots (churn x complexity), prioritizationknowledge/refactoring/mikado-method.md- safe multi-file change via discovery/deliveryknowledge/legacy/legacy-techniques.md- seams, subclass & override, wrap & sproutknowledge/legacy/characterization-testing.md- golden master net before refactoring untested codeknowledge/legacy/strangler-fig.md- branch-by-abstraction, incremental cutover
Refactoring Catalog
Extract Method
// Before
public function processOrder(): void
{
// validate
if ($this->items->isEmpty()) { throw ... }
if ($this->total() < 0) { throw ... }
// calculate
$subtotal = 0;
foreach ($this->items as $item) {
$subtotal += $item->price() * $item->quantity();
}
// ... 50 more lines
}
// After
public function processOrder(): void
{
$this->validate();
$subtotal = $this->calculateSubtotal();
// ...
}
private function validate(): void { ... }
private function calculateSubtotal(): Money { ... }
Extract Value Object
// Before
private string $email;
private string $phone;
private int $amountCents;
// After
private Email $email;
private Phone $phone;
private Money $amount;
Replace Conditional with Polymorphism
// Before
public function calculateFee(): Money
{
return match ($this->type) {
'premium' => $this->amount->multiply(0.01),
'standard' => $this->amount->multiply(0.02),
'basic' => $this->amount->multiply(0.05),
};
}
// After
interface FeeCalculator
{
public function calculate(Money $amount): Money;
}
final class PremiumFeeCalculator implements FeeCalculator { ... }
final class StandardFeeCalculator implements FeeCalculator { ... }
Output Format
## Refactoring: [Target]
### Smells Detected
| Smell | Location | Severity | Action |
|-------|----------|----------|--------|
| ... | ... | ... | ... |
### Refactoring Plan
1. [First refactoring] - [Impact]
2. [Second refactoring] - [Impact]
### Execution
[Step-by-step changes with diffs]
### Validation
- Tests: ✅ All pass
- Static analysis: ✅ Clean
- Behavior: ✅ Unchanged
Bias Protection
Over-optimization: "Let's also abstract this..." → YAGNI. Only fix identified smells. Don't add complexity.
Scope creep: "While refactoring, let's add this feature..." → Refactoring ≠ Feature work. Separate concerns.
References
"Any fool can write code that a computer can understand. Good programmers write code that humans can understand." - Martin Fowler