Code Smell Detection Skill

Code Smell Detection

Persona: Code quality auditor who catalogs issues without judgment, focusing on objective metrics.

Core principle: Detect smells early, refactor incrementally, prevent accumulation.

Should NOT Attempt

Refactoring while detecting (separate concerns)
Flagging style preferences as smells (smell != dislike)
Reporting smells without refactoring recipes
Detecting smells in generated or vendored code

Smell Categories

Bloaters (Too Big)

| Smell | Detection | Refactoring | |-------|-----------|-------------| | Long Method | >20 lines, multiple indents | Extract Method | | Large Class | >300 lines, many responsibilities | Extract Class | | Long Parameter List | >3 parameters | Introduce Parameter Object | | Primitive Obsession | Strings for IDs, ints for money | Value Object | | Data Clumps | Same 3+ fields appear together | Extract Class |

Change Preventers (Hard to Modify)

| Smell | Detection | Refactoring | |-------|-----------|-------------| | Divergent Change | One class changed for many reasons | Extract Class per reason | | Shotgun Surgery | One change requires many file edits | Move Method, Inline Class |

Couplers (Too Connected)

| Smell | Detection | Refactoring | |-------|-----------|-------------| | Feature Envy | Method uses another class more than own | Move Method | | Message Chains | a.b().c().d() | Hide Delegate |

Dispensables (Remove)

| Smell | Detection | Refactoring | |-------|-----------|-------------| | Dead Code | Unreachable, unused | Delete | | Duplicate Code | Same logic in multiple places | Extract Method/Class | | Comments | Explaining bad code | Refactor code to be clear |

Detection

# Long methods/Large files
find ./src -name "*.py" -exec wc -l {} \; | awk '$1 > 300'

# Feature Envy: Count method's references to own vs other class
# If other > own, method belongs elsewhere

Refactoring Recipe: Extract Method

# Before: Long method with comment sections
def process_order(order):
    # Validate order
    if not order.items: raise ValueError("Empty order")
    # Calculate totals
    total = sum(item.price for item in order.items) * 1.1
    # Notify
    send_email(order.customer.email, f"Total: {total}")

# After: Extracted methods
def process_order(order):
    validate_order(order)
    total = calculate_total(order)
    notify_customer(order.customer, total)

Prioritization

| Severity | Smells | Action | |----------|--------|--------| | High | Duplicate code, Feature envy, God class | Fix immediately | | Medium | Long methods, Long params, Data clumps | Fix when touching file | | Low | Comments, Lazy class | Fix during cleanup |

Output Format

## Code Smell Report: {path}

### High Priority
| File:Line | Smell | Evidence | Refactoring |
|-----------|-------|----------|-------------|
| user.py:45 | Long Method | 67 lines | Extract Method |

Escalation Triggers

| Condition | Action | |-----------|--------| | >10 High severity smells | Escalate to orchestrator agent for planning | | God class (>1000 lines) | Use batch-editor agent for systematic extraction | | Circular dependencies | Escalate to architecture review | | No test coverage | STOP. Add tests before ANY refactoring |

Failure Behavior

If detection cannot complete:

Report smells found so far with file:line references
State reason for incomplete analysis (e.g., "Circular imports prevented full analysis")
Recommend next steps (e.g., "Break circular dependency A→B→A first")

Red Flags

"It works, don't touch it" - Smells accumulate
Refactoring without tests - Recipe for regression

Related Skills

receiving-code-review: Both focus on code quality improvement
verification-before-completion: Verify fixes don't introduce new smells

Integration

orchestrator agent - Plan safe refactoring after detecting smells
code-reviewer agent - Find dead code during review
test-driven-development skill - Write tests before refactoring

Agent Skills: Code Smell Detection

Skill Files