cm-clean-code

name	cm-clean-code
description	Code hygiene gate — detect and eliminate dead code, duplicates, naming mess, and code smells. TRIZ-powered. Run after features, before PRs, during debt sprints.
token_budget	1200
compressed	true
deprecated	false

Clean Code — Code Hygiene Gate

TL;DR

• Use after a feature lands, before PR review
• Detects: dead code, duplicates, naming smells, dependency drift
• Powered by TRIZ contradictions and Boy Scout rule
• Next: cm-code-review

Code that works is not enough. Code must be CLEAN. Inspired by Clean Code (Robert C. Martin) + Refactoring (Martin Fowler) + TRIZ.

When to Use

ALWAYS when:

• After completing a feature (mandatory hygiene pass before PR)
• After cm-reactor migration (cleanup dead code from migration)
• Before code review (cm-code-review) — clean FIRST, review AFTER
• During technical debt sprints
• When code smells are detected (see Detection section)
• After AI-generated code sessions (AI tends to leave mess)
• When file grows beyond 300 lines

Run automatically after:

• cm-execution completes a task batch
• cm-reactor Phase 5 (post-migration cleanup)
• cm-tdd Refactor phase (Red → Green → Refactor)

Skip when:

• Quick hotfix (patch first, clean later — but schedule the cleanup!)
• Prototype/spike code (will be thrown away)

TRIZ Principles Applied

#	Principle	How Applied
#1	Segmentation	Break large files/functions into focused units
#10	Prior Action	Clean BEFORE it rots — don't wait for tech debt sprint
#6	Universality	One function should serve one purpose (SRP)
#27	Cheap Short-living	Quick small cleanups > expensive large refactors
#2	Taking Out	Extract what doesn't belong — separate concerns

The 7-Point Hygiene Checklist

Run this checklist on every file touched. Each point has auto-detect criteria:

┌───┬──────────────────────┬──────────────────────────────┬────────────────────────┐
│ # │ Check                │ Auto-Detect                  │ Action                 │
├───┼──────────────────────┼──────────────────────────────┼────────────────────────┤
│ 1 │ Dead Code            │ Unused exports, unreachable  │ DELETE — don't comment │
│   │                      │ branches, commented-out code │ out, DELETE            │
├───┼──────────────────────┼──────────────────────────────┼────────────────────────┤
│ 2 │ Unused Imports       │ Import not used in file      │ REMOVE import line     │
├───┼──────────────────────┼──────────────────────────────┼────────────────────────┤
│ 3 │ Magic Numbers        │ Literal numbers in logic     │ EXTRACT to named const │
│   │ & Strings            │ Repeated string literals     │                        │
├───┼──────────────────────┼──────────────────────────────┼────────────────────────┤
│ 4 │ Naming               │ Single-letter vars (not i,j) │ RENAME to describe     │
│   │                      │ Abbreviations, inconsistent  │ intent clearly         │
├───┼──────────────────────┼──────────────────────────────┼────────────────────────┤
│ 5 │ Single Responsibility│ Function does 2+ things      │ EXTRACT into separate  │
│   │ (SRP)                │ Class has 5+ public methods  │ focused units          │
├───┼──────────────────────┼──────────────────────────────┼────────────────────────┤
│ 6 │ DRY (Don't Repeat)   │ Similar code blocks in 2+    │ EXTRACT shared logic   │
│   │                      │ places                       │ into reusable function │
├───┼──────────────────────┼──────────────────────────────┼────────────────────────┤
│ 7 │ Nesting Depth        │ if/for/while nested > 3      │ EXTRACT, early return, │
│   │                      │ levels deep                  │ guard clauses          │
└───┴──────────────────────┴──────────────────────────────┴────────────────────────┘

The Process

Phase 1: SCAN — Detect Code Smells

Goal: Find what's dirty before cleaning.

Automated scan (file-by-file):

For each file modified in current task:

  1. SIZE CHECK:
     IF lines > 300 → FLAG "Large file — consider splitting"
     IF any function > 50 lines → FLAG "Long function — extract methods"
  
  2. IMPORT CHECK:
     Scan imports → cross-reference with usage in file body
     Unused import → FLAG for removal
  
  3. DEAD CODE CHECK:
     Commented-out code blocks → FLAG for deletion
     Functions not called anywhere → FLAG (verify with codeintell)
     Unreachable code after return/throw → FLAG
  
  4. DUPLICATION CHECK:
     Similar code blocks (>5 lines identical/near-identical) → FLAG
     Copy-paste patterns → FLAG
  
  5. NAMING CHECK:
     Single-char variables (except loop vars i,j,k) → FLAG
     Inconsistent casing (camelCase vs snake_case in same file) → FLAG
     Generic names (data, result, temp, item, value, obj) → FLAG
  
  6. COMPLEXITY CHECK:
     Nesting > 3 levels → FLAG
     Function with > 4 parameters → FLAG
     Cyclomatic complexity > 10 → FLAG

Output: Smell Report

## Clean Code Scan: [filename]

| # | Smell | Line | Severity | Auto-fix? |
|---|-------|------|----------|-----------|
| 1 | Unused import: lodash | 3 | Low | ✅ Yes |
| 2 | Magic number: 86400 | 47 | Medium | ✅ Yes |
| 3 | Long function: processData (78 lines) | 23-101 | High | 🔧 Manual |
| 4 | Dead code: commented block | 112-125 | Low | ✅ Yes |

**Total smells: 4 | Auto-fixable: 3 | Manual: 1**

Phase 2: CLEAN — Apply Fixes

Goal: Fix each smell, one at a time, with tests passing between each fix.

Rules (from refactoring.guru + Martin Fowler):

1. Tests first: Ensure tests exist and pass BEFORE cleaning
2. One change at a time: Fix one smell → run tests → commit → next smell
3. Behavior preservation: Clean code MUST NOT change functionality
4. No feature additions: Cleaning and feature work are SEPARATE commits

Fix patterns:

Fix 1: Dead Code & Unused Imports

Action: DELETE (not comment out)
Rationale: Version control is your backup, not comments
Commit: "clean: remove dead code in [file]"

Fix 2: Magic Numbers → Named Constants

Before: if (retryCount > 3) { ... }
         setTimeout(fn, 86400000)

After:   const MAX_RETRIES = 3;
         const ONE_DAY_MS = 86_400_000;
         if (retryCount > MAX_RETRIES) { ... }
         setTimeout(fn, ONE_DAY_MS)

Commit: "clean: extract magic numbers in [file]"

Fix 3: Extract Method (TRIZ #2 Taking Out)

Before: 50+ line function doing validation + calculation + persistence

After:  function processOrder(order) {
          validateOrder(order);
          const total = calculateTotal(order);
          return persistOrder(order, total);
        }

Commit: "clean: extract methods from [function] in [file]"

Fix 4: Reduce Nesting (Guard Clauses)

Before: function foo(x) {
          if (x) {
            if (x.valid) {
              if (x.items.length > 0) {
                // actual logic buried 3 levels deep
              }
            }
          }
        }

After:  function foo(x) {
          if (!x) return;
          if (!x.valid) return;
          if (x.items.length === 0) return;
          // actual logic at top level
        }

Commit: "clean: reduce nesting with guard clauses in [file]"

Fix 5: DRY — Extract Shared Logic

Before: // In file_a.ts
        const formatted = `${date.getFullYear()}-${date.getMonth()+1}-${date.getDate()}`
        
        // In file_b.ts (same pattern)
        const formatted = `${d.getFullYear()}-${d.getMonth()+1}-${d.getDate()}`

After:  // In utils/date.ts
        export function formatDate(date: Date): string { ... }
        
        // Both files import formatDate()

Commit: "clean: extract shared date formatting to utils"

Fix 6: Improve Naming

Before: const d = getData();
        const r = process(d);
        const x = r.filter(i => i.v > 0);

After:  const orders = fetchPendingOrders();
        const processedOrders = applyDiscounts(orders);
        const validOrders = processedOrders.filter(order => order.total > 0);

Commit: "clean: improve naming in [file]"

Phase 3: VERIFY — Confirm Cleanliness

Goal: Ensure cleanup didn't break anything and meets standards.

Verification Checklist:
  □ All tests pass (exact same test results as before)
  □ No functionality changed (behavior preservation)
  □ 7-Point Checklist passes on all modified files
  □ No new code smells introduced
  □ Commit history is clean (one commit per fix type)

Re-run Phase 1 scan on cleaned files:

• If smells remain → iterate Phase 2-3
• If clean → mark task as done

---

SOLID Quick Reference

Apply SOLID during Phase 2 when restructuring:

Principle	Meaning	Quick Test
S — Single Responsibility	One class/function = one reason to change	"Can you describe what it does in one sentence without 'and'?"
O — Open/Closed	Open for extension, closed for modification	"Can you add behavior without changing existing code?"
L — Liskov Substitution	Subtypes must be substitutable	"Can you swap implementations without breaking callers?"
I — Interface Segregation	No client forced to depend on unused methods	"Does every consumer use all methods of this interface?"
D — Dependency Inversion	Depend on abstractions, not concretions	"Do high-level modules import from low-level modules?"

---

Red Flags — STOP

Thought	Reality
"It works, don't touch it"	Working ≠ maintainable. Clean it now or pay 10x later
"I'll clean it up later"	Later never comes. Clean after each feature
"Cleaning wastes time"	10 min cleaning saves 2 hours debugging later
"It's just one small hack"	Hacks compound. Today's hack is tomorrow's tech debt
"AI generated it, must be clean"	AI creates functional code, rarely clean code
"Let me refactor AND add features"	Separate commits. Never mix.
"Comments explain the complex code"	If code needs comments, simplify the CODE

Commit Convention

clean: remove dead code in [file/module]
clean: extract magic numbers in [file]
clean: improve naming in [file]
clean: reduce nesting in [function]
clean: extract [method] from [function]
clean: remove unused imports in [file]
clean: apply DRY — extract shared [logic] to [util]

Integration

Skill	Relationship
cm-tdd	Tests MUST pass before and after cleaning (Red → Green → Refactor)
cm-reactor	Reactor triggers cm-clean-code in Phase 5 (post-migration cleanup)
cm-execution	After task batch → run hygiene pass
cm-code-review	Clean FIRST → review AFTER (reviewers see clean code)
cm-quality-gate	Cleanliness as a quality dimension
cm-debugging	After bug fix → clean the surrounding code
cm-continuity	Record cleanup decisions and patterns learned
cm-codeintell	Verify dead code with call graph before deleting

Lifecycle Position

cm-execution → cm-clean-code → cm-code-review → cm-quality-gate → cm-safe-deploy
   (build)      (hygiene)        (review)          (verify)         (ship)

Karpathy Discipline — Simplicity First

Combat over-engineering at the source, not after the fact:

• No features beyond what was asked. No "while I'm here" additions.
• No abstractions for single-use code. Three similar lines beats a premature interface.
• No flexibility/configurability that wasn't requested. YAGNI is the default.
• No error handling for impossible scenarios. Trust internal callers; validate at boundaries.
• 200 lines that should be 50 → rewrite, don't ship.

Senior-engineer test: "Would a senior call this overcomplicated?" If yes, simplify before commit.

The Bottom Line

Clean code = investment. Dirty code = debt. Pay as you go, or pay with interest later.