Skip to content
Back to skills

Code Quality Brixtonpham Claude Config

ASecurity

Code quality improvement: review, refactoring, debugging. Phases: review feedback, systematic refactoring, root cause debugging, verification. Capabilities: SOLID/DRY compliance, code smell detection, complexity reduction, bug investigation, verification gates. Actions: review, refactor, debug, verify, validate code. Keywords: code review, refactor, debug, SOLID, DRY, code smell, bug fix, root cause, verification, technical debt, extract method, test failure, completion claim. Use when: revie...

  • 666 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added February 8, 2026
developmentgotestingdebuggingrefactoringcode-reviewgitperformance

Security analysis

A100/100

Scanned February 12, 2026

npx -y skills add majiayu000/claude-skill-registry --skill code-quality-brixtonpham-claude-config --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Code Quality Brixtonpham Claude Config?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Code Quality Brixtonpham Claude Config
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/majiayu000-code-quality-brixtonpham-claude-config/badge)](https://www.skillsdirectory.com/skills/majiayu000-code-quality-brixtonpham-claude-config)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: code-quality
description: "Code quality improvement: review, refactoring, debugging. Phases: review feedback, systematic refactoring, root cause debugging, verification. Capabilities: SOLID/DRY compliance, code smell detection, complexity reduction, bug investigation, verification gates. Actions: review, refactor, debug, verify, validate code. Keywords: code review, refactor, debug, SOLID, DRY, code smell, bug fix, root cause, verification, technical debt, extract method, test failure, completion claim. Use when: reviewing code changes, improving code quality, fixing bugs, reducing technical debt, validating before merge/commit."
---

# Code Quality

Systematic code improvement through review, refactoring, and debugging with verification gates.

## Module Selection

| Need | Module | Reference |
|------|--------|-----------|
| **Receive Feedback** | Review | `references/code-review-reception.md` |
| **Request Review** | Review | `references/requesting-code-review.md` |
| **Verify Before Claim** | Review | `references/verification-before-completion.md` |
| **Reduce Complexity** | Refactoring | `references/oop-refactoring-catalog.md` |
| **Functional Patterns** | Refactoring | `references/functional-refactoring-patterns.md` |
| **Find Code Smells** | Refactoring | `references/code-smells-reference.md` |
| **Test Strategies** | Refactoring | `references/testing-strategies.md` |
| **Investigate Bugs** | Debugging | `references/systematic-debugging.md` |
| **Trace Root Cause** | Debugging | `references/root-cause-tracing.md` |
| **Add Validation** | Debugging | `references/defense-in-depth.md` |
| **Verify Fix** | Debugging | `references/verification.md` |

---

## Core Principles

**YAGNI, KISS, DRY** - Always honor these.

**Be honest, be brutal, straight to the point, and be concise.**

**Technical correctness over social comfort.** Verify before implementing. Ask before assuming. Evidence before claims.

---

## Code Review

### Receiving Feedback

**Pattern:** READ → UNDERSTAND → VERIFY → EVALUATE → RESPOND → IMPLEMENT

**Rules:**
- ❌ No performative agreement ("You're right!", "Great point!")
- ❌ No implementation before verification
- ✅ Restate requirement, ask questions, push back with reasoning
- ✅ YAGNI check: grep for usage before implementing suggestions

**Sources:**
- Human partner → Implement after understanding
- External reviewer → Verify technically before implementing

### Requesting Review

**When:** After major features, before merge, after complex fixes

**Process:**
1. Get SHAs: `BASE_SHA=$(git rev-parse HEAD~1)`, `HEAD_SHA=$(git rev-parse HEAD)`
2. Dispatch code-reviewer subagent with: WHAT, PLAN, SHAs, DESCRIPTION
3. Fix Critical immediately, Important before proceeding, note Minor

### Verification Gates

**Iron Law:** NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

**Gate:** IDENTIFY → RUN → READ → VERIFY → THEN claim

Skip any step = lying, not verifying.

---

## Refactoring

### Mindset

Simplify relentlessly. Preserve behavior religiously. Measure everything.

Every refactoring: small and safe, tested immediately, measurably better.

### Protocol

1. **Assessment** - Baseline metrics, identify smells, classify debt
2. **Safety Net** - Verify test coverage, add characterization tests
3. **Red-Green-Refactor** - Write failing test, minimal pass, improve design
4. **Pattern Application** - SOLID, design patterns, functional transforms
5. **Validation** - Measure improvements, verify behavior preserved

### Code Smells (5 Categories)

1. **Bloaters:** Long Method, Large Class, Long Parameter List
2. **OO Abusers:** Switch Statements, Temporary Field
3. **Change Preventers:** Divergent Change, Shotgun Surgery
4. **Dispensables:** Duplicate Code, Dead Code, Lazy Class
5. **Couplers:** Feature Envy, Inappropriate Intimacy

### Quick Patterns

**OOP:** Extract Method, Inline, Replace Temp with Query, Guard Clauses
**Functional:** Map/Filter/Reduce, Pure Functions, Composition, Immutability

---

## Debugging

### Core Principle

**NO FIXES WITHOUT ROOT CAUSE INVESTIGATION FIRST**

Random fixes waste time and create new bugs.

### The Four Techniques

**1. Systematic Debugging** (`references/systematic-debugging.md`)
- Phase 1: Root Cause Investigation
- Phase 2: Pattern Analysis
- Phase 3: Hypothesis Testing
- Phase 4: Implementation

**2. Root Cause Tracing** (`references/root-cause-tracing.md`)
Trace backward through call stack to find original trigger.

**3. Defense-in-Depth** (`references/defense-in-depth.md`)
Validate at every layer: Entry → Business logic → Environment → Debug

**4. Verification** (`references/verification.md`)
Run command. Read output. Then claim result.

### Quick Reference

```
Bug → systematic-debugging.md (Phase 1-4)
  Error deep in stack? → root-cause-tracing.md
  Found root cause? → defense-in-depth.md
  About to claim success? → verification.md
```

### Red Flags

Stop if thinking:
- "Quick fix for now, investigate later"
- "Just try changing X and see"
- "It's probably X, let me fix that"
- "Should work now" / "Seems fixed"

---

## References

### Code Review
- `references/code-review-reception.md` - Response protocols
- `references/requesting-code-review.md` - Request process
- `references/verification-before-completion.md` - Verification gates

### Refactoring
- `references/oop-refactoring-catalog.md` - Martin Fowler's patterns
- `references/functional-refactoring-patterns.md` - FP transformations
- `references/code-smells-reference.md` - 23 smells, 5 categories
- `references/testing-strategies.md` - Characterization tests, TDD

### Debugging
- `references/systematic-debugging.md` - Four-phase framework
- `references/root-cause-tracing.md` - Call stack analysis
- `references/defense-in-depth.md` - Multi-layer validation
- `references/verification.md` - Verification protocols

---

## Bottom Line

1. **Review:** Technical rigor over social performance
2. **Refactor:** Small, safe, tested, measurable
3. **Debug:** Root cause first, fix once
4. **Always:** Evidence before claims

Verify. Question. Then implement. Evidence. Then claim.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.