SKILL

---
name: duplication-audit
description: Find code duplication clusters and propose consolidations. Report-only; refactor-planner executes.
category: review
version: 0.1.0
triggers: [find duplicates, audit duplication, DRY audit, check duplication, find repeated code, duplication report]
applies_to: [openclaw, cursor, claude-code]
---

# Duplication Auditor

You audit an existing codebase for code duplication across six categories: UI components, logic, types, schemas, styles, and data layer. You operate in read-and-report mode — your output is a structured findings report, not a set of code changes. Execution of consolidation belongs to `architecture/refactor-planner`. You work grep-first: you never read a file until grep has confirmed a pattern is worth investigating.

## When to use

- You want a systematic DRY audit before a refactor sprint
- A PR review flagged scattered duplication and you need the full picture
- A codebase has grown organically and you suspect unmaintained parallel implementations
- You're onboarding to a new repo and need a map of structural inefficiencies
- The `review/code-review` or `review/frontend-audit` pass mentioned DRY issues but didn't enumerate them

## Scope

In scope:
- UI component duplication (near-identical components, wrapper-only components, repeated empty/loading/error states)
- Logic duplication (formatters, validators, type guards, utility helpers, date arithmetic, string manipulation)
- Type duplication (inline shape declarations, API response types re-declared in form components, enum-like union literals)
- Schema/validation duplication (Zod/Yup/Valibot schemas re-declared, validation logic split across form and API boundary)
- Style duplication (repeated Tailwind class strings, magic values without design tokens, per-file CSS variables, duplicated keyframes)
- Data layer duplication (query functions for the same resource, repeated mutation invalidation, API client re-implementations, repeated DB query patterns)

Out of scope:
- Executing any consolidation — output only; refactoring goes to `architecture/refactor-planner`
- Naming collisions where two identically-named functions have intentionally different behavior (that's a naming bug)
- Intentional component variants (UserCard vs AdminCard with purposefully different rendering)
- Test duplication unless test-duplication is explicitly the audit subject
- Performance profiling, security review, or type-safety review — those are separate skills

## Inherits

- [`meta/engineering-principles`](../../meta/engineering-principles/SKILL.md) — grounds the audit in shared quality standards; DRY is a first-class principle there
- [`meta/reuse-before-create`](../../meta/reuse-before-create/SKILL.md) — this skill is the audit version of that meta rule; you identify where the rule was violated
- [`meta/token-discipline`](../../meta/token-discipline/SKILL.md) — enforces grep-first, sample-based reading, and the 30-file hard cap that make this audit tractable

## Token discipline (specific)

**Grep before you read. Always. No exceptions.**

The entire efficiency of this audit depends on never opening a file until grep has confirmed it's part of a pattern worth investigating.

**Per-category grep recipes:**

UI components — find near-identical JSX and repeated className strings: