dry

name	dry
description	Finds and ranks meaningful duplication and DRY violations in a chosen scope. Use when deciding what repeated logic should be consolidated.
user-invocable	true
disable-model-invocation	false

Dry

Core Contract

Read-only DRY review: find meaningful duplication and propose concrete consolidation. Do not edit files in this skill. Prefer true shared concepts; avoid coupling independent logic because it looks similar. Do not intentionally skip valid issues because there are many; cover the full scoped surface. Follow CLAUDE.md / AGENTS.md on conflict.

Required Inputs

1. Scope (confirm via scope unless already explicit).
2. File types to include/exclude.
3. Pass depth (high-signal default vs exhaustive).

Workflow

1. Confirm scope and repository sharing conventions.
2. Gather candidates from diff or target directories.
3. Classify repetition: true duplication, near-duplication, or incidental similarity.
4. Propose smallest useful consolidation (reuse helper, extract helper, or create small focused file).
5. Score each finding on issue severity and improvement value (expected maintenance/testability/risk payoff).
6. Group findings into High/Medium/Low priority using the combined score.

Safety Rules

• Never edit, create, move, stage, or commit files. This skill only reads and reports. Proposing a new file is a proposal, not an action.
• Do not recommend merging code that merely looks similar; false coupling is a regression. When in doubt, leave it separate and say why.
• Do not propose a sprawling catch-all utils/helpers file. New files must be narrow and named for a single concept.
• Do not over-index on tiny repetitions when larger consolidations exist, but do not suppress real issues due to volume; include all material findings and keep nits clearly low priority.
• Apply the rule of three as a guide, not a law: two occurrences can justify extraction when both must change together, and three near-copies can be fine if they're independent.

Output Style

Report scope/exclusions, then findings grouped into High, Medium, and Low priority. For each finding include file:line evidence, duplication type, issue severity, improvement value, and proposed consolidation. Call out notable non-findings left separate and the highest-payoff next step.