deduplicate-code

name	deduplicate-code
description	Use when the user wants to find and remove code duplication, says "deduplicate", "art-dupl", or wants to reduce duplication to ZERO. This skill is all about JUDGMENT and ZERO HARMFUL duplication. It does NOT mean zero report lines.
metadata	{"tags":"deduplication, quality, cleanup, art-dupl"}

Deduplicate Code

Run

art-dupl --semantic --sort total-tokens -t 5 --html

View the HTML output directly — do not save it as a file. Threshold (-t) counts duplicated statements (not AST nodes); 5 skips one-liner idioms while catching real clones. Use whatever threshold the user specifies. Generated code (sqlc, protobuf, mockgen, stringer, templ) and test-file noise are auto-excluded by default — add --exclude-pattern only when a path genuinely slips past the detector.

Iterate to Zero

For every clone group: read it, then extract, accept, or exclude. Re-run art-dupl after each refactor — keep going until only intentional duplication remains. Run tests after every change. Do not stop at "good enough"; stop when the report is clean or every remaining clone has a defensible reason to exist.

Judgment: Harmful vs Acceptable

Eliminate — the duplication is a real maintenance burden:

• Same logic, different names (a semantic clone)
• Must change in N places to keep behavior consistent
• A clear domain name exists for the shared concept
• Shared setup, fixtures, or boilerplate across multiple packages

Accept — the similarity is intentional or idiomatic:

• Different business rules or domain contexts behind a similar shape
• Table-driven tests, standard assertions (Expect(err).ToNot(HaveOccurred()))
• Generated code (never hand-edit — exclude it, or --include-generated to inspect)
• An abstraction would take more parameters than the duplicated code has lines

When accepting, leave a one-line rationale so the next reader knows it was deliberate.

Done

Zero harmful duplication — not zero report lines. Every remaining clone is there on purpose, every change is verified by tests.