drydry

Invoke quick mode by naming or pasting the two candidates. The Skill reads both, writes a drift hypothesis describing what breaks if they diverge, produces a runnable grep or rg verifier command, and returns a verdict: converge, needs design, or not a duplicate.

Pass a directory, package, or project root with the audit keyword. The Skill formulates a six-to-ten item checklist from the actual code, runs a contrarian review of that checklist, sweeps for hits, triages findings, and writes findings and checklist markdown artefacts to the project root.

Quick mode answers a scoped duplicate question inline with no artefact, suited to one or two files or snippets. Audit mode sweeps an entire scope, applies the full eight-chapter discipline including contrarian passes, and produces persistent findings documents.

Yes. The checklist formulation prompts cover user-facing copy variants, parallel UX surfaces, and design-system patterns alongside code. Domain hints such as Swift, Rails, prose, or design tokens steer the audit toward the relevant duplication shapes.

The verifier-burden discipline requires every finding to include a runnable grep, rg, or find command that re-confirms the duplication without the LLM. Hits that cannot be independently re-verified are dropped to prevent hallucinated duplicates.

Use drydry when asking whether something specific already exists. Use drydry:parity-audit when the question is whether all entry points of the same action still behave identically, such as sweeping every path that triggers the same underlying operation.