copy-and-ia-reviewlisted

## What It Does Audits user-facing copy and information architecture against `patterns.md`: voice/tone vocabulary, error message format, button label conventions (`Save` vs `Submit` vs `Done`), empty-state copy, navigation labels. Catches inconsistent capitalization, ad-hoc terminology, and IA drift (a label that means different things on different routes). ## Use This When Use for every design-test run. Runs last in the skill sequence — its findings are usually lower severity than token/component/a11y, but they accumulate into a coherent UX signal. ## Inputs - Read `docs/instructions/design-system/patterns.md` for declared voice/tone rules and IA vocabulary. - Read `references/copy-checklist.md`. ## Procedure Extraction and rule-checking are deterministic — drive them with the scripts. The LLM picks severity and writes findings. 1. Run `scripts/extract-user-strings.sh [search-root]` → `<category>\t<file>:<line>\t<string>` rows. Categories: `aria-label | placeholder | title | jsx-text`. This is the inventory of every user-facing string. 2. Run `scripts/check-action-verbs.sh --verbs <a,b,c> --root <dir>` to flag button labels outside the declared action-verb set. The match is case-insensitive; the offending label is reported as-typed. 3. Run `scripts/check-terminology.sh --preferred <Word> --avoid <a,b,c> --root <dir>` once per glossary entry. Each match is a `copy` finding citing the preferred term. 4. IA check: navigation labels and route names form a consistent hier