AI workflowJun 3, 2026
UX Writing Reviewer: a Claude skill that polices your copy
A Claude Code skill that audits every word on a screen: terminology conflicts, inconsistent verbs, tone drift, weak CTAs, missing states. It scores the screen 0–10, then rewrites the whole thing as one unified system.
Brief
Package the most repetitive part of content design as a reusable Claude skill. Take all the text on a screen, audit it for consistency against the product's language, score it, and hand back a full rewrite. Not notes. The actual copy.
Initial idea
Copy reviews tend to die in one of two ways: they catch issues screen by screen and never see the system, or they end with vague advice like "make it friendlier." The fix for both is structural. Audit across fixed dimensions, compare against other screens when provided, and always end with a complete rewrite plus the language rules that justify it. That's a contract a skill can enforce every time.
Decisions
- 01The audit runs across five fixed dimensions: terminology, actions & verbs, tone of voice, structural patterns, and cross-screen consistency. 'Vendor' vs 'Supplier' on two screens is a bug, not a style choice.
- 02Every piece of text gets classified first: headings, labels, buttons, helper texts, placeholders, empty states, error messages, tooltips, system feedback. You can't audit what you haven't inventoried.
- 03The score is 0–10, broken into four axes: Clarity, Consistency, Actionability, and System Alignment. No single vague grade.
- 04The output's centerpiece is a full rewrite, organized by text category. Not a list of suggestions. If the skill flags it, the skill fixes it.
- 05Every review extracts Language System Rules ('always use Add for creation actions', 'error messages must include guidance'), so each audit leaves the product's language system stronger than it found it.
- 06It also reports what's missing, not just what's wrong: absent error, empty, and success states count as issues.
- 07The quality bar is written in: think system-first not screen-first, clarity over cleverness, no generic suggestions, production-ready output.
Conclusion
The reviewer turns copy QA from a subjective pass into a repeatable audit with a real deliverable: the rewritten screen plus the rules that keep the next screen consistent. It pairs naturally with the Affordance Auditor. One judges whether elements look actionable; the other checks whether the words on them say the right thing.
The skill operates as a senior UX writer, product designer, and design-system guardian in one pass. Give it a screen name, the screen's goal, and every piece of text on it. Optionally include other screens' copy, product language rules, or tone guidelines for cross-comparison.
It classifies every string, audits across the five dimensions, and detects ambiguity, redundancy, weak CTAs, and missing states. Then it scores the screen and rewrites all of it, organized by headings, labels, buttons, helper texts, placeholders, empty states, errors, tooltips, and system feedback.
Download it below, unzip into ~/.claude/skills, and ask Claude to review the copy on any screen. It triggers on requests about UX writing audits, terminology conflicts, tone deviations, or missing UI states.
Download the skill
The complete skill, ready for your setup. Unzip into ~/.claude/skills and it starts triggering in your conversations.