What is the ux-copy skill?
The ux-copy skill is a slash command skill in Anthropic's knowledge-work-plugins repository. Give it a screen, flow or draft line and it writes or reviews microcopy: button labels, error messages, empty states and confirmation dialogs, returning a recommended line, tone alternatives, rationale and localization notes.
The skill's CTA rules are specific: start with a verb, name the outcome, and write Create account instead of Submit. The same file covers error messages, empty states, confirmation dialogs, tooltips, loading states and onboarding text.
Before writing, it asks for four inputs: the screen or flow, the user's state of mind, the tone and any constraints such as character limits. The reply offers a recommended line plus three alternatives and localization notes.
Patterns it applies
- Error messages follow what happened, why, and how to fix it.
- Empty states follow what this is, why it is empty, and how to start.
- Confirmation dialogs name the action, such as Delete 3 files, and describe the consequence.
- Dialog buttons carry the action label, such as Delete files and Keep files, instead of OK and Cancel.
- Onboarding uses progressive disclosure, one concept at a time.
Five principles
- Clear: no jargon and no ambiguity.
- Concise: the fewest words that carry the full meaning.
- Consistent: the same term for the same thing everywhere.
- Useful: every word helps the user reach their goal.
- Human: written like a helpful person, not a robot.
Where it fits
Good for
- Naming a button, wording a failure or filling an empty screen.
- Reviewing existing copy against a consistent checklist.
- Drafts that a brand reviewer then adjusts. Related: brand guidelines.
Not for
- Long form marketing pages. Related: page CRO and landing page design.
- Final translation. It adds localization notes, but a translator still does the work.
- Testing copy with real users, which the skill does not do.
Use it with UiChemy
Page copy on a WordPress site is yours to edit after UiChemy Figma to WordPress converts a design, so this skill can draft the button and error wording beforehand. It writes text only, not pages. Click Get Started on uichemy.com.
How to install UX Copy Skill: Microcopy, Errors and Empty States
git clone https://github.com/anthropics/knowledge-work-plugins.gitClone the repository
Run
git clone https://github.com/anthropics/knowledge-work-plugins.git. The skill sits in design/skills/ux-copy.Copy the skill folder
Copy design/skills/ux-copy into the folder your agent reads skills from. The pack shows a single SKILL.md in it.
Optional marketplace route
For Claude Code the README documents
claude plugin marketplace add anthropics/knowledge-work-plugins, followed by installing a plugin, after which skills activate.Ask for copy
Type
/ux-copyfollowed by the context, for example the error shown when a payment fails, or paste a line to review.
Check that it loaded
Run /ux-copy with the context empty inbox screen. The reply should list a recommended line, three alternatives with tone, a rationale and localization notes.
Key features
- Write or reviewOne command handles both jobs: drafting new copy for a context, or reviewing a line you already have.
- Context intakeAsks for the screen or feature, what the user is trying to do and feeling, the tone and any character or platform limits.
- CTA rulesStart with a verb, be specific, and match the outcome to the label, with examples such as Save changes and Download report.
- Error message structureUses a three part form of what happened, why, and how to fix it, shown with a declined payment example.
- Empty state structureExplains what the area is, why it is empty and how to start, shown with a no projects yet example.
- Confirmation dialog rulesMakes the action clear, states the consequence and labels buttons with the action itself.
- Tone by situationSuccess is celebratory but restrained, errors are empathetic, warnings are clear and actionable, and neutral text is informative.
- Alternatives tableReturns options A, B and C, each with copy, tone and the situation it suits best.
- Localization notesFlags idioms to avoid, character expansion and cultural context that translators should know.
- Connector hooksWith a knowledge base connected it pulls brand voice guidelines and existing terminology, and with a design tool it reads the screen context and layout limits.
Frequently asked questions
What is the ux-copy skill?
It is a skill in Anthropic's knowledge-work-plugins repository for writing or reviewing interface text. It covers microcopy, error messages, empty states and calls to action, and returns a recommended line with alternatives, a rationale and localization notes.
How do I install ux-copy?
Clone anthropics/knowledge-work-plugins and copy the design/skills/ux-copy folder to where your agent loads skills. The README also documents a marketplace route for Claude Code, using claude plugin marketplace add anthropics/knowledge-work-plugins and then a plugin install.
What should I tell it before it writes?
Give the screen or flow, what the user is trying to do, the tone you want and any limits such as character counts. The skill's tips add that sharing your brand voice, for example professional but warm, helps it match.
Which agents does it work with?
The repository says it is built for Claude Cowork and also compatible with Claude Code. The skill is a plain markdown file, so other agents that read SKILL.md may load it, but the repository does not list them.
How does it write error messages?
It uses a three part structure: what happened, why it happened and how to fix it. The skill's own example is a declined card, which names the bank decision and tells the user to try a different card or call the bank.
Does it connect to a brand style guide?
Only through a connector. If a knowledge base is connected, the skill can pull brand voice guidelines and check existing terminology. Otherwise you paste your guidance into the request, as the skill's tips suggest.
What license does it use?
The repository metadata lists Apache-2.0. The README describes the plugins as markdown and JSON files with no build steps, which you can fork and edit to match your own terminology and voice.