agentleFS
uprinteruprinter/ai-agents-team/.claude/skills/ux-copy/SKILL.md

ux-copy

Skill1 starsChanged 23 days agoMITView on GitHub

Write and review user-facing text for brevity and plain language — every string should read in about 2 seconds, no jargon, no exposed implementation detail. Use when writing or reviewing UI copy — headings, body microcopy, button labels, form fields, error/empty/loading states, confirmation and outcome text. Does not apply to legal/consent text (T&Cs, privacy policy body, GDPR consent language), which may run longer for legal completeness.

15 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 13 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Plain words only: If a non-technical user wouldn't say it themselves, don't write it.

    line 23
  2. Distinctive:

    One idea per sentence: If a string does two things, it's two strings — a heading plus one short line, or a label plus placeholder — not one …

    line 31
  3. Distinctive:

    Read it aloud. Past ~2 seconds (legal text excepted), cut it.

    line 61
  4. Distinctive:

    Before (rewritten in this codebase's history to sound reassuring — but a non-technical user has to read four sentences to find "leave your …

    line 72
  5. Distinctive:

    After (same substance, same required disclosure, each string readable in ~2 seconds):

    line 82

“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

UX Copy — plain, short, no leaks

Most user-facing text fails not because it's wrong, but because it makes the user work to read it: long sentences, hedged qualifiers, and implementation detail that belongs in a commit message, not a UI.

The 2-second rule

Any user-facing string outside legal/consent text must be readable in about 2 seconds — roughly one short sentence, ~12 words, one idea. If a string needs a second sentence to land, that's a sign to cut, not to keep writing. Legal/consent copy (T&Cs, privacy policy body, GDPR consent language) is exempt — precision there matters more than speed — but still prefer short paragraphs and plain words over dense legal-brochure prose.

Rules

  1. Plain words only. If a non-technical user wouldn't say it themselves, don't write it. "Log in" not "authenticate." "Save" not "persist."
  2. Never expose the mechanism. Users don't need to know about queues, webhooks, retries, endpoints, databases, async processing, feature flags, or "our system." Describe the outcome they get, not how it's built or what's happening (or not yet built) behind it.
  3. Name the outcome, not the mechanic, on every button/link/CTA. "Send my report" beats "Submit." "Get updates" beats "Subscribe."
  4. One idea per sentence. If a string does two things, it's two strings — a heading plus one short line, or a label plus placeholder — not one long sentence joined by "and" or "which."
  5. Errors: what happened, what to do — two short clauses, no more. Skip technical causes and blame. "Couldn't save your changes. Try again." not "A network timeout occurred while persisting your changes to the server; please retry your request."
  6. Cut hedges and filler. "Please note that," "in order to," "we just wanted to let you know" — delete them; the sentence loses nothing.
  7. A "not built yet" caveat is not a paragraph. This team's scope-integrity rule (product-owner.md §7, senior-software-engineer.md "Scope Integrity") sometimes requires disclosing that something isn't automatic yet, so users aren't misled — that's a product decision, not a copy-length one. Keep the disclosure, but say it once in a short clause; don't let honesty become an essay. Whether the disclosure itself is still required is for the product owner to decide, not for a copy pass to remove unilaterally.
  8. Cut text the page already says (page level; → RP-57). The 2-second rule covers one string. This rule covers the total. Remove any text that an image, icon, button, card or neighbouring element already says: a caption under a badge that shows the same words, a link row that repeats the buttons above it, a label on a card that is already a link, a facts list that repeats the headline. Keep the accessible name in alt or aria-label. A section stays only if a first-time reader or a hiring reader needs it to act or decide. When a page reads as a wall of text, cut sections. Do not restyle them.

Self-check before shipping any string

  • Read it aloud. Past ~2 seconds (legal text excepted), cut it.
  • Underline any word a non-technical friend wouldn't use in conversation — replace it or delete it.
  • Count sentences in the string. More than one → split or cut.
  • Page check: list every block of text on the page. For each, name what it tells that no image, icon or other element already tells. No answer, delete it.
  • Would a user actually read this, or skim past it? If they'd skim, it's too long regardless of how important the content feels to the team.

Example

Before (rewritten in this codebase's history to sound reassuring — but a non-technical user has to read four sentences to find "leave your email"):

"Observing Planner is new and still changing. Leave your address and I'll write when something worth knowing changes — including if clear-sky alerts start working. One email to confirm, then a few updates a year, written by me. Nothing here is automatic — clear-sky alerts don't exist yet, and this list is how you'd hear if they start working. One click unsubscribes and deletes your address."

After (same substance, same required disclosure, each string readable in ~2 seconds):

Heading: "Get occasional updates" Body: "A few emails a year, written by me — nothing automatic yet." Button: "Email me updates" Fine print (below, smaller): "One click to unsubscribe and delete your address."

The "nothing automatic yet" clause is kept (per rule 7) because it's a scope-integrity disclosure, not filler — it's just said once, briefly, instead of across three sentences.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.