UX CopyTone of VoiceAnthropicAI Agent Skill
ux-copy
Anthropic's official UX copy standard: concise, consistent microcopy that always gives the user a clear next action.
Install Options
Install into Claude Code, Cursor, Codex, or Antigravity with:
$ npx skills add anthropics/knowledge-work-plugins@ux-copy
Real Work Situation Solved
Calibrate AI response voice and tone to match your brand identity
“The AI swings between overly formal, excessively verbose, and too casual - how do you enforce a consistent tone?”
5-Step Action Framework
- 1Define a 3-pair Tone of Voice rubric ('Is X, but Not Y' - e.g. Helpful but Not Verbose, Empathetic but Not Over-apologetic).
- 2Set strict length boundaries (e.g. max 3 sentences for direct answers, max 150 words for explanations).
- 3Provide 5 curated Good vs. Bad input/output pairs directly in the system prompt.
- 4Maintain a negative lexicon: ban generic corporate cliches and robotic filler phrases.
- 5Stress-test against 10 frustrated user scenarios to verify measured empathy without excessive apologies.
Before
"I am terribly sorry for this immense inconvenience, please allow me to assist you with... (3 paragraphs)"
After
"Order #1234 is scheduled for redelivery by 5 PM today. Would you like to update your delivery address?"
Related Course Lesson
AI Literacy • Lesson 5
SKILL.md Source Instructions
SKILL.md • Read-only preview# UX Copy Skill Write state feedback and error messages in active voice, keep every message under a strict length cap, and always pair a problem statement with a next action.