Skill: Microcopy & Tone of Voice

The words are part of the product. A calm, restrained visual system paired with exclamatory, chirpy, or hedged copy breaks trust faster than any visual bug. Concept2Cure users are regulatory professionals — they want facts, not cheerleading.

This skill pairs with claude-ui-design-principles (principle 9: "Trust Through Restraint") and regulatory-compliance-ux (reviewer-grade language).

Activation

This skill activates when:

• Writing any user-facing string: button label, toast, error message, empty state, confirmation dialog, tooltip, help text, placeholder, alert
• Reviewing a PR that touches UI copy
• Generating assistant / AnA chat response phrasing
• Localizing or rewording existing copy

Core Principles

1. Factual Over Emotional

State what happened. Do not celebrate it, apologize for it, or editorialize.

• ❌ "Nice work! Your document has been saved."
• ❌ "Oops! Something went wrong."
• ✅ "Saved at 14:32."
• ✅ "Couldn't save. Network error."

2. Specific Over Generic

Name the artifact, the action, the reason. Generic copy is a tell that the system doesn't know what's happening.

• ❌ "Action failed."
• ❌ "An error occurred."
• ✅ "Couldn't publish Module 3 — the CTD section is locked by Jane Smith."

3. Active Voice, Present Tense

Users own their actions; the system responds in the present.

• ❌ "The document was saved by the system."
• ❌ "A new version has been created for you."
• ✅ "Saved. Now on v4.2."

4. No Hedging, No Cheerleading

Cut "successfully", "great", "perfect", "awesome", "", "let's", "we'll". Cut "might", "may want to", "possibly" unless the hedge is factually true.

• ❌ "Successfully published!"
• ❌ "Great! You might want to review the output."
• ✅ "Published to IND-2026-0047."
• ✅ "Review the output before signing."

5. No Emoji, No Exclamations

Emoji and ! signal informality. Regulatory users read every word as potential audit trail text. The default is zero of both.

• ❌ "Done! 🎉"
• ❌ "Warning!"
• ✅ "Done."
• ✅ "Warning: this will overwrite the submission draft."

6. Errors Explain + Act

Every error answers: what broke, why, and what the user can do. No dead-end errors.

• ❌ "Error 500"
• ❌ "Something went wrong. Please try again."
• ✅ "Couldn't load the precedent set. The regulatory intelligence service is unavailable. Retry or continue without precedent data."

7. Disabled / Empty States Explain

A disabled button or empty list without context is a dead-end. Say why, and what unblocks it.

• ❌ Empty list: "No items."
• ❌ Disabled button with no tooltip
• ✅ Empty list: "No precedents match this indication. Broaden the search or switch agency."
• ✅ Tooltip on disabled button: "Requires Medical Director role. You are QA Reviewer."

8. Confirmation Dialogs Are Informative, Not Performative

"Are you sure?" adds friction without information. A confirmation dialog should state: the action, the target, the consequence.

• ❌ "Are you sure you want to delete this?"
• ✅ "Delete draft section 3.2.P? This cannot be undone. The section has no published versions."

9. Match Regulatory Vocabulary

Use the language regulators use. Don't soften it for consumers.

• ❌ "Hey, your submission looks good!"
• ❌ "Your paperwork is ready."
• ✅ "IND-2026-0047 is submission-ready. All required CTD sections are locked and signed."

10. AnA (Chat Assistant) Voice

AnA is a regulatory colleague, not a consumer chatbot. Concise. Factual. Offers tradeoffs. Asks when uncertain.

• ❌ "Sure thing! I'd be happy to help you with that! 😊"
• ❌ "I think this might possibly be a good approach, but you should check with your team."
• ✅ "Here's the 510(k) precedent set for your device class. Two have been cleared in the last 18 months — cited below. Want me to draft the Substantial Equivalence section?"

Copy Audit — Quick Pass

Run these checks on any new string before merging:

• No exclamation marks (unless it's a genuine warning like "Warning: destructive action")
• No emoji
• No "successfully", "great", "awesome", "", "let's", "we'll"
• No generic errors — every error names what broke and what to do
• Disabled / empty / loading states explain themselves
• Specific over generic — names the artifact, version, or actor
• Active voice, present tense
• Reads as if a regulatory reviewer could quote it in an audit response

Forbidden Patterns

Forbidden	Use Instead
"Oops!", "Whoops!", "", ""	Factual error + cause + remedy
"Successfully saved!"	"Saved."
"Are you sure?"	"[Action] will [consequence]. [Irreversible?]"
"Something went wrong"	"Couldn't [verb]. [Cause]. [What to do]."
"No items."	"No [thing] match [filter]. [How to change the filter]."
Emoji in regulated UI	Plain text
! at end of confirmation / toast	No terminal punctuation or .
"Please wait..."	<LoadingState label="Loading [thing]" />
AnA: "I'd be happy to help!"	AnA: Direct answer or direct question

Completion Gate

Copy is NOT done until:

• The audit checklist above passes
• Error messages pass the "what broke / why / what do I do" test
• Empty and disabled states explain themselves
• No emoji, no exclamations, no hedging
• Copy reads at the register of the audience (regulatory professional)

When in doubt, read the string aloud in the voice of a calm, competent reviewer. If it sounds chirpy, rewrite.