Brand Voice Writer
Teach your agent your brand voice once — get consistent copy forever.
A fill-in-the-blanks skill that encodes your brand voice as executable writing rules: tone sliders, vocabulary do/don't lists, structural habits, and calibrated before/after examples. Every piece of copy your agent writes sounds like you, across channels and team members.
SKILL.md
---
name: brand-voice
description: Our brand voice rules. Use for ALL user-facing copy — marketing, product, docs, social.
---
# Brand Voice: [Your Company]
Apply to every piece of user-facing writing.
## Voice summary
We sound like: [e.g. "a sharp colleague who respects your time — direct, warm, occasionally dry"]
We never sound like: [e.g. "a press release, a hype video, or a terms-of-service document"]
## Tone sliders (0-10)
- Formality: [4] — contractions yes, slang no
- Playfulness: [3] — dry humor OK in blog/social, none in errors or billing
- Technicality: [6] — assume smart readers, define jargon on first use
## Vocabulary
**We say:** [use, build, fast, "here's how"]
**We never say:** [leverage, utilize, seamless, revolutionary, unlock, supercharge, "we're excited to announce"]
**Product terms (exact):** [YourProduct (never "the platform"), workspace (never "org")]
## Structure habits
- Lead with the reader's outcome, not our feature
- Sentences average under 20 words; one idea each
- Headings are statements ("Ship faster with previews"), not labels ("Features")
- CTAs are specific verbs ("Start a project"), never "Learn more" alone
## Calibration examples
❌ "We're excited to announce our revolutionary new AI-powered workflow solution!"
✅ "You can now automate review assignments. Here's how it works."
❌ "Oops! Something went wrong."
✅ "We couldn't save your changes — check your connection and try again. Your draft is safe."
## Per-channel adjustments
- Error messages: formality +2, playfulness 0, always include next step
- Social: playfulness +2, may use fragments
- Docs: technicality +2, second person, present tenseTry it out
Open in Playground →Run this skill on a real model without leaving the page. Every run is saved to your history for this skill.
How to use
- 1Replace every [bracketed] placeholder with your actual voice decisions — the calibration examples matter most.
- 2Save as .claude/skills/brand-voice/SKILL.md (or your agent's rules directory).
- 3Add 3-5 real before/after pairs from your own edits; they teach more than any adjective list.
- 4Update it when your voice evolves — it's your living style guide, now executable.
Examples
Same announcement, voice applied
Input
Draft copy: 'We are thrilled to unveil our seamless new integration experience!'
Output
Rewritten per skill: 'YourProduct now connects to Slack in two clicks. Set it up →' — outcome first, banned words removed, specific CTA.
Pro tips
- The 'we never sound like' line does more work than the 'we sound like' line — be specific about what to avoid.
Frequently asked questions
How is this better than pasting a style guide into each prompt?+
It loads automatically for every writing task, stays in version control with your team's edits, and the do/don't + example format is dramatically more instructable than prose guidelines.
Can non-engineers maintain it?+
Yes — it's a markdown file. Many teams have marketing own the content while engineering owns the file location.