Brand Voice
Applies a configured voice and tone guide to drafts. The voice itself lives in references/brand-voice.md so it can be customized without touching the skill logic.
Workflow
- Read
references/brand-voice.mdin full before editing anything. - Identify the input: a draft pasted by the user, an uploaded file, or a section of an existing document.
- Run the rewrite pass below.
- Output the rewritten version, then a short audit of what changed.
Rewrite pass
For each paragraph in the input:
- Check it against the voice attributes in the reference file.
- Check it against the banned-words list.
- Check it against the preferred-phrasings list.
- Rewrite if any check fails. Leave alone if all pass.
- Preserve the original meaning and key facts. Only change how it is said.
What to preserve
- Numbers, dates, names, quotes (exact)
- Section structure (headings, ordering, list items)
- Technical accuracy (do not soften precise language into vague language)
- Length within plus or minus 15% unless the user asked for shorter or longer
What to change
- Word choice that violates the banned-words list
- Sentence rhythm that conflicts with the stated voice (e.g., long flowing prose when the voice is "punchy")
- Formality level (casual versus formal)
- Filler and hedging that is not in the voice
- Default-AI patterns that creep in (em-dashes, "not just X but Y", tricolon abuse)
Output format
Render in two parts:
## Rewritten
<the full rewritten text>
## Changes
- <one-line summary of the most important change>
- <next change>
- (up to 5 total)
When the reference file is missing or empty
If references/brand-voice.md is missing, or contains only the placeholder template with no user content, do not guess at the voice. Tell the user the reference file needs content first, and offer to interview them with three specific questions:
- What three adjectives describe your voice? (e.g., direct, warm, technical)
- What words or phrases do you ban?
- Paste two sentences in your voice and two that fail.