brand-voicelisted

# Brand voice A voice guide that lives in a document gets read once. A voice guide that lives in a skill gets applied to every sentence the agent writes. ## When to use this Any time you write or edit prose that a reader outside the team will see: page copy, UI strings, emails, changelogs, documentation, social posts. Also when reviewing a pull request that touches user-facing text. Do not apply it to code comments or internal notes unless the project says so. ## How to run it 1. **Read the voice file.** `references/voice.md` in this skill is the template. A project overrides it by putting its own `VOICE.md` at the repo root; that file wins on every rule it defines. 2. **Write the text.** 3. **Check it before you hand it over.** ```bash python3 scripts/check_voice.py <file-or-glob> ``` The checker is mechanical: it catches the rules that can be caught mechanically (forbidden characters, banned phrases, pronoun drift). Everything it cannot catch is on you. 4. **Fix, do not annotate.** A note saying "this uses 'we' but the guide says 'I'" is not the deliverable. The corrected sentence is. ## The rules that get broken most **Pronouns drift under pressure.** A solo operator writing "we offer" is the single most common failure, and it happens on the third paragraph, not the first: the writing slips into corporate register when the topic gets abstract. Reread specifically for this. **Punctuation the guide forbids comes back through rewriting