brand-voice

# Brand Voice This skill complements the `brand-voice` plugin. The plugin runs as **post_receive middleware** after the assistant finishes a turn: it inspects the final text against banned phrases, banned patterns, required phrases, and an optional LLM classifier, then either allows, rewrites, or blocks the response. The plugin is a safety net. This skill is the up-front discipline that keeps the safety net from firing. ## When to use Activate this guidance whenever the response will reach an external audience: customers, partners, stakeholders, public comments, marketing copy, sales follow-ups, support replies. It is also a sensible default for internal messages on user-facing channels (Slack to a customer team, email to a vendor). ## Working rules 1. **Read the configured voice before drafting.** Run `/brand-voice` (the plugin command) to see the active mode, banned phrases, banned patterns, and required phrases. Treat that output as authoritative. If no voice is configured, write in clear neutral business prose and stop second-guessing tone. 2. **Lead with substance, not voice.** Brand voice is the polish on top of a correct, concrete answer. Never trade accuracy for tone. 3. **Avoid the banned set.** Do not use banned phrases or anything that matches the banned regex patterns. If a banned phrase is the most natural word, find a synonym or restructure the sentence — do not simply alias the banned word with a similar one. 4. **Honor required ph...