voice-profilelisted

# Voice profile Match a person's writing voice instead of defaulting to generic AI prose. A *voice profile* is a markdown file of that person's rules — a Do list, a Don't list, a Banned-words list, and before/after pairs — built from their own writing. This skill applies an existing profile, builds a new one, or falls back to baseline expository rules when no profile exists. This skill is the personal-voice layer. The `humanizer` skill removes generic AI tells (em-dash overuse, rule of three, AI vocabulary, sycophancy); run it first or alongside. `voice-profile` adds one person's specific habits and expository concision on top. ## Which mode ```dot digraph { rankdir=LR; "asked to write/revise prose" -> "profile exists?"; "profile exists?" -> "APPLY" [label="yes"]; "profile exists?" -> "DEFAULT" [label="no, just write"]; "asked to capture a voice" -> "BUILD"; } ``` ## Find the profile Look in this order; the first hit wins: 1. `$WRITING_VOICE_PROFILE` (a file path) 2. `.claude/voice_profile.md` in the repo 3. `~/.claude/voice_profile.md` ## Apply a profile 1. Read the profile in full. Do not work from a one-line summary — load the actual rules. 2. Draft or revise the prose, following its Do / Don't lists and avoiding its banned words. 3. Audit the draft before returning it: `voice_audit.py --profile <profile> <file>` (in `scripts/`) flags banned words with line numbers. Read each hit — some banned words have valid uses in context — and fix the real ones.