personal-voicelisted

Make all writing output match the user's own voice, calibrated from 5 real writing samples plus their explicit rules on sentence length, rhythm, forbidden phrases, and tone. Has two modes: a one-time calibration that builds a voice profile from samples, and an application mode where every draft is written and verified against that profile. Use this skill whenever the user asks for writing "in my voice", "like I'd write it", "sound like me", wants to set up or update their voice profile, or whenever drafting anything that will be published or sent under the user's name (emails, posts, articles, bios) once a profile exists. Do NOT use to imitate the voice of any person other than the user, and do not use for content where a house style overrides personal style (legal filings, API docs).
kshyam/opencode-and-claude-skills · ★ 0 · AI & Automation · score 60

Install: claude install-skill kshyam/opencode-and-claude-skills

# Personal Voice Generic AI writing fails people twice: it doesn't sound like them, and everyone who knows them can tell. Style rules alone can't fix this because voice lives in measurable habits nobody thinks to state: how long sentences run, where the rhythm breaks, which words never appear. This skill extracts those habits from real samples, merges them with the rules the user CAN state, and holds every future draft to the result. One boundary, stated once: this skill models the user's own voice at their request. Refuse to build a profile of another person from their writing in order to pass as them. ## Mode selection Check `references/voice-profile.md` in this skill's folder. - **Exists** → application mode: read it and write to it. - **Missing** → calibration mode: build it. Also enter calibration when the user asks to recalibrate or update their profile. ## Calibration mode ### 1. Collect the inputs Ask for **5 samples of the user's real writing**, with this guidance, because sample quality caps profile quality: - 150+ words each where possible; short fragments hide rhythm. - Final writing the user was happy with, not rough drafts, and not text heavily edited by someone else (including AI): that would calibrate to the editor. - Ideally matching the contexts they'll generate in. Five LinkedIn posts produce a LinkedIn voice; if they want emails too, ask for at least one email. If all samples are one genre, say plainly that the profile covers that genre