| name | voice-os |
| description | Always-on voice alignment spine — load before ANY writing task that carries Farrice's own name: LinkedIn posts, Substack/Parallax editions, Notes/Threads, emails, DMs, replies, memos, comments, docs, drafts, brand copy. Answers "does this sound like me" before craft ever runs. Owns a 4-mode fidelity dial (MIRROR/BLEND/STRETCH/OFF), the identity + voice-law floor, the banned-moves gate, and the felt-verdict calibration loop. Trigger words: voice, tone, style, sound like me, in my voice, my voice, does this sound like Farrice, voice check, voice pass, calibrate the voice, on-brand, personal voice, writing voice, ghostwrite my voice. Do NOT use for: client deliverables where Voice OS is explicitly OFF (Jen, Andrea, any client brand — their own voice docs govern), or for craft/structure work with no identity stakes (pure research synthesis, deterministic scripts). |
| expert | Voice OS (spine over VOICE-CARD.md — no craft of its own) |
| domain | Cross-cutting Identity x Voice Fidelity x Calibration |
| version | 1.0 |
| format | spine |
| tier | system |
Voice OS
A piece can pass every craft gate — structure sound, hook sharp, no slop — and still not sound like Farrice. Voice OS is the layer that catches that failure before delivery. It owns no craft. It owns identity: who Farrice is, how he actually talks, what he never says, and how far a piece is allowed to drift from his fingerprint before it stops being his. Every writing task that carries his name loads this spine first, then hands craft authority to the expert skill that actually writes.
What Voice OS Is
Voice OS is the always-on voice alignment layer sitting between intent and craft. It is not a writer, not a persona, not a style guide to quote back at a reader — it is grounding an expert absorbs before drafting. Its one job: keep every piece inside Farrice's identity spine and voice law while craft experts (writers-room, /parallax, /ghostwrite, How-I-Write OS) do the actual writing.
Load order (every session that touches Farrice's own voice):
- This SKILL.md — orientation, the Dial, the decision table.
_active/farrice-brand/voice/VOICE-CARD.md — always. The canonical card: §1 Identity Spine, §2 Voice Law, §3 Stylometrics, §4 Channel Registers, §5 Banned Moves, §6 Calibration Bank, §7 The Dial, §8 Loop Protocol.- The relevant §4 channel register only — LinkedIn, Substack edition, Notes/Threads, email/DM, or client-facing docs. Don't load registers you're not writing for.
genius.md (Tier 2, craft/complex work) — the texture layer: signature-move mechanics traced to source, the two-stage rhythm behind the stylometrics, the recognition test, and the sourced anti-pattern bank. Load this whenever the deliverable is more than a mechanical mode-check — i.e. any time craft judgment is needed, not just gate enforcement.
If VOICE-CARD.md doesn't exist yet, say so and stop — do not improvise a voice from memory. Run python3 execution/voice_ratchet.py status to confirm state before assuming the card is missing; the ratchet script reports "not found" cleanly if it truly isn't compiled.
The Dial (voice fidelity modes)
Four modes, one hard floor that never lifts. The dial answers "how much of this piece is Farrice's literal fingerprint vs. an elevated/explored version of it."
| Mode | Fidelity | When | What's fixed | What's free |
|---|
| MIRROR | 95% | Personal comms, DMs, replies, anything conversational under his name | Everything — full fingerprint, including imperfect texture (run-ons, real speech rhythm) |
