What problem does it solve? Writing tweets, posts, launch copy, or emails that genuinely sound like a specific person is hard to do consistently, and improvising a voice from memory produces generic or off-register drafts. This Skill grounds ghostwriting in a validated voice profile stored in the GBrain knowledge base, so every draft matches the subject's actual fingerprint instead of a guess. ## Core Features & Use Cases - Profile-gated drafting: Loads the subject's validated voice profile (people/<slug>-voice) plus first-party context from the brain, then drafts 2-3 options in a single register with a hard voice-fidelity self-check before anything is shown. - Voice profile builder: When no validated profile exists, drafting hard-stops and the Skill walks a corpus-to-fingerprint build: consent recording, gathering 20+ first-party writing samples spanning 6+ months, extracting six fingerprint dimensions (tells, length, registers, signature moves, banned boilerplate, worldview), and a blind validation test. - Safety contract: Never auto-posts, never invents facts or metrics, never drafts without a validated profile, and keeps private details private. - Use Case: A founder asks you to draft a launch tweet as their CTO. The Skill loads the CTO's validated voice profile, drafts three options in the correct register and length band, runs the self-check against the profile's named tells, and presents the options with verdicts for human approval. ## Quick Start Ask the agent to draft a tweet as a specific person about a topic, for example: draft a tweet as alice-example announcing the acme-example launch.
