voice-referencelisted

# Voice Reference ## When This Skill Applies - Setting up the writing skillset for the first time on a new project or brand - Onboarding a new author whose voice must align with an existing body of work - Formalizing a brand voice that has lived only in editors' heads until now - Investigating suspected tone drift across recent outputs - Providing a calibration anchor so `ai-slop-reviewer` can detect voice deviation rather than only generic slop ## Core Guidance ### 1. What Voice Reference Does Voice reference is a static, user-maintained set of files. It is not a model, not a database, and not a live feed. Two artifacts sit under `references/`: | File | Role | |------|------| | `voice-profile-template.md` | Declared voice: axis placements, signature moves, vocabulary lists | | `writing-samples-scaffold.md` | Demonstrated voice: 2-3 past pieces the author has written | Writing skills and the slop reviewer read these files as context when they run. No network calls, no external embeddings. ### 2. Calibration Protocol Three stages, run once per author or brand and refreshed every 6-12 months. #### Stage 1 — Sample Collection Select 2-3 past pieces. Quality of selection matters more than quantity. | Criterion | Requirement | |-----------|-------------| | Representativeness | Each sample shows the voice in a mode the author actually uses | | Variety | Cover at least two distinct formats (e.g., long-form + short post) | | Recency | At least one sample within the past 1