voice-profiler

# Voice Profiler ## Tone & Voice Diagnostic observation skill. output is structured pattern data, not advisory prose. **Hard rule**: profile based on actual samples only, never inferred from assumptions. Each dimension needs ≥3 supporting examples. **NEVER** fabricate dimensions to fill perceived gaps. ## Source & Pairs with - `memory/Voice-Profile.md`. destination + prior-state baseline (read BEFORE re-profiling to track deltas) - `memory/SOUL.md`. voice constitution; profile observations feed into SOUL refinement (rare) - Telegram message history (logs/session.jsonl + plugin storage). primary writing sample source - `outputs/` directories. user-written/edited content samples (NOT generated samples) - Sister skills: `skills-shared/anti-ai-slop/SKILL.md` (different abstraction: voice vs slop), `skills/save/SKILL.md` (captures observations during sessions) - A voice-profile-update script (one you provide). automation helper if scripted ## Example output (dimension entry) ```markdown dimension:: sentence-length pattern:: short-fragment dominant. the operator's messages avg 7-12 words, with ~30% under 5 words. Long sentences appear when explaining technical concepts (avg 20-25 words for those). examples: - "Yes." (1 word, dismissal) - "Go." (1 word, authorization) - "ya fix the broken tasks / the wallet payout, nudged already, still in production they said" (15 words, multi-clause directive. note slash-as-delimiter) confidence:: high ``` ## Conversation cont...