Use this when
learn style, analyze author voice, writing baseline, infer tone from posts.
Process
Use 5-10 representative posts from the same author, brand, or editorial voice (fewer is fine, but warn that the profile will be less stable).
Extract measurable signals from the sample: sentence-length mean and variation, vocabulary richness (type-token ratio), transition-word and passive-voice sentence rates, paragraph-length distribution, first-person usage rate, heading-as-question ratio, and 2- to 3-word signature phrases (stopwords removed). Derive tone descriptors from these measured metrics rather than guessing.
Feed the result into two places: drop a markdown summary into VOICE.md for durable project context, and hand the structured values to writing-persona to seed or update sentence-length, passive-voice, readability, and tone settings.
When Wren drafts against this baseline: keep average sentence length near the learned mean and match its variation unless the user asks for a tighter or looser cadence; preserve signature phrases only when they fit the topic naturally; treat the measured AI-trigger-phrase rate as a ceiling, not a target, when the author rarely uses those terms; use the measured first-person and heading-question rates to judge how personal or question-led the draft should feel.
If a path is missing or a file type is unsupported, skip it and note the gap in the profile rather than failing the whole run. On an empty sample, return zeroed metrics rather than fabricating a baseline.