Аудит академического текста на AI-признаки

documentation

humanize is a Claude Code skill that performs a read-only audit of `.tex`, `.qmd`, and `.md` files for AI-voice tells in academic prose. It scans for boilerplate transitions (Moreover, Furthermore, It is important to note that), AI-cliché lexicon (delve, tapestry, robust framework, navigate the complexities), em-dash overuse, symmetric paragraph shapes repeating a topic→examples→summary cadence, tricolon abuse, and stacked epistemic hedges. Each finding is reported with its category and severity level (LOW, MED, or HIGH) — the skill produces a structured report and does not rewrite anything, leaving all edits to the author. It is intended for use before journal submission, preprint posting, or after any AI-assisted drafting session when stripping the model's lexical fingerprint matters.