What problem does it solve? Academic writers who use AI assistance risk submitting prose that reads as AI-drafted, triggering reviewer suspicion and journal policy issues. This Skill detects the statistically conspicuous patterns of LLM-generated text so authors can strip them before submission. ## Core Features & Use Cases - Ten detection categories: Flags boilerplate transitions, AI-cliché lexicon, em-dash overuse, symmetric paragraph shapes, tricolon abuse, hedging stacking, "not only X but also Y" frames, formulaic openers, hyphenation excess, and sycophantic framing. - Severity-ranked reporting: Each finding is rated HIGH, MED, or LOW with line numbers, current text, and suggested fixes, written to a report file in quality_reports/. - Detect-and-flag only: The Skill never rewrites your prose, preserving authorial voice while making the tells visible. - Use Case: Before submitting a journal article drafted with AI assistance, run the audit on your .tex or .qmd manuscript to find every "Moreover", "delve into", and hedging stack, then edit them out manually. ## Quick Start Ask the assistant to run the humanize audit on your manuscript file, for example by saying "humanize my paper.tex and report only high-severity AI tells".
