I've been running this skill against real drafts and kept hitting the same gaps, so this commit fills them in.
The biggest addition is a "chatbot chatter" category: "Certainly!" / "Great question!" openers, "I hope this helps!" closers, meta-commentary, labeled option menus, compliance echoes, and leftover paste artifacts like utm_source=chatgpt.com and [cite: 1]. That stuff wraps around the actual prose, so it's worth catching before touching any word choice.
Also new:
- categories for throat-clearing openers, hedge openers, formulaic closers, hook transitions, and rhetorical Q&A fragments
- a final "pub test": read the result aloud, and if you can't say it to a friend over a beer, keep rewriting
- DETAILED_PATTERNS.md gets a table of contents and two new sections (chatbot chatter, paste artifacts)
- REPLACEMENTS.md gets a contents line, a bunch of new replacement rows, and a section on human tells worth keeping, not just patterns to strip
- the frontmatter description now also triggers on "AI slop" and "ChatGPT mannerisms" phrasing
The rest is mostly rewording into consistent imperatives, plus a note about varying paragraph shape rather than only sentence length.