AI-tell blacklist for copy. Clients reject drafts that read as AI-written, and the worst tells are sentence shapes that word lists miss. Closes #613.
- copywriting (2.0.2 → 2.1.0): new No AI Tells section and
references/ai-tells.md, a researched blacklist (Wikipedia's AI-cleanup guide, excess-vocabulary studies, slop benchmarks, practitioner catalogs). Bans contrast reveals ("it's not X, it's Y," "not because X, because Y"), negation lists ("no X, no Y, no Z"), trailing pile-ons (a claim followed by more comma clauses), self-answered questions, colon reveals, stock openers and marketing phrases, and em dashes in short copy. Caps fragments, lists of three, and vague vocabulary. Rewrite rules (fix from the facts, swap test, no invented proof) and a self-check to run before every delivery. Rhetorical-question advice now warns against answering your own question. The skill now follows its own rules: em dashes and a contrast reveal removed from SKILL.md, and conflicting advice removed from the references ('When it comes to,' self-answering question transitions, the 'Everything you need to' formula). Defines short copy, long copy, and sections, and lists what stays fine (FAQ questions, one 'No card required,' real three-feature lists). The rules are mirrored in the description. New triggers: 'this sounds like AI,' 'AI slop,' 'make it sound human.' New eval (id 9). - copy-editing (2.0.0 → 2.1.0): new AI-Tell Check run on every edit, inline because installed skills can't load copywriting's references; a "Reads as AI-Written" problem entry and checklist section. The word-swap table no longer suggests synonyms (robust → strong) that the check forbids. New control eval (id 8) checks that clean copy is left alone. Rule mirrored in the description. New triggers: 'this sounds like AI,' 'AI slop,' 'de-slop this,' 'make it sound human.' New eval (id 7).