What problem does it solve? Generated documents often contain generic "AI-sounding" prose — stock vocabulary like "delve" and "seamless", hedge-word clusters, vague attribution, and boilerplate transitions — that makes deliverables read as machine-written. This Skill deterministically detects those patterns with exact file:line locations so you can review and fix them deliberately instead of guessing. ## Core Features & Use Cases - Deterministic pattern scanning: Runs a dependency-free Node.js script (check-tone.cjs) that flags stock vocabulary, promotional puffery, hedge clusters, weasel attribution, the "not just X, it's Y" construction, dangling "-ing" clauses, bold-lead-in bullet lists, em-dash overuse, uniform sentence rhythm, and repeated paragraph openers. - Structured report before rewriting: Findings are grouped by tier with file:line locations; nothing is rewritten until explicitly requested, and then only the flagged spans with before/after diffs. - Qualitative second pass: A reference guide covers what regex cannot catch — staging instead of stating, mechanical three-part structure, inflation without watched words, drafting leftovers, and absence of voice. - Use Case: After generating a status report or runbook with Claude, run the tone check to get a list of generic phrases with locations, then request targeted rewrites of only the flagged sentences while preserving all facts, numbers, and document structure. ## Quick Start Ask the AI to check whether the generated document at docs/report.md sounds like AI writing and report the flagged patterns before changing anything.
