What problem does it solve? Korean text produced by ChatGPT, Claude, or Gemini carries recognizable machine patterns—translationese, mechanical parallelism, formulaic closings, uniform sentence rhythm—that make it instantly identifiable as AI-written. This Skill detects those patterns and rewrites only the style and rhythm while leaving the content untouched. ## Core Features & Use Cases - 40+ AI-tell pattern taxonomy: Detects 10 categories of Korean AI signatures (translationese, excessive hedging, connector overuse, comma-after-connective-endings, hanja nominalizer density) with severity levels S1–S3. - Two execution modes: Fast mode runs a single monolith agent call for texts under 8,000 characters; Strict mode runs a 5-agent pipeline (detector, rewriter, fidelity auditor, naturalness reviewer) with up to 3 rewrite loops. - Quantitative metrics: Python scripts compute comma rates, lexical diversity, passive constructions, and pronoun density against human/AI baselines from published research (KatFish, LREAD). - Use Case: Paste a ChatGPT-drafted Korean column and receive a rewritten version with a change-rate metric, quality grade, and per-category detection summary—meaning preserved, AI tells removed. ## Quick Start Ask the agent to remove the AI tone from your Korean draft, for example: "이 ChatGPT 초안에서 AI 티 없애줘."
