What problem does it solve? AI-generated text carries recognizable tells—excessive em-dashes, hollow marketing words like "leverage" and "혁신적", rigid 3-bullet structures, and formulaic closings—that make content feel robotic. This Skill scans text for those patterns, rewrites it in a natural human voice, and creates a per-project STYLE.md so future writing stays consistent. ## Core Features & Use Cases - LLM Tell Detection: Identifies Korean and English patterns such as "~을 도와드립니다", "In conclusion", emoji overload, and em-dash overuse, logging each finding with file, line, and suggested fix. - Three Output Modes: audit (findings only), rewrite (in-place correction plus findings and STYLE.md), and init (STYLE.md only for new projects). - CI Regression Prevention: Ships a grep-based voice-lint.sh script that fails builds when banned vocabulary from STYLE.md reappears. - Use Case: Before publishing release notes or store listing copy, run an audit to strip phrases like "seamless" and "강력한", then rewrite in your project's defined persona so the text reads like a person wrote it. ## Quick Start Ask the assistant to check this text for AI-sounding phrasing and rewrite it to sound like a real person wrote it.
