If you're working with Korean LLM outputs and they feel stiff or obviously machine-generated, this skill rewrites them to sound natural. It's built on linguistic research from the KatFishNet paper and checks for 40 specific patterns like comma overuse, rigid spacing around dependent nouns, vocabulary tells, and translation-ese constructions. Each pattern gets a severity rating and the tool explains what it caught before rewriting. It includes safeguards to prevent over-editing and meaning drift. Works best on longer Korean text where patterns are easier to spot. The reference system is thorough, maybe more than you need for quick edits, but helpful if you're curious why Korean AI text reads the way it does.
npx skills add https://github.com/daleseo/korean-skills --skill humanizer