I spent a month cleaning Korean text data every day.
The same problem kept showing up. A sentence could be grammatically correct and still feel like English translated word for word.
For example:
Before:
고객 문의 분류 체계 고도화를 통해 주간 평균 처리 시간을 18시간에서 11시간으로 크게 개선했습니다.
After:
고객 문의 분류 기준을 정리해 주간 평균 처리 시간을 18시간에서 11시간으로 줄였습니다.
The facts did not change. The second version simply sounds closer to how a Korean speaker would write it.
I kept notes on patterns like this and turned them into K-Humanizer, a free and open-source Agent Skill for Cursor.
It is meant for non-native Korean writers and global teams writing for Korean users. Before editing, it checks the reader, the context, and how formal the text should be. It also tries to preserve facts, numbers, terminology, and uncertainty.
Install:
npx skills add evergreentree97/K-Humanizer --skill k-humanizer --full-depth
Example:
Use $k-humanizer to rewrite this Korean product copy for Korean users. Keep the meaning, numbers, and product terms unchanged.
Repository:
It is MIT-licensed and currently at v0.1. I would like to see examples where it misses translation-like phrasing or makes the Korean too formal.