This edition of Humanize Korean is made for GitHub Copilot CLI. It reads a Korean text written by an AI tool, finds the habits that mark it as machine writing, and rewrites only those spans so style, rhythm and wording read naturally while facts, numbers, names and quotes stay unchanged.
Everything happens in one model call. The skill loads a short rule book, scans for patterns across ten categories, rewrites in a fixed order, checks six self-test items and saves one final file with a summary block at the end.
What You Get
- A final.md file in a run folder with the rewritten text and one summary block
- A short reply with status, main detections, highlights and remaining warnings
- A grade from A to D based on remaining severe patterns and the change rate
- Options for genre, strength and minimum severity
Common Use Cases
- Polishing an AI-written Korean paragraph from Copilot CLI
- Keeping a formal register intact while removing translation-style phrasing
- Getting a quick cleanup without the multi-call diagnosis path
What Makes It Unique
It is the lightest edition of the project. The full version routes text through scoring scripts and several agents, and this one runs a single pass with a seven-point rule list, so it fits Copilot CLI, which has no multi-call path. It keeps the register rule in both directions, so formal text stays formal and spoken text stays spoken.
Under The Hood
The skill is one instruction file that points to rule books kept elsewhere in the repository, so those files must be present next to it for the pass to work. It reads the pasted text or a file path you give it, and it creates a workspace folder under the current directory where it writes the final file. It does not run scripts, fetch web pages or call outside APIs.