humanize-korean is a specialized Claude Code skill designed to strip away the 'AI signature' from Korean text produced by LLMs like ChatGPT and Claude. It employs a sophisticated 10-category taxonomy to detect and surgically fix over 40 mechanical patterns—such as awkward translation-style phrasing, robotic list structures, and characteristic AI idioms—while strictly preserving all factual data, quotes, and technical terms. By utilizing quantitative metrics and genre-specific rules, it ensures that blogs, reports, and business documents maintain a natural, professional tone without the risk of over-editing or meaning distortion.
主な機能
01Strict preservation of facts, figures, proper nouns, and direct quotes during rewriting.
02Safety guardrails that prevent over-editing by triggering warnings or rollbacks if change rates exceed thresholds.
03219 GitHub stars
04Detection and removal of 40+ AI-specific Korean linguistic patterns and translation artifacts.
05Quantitative before-and-after metrics via a built-in Python analysis engine for quality assurance.
06Automatic genre detection (Column, Report, Blog, Official) to maintain context-appropriate register.
ユースケース
01Localizing translated content by eliminating awkward grammar and excessive English citations.
02Refining business proposals and reports to remove robotic phrasing and repetitive list structures.
03Polishing AI-generated blog posts or newsletters to improve readability and avoid AI detection.