daleseo/humanizer-ko

Humanizer ν•œκ΅­ πŸ‡°πŸ‡·

1 stars0 forksUpdated Jan 26, 2026
npx skills add daleseo/humanizer-ko

README

humanizer-ko

An AI agent skill that detects and corrects Korean AI writing patterns to make text sound naturally human-written.

ν•œκ΅­μ–΄ AI μž‘λ¬Έ νŒ¨ν„΄μ„ κ°μ§€ν•˜κ³  μžμ—°μŠ€λŸ¬μš΄ μΈκ°„μ˜ κΈ€μ“°κΈ°λ‘œ λ³€ν™˜ν•˜λŠ” AI μ—μ΄μ „νŠΈ μŠ€ν‚¬μž…λ‹ˆλ‹€.

Based on scientific research (KatFishNet paper, 94.88% AUC accuracy) β€’ 과학적 연ꡬ 기반 (KatFishNet λ…Όλ¬Έ, 94.88% AUC 정확도)


English

What is humanizer-ko?

humanizer-ko is an AI agent skill that transforms artificial-sounding Korean text into natural human writing based on scientific linguistic research. It works with Claude Code, Cursor, Windsurf, and other AI coding agents that support the Agent Skills format. The skill analyzes and corrects the distinctive patterns that large language models (LLMs) like ChatGPT, Claude, and Gemini produce when writing in Korean.

The Problem: LLM-generated Korean text exhibits measurable linguistic patterns that differ from human writing. For example, LLMs use commas in 61% of sentences while humans use them in only 26%. LLMs also show overly consistent spacing (standard deviation 0.02 vs natural human variation), higher noun ratios, and poverty in verb/adjective usage.

The Solution: Based on the KatFishNet paper (ArXiv 2503.00032v4), this skill detects 19 linguistic patterns across 5 categories and rewrites text to sound naturally human-written while preserving the original meaning and formality level.

Key Features

  • Scientifically grounded: Based on empirical linguistic research (KatFishNet paper)
  • Korean-specific: Analyzes patterns unique to Korean language, not translated from English/Chinese
  • 19 detection patterns: Organized into 5 categories by priority
  • Preserves meaning: Maintains original intent and formality level
  • High accuracy: 94.88% AUC for punctuation, 82.99% for POS, 79.51% for spacing

Quick Start

Installation:

Using Skills CLI (recommended):

npx skills add daleseo/humanizer-ko

Or manual installation:

mkdir -p ~/.claude/skills
git clone https://github.com/daleseo/humanizer-ko.git ~/.claude/skills/humanizer-ko

Usage:

/humanizer-ko

[Paste Korean text to humanize]

Detection Patterns

The skill analyzes 5 categories of AI writing patterns:

1. Punctuation (6 patterns) - Highest priority (94.88% AUC)

  • Excessive comma usage (LLM 61% vs human 26%)
  • English-style comma placement
  • Commas after connective endings
  • Sentence-ending comma patterns
  • Unnecessary list commas
  • Em-dash overuse

2. Spacing (3 patterns) - High priority (79.51% AUC)

  • Rigid bound noun spacing (standard deviation <0.05)
  • Excessive auxiliary verb spacing
  • Numeral-bound noun spacing

3. Part-of-Speech Diversity (3 patterns) - High priority (82.99% AUC)

  • Noun overuse (>35% ratio)
  • Verb/adjective poverty (<25% ratio)
  • POS n-gram monotony

4. Vocabulary (3 patterns) - Medium priority

  • AI vocabulary overuse (μ€‘μš”ν•˜λ‹€, 효과적, ν˜μ‹ μ , 지속가λŠ₯ν•œ, etc.)
  • Unnecessary Sino-Korean terms (μ§„ν–‰ν•˜λ‹€β†’ν•˜λ‹€, μ‹€μ‹œν•˜λ‹€β†’ν•˜λ‹€)
  • English idiom calques (~의 쀑심에, ~λ₯Ό 톡해)

5. Sentence Structure (4 patterns) - Medium priority

  • Lack of rhythm (uniform sentence length)
  • Rule of three overuse
  • Connector overuse (κ·ΈλŸ¬λ‚˜, λ˜ν•œ, λ”°λΌμ„œ)
  • Honorific uniformity

Example

Before (AI-generated):

인곡지λŠ₯ 기술의 λ°œμ „μ€ λΉ λ₯΄κ²Œ μ§„ν–‰λ˜κ³  있으며, λ‹€μ–‘ν•œ μ‚°μ—… 뢄야에 적용되고 μžˆμŠ΅λ‹ˆλ‹€. μ΄λŸ¬ν•œ κΈ°μˆ μ€ 효과적으둜 업무 νš¨μœ¨μ„±μ„ ν–₯μƒμ‹œν‚€κ³ , ν˜μ‹ μ μΈ μ†”λ£¨μ…˜μ„ μ œκ³΅ν•˜λ©°, 지속가λŠ₯ν•œ μ„±μž₯을 κ°€λŠ₯ν•˜κ²Œ ν•©λ‹ˆλ‹€.

After (humanized):

인곡지λŠ₯ κΈ°μˆ μ€ λΉ λ₯΄κ²Œ λ°œμ „ν•˜κ³  있으며 μ—¬λŸ¬ μ‚°μ—… 뢄야에 적용되고 μžˆμŠ΅λ‹ˆλ‹€. 이 κΈ°μˆ μ€ 업무 νš¨μœ¨μ„ 높이고 μƒˆλ‘œμš΄ μ†”λ£¨μ…˜μ„ μ œκ³΅ν•˜λ©° μž₯기적 μ„±μž₯을 κ°€λŠ₯ν•˜κ²Œ ν•©λ‹ˆλ‹€.

Key changes: Removed 6 commas, replaced AI vocabulary (λ‹€μ–‘ν•œβ†’μ—¬λŸ¬, νš¨κ³Όμ μœΌλ‘œβ†’λ†’μ΄κ³ , ν˜μ‹ μ μΈβ†’μƒˆλ‘œμš΄, 지속가λŠ₯ν•œβ†’μž₯기적), replaced unnecessary Sino-Korean terms (μ§„ν–‰λ˜κ³ β†’λ°œμ „ν•˜κ³ , ν–₯μƒμ‹œν‚€κ³ β†’λ†’μ΄κ³ ).

Resources

License

MIT License - Free to use, modify, and distribute.


ν•œκ΅­μ–΄

humanizer-koλž€?

humanizer-koλŠ” 과학적 μ–Έμ–΄ν•™ 연ꡬλ₯Ό 기반으둜 μΈμœ„μ μΈ ν•œκ΅­μ–΄ ν…μŠ€νŠΈλ₯Ό μžμ—°μŠ€λŸ¬μš΄ μΈκ°„μ˜ κΈ€μ“°κΈ°λ‘œ λ³€ν™˜ν•˜λŠ” AI μ—μ΄μ „νŠΈ μŠ€ν‚¬μž…λ‹ˆλ‹€. Claude Code, Cursor, Windsurf λ“± Agent Skills ν˜•μ‹μ„ μ§€μ›ν•˜λŠ” λ‹€μ–‘ν•œ AI μ½”λ”© μ—μ΄μ „νŠΈμ—μ„œ μ‚¬μš©ν•  수 μžˆμŠ΅λ‹ˆλ‹€. ChatGPT, Claude, Gemini 같은 λŒ€ν˜• μ–Έμ–΄ λͺ¨λΈ(LLM)이 ν•œκ΅­μ–΄λ‘œ μž‘μ„±ν•  λ•Œ λ‚˜νƒ€λ‚˜λŠ” νŠΉμ§•μ μΈ νŒ¨ν„΄μ„ λΆ„μ„ν•˜κ³  κ΅μ •ν•©λ‹ˆλ‹€.

문제: LLM이 μƒμ„±ν•œ ν•œκ΅­μ–΄ ν…μŠ€νŠΈλŠ” μΈκ°„μ˜ 글쓰기와 λ‹€λ₯Έ μΈ‘μ • κ°€λŠ₯ν•œ 언어학적 νŒ¨ν„΄μ„ λ³΄μž…λ‹ˆλ‹€. 예λ₯Ό λ“€μ–΄, LLM은 λ¬Έμž₯의 61%μ—μ„œ μ‰Όν‘œλ₯Ό μ‚¬μš©ν•˜λŠ” 반면 μ‚¬λžŒμ€ 26%만 μ‚¬μš©ν•©λ‹ˆλ‹€. λ˜ν•œ LLM은 μ§€λ‚˜μΉ˜κ²Œ 일관적인 띄어쓰기(ν‘œμ€€νŽΈμ°¨ 0.02 vs μΈκ°„μ˜ μžμ—°μŠ€λŸ¬μš΄ λ³€ν™”), 높은 λͺ…사 λΉ„μœ¨, 동사/ν˜•μš©μ‚¬ λΉˆκ³€μ„ λ³΄μž…λ‹ˆλ‹€.

ν•΄κ²°μ±…: KatFishNet λ…Όλ¬Έ(ArXiv 2503.00032v4)을 기반으둜 5개 μΉ΄ν…Œκ³ λ¦¬μ˜ 19κ°€μ§€ 언어학적 νŒ¨ν„΄μ„ κ°μ§€ν•˜κ³ , μ›λ¬Έμ˜ μ˜λ―Έμ™€ 격식 μˆ˜μ€€μ„ μœ μ§€ν•˜λ©΄μ„œ μžμ—°μŠ€λŸ¬μš΄ μΈκ°„μ˜ κΈ€μ“°κΈ°λ‘œ μž¬μž‘μ„±ν•©λ‹ˆλ‹€.

μ£Όμš” νŠΉμ§•

  • 과학적 기반: 싀증적 μ–Έμ–΄ν•™ 연ꡬ 기반 (KatFishNet λ…Όλ¬Έ)
  • ν•œκ΅­μ–΄ νŠΉν™”: μ˜μ–΄/쀑ꡭ어 λ²ˆμ—­μ΄ μ•„λ‹Œ ν•œκ΅­μ–΄ 고유 νŒ¨ν„΄ 뢄석
  • 19κ°€μ§€ νŒ¨ν„΄: μš°μ„ μˆœμœ„λ³„ 5개 μΉ΄ν…Œκ³ λ¦¬λ‘œ ꡬ성
  • 의미 보쑴: μ›λ¬Έμ˜ μ˜λ„μ™€ 격식 μˆ˜μ€€ μœ μ§€
  • 높은 정확도: λ¬Έμž₯λΆ€ν˜Έ 94.88% AUC, ν’ˆμ‚¬ 82.99%, 띄어쓰기 79.51%

λΉ λ₯Έ μ‹œμž‘

μ„€μΉ˜:

Skills CLI μ‚¬μš© (ꢌμž₯):

npx skills add daleseo/humanizer-ko

λ˜λŠ” μˆ˜λ™ μ„€μΉ˜:

mkdir -p ~/.claude/skills


...
Read full README

Publisher

daleseodaleseo

Statistics

Stars1
Forks0
Open Issues0
LicenseMIT License
CreatedJan 26, 2026