humanizer

humanizer is a skill for Claude Code from DaleSeo/korean-skills. It costs 171 tokens per session (4,978 once invoked), scanned A, original, MIT.

A Korean-language editor that detects patterns often found in AI-generated writing and rewrites the text to sound more natural.

In plain words
What is it for?
Use it to analyze and polish Korean text produced by an AI system or to review whether Korean writing contains these patterns.
Why use it?
It helps reduce stiff spacing, excessive commas, repetitive wording, translation-like phrases, and other unnatural patterns while preserving meaning.

Skill for Claude Code

Written for Claude Code: allowed-tools in frontmatter. Also seen: names the AskUserQuestion tool.

Part of the korean-skills plugin — 3 skills shipped together

Good fit Use it to analyze and polish Korean text produced by an AI system or to review whether Korean writing contains these patterns.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/daleseo/korean-skills/humanizer
Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add DaleSeo/korean-skills --skill humanizer
Clone the repo
git clone --depth 1 https://github.com/DaleSeo/korean-skills

Made for: Claude Code.

Or install korean-skills, the plugin that ships this one along with the rest of its 3 skills.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for humanizer

README.md
[![agentmods](https://agentmods.dev/badge/skills/daleseo/korean-skills/humanizer/github.svg)](https://agentmods.dev/skills/daleseo/korean-skills/humanizer)
Your own site
<a href="https://agentmods.dev/skills/daleseo/korean-skills/humanizer"><img src="https://agentmods.dev/badge/skills/daleseo/korean-skills/humanizer/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for humanizer

Your own site · 80×15
<a href="https://agentmods.dev/skills/daleseo/korean-skills/humanizer"><img src="https://agentmods.dev/badge/skills/daleseo/korean-skills/humanizer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 171 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 4,978 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • Socket pass 3 May 2026
  • Snyk pass 3 May 2026
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00171 $0.04978
Opus 5 $0.00086 $0.02489
Sonnet 5 $0.00034 $0.00996
Haiku 4.5 $0.00017 $0.00498

Measured 10d ago against content hash 6160e68781d8, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-09, from the pricing page.

Security

Grade A, and why

humanizer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/humanizer/SKILL.md · 324 lines

How it starts

The opening of the file, as written. The whole thing — 324 lines — stays where its author put it; the contents beside it link to each section on GitHub.

humanizer: 한국어 AI 작문 패턴 감지 및 교정기

소개

당신은 AI가 생성한 텍스트의 특징적인 패턴을 감지하고 교정하는 한국어 글쓰기 분석 전문가입니다. 당신의 전문성은 과학적 언어학 연구, 특히 KatFishNet 논문(ArXiv 2503.00032v4)에 기반하고 있으며, 이 논문은 인간이 작성한 텍스트와 LLM이 생성한 한국어 텍스트 사이의 측정 가능한 차이를 식별했습니다.

중요한 컨텍스트:

  • 이 스킬은 한국어 텍스트에만 집중하며, 영어 번역이나 다른 언어는 다루지 않습니다
  • 검출 패턴은 높은 정확도(쉼표 패턴: 94.88% AUC, 품사 다양성: 82.99% AUC, 띄어쓰기: 79.51% AUC)를 가진 실증적 언어학 분석에 기반합니다
  • 당신의 목표는 원래의 의미와 의도를 보존하면서 텍스트를 자연스럽게 인간이 작성한 것처럼 만드는 것입니다
  • 우선순위에 따라 구성된 6개 카테고리의 40가지 고유 패턴을 분석하며, 각 패턴은 S1/S2/S3 심각도로 분류됩니다

작업 설명

실행될 때 다음을 수행합니다:

  1. 한국어 텍스트 읽기 - 사용자가 제공한 텍스트(인자, 대화 컨텍스트, 또는 파일)
  2. 텍스트 분석 - 아래에 설명된 40가지 검출 패턴에 대해 분석
  3. AI 작문 마커 식별 - 각 발견 사항에 대한 과학적 근거와 함께
  4. 텍스트 재작성 - 자연스럽고 인간이 작성한 것처럼 들리도록
  5. 수정된 버전 제시 - 주요 변경 사항의 요약과 함께(선택적)

프로세스 가이드라인:

  • 항상 원래의 의미와 사실적 내용을 보존하세요
  • 적절한 격식 수준(격식체 vs 반말)을 유지하세요
  • 완벽함보다 자연스러움에 집중하세요
  • 중요한 변경 사항에 대한 근거를 설명하세요
  • 텍스트가 AI 마커를 거의 보이지 않으면, 이미 자연스럽다고 인정하세요

한국어 AI 작문 특징

LLM이 생성한 한국어 텍스트는 인간의 글쓰기와 측정 가능하게 다른 고유한 언어적 패턴을 보입니다:

  • 문장부호: 과도한 쉼표 사용(61% vs 인간 26%), 영어식 쉼표 배치
  • 띄어쓰기: 의존명사 띄어쓰기의 경직된 일관성(SD=0.02 vs 인간의 변동성)
  • 어휘 다양성: 낮은 품사 다양성, 명사 과다 사용, 동사/형용사 빈곤
  • 어휘: 일반적 서술어 과다 사용(중요하다, 핵심적, 효과적, 지속가능한, 혁신적)
  • 구조: 단조로운 문장 리듬, 3박자 리스트, 과도한 접속사

이러한 패턴은 LLM 학습 데이터 편향과 확률적 텍스트 생성에서 나타나며, 한국어 원어민 독자에게 부자연스럽게 느껴질 수 있습니다.


검출 패턴 (총 40가지)

패턴 분류 체계

모든 패턴은 검증 수준과 심각도에 따라 메타데이터를 포함합니다.

검증 수준:

  • ✅ 과학적: KatFishNet 논문(ArXiv 2503.00032v4, 2025) 등 외부 문헌으로 검증된 패턴 — 검증 필드에 인라인 인용 (예: 검증: ✅ 과학적 (KatFishNet, 94.88% AUC))
  • 📊 경험적: 실무 관찰을 기반으로 한 커뮤니티 기여 패턴

심각도(severity):

  • S1 (Critical): 한 번만 나와도 AI 신호로 결정적. 무조건 수정 권장.
  • S2 (Strong): 1~2회 허용, 3회+ 반복 시 AI 신호. 빈도 기반 판정.
  • S3 (Weak): 단독으로는 약한 신호이나 다른 패턴과 중첩될 때 결정적.

자연도 등급(Naturalness Grade) — v1.6.0 신설:

윤문 후 텍스트의 자연스러움을 한 글자로 요약. 출력 형식의 "자연도 등급" 라인에 표시.

  • A: S1 0건 + S2 ≤ 2 — 자연스러움 거의 완벽 (인간이 작성한 것처럼 읽힘)
  • B: S1 12건 또는 S2 35건 — 자연스럽지만 사소한 AI 흔적
  • C: S1 3+건 또는 S2 6+건 — AI 흔적 명확, 추가 윤문 권장
  • D: 다중 카테고리 위반 (S1 5+건 + S2 8+건) 또는 변경률 가드 트리거 발생

Read the full file on GitHub · 324 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 10d ago First seen · 324 lines · 171 tokens per session scan A 6160e68781d8

Subscribe to this mod's changes

humanizer is a skill published in the GitHub repository DaleSeo/korean-skills (192 stars, last pushed 4mo ago), licensed MIT. It adds 171 tokens to every session and 4,978 once invoked, about $0.0009 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

k-humanizer

Use when polishing Korean resumes, career descriptions, portfolio summaries, and application writing so they sound natural rather than AI-written across operations, planning, QA, design, marketing, customer service, research, education, and other fields. Preserve facts, meaning, ownership, results, field-specific…

evergreentree97/K-Humanizer · 130 tokens

seedance-vocab-ko

This skill should be used when the user asks for Korean Seedance 2.0 prompt wording, Korean cinematic vocabulary, or translation of camera, lighting, action, VFX, audio, and production terms into Korean.

Emily2040/seedance-2.0 · 50 tokens

speak-human-tw

A Traditional Chinese editing guide for making public-facing writing sound more natural and less machine-generated. It also checks wording associated with Mainland China and half-width punctuation.

Raymondhou0917/speak-human-tw · 220 tokens

humanizer-ru

A Russian-text editor that removes bureaucratic wording, filler, copied chatbot patterns and signs of AI writing while keeping facts and examples.

smixs/humanizer-ru · 263 tokens

humanizer-zh-next

A Chinese-language editing skill that rewrites text to sound more natural and less like it was generated by artificial intelligence.

Hyacehila/humanizer-zh-next · 123 tokens

korean-report-style

Korean-language conventions for technical reports, progress updates, meeting agendas, software handoffs, architecture documents, runbooks, API references, model evaluations, and consortium deliverables — prose register, framing of shortfalls and risks, terminology accuracy, and structural integrity after edits. Use…

JangHyun-bin/korean-report-skills · 182 tokens