Skip to content

kgraph57/akanuke

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 

Repository files navigation

Akanuke

A Claude Code skill that removes signs of AI-generated writing from text — in both English and Japanese.

What It Does

LLMs produce text with predictable tells: inflated significance, promotional tone, excessive hedging, sycophancy, em-dash overuse, rule-of-three lists, AI vocabulary words, and more. This skill detects those patterns and rewrites text to sound like a human wrote it.

  • English: 24 patterns based on Wikipedia's "Signs of AI writing"
  • Japanese: 30 patterns across 6 categories, adapted for Japanese-specific LLM artifacts (formatting leaks, monotonous rhythm, instruction-manual tone, fence-sitting, abstract filler, cliched metaphors)

Language is auto-detected. No configuration needed.

Installation

Clone directly into your Claude Code skills directory:

mkdir -p ~/.claude/skills
git clone https://github.com/kgraph57/akanuke.git ~/.claude/skills/akanuke

Or copy just the skill file:

mkdir -p ~/.claude/skills/akanuke
curl -o ~/.claude/skills/akanuke/SKILL.md \
  https://raw.githubusercontent.com/kgraph57/akanuke/main/SKILL.md

Usage

Invoke the skill in Claude Code:

/humanize [paste your text]

Or just ask naturally:

Humanize this: [your text]
人間っぽくして: [your text]
AI臭を消して: [your text]

Options

Flag Description
--report / レポートも Append a diagnostic report with pattern counts, severity, and AI score
--formal Business document mode (polished but not robotic)
--academic Scholarly writing mode (objective, no unsupported claims)
--medical Healthcare content mode (preserves terminology, adjusts to evidence level)

Patterns Detected

English (24 patterns)

Category Patterns
Content Significance inflation, notability name-dropping, superficial -ing analyses, promotional language, vague attributions, formulaic challenges sections
Language AI vocabulary, copula avoidance, negative parallelisms, rule of three, synonym cycling, false ranges
Style Em dash overuse, boldface overuse, inline-header lists, title case headings, emojis, curly quotes
Communication Chatbot artifacts, knowledge-cutoff disclaimers, sycophantic tone
Filler Filler phrases, excessive hedging, generic positive conclusions

Japanese (30 patterns, 6 categories)

Category Patterns
Leftover formatting Asterisk bold, em dashes, over-bracketing 「」, nested 「『』」, colon-space, over-parenthesizing, parenthetical disclaimers, slash-separated concepts
Monotonous rhythm Same endings, excessive conjunctions, flat temperature, neat closings, repeated not-A-but-B
Instruction-manual tone Long preambles, hollow "conclusion first", structure re-announcements, STEP labels, closing cliches
Fence-sitting Insurance clauses, forced neutrality, weak negation, case-by-case escape
Abstract filler Abstract-only prose, strong claims with zero evidence, synonym barrages, AI vocabulary (JP)
Cliched metaphors Tool metaphors (羅針盤, 地図), body metaphors (土台, 柱), machine metaphors (車の両輪), cooking metaphors (レシピ)

Examples

English

Before:

AI-assisted coding serves as an enduring testament to the transformative potential of large language models, marking a pivotal moment in the evolution of software development. It's not just about autocomplete; it's about unlocking creativity at scale.

After:

AI coding assistants speed up some tasks. In a 2024 study by Google, developers using Codex completed simple functions 55% faster, but showed no improvement on debugging or architectural decisions.

Japanese

Before:

現代の医療現場において、生成AIの活用は非常に重要なテーマとなっています。以下では、その具体的な活用方法について3つの観点から解説します。これは単なる「ツール」ではなく、医療の「羅針盤」とも言えるでしょう。参考になれば幸いです。

After:

医療の現場で生成AIが使える場面は、思ったより広い。診断支援では鑑別診断のリストアップや見落としの防止に向いている。万能ではないが、定型的な作業を肩代わりさせるだけでも、浮いた時間を患者や研究に回せる。

Why AI Text Sounds Like AI

  1. Training data bias — Manuals, FAQs, regulations, and training materials are overrepresented. LLMs learn them as templates.
  2. Safety alignment side effects — "Don't say anything harmful" becomes "don't say anything at all." Every position gets hedged.
  3. English-centric architecture — Markdown formatting, bulleted-list culture, and English logical structures leak into other languages.

Credits

License

MIT

About

Claude Code skill: remove AI writing patterns from English and Japanese text

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors