Skip to content

Aboudjem/humanizer-skill

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Humanizer

The open-source skill that makes AI text undetectable.

Drop-in Claude Code skill • 30 AI patterns • 5 voice profiles • Zero dependencies

MIT License 30 AI Patterns 5 Voice Profiles Zero Dependencies Claude Code Skill


"AI detectors don't catch good writing. They catch lazy writing."

Most "humanizers" swap words. This one rewrites with a pulse. It injects the burstiness, perplexity, and voice that make text read like a thinking human wrote it.

Built from 90+ research sources including Wikipedia's Signs of AI Writing, academic papers from NeurIPS/ICLR/ACL 2024, the Washington Post's 328,744-message ChatGPT analysis, and battle-tested techniques from Reddit, HackerNews, and professional editorial firms.


Why this exists

Every AI detector measures the same things:

Signal Human AI What it means
Perplexity >85 (surprising words) <85 (predictable) AI picks the most likely next word. Humans don't.
Burstiness HIGH (3-word sentence, then 40) LOW (every sentence ~18 words) AI writes in monotone rhythm.
Type-Token Ratio 55.3 (diverse vocab) 45.5 (repetitive) AI recycles the same words.

Word-swapping tools don't fix these. You need structural transformation. That's what this does.


Install

Tip

The skill is a single file. No config, no setup, no dependencies. Claude Code picks it up automatically.

Option 1: Project-scoped (recommended, travels with your repo):

git clone https://github.com/Aboudjem/humanizer-skill.git
cp -r humanizer-skill/skills/humanizer .claude/skills/
rm -rf humanizer-skill

Option 2: Global (available in every project):

mkdir -p ~/.claude/skills/humanizer
curl -sL https://raw.githubusercontent.com/Aboudjem/humanizer-skill/main/skills/humanizer/SKILL.md \
  -o ~/.claude/skills/humanizer/SKILL.md

Option 3: Inside an existing plugin:

cp -r skills/humanizer /path/to/your-plugin/skills/

Note

Claude Code detects skills in .claude/skills/, ~/.claude/skills/, or any plugin's skills/ directory. No restart needed.


Usage

# Rewrite AI-sounding text (default)
/humanizer "Your AI-generated text here"

# Scan for AI patterns without changing anything
/humanizer "text" --mode detect

# Edit a file in place
/humanizer --mode edit --file docs/blog-post.md

# Pick a voice
/humanizer "text" --voice casual
/humanizer "text" --voice blunt --aggressive

Before & After

Technical docs

Before: This comprehensive guide delves into the intricacies of our authentication system. The platform leverages cutting-edge JWT technology to provide a seamless, secure, and robust authentication experience. Additionally, it features a pivotal role-based access control system that serves as a testament to our commitment to security.

After (--voice technical): The auth system uses JWTs. Tokens expire after 15 minutes; refresh tokens last 7 days. Role-based access control restricts API endpoints by user role: admin, editor, and viewer each see different data. The token rotation logic is in src/auth/refresh.ts if you need to change the expiry windows.

Killed 9 AI patterns. Added real details. Cut word count by 40%.


Blog post

Before: In today's rapidly evolving technological landscape, artificial intelligence is reshaping how we think about creativity. This groundbreaking shift represents a pivotal moment in human history, one that underscores the intricate interplay between innovation and artistic expression.

After (--voice casual): I've been messing around with AI image generators for about six months now, and I still can't decide if I love them or if they make me uneasy. The outputs are technically impressive. I got a portrait last week that looked better than anything I could paint in a year. But there's something missing. It's like eating a perfect-looking meal that has no flavor.

Replaced abstract commentary with personal experience. Sentence lengths: 8, 31, 22, 4, 13 words. That's burstiness.


LinkedIn post

Before: Excited to announce that I've taken on a pivotal new role at TechCorp! This incredible opportunity represents a significant milestone in my professional journey. I'm deeply grateful to my amazing network for their unwavering support. #NewBeginnings #Innovation #Leadership #Grateful

After (--voice professional): Started a new job at TechCorp this week. I'm leading their developer tools team, 12 engineers building internal tooling that currently serves about 400 developers. First week has been drinking from the firehose: new codebase, new faces, new coffee machine I can't figure out. Nervous and excited in roughly equal measure.

No emojis. No hashtags. Real details instead of "pivotal milestone." The coffee machine line is more human than any amount of gratitude signaling.


How it works

Input text
    |
    v
[1. DETECT] ── Scan for 30 AI patterns across 5 categories
    |
    v
[2. STRIP] ─── Remove significance inflation, AI vocabulary,
    |             filler phrases, chatbot artifacts
    v
[3. INJECT] ── Apply voice profile, burstiness variation,
    |             perplexity increase, soul injection
    v
[4. VERIFY] ── Sentence variance, blacklist words,
    |             parallel structure, "who wrote this?" test
    v
Output + change summary

Three modes

Mode What it does When to use
rewrite Full transformation with voice injection Content creation, blog posts, social media
detect Scan-only report with pattern counts and severity Auditing existing content, learning what to fix
edit In-place file editing with minimal changes Documentation cleanup, README polishing

Important

rewrite is the default mode. You don't need to specify it.


Voice profiles

Voice Personality Best for
casual Contractions, first person, fragments, "And" starters Blog posts, social media, community docs
professional Selective contractions, dry wit, concrete examples Business comms, reports, formal docs
technical Precise terms, code-like clarity, deadpan humor API docs, READMEs, architecture docs
warm "We/our" language, empathy, shorter paragraphs Tutorials, onboarding, support content
blunt Shortest sentences, no hedging, active voice only Reviews, internal comms, direct feedback

All 30 patterns

Content Patterns (P1-P8), the worst offenders
# Pattern What to look for
P1 Significance Inflation "marking a pivotal moment", "is a testament to"
P2 Notability Name-Dropping "featured in", "active social media presence"
P3 Superficial -ing Phrases "highlighting", "ensuring", "fostering"
P4 Promotional Language "cutting-edge", "seamless", "world-class", "nestled"
P5 Vague Attributions "Experts argue", "Research suggests" (no citation)
P6 Formulaic Challenges "Despite challenges, continues to thrive"
P7 AI Vocabulary "delve", "leverage", "multifaceted", "tapestry"
P8 Copula Avoidance "serves as" instead of "is"
Language & Style (P9-P18), structural tells
# Pattern What to look for
P9 Negative Parallelisms "It's not just X, it's Y"
P10 Rule of Three Forced triads: "innovation, inspiration, and insights"
P11 Synonym Cycling "protagonist" then "main character" then "central figure"
P12 False Ranges "From X to Y" on non-spectrums
P13 Em Dash Ban Zero em dashes allowed, replace with commas/hyphens
P14 Boldface Overuse Bold on every noun, emoji headers
P15 Structured List Syndrome **Header:** description bullets for prose content
P16 Title Case Headings "Strategic Negotiations And Global Partnerships"
P17 Typographic Tells Curly quotes, consistent Oxford comma
P18 Formal Register Overuse "it should be noted that", "it is essential to"
Communication (P19-P21), chatbot residue
# Pattern What to look for
P19 Chatbot Artifacts "I hope this helps!", "Certainly!"
P20 Knowledge-Cutoff Disclaimers "As of [date]", "based on available information"
P21 Sycophantic Tone "Great question!", "That's an excellent point!"
Filler & Hedging (P22-P30), dead weight
# Pattern What to look for
P22 Filler Phrases "In order to", "Due to the fact that", "It's worth noting"
P23 Excessive Hedging "could potentially possibly"
P24 Generic Conclusions "The future looks bright", "poised for growth"
P25 Hallucination Markers Fabricated-feeling dates, phantom citations
P26 Perfect/Error Alternation Inconsistent quality = partial AI edit
P27 Question-Format Titles "What makes X unique?", "Why is Y important?"
P28 Markdown Bleeding **bold** in emails, Word docs, social posts
P29 "Comprehensive Overview" "This guide delves into...", "Let's dive in"
P30 Uniform Sentence Length Every sentence 15-25 words, no variation

The science

Note

Every technique in this skill is grounded in published research. No guesswork.

Technique Source Finding
Burstiness injection GPTZero1 Human sentence length varies wildly. AI doesn't.
Perplexity increase GPTZero1 AI picks the most statistically likely next word.
Vocabulary diversity SSRN stylometric study2 Human TTR: 55.3 vs AI: 45.5
Kill negative parallelism Washington Post3 "It's not X, it's Y" confirmed as #1 AI tell across 328K messages
Structural paraphrasing RAID benchmark, ACL 20244 Drops DetectGPT accuracy from 70.3% to 4.6%
Intrinsic dimension NeurIPS 20235 Human text ~9 dimensions vs AI ~7.5

Comparison

Feature Humanizer QuillBot Undetectable.ai Prompt hacks
Open source Yes No No N/A
Pattern detection 30 0 0 0
Voice profiles 5 0 3 Manual
Works offline Yes No No No
Burstiness injection Yes No Partial No
Works in English Yes Multi Multi Manual
File editing mode Yes No No No
Explains changes Yes No No No
Price Free $20/mo $10/mo Free

File structure

your-project/
  .claude/
    skills/
      humanizer/
        SKILL.md    # <- the entire skill, one file

Tip

The skill works in .claude/skills/ (project), ~/.claude/skills/ (global), or any plugin's skills/ directory.


Contributing

Found a new AI pattern? Have a better fix? PRs welcome.

  1. Fork the repo
  2. Add your pattern to SKILL.md (follow the P1-P30 format)
  3. Include a before/after example
  4. Open a PR

See CONTRIBUTING.md for details.


Sources

Built from 90+ sources across academic research, editorial expertise, and community intelligence:

  • Wikipedia: Signs of AI writing, 24 pattern categories with real examples
  • Wikipedia FR: Identifier l'usage d'une IA generative, additional AI pattern research
  • RAID Benchmark (ACL 2024), 6M+ generations, 12 detectors evaluated
  • NeurIPS 2023, intrinsic dimension analysis (Tulchinskii et al.)
  • Washington Post, 328,744 ChatGPT message analysis
  • Stanford HAI, ESL false positive study
  • Max Planck Institute, AI vocabulary frequency spikes
  • Softaworks agent-toolkit humanizer by @blader
  • William Strunk Jr., The Elements of Style
  • Gary Provost, David Ogilvy, Ann Handley, professional writing craft

If this skill saved your writing from sounding like a chatbot, consider giving it a star.
It helps others find it.


MIT License

Footnotes

  1. GPTZero detection methodology: perplexity and burstiness as core signals 2

  2. SSRN stylometric study comparing type-token ratios across human and AI corpora

  3. Washington Post analysis of 328,744 ChatGPT messages identifying distinctive AI constructs

  4. RAID: A Shared Benchmark for Robust Evaluation of Machine-Generated Text Detectors (ACL 2024, 6M+ generations)

  5. Tulchinskii et al., NeurIPS 2023: Intrinsic dimensionality estimation for AI text detection

About

Make AI text undetectable. 30 AI patterns, 5 voice profiles, zero dependencies. Drop-in Claude Code skill that transforms AI-generated text into natural human writing using burstiness, perplexity, and voice injection.

Topics

Resources

License

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors