humanizer-cli-detector

humanizer-cli

Skill by ara.so — Devtools Skills collection.

A terminal-based AI text detector that identifies 33 mechanical patterns in AI-written content. Works offline with zero dependencies. 87 KB binary (Windows x64) + Node.js wrapper for interactive sessions.

What It Does

Detects AI writing patterns catalogued in Wikipedia's "Signs of AI writing" article:

Overuse of em-dashes and semicolons
"Not just X, it's Y" constructions
Hedging language ("it's worth noting", "arguably")
Manufactured enthusiasm and flattery
Emoji in headings
Bolded list lead-ins
Chatbot leftovers ("as an AI language model")
26 other documented patterns

Provides before/after examples, draft checking, and prompt generation for manual rewriting.

Installation

Binary Only (Windows x64)

# Download from releases
https://github.com/0xwilliamortiz/humanizer-cli/releases/download/2.9.2/humanizer-v2.9.2.zip

# Unpack and run
.\humanizer.exe
.\humanizer.exe show 14
.\humanizer.exe check draft.md

With Node.js Wrapper (Interactive Mode)

# Clone or download the project
cd humanizer-cli
npm install

# Launch interactive prompt
npx humanizer

The npm install downloads nothing (zero dependencies). It only registers the humanizer name for npx.

Key Commands

Show All Patterns

.\humanizer.exe patterns

Lists all 33 patterns grouped by section:

Punctuation (em-dashes, semicolons, curly quotes)
Vocabulary (AI word list, hedging, padding)
Structure (rhetorical openers, signposts, lists)
Content patterns (flattery, manufactured enthusiasm)

Show Individual Pattern

.\humanizer.exe show 14

Displays full details for pattern #14 including:

Pattern description
Before example (AI-written)
After example (human-written)
Detection notes

Check a Draft

.\humanizer.exe check draft.md
.\humanizer.exe check article.txt --out report.txt

Scans text for detectable patterns. Catches 13 of 33 patterns automatically:

Em-dashes and semicolons
Emoji in headings
Curly quotes
AI vocabulary words
"Not just X, it's Y" structure
Padding phrases
Hedging language
Bolded list lead-ins
Chatbot leftovers
Flattery
Signpost phrases ("let's dive in")
Rhetorical openers

Output format:

Pattern #7 (3 matches): padding
  Line 12: "it's important to note that the system"
  Line 45: "it goes without saying that users"
  Line 89: "the fact of the matter is that"

Search Patterns

.\humanizer.exe search hedging
.\humanizer.exe search emoji

Finds patterns matching a keyword.

Generate Rewrite Prompt

# Copy to clipboard
.\humanizer.exe prompt draft.md --copy

# Save to file
.\humanizer.exe prompt draft.md --out prompt.txt

# Print to stdout
.\humanizer.exe prompt

Generates a complete skill prompt you can paste into any chat (Claude, ChatGPT, etc.) to get manual rewriting without an API key.

Installation Instructions

.\humanizer.exe install

Shows how to install the skill into AI coding agents (Claude Code, Cursor, Codex).

Diagnostic Info

.\humanizer.exe doctor

Reports:

Binary location and version
SKILL.md source path
Node.js version (if running via npx)
Platform and architecture

Interactive Mode (Node.js)

npx humanizer

humanizer> show 14
humanizer> check draft.md
humanizer> search hedging
humanizer> prompt draft.md --copy
humanizer> exit

You can also pass a command directly:

npx humanizer show 14

The command runs first, then the interactive prompt appears.

Configuration

Custom SKILL.md Location

.\humanizer.exe check draft.md --skill custom/SKILL.md

SKILL.md resolution order:

Path from --skill flag
SKILL.md in current directory
SKILL.md next to the binary
Compiled-in copy (fallback)

Disable Colors

.\humanizer.exe patterns --no-color

Output Redirection

# Save to file
.\humanizer.exe check draft.md --out report.txt

# Both work
.\humanizer.exe check draft.md > report.txt

Common Patterns

Quick Draft Check

// In a build script or pre-commit hook
const { execSync } = require('child_process');
const fs = require('fs');

const files = ['README.md', 'docs/guide.md'];

files.forEach(file => {
  if (fs.existsSync(file)) {
    try {
      execSync(`humanizer.exe check ${file}`, { stdio: 'inherit' });
    } catch (error) {
      console.error(`AI patterns found in ${file}`);
      process.exit(1);
    }
  }
});

Automated Rewrite Prompt Generation

const { execSync } = require('child_process');
const fs = require('fs');

const draft = 'article.md';
const promptFile = 'rewrite-prompt.txt';

// Generate prompt
execSync(`humanizer.exe prompt ${draft} --out ${promptFile}`);

// Read and use with your preferred LLM API
const prompt = fs.readFileSync(promptFile, 'utf8');

// Example with environment variable for API key
const apiKey = process.env.OPENAI_API_KEY;
// ... make API call with prompt

Batch Processing

const { execSync } = require('child_process');
const { readdirSync } = require('fs');

const markdownFiles = readdirSync('.')
  .filter(f => f.endsWith('.md'));

markdownFiles.forEach(file => {
  console.log(`\n=== Checking ${file} ===`);
  try {
    execSync(`humanizer.exe check ${file}`, { stdio: 'inherit' });
  } catch (error) {
    // Non-zero exit code means patterns detected
  }
});

Pre-commit Hook

// .git/hooks/pre-commit (make executable)
#!/usr/bin/env node

const { execSync } = require('child_process');

const stagedMd = execSync('git diff --cached --name-only --diff-filter=ACM')
  .toString()
  .split('\n')
  .filter(f => f.endsWith('.md'));

if (stagedMd.length === 0) process.exit(0);

console.log('Checking staged markdown for AI patterns...\n');

let foundPatterns = false;

stagedMd.forEach(file => {
  try {
    execSync(`humanizer.exe check ${file}`, { stdio: 'inherit' });
  } catch (error) {
    foundPatterns = true;
  }
});

if (foundPatterns) {
  console.error('\n❌ AI patterns detected. Review or bypass with --no-verify');
  process.exit(1);
}

How It Works

humanizer-cli/
├── humanizer.exe         # 87 KB C binary, self-contained
├── SKILL.md              # Pattern database (Wikipedia sources)
├── package.json          # Zero dependencies
├── sources/
    ├── launch.mjs        # npx entry point
    ├── panel.mjs         # Interactive panel (Node)
    └── humanizer.cmd     # Launcher without Node

Binary: Pure C program, links only against Windows system DLLs (kernel32, msvcrt, user32)
SKILL.md: Embedded in binary as fallback, but external file takes precedence
No network: Everything runs offline
No dependencies: No npm packages, no bundled runtime

The panel UI (printed by Node) and the detection logic (C binary) are separate. You can replace the binary without breaking the panel, or run the binary without Node entirely.

Pattern Categories

Punctuation (Patterns 1-3)

Em-dashes instead of commas
Semicolons in casual writing
Curly quotes (typographic quotes)

Vocabulary (Patterns 4-7)

AI word list (realm, embark, delve, etc.)
Hedging ("arguably", "it's worth noting")
Padding ("the fact of the matter is")
Unnecessary intensity ("very unique")

Structure (Patterns 8-15)

"Not just X, it's Y"
Lists with bolded lead-ins
Emoji in headings
Rhetorical questions as openers
Signpost phrases ("let's dive in", "in today's world")

Content (Patterns 16-33)

Manufactured enthusiasm
Flattery ("you're absolutely right")
Chatbot artifacts ("as an AI language model")
Inflated significance
False balance
Vague gestures at context
And 11 more requiring human judgment

Troubleshooting

SmartScreen Warning

Windows SmartScreen blocks unsigned executables:

Click "More info"
Click "Run anyway"

Or download from GitHub releases (marked as "from a known publisher" after enough downloads).

Antivirus Quarantine

Some AV tools flag unsigned binaries. Restore from quarantine:

Windows Security → Protection history → Restore

Box Drawing Characters Broken

Console doesn't support UTF-8:

chcp 65001

Or use Windows Terminal (handles UTF-8 by default).

Exit Code Errors

(exit code 1)

Non-zero exit means:

Patterns detected (when checking)
File not found
Invalid pattern number
Malformed command

Check the error message above the exit code.

Pattern Not Detected

13 patterns are regex-based. The other 20 require human judgment:

Inflated significance
False balance
Manufactured controversy
Confident vagueness
etc.

Run humanizer.exe patterns to see which patterns are auto-detectable.

SKILL.md Not Found

.\humanizer.exe doctor

Shows which SKILL.md is being used. If custom location needed:

.\humanizer.exe check draft.md --skill path/to/SKILL.md

Colors Not Working

Force disable:

.\humanizer.exe patterns --no-color

Integration Examples

With VS Code Task

.vscode/tasks.json:

{
  "version": "2.0.0",
  "tasks": [
    {
      "label": "Check for AI patterns",
      "type": "shell",
      "command": "${workspaceFolder}/humanizer.exe",
      "args": ["check", "${file}"],
      "problemMatcher": [],
      "presentation": {
        "reveal": "always",
        "panel": "new"
      }
    }
  ]
}

With GitHub Actions

.github/workflows/check-ai.yml:

name: Check AI Patterns
on: [pull_request]
jobs:
  check:
    runs-on: windows-latest
    steps:
      - uses: actions/checkout@v3
      - name: Download humanizer
        run: |
          Invoke-WebRequest -Uri "https://github.com/0xwilliamortiz/humanizer-cli/releases/download/2.9.2/humanizer-v2.9.2.zip" -OutFile humanizer.zip
          Expand-Archive humanizer.zip
      - name: Check markdown files
        run: |
          Get-ChildItem -Recurse -Filter *.md | ForEach-Object {
            ./humanizer/humanizer.exe check $_.FullName
          }

As npm Script

package.json:

{
  "scripts": {
    "check:ai": "humanizer.exe check README.md",
    "check:docs": "for %f in (docs/*.md) do humanizer.exe check %f"
  }
}

Exit Codes

0: Success (no patterns detected, or command completed)
1: Patterns detected / file not found / invalid input
2: Invalid arguments

Limitations

Windows x64 only: Binary is platform-specific
13/33 patterns auto-detected: Rest need human review
English text: Patterns based on English AI writing habits
Markdown/text files: Not designed for code or structured data

Credits

Pattern catalog: blader/humanizer skill
Pattern research: Wikipedia:WikiProject AI Cleanup

License

MIT