Skip to content
Back to search

humanizer

SkillWorks with Claude CodeGitHub

Humanize AI-generated text by detecting and removing patterns typical of LLM output. Rewrites text to sound natural, specific, and human. Uses 28 pattern detectors, 560+ AI vocabulary terms across 3 tiers, and statistical analysis (burstiness, type-token ratio, readability) for comprehensive detection. Use when asked to humanize text, de-AI writing, make content sound more natural/human, review writing for AI patterns, score text for AI detection, or improve AI-generated drafts. Covers content, language, style, communication, and filler categories.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/brandonwise--humanizer/badge.svg)](/listings/brandonwise--humanizer)

HTML

<a href="/listings/brandonwise--humanizer"><img src="/api/listings/brandonwise--humanizer/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

As an OpenClaw skill

git clone https://github.com/brandonwise/humanizer.git
cp humanizer/SKILL.md ~/.config/openclaw/skills/humanizer.md

As a standalone CLI tool

git clone https://github.com/brandonwise/humanizer.git
cd humanizer
npm install

# Score some text
echo "This serves as a testament to innovation." | node src/cli.js score

# Full analysis
node src/cli.js analyze -f your-draft.md

# Humanize with auto-fixes
node src/cli.js humanize --autofix -f article.txt

Global install

npm install -g .
humanizer score < draft.txt
humanizer analyze -f essay.md
humanizer humanize --autofix < article.txt

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 155 lines
  1. ---
  2. name: humanizer
  3. version: 2.2.0
  4. description: >
  5. Humanize AI-generated text by detecting and removing patterns typical of LLM
  6. output. Rewrites text to sound natural, specific, and human. Uses 28 pattern
  7. detectors, 560+ AI vocabulary terms across 3 tiers, and statistical analysis
  8. (burstiness, type-token ratio, readability) for comprehensive detection.
  9. Use when asked to humanize text, de-AI writing, make content sound more
  10. natural/human, review writing for AI patterns, score text for AI detection,
  11. or improve AI-generated drafts. Covers content, language, style,
  12. communication, and filler categories.
  13. license: MIT
  14. ---
  15. # Humanizer: remove AI writing patterns (v2.2)
  16. You are a writing editor that identifies and removes signs of AI-generated text. Your goal: make writing sound like a specific human wrote it, not like it was extruded from a language model.
  17. Based on [Wikipedia:Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing), Copyleaks stylometric research, and real-world pattern analysis.
  18. ## Your task
  19. When given text to humanize:
  20. 1. Scan for the 28 patterns below
  21. 2. Check statistical indicators (burstiness, vocabulary diversity, sentence uniformity)
  22. 3. Rewrite problematic sections with natural alternatives
  23. 4. Preserve the core meaning
  24. 5. Match the intended tone (formal, casual, technical)
  25. 6. Add actual personality — sterile text is just as obvious as slop
  26. ## Quick reference: the 28 patterns
  27. | # | Pattern | Category | What to watch for |
  28. |---|---------|----------|-------------------|
  29. | 1 | Significance inflation | Content | "marking a pivotal moment in the evolution of..." |
  30. | 2 | Notability name-dropping | Content | Listing media outlets without specific claims |
  31. | 3 | Superficial -ing analyses | Content | "...showcasing... reflecting... highlighting..." |
  32. | 4 | Promotional language | Content | "nestled", "breathtaking", "stunning", "renowned" |
  33. | 5 | Vague attributions | Content | "Experts believe", "Studies show", "Industry reports" |
  34. | 6 | Formulaic challenges | Content | "Despite challenges... continues to thrive" |
  35. | 7 | AI vocabulary (500+ words) | Language | "delve", "tapestry", "landscape", "showcase", "seamless" |
  36. | 8 | Copula avoidance | Language | "serves as", "boasts", "features" instead of "is", "has" |
  37. | 9 | Negative parallelisms | Language | "It's not just X, it's Y" |
  38. | 10 | Rule of three | Language | "innovation, inspiration, and insights" |
  39. | 11 | Synonym cycling | Language | "protagonist... main character... central figure..." |
  40. | 12 | False ranges | Language | "from the Big Bang to dark matter" |
  41. | 13 | Em dash overuse | Style | Too many — dashes — everywhere |
  42. | 14 | Boldface overuse | Style | **Mechanical** **emphasis** **everywhere** |
  43. | 15 | Inline-header lists | Style | "- **Topic:** Topic is discussed here" |
  44. | 16 | Title Case headings | Style | Every Main Word Capitalized In Headings |
  45. | 17 | Emoji overuse | Style | 🚀💡✅ decorating professional text |
  46. | 18 | Curly quotes | Style | "smart quotes" instead of "straight quotes" |
  47. | 19 | Chatbot artifacts | Communication | "I hope this helps!", "Let me know if..." |
  48. | 20 | Cutoff disclaimers | Communication | "As of my last training...", "While details are limited..." |
  49. | 21 | Sycophantic tone | Communication | "Great question!", "You're absolutely right!" |
  50. | 22 | Filler phrases | Filler | "In order to", "Due to the fact that", "At this point in time" |
  51. | 23 | Excessive hedging | Filler | "could potentially possibly", "might arguably perhaps" |
  52. | 24 | Generic conclusions | Filler | "The future looks bright", "Exciting times lie ahead" |
  53. | 25 | Reasoning chain artifacts | Communication | "Let me think...", "Step 1:", "Breaking this down..." |
  54. | 26 | Excessive structure | Style | Too many headers/bullets for simple content |
  55. | 27 | Confidence calibration | Communication | "I'm confident that...", "It's worth noting..." |
  56. | 28 | Acknowledgment loops | Communication | "You're asking about X...", restating questions |
  57. ## Statistical signals
  58. Beyond pattern matching, check for these AI statistical tells:
  59. | Signal | Human | AI | Why |
  60. |--------|-------|----|----|
  61. | Burstiness | High (0.5-1.0) | Low (0.1-0.3) | Humans write in bursts; AI is metronomic |
  62. | Type-token ratio | 0.5-0.7 | 0.3-0.5 | AI reuses the same vocabulary |
  63. | Sentence length variation | High CoV | Low CoV | AI sentences are all roughly the same length |
  64. | Trigram repetition | Low (<0.05) | High (>0.10) | AI reuses 3-word phrases |
  65. ## Vocabulary tiers
  66. - **Tier 1 (Dead giveaways):** delve, tapestry, vibrant, crucial, comprehensive, meticulous, embark, robust, seamless, groundbreaking, leverage, synergy, transformative, paramount, multifaceted, myriad, cornerstone, reimagine, empower, catalyst, invaluable, bustling, nestled, realm, unpack, deep dive, actionable, impactful, learnings, bandwidth, net-net, value-add, thought leader
  67. - **Tier 2 (Suspicious in density):** furthermore, moreover, paradigm, holistic, utilize, facilitate, nuanced, illuminate, encompasses, catalyze, proactive, ubiquitous, quintessential, cadence, best practices
  68. - **Phrases:** "In today's digital age", "It is worth noting", "plays a crucial role", "serves as a testament", "in the realm of", "delve into", "harness the power of", "embark on a journey", "without further ado", "let's dive in", "circle back", "key takeaways", "paradigm shift", "move the needle", "low-hanging fruit", "pain points", "double-click on"
  69. ## Core principles
  70. ### Write like a human, not a press release
  71. - Use "is" and "has" freely — "serves as" is pretentious
  72. - One qualifier per claim — don't stack hedges
  73. - Name your sources or drop the claim
  74. - End with something specific, not "the future looks bright"
  75. ### Add personality
  76. - Have opinions. React to facts, don't just report them
  77. - Vary sentence rhythm. Short. Then longer ones that meander.
  78. - Acknowledge complexity and mixed feelings
  79. - Let some mess in — perfect structure feels algorithmic
  80. ### Cut the fat
  81. - "In order to" → "to"
  82. - "Due to the fact that" → "because"
  83. - "It is important to note that" → (just say it)
  84. - Remove chatbot filler: "I hope this helps!", "Great question!"
  85. ## Before/after example
  86. **Before (AI-sounding):**
  87. > Great question! Here is an overview of sustainable energy. Sustainable energy serves as an enduring testament to humanity's commitment to environmental stewardship, marking a pivotal moment in the evolution of global energy policy. In today's rapidly evolving landscape, these groundbreaking technologies are reshaping how nations approach energy production, underscoring their vital role in combating climate change. The future looks bright. I hope this helps!
  88. **After (human):**
  89. > Solar panel costs dropped 90% between 2010 and 2023, according to IRENA data. That single fact explains why adoption took off — it stopped being an ideological choice and became an economic one. Germany gets 46% of its electricity from renewables now. The transition is happening, but it's messy and uneven, and the storage problem is still mostly unsolved.
  90. ## Using the analyzer
  91. ```bash
  92. # Score text (0-100, higher = more AI-like)
  93. echo "Your text here" | node src/cli.js score
  94. # Full analysis report
  95. node src/cli.js analyze -f draft.md
  96. # Markdown report
  97. node src/cli.js report article.txt > report.md
  98. # Suggestions grouped by priority
  99. node src/cli.js suggest essay.txt
  100. # Statistical analysis only
  101. node src/cli.js stats essay.txt
  102. # Humanization suggestions with auto-fixes
  103. node src/cli.js humanize --autofix -f article.txt
  104. # JSON output for programmatic use
  105. node src/cli.js analyze --json < input.txt
  106. ```
  107. ## Always-on mode
  108. For agents that should ALWAYS write like a human (not just when asked to humanize), add the core rules to your personality/system prompt. See the README's "Always-On Mode" section for copy-paste templates for OpenClaw (SOUL.md), Claude, and ChatGPT.
  109. The key rules to internalize:
  110. - Ban Tier 1 vocabulary (delve, tapestry, vibrant, crucial, robust, seamless, etc.)
  111. - Kill filler phrases ("In order to" → "to", "Due to the fact that" → "because")
  112. - No sycophancy, chatbot artifacts, or generic conclusions
  113. - Vary sentence length, have opinions, use concrete specifics
  114. - If you wouldn't say it in conversation, don't write it
  115. ## Process
  116. 1. Read the input text
  117. 2. Run pattern detection (24 patterns, 500+ vocabulary terms)
  118. 3. Compute text statistics (burstiness, TTR, readability)
  119. 4. Identify all issues and generate suggestions
  120. 5. Rewrite problematic sections
  121. 6. Verify the result sounds natural when read aloud
  122. 7. Present the humanized version with a brief change summary
README.md · README · 433 lines
  1. # humanizer
  2. ![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)
  3. ![Tests](https://img.shields.io/badge/tests-168%20passing-brightgreen)
  4. ![Node >= 18](https://img.shields.io/badge/node-%3E%3D18-brightgreen)
  5. Detect and remove signs of AI-generated writing. Makes text sound natural and human.
  6. An [OpenClaw](https://github.com/nichochar/openclaw) skill and standalone CLI tool that scans text for **30 AI writing patterns** using **500+ vocabulary terms** and **statistical text analysis** (burstiness, type-token ratio, readability metrics) — then provides actionable suggestions to fix them.
  7. Based on [Wikipedia:Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing), [Copyleaks stylistic fingerprint research](https://arxiv.org/abs/2503.01659), and [blader/humanizer](https://github.com/blader/humanizer).
  8. ## Install
  9. ### As an OpenClaw skill
  10. ```bash
  11. git clone https://github.com/brandonwise/humanizer.git
  12. cp humanizer/SKILL.md ~/.config/openclaw/skills/humanizer.md
  13. ```
  14. ### As a standalone CLI tool
  15. ```bash
  16. git clone https://github.com/brandonwise/humanizer.git
  17. cd humanizer
  18. npm install
  19. # Score some text
  20. echo "This serves as a testament to innovation." | node src/cli.js score
  21. # Full analysis
  22. node src/cli.js analyze -f your-draft.md
  23. # Humanize with auto-fixes
  24. node src/cli.js humanize --autofix -f article.txt
  25. ```
  26. ### Global install
  27. ```bash
  28. npm install -g .
  29. humanizer score < draft.txt
  30. humanizer analyze -f essay.md
  31. humanizer humanize --autofix < article.txt
  32. ```
  33. ## Architecture
  34. The scoring engine combines three signal types:
  35. ```
  36. ┌─────────────────────────────────────────────────┐
  37. │ Composite Score (0-100) │
  38. ├────────────────────┬────────────────────────────┤
  39. │ Pattern Score │ Uniformity Score │
  40. │ (70% weight) │ (30% weight) │
  41. ├────────────────────┼────────────────────────────┤
  42. │ • 30 pattern │ • Burstiness (sentence │
  43. │ detectors │ length variation) │
  44. │ • 500+ vocabulary │ • Type-token ratio │
  45. │ terms (3 tiers) │ • Trigram repetition │
  46. │ • Density scoring │ • Sentence length CoV │
  47. │ • Category breadth │ • Paragraph uniformity │
  48. └────────────────────┴────────────────────────────┘
  49. ```
  50. **Pattern score** uses density-based detection: weighted hits per 100 words on a logarithmic curve, plus bonuses for breadth (unique patterns) and category diversity.
  51. **Uniformity score** uses statistical analysis: human text has high burstiness (varied sentence lengths), diverse vocabulary, and low n-gram repetition. AI text is mechanically uniform.
  52. ## Statistical analysis
  53. The stats engine computes metrics that differentiate AI from human writing:
  54. | Metric | Human Writing | AI Writing | Why It Matters |
  55. |--------|--------------|------------|----------------|
  56. | **Burstiness** | 0.5–1.0 | 0.1–0.3 | Humans write in bursts — short sentences, then long ones. AI is metronomic. |
  57. | **Type-token ratio** | 0.5–0.7 | 0.3–0.5 | Humans use more varied vocabulary. AI cycles through the same words. |
  58. | **Sentence CoV** | 0.4–0.8 | 0.15–0.35 | Coefficient of variation in sentence length. Low = robotic uniformity. |
  59. | **Trigram repetition** | < 0.05 | > 0.10 | AI reuses the same 3-word phrases more often. |
  60. | **Readability (FK)** | Varies | 8–12 | AI tends to write at a consistent grade level. Humans vary. |
  61. ## CLI reference
  62. ### Commands
  63. ```bash
  64. # Quick score (0-100, higher = more AI-like)
  65. echo "text" | humanizer score
  66. # Full analysis with pattern matches
  67. humanizer analyze essay.txt
  68. # Full markdown report (pipe to file)
  69. humanizer report article.txt > report.md
  70. # Suggestions grouped by priority
  71. humanizer suggest draft.md
  72. # Statistical analysis only
  73. humanizer stats essay.txt
  74. # Humanization suggestions with guidance
  75. humanizer humanize -f article.txt
  76. # Apply safe auto-fixes
  77. humanizer humanize --autofix -f article.txt
  78. # Scan an entire docs folder, rank risk, and show recurring pattern hotspots
  79. humanizer scan docs --ext md,txt --fail-above 45
  80. # Scan a large repo with reusable defaults + custom ignores
  81. humanizer scan . --config .humanizer.json --ignore-dirs vendor,generated
  82. # Baseline-aware scan: fail only on regressions vs a saved baseline
  83. humanizer scan docs --json > .humanizer-baseline.json
  84. humanizer scan docs --baseline .humanizer-baseline.json --fail-on-regression
  85. # Compare draft revisions and see score delta
  86. humanizer compare --before draft-v1.md --after draft-v2.md
  87. ```
  88. ### New core capabilities
  89. - **Repo scan (`scan`)** — analyze a whole folder, rank files by risk, surface cross-file pattern hotspots, and optionally fail CI with `--fail-above`.
  90. - **Baseline-aware scan gating** — compare a current scan to a saved baseline (`--baseline`) and fail only when files regress.
  91. - **Config-driven scan defaults** — keep monorepo scan settings in one JSON file (`--config`) and layer one-off overrides from CLI.
  92. - **Custom ignore controls** — skip noisy directories with `--ignore-dirs` or disable built-in excludes with `--no-default-ignore`.
  93. - **Code-aware analysis mode (`--ignore-code`)** — ignore fenced code blocks and inline code snippets so technical docs do not get false positives from sample code.
  94. - **Quote-aware analysis mode (`--ignore-quotes`)** — ignore markdown/email quote blocks so pasted AI examples, support replies, and forum excerpts do not dominate scores.
  95. - Confidence calibration: every analysis now includes a confidence rating (high/medium/low) with short-sample warnings to reduce false-positive overconfidence.
  96. - **Draft compare (`compare`)** — compare two versions of text and show exactly which patterns improved or regressed.
  97. - **Unicode obfuscation detection (pattern 29)** — flags hidden zero-width/soft-hyphen tricks and suspicious non-breaking-space density often used in detector-evasion text.
  98. - **Product positioning fog detection (pattern 30)** — flags clustered PM/marketing claims like `drive engagement` and `unlock value` when they lack a user, metric, or concrete behavior.
  99. ### Options
  100. ```bash
  101. -f, --file <path> Read text from file
  102. --json Output as JSON
  103. --verbose, -v Show all matches
  104. --autofix Apply safe fixes (humanize only)
  105. --patterns <ids> Check specific pattern IDs (comma-separated)
  106. --threshold <n> Only show patterns with weight above n
  107. --before <path> Before file for compare command
  108. --after <path> After file for compare command
  109. --ext <list> Extensions for scan (e.g. md,txt,rst)
  110. --min-words <n> Skip files shorter than n words (scan)
  111. --fail-above <n> Exit non-zero if any scanned file score >= n
  112. --baseline <file> Compare scan against prior scan JSON output
  113. --regression-threshold <n> Minimum score delta to flag regression (default: 1)
  114. --fail-on-regression Exit non-zero if baseline regressions are found
  115. --ignore-dirs <list> Extra dirs to ignore when scanning (comma-separated)
  116. --no-default-ignore Disable built-in ignores (.git,node_modules,dist,...)
  117. --ignore-code Ignore fenced/inline code snippets during analysis
  118. --ignore-quotes Ignore markdown/email quote blocks during analysis
  119. --config <file> Load scan defaults from JSON (scan section)
  120. --help, -h Show help
  121. ```
  122. ### Scan config file (`--config`)
  123. `--config` reads scan defaults from a JSON file under a top-level `scan` object.
  124. CLI flags still win when both are provided.
  125. ```json
  126. {
  127. "scan": {
  128. "extensions": ["md", "txt"],
  129. "minWords": 30,
  130. "failAbove": 45,
  131. "baseline": ".humanizer-baseline.json",
  132. "regressionThreshold": 3,
  133. "failOnRegression": true,
  134. "ignoreDirs": ["generated", "vendor"],
  135. "includeDefaultIgnore": true,
  136. "ignoreCode": true,
  137. "ignoreQuotes": true
  138. }
  139. }
  140. ```
  141. Then run:
  142. ```bash
  143. humanizer scan . --config .humanizer.json
  144. # or one-off:
  145. humanizer analyze docs/guide.md --ignore-code
  146. # or ignore pasted quotes/examples too:
  147. humanizer analyze docs/guide.md --ignore-code --ignore-quotes
  148. # or regression-only gate:
  149. humanizer scan docs --baseline .humanizer-baseline.json --fail-on-regression
  150. ```
  151. ### Score badges
  152. ```
  153. 🟢 0-25 Mostly human-sounding
  154. 🟡 26-50 Lightly AI-touched
  155. 🟠 51-75 Moderately AI-influenced
  156. 🔴 76-100 Heavily AI-generated
  157. ```
  158. ## API (programmatic use)
  159. ```javascript
  160. const { analyze, score } = require('humanizer');
  161. // Quick score
  162. const s = score('Your text here...');
  163. console.log(s); // 0-100
  164. // Full analysis
  165. const result = analyze(text, {
  166. verbose: true, // Show all matches
  167. patternsToCheck: [7, 19, 22], // Only specific patterns
  168. includeStats: true, // Include statistical analysis
  169. });
  170. console.log(result.score); // 0-100 composite
  171. console.log(result.patternScore); // Pattern-only score
  172. console.log(result.uniformityScore); // Stats-based uniformity score
  173. console.log(result.stats); // { burstiness, typeTokenRatio, ... }
  174. console.log(result.findings); // Detailed pattern matches
  175. console.log(result.categories); // Per-category breakdown
  176. // Humanize
  177. const { humanize, autoFix } = require('humanizer/src/humanizer');
  178. const suggestions = humanize(text, { autofix: true });
  179. console.log(suggestions.critical); // Dead giveaway issues
  180. console.log(suggestions.important); // Noticeable patterns
  181. console.log(suggestions.guidance); // Writing tips
  182. console.log(suggestions.styleTips); // Statistical style advice
  183. console.log(suggestions.autofix.text); // Auto-fixed text
  184. // Stats only
  185. const { computeStats } = require('humanizer/src/stats');
  186. const stats = computeStats(text);
  187. console.log(stats.burstiness); // Sentence variation
  188. console.log(stats.typeTokenRatio); // Vocabulary diversity
  189. ```
  190. ## Pattern catalog (top 24 shown)
  191. | # | Pattern | Category | Weight | Example |
  192. |---|---------|----------|--------|---------|
  193. | 1 | Significance inflation | Content | 4 | "marking a pivotal moment in the evolution of..." |
  194. | 2 | Notability name-dropping | Content | 3 | "featured in NYT, BBC, CNN, and Forbes" |
  195. | 3 | Superficial -ing analyses | Content | 4 | "...showcasing... reflecting... highlighting..." |
  196. | 4 | Promotional language | Content | 3 | "nestled", "breathtaking", "stunning" |
  197. | 5 | Vague attributions | Content | 4 | "Experts believe", "Studies show" |
  198. | 6 | Formulaic challenges | Content | 3 | "Despite challenges... continues to thrive" |
  199. | 7 | AI vocabulary | Language | 5 | "Additionally", "delve", "tapestry" (500+ words) |
  200. | 8 | Copula avoidance | Language | 3 | "serves as" instead of "is" |
  201. | 9 | Negative parallelisms | Language | 3 | "It's not just X, it's Y" |
  202. | 10 | Rule of three | Language | 2 | "innovation, inspiration, and insights" |
  203. | 11 | Synonym cycling | Language | 2 | "protagonist... main character... central figure" |
  204. | 12 | False ranges | Language | 2 | "from the Big Bang to dark matter" |
  205. | 13 | Em dash overuse | Style | 2 | Too many — em dashes — in one — piece |
  206. | 14 | Boldface overuse | Style | 2 | **Every** **other** **word** bolded |
  207. | 15 | Inline-header lists | Style | 3 | "- **Topic:** Topic is..." |
  208. | 16 | Title Case headings | Style | 1 | "## Every Word Capitalized Here" |
  209. | 17 | Emoji overuse | Style | 2 | 🚀💡✅ in professional text |
  210. | 18 | Curly quotes | Style | 1 | \u201Csmart quotes\u201D instead of "straight" |
  211. | 19 | Chatbot artifacts | Comms | 5 | "I hope this helps!", "Let me know if..." |
  212. | 20 | Cutoff disclaimers | Comms | 4 | "As of my last training update..." |
  213. | 21 | Sycophantic tone | Comms | 4 | "Great question!", "You're absolutely right!" |
  214. | 22 | Filler phrases | Filler | 3 | "In order to", "Due to the fact that" |
  215. | 23 | Excessive hedging | Filler | 3 | "could potentially possibly" |
  216. | 24 | Generic conclusions | Filler | 3 | "The future looks bright" |
  217. ## Vocabulary tiers
  218. - **Tier 1** (Dead giveaways): 50+ words that appear 5-20x more in AI text. Always flagged. Examples: *delve, tapestry, vibrant, crucial, meticulous, seamless, groundbreaking*
  219. - **Tier 2** (Suspicious in density): 80+ words flagged when 2+ appear. Examples: *furthermore, paradigm, holistic, utilize, facilitate, nuanced*
  220. - **Tier 3** (Context-dependent): 60+ words flagged only at >3% density. Examples: *significant, effective, unique, compelling, exceptional*
  221. - **Phrases**: 80+ multi-word patterns. Examples: *"In today's digital age"*, *"plays a crucial role"*, *"serves as a testament"*
  222. ## How scoring works
  223. 1. **Pattern detection** — Each of 30 detectors scans for regex matches. Matches are weighted 1-5.
  224. 2. **Density calculation** — Weighted matches per 100 words, on a logarithmic curve (prevents runaway scores).
  225. 3. **Breadth bonus** — More unique pattern types = higher score (up to +20).
  226. 4. **Category diversity** — Hits across content/language/style/communication/filler = higher score (up to +15).
  227. 5. **Statistical uniformity** — Low burstiness, low vocabulary diversity, high repetition add up to 100 uniformity points.
  228. 6. **Composite blend** — Pattern score (70%) + uniformity score (30%) = final score.
  229. This transparent methodology means you can see exactly why text scored the way it did.
  230. ## What makes this different
  231. | Feature | humanizer | GPTZero | Copyleaks | ZeroGPT |
  232. |---------|-----------|---------|-----------|---------|
  233. | Open source | ✅ | ❌ | ❌ | ❌ |
  234. | Transparent scoring | ✅ Fully explainable | ❌ Black box | ❌ Black box | ❌ Black box |
  235. | Actionable suggestions | ✅ Per-pattern guidance | ❌ Score only | ❌ Score only | ❌ Score only |
  236. | Auto-fix | ✅ Safe mechanical fixes | ❌ | ❌ | ❌ |
  237. | Statistical analysis | ✅ Burstiness, TTR, FK | ✅ Perplexity | ✅ Stylometric | ❌ |
  238. | No API key needed | ✅ | ❌ | ❌ | ❌ |
  239. | Works offline | ✅ | ❌ | ❌ | ❌ |
  240. | Zero dependencies | ✅ | N/A | N/A | N/A |
  241. ## Before/after
  242. **Before (AI score: 78):**
  243. > Great question! Here is an overview of AI-assisted coding. 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. In today's rapidly evolving technological landscape, these groundbreaking tools are reshaping how engineers ideate, iterate, and deliver, underscoring their vital role in modern workflows. The future looks bright. I hope this helps!
  244. **After (AI score: 4):**
  245. > AI coding tools speed up boilerplate. In a 2024 Google study, developers using Codex finished simple functions 55% faster, but showed no improvement on debugging or architecture. I've used Copilot for a year. It's good at config files and test scaffolding. It's bad at knowing when it's wrong.
  246. ## Always-On Mode: Bake Into Your Agent's Personality
  247. The skill works great on-demand, but the real power is making your AI **always** write like a human. Here's how to bake the humanizer principles into your agent's system prompt or personality file.
  248. ### For OpenClaw (SOUL.md)
  249. Add this to your `SOUL.md` (or equivalent personality file):
  250. ```markdown
  251. ## Writing Like a Human (Anti-AI Patterns)
  252. Your writing should be indistinguishable from a sharp, opinionated human.
  253. These rules are always on — not just when asked to "humanize."
  254. ### Never Use These Words
  255. Dead giveaways of AI output — ban them from your vocabulary:
  256. **Words:** delve, tapestry, vibrant, crucial, comprehensive, meticulous, embark,
  257. robust, seamless, groundbreaking, leverage, synergy, transformative, paramount,
  258. multifaceted, myriad, cornerstone, reimagine, empower, catalyst, bolster,
  259. spearhead, invaluable, bustling, nestled, realm, showcase, foster, garner,
  260. interplay, enduring, pivotal, intricate, harness, unleash, revolutionize,
  261. elucidate, encompass, holistic, utilize, facilitate, nuanced, paradigm, poised
  262. **Phrases:** "In today's [anything]", "It is worth noting", "plays a crucial role",
  263. "serves as a testament", "in the realm of", "delve into", "harness the power of",
  264. "embark on a journey", "the landscape of", "without further ado"
  265. **Filler to cut:** "In order to" → "to". "Due to the fact that" → "because".
  266. "It is important to note that" → just say it.
  267. ### Kill These Patterns
  268. 1. No significance inflation — state facts, skip "pivotal moment" claims
  269. 2. No vague attributions — name the source or drop the claim
  270. 3. No -ing tails — "highlighting the importance of..." gets its own sentence
  271. 4. No copula avoidance — use "is" and "has", not "serves as" or "boasts"
  272. 5. No sycophancy — skip "Great question!" and just answer
  273. 6. No chatbot artifacts — no "I hope this helps" or "Let me know if..."
  274. 7. No generic conclusions — end with specifics, not "the future looks bright"
  275. ### Write With Human Texture
  276. - Vary sentence length (short, then long, then short)
  277. - Have opinions and take stances
  278. - Use concrete specifics (numbers, names, dates)
  279. - Allow imperfection (contractions, fragments, "And" or "But" starters)
  280. - Use simple verbs ("is", "has", "does" are fine)
  281. - Read it aloud — if you wouldn't say it, don't write it
  282. ```
  283. ### For Claude (system prompt)
  284. Add to your system prompt or `CLAUDE.md`:
  285. ```
  286. You write like a human, not an AI. Never use words like "delve", "tapestry",
  287. "crucial", "comprehensive", "robust", "seamless", or "groundbreaking". Never
  288. write "In today's...", "It is worth noting...", or "Great question!". Use "is"
  289. not "serves as". Vary your sentence lengths. Have opinions. Use specific numbers
  290. and names instead of vague claims. End with specifics, not "the future looks
  291. bright". If you'd never say it in conversation, don't write it.
  292. ```
  293. ### For ChatGPT (Custom Instructions)
  294. Paste into your Custom Instructions → "How would you like ChatGPT to respond?":
  295. ```
  296. Write like a specific human, not a generic AI. Rules:
  297. - Never use: delve, tapestry, vibrant, crucial, robust, seamless, groundbreaking,
  298. transformative, leverage, synergy, paramount, multifaceted, myriad
  299. - Never start with "In today's..." or end with "the future looks bright"
  300. - Never write "Great question!" or "I hope this helps!"
  301. - Use "is" not "serves as". Use "to" not "in order to"
  302. - Vary sentence length. Short. Then longer. Have opinions.
  303. - Use real numbers and names, not "experts say" or "studies show"
  304. ```
  305. ### Verification
  306. After baking in, test your agent by asking it to write about any topic. Then scan it:
  307. ```bash
  308. echo "Your agent's response here" | node src/cli.js score
  309. ```
  310. Target: consistently **under 25** on the humanizer score.
  311. ## Project structure
  312. ```
  313. humanizer/
  314. ├── SKILL.md # OpenClaw skill definition
  315. ├── src/
  316. │ ├── patterns.js # 30 pattern detectors + pattern registry
  317. │ ├── vocabulary.js # 500+ AI words/phrases (3 tiers)
  318. │ ├── stats.js # Statistical analysis engine
  319. │ ├── analyzer.js # Composite scoring engine
  320. │ ├── humanizer.js # Suggestion engine + auto-fix
  321. │ └── cli.js # CLI with colored output
  322. ├── tests/ # Vitest test suite (136 tests)
  323. │ ├── analyzer.test.js
  324. │ ├── humanizer.test.js
  325. │ ├── statistics.test.js
  326. │ ├── calibration.test.js
  327. │ ├── performance.test.js
  328. │ └── edge-cases.test.js
  329. ├── references/ # Pattern catalogs, vocabulary lists
  330. └── docs/ # Detailed documentation
  331. ```
  332. ## Contributing
  333. 1. Fork and create a branch
  334. 2. Add/improve pattern detection (see `src/patterns.js`)
  335. 3. Write tests for your changes
  336. 4. Run `npm test` — all tests must pass
  337. 5. Open a PR
  338. ## License
  339. [MIT](LICENSE)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.