Compatible with Claude, ChatGPT, Gemini, and Claude Code.
Humanizer
Strips out AI patterns. The text sounds like you.
Free, no account, no email — the command is all you need.
Already have it?
Nothing to install if it's already active
If this skill is already added in your Claude, there's no need to install it again. Just ask for what you want to do.
Not sure? Ask Claude: "which skills do you have available?"
“Use the humanizer skill to rewrite this text and strip out the AI-sounding phrasing: …”
1 · The simplest way
Copy and paste into any AI chat
This button copies the whole skill. Paste it into Claude, ChatGPT, or Gemini, then describe what you need. No installation required.
In a chat, Word or Excel files can't always be generated: you'll receive the document as structured text instead, and Claude will flag it.
2 · Command line
/humanizer installs in 3 steps
For Claude Code. The skill installs on your computer and stays available across all your projects.
Prerequisites: Claude Code + Node.js
The skill runs inside Claude Code, Anthropic's AI assistant installed on your computer. Node.js is the engine that runs the install command. Both are free.
Open PowerShell
Click Start, type "PowerShell", then press Enter.
Paste the command
Right-click inside the PowerShell window (or press Ctrl + V) to paste the command below, then press Enter. Wait a few seconds.
npx educasium --skill humanizer --globalWhat you should see:
npx educasium --skill humanizer --global
Fetching https://raw.githubusercontent.com/Chappygo-OS/claude-meta-skills/main/dist/humanizer.zip
…
Installed "humanizer". Restart Claude Code to pick it up.
Restart Claude Code
In Claude Code, type /restart (or quit and relaunch Claude Code). The skill now appears in the command list.
Verify it's installed (30 seconds)
Write in Claude Code:
/humanizer Our innovative solution will let you revolutionize your processes starting today.
If Claude runs the skill without an error message and produces the requested result, everything works. If not, the most likely causes are listed below.
Not working?
The most common causes
“npx” is not recognized
Your computer probably doesn't have Node.js. Download the LTS version from nodejs.org, run the installer, close and reopen PowerShell, then paste the command again.
The skill does not appear
Restart Claude Code (type /restart in the conversation, or quit and relaunch it). Check the command spelling: /humanizer. If needed, re-run the installation with --global at the end.
Describe the exact error message as it appears — the team will answer and guide you.
Contact the teamDon't use a terminal?
Using Claude.ai in your browser?
The assisted install works through a .zip file imported into Claude.ai. It takes longer and asks for your email.
What the skill does
View the source codeIn detail.
Paste any text. Claude scans 20 AI patterns (hollow openers, fluffy words, rule-of-three lists, em-dash overuse, generic closers) and rewrites the problematic passages. Shows exactly what changed and why.
When to use it
You drafted something with AI and it still sounds like a press release.
Example usage
/humanizer Here's my text to humanize...In depth
How this skill really works
Humanizer takes an already-written text — a LinkedIn post, an email, an article, an AI-generated draft — and rewrites it in two passes: it strips out the tells that signal an AI origin, then it adds personality back in (opinion, uneven rhythm, specificity). Output is always structured as two blocks: the ready-to-use humanized text, then a list of what changed and why, capped at 3-5 bullets.
If the user supplies a sample of their own real writing, the skill first extracts a voice profile — sentence length, use of I/we/you, vocabulary level, use of bullets — and anchors the rewrite to that profile instead of a generic style. On explicit request ("how AI does this sound?"), it can also return an AI-signature score out of 10, before and after the rewrite.
It's built on Wikipedia's "Signs of AI writing" guide (maintained by WikiProject AI Cleanup), whose full taxonomy of 8 categories rated by severity level (high / medium / low) is reproduced in references/patterns.md — that grid drives the detection, not a subjective sense that something "sounds AI".
When to use it
- After a first AI-assisted draft (article, LinkedIn post, email), before publishing.
- When the text chains typical AI words (leverage, seamless, robust, delve, cutting-edge...) or promotional adjectives (vibrant, nestled, breathtaking...).
- When the text overuses em dashes, perfectly symmetrical bullet lists, or generic closers ("the future looks bright").
- When the text chains artificial transition connectors ("Furthermore," "In conclusion," "Moreover,").
- When the user wants the text to match their own voice, by supplying a sample of their real writing.
- When the improvement needs to be objectified with a before/after AI-signature score.
When not to use it
- Not for checking the text's factual accuracy — that's the role of /fact-checker, to run first if the content needs validating on substance.
- Not for generating content from a blank page — the skill edits and rewrites text the user already supplied; it doesn't produce the underlying ideas or argument.
The method
The 8-category severity grid
references/patterns.md documents the full taxonomy of AI-writing markers across 8 categories, each rated high, medium, or low severity. It's this grid — not a subjective impression — that drives which passages get rewritten first.
- Filler openers (high)
- Opening lines that say nothing ("Great question!", "In today's fast-paced world...") — deleted entirely, the text starts with the first real content.
- Hype and jargon words (high)
- leverage, seamless, cutting-edge, robust, innovative, delve... replaced with a concrete description of what the thing actually does.
- Vague intensifiers (medium)
- "incredibly fast", "many businesses" — replaced with a specific number, name, or source.
- Structural tells (medium)
- Perfectly symmetrical bullet lists, systematic three-part structures, more than one em dash per 300 words.
- Hedge phrases (medium)
- "It's worth noting that", "As we know" — removed unless the hedge is genuinely warranted by real uncertainty.
- Transition padding (high)
- "Furthermore," "In conclusion," "Moreover," — deleted outright; "but" and "so" stay when the contrast or causation is genuine.
- Passive voice overuse (medium)
- Fixed once passive voice exceeds 20% of sentences, except when the actor is unknown or irrelevant.
- Future-tense closes (low)
- Generic promises ("this will change everything") replaced with a concrete present-tense capability, or cut.
Steps
Identify AI patterns
Screen the text against the 20 documented patterns, grouped into 4 families: content, language and grammar, style, communication and filler.
Detect the voice profile
If the user supplies a sample of their writing, extract their natural patterns first — sentence length, preferred connectors, vocabulary level, use of I/we/you, humor, directness — before any rewriting.
Rewrite problematic sections
Replace detected AI-isms with natural alternatives, following the rewrite rules per marker category.
Preserve meaning
Keep the core message, argument structure, personal anecdotes, named people/companies/data, and existing humor or personality signals intact.
Add soul
Don't just remove bad patterns: state an opinion rather than reporting neutrally, deliberately vary sentence rhythm, acknowledge complexity or mixed feelings, use "I" when it fits, be specific about feelings rather than vague.
Rules
- Always render output in the required format: a "## Humanized" block (ready-to-use text, no annotations) followed by "## What changed" (3-5 bullets max).
- Limit em dashes to 1 per 300 words.
- Switch to active voice once passive exceeds 20% of sentences, except when the actor is unknown or irrelevant.
- Never touch the elements marked to preserve: core message, argument structure, personal anecdotes, named people/companies/data, humor and personality.
- If the user supplies a sample of their writing, anchor the rewrite to that voice profile instead of a generic style.
- On explicit request ("how AI does this sound?"), provide a before/after AI-signature score out of 10, with the top offenders listed.
Mistakes to avoid
- Removing concrete data, named people/companies, or personal anecdotes instead of only AI-isms.
- Rewriting an already-clean text to artificially bolt on superficial "human" tics without checking whether it actually needs it.
- Softening a hollow opener instead of deleting it outright — a watered-down "In today's fast-paced world..." is still a hollow opener.
- Making every sentence the same length and structure after the rewrite — that recreates one of the very tics the skill is meant to eliminate.
- Keeping a neutral, opinion-free tone when the "Add soul" step explicitly calls for expressing an opinion or nuance.
What this skill doesn't do
- Does not verify the factual accuracy of claims in the text — that step stays with /fact-checker, to run first if needed.
- Does not generate content from nothing — it needs existing text (a draft, notes, AI output) to rewrite.
- Does not translate the text or change its language.
- Does not touch the final document's formatting (Word/PDF layout, styles) — it only produces the rewritten text and its changelog.
- Does not guarantee a humanized text becomes undetectable by an AI detector — it removes patterns statistically identified in a public guide, with no undetectability measurement provided or tested by the skill.
Skill contents — 4 fichiers · 29.5 KB(SKILL.md, references/examples.md, references/patterns.md, references/voice-profiles.md)
Download for free