What Is the Humanizer Skill? Making Claude-Written Text Sound Natural, Installation, Turkish Support and Cost

You paste an announcement, blog post or email written with AI; Humanizer looks for 26 patterns that make text feel "written by a robot" and rewrites it while keeping the meaning and facts. It is a free skill made of a single instruction file; you install it in Claude Code as a plugin, in claude.ai as a ZIP, and in Codex with a command. It writes for the human reader, not to get past AI detectors.

What it does: Removes AI patterns from text
Type: Skill (single instruction file)
Where: Claude Code, claude.ai, Claude Desktop, Codex
Turkish text: Not stated in the official sources
Beats detectors?: No, that's not the goal
Latest version: 3.1.0 (28.09.2026)
Cost: Free, MIT licensed

What is Humanizer for?

An announcement, blog post or email you have Claude or ChatGPT write often comes out with the same patterns: "This isn't just X, it's Y", one-line aphorisms wrapping up every paragraph, headings bolded for decoration, intros that start with "Great question!". Humanizer is an instruction file (a skill) that looks for these patterns, flags them and rewrites the text. A skill is a ready-made instruction set that tells the assistant how to do a specific job.

Making AI text sound naturalWhat you get: For pasted text: a first draft, a short list of remaining patterns, and the final text; in file mode, only the final prose is written to the fileNeeds a separate setup

You give it the textPaste it or type the file name.

It flags the patternsIt rewrites starting from the strongest signs and checks the draft against the patterns.

It delivers the resultA first draft, a short list of remaining patterns, and the final text.

It doesn't add names, numbers, dates, quotes or sources that aren't in the text or that you don't have; it asks for missing information.

Who it's for: anyone who drafts with AI and wants the final version closer to their own voice: content creators, consultants, teachers, business owners. What it doesn't do: it doesn't pad the text with made-up facts and doesn't promise to get past AI detection tools (see below).

What is a skill?

A skill is a ready-made instruction file that tells the assistant how to do a job. Once installed, the assistant specializes in that job: for Humanizer, the job is finding AI patterns in text and rewriting it.

SKILL.mdname: humanizerdescription: AI patterns in text…────────1. Start with the strongest signs2. Keep the meaning, add no facts3. Check the draft against the patterns…

When it kicks in: the assistant loads it itself when you type /humanizer or say "humanize this text".
What's inside: only this instruction file; no program, hooks or internet connection.

The file preview is a shortened summary; the full text is in Humanizer's GitHub repository.

How does Humanizer change text?

AI's first draftIllustrative

Great question! Brewing coffee isn't just a habit, it's an art. The right grind, the right temperature and the right time (all three together) are the key to the perfect cup. In short: coffee is a journey.

With the patterns removedIllustrative

For a good cup, watch three things: grind the beans to suit your brewing method, let the water sit for a moment after it boils, and keep the brewing time short, adjusting to taste.

The highlighted parts are among the patterns Humanizer looks for. The example was written to illustrate the patterns; it is not actual output from the skill.

How do you install Humanizer?

No account, API key or separate program is needed; the assistant that runs the skill is enough. The installation path depends on where you use it; the commands are taken verbatim from the official README.

1. Add the marketplace› /plugin marketplace add blader/humanizer

2. Install the plugin› /plugin install humanizer@humanizer

Inside a session, the second command doesn't install right away; it opens the plugin's details panel. Choose the scope there: all your projects, this project only, or only you in this project.
If the panel says "Run /reload-plugins to apply", run that command.
The plugin is called as /humanizer:humanizer. The README requires Claude Code 2.1.142 or newer for the plugin route.

For an older Claude Code version› npx skills add blader/humanizer --global --agent claude-code

This marketplace counts as third-party; according to the Claude Code documentation, auto-update is off by default for third-party marketplaces, so you update to new versions yourself from the /plugin panel.

For Codex› npx skills add blader/humanizer --global --agent codex

The README says to drop --global if you want to install it only in the current project. The Codex documentation describes invoking skills manually with $.

All agents the Skills CLI knows› npx skills add blader/humanizer --global --agent '*'

The README says this command also installs it in agents such as Gemini CLI, GitHub Copilot and Windsurf; those agents' own documentation doesn't mention Humanizer. For an agent the Skills CLI doesn't recognize, you need to copy the SKILL.md file into the agent's skills folder.

The Skills CLI sends anonymous usage data during installation (skill name, files, timestamp; no personal data). To turn it off, put DISABLE_TELEMETRY=1 in front of the command.

How do you use Humanizer?

After installing, you can call the skill manually; according to the Claude Code documentation, Claude can also load the skill on its own for a task that matches its description, and Humanizer has no setting to turn that off. So just saying "humanize this text" is enough.

Pasting the text"/humanizer Rewrite the following text: [paste the text here]"

Editing a file"Humanize the prose in docs/launch-post.md"In file mode only the prose changes; code, data, YAML front matter and link URLs stay the same. The file is overwritten, so keep it under version control to be able to undo the change.

In your own voice"Below are two paragraphs of my own writing. Rewrite the draft to match this rhythm and word choice: [sample] [draft]"README: it follows the sample's sentence length, word choice and punctuation (including how often dashes are used). You provide the sample each time; nothing says it learns it permanently.

Calling it manually in Codex› $humanizer

On your first try, start with a short paragraph and compare the result with the original; check in particular that names, numbers and dates stayed the same.

Which patterns does it look for?

The skill looks for 26 patterns in six groups. The first five patterns count as the strongest signs and are fixed on sight; some count as "weak on their own" and are fixed only together with other signs. The summary below is compiled from SKILL.md's own groups.

Group	What it looks for
Staging instead of telling	The "This isn't just X, it's Y" contrast, a one-line closer that repeats the previous sentence, profound-sounding lines, "let's dive in" warm-ups before getting to the point
Forced rhythm	Forced lists of three, consecutive sentences starting with the same word, em dashes everywhere, stacked hedges
Inflation and borrowed authority	Overused fancy words, presenting an ordinary fact as a turning point, nameless authority like "according to experts"
Formatting	Bold text as decoration, capital letters and emoji in every heading, curly quotes instead of straight quotes
Chat leftovers	"Great question!" and "I hope this helps" wrappers, a first sentence that repeats the heading, sentences in which the document describes itself (these are deleted)
Writing for the wrong reader	In a reply within a conversation, re-explaining background the reader already knows and leaving the decision for the end

The pattern lists are written with English phrases. The one exception: for the "This isn't just X, it's Y" construction, SKILL.md says "the same construction is handled the same way in every language".

Does Humanizer get past AI detectors?

No, that's not the goal. Humanizer's README says so clearly: the tool edits for the reader, getting past AI detection tools is not the aim, and these tools will still flag most of the output.

Reported by users At least five users on GitHub reported that the output still looks like "AI" in a detector; in his replies the developer points to the README.
So use the tool to make the text better for the reader; don't run the result through a detector and rely on it.

Source: README introduction and GitHub issue #229 (developer replies). Rules on disclosing AI use (school, workplace, publishing) vary by institution; check the rule with your own institution.

Is it the same as paid "AI humanizer" sites?

No. When you search for "AI humanizer", you mostly get other products with the same name that have nothing to do with this skill. Humanizer has no connection to them.

Humanizer is not a website or app, but a single instruction file you install in your assistant.
No account, credits or subscription system; the skill is free and MIT licensed.
No promise to beat detectors; the README says it doesn't consider that a goal.
It doesn't send text to a Humanizer server; the text goes to the provider of the assistant you use (Anthropic, OpenAI, etc.) like any of your messages.

On October 14, 2026, the titles of the top ten results of a Turkish-language YouTube search for "ai humanizer" were promotions of other products. Those products are not reviewed on this page; check their prices and claims on their own sites.

Which programs does Humanizer work in?

Claude Code (plugin or skill)Cleaning text, a README or documentation written in Claude Code of AI patterns.Terminal, Code editor, Desktop · Claude subscription or Anthropic Console account; Claude Code 2.1.142 or newer for the plugin · details

claude.ai and Claude Desktop (ZIP upload)Rewriting text you paste into the chat with the same instructions.Web, Desktop · Code execution and file creation on in settings; admin permission on Team/Enterprise · details

CodexEditing text in Codex with the same instructions.Terminal, Code editor · Codex plan not clear in the official sources; install with the command in the README · details

Other agents the Skills CLI knowsUsing the same skill in agents such as Gemini CLI, GitHub Copilot and Windsurf.Terminal, Code editor · Stated only in the README; these agents' own documentation doesn't mention it · details

If you install the plugin in Claude Code on your computer, it doesn't load in Claude Code's cloud sessions (claude.ai/code); there, the skills enabled in your claude.ai account apply.

Does Humanizer work on Turkish text?

Fixing Turkish textUnclear	Not stated in the official sources. The README and SKILL.md don't mention Turkish; no results for Turkish are published.
Turkish-specific pattern listNo	The pattern lists are written with English phrases (words like "delve", "testament"). There is no separate list for Turkish writing habits.
Adaptations for other languagesNo	The developer doesn't merge adaptations such as French, Spanish or Chinese into this repository; he wrote that he prefers them to stay in separate community repositories (issue #163). There is no request or adaptation for Turkish in this repository.
InterfaceNo	The skill has no interface of its own; you use Claude's or Codex's interface.

If you want to try it on Turkish text, start with a short paragraph, compare the result with the original and make sure the meaning hasn't changed. If it doesn't help, you lose nothing since it's free; but don't assume it has been proven to work.

Does it access my files or the internet? Security

What it reads and writes

Reads: the text you paste, the file you name, your own writing sample if you give one
Writes: a reply in the chat; in file mode, only the prose of the file you name
Doesn't delete; it has no hook, MCP server, agent or any part that runs commands

Does it go online?

SKILL.md has no step that runs commands, downloads files or calls URLs. But the text you give goes to the assistant's provider like every message you write to the assistant. If it's private text, that provider's data rules apply.

Recorded security warnings

No security advisory has been published on GitHub. skills.sh audit page (September 28, 2026): two audits "passed", Snyk "warning" (medium; the reason is not the skill's instructions but a CI workflow file in the repository). This is not a security certification.

Updates

If you installed it from the marketplace, it doesn't update automatically; you update it yourself via /plugin › Installed › Update now. If you installed it with the Skills CLI, you run the command again for a new version.

This information is compiled from the files in the official repository; it is not a security audit. Before pasting confidential or client text, check the data rules of the assistant you use.

Is Humanizer free?

The skill itself is free and open source (MIT license); no account, credits or subscription. Whether a paid plan is needed depends not on the skill but on the assistant you run it in.

Where you use it	What you need
claude.ai and Claude Desktop	According to Claude's support page, skills are available on the Free, Pro, Max, Team and Enterprise plans; code execution and file creation must be on
Claude Code	A Claude subscription or a pay-as-you-go Anthropic Console account; a paid subscription in the desktop app
Codex	Which Codex plans have skills enabled is not stated here; check the Codex documentation

Allowed

Use, copy, modify and distribute, including commercially (MIT)

Not allowed

Removing the copyright and license notice from copies
The software is provided "as is"; the developer gives no warranty

Source: LICENSE (MIT, copyright holder Siqi Chen), the Claude help center and Claude Code documentation. The rewritten text is your text; the terms of the assistant you run it in also apply. This is not legal advice.

Common problems and fixes

What happens	What to do
The output still shows as "AI" in a detector	Expected: the skill wasn't made to beat detectors. Use it to improve the text for the reader.
"unexpected key in SKILL.md frontmatter" or "symbolic link" error when uploading to claude.ai	An issue seen in older versions; the developer fixed it in versions 2.8.3, 2.9.1 and 2.11.2. Download the current version (3.1.0) as a ZIP and upload it.
The rewritten text has a date, name or number that wasn't in the original	The current instructions forbid adding facts (added in 2.9.0), but still compare the result with the original; on first use, check numbers and names in particular.
The `/plugin install` command didn't install the plugin right away	Inside a session this command opens the details panel; you need to choose the scope and confirm. Then run `/reload-plugins`.
The `/humanizer:humanizer` command isn't recognized	The plugin route requires Claude Code 2.1.142 or newer. If you're on an older version, install with `npx skills add blader/humanizer --global --agent claude-code`; the skill responds to `/humanizer`.
A new version is out but you still have the old one	A plugin installed from the marketplace doesn't update automatically: /plugin › Installed › Humanizer › Update now.

How do you uninstall Humanizer?

/plugin › Installed › Humanizer › Uninstall; or the command below in the terminal.

Uninstall the plugin› claude plugin uninstall humanizer@humanizer

Also remove the marketplace (removes all plugins installed from it)› claude plugin marketplace remove humanizer

If you installed it with the Skills CLI: the skills.sh documentation doesn't give an uninstall command. According to the Claude Code documentation, personal skills live in the ~/.claude/skills/<skill-name>/ folder.

According to the Codex documentation, to disable it without deleting, add path and enabled = false to the skill's [[skills.config]] entry in ~/.codex/config.toml and restart Codex.

Frequently asked questions

What is Humanizer and what does it do?

A free skill made of a single instruction file that cleans AI-written text of the patterns that make it look "written by a robot". It checks 26 patterns, keeps the meaning, doesn't invent facts, and rewrites the text.

How do you install Humanizer in Claude Code?

In Claude Code, type /plugin marketplace add blader/humanizer, then /plugin install humanizer@humanizer. The second command opens the details panel; it installs once you choose the scope. The plugin is called as /humanizer:humanizer and, according to the README, requires Claude Code 2.1.142 or newer.

How do you use Humanizer in claude.ai?

Download the repository from GitHub as a ZIP and upload it via Customize > Skills > Create skill > Upload a skill. "Code execution and file creation" must be on in the settings. According to Claude's support page, skills are available on all plans, Free included.

Does Humanizer work on Turkish text?

Not stated in the official sources. The README and SKILL.md don't mention Turkish, and the pattern lists are written with English phrases. The developer keeps adaptations for other languages in separate community repositories. For Turkish text, try a short paragraph first.

Are Humanizer and paid "AI humanizer" sites the same thing?

No. Humanizer is free, MIT licensed and only an instruction file; there is no account or credit system and no promise to beat detectors. The paid sites that come up when you search for "AI humanizer" are different products.

Is Humanizer free?

Yes, the skill is free and open source (MIT). Whether you need a paid plan depends on where it runs: claude.ai includes Free, while Claude Code requires a paid Claude plan or an API account.

Does Humanizer get past AI detectors?

No, that's not the goal. The README says the tool edits for the reader, beating detectors is not the aim, and detectors will still flag most of the output.

Do I have to type /humanizer every time?

No. According to the Claude Code documentation, you can call the skill manually, or Claude loads it on its own for a matching task; Humanizer has no setting to turn off automatic loading. Saying "humanize this text" is enough.

Related guides

Sources

Not clear in the official sources: Humanizer's results on Turkish text (not stated in the official sources) · which Codex plans it works on · whether it works in Gemini CLI, Copilot and Windsurf (only the README says so) · whether the ZIP downloaded from GitHub currently uploads to claude.ai without issues. Each is covered in its section along with what you can do.