Humanizer skill for Claude

Drafts written with AI help often share the same habits: a dramatic one-line closer, three adjectives where one would do, "it's not just X, it's Y". Humanizer is a style editor packaged as a skill. You give it a draft, and it rewrites those patterns so the text reads like a careful person wrote it, while keeping every claim, number and quote. This page covers what it checks, how to install it and where it stops. For how skills work in general, see our Claude Skills manual.

What the skill does

The author says Humanizer is built on Wikipedia's "Signs of AI writing", the guide Wikipedia editors use to spot machine-generated text. It checks 26 patterns in six groups:

• Staging instead of stating: "not X but Y" lines, one-line closers, sayings that sound deep, run-ups like "Here's the thing".
• Rhythm by rule: forced triads, repeated sentence openings, dashes as the default connector, stacked qualifiers.
• Inflation and borrowed authority: overused words, inflated significance, sales language, vague "experts say" claims.
• Formatting by rule: bold used as decoration, decorative headings and curly quotation marks.
• Leftovers from chat: assistant phrases and disclaimers left in the text.
• Writing for the wrong reader: re-explaining what the reader already knows.

The SKILL.md runs in four steps: mark the patterns it finds, strongest first; draft a rewrite; audit the draft by asking what still sounds generated and whether every original claim survived; then write the final version. It keeps names, numbers, dates, quotes and citations from the source. If a sentence needs a detail that is missing, it asks you instead of inventing one. When you point it at a file, it changes only the prose and leaves code, data, front matter and link targets alone.

Who publishes it

A community author, GitHub user blader, in the blader/humanizer repository under the MIT license. Anthropic does not publish it. As with any third-party skill, read SKILL.md before you install it.

Where it works

The README lists Claude Code, Claude.ai, Claude Desktop, OpenAI Codex, and other agents through the Skills CLI, including Gemini CLI, GitHub Copilot and Windsurf.

How to install it

Claude Code (version 2.1.142 or newer):

/plugin marketplace add blader/humanizer
/plugin install humanizer@humanizer

The plugin answers to /humanizer:humanizer. On older versions, the README gives npx skills add blader/humanizer --global --agent claude-code.

Claude.ai and Claude Desktop: download the repository as a ZIP (Code, then Download ZIP) and upload it as a skill. In Claude.ai that is Customize > Skills, then "+", "+ Create skill" and "Upload a skill". Skills need code execution turned on.

Codex: npx skills add blader/humanizer --global --agent codex. Leave off --global to install it for the current project only.

How to use it

Call it and paste your text, or ask in plain words: "Please humanize this text". When you paste text, it shows the first rewrite, a short critique of what still sounds artificial, and the final version. To make the edit sound like you, add two or three paragraphs of your own writing. The skill then follows your sentence length, word choice and punctuation, including dashes if you use them.

/humanizer

Here's a sample of my writing for voice matching:
[2-3 paragraphs you wrote yourself]

Now edit this draft. Keep every fact, number and quote:
[your draft]

When it helps

• Cleaning up your own first draft of a blog post, newsletter or product announcement written with AI help.
• Editing documentation or reports into plain, neutral prose.
• Bringing a team's drafts closer to one house voice by giving it a writing sample.

If you do not want to install anything, our prompts for making AI text sound natural do a lighter version of the same edit in any chat, and the writing prompts cover drafting from the start.

Limits

• It edits style only. It does not check facts. If the draft is wrong, the rewrite is wrong in better prose.
• It is written for human readers. The author says the goal is clearer prose for people, and that passing AI detectors is not one of its aims. Judge the result by whether it reads well. For tools built for other writing jobs, see our AI writing tools guide.
• Some patterns are judgment calls. The README marks several patterns as weak on their own, because a careful writer may use any of them on purpose. Review the edit before you publish it.

A similar skill: stop-slop

Stop-slop, by Hardik Pandya (hardikpandya/stop-slop, MIT), is a smaller skill with the same aim. It ships SKILL.md with three reference files: phrases to cut, structures to avoid, and before-and-after examples. The README says to add the folder as a skill in Claude Code, upload the files to a Claude project's knowledge, or paste the core rules into custom instructions.

More skills, from Anthropic and the community, are in the Claude Skills catalog.

FAQ

Is the humanizer skill made by Anthropic?

No. It is a community skill by GitHub user blader, released under the MIT license.

Will it change the meaning of my text?

Its rules say to keep every supported claim, number, date, quote and citation, and to ask when a detail is missing. Still compare the result with your original.

Can it write in my style?

Yes, if you give it a sample. The README suggests two or three paragraphs of your own writing.