Humanizer

# Humanizer

Rewrites AI-sounding text so it reads like a person wrote it, keeping the meaning and facts unchanged. Checks 26 patterns from Wikipedia's AI-writing guide. For writers using Claude Code, Codex or other agents.

Made by blader on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

## About

What it is: Humanizer is a skill that edits AI-written prose until it sounds human, and it keeps your meaning intact. It is built on Wikipedia's guide to the signs of AI writing, which editors use to spot machine-written text. It checks for 26 patterns in all.

Who it suits: Anyone who drafts with AI and then has to publish the result under their own name. It edits for human readers, so it fits writers, marketers and teams that want a clean final version.

```table
Industry | Who uses it | How they use it | Example
Marketing | Content and social teams | Run AI drafts through it before publishing | A launch announcement with the hype stripped out
Software | Developers writing docs and posts | Point it at a file so only the prose changes | A README paragraph that reads less like a template
Publishing | Editors and bloggers | Match a rewrite to their own voice using a sample | A blog post tuned to the author habits
Business | Founders and managers | Clean up emails and updates written with a chatbot | A team update that sounds like the manager
```

Anyone who wants to hide AI use from detectors can skip it, because the author says that is not a goal.

Where to use it: It works in agents that support skills, and you call it with a slash command once installed.

- Claude Code, as a plugin
- Codex
- Claude.ai and Claude Desktop, as an uploaded skill
- Windsurf, GitHub Copilot, Gemini CLI and more, through the Skills command line tool

```table
Title: Jobs the skill does
Use case | Who | What they get back
Rewrite pasted text | Writers | A first rewrite, a short critique and a final version
Edit a file | Developers and editors | The prose changed, with code, data and link targets left alone
Match your voice | Anyone with a writing sample | A rewrite that follows your rhythm, word choice and punctuation
```

How to set it up: Choose the route for the tool you use. The plugin route in Claude Code needs version 2.1.142 or newer.

- Claude Code 2.1.142 or newer for the plugin route
- Node.js, for the npx routes

1. In Claude Code, run the two plugin lines below. The plugin then answers to /humanizer:humanizer.
2. In older Claude Code versions, use the npx command for Claude Code below.
3. In Codex, use the Codex npx command. For all supported agents, use the last command.
4. For Claude.ai or Claude Desktop, grab the project as a ZIP archive and add it under skills in your settings.
5. If your agent is not on that list, place the SKILL.md file in the folder where it keeps skills.

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

```bash
npx skills add blader/humanizer --global --agent claude-code
```

```bash
npx skills add blader/humanizer --global --agent codex
```

```bash
npx skills add blader/humanizer --global --agent '*'
```

Leaving off the --global option installs it only in the current project.

How to use it: Once installed, the skill answers to /humanizer. Paste your text after it, or ask in everyday words, or give a file path.

```text
/humanizer

[paste your text here]
```

```table
Title: Ways to call it
Tool | What it does
/humanizer | Runs the skill on the text you paste after it
Plain request | Asks for the same rewrite in everyday words
File request | Rewrites only the prose in a file and leaves code, data, frontmatter and link targets alone
Voice sample | Matches the rhythm, word choice and punctuation of two or three paragraphs you supply
```

Example: You paste a chatty product update full of hype and fragments, then run the skill. It shows a first rewrite, a short note on anything that still sounds artificial, and a final version. The final text keeps your facts and drops the staged phrasing.

For newcomers and experts: If you are new, paste one paragraph you wrote with an AI and run the skill. Compare the final version with the original and see which phrases changed.

If you know the field, the premise is that a language model picks the most likely wording for the widest audience, while a person writes for one reader. Each tell is a form of that default choice. The skill lists the tells it finds from strongest to weakest and writes a new draft that may restructure the text. It then compares that draft with the pattern list and with the claims in your original.

```flow
It marks the tells, strongest first
It drafts a rewrite
It critiques what still sounds artificial
It writes the final version
```

Good to know:

- License: MIT.
- It does not invent details. A name, number, date or quote must come from the source or you, and it asks when one is missing.
- Detectors still flag most of its output, so do not use it to get past them.
- The author reports that blind-test judges preferred its rewrite 16 times out of 16.
- Personal writing keeps your opinions and quirks, while technical prose stays neutral and plain.
- It reads and changes only the files you point it at.

## On GitHub, 15 November 2026

- Stars: 69,148
- Forks: 5,439
- Watching: 288
- Open issues and pull requests: 4
- Language: Python
- License: MIT
- Topics: agent-skills, ai-humanizer, ai-writing, chatgpt, claude, claude-code, codex, cursor, humanize-ai-text, humanizer, llm, prompt-engineering

## Links

- Code: https://github.com/blader/humanizer
- Website: https://skills.sh/blader/humanizer
- Releases: https://github.com/blader/humanizer/releases
- Issues: https://github.com/blader/humanizer/issues
- Star history: https://www.star-history.com/#blader/humanizer&Date

## Get the code

```
git clone https://github.com/blader/humanizer.git
```

Then follow the setup in its README. Nothing in this kit runs by itself.

README.md  6 KB