Humanizer

Humanizer is an agent skill, distributed as a plain Markdown file, that rewrites AI-generated text to remove the tells that make it sound like an LLM wrote it. It targets developers and writers who use AI agents (Claude Code, Cursor, Codex, or any harness that supports skill-style instructions) and want output that reads like a person wrote it, not a chatbot.

The skill is built on a catalog of 33 patterns adapted from Wikipedia's "Signs of AI Writing" guide, maintained by WikiProject AI Cleanup. Each pattern comes with a before/after example, covering things like promotional language, hedging, em dashes, rule-of-three lists, and chatbot artifacts like "I hope this helps!" Because it's just a SKILL.md file, it works in any harness that reads skill directories, and it can be installed globally or per-project.

Key features

• 33 documented patterns: detects content, language, style, and communication tells common in AI writing, each with a before/after example.
• Two-pass rewrite: runs a final "obviously AI generated" audit and a second rewrite to catch lingering issues from the first draft.
• No-fabrication rule: rewrites never add facts, names, dates, or citations not present in the source text; specificity has to come from the source or the author.
• Voice calibration: accepts 2-3 paragraphs of your own writing and adapts the rewrite to match your sentence rhythm and word choices instead of producing generic clean output.
• Multiple invocation modes: paste text directly, point it at a file to rewrite in place, or embed a request inline.
• Portable format: it's plain Markdown (SKILL.md), so it runs in any harness that supports skill-style instructions, no runtime dependencies beyond the agent itself.

Ideal use cases

Use Humanizer when you've drafted something with an AI agent (a blog post, docs, a launch announcement, marketing copy) and want to strip out AI-isms like significance inflation, boilerplate hedging, emoji overuse, or sycophantic chatbot phrasing before publishing. It's a good fit for teams that write a lot of AI-assisted first drafts and need a consistent cleanup pass, and for anyone who wants their agent's writing to match their own voice via the calibration feature.

It's not a fit if you need fact-checking, translation, or content generation from scratch; it only rewrites text that already exists and refuses to add new facts. It also won't help if your harness doesn't support skill-style instructions at all, since the whole thing runs as an agent skill rather than a standalone script or API you call directly. If you're looking for a hosted humanizer-as-a-service with a UI, this isn't that either. It's a prompt/instruction set that runs inside your existing AI coding agent.

Installation

The most common install path is the cross-agent skills CLI, which works globally across supported harnesses:

npx skills add blader/humanizer --global

Update it later with:

npx skills update humanizer --global

To install into every supported agent harness at once:

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

Or target a specific harness by name:

npx skills add blader/humanizer --global --agent <agent-name>

Omit --global to install it project-locally instead, which lets you commit it and share it with collaborators. Restart your agent session or reload skills after installing.

Claude Code users can install it as a plugin instead:

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

It's then invoked as /humanizer:humanizer.

For any other harness, install manually since the runtime artifact is just SKILL.md. Clone the repo into your harness's skill directory:

git clone https://github.com/blader/humanizer.git /path/to/your/skills/humanizer

Or copy the file directly if you already have the repo:

mkdir -p /path/to/your/skills/humanizer
cp SKILL.md /path/to/your/skills/humanizer/

Once installed, invoke it with a slash command or a direct request, such as /humanizer followed by pasted text, or Humanize the prose in docs/launch-post.md to rewrite a file in place.