Share

Install It Now

Run this to install it globally, so it is available across every project:

npx skills add johnakande/ai-humanizer-skill --global

Leave off --global to install it in just the current project, or add --agent <name> to choose which agents receive it. Manual install and Claude-plugin instructions are both in the repo README.

What It Does

AI Humanizer Skill rewrites text to a specific house style: plain language, active voice, no AI-tell vocabulary, no em dashes, no generic headings, no visible drafting process shown by default. It is plain Markdown, so it works with any agent that supports skills, not just Claude. I use it as part of the same Claude-based writing workflow I've written about in a separate walkthrough.

The rule set is original house style built around 13 named patterns, adapted in part from Wikipedia's documented list of AI writing signs and reframed for marketing and GTM copy instead of encyclopedia prose. It never invents facts: a name, number, date, or quote has to come from your source text or from you. If a sentence is missing a detail it needs, the skill asks for it or rewrites around it instead of guessing.

Work With John

Your site should be your best salesperson. If it is not, that is a fixable problem.

I work with US service businesses and B2B brands to build SEO systems that produce consistent, compounding leads. I will tell you exactly what is broken. No pitch.

Book a free strategy call

How to Use It

Step 1: Install. Run the command above, or clone the repo manually into your agent's skills directory for project-level control.

Step 2: Call it. Type /ai-humanizer-skill followed by your text, or just ask in plain language: "Rewrite this in my house style: [your text]." Give it a file path instead and it rewrites the prose in place, leaving code, data, frontmatter, and link targets untouched.

Step 3 (optional): Match your voice. Paste 2-3 paragraphs of your own past writing before the text you want rewritten. The skill reads the sample for sentence length, word choice, and paragraph rhythm, then applies that voice within the house style instead of defaulting to a generic tone.

Step 4: Review. By default it hands back only the finished text, no visible draft or list of what changed. Ask explicitly if you want to see its reasoning or the intermediate draft.

Key Features

13 named failure patterns, not a vague "sound more human" prompt. Every rewrite is checked against a documented rule set covering sentence rules, word rules, structure rules, and tone rules, so the same standard applies every time instead of varying by mood or model.

A silent draft-and-check process. It writes, checks the draft against every rule, and hands back only the finished version, with no meta-commentary about what it fixed unless you ask.

A fact-preservation guardrail. It never invents a name, number, date, or quote to make a sentence read more smoothly. Missing a detail means it asks or rewrites around the gap.

File-mode rewriting. Point it at a real file and it edits only the prose, leaving code blocks, YAML frontmatter, and links exactly as they were.

Portable across agents. It is Markdown with no host-specific installer, so it runs the same in Claude Code, Codex, or any other agent that reads Markdown skills.

Use Cases

Cleaning up a Claude or ChatGPT draft before it goes out as an email, landing page, or blog post. Passing generated documentation, changelog entries, or release notes through a final editorial check before a repo goes public. Running as a QA step at the end of an agent workflow, so nothing with an obvious "written by AI" tell ships without a human noticing. Matching a specific brand voice across multiple writers or multiple AI-assisted drafts, using the voice-sample step.

Best Fit

  • You want AI-drafted copy that does not read like it was AI-drafted, without turning into a generic paraphraser.
  • You use Claude, Codex, or another agent-based workflow and want a repeatable editing pass instead of re-prompting "make this sound human" every time.
  • You need every fact, name, and number preserved exactly as given, not smoothed over with invented specifics.

Not a Fit

  • You are trying to get around a school, workplace, or platform's AI-disclosure policy. This is an editing tool for responsible use, not a way to evade a rule.
  • You want a guarantee that an AI-detection tool will call the output human. Nobody can promise that, and I would not trust any tool that does.
  • You need legal, medical, financial, or compliance language preserved in an exact, pre-approved template. Use your compliance process for that text, not this skill.

Requirements

Free and open source under the MIT license. No API key of its own, since it runs inside whatever agent calls it. Works with any tool that reads Markdown-based skills.

Building small tools like this instead of just using someone else's is the same instinct behind our AI automation work, operator-built infrastructure rather than template-level configuration.