Strip the AI Tells from Your Prose with the Humanizer Skill

Your agent leans on "It's not just X, it's Y" in every other paragraph and ends each section with a one-line closer. Humanizer, a 49k-star skill, finds those tells and rewrites them out — without changing what the text says.

Why This Skill Matters

The README grounds every AI tell in one explanation: a model defaults to whatever fits the widest range of readers, while a person writes for one reader on one subject. The 25 patterns are sourced from Wikipedia's "Signs of AI writing" page, which WikiProject AI Cleanup maintains, and version 3.0.0 consolidated an earlier list of 35 into these 25, grouped into five sections and numbered by strength and frequency.

The numbering carries meaning. The first five patterns justify an edit on a single sighting — "not X but Y" framing, one-line closers, sayings that sound deep, staged run-ups, and arguments with no one. Patterns marked weak-alone count only when several tells share a passage, because a careful writer may use any one of them on purpose. And the skill does not invent: a name, number, date, quote, or citation must come from your source, and a rewrite that needs a missing detail asks instead of making one up.

Installation

Install with the Skills CLI:

npx skills add blader/humanizer --global

Leave off --global to install it in the current project only, and pass --agent <name> or --agent '*' to choose which agents receive it. The skill answers to /humanizer. On Claude Code 2.1.142 or newer you can install the plugin instead:

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

The plugin answers to /humanizer:humanizer. In Claude Desktop, download the repository as a ZIP and upload it as a skill; for a manual install anywhere else, copy SKILL.md into the agent's skill folder.

Real Workflow: De-AI a Launch Announcement

Paste the draft straight into the skill:

/humanizer

[paste your draft here]

Humanizer marks every tell it finds, strongest first, drafts a rewrite without treating your original structure as fixed, checks the draft against the patterns and your original claims, and only then writes the final version. On pasted text you see all three steps: the first rewrite, a short critique of anything that still sounds artificial, and the final version.

Real Workflow: Match Your Own Voice

A generic de-AI pass can flatten your style, so include a sample when it matters:

/humanizer

Here's a sample of my writing for voice matching:
[2-3 paragraphs of your own writing]

Now humanize this text:
[the AI draft]

The skill follows the sample's rhythm, word choice, punctuation, and deliberate quirks — dashes included, if you use them. For files, skip the paste entirely: say Humanize the prose in docs/launch-post.md, and it changes only the prose, leaving code, data, frontmatter, and link targets alone.

Tips

• Feed it source notes. The README's worked example stays honest because the writer's notes carry the facts — the month, the neighborhood, how long the tram took. Without notes, Humanizer asks instead of inventing.
• Learn the top five patterns first; they justify an edit on a single sighting, so you will start spotting them in your own drafts.
• Treat weak-alone flags as hints, not verdicts — a single dash or passive is a style choice until several tells pile up in one passage.
• Read the critique between drafts. It names what still sounds artificial, which is the fastest way to learn the patterns.

When Not to Use This

Humanizer edits prose. Text that must stay verbatim — quotations foremost — is out of scope, because a rewrite is a rewrite no matter how careful. Its job is rewriting: polishing phrasing and checking the rewrite against your claims; research and fact-finding stay with you. And it is the wrong tool for code or data files beyond the prose around them.

See the leaderboard for more skills.