Humanizer

// Overview

Humanizer is a Markdown skill you install into a coding agent and call with `/humanizer`. It takes a draft that reads as machine-written and rewrites it so it reads as though a person wrote it, while keeping every claim in the original. The project's stated premise is that a language model writes whatever fits the widest range of readers, and that this averaging is what produces the recognisable register — so the fix is a named checklist applied after the draft, not another instruction to the model that produced it.

The checklist is the substance of the tool. Twenty-five patterns are grouped into five sections, ordered by how strongly each one signals machine authorship: staging instead of stating (contrast constructions, one-line closers, artificial run-ups, strawmen), rhythm by rule (forced triads, repeated openings, dash overuse, stacked qualifiers, passive voice), inflation and borrowed authority (overused vocabulary, exaggerated importance, vague associations, sales language), formatting by rule (decorative bold, fancy headings, curly quotes) and leftovers (chatbot wrappers, knowledge disclaimers, repeated headings). The first five patterns justify an edit on their own; the rest have to appear more than once before the skill acts.

Running the skill is a loop rather than a single pass: it marks what it finds, drafts a rewrite without treating the original structure as fixed, validates the result against the same list, then produces the final text. Supplying two or three paragraphs of your own writing first makes it match your voice instead of a generic plain register. Humanizer is MIT-licensed, installs through the Skills CLI, the Claude Code plugin marketplace or a ZIP upload in Claude Desktop, and its README is explicit that the skill does not invent content while rewriting.

// What it does

• `/humanizer` slash command that rewrites pasted text, a whole file, or the draft you name in plain language
• A fixed list of 25 AI-writing patterns in five sections, ordered by how strongly each signals machine authorship
• Tiered severity: the top five patterns justify an edit alone; the remaining twenty need repeat occurrences
• Voice matching — give it two or three paragraphs of your own writing and the rewrite follows your style
• Mark, draft, validate, finalise loop, so the rewrite is checked against the same list that produced it
• Installs through the Skills CLI, the Claude Code plugin marketplace, or as a ZIP upload in Claude Desktop
• A repository validator derives the pattern count from the headings and caps the skill file at 400 lines

// Getting started

Humanizer is a skill, not a service — there is nothing to run or host. Install it into an agent that supports skills, then call it on a draft.

npx skills add blader/humanizer --global

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

/humanizer
Please humanize this text: <your draft>
Humanize the prose in docs/launch-post.md

Here are 3 paragraphs I wrote: <samples>
Now humanize this draft in that voice: <your draft>

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Clean up a blog post, changelog or README that an agent drafted before it is published
• Give a team one shared, versioned definition of what "sounds like AI wrote it" instead of per-reviewer taste
• Rewrite agent output in your own voice by seeding the skill with a few samples of your writing
• Read the pattern list as a style reference — it names the habits rather than only removing them

// How Humanizer compares

Humanizer alongside other open-source agent skills & plugins tools AI/TLDR tracks, ranked by GitHub stars.