Humanizer: The Agent Skill That Makes AI Writing Sound Like a Person Wrote It

Humanizer makes AI-written text sound like a person wrote it, without changing what it says. It is an agent skill, built on Wikipedia's Signs of AI writing, the guide Wikipedia editors use to catch AI-generated text, and it works in Claude Code, Codex and any other agent that supports skills. The README opens on a real before-and-after example, a product announcement about shared drafts. The before version is thrilled, drowning in files named final_v7.docx, and ends with "Let that sink in." The after version says the same thing in plain sentences: "Shared drafts are out today. For months our own team passed around files named final_v7.docx, so we built a way for two people to edit the same doc at once, with each person's changes showing up live for the other." Nothing in the meaning changed. Everything that marked it as AI writing is gone. In a blind test, judges preferred Humanizer's rewrite over the original AI text 16 times out of 16.

Humanizer checks for 26 patterns in all, numbered by strength and frequency, and the first five justify an edit on a single sighting. First is "Not X but Y": "It's not just a feature, it's a shift." Second is the one-line closer: "Let that sink in." Third is the saying that sounds deep: "At its core, what really matters is..." Fourth is the staged run-up: "Here's the thing." "Honestly?" Fifth is arguing with no one: "I'm not saying X, but..." The other 21 cover rhythm by rule, like forced triads and repeated sentence openings, inflation and borrowed authority, like overused AI words and inflated significance, formatting by rule, like bold as decoration and decorative headings, and leftovers from the chat, like "Great question!" wrappers and knowledge-limit disclaimers. The idea behind all of them is one observation from the README: a language model writes whatever is most likely to come next, so by default it makes the choice that fits the widest range of readers, while a person chooses for one reader and one subject. Before you paste a draft into Humanizer, an AI Text Cleaner strips the em dashes, curly quotes and hidden characters that AI output carries, so the skill starts from clean text.

Humanizer does not rewrite everything into the same voice. If you want the rewrite to sound more like you, include a sample: two or three paragraphs of your own writing, then the AI text to humanize. It follows the sample's rhythm, word choice, punctuation and deliberate quirks, including dashes if you use them. That is the difference between a rewrite that sounds human and one that sounds like a different person. It shows its work too. When you paste text, you see the first rewrite, a short critique of anything that still sounds artificial, and the final version. Point it at a file and it changes only the prose, leaving code, data, frontmatter and link targets alone. Personal writing keeps the writer's opinions and quirks, and technical and reference prose stays neutral and plain. One rule runs through all of it: it does not make things up. A name, number, date, quote, citation or other factual detail must come from the source or the writer, and if a sentence needs a detail that is missing, Humanizer asks instead of inventing one.

This is the part the README states plainly, and it matters. Humanizer edits for human readers. Getting past AI detectors is not a goal, and detectors still flag most of its output. That is not a failure of the tool, it is the point of it: the goal is prose a person enjoys reading, not prose that fools a classifier. The skill is honest about what it does and what it does not do, and the article should be too. If you need a detector to pass, this is not the tool for that job. If you need a launch post, a doc or an email that reads like you wrote it, that is exactly what it is for. After the rewrite, an AI Proofreader catches any grammar, spelling or punctuation slips that survived, so the final text is clean as well as human sounding.

Once installed, the skill answers to /humanizer. In Claude Code, add the marketplace and install the plugin, which needs Claude Code 2.1.142 or newer. In Codex, run npx skills add blader/humanizer --global --agent codex. For Claude.ai and Claude Desktop, download the repository as a ZIP and upload it as a skill in Settings. For any other agent, npx skills add blader/humanizer --global --agent '*' installs it everywhere the Skills CLI supports, including Gemini CLI, GitHub Copilot and Windsurf, and leaving off --global installs it only in the current project. Then call it directly with /humanizer and paste your text, ask in plain language, or give it a file path and let it rewrite the prose inside. It has 66,259 stars, up about 3,000 in the past week, which says something about how many people are tired of reading text that sounds like it was written by a committee of one.