Last updated: 11/18/2026Last verified: 2026-11-14
Humanizer is an MIT-licensed agent skill that rewrites text so it reads like a person wrote it, built on Wikipedia’s “Signs of AI writing” guide, and it runs in Claude Code, Codex, pi and any other agent that supports skills. The maintainer’s skills.sh listing recorded 13.1K installs on 2026-11-14.
What is Humanizer?
Humanizer is an MIT-licensed agent skill that rewrites AI-sounding text using 26 numbered patterns drawn from Wikipedia’s “Signs of AI writing” guide. It runs inside Claude Code, Codex, pi and other agents that support skills rather than as a standalone writing app: paste text, or point it at a file, and it marks the tells it finds before rewriting. The current release is v4.0.0, published 2026-11-10; the GitHub API recorded 69,016 stars and 5,417 forks on 2026-11-14, and the project’s skills.sh page listed 13.1K installs on the same date. Those are GitHub’s and the project’s own figures, not an independent evaluation, and this listing did not install the skill or run a rewrite.
Key Features
- Pattern-based editing: Checks prose against 26 numbered AI-writing patterns, prioritizes the strongest signals, and marks the tells it identifies.
- Agent compatibility: Supports Claude Code, Codex, pi, and other agents that can run skills.
- Flexible installation: Offers one-command installation for supported agent environments or ZIP upload through Claude.ai and Claude Desktop settings.
- Multiple input options: Invoke /humanizer with pasted text, a plain request, or a file path.
- Visible revision process: Text mode presents an initial rewrite, a short critique of remaining artificial-sounding language, and a final version.
- Prose-focused file editing: File mode is designed to change prose while leaving code, data, frontmatter, and link targets unchanged.
- Optional voice matching: Supply two or three paragraphs of your own writing to guide rhythm, word choice, punctuation, and stylistic quirks.
- Open-source licensing: The project is distributed under the MIT license.
- Release and licence: v4.0.0 (2026-11-10) under the MIT licence; the Claude Code plugin requires Claude Code 2.1.142 or newer, with npx skills add or pi install as alternatives.
- File scope: the README says file mode reads the whole current file before editing and changes only prose, leaving code, data, frontmatter and link targets in place.
Best For
Pricing
Humanizer is free and MIT-licensed; the repository README is the source for both. There is no paid tier and no hosted editor. Running it still requires an agent and model you already have, and those can carry their own costs.
Pros & Cons
Pros
- Uses an explicit set of 26 editing patterns rather than an unspecified instruction to make text sound more human.
- Fits into several agent environments instead of requiring a standalone editor.
- Text mode makes the revision process visible through a draft, critique, and final version.
- Supports voice matching from a short sample of the user’s writing.
- Offers a file-editing mode intended to preserve non-prose elements.
- Free availability and MIT licensing make it accessible for users who already have a compatible setup.
Cons
- Requires a compatible agent environment or a supported Claude interface rather than functioning as an independent writing app.
- v4.0.0 reversed the earlier no-invention rule: a rewrite may now add a specific detail that was not in your text, and the maintainer says to stay on v3.1.0 if you rely on the old behaviour.
- The README states it does not try to get past AI detectors and that detectors still flag most of its output; the 16-of-16 blind test it cites is the maintainer’s own issue report (#229), not an independent test.
- File mode’s preservation of code, data, frontmatter and link targets is documented behaviour, not something this listing tested.
- Total cost depends on the host agent and model subscription; only the skill itself is free.
Alternatives
Grammar, clarity and tone suggestions while you type in an editor or browser extension, where Humanizer rewrites whole passages inside an agent.
Academic-focused proofreading for manuscripts and essays, instead of pattern-based style rewriting.
Generates marketing copy from prompts and a brand voice, where Humanizer edits text you already wrote.
FAQ
Details
Platform
Features
- Rewrites AI-sounding prose against 26 numbered patterns taken from Wikipedia’s Signs of AI writing, strongest first, marking every tell it finds
- Installs with one command for Claude Code, Codex and pi, or as a ZIP uploaded in Claude.ai and Claude Desktop settings; then run /humanizer on pasted text, a plain request, or a file path
- Text mode shows the first rewrite, a short critique of what still sounds artificial and the final version; file mode changes only prose and leaves code, data, frontmatter and link targets alone
- Optional voice matching: paste two or three paragraphs of your own writing and the rewrite follows its rhythm, word choice, punctuation and quirks
Languages
Known limitations
- The v4.0.0 release notes (2026-11-10) reverse the earlier no-invention rule: a rewrite may now add a specific detail that was not in your text, and the maintainer says to stay on v3.1.0 if you rely on the old rule
- It edits for human readers rather than to evade AI detectors, and the README says detectors still flag most of its output
- It is a skill rather than a standalone app: it runs inside an agent you already use, the Claude Code plugin needs version 2.1.142 or newer, and there is no hosted editor of its own
- The 69,016 stars and 5,417 forks are GitHub API figures from 2026-11-14; the 13.1K installs come from the project’s skills.sh listing and the 16-of-16 blind-test result is the maintainer’s own issue report, not an independent test by this listing






