Humanizer is a skill, a single instruction file your AI agent loads, that edits text to remove the habits that make writing read as machine-made. It is built on Wikipedia's Signs of AI writing guide and checks 26 patterns, strongest first: "not X but Y" constructions, one-line dramatic closers, sayings that sound deep, staged run-ups like "Here's the thing", forced groups of three, stacked qualifiers and the overused vocabulary everyone now recognises.
It shows its work: a first rewrite, a short critique of what still sounds artificial, then the final version. Give it two or three paragraphs of your own writing and it matches your rhythm and quirks instead of a generic style. Pointed at a file, it changes only the prose and leaves code, data and links alone, and it is told not to invent names, numbers or claims.
The project is clear that it edits for human readers and that beating AI detectors is not a goal; the README says detectors still flag most of its output. It is MIT-licensed, about 1,100 lines in total, and one of the most-installed skills on GitHub at about 60,800 stars.
- Repository: github.com/blader/humanizer
- Licence: MIT (MIT License)
- Language: Python. Stars: 60.8K. Forks: 4,831. Last push: Oct 16, 2026.
- Scan: safe, Oct 26, 2026, commit 8c307f3
Who it is for
Writers, marketers and developers who draft with Claude, ChatGPT or Codex and want the result to read like a person wrote it, especially for posts, docs and emails that go out under their name.
Getting started
1. Claude Code: add the marketplace
/plugin marketplace add blader/humanizer2. Claude Code: install the skill, then call it with /humanizer
/plugin install humanizer@humanizer3. Codex, Gemini CLI, Copilot and other agents, through the Skills CLI
npx skills add blader/humanizer --global --agent '*'The /plugin commands are typed inside Claude Code (2.1.142 or newer), not a shell. On Claude.ai or Claude Desktop, download the repository as a ZIP and upload it as a skill in Settings.
Safety scan
We cloned blader/humanizer at commit 8c307f3 on Oct 26, 2026 and ran the checks described on the GitHub Tools page: credential patterns, decode-and-execute code, install-time scripts, committed binaries, risky CI workflows, every host the code talks to, known vulnerabilities in pinned dependencies, and project hygiene. A person read every hit. This is what we found.
- Every check came back empty: no secrets, no pattern hits, no committed binaries, no install hooks and no network hosts across 14 files and about 1,100 lines, 780 of them Markdown.
- The only code is scripts/validate-package.py (147 lines), which the project's own CI runs to check the package files. Nothing in the skill executes when you use it; your agent reads SKILL.md as instructions.
- No lockfiles, so no dependency advisories. The text you humanize goes only to the model your agent already uses.
- 1 workflow, triggered by pull_request rather than pull_request_target, with read-only permissions and every action pinned to a commit. Licence present; no security policy, Dependabot, CodeQL or contributing guide.
What the scanner counted
| Check | Result |
|---|---|
| Secrets | None found. |
| Suspicious code | None found. |
| Install-time code | None: nothing runs at install beyond the package manager itself. |
| Committed binaries | None. |
| CI workflows | 1 workflow. None use pull_request_target. No third-party actions. |
| Network hosts | 0 distinct hosts referenced from source. No URLs to bare IP addresses. |
| Known vulnerabilities | No lockfile to check: dependencies are declared as ranges, so what gets installed is whatever is current on the day. |
| Project hygiene | Has licence file. Missing security policy, automated dependency updates, CodeQL, contributing guide. |
| OpenSSF Scorecard | Not scored: the project is not in Scorecard's weekly index. |
By the numbers
| Stars | 60.8K |
|---|---|
| Forks | 4,831 |
| Contributors | 23 |
| Commits | 101 |
| Open issues | 2 |
| Open pull requests | 8 |
| Releases | 6 |
| Latest release | v3.1.0 |
| Licence | MIT |
| Main language | Python |
| Project age | 9 months |
| Last push | Oct 16, 2026 |
| Tracked files | 14 |
| Lines of code | 1,089 |
| Checkout size | 66 KB |
Lines by language: Markdown 780, Python 147, YAML 114, JSON 48.
Questions
Is Humanizer free?
Yes. Humanizer is MIT-licensed and free in every supported agent. The only cost is the model usage of the rewrite, which runs on whatever model your agent is already set to.
Will Humanizer get text past AI detectors?
No, and it does not try to. The project says it edits for human readers and that detectors still flag most of its output. Use it to make writing clearer and more natural, not to disguise where it came from.
Can Humanizer match my own writing style?
Yes. Paste two or three paragraphs of your own writing before the text to rewrite, and it follows the sample's rhythm, word choice and punctuation, including dashes if you use them.
This post is part of GitHub Tools, where every repository is cloned and scanned before it is written up. The scan is a snapshot of one commit on one day; the repository has moved on since, so check it before you install.