Skill：untell

untell — free, open-source AI humanizer that closes the loop: iteratively rewrites AI text against live detector scores (GPTZero, ZeroGPT, Originality.ai, Turnitin) while preserving meaning, citations & facts. Claude Code skill + CLI. Live-proven 100%→0% on ZeroGPT. MIT.

适用场景

招聘人事

适合角色

介绍

以下内容整理自开源仓库 `ssamba1/untell` 的 README / SKILL.md 摘要：

<div align="center">

<a href="https://ssamba1.github.io/untell/"><img src="docs/og.png" alt="untell — the open-source AI humanizer that closes the loop: rewrites AI text against live detector scores while keeping meaning, citations and facts intact" width="820"></a>

# untell — the open-source AI humanizer that *closes the loop*

### Iteratively rewrite AI-generated text against live AI-detector scores until it reads human — while keeping your meaning, citations, and facts intact.

A **closed-loop, detector-feedback** AI humanizer, shipped as a **Claude Code skill** *and* a Python CLI.
Free. Open source. Honest about what it can and can't do.

[![CI](https://github.com/ssamba1/untell/actions/workflows/ci.yml/badge.svg)](https://github.com/ssamba1/untell/actions/workflows/ci.yml)
[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](LICENSE)
[![Python 3.9+](https://img.shields.io/badge/python-3.9%2B-blue.svg)](https://www.python.org/)
[![Claude Code skill](https://img.shields.io/badge/Claude%20Code-skill-8A2BE2.svg)](#-quick-start)
[![Zero-dependency lite tier](https://img.shields.io/badge/install-zero--dependency-brightgreen.svg)](#tiers)
[![PRs welcome](https://img.shields.io/badge/PRs-welcome-orange.svg)](CONTRIBUTING.md)
[![Live site](https://img.shields.io/badge/site-ssamba1.github.io%2Funtell-2ea44f.svg)](https://ssamba1.github.io/untell/)
[![good first issues](https://img.shields.io/github/issues/ssamba1/untell/good%20first%20issue.svg?label=good%20first%20issues&color=7057ff)](https://github.com/ssamba1/untell/labels/good%20first%20issue)

**Optimize against real detectors — with the detector *in the loop*, not blind guessing.** Out of the box it
beats the **free web checkers** (ZeroGPT 100%→0%, measured live on 2026-06-25 against the formulaic
demo paragraphs — a third-party site that can change without notice, so re-verify with
`--browser zerogpt` rather than trusting the figure). To actually beat **GPTZero · Originality.ai ·
Turnitin-class · Copyleaks**, you wire *their* API into the loop (key-gated, paid) — the bundled **local
proxies alone do *not* predict those, and we [say so plainly](#honest-caveats)** rather than fake a "99% human."
[Why this is the most complete open humanizer →](#-why-this-is-the-best-open-source-ai-humanizer)

</div>

---

## TL;DR

Most "AI humanizers" do **one blind paraphrase pass** and plateau at 60–80% detector bypass. This one runs a
**loop**: it *scores* your text against an ensemble of real AI detectors, *rewrites* using each detector's
score as feedback (targeting the exact sentences that read as AI), and *re-scores* — repeating until the
hardest detector stops flagging it **and** a semantic-similarity gate confirms the meaning is unchanged.

That iterative, detector-feedback approach is 

---

安装方式

1. 打开仓库：https://github.com/ssamba1/untell
2. 查找 SKILL.md 或 README 中的安装章节
3. 克隆或下载所需 Skill 子目录到本地 skills 路径（如 `.cursor/skills/` 或 `.claude/skills/`）
4. 重启 / 重载 Agent 客户端
5. 用一条真实任务验证 Skill 是否被正确加载

适用平台

Claude

打开详情 · 返回技巧库