phuryn/work-humanizer

选用指南

AI 根据 README 整理 · 2026/10/17 09:31:13

使用方式

作为Agent Skill使用，需复制到代理的skills文件夹，或通过Skills CLI安装。

查看 README 依据

One brief, checked by hand against it, so read it as an illustration rather than a measurement. Data: [`evidence/brief-to-email/`](evidence/brief-to-email/). ## Install Copy the `skills/work-humanizer` folder into your agent's skills folder: | Agent | Folder | |---|---| | Claude Code | `~/.claude/skills/work-humanizer` (all projects) or `.claude/skills/work-humanizer` (one project) | | Codex | `~/.codex/skills/work-humanizer` | | Other agents | the agent's skills folder; work-humanizer is a standard `SKILL.md` plus two supporting files |

Or with the [Skills CLI](https://skills.sh): `npx skills add phuryn/work-humanizer`. ## Use ```text Use work-humanizer on this: <paste text> Use work-humanizer on this and interview me first: <paste draft> Use work-humanizer on docs/launch-post.md ``` **For the most human result, give it your own material.** Paste your notes, a voice-memo transcript or a rough draft you wrote yourself. work-humanizer builds the piece on your argument, in your order, and keeps your sentences where it can. If you also have an AI draft, it only uses it as a checklist of points.

部署要求

需要支持SKILL.md的代理环境，如Claude Code、Codex或Cursor。

查看 README 依据

questions first. An Agent Skill: works in any agent that loads `SKILL.md`, including Claude Code, Codex and Cursor.  ## What the test found 12 new AI-written texts, rewritten by each tool on the same model, judged blind by four AI models from different companies, with every meaning change checked against the original.

已知限制

测试样本较小，且未包含合同等法律文件；AI检测器无法完全规避。

查看 README 依据

Caveats: small samples (12 texts on the scoreboard, 56 across all sets, 4 for the notes result); the texts were sales, marketing, support, product and internal writing, with no contracts or other legal documents; AI judges rather than people, and the judges disagree on what "human" means; one sample per text per tool; one Pangram version on one day; a simulated writer in interview mode; and the people who built the skill designed the test. Full method, data, judge verdicts, fact checks and scripts: [`evidence/`](evidence/README.md). Read it before quoting any number. ## Repository ```

- **The one-line prompt reads more human, and that's the catch.** It beat the AI draft on 90% of judgments to work-humanizer's 75% because it rewrites freely: it changes more and adds things the writer never said. Judges who could see the original sent it once in 96 picks. - **No rewrite gets past an AI detector.** On Pangram, work-humanizer, blader/humanizer and the one-line prompt all leave the text reading as AI (2 to 4% human). Emulate-1, sold as undetectable, reaches 87% by changing the most.