Humanizer-zh

Humanizer-zh is a Chinese-language adaptation of a Claude Code skill designed to help editors identify and remove AI-generated writing patterns from text. The goal is not to fool automated detectors but to genuinely improve writing quality, making AI-assisted drafts read more like natural human writing.

The tool catalogs 24 specific patterns commonly found in AI-generated text, grouped into four categories. Content patterns include things like vague attributions, excessive emphasis on significance or legacy, and generic challenge-and-outlook sections. Language patterns cover overused AI vocabulary, avoiding the verb "to be," and the habit of grouping things in threes. Style patterns flag excessive use of bold text, over-reliance on em-dashes, and certain formatting habits like heavy use of inline headers and title-case headings. Communication patterns identify filler phrases, flattering language, unnecessary knowledge-cutoff disclaimers, and overly qualified statements.

To use it, a writer pastes their AI-generated text and invokes the skill with the /humanizer-zh command in Claude Code. The tool scans the text against the 24 patterns and rewrites flagged sections with more concrete, specific, and direct language. The README includes before-and-after examples showing how abstract or inflated phrasing gets replaced with factual, human-sounding alternatives.

Installation can be done with a single npx command or by cloning the repository into Claude Code's skills directory. The project translates and adapts the English-language humanizer skill specifically for Chinese writing contexts, with examples and vocabulary lists calibrated for Chinese-language AI tells rather than English ones. Source material draws from a Wikipedia guide on signs of AI writing and a related community stop-slop project.