# Humanizer-zh Skill > Humanizer-zh is an AI humanizer that rewrites Chinese text to eliminate AI patterns, delivering natural and human-like writing. Start free in seconds. - Canonical: https://nanoskill.ai/skills/chinese-ai-humanizer-skill - Markdown: https://nanoskill.ai/skills/chinese-ai-humanizer-skill.md - Author: op7418 - Published: 2026-07-01T03:00:00.000Z - Updated: 2026-07-25T04:33:14.644Z - Language: en - Source type: github - Popularity signal: 11332 ## Sources - https://github.com/op7418/Humanizer-zh ## Install ```shell npx skills add https://github.com/op7418/Humanizer-zh ``` ## About Humanizer-zh is an AI humanizer that helps editors, marketers, and researchers transform AI-generated Chinese text into authentic, human-quality writing. It removes robotic patterns like overused buzzwords and awkward structures, turning stiff drafts into natural content that connects with real readers. Unlike basic rewriters, it identifies 24 specific AI writing traces grouped into content, language, style, and communication categories. Adapted for Chinese conventions, it corrects issues from tricolon abuse to title-case titles, ensuring the output feels genuinely handcrafted. Whether you're refining marketing copy, polishing academic abstracts, or blogging, Humanizer-zh works directly inside Claude Code with a simple slash command. Install in one line, then tell it which text to humanize — the tool rewrites while preserving core meaning and adding human warmth. ## Key features - **24 AI Pattern Detection**: The AI humanizer identifies and fixes 24 specific AI writing traces across content, language, style, and communication. - **Chinese Language Optimization**: Tailored for Chinese text, adapting patterns like title case, tricolon, and terminology to local writing conventions. - **One-Command Installation**: Install instantly with npx skills add and start humanizing text without complex setup or API keys. - **Natural Rewriting Guidelines**: Combines pattern detection with human-centric rewriting principles to ensure output is vivid, opinionated, and not just “clean”. - **AI Vocabulary Warnings**: Flags overused AI buzzwords like “additionally”, “showcases”, and “tapestry”, helping you avoid robotic phrasing. ## Use cases - **HUMANIZE MARKETING COPY**: Transform AI-generated promotional text into genuine, engaging copy that resonates with real readers. - **REFINE ACADEMIC WRITING**: Remove AI traces from research summaries and abstracts while preserving scientific precision. - **POLISH BLOG POSTS**: Convert stiff AI drafts into conversational, opinionated articles that readers enjoy. ## Result preview Explore a production-ready Chinese AI humanization report powered by this Skill. ![Editorial case study page titled '01 Source Copy' analyzing a deliberately typical piece of AI-generated product writing for FlowMind. The page explains that the draft is grammatical and complete but contains generated-writing signals, including inflated product significance and limited connection to recognizable workday scenarios. Three numbered source paragraphs describe FlowMind using broad AI marketing language about information overload, intelligent transformation, an intelligent second brain, automated workflows, AI agents, efficiency, creativity, and business growth. The layout uses a clean white report style with large bold headings, blue section labels, and spacious paragraph formatting to present the original copy before critique and rewriting.](https://file.nanoskill.ai/humanizer-outcome1.png) ![Editorial case study page titled '02 Source Critique' explaining why the original FlowMind product copy reads like AI-generated writing. The page uses a clean report layout with a section label 'Why it reads like AI copy' and six pink critique blocks. Each block names a common AI-writing pattern and explains its weakness: 'Grand opening' critiques vague dramatic framing such as rapidly evolving era and unprecedented challenges; 'Template triplets' points out repeated three-part phrasing that sounds complete without adding meaning; 'Abstract value claims' identifies broad promises like redefine, unleash creativity, and accelerate growth without observable action; 'Manufactured elevation' critiques overuse of not-just-a-tool framing; 'Missing boundaries' notes that the draft claims agents complete work automatically without explaining human review or approval; and 'Generic ending' warns that a possibility-filled future sounds interchangeable with almost any AI product. The slide functions as a structured diagnostic analysis of machine-written marketing copy before rewriting it into more credible human-centered product language.](https://file.nanoskill.ai/humanizer-outcome2.png) ![Editorial case study page titled '03 Editorial Strategy' presenting a framework for rewriting AI-generated product copy into more natural and recognizable human language. The section subtitle says to move from feature proclamation to a recognizable day at work. Four horizontal strategy blocks explain how to improve the FlowMind copy: 'Start with recognition' recommends using familiar workday scenes such as tabs, meetings, chat, and loose ends instead of abstract lectures about a changing world; 'Turn capabilities into actions' suggests making features tangible through specific actions like extracting key details, linking related material, and finding project decisions; 'State the boundary' explains that AI agents can run workflows but should pause when human judgment is needed, building trust through clear limits; and 'Leave a point of view' frames FlowMind as a tool for reducing context hunting and repetitive handoffs so users can protect their attention. A final tone note defines the desired voice as professional without stiffness, intelligent without showing off, ambitious about efficiency, and respectful of human judgment.](https://file.nanoskill.ai/humanizer-outcome3.png) ![Editorial case study page titled '04 Before and After' comparing AI-generated FlowMind product copy with a more humanized rewrite. The page focuses on paragraphs 1–2 with the note 'Ground the problem in real work.' Two stacked comparison tables use pink 'Before' rows and green 'After' rows. The first before example uses broad AI marketing language about a rapidly evolving era, information overload, intelligent transformation, and a seamless productivity platform. The after version rewrites it into a recognizable workday scene involving open tabs, meeting notes, loose ends in chat, scattered information, and FlowMind bringing relevant pieces into one workspace. The second before example describes FlowMind as an intelligent second brain that gathers information from documents, meetings, emails, and conversations. The after version makes the value more concrete by explaining how documents, meeting notes, email, and chat feed into FlowMind, helping users find decisions, customer comments, project context, and latest versions more easily.](https://file.nanoskill.ai/humanzier-outcome4.png) ## Result walkthrough ### Install Add the Chinese AI Humanizer Skill to your AI agent. ![AI agent installation log showing setup of the Humanizer-zh skill from the op7418/Humanizer-zh GitHub repository using an npx skills add command. The interface explains that an existing English humanizer skill is already installed, then identifies Humanizer-zh as a Chinese-language humanizer designed to detect and remove AI writing patterns in Chinese text. The workflow handles interactive agent selection with a -y flag, verifies accessibility, and confirms installation into the Hermes Agent skills directory. The summary notes that the skill covers 24 AI-writing patterns across content structure, grammar, style, and communication, including inflated significance, promotional language, vague attribution, AI vocabulary, em dash overuse, rule-of-three phrasing, and filler expressions, with a five-dimension quality scoring system totaling 50 points.](https://file.nanoskill.ai/humanizer-install.png) ### Humanizer Content Provide AI-generated Chinese text to rewrite with a more natural, native writing style. ![AI agent workflow screen showing a detailed prompt for using a Chinese humanizer skill to transform AI-generated Chinese text into natural, human-like writing for FlowMind, an AI productivity startup. The prompt defines the role as an editorial language director and Chinese content optimization specialist, with requirements to remove mechanical phrasing, translation artifacts, repetitive structures, overly formal tone, and template-like expressions while preserving meaning and improving readability, rhythm, emotional resonance, and cultural naturalness. The workflow log below shows the agent loading the humanizer-zh skill, creating a PDF with before-and-after comparisons, resolving quote escaping issues, switching to a Unicode Chinese font, and verifying the generated PDF output.](https://file.nanoskill.ai/humanizer-task.png) ### Refine Expressions Receive polished content with improved fluency, tone, and readability while preserving the original intent. ![A clean editorial case study cover page for FlowMind focused on copy humanization. The page headline reads 'Make the copy sound human. Make the product easier to understand.' Supporting text describes it as an English content humanization case study covering brand narrative, source critique, rewrite, and editorial notes. A centered label reads 'EDITORIAL CASE STUDY'. A highlighted project brief box explains the goal of refining FlowMind's core website copy while preserving the product promise of organizing information, automating workflows, and collaborating with AI agents, while removing machine-written tech marketing language. The footer includes an editorial principle emphasizing that specificity, action, and credibility matter more than sweeping claims or noise.](https://file.nanoskill.ai/humanizer-outcome.png) ## Skill definition # Humanizer-zh: AI Writing Trace Removal Tool (Chinese Version) > **Disclaimer:** > - The core files of this project are translated from [blader/humanizer](https://github.com/blader/humanizer/tree/main) > - The utility parts (core rules, quick checklist, quality scoring) reference [hardikpandya/stop-slop](https://github.com/hardikpandya/stop-slop) > - The original project is based on Wikipedia's [Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) guide --- ## Project Introduction Humanizer-zh is a tool for removing AI-generated traces from text, helping you rewrite AI-generated content to sound more natural and human-written. This project is suitable for: - Editing and reviewing AI-generated content - Improving the human-like quality of articles - Learning to identify common AI writing patterns ## Installation ### Method 1: One-click install via npx (Recommended) ```bash npx skills add https://github.com/op7418/Humanizer-zh.git ``` This is the simplest installation method, automatically installing the skill to the correct directory. ### Method 2: Clone via Git ```bash # Clone to Claude Code's skills directory git clone https://github.com/op7418/Humanizer-zh.git ~/.claude/skills/humanizer-zh ``` ### Method 3: Manual Installation 1. Download this project's ZIP file or clone to local 2. Copy the `Humanizer-zh` folder to Claude Code's skills directory: - **macOS/Linux**: `~/.claude/skills/` - **Windows**: `%USERPROFILE%\.claude\skills\` 3. Ensure the folder structure is as follows: ``` ~/.claude/skills/humanizer-zh/ ├── SKILL.md # Skill definition file (Chinese version) └── README.md # Documentation ``` ### Verify Installation After restarting Claude Code or reloading skills, enter in the conversation: ``` /humanizer-zh ``` If installed successfully, the skill will be activated. ## Usage ### Basic Usage In Claude Code, you can use Humanizer in the following ways: #### 1. Invoke the skill directly ``` /humanizer-zh Please humanize the following text: [Paste your AI-generated text] ``` #### 2. Use in conversation ``` Please use humanizer to rewrite this passage to make it more natural: This project serves as proof of our team's commitment to innovation. Additionally, it showcases our key role in the ever-evolving technological landscape. ``` #### 3. Process file content ``` /humanizer-zh Please humanize the content in article.md ``` ### Usage Scenario Examples #### Scenario 1: Rewrite marketing copy **Input:** ``` /humanizer-zh Located in the picturesque center of Hangzhou, this café boasts rich cultural heritage and breathtaking decor. It serves as the focal point of the city's coffee culture, offering customers a seamless, intuitive, and vibrant experience. ``` **Output Example:** > This café has been in downtown Hangzhou for three years, known for its pour-over coffee and renovated old building space. #### Scenario 2: Rewrite academic abstract **Input:** ``` /humanizer-zh This study delves into the critical role of machine learning in medical diagnosis, highlighting its importance in the ever-evolving healthcare landscape. Additionally, it lays a solid foundation for the future development of this field. ``` **Output Example:** > This study analyzed the application of machine learning in medical diagnosis, focusing on early lung cancer screening. It used 5,000 medical records from 2019–2023. #### Scenario 3: Rewrite blog posts **Input:** ``` /humanizer-zh Artificial intelligence is not just a technology; it is a revolution in the way we think about the future. Industry experts believe it will have a lasting impact on society as a whole. ``` **Output Example:** > I keep thinking about how AI will change the way we work. Last week, chatting with a few friends in product development, some were excited, some worried about job loss—the truth probably lies somewhere boring in between. ## Detectable AI Writing Patterns This tool can identify and fix **24 types** of AI writing traces, divided into four categories: ### 📝 Content Patterns (6 types) 1. Overemphasis on significance, legacy, and broader trends 2. Overemphasis on popularity and media coverage 3. Superficial analysis ending with -ing 4. Promotional and advertising language 5. Vague attribution and ambiguous wording 6. Token "challenges and future outlook" sections ### 🔤 Language and Grammar Patterns (6 types) 7. Overused "AI vocabulary" 8. Avoidance of "to be" (copula avoidance) 9. Negative parallelism 10. Overuse of the rule of three 11. Deliberate synonym swapping (synonym cycling) 12. False scope ### 🎨 Style Patterns (6 types) 13. Overuse of dashes 14. Overuse of bold 15. Inline heading vertical lists 16. Title case in headings 17. Emojis 18. Curly quotes ### 💬 Communication Patterns and Filler Words (6 types) 19. Collaborative communication traces 20. Knowledge cutoff disclaimers 21. Ingratiating / sycophantic tone 22. Filler phrases 23. Over-qualification 24. Generic positive conclusions ## File Description - **`SKILL.md`** - Chinese version skill definition file - **`README.md`** - This documentation **Note:** For the English original, see [blader/humanizer](https://github.com/blader/humanizer) ## Manual Usage Method ### Basic Workflow 1. **Identify AI patterns** - Scan text against the 24 patterns listed in `SKILL.md` 2. **Rewrite problematic segments** - Replace AI traces with natural expressions 3. **Preserve core meaning** - Ensure information completeness 4. **Maintain appropriate tone** - Match the intended style of the text 5. **Inject authentic personality** - Give the text a "human touch" ### Key Principles #### ✨ Not just "clean," but "vivid" Avoiding AI patterns is just the foundation; good writing needs an authentic human voice: - **Have an opinion** - Don't just report facts; react to them - **Vary rhythm** - Mix long and short sentences - **Acknowledge complexity** - Real people have mixed feelings - **Use "I" appropriately** - First person is a sign of honesty - **Allow some messiness** - Perfect structure can feel mechanical - **Be specific about feelings** - Replace abstract summaries with concrete details #### Example Comparison **Before rewriting (AI flavor):** > The new software update serves as proof of the company's commitment to innovation. Additionally, it provides a seamless, intuitive, and powerful user experience—ensuring users can efficiently achieve their goals. This is not just an update, but a revolution in the way we think about productivity. **After rewriting (humanized):** > The software update added batch processing, keyboard shortcuts, and offline mode. Early feedback from test users is positive—most report faster task completion. **Changes:** - Removed exaggerated symbolism ("serves as proof") - Removed AI vocabulary ("additionally," "seamless") - Removed rule of three ("seamless, intuitive, and powerful") - Removed negative parallelism ("not just... but...") - Added specific features and real feedback ## Common AI Vocabulary Warning List The following words appear with unusually high frequency in AI-generated text: - Additionally, crucial, delve into, emphasize - Enduring, enhance, foster, gain - Highlight, interplay, complex/complexity - Landscape (abstract noun), pivotal, showcase - Tapestry (abstract noun), testament, underscore - Valuable, vibrant ## Contributing If you find translation issues or want to improve the documentation, please submit an Issue or Pull Request. ### Chinese Context Specifics During translation and adaptation, we considered the characteristics of Chinese writing: - Some English patterns manifest differently in Chinese (e.g., title case issues) - Added examples suitable for Chinese context - Adjusted some expressions to align with Chinese conventions ## References - [Wikipedia: Signs of AI writing](https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing) - Original guide source - [WikiProject AI Cleanup](https://en.wikipedia.org/wiki/Wikipedia:WikiProject_AI_Cleanup) - Wikipedia AI Cleanup Project - [blader/humanizer](https://github.com/blader/humanizer) - Original English project - [hardikpandya/stop-slop](https://github.com/hardikpandya/stop-slop) - Inspiration for the utility sections ## License This translation project follows the license of the original project. The core content is based on observations and summaries from the Wikipedia community. --- **Tip:** This tool is not designed to "deceive" AI detectors but to genuinely improve writing quality. The best way to "de-AI" text is to give it authentic human thought and voice. ## FAQ ### What is Humanizer-zh? Humanizer-zh is a tool for removing AI-generated traces from Chinese text, making it sound natural and human-written. It detects and fixes 24 specific AI writing patterns. ### How does Humanizer-zh detect AI writing? It scans text for patterns such as overused vocabulary, tricolon structure, vague attributions, and stylistic tics that are common in AI-generated content. ### Which AI patterns does it fix? It covers 24 patterns across four categories: content (e.g., exaggerated significance), language (e.g., 'AI vocabulary'), style (e.g., dash overuse), and communication (e.g., filler phrases). ### Will Humanizer-zh help me bypass AI detectors? No. It is designed to genuinely improve writing quality, not to deceive detection tools. Focus on making text sound authentically human. ### How do I use Humanizer-zh in Claude Code? After installation with npx skills add, you can invoke it via /humanizer-zh followed by your text or file name. It then rewrites the content. ### Is the tool only for Chinese text? Yes, Humanizer-zh is specifically adapted for Chinese. The original English version is available separately at blader/humanizer.