Humanize Chinese Writing

中文文本与 HTML 内容的去 AI 味、人性化改写、素材型长文、页面信息密度检查与分析稿审阅。用于用户说“去AI味”“像人写的”“自然一点”“保留原意改写”，或要求减少黑话、套话、模板腔、讲义腔、机器感，按个人文风润色中文稿、长文、HTML 页面文案、仪表盘和分析报告。优先保留事实、观点、语气、引用、链接、数据口径与代码结构，不以规避检测器为目标。

5 starsWriting & Content

Install

skills CLI (Claude Code, Cursor, Codex and more)

npx skills add fy-agent/humanize-chinese-writing --skill humanize-chinese-writing

Manual install

git clone --depth 1 https://github.com/fy-agent/humanize-chinese-writing.git /tmp/humanize-chinese-writing
mkdir -p ~/.claude/skills
cp -r /tmp/humanize-chinese-writing/. ~/.claude/skills/humanize-chinese-writing

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# 去 AI 味中文写作

把文字改得像一个具体的人在具体情境里说话，而不是把它改成“更像人类”的统一模板。

## 第一性原理

将“AI 味”判断为一种语境失配：模型倾向选择对大多数场景都通用的平均表达，于是把具体处写成抽象话，把段落排成整齐的模板，把每个逻辑空隙都填满，并用助手式礼貌和漂亮结论包住内容。

将“人性化”定义为四件事的同时成立：

1. 内容有出处。保留原文事实、观点和证据，不用虚构经历或细节填空。
2. 说话有位置。让作者的立场、取舍、犹豫和语气来自输入或用户提供的样本。
3. 组织有选择。只在有理由的地方展开、跳过、合并或停住，不把每段都写成完整小论文。
4. 表达有节奏。打散过度均匀的句长、段厚和转折，但不靠俚语、错别字或随机跑题表演“真实”。

按以下优先级取舍：事实与原意 > 作者声音与任务语气 > 清晰度 > 节奏变化 > 新鲜感。后面的目标不能破坏前面的目标。

## 先判断任务模式

根据用户请求选择一种模式，不要默认做大改：

- **保真改写（默认）**：只修补让人感觉“模型在组织答案”的痕迹，保留信息厚度和核心判断。
- **自由重写**：只有用户明确允许重排、增删或改变叙述角度时才使用；仍不得编造事实。
- **从零生成**：先使用用户给出的受众、场景、作者立场、具体事实和目标长度。缺少个人材料时写清楚、写具体，但不要假装有个人经历。
- **风格匹配**：用户提供写作样本时，先抽取样本的句长、常用词、段落开法、标点、第一人称、幽默和收尾方式，再按样本写。样本优先于本 Skill 的通用偏好。
- **素材型长文/公众号**：先确认素材、作者立场和文章要解决的问题，再选择一种真实的推进方式，例如调查/体验、现象解读、工具分享或方法经验，不套万能大纲。用 H/K/R 做轻量选题检查，分别看是否有兴趣点、信息增量和读者共鸣；至少占两项才值得扩写。缺一手经历时，不得补造经历，用已给材料、可核验来源或明确的未知代替。
- **审阅/找 AI 味**：只指出最影响读感的 Top 5 问题，逐条引用原句、说明成因并给出修法；不要丢出一份黑名单。
- **HTML/分析页面**：先检查页面是否有实质信息，再润色文案。为页面锁定一个主问题、证据范围、当前判断和下一步动作；删掉只占空间的黑话标题、口号、空卡片和重复说明。
- **文件处理**：只改 prose 或 HTML 可见文案；保留标签、属性、CSS/JS、frontmatter、表格数据、引用内容、数据字段和链接目标。只有用户明确要求写回时才改文件。

如果任务是高保真翻译，优先保留原文信息顺序、标题和语气，不把译文擅自改成评论文章。

## 执行流程

### 0. 先过页面与分析的内容密度门

只对 HTML、仪表盘、报告页和文本分析任务执行本节。先回答“这页要帮谁做什么决定”，再决定标题、卡片和版式。

1. 为页面或分析稿写出一个主问题。首屏直接给出状态、结论、变化或待处理事项，不用“赋能、洞察、升级、探索”这类词代替答案。
2. 给每个可见模块指定唯一职责：事实、证据、判断、建议、状态或恢复动作。没有职责的模块直接删除；重复职责的模块合并。
3. 给 KPI、图表和结论补足对象、时间范围、口径、对比或来源。只有“增长明显”“用户活跃”“情况良好”的卡片不算信息。
4. 把**事实、判断、建议**分开写：事实说明来源和范围；判断标明是基于事实的推断；建议写清动作、对象或时点。不要把猜测包装成数据结论。

Read the full skill on GitHub

Details

Skill name: humanize-chinese-writing
Repository: fy-agent/humanize-chinese-writing
Repo updated: 11 August 2026

FAQ

What does the Humanize Chinese Writing skill do?

How do I install the Humanize Chinese Writing skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add fy-agent/humanize-chinese-writing --skill humanize-chinese-writing. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Humanize Chinese Writing popular?

It lives in fy-agent/humanize-chinese-writing, which has 5 GitHub stars and 0 forks.

Related skills

Machiavelli Digital Twin / 马基雅维利分身

niflhum/machiavelli-skill

A conversational digital twin of Niccolò Machiavelli, distilled from his complete works. Delivers power analysis, strategic decision-making frameworks, and risk assessment with Machiavelli's signature style - based on how things actually are, not how they ought to be. / 以马基雅维利的思维方式、分析框架和表达风格,为用户提供局势分析、决策建议和风险评估。 Three modes: /machia (dialogue), /machia-a (analysis), /machia-w (writing), /machia-e (eastern). Not for: pure emotional venting, everyday trivia, technical questions.

Writing & Content

Grantproof

ellewangyuan/grantproof

Surfaces common writing-style mistakes in federal R&D grant prose — NSF SBIR/STTR, DoD SBIR, NIH SBIR, DOE, DARPA, NASA, NOAA, and similar. For each mistake, identifies where the issue is, explains why reviewers downgrade it, and shows a sample rewrite structure the founder can apply. Use when the user pastes draft writing for an R&D grant pitch or proposal.

Writing & Content

Mr Li Writer

NocodeMrLi/mr-li-writer-skill

面向中文内容创作的编辑判断、研究、策划、写作与平台原生交付工作流。根据标题、素材、参考资料、核心想法或模糊灵感，先识别读者任务、材料条件和创新必要性，再按需完成语境扫描、可写性判断与方向校准，并区分发布平台与内容目标，进行事实检索、观点设计、平台原生正文写作、标题策划和对应交付。适用于公众号文章、知乎回答、小红书原生笔记、官网/网页、个人博客、行业解读、实用指南和基于素材的扩展研究写作。

Writing & Content

Godot Plugin

TomLeeLive/openclaw-godot-skill

Control Godot Editor via OpenClaw Godot Plugin. Use for Godot game development tasks including scene management, node manipulation, input simulation, debugging, and editor control. Can modify, save, and delete scenes/nodes and capture project metadata and screenshots — destructive operations should be confirmed with the user. Use only on explicit Godot Editor requests like inspecting scenes, creating nodes, taking screenshots, testing gameplay, or controlling the editor, in trusted local projects.

Writing & Content

Mv3 Doctor

n0liu/mv3-doctor

Use when building, modifying, migrating, or reviewing a Chrome extension — anything involving manifest.json, a background service worker, content scripts, or Manifest V3. Encodes the MV3 rules that models routinely get wrong, and ships a validator to check the result.

Writing & Content

Game Terminology Vietnamization

dungnotnull/game-terminology-vietnamization-agent-skill

Vietnamization & Localization of Deep Game Terminology (No Misleading Translations) — A production-grade Game Localization & Specialized Terminology Translation analysis harness with evidence-backed outputs, quality gates, and continuous knowledge updates. Use this skill whenever the user asks about game terminology translation, game localization, Vietnamization of game terms, specialized terminology translation, localization quality assurance, or needs analysis of game terminology in Vietnamese/English contexts.

Writing & Content

Browse the most popular Claude skills, all writing & content skills, or write your own with the skills generator.