Writing Humanizer
Remove AI writing patterns from text, making it sound more natural and human. Focused on Traditional Chinese (Taiwan).
One skill, seven hosts. The same skills/writing-humanizer skill runs in Claude Code, OpenAI Codex, Cursor, Google Antigravity CLI, OpenCode, Kimi CLI, and pi — only the per-host manifest differs.
Installation
Claude Code
Via marketplace (recommended):
/plugin marketplace add shyuan/shyuan-marketplace
/plugin install writing-humanizer@shyuan-marketplace
Or direct:
/plugin install shyuan/writing-humanizer
OpenCode
Add to the plugin array in your opencode.json (global or project-level), then restart OpenCode:
{
"plugin": ["writing-humanizer@git+https://github.com/shyuan/writing-humanizer.git"]
}
See .opencode/INSTALL.md for version pinning and troubleshooting.
Codex / Cursor / Antigravity CLI / Kimi CLI
Each of these hosts reads its own manifest bundled in this repo. Install with the host's own plugin/marketplace command pointing at this repo, or clone it into the host's plugins location and let the host discover the manifest:
git clone https://github.com/shyuan/writing-humanizer.git
| Host | Manifest the host reads |
|---|
| OpenAI Codex | .codex-plugin/plugin.json |
| Cursor | .cursor-plugin/plugin.json |
| Google Antigravity CLI | plugin.json (repo root) |
| Kimi CLI | .kimi-plugin/plugin.json |
pi
pi loads the skill through the "pi" field in package.json (extension at .pi/extensions/writing-humanizer.ts). Install the package into your pi setup so pi discovers the bundled skills/.
What You Get
Skill: writing-humanizer
A comprehensive skill that detects and rewrites 31 categories of AI writing patterns:
Content Patterns (1-6)
- Significance inflation, notability name-dropping, superficial analysis
- Promotional language, vague attributions, formulaic "challenges & outlook"
Language Patterns (7-12)
- AI vocabulary overuse, copula avoidance, negative parallelisms
- Rule of three, synonym cycling, false ranges
Style Patterns (13-17)
- Em dash overuse, boldface overuse, inline-header lists
- Emoji decoration, quotation mark issues
Communication Patterns (18-24)
- Chatbot artifacts, knowledge-cutoff disclaimers, sycophantic tone
- Filler phrases, excessive hedging, generic conclusions, throat-clearing openers
Structure & Rhetoric Patterns (25-31) — Chinese-essay-specific
- Outline-as-essay, bold 4-char-label lists, significance-stamping endings
- In-sentence keyword bolding, meta-discourse/roadmap announcements, sublimating moralizing conclusions
- Reifying an abstract process into "a line/route" and pointing back to it repeatedly
Key Features
- AI vocabulary watchlist with suggested replacements for Traditional Chinese (Taiwan)
- Common AI sentence templates specific to Chinese writing
- Two-pass rewriting process with self-audit
- Quality scoring system (50 points across 5 dimensions)
Quick Start
After installation, invoke the skill (slash-command syntax varies by host — this is the Claude Code form):
/writing-humanizer:writing-humanizer <paste your text here>
Or simply ask your agent to humanize text — the skill is discovered from its description and loaded on demand.
References
License
MIT
Writing Humanizer（中文說明）
去除文章中的 AI 寫作痕跡，使文字更自然、更有人味。以台灣正體中文為主。
一份 skill，七個 host。同一個 skills/writing-humanizer skill 可在 Claude Code、OpenAI Codex、Cursor、Google Antigravity CLI、OpenCode、Kimi CLI、pi 上運作——差別只在各 host 的 manifest。
安裝方式
Claude Code
透過 Marketplace（推薦）：
/plugin marketplace add shyuan/shyuan-marketplace
/plugin install writing-humanizer@shyuan-marketplace
或直接安裝：
/plugin install shyuan/writing-humanizer
OpenCode
在你的 opencode.json（全域或專案層級）的 plugin 陣列加入以下設定，然後重啟 OpenCode：
{
"plugin": ["writing-humanizer@git+https://github.com/shyuan/writing-humanizer.git"]
}
版本鎖定與排錯請見 .opencode/INSTALL.md。
Codex / Cursor / Antigravity CLI / Kimi CLI
這幾個 host 各自讀取 repo 內附的 manifest。用該 host 自己的 plugin／marketplace 安裝指令指向本 repo，或將 repo clone 到該 host 的 plugins 位置，讓它探索對應 manifest：
git clone https://github.com/shyuan/writing-humanizer.git
| Host | 讀取的 manifest |
|---|
| OpenAI Codex | .codex-plugin/plugin.json |
| Cursor | .cursor-plugin/plugin.json |
| Google Antigravity CLI | plugin.json（repo 根目錄） |
| Kimi CLI | .kimi-plugin/plugin.json |
pi
pi 透過 package.json 的 "pi" 欄位載入 skill（extension 在 .pi/extensions/writing-humanizer.ts）。將此套件安裝進你的 pi 環境，pi 便會探索到內附的 skills/。
安裝後你會得到
Skill：writing-humanizer
一個完整的 skill，可偵測並改寫 31 類 AI 寫作模式：
內容模式（1-6）
- 誇大象徵意義、過度強調知名度、膚淺分析
- 宣傳性語言、模糊歸因、公式化的「挑戰與展望」
語言模式（7-12）
- AI 詞彙過度使用、繫動詞迴避、否定式排比
- 三段式法則、同義詞循環、虛假範圍
風格模式（13-17）
- 破折號過度使用、粗體過度使用、內嵌標題列表
- 表情符號裝飾、引號問題
交流模式（18-24）
- 聊天機器人痕跡、知識截止免責聲明、諂媚語氣
- 填充短語、過度限定、通用積極結論、開場白贅詞
結構與修辭模式（25-31） — 中文論說文專屬
- 大綱骨架代替文章、四字標籤排比清單、意義蓋章式收尾
- 句內關鍵詞粗體、元論述導讀宣告、升華訓誡式結尾
- 抽象事物具象成「一條線／路線」並反覆回指
主要特色
- 台灣正體中文 AI 詞彙警示列表與替代建議
- 中文常見 AI 句式模板辨識
- 兩輪改寫流程（初稿 + 自我審查再修改）
- 質量評分系統（5 個維度，滿分 50 分）
快速入門
安裝完成後，呼叫 skill（slash 指令語法因 host 而異，以下為 Claude Code 形式）：
/writing-humanizer:writing-humanizer <貼上你要處理的文字>
或直接請你的 agent 幫你去除 AI 痕跡——skill 會依其 description 被探索並按需載入。
參考資料
授權條款
MIT
