A style guide for Traditional Chinese (Taiwan) writing and translation that strips AI-generated prose habits, packaged as a Claude Code skill.

Stars

10

7-day growth

No data

Forks

0

Open issues

0

License

CC-BY-SA-4.0

Last updated

2026-07-29

AI repository intelligence
FR-AI / ANALYSIS

Why it is worth attention

It tackles a growing problem—detectable AI writing patterns—with concrete, testable rules drawn from real translation work, and it is packaged as an actionable Claude Code skill that auto-applies to tasks.

Who it is for

  • Traditional Chinese writers and translators
  • Claude Code users
  • AI content editors and reviewers
  • Taiwan-based content creators

Use cases

  • Editing AI-generated Traditional Chinese text to remove machine-like patterns
  • Translating English to Traditional Chinese while avoiding calque and AI habits
  • Writing articles or reports with a natural, human-sounding Traditional Chinese style
  • Integrating a style guide as an automated skill in Claude Code workflows

Strengths

  • Rule-based with executable tests (deletion test, straw-man test) for borderline cases
  • Covers both common AI habits and Taiwan-specific vocabulary/usage
  • Includes a mandatory ten-item self-check checklist before output
  • Grounded in a real book translation project (16th-century music treatise)

Considerations

  • Only applies to Traditional Chinese (Taiwan variety)
  • Requires Claude Code to benefit from automatic triggering
  • Some rules may be subjective or context-dependent

README quick start

de-ai-tone

去 AI 味:繁體中文寫作與翻譯的風格規範,做成 Claude Code skill。 A style guide for Traditional Chinese (Taiwan) writing and translation that strips the telltale habits of AI-generated prose, packaged as a Claude Code skill.


中文說明

這是什麼

AI 生成的中文有一組可辨識的統計慣性:假對比、破折號濫用、三項式排比、空轉的話語標記、每段收尾的微型總結。單獨看都不算錯,堆疊起來就是「AI 味」。本規範逐條列出這些慣性,禁用或限用,並附生成後的強制自檢清單。核心原則只有一條:每個修辭手段都必須賺回自己的位置。

裝進 Claude Code 之後,任何以繁體中文產出連續文字的任務(翻譯、撰稿、長篇回覆)會自動套用。

規範涵蓋

  • 禁用與限用清單(第 1–10 條):「不是 X 而是 Y」假辯證、破折號、三項式排比、空轉話語標記、貼標語彙(賦能、打造、旨在)、結構強迫症(濫用列點、加粗、小標題)、開場與收場公式、修辭性設問、贅詞「一個」、框式介詞。
  • 翻譯專用(第 11–12 條):歐化句式的還原——關係子句拆句、「被」字句改寫、代名詞刪減,以及慣用直譯句型的替代方案。忠實優先於流暢:原文刻意的修辭不在清理之列。
  • 臺灣用語規範(第 13–15 條):支語過濾對照表,按資訊科技、互聯網黑話、音樂與音訊工程、數學科學、學術日常分組;表外詞彙的判斷原則;英文命名禁止無意義的拼音音譯。
  • 標點規範(第 16 條):中文行文一律全形,唯英文與英文之間的標點用半形。
  • 正向原則與自檢:動詞驅動、短句節奏、具體優於抽象、信任讀者、有立場;產出前強制執行十項逐條掃描。

安裝

git clone https://github.com/allenloves/de-ai-tone.git ~/.claude/skills/de-ai-tone

Claude Code 讀取 SKILL.md 的 description 自動觸發;也可在對話中以 /de-ai-tone 手動載入。

背景

規範出自實際的翻譯與寫作工作,其中包括一部十六世紀音樂理論論著的英譯中全書專案。條目的演進方式是:發現一個實例,收一條規則,配一個可執行的判準(刪除測試、稻草人測試)。


English

What this is

AI-generated Chinese prose carries a recognizable set of statistical habits: fake dialectic contrasts, em-dash overuse, triadic parallelism, throat-clearing discourse markers, a miniature summary at the end of every paragraph. None of them is wrong in isolation; stacked, they read unmistakably as machine output. This guide enumerates those habits and bans or rations them, ending with a mandatory post-generation checklist. The whole thing rests on one principle: every rhetorical device must earn its place.

Installed as a Claude Code skill, it applies automatically to any task that produces running text in Traditional Chinese.

What it covers

  • Bans and rations (rules 1–10): fake "not X but Y" dialectics, em-dashes, triadic parallelism, empty discourse markers, buzzword vocabulary, structural compulsions (bullet points, bold, headings), formulaic openings and closings, rhetorical self-questioning, article-calque "一個", frame prepositions calqued from English.
  • Translation-specific (rules 11–12): undoing Europeanized syntax — splitting relative clauses, rewriting passive constructions, trimming pronouns — plus alternatives to stock calque patterns. Fidelity outranks fluency: deliberate rhetoric in the source is not cleaned up.

Description

去 AI 味:繁體中文寫作與翻譯的風格規範(Claude Code skill)

Related repositories

Similar projects matched by category, topics, and programming language.

MoonshotAI
Featured
MoonshotAI GitHub avatar

Kimi-K3

Kimi K3 is an open-weight, 2.8T-parameter native multimodal agentic model with a 1M-token context window, designed for frontier coding, knowledge work, and reasoning tasks.

AI & Machine LearningAI Agents
3,348
lopopolo
Featured
lopopolo GitHub avatar

harness-engineering

Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.

AI & Machine LearningAI Agents
2,390
slvDev
Featured
slvDev GitHub avatar

esp32-ai

A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.

AI & Machine LearningLarge Language Models
1,960