cuimao-translator

Installation
SKILL.md

萃猫翻译 · Cuimao Translator

Skill by ara.so — Daily 2026 Skills collection.

A Claude Code skill that translates English PDF books into idiomatic, natural-sounding Chinese. Designed for book-length texts, it avoids "translation smell" through mode selection, style presets, and automated Europeanized-Chinese detection.


Installation

git clone https://github.com/Cuimao777/cuimao-translator.git ~/.claude/skills/cuimao-translator

Restart Claude Code. The skill auto-activates when you say phrases like "翻译这个PDF" or "translate this PDF".


File Structure

cuimao-translator/
├── SKILL.md                      # Main skill file (auto-loaded by Claude Code)
├── README.md
└── references/
    ├── glossary-en-zh.md         # 100+ English-Chinese term pairs
    ├── translation-guide.md      # Sentence patterns, diagnostics, register guides
    └── refined-workflow.md       # 5-step refined translation pipeline

Three Translation Modes

Mode Trigger Words Pipeline Best For
⚡ Quick 快翻 速译 quick Read → Translate Skimming, short snippets
📖 Normal 翻译 (default) Analyze → Translate Most books, articles
💎 Refined 精翻 精细翻译 出版级 Analyze → Draft → Review → Revise → Polish Publication-quality output

You can upgrade after Normal completes: reply 继续润色 or refine to enter the Refined pipeline.


Five Style Presets

Style Key Use Case Characteristics
storytelling Novels, biography, narrative non-fiction Natural transitions, Chinese rhythm
academic Textbooks, research papers Formal register, precise terminology
literal Manuals, legal docs, contracts Close to source structure, minimal rewrite
elegant Literary fiction, poetry, essays Four-character idioms, musicality
conversational Self-help, memoirs, dialogue-heavy Relaxed, reader-friendly tone

If not specified, the skill auto-detects from content. storytelling is the safe default for most books.


How to Use

Basic Usage

Just say it naturally in Claude Code chat:

翻译这个 PDF
translate this PDF book to Chinese
帮我把这本书汉化,用 storytelling 风格

Specify Mode + Style

精翻这个PDF,academic 风格
快翻这几页,conversational 风格
出版级翻译,elegant 风格

Upgrade After Normal Translation

继续润色
refine

Translation Philosophy: 信 → 达 → 雅

信 (Faithful)

  • Every English sentence has a corresponding Chinese sentence
  • Facts, numbers, logic, proper nouns are preserved exactly
  • No summarizing, skipping, or embellishment

达 (Fluent)

  • Output reads as if written by a native Chinese author
  • Topic-comment sentence structure (Chinese-native order)
  • Split long sentences into 7–15 character natural chunks
  • Active voice preferred; complement clauses over adverbial chains

雅 (Culturally Adapted)

  • English idioms/slang → find Chinese equivalent expressions
  • Untranslatable wordplay → translate meaning + add note
  • Unknown cultural references → brief translator's note

Europeanized Chinese Auto-Detection (六重检测)

After every translation, the skill self-checks for six common "translation smell" patterns:

# Problem Bad Example Fix
1 被字句泛滥 他被表扬了 他受到了表扬
2 连接词堆砌 因为…所以…虽然…但是… Remove redundant connectors
3 定语过长 那个在昨天下午三点去买东西的人 Split into multiple short sentences
4 之一滥用 最重要的问题之一 极其重要的问题
5 的的的 ≥3 in one sentence Restructure the sentence
6 名词化过度 政策的实施 实施政策

Real Example

English Input:

The old man had taught the boy to fish and the boy loved him.
He was an old man who fished alone in a skiff in the Gulf Stream
and he had gone eighty-four days now without taking a fish.

Chinese Output (storytelling style):

老头儿教过那孩子打鱼,孩子也爱他。
他是个独自在湾流里的一只小船上打鱼的老头儿,
如今已经接连八十四天一条鱼也没打着了。

Working with the References

Glossary (references/glossary-en-zh.md)

Contains 100+ curated English→Chinese term mappings. The skill applies these automatically for consistency across an entire book. When translating a new domain-specific text, you can extend it:

<!-- Add domain terms at the bottom of glossary-en-zh.md -->
| machine learning | 机器学习 |
| gradient descent | 梯度下降 |
| overfitting | 过拟合 |
| hyperparameter | 超参数 |

Translation Guide (references/translation-guide.md)

Covers:

  • Sentence transformation patterns (passive → active, noun phrase → verb phrase)
  • Register-specific vocabulary lists
  • How to handle culturally-specific content per genre

Refined Workflow (references/refined-workflow.md)

Documents the 5-step Refined pipeline in detail:

  1. Analyze — genre, register, key terms, author voice
  2. Draft — first-pass translation
  3. Review — check against 六重检测
  4. Revise — fix flagged issues
  5. Polish — rhythm, flow, final read-aloud check

Practical Patterns

Translate a Specific Chapter

翻译第三章,storytelling 风格
translate chapter 5 only, academic style

Batch Translation with Consistency

For multi-chapter books, mention the book title so the skill maintains term consistency:

这是《Thinking, Fast and Slow》的第二章,请继续保持之前的翻译风格和术语

Custom Glossary Injection

Paste a term list before the content:

请使用以下术语表翻译:
- neural network → 神经网络
- token → 词元
- prompt → 提示词

[paste content here]

Output Format Control

翻译后按段落对照输出(英文段落 + 中文段落交替)
只输出中文译文,不要对照

Trigger Reference

Trigger Effect
翻译 / translate Normal mode, auto-detect style
汉化 / 英译中 Normal mode
速译 / 快翻 / quick Quick mode
精翻 / 精细翻译 / 出版级 Refined mode
继续润色 / refine Upgrade current translation to Refined
storytelling / 叙事 Force storytelling style
academic / 学术 Force academic style
literal / 逐句 Force literal style
elegant / 文学 Force elegant style
conversational / 口语 Force conversational style

Troubleshooting

Skill not activating after install:

  • Restart Claude Code completely
  • Verify file is at ~/.claude/skills/cuimao-translator/SKILL.md
  • Check file permissions: chmod -R 644 ~/.claude/skills/cuimao-translator/

Translation sounds too formal / too casual:

  • Explicitly specify a style preset: 用 conversational 风格重译
  • For mixed-register texts, specify per section

Terminology inconsistent across chapters:

  • Start each session by referencing the book title
  • Add domain terms to references/glossary-en-zh.md before translating

Output contains Europeanized Chinese patterns:

  • Run: 检查译文中的翻译腔,按六重检测标准修正
  • For Refined mode, this check runs automatically

PDF content not parsing correctly:

  • Extract text from PDF first, then paste into chat
  • For scanned PDFs, run OCR before translation

License

MIT — see LICENSE

Related skills
Installs
21
GitHub Stars
1
First Seen
4 days ago