yiyue31-translator

name	yiyue31-translator
description	当用户输入"翻译"，"translate" "translate article", "translate to Chinese", "改成中文"，"convert to Chinese"等指令时启用。当用户提供url、文件路径、直接粘贴内容，并表达翻译意图时启用。
version	2.0.0
author	Yiyue31

Tech Article Translator Skill

Description

Professional translator that converts English articles to Chinese with topic-based glossary management, translation style selection, and a generate-evaluate loop for translation quality assurance.

---

Directory

{skill-dir} = this SKILL.md's directory path. It means the directory where this SKILL.md is located.

---

User Input

Required Input

User should provide one of the following:

1. Article URL: URL of the English article
2. File path: Local file path
3. Article content: Directly pasted English article content

Optional Input

4. Translation style: Literal or Free, default Literal
5. Output filename: Specify the save filename, default uses translated title

---

Reusable Sub-workflows

Generate-Evaluate Loop

Iterative refinement: generate → evaluate → revise. Used by Step 2 and Step 4.

Parameters (caller provides):

• {max-rounds}: Maximum iterations
• {threshold}: Passing score out of 10
• {eval-prompt}: Path to evaluation prompt file
• {round-file}: Save pattern for round N output (include {N} placeholder for round number)
• {final-file}: Final output path on pass
• {eval-file}: Save pattern for evaluation report (include {N} placeholder)
• {eval-context}: Additional context to pass to evaluator (original article, glossary, style, etc.)
• {timer-tag}: Optional timer tag for global timeout. When set, also provide {timeout-seconds}.

Procedure:

1. If {timer-tag} provided: node {skill-dir}/scripts/timer.js start --tag {timer-tag}.
2. Each round N (1 to {max-rounds}):
   • If {timer-tag} provided: node {skill-dir}/scripts/timer.js check --tag {timer-tag} --timeout {timeout-seconds}. If "expired": true → use best candidate as final → exit.
   • Round 1: full generation per caller spec. Round 2+: revise based on previous evaluation Issues table + original input.
   • Save to {round-file} (replace {N} with round number).
   • Evaluate: call subagent with {eval-prompt}, providing round output and {eval-context} as input. Save report to {eval-file} (replace {N}).
   • Extract score: parse the **[X/10]** total line from evaluation report. Score ≥ {threshold} → copy to {final-file} → PASS, exit loop. Score < {threshold} → track as best candidate if highest so far, next round.
3. Rounds exhausted: copy best candidate to {final-file}, report score.

Evaluate Once

Single-shot evaluation, no loop.

Parameters: {eval-prompt}, {input-file}, {output-file}, {threshold} (default 8.0).

Procedure: Call subagent with {eval-prompt}, providing {input-file} content. Save report to {output-file}. Extract total score from **[X/10]** line. Score ≥ {threshold} → PASS, else FAIL.

---

Translation Workflow

Step 1: Retrieve Article Content

Retrieve article content based on input type:

• URL input: Use web-access skill to fetch the article. Alternatively, use wget or curl.
• File path input: Use the Read tool.
• Direct paste: Process directly.
• Missing content: Ask the user.

Processing:

1. Extract title (priority: heading → filename → first sentence words → untitled-{timestamp}). Sanitize filesystem-unsafe characters (/ \ : * ? " < > |).
2. File naming rule: lowercase, words connected with hyphens.
3. If {title}/translation/ exists, delete it.
4. If not markdown, convert preserving structure. Unconvertible elements: keep original with comments.
5. If unsure about conversion, ask user to confirm.
6. Save to {title}/translation/original-{title}.md.

Step 2: Topic Analysis & Glossary (Generate-Evaluate Loop)

Analysis requirements:

• Topic identification: Read glossary list from {skill-dir}/glossary/. Extract title, h2/h3 headings, keywords. Match against glossary topics. AI determines best topic — notify user, no confirmation needed.
• Language check: If primarily Chinese (>50%) or non-English, ask user to confirm.
• Glossary: Load {skill-dir}/glossary/{topic}.md, identify terms to keep in English. Scan for terms not in glossary.
• Hyperlinks: Extract from original article.

Run Generate-Evaluate Loop:

• {max-rounds}: 3, {threshold}: 8.0
• {eval-prompt}: {skill-dir}/references/evaluate-topic-prompt.md
• {round-file}: {title}/translation/analysis-topic-round{N}-{title}.md
• {final-file}: {title}/translation/analysis-topic-{title}.md
• {eval-file}: {title}/translation/evaluation-topic-round{N}-{title}.md
• {eval-context}: original article content, glossary file list from {skill-dir}/glossary/

After loop passes — user interaction:

Display to user:

## Article Analysis

**Original Title**: {Original English Title}
**Language**: {from} → {to}
**Identified Topic**: {Topic Name}
**Glossary Loaded**: {skill-dir}/glossary/{topic}.md ({N} terms)
**New Terms**: {list of new terms found}

1. Use AskUserQuestion for new terms: add all, add partial, skip, or custom input. Append confirmed terms to {skill-dir}/glossary/{topic}.md.
2. Ask translation style:

Option	Description
Literal	Word-by-word, preserve original structure, suitable for technical docs
Free	Adapt to Chinese conventions, restructure sentences, suitable for blogs

Default: Literal.

Step 3: Special Phrases Extraction

Exclusions: Skip terms already in the glossary.

Targets and translation rules:

• Golden quotes: Insightful, summarizing, or strongly opinionated sentences. Preserve expressiveness, append original as annotation.
• Hyphenated phrases: Technical phrases with "-" (e.g., "agent-based", "one-feature-at-a-time"). Keep original, append original as annotation.
• Idioms and slang: (e.g., "hit the ground running", "low-hanging fruit"). Translate naturally, append original as annotation.

Output format:

Location	Original	Chinese Translation (English original)
1.2	"This approach allows us to hit the ground running."	"这种方法让我们能够立刻投入工作（hit the ground running）"

Save to: {title}/translation/special-phrases-{title}.md.

Step 4: Translate (Generate-Evaluate Loop)

Translation rules:

• Accuracy first: Facts, data, logic must perfectly match the original. No additions, deletions, or subjective alterations. Translate from the reader's perspective.
• Terminology: Use glossary-specified translations. On first occurrence: keep English with Chinese annotation (e.g., Prompt(提示词)).
• Rhetoric: Translate metaphors and idioms by intent, not word-by-word. Replace with equivalent Chinese expressions when cultural connotations differ.
• Format: Preserve all Markdown formatting. For detailed rules, refer to {skill-dir}/references/markdown-format-checklist.md.
• Image text: Verify image text language matches translation.
• Translator notes: Add concise notes only when necessary: Chinese（English original, explanation）.
• Special phrases: Follow Chinese Translation column from Step 3.
• Golden quotes, idioms, slang: Bold: **{golden quote}**.
• Links: Preserve URLs, translate link text.

Free translation additional rules (skip when Literal):

• Translate core intent, restructure freely when literal translation is stiff.
• Preserve emotional connotations, not just dictionary definitions.
• Use natural Chinese word order.

Run Generate-Evaluate Loop:

• {max-rounds}: 5, {threshold}: 8.0
• {eval-prompt}: {skill-dir}/references/evaluate-translation-prompt.md
• {round-file}: {title}/translation/translated-round{N}-{title}-zh.md
• {final-file}: {title}/translation/translated-{title}-zh.md
• {eval-file}: {title}/translation/evaluation-round{N}-{title}.md
• {timer-tag}: {title}, {timeout-seconds}: 1800
• {eval-context}: original article, glossary, translation style (Literal/Free), special phrases table from Step 3

After each round save, run word count: node {skill-dir}/scripts/word-counter.js {title}/translation/translated-round{N}-{title}-zh.md, display results.

After loop completes:

• Add YAML frontmatter to final translation file:

---
title: {translated title}
source_title: {original English title}
source_url: {url or empty}
source_author: {author or empty}
translated_at: {date}
translation_style: literal or free
topic: {topic}
language: English → Chinese
word_count: {from word-counter}
---

• Run final word count: node {skill-dir}/scripts/word-counter.js {title}/translation/translated-{title}-zh.md
• Inform user, provide file path.

---

Glossary

File location: {skill-dir}/glossary/{topic}.md

# Topic: {Topic Name}

| English Term | Translation |
|--------------|-------------|
| AI | [KEEP] |
| Agent | 智能体 |
| API Gateway | API 网关 |

• [KEEP]: Do not translate, keep English
• Chinese translation: Always translate as specified