baoyu-translate skill

Translator

Three-mode translation skill: quick for direct translation, normal for analysis-informed translation, refined for full publication-quality workflow with review and polish.

User Input Tools

When this skill prompts the user, follow this tool-selection rule (priority order):

1. Prefer built-in user-input tools exposed by the current agent runtime — e.g., AskUserQuestion, requestuserinput, clarify, ask_user, or any equivalent.
2. Fallback: if no such tool exists, emit a numbered plain-text message and ask the user to reply with the chosen number/answer for each question.
3. Batching: if the tool supports multiple questions per call, combine all applicable questions into a single call; if only single-question, ask them one at a time in priority order.

Concrete AskUserQuestion references below are examples — substitute the local equivalent in other runtimes.

Script Directory

Scripts in scripts/ subdirectory. {baseDir} = this SKILL.md's directory path. Resolve ${BUNX} runtime: if bun installed → bun; if npx available → npx -y bun; else suggest installing bun. Replace {baseDir} and ${BUNX} with actual values.

Preferences (EXTEND.md)

Check EXTEND.md in priority order — the first one found wins:

EXTEND.md supports: default target language, default mode, target audience, custom glossaries (inline or file path), translation style, chunk settings.

Schema: references/config/extend-schema.md.

First-Time Setup (BLOCKING)

CRITICAL: When EXTEND.md is not found, you MUST run the first-time setup before ANY translation. This is a BLOCKING operation.

Full reference: references/config/first-time-setup.md

Use AskUserQuestion with all questions (target language, mode, audience, style, save location) in ONE call. After user answers, create EXTEND.md at the chosen location, confirm "Preferences saved to [path]", then continue.

Defaults

All configurable values in one place. EXTEND.md overrides these; CLI flags override EXTEND.md.

Modes

Default mode: Normal (can be overridden in EXTEND.md default_mode setting).

Style presets — control the voice and tone of the translation (independent of audience):

Custom style descriptions are also accepted, e.g., --style "poetic and lyrical".

Auto-detection:

• "快翻", "quick", "直接翻译" → quick mode
• "精翻", "refined", "publication quality", "proofread" → refined mode
• Otherwise → default mode (normal)

Upgrade prompt: After normal mode completes, display:

Translation saved. To further review and polish, reply "继续润色" or "refine".

If user responds, continue with review → polish steps (same as refined mode Steps 4-6 in refined-workflow.md) on the existing output.

Audience presets:

Custom audience descriptions are also accepted, e.g., --audience "AI感兴趣的普通读者".

Workflow

Step 1: Load Preferences

1.1 Check EXTEND.md (see Preferences section above)

1.2 Load built-in glossary for the language pair if available:

• EN→ZH: references/glossary-en-zh.md

1.3 Merge glossaries: EXTEND.md glossary (inline) + EXTEND.md glossary_files (external files, paths relative to EXTEND.md location) + built-in glossary + --glossary file (CLI overrides all)

Step 2: Materialize Source & Create Output Directory

Materialize source (file as-is, inline text/URL → save to translate/{slug}.md), then create output directory: {source-dir}/{source-basename}-{target-lang}/. Detect source language if --from not specified.

Full details: references/workflow-mechanics.md

Output directory contents (all intermediate and final files go here):

Step 3: Assess Content Length

Quick mode does not chunk — translate directly regardless of length. Before translating, estimate word count. If content exceeds chunk threshold (default 4000 words), proactively warn: "This article is ~{N} words. Quick mode translates in one pass without chunking — for long content, --mode normal produces better results with terminology consistency." Then proceed if user doesn't switch.

For normal and refined modes:

3.1 Long Content Preparation (normal/refined modes, >= chunk threshold only)

Before translating chunks:

1. Extract terminology: Scan entire document for proper nouns, technical terms, recurring phrases
2. Build session glossary: Merge extracted terms with loaded glossaries, establish consistent translations
3. Split into chunks: Use ${BUN_X} {baseDir}/scripts/main.ts [--max-words ] [--output-dir ]

• Parses markdown blocks (headings, paragraphs, lists, code blocks, tables, etc.)
• Splits at markdown block boundaries to preserve structure
• If a single block exceeds the threshold, falls back to line splitting, then word splitting

1. Assemble translation prompt:

• Main agent reads 01-analysis.md (if exists) and assembles shared context using Part 1 of references/subagent-prompt-template.md — inlining: target style, content background, merged glossary, and translation challenges
• Save as 02-prompt.md in the output directory (shared context only, no task instructions)

1. Draft translation via subagents (if Agent tool available):

• Spawn one subagent per chunk, all in parallel (Part 2 of the template)
• Each subagent reads 02-prompt.md for shared context, receives chunk position info (chunk N of M + brief context of where it sits in the argument), translates its chunk, saves to chunks/chunk-NN-draft.md
• Consistency is guaranteed by the shared 02-prompt.md (glossary, figurative language mapping, comprehension challenges, source voice, and translation challenges from analysis)
• If no chunks (content under threshold): spawn one subagent for the entire source file
• If Agent tool is unavailable, translate chunks sequentially inline using 02-prompt.md

1. Merge: Once all subagents complete, combine translated chunks in order. If chunks/frontmatter.md exists, prepend it. Save as 03-draft.md (refined) or translation.md (normal)
2. All intermediate files (source chunks + translated chunks) are preserved in chunks/

After chunked draft is merged, return control to main agent for critical review, revision, and polish (Step 4).

Step 4: Translate & Refine

Translation principles (apply to all modes):

• Rewrite, not translate: Rewrite content into natural, engaging target language as if a skilled native writer composed it from scratch. Quality test: "Does this read like it was originally written in the target language?"
• Accuracy first: Facts, data, and logic must match the original exactly
• Natural flow: Use idiomatic target language word order. Break long source sentences into shorter, natural ones. Interpret metaphors and idioms by intended meaning, not word-for-word
• Terminology: Use standard translations consistently. First occurrence of specialized terms: annotate with original in parentheses
• Preserve format: Keep all markdown formatting (headings, bold, italic, images, links, code blocks)
• Proactive interpretation: For jargon or concepts the target audience may lack context for, add concise explanations in bold parentheses （解释）. Keep annotations few — only where genuinely needed for comprehension
• Frontmatter: If source has YAML frontmatter, rename source-metadata fields with source prefix (camelCase: url→sourceUrl, title→sourceTitle, etc.), add translated values as new top-level fields (skip title if body has H1), keep other fields as-is

Quick Mode

Translate directly → save to translation.md. Apply all translation principles above.

Normal Mode

1. Analyze → 01-analysis.md (domain, tone, terminology, translation challenges)
2. Assemble prompt → 02-prompt.md (translation instructions with context, glossary, challenges)
3. Translate (following 02-prompt.md) → translation.md

After completion, prompt user: "Translation saved. To further review and polish, reply 继续润色 or refine."