Book Translation Guide for Agents

Prerequisites

IMPORTANT: This skill only works if the book has already been parsed to Markdown format. The source file must be a .md file with proper heading structure using # symbols.

If the book is in PDF, EPUB, or other formats, it must first be converted to Markdown before using this translation workflow.

Prompt Example (ZIP input: PDF + assets)

Use this starter prompt when your input is a ZIP file containing a book PDF and asset files:

Translate this book to Traditional Chinese using:
- skills/book-translation-guide/SKILL.md
- agents/markdown-zh-tw-translator.md

https://github.com/madeyexz/ian-skills-agents

Input ZIP:
/path/to/book-package.zip

Requirements:
1. Unzip into a working folder.
2. If the source is PDF, convert it to markdown first (chaptered `.md` files).
3. Preserve markdown image attachments by keeping/copying asset folders and not changing image link targets.
4. Split oversized markdown chapters to ~40-60KB each.
5. Translate with sub-agents in batches of 5 files.
6. Merge translated parts and verify completeness with tail checks.
7. Output import-ready folders:
   - Book-Chapters/
   - Book-Chapters-ZH-TW/ (translated markdown + assets)

Overview

This guide provides a systematic approach to translating entire books between any language pair using Claude Code translator agents.

Supported translations include:

• English → Traditional Chinese, Japanese, Korean, Spanish, French, German, etc.
• Any source language → Any target language

The workflow is language-agnostic; you simply need a translator agent configured for your target language.

Step 1: Analyze Document Structure

First, understand the book's structure:

# Check file size and line count
wc -l "book.md" && ls -la "book.md"

# Find all chapter headings (lines starting with #)
grep "^# " "book.md"

This reveals:

• Total length of the book
• Number of chapters and their titles
• Starting line numbers for each chapter

Step 2: Create Directory Structure

mkdir -p "Book-Chapters"            # For original chapter files
mkdir -p "Book-Chapters-Translated" # For translated chapter files

You can name the translated folder based on your target language:

• Book-Chapters-ZH-TW (Traditional Chinese)
• Book-Chapters-JA (Japanese)
• Book-Chapters-ES (Spanish)
• etc.

Step 3: Split into Chapters

Use sed to extract chapters by line numbers:

sed -n 'START,ENDp' "book.md" > "Book-Chapters/00-Chapter-Name.md"

Example for a typical book:

sed -n '1,55p' "book.md" > "Book-Chapters/00-Front-Matter.md"
sed -n '56,593p' "book.md" > "Book-Chapters/01-Chapter-One.md"
sed -n '594,841p' "book.md" > "Book-Chapters/02-Chapter-Two.md"
# Continue for all chapters...

CRITICAL: Keep each file under 50-60KB to avoid token limits.

Step 4: Handle Oversized Chapters

If a chapter exceeds 60KB, split it further:

# Check chapter structure
grep "^# " "07-Large-Chapter.md"

# Split into parts at a logical section break
sed -n '1,231p' "07-Large-Chapter.md" > "07a-Part1.md"
sed -n '232,471p' "07-Large-Chapter.md" > "07b-Part2.md"

After translation, merge the parts:

cat "Book-Translated/07a-Part1.md" "Book-Translated/07b-Part2.md" > "Book-Translated/07-Large-Chapter.md"
rm "Book-Translated/07a-Part1.md" "Book-Translated/07b-Part2.md"

Step 5: Batch Translation Strategy

Translate 5 chapters at a time, waiting for completion before proceeding:

Batch 1: Chapters 00-04 → Wait for completion
Batch 2: Chapters 05-09 → Wait for completion
Batch 3: Chapters 10-14 → Wait for completion
...continue until done

Benefits:

• Avoids rate limits
• Allows early detection of issues
• Ensures consistent translation quality

Step 6: Invoke Translation Agent

For each chapter, call the appropriate translation agent based on your target language:

Example: English → Traditional Chinese (Taiwan)

Task(
  subagent_type: "markdown-zh-tw-translator",
  description: "Translate Chapter X to ZH-TW",
  prompt: "Translate the markdown file at /path/to/Book-Chapters/XX-Chapter.md
           from English to Traditional Chinese (繁體中文).

           Save the translated file to /path/to/Book-Chapters-ZH-TW/XX-Chapter.md

           Use your translation guidelines to ensure high quality translation
           while preserving markdown formatting."
)

Example: English → Japanese

Task(
  subagent_type: "markdown-ja-translator",
  description: "Translate Chapter X to Japanese",
  prompt: "Translate the markdown file at /path/to/Book-Chapters/XX-Chapter.md
           from English to Japanese (日本語).

           Save the translated file to /path/to/Book-Chapters-JA/XX-Chapter.md

           Use your translation guidelines to ensure high quality translation
           while preserving markdown formatting."
)

Example: French → English

Task(
  subagent_type: "markdown-en-translator",
  description: "Translate Chapter X to English",
  prompt: "Translate the markdown file at /path/to/Book-Chapters/XX-Chapter.md
           from French to English.

           Save the translated file to /path/to/Book-Chapters-EN/XX-Chapter.md

           Use your translation guidelines to ensure high quality translation
           while preserving markdown formatting."
)

Run 5 agents in parallel for each batch.

Step 7: Verify Translation Completeness

After all chapters are translated, perform side-by-side verification comparing the last 10 lines of each chapter between original and translated versions.

Comprehensive Side-by-Side Verification

For each chapter, compare original and translated endings in parallel:

echo "=== 01-Chapter ===" && \
echo "-- Original --" && \
tail -10 "Book-Chapters/01-Chapter.md" && \
echo -e "\n-- Translated --" && \
tail -10 "Book-Chapters-Translated/01-Chapter.md"

Run this for all chapters in parallel to verify:

• Translation reaches the end of each chapter (no truncation)
• Final paragraphs match in meaning
• Reference links and citations are preserved
• Formatting (headers, lists, links) is intact

Bulk Verification Script

for chapter in Book-Chapters/*.md; do
  name=$(basename "$chapter")
  echo "=== $name ==="
  echo "-- Original --"
  tail -10 "$chapter"
  echo -e "\n-- Translated --"
  tail -10 "Book-Chapters-Translated/$name"
  echo -e "\n"
done

What to Check

Quick Verification (Single Chapter)

# Compare original ending
tail -5 "Book-Chapters/01-Chapter.md"

# Compare translated ending
tail -5 "Book-Chapters-Translated/01-Chapter.md"

Translation Quality Standards

General principles that apply to all languages:

Language-Specific Considerations

Traditional Chinese (Taiwan):

• Use Taiwan conventions: 軟體, 程式, 資料, 網路

Japanese:

• Use appropriate politeness levels
• カタカナ for loanwords

Spanish:

• Consider regional variations (Spain vs Latin America)
• Appropriate formality (tú vs usted)

German:

• Compound word handling
• Formal vs informal address (Sie vs du)

Complete Workflow Diagram

Original File (book.md)
        ↓
[1] Analyze Structure (grep chapter headings)
        ↓
[2] Create Directories
        ↓
[3] Split into Chapters (sed by line numbers)
        ↓
[4] Check File Sizes (split if >60KB)
        ↓
[5] Batch Translate (5 chapters per batch)
        ↓
[6] Merge Split Chapters (if any)
        ↓
[7] Verify Completeness (tail comparison)
        ↓
[8] Done - All chapters translated

Troubleshooting

Rate Limit Errors

• Reduce batch size from 5 to 3 chapters
• Wait longer between batches

Token Limit Exceeded

• Split the chapter into smaller parts
• Target 40-50KB per file instead of 60KB

Incomplete Translation

• Check if agent hit limits mid-translation
• Re-run translation for that specific chapter
• Verify by comparing file endings

Missing Chapters

• List output directory to identify gaps
• Re-translate missing files individually

Wrong Language Conventions

• Create a custom translator agent with specific rules for your target language/region

Example: Real Book Translations

Notes

1. Always verify the source markdown has proper # heading structure
2. Back up original files before starting
3. Keep a checklist of chapters to track progress
4. For books >1MB, consider translating over multiple sessions
5. Create or configure a translator agent for your target language before starting
6. The workflow is the same regardless of language pair - only the translator agent changes