deusyu/translate-book

Agent skill for Codex, Claude Code, and OpenClaw that translates entire books (PDF/DOCX/EPUB) into any language using parallel subagents.

What it solves

Translating entire books (PDF, DOCX, EPUB) using LLMs often leads to context window overflow, output truncation, and inconsistent terminology (where the same name is translated differently in different chapters). This project provides an automated pipeline to translate large documents in parallel while maintaining linguistic consistency and structural integrity.

How it works

The system operates as an agent skill for runtimes like Codex, Claude Code, and OpenClaw. It follows a multi-step pipeline:

  1. Conversion: Uses Calibre and Pandoc to convert the input book into Markdown chunks (~6,000 characters each).
  2. Glossary Generation: Samples the book to extract proper nouns and recurring terms, creating a canonical glossary to ensure consistent translation across all chunks.
  3. Parallel Translation: Deploys multiple sub-agents (8 by default) to translate chunks concurrently. Each sub-agent receives the relevant glossary terms and short excerpts from neighboring chunks to resolve pronouns and entities.
  4. Validation: Uses SHA-256 hashes (via a manifest file) to verify that every source chunk has a corresponding, non-empty translation.
  5. Reconstruction: Merges the translated chunks and uses Pandoc and Calibre to export the final result into HTML, DOCX, EPUB, or PDF formats.

Who it’s for

  • Users who need to translate long-form documents or e-books while preserving formatting.
  • Developers using AI agent frameworks (Codex, Claude Code, OpenClaw) who want to add high-capacity translation capabilities to their workflow.

Highlights

  • Parallel Processing: Uses isolated sub-agents to prevent context accumulation and speed up translation.
  • Consistency Tools: Implements a glossary system and neighbor-context excerpts to prevent terminology drift and pronoun errors.
  • Resumable Workflow: Tracks state via run_state.json, allowing the system to resume interrupted runs or selectively re-translate only the chunks affected by glossary changes.
  • Multi-Format Support: Handles a wide range of inputs and outputs, including PDF, EPUB, and DOCX.

Related

  • Project
    hydropix/TranslateBooksWithLLMs

    A desktop application that uses LLMs to translate books, subtitles, and documents while preserving original formatting and maintaining consistency across long texts.

  • Project
    yihong0618/bilingual_book_maker

    An AI-powered translation tool that creates bilingual versions of EPUB, TXT, MD, SRT, and PDF files using various LLMs and translation services.

  • Project
    virgiliojr94/book-to-skill

    A Python tool that converts PDFs, EPUBs, DOCX, Markdown, etc., into an Agent‑Skill bundle (SKILL.md + per‑chapter markdown files). The skill can be loaded by Copilot CLI, Amp, Claude Code, letting the LLM answer questions from the original book on‑demand while using 24‑51× fewer tokens than feeding the whole book as context.

  • Project
    openclaw/agent-skills

    A centralized library of shared workflows and helper scripts for coding agents to ensure consistent operational processes across multiple repositories.

  • Project
    fkirc/attranslate

    A CLI tool for syncing translation files that allows AI coding agents to efficiently translate new strings without overwriting manual corrections.