How a Half‑Hour Skill Pack Turned Claude Code into a One‑Click Book Translator

Problem

A 542‑page memoir (≈13.3 k English words) and a 336‑page AI‑competition history (≈9.7 k words) needed Chinese translation. Directly feeding the entire text to Claude Code would cost several hundred dollars in API tokens and require 7–8 hours of waiting.

Open‑source tool discovery

A ten‑minute GitHub search uncovered LinguaGacha (≈1.5 k stars), an open‑source pipeline that supports batch translation, terminology injection, checkpoint‑resume, and high concurrency for any OpenAI‑compatible model.

Workflow orchestration

Claude Code extracted pure text from the .docx file.

Multiple sub‑agents fetched terminology (person names, technical terms, places) from authoritative sources and produced a JSON glossary of 185 entries.

Claude Code generated a LinguaGacha configuration that split the book into 1 800+ translation tasks.

LinguaGacha performed the bulk translation using the chosen model.

Claude Code reformatted the output, rebuilt chapters, and produced a final Word document with appropriate fonts and line spacing.

Concurrency limitation and mitigation

Running 50 parallel translation agents caused HTTP 429 rate‑limit errors after about 800 entries. The issue was resolved by switching the model to gemini‑2.5‑flash and enabling LinguaGacha’s built‑in checkpointing, which resumed from the last successful entry without re‑translating completed segments.

Packaging as a reusable Skill

All scripts, configuration files, terminology JSON, and prompt templates were bundled into a Skill . The Skill was pushed to a GitHub repository with a single command: export GITHUB_TOKEN="ghp_xxxxxx" The repository URL is https://github.com/147228/south-asia-research-skills. This demonstrated that an AI agent can autonomously create a repository, push code, and make the workflow reusable.

Second book (ePub) using the same Skill

The same Skill was invoked with a one‑line command:

"C:\...\Supremacy_AI,_ChatGPT,...epub" 翻译这本书，学术风格，20并发

The process produced a 200 k‑word Chinese translation in minutes, automatically handling term consistency, formatting, and email delivery.

Results and metrics

First book: 1 821 entries successfully translated, 1 878 English entries retained, 53 format‑error entries skipped.

Final Chinese output: ~240 k words, generated in half a day.

Second book: complete Chinese ePub generated with a single command.

Key technical takeaways

Briefly searching for existing open‑source tools can avoid days of custom development.

Encapsulating the entire process as a Skill reduces repeat effort to a single command.

Providing each agent with the necessary tokens (GitHub, API keys) creates a fast, composable ecosystem.

Domain expertise (terminology awareness, formatting expectations) remains the most valuable input for AI agents.