This tackles the context window problem that kills most book translation attempts. Instead of feeding a 300-page PDF into one session and watching it choke or lose coherence halfway through, it splits the book into ~6000 character chunks, spins up 8 parallel Claude subagents that each translate one chunk in isolation, then validates and merges everything back into DOCX, EPUB, and PDF. The manifest.json tracks SHA-256 hashes so you can resume if something fails without re-translating finished chunks. Requires Calibre and Pandoc locally, which is a bit of setup friction, but the pipeline design is solid. Works with PDF, DOCX, and EPUB inputs. If you've ever tried to translate technical documentation or a full book and hit the context limit wall, this is the workaround.
npx skills add https://github.com/aradotso/trending-skills --skill translate-book-parallel