Kobo Extract

Converts an EPUB → one Markdown file per chapter, ready for LLM ingestion.

The downstream goal is knowledge → skill: feed the chunked Markdown to Claude, distill into an outline, then refine into a reusable skill (typically via dev-workflow:skill-creator-advance).

Why a custom converter?

Many CJK EPUBs use Kobo-specific markup (<span class="koboSpan">) instead of semantic <h1> for chapters. Running pandoc directly produces zero markdown headings, which kills downstream chunking. This skill: