What problem does it solve? PDFs often convert to empty or mangled HTML: scanned pages have no text layer, headings and lists lose their structure, and repeated page headers pollute the output. This Skill guides you through the extract-pdf package so you can pick the right parsing engine and OCR strategy for each document. ## Core Features & Use Cases - Structured PDF-to-HTML conversion: convertPDFToHTML() infers headings from text-height statistics, detects lists, code and quote blocks, links footnotes, and strips repeated page headers. - Two independent switches: choose the parser with method (ts-block-algorithm, liteparse, liteparse-wasm) and where OCR runs with processor (frontend, hybrid, docling, or a remote processor URL). - Hybrid OCR with Granite Docling: scanPagesForOCR flags only pages with figures, tables, or no text layer, so expensive vision-model OCR runs only where needed; a bundled Hono server lets you self-host the model on Cloudflare Workers. - Use Case: A research pipeline ingests academic papers as PDFs. Use hybrid mode to convert text-layer pages instantly while OCR'ing only the figure-heavy pages, producing clean HTML for downstream indexing. ## Quick Start Ask the assistant to convert a PDF file or URL to structured HTML using extract-pdf, enabling hybrid OCR if the document contains scanned or table-heavy pages.
