wtf is ky-markdown-rebuilder?

KY Markdown Rebuilder is a skill for AI coding assistants like Codex and Claude Code that rebuilds visually complex documents into reliable Markdown, aligned page by page. It is not a simple file to Markdown converter. Instead it extracts text, renders each original page as an image, checks that extracted content against the actual page layout, and runs scripts to verify page counts, formatting, and image links before calling the result done. It is meant for documents where plain text extraction usually goes wrong, such as PDFs, PowerPoint slides, scanned pages, long screenshots, reports, whitepapers, and course materials, especially ones with tables, flowcharts, diagrams, timelines, or multi column layouts. Ordinary conversion tools tend to keep only the text and lose reading order, mix content from adjacent pages together, break apart tables, and drop the relationships shown in diagrams. This project keeps each page or slide as its own section, checks visually dense pages at full resolution instead of guessing from thumbnails, and falls back to Tesseract OCR only when a scanned page cannot be handled by a vision model. Amounts, units, or small text that cannot be reliably read are explicitly flagged rather than guessed. It offers four output modes: a fast mode for mostly linear text, a page aligned mode that produces one Markdown block per page, a deep visual mode for high stakes or visually dense slide decks that checks every page image individually, and a transcribe mode for scanned books or tables that copies content one to one without added commentary. By default it produces a single calibrated Markdown file, with an optional outline file or a translated version available on request. Installation means cloning the repository into the Codex or Claude Code skills folder, and the project includes a preflight script that checks whether tools like pdftoppm, soffice, markitdown, and tesseract are available before running. A bundled output checker script verifies page continuity, formatting, and link correctness before a result is considered ready to deliver.