Pdf2md Skill

pdf2md-skill

A Claude Code / Cursor skill for converting any PDF into clean, structured Markdown — with equations, tables, and figures intact.

What This Does

**pdf2md-skill** makes PDFs readable for AI agents. Instead of losing tables, mangling equations, or dropping figures, it uses LLM vision to "see" each page and reconstruct the full document as agent-friendly Markdown.

No OCR. No layout heuristics. The LLM is both the reader and the formatter.

Key Features

• Vision-Based Extraction — Renders pages as images, uses LLM vision to read them. No OCR dependency. Works on scanned documents too.
• LaTeX Equation Precision — Distinguishes bold vectors (\mathbf{x}) from scalars (x), validates bracket nesting, preserves equation numbering.
• Smart Table Extraction — Handles multi-level headers, merged cells, preserves bold fidelity from the original PDF.
• Float-Aware Placement — Figures and tables placed after their first textual reference, not at their visual position on the page.
• Semantic Heading Hierarchy — Two-pass approach determines heading levels by document structure, not just font size.
• Cross-Page Continuity — Catches split paragraphs and dropped sentences at page boundaries.
• Metadata Extraction — Authors, affiliations, arXiv IDs structured in a clean blockquote header.

Installation

For Claude Code Users

git clone https://github.com/yuhaoliu7456/pdf2md-skill.git ~/.claude/skills/pdf2md-skill

Or copy manually:

mkdir -p ~/.claude/skills/pdf2md-skill
cp SKILL.md ~/.claude/skills/pdf2md-skill/

For Cursor Users

git clone https://github.com/yuhaoliu7456/pdf2md-skill.git .cursor/skills/pdf2md-skill

Then invoke by asking your agent to convert a PDF to markdown.

Usage

> "Convert this paper to markdown: ~/papers/attention-is-all-you-need.pdf"

> "Parse this PDF and extract all the equations: /path/to/document.pdf"

The skill will:

1. Render each