MinerU Document to Markdown

Problem

When documents are PDF, DOC/DOCX, PPT/PPTX, images, or HTML, copied text often loses heading hierarchy, table boundaries, and reading order. That makes it harder to edit, summarize, or pass to an LLM. This skill converts documents to Markdown and addresses unstable conversion results, scattered ad-hoc requests, and the need for one workflow across tokenized and token-free modes.

How It Works

It uses two MinerU parsing paths:
- Precise mode: triggered by MINERU_API_TOKEN or --mode precise. Local files are uploaded via batch URLs and polled; the skill downloads full_zip_url and extracts full.md. Remote URLs are submitted to batch extraction and polled for results.
- Lightweight Agent mode: triggered without a token or with --mode agent. Local and remote inputs are submitted through the Agent API, polled, and the final markdown_url is downloaded.

Output defaults to outputs/mineru/, and logs go to logs/mineru_markdown.log. full.md is the main artifact; the .zip is kept for debugging or reuse. After conversion, continue with the Markdown instead of requesting another conversion.

Boundaries

Use it for document-to-Markdown conversion, not for xlsx/csv tables or when pasted text only needs summarization. Do not mix local files and remote URLs in one command. Large files are more constrained in Agent mode; on failure, inspect MinerU errors first and consider splitting the document.