document-converter_skill

Overview

This skill converts documents bidirectionally between Markdown, DOCX, and PDF formats with automatic tool selection and fallback handling. It supports batch processing, concurrent conversions, image extraction, and filename sanitization. Use it to extract structured Markdown from Word or PDF files or to generate high-fidelity DOCX/PDF from Markdown.

How this skill works

The skill inspects input files, detects conversion direction, and selects the optimal toolchain based on availability and configured mode (online with Gemini API or offline). It runs conversions with timeouts, retries using fallback tools when needed, extracts and embeds images, and validates outputs for basic structure and missing assets. Progress and quality metrics are logged for each run.

When to use it

• Extract structured Markdown from DOCX or PDF files for editing or analysis
• Generate DOCX or PDF deliverables from Markdown sources with high fidelity
• Batch-convert large document collections while preserving tables and images
• Convert scanned or layout-complex PDFs with configurable timeouts and OCR options
• Integrate conversion into automation or agent pipelines for document preprocessing

Best practices

• Enable the Gemini API key for best PDF→Markdown fidelity when available
• Use MarkItDown for DOCX→Markdown to preserve tables and Unicode reliably
• Increase PDF timeouts for large or image-heavy documents (default 300s)
• Run batch conversions on a machine with enough disk and concurrency limits set
• Review conversion.log for validation warnings and failed retries

Example use cases

• Convert a folder of research DOCX files to Markdown for note-taking and version control
• Extract text and code blocks from PDF reports into Markdown for analysis
• Produce client-ready DOCX or PDF from a Markdown manuscript with Pandoc and Typst
• Transform scanned PDFs using OCR-enabled fallbacks before downstream processing
• Automate preprocessing of documents for an NLP pipeline or knowledge base ingestion