About this skill

Problem

LLM workflows often start with messy inputs: PDF, DOCX, PPTX, XLSX, scanned images, audio, web pages, and YouTube transcripts. Manual cleanup can break structure, while raw binary or HTML input makes token cost and model comprehension less predictable.

How It Works

MarkItDown转换 wraps Microsoft's MarkItDown to convert common documents into structured Markdown for review, retrieval, analysis, and regeneration:

Office files: PDF, DOCX, PPTX, XLSX, with tables and key formatting preserved
Images and scans: JPEG, PNG, GIF, WebP, with EXIF metadata and OCR support
Media and data: WAV, MP3 transcription, plus HTML, CSV, JSON, XML, EPUB, and ZIP
Optional AI: image captions via OpenRouter and Azure Document Intelligence for complex PDFs

Boundaries

OCR, speech transcription, and AI image descriptions require extra dependencies, APIs, or compute. Large files and batch jobs need pagination, concurrency limits, and error handling. This skill standardizes documents to Markdown; it does not perform final research or report writing.

Screenshots

Use Cases

Convert client DOCX, PDF, and PPTX files into unified Markdown for LLM-based clause review.
Extract tables and OCR text from spreadsheets and images, then format them for analytical models.
Turn YouTube links and WAV audio into transcripts, then archive them as Markdown knowledge base entries.
Use AI image descriptions to enrich PPTX slide notes and output Markdown for presentation scripting.

Best For

Document engineers: need to normalize office files into LLM-parseable Markdown.
Data analysts: need to extract tables and OCR text from spreadsheets and scans for analysis.
LLM application engineers: need clean, token-efficient inputs for RAG or workflow pipelines.
Research assistants: need to batch-convert PDFs, EPUBs, and YouTube transcripts into review drafts.