LiteParse

Rule

Use pdftotext -layout first for text PDFs. Use lit only when a PDF is scanned or image-only and needs OCR, and then use lit parse --format text, not markdown: markdown splits labels from values and table rows.

lit screenshot and lit parse --format json (bounding boxes, page coordinates) stay fine for any PDF when you need rendered pages or positions.

Route DOCX to documents:documents, XLSX/CSV to spreadsheets:Spreadsheets, PPTX to presentations:Presentations, and PDF authoring/editing or fillable forms to pdf:pdf. Use Browser or a structured HTML parser for web and HTML content.

Commands

Write output to a temp file, then read the file.

Text PDF (pdftotext comes from poppler):

pdftotext -layout "input.pdf" "input.extracted.txt"

Scanned or image-only PDF, screenshots, and coordinates:

lit parse "input.pdf" --format text -o "input.extracted.txt" --quiet
lit parse "input.pdf" --format json -o "input.extracted.json" --quiet
lit screenshot "input.pdf" -o "screenshots" --target-pages 1 --quiet

A text PDF is one where pdftotext returns the page's words; if it returns nothing or near nothing, treat the PDF as scanned.

Useful options:

Helper Script

For repeatable extraction, run:

python "$env:USERPROFILE\.codex\skills\liteparse\scripts\parse_document.py" "input.pdf"

The script defaults to --format text and writes an adjacent .liteparse.txt file unless --output is provided. --format json and --format markdown stay selectable.

Fallbacks

• If lit is missing, install with uv tool install liteparse.
• Do not use LiteParse as a generic Office converter; invoke the matching bundled artifact skill.
• If visual layout matters, generate screenshots and inspect rendered pages before trusting text order.