Extract text, metadata, and annotations from PDF files with provenance sidecars.
npx skills add https://github.com/ppavlidis/skillz --skill pdf-extract-ppavlidis
Please help me install this Agent Skill. Skill: pdf-extract Source: https://github.com/ppavlidis/skillz/tree/main/skills/pdf-extract Command: npx skills add https://github.com/ppavlidis/skillz --skill pdf-extract-ppavlidis
extract.py all paper.pdf produces a single JSON bundle with text, metadata, and annotations, plus a verifiable provenance record. ## Quick Start Ask the AI to extract all annotations and metadata from your PDF file, for example: "Extract the highlights and DOI from paper.pdf using the pdf-extract skill."High-intent search queries and answers about installing and using this skill.
Run `python scripts/extract.py annotations paper.pdf` with pymupdf installed. It captures Highlight, Underline, StrikeOut, Squiggly, Text, FreeText, and other annotation types, resolving the marked text from the page's word stream into a JSON file.
Run `python scripts/extract.py metadata paper.pdf`. The DOI is looked up in XMP metadata first, then the docinfo subject, then a regex search of the first page. Title falls back to the largest-font text block on page 1 if docinfo is empty.
pymupdf is the primary library here and is required for annotation extraction; pypdf is an optional fallback that handles text and metadata only. pymupdf is AGPL-licensed, while pypdf is MIT for fully permissive operation.
A null highlighted_text means the PDF has no text layer, which happens with scanned image-only documents. Run an OCR tool such as ocrmypdf first to add a text layer, then re-run the extraction.
Yes, pass the --pages flag to the text operation, accepting formats like "3", "2-5", or "1,3,5-7". Use --format text for plain output with page markers or the default JSON for per-page structured output.
Encrypted PDFs raise an error and must be decrypted first with qpdf. The year heuristic grabs the first four-digit number in 1900-2099 on page 1, which can catch grant numbers, and DOIs split across line breaks are not reassembled.
Monjyu1101Extract text from PDFs and scanned documents using pymupdf and marker-pdf.
PancrePal-xiaoyibaoRetrieve and verify biomedical literature metadata from the PubMed database.
ZK-TheoryReviews computational result files for reproducibility, seed recording, and provenance traceability.
juliuss1907Extract text from PDFs, scanned documents, and images using OCR.