literature-pdf-abstract

Extract traceable metadata and abstract-level notes from local PDF first pages or covers without claiming full-text review.

SKILL.md

GitHub

PDF First Pages and Abstracts

Read shared contract and record schema. This module handles PDF evidence, not Zotero metadata screening or WoS citation analysis.

  1. Run ingest.py --input INPUT --output OUTPUT --backend pdfium. Check actual page counts and text quality.
  2. Find the article page rather than assuming PDF page 1; publisher/ResearchGate covers may precede it. Render ambiguous pages with render_pages.py; use OCR only when needed and compare author affiliations visually.
  3. Read title, author line, affiliations, abstract and original keywords. Write curated*.json with source_sha256, source_page and per-field evidence. Themes are model-inferred; original keywords stay separate. Do not record editors or funding bodies as affiliations.
  4. Run build_catalog.py with --scope first_page or abstract, then verify_catalog.py. Empty institution lists are valid when the provided page lacks them.
  5. Deliver structured JSON and per-paper notes. Route requested XLSX/CSV to literature-export and figures to literature-figures. Full-text requests belong to literature-deep-read.