What problem does it solve? Reading a scientific paper deeply enough to reconstruct its argument, trace every conclusion back to figures, tables, and experiments, and separate what the authors proved from what they merely claimed is slow and error-prone. This Skill turns one paper (PDF, DOI, arXiv page, publisher article, pasted text, or a nature-reader source map) into a fixed Sections 01-16 Paper Card with explicit evidence pointers instead of a shallow summary. ## Core Features & Use Cases - Evidence-grounded deep reading: Builds an evidence inventory and claim-evidence matrix before drafting, then produces the fixed 16-section card covering research question, method logic, essential formulas, experiment-to-claim evidence chain, conclusion boundaries, author-stated limitations, critical analysis, and testable research ideas. - Three locator modes with deterministic tooling: Runs the bundled prepare_paper.py to normalize PDFs or source maps into source_bundle.json, selects page-grounded, structure-grounded, or source-limited mode based on extraction reliability, and never fabricates page numbers. - Automated groundedness audit: Runs the bundled audit_paper_card.py to verify section structure, source pointers, figure/table/equation coverage, and scope compliance, blocking delivery on audit errors. - Use Case: A researcher downloads an arXiv PDF and asks for a Chinese Paper Card focused on method modules and conclusion boundaries; the Skill prepares the source bundle, classifies the paper type (e.g., methods with a secondary resource lens), writes paper-card.md with PDF page pointers, and delivers an audit-report.json confirming traceability. ## Quick Start Use nature-paper-card to deep-read this paper PDF and generate a complete Paper Card with the experiment-to-claim evidence chain and testable follow-up research ideas.
