nature-paper-card

nature-paper-card builds a source-grounded, deep-reading "Paper Card" for a single scientific paper — from a PDF, DOI, arXiv page, publisher article, or pasted text. Rather than a translated abstract or generic summary, it produces a fixed 16-section analysis covering bibliographic position, research question, background path, core pain points, key insight, method and module logic, essential formulas, an experiment-to-claim evidence chain, conclusion-boundary auditing, author-stated limitations, critical analysis, learned knowledge, knowledge connections, and testable research ideas.

It operates as a router with strict grounding discipline. It loads a manifest and core layer, establishes the exact source boundary (full paper, unreliable layout, abstract-only, or an existing nature-reader artifact), and classifies the paper into one primary type (methods, discovery, resource, clinical, materials, or review) plus at most one secondary lens. For PDF or source-map input it mandates running the bundled scripts/prepare_paper.py to extract and validate content, then chooses one locator mode — page-grounded, structure-grounded, or source-limited — so citations never overstate what was actually seen; unlocated content is cited structurally rather than falsely attributed to page 1. Schema files enforce evidence tables that separate measured effects from expected ones and author-acknowledged limitations from agent analysis, marking anything unsupported as not assessable and labeling agent-generated ideas as candidates.

Target users are researchers, graduate students, and reviewers who want a rigorous, evidence-first reading aid rather than a promotional summary. The bundled script only extracts and analyzes supplied papers, the skill forbids writing ad-hoc replacement scripts during a run, and there is no network exfiltration or destructive behavior, making it a safe, read-only analysis tool.