Academic Paper Intensive Reader

Problem

Close reading a paper often goes beyond the abstract. The hard part is turning methods, experimental setup, ablation results, and limitations into a reviewable structure. Simply asking a model to summarize a PDF can introduce problems: invented numbers that are not in the source, formula explanations that drift away from the original meaning, external background knowledge mixed into the paper, and notes that are hard to reuse in reports or follow-up questions.

How It Works

The skill follows a fixed workflow for paper inputs:

• Input recognition: supports PDF, DOI/PMID, and pasted text; for PDFs it prioritizes the abstract, introduction, methods, experiments, discussion, and conclusion; for DOI/PMID it first retrieves metadata and the abstract.
• Structured close reading: outputs the research overview, scientific background and problem, core methods and innovations, experimental design and key results, conclusions and research value, limitations and critical analysis, and follow-up research directions.
• Grounding constraints: conclusions must follow the logic of the source; missing content is marked as “not mentioned in the original”; technical terms are kept in English when appropriate, with Chinese explanations on first use.
• Markdown output: standard Markdown is used so the notes can be imported into note tools, adapted for reports, or expanded with follow-up questions on sections, formulas, and experimental details.

Boundaries and Caveats

It is better suited for generating close-reading notes from an existing paper than for acting as a literature database, fact-checking system, or automated citation tool. If the input lacks methods or experimental data, the skill can only mark the gaps as unknown and cannot fabricate missing evidence. If the user asks to focus on methods, experimental design, or a report outline, it prioritizes that custom request before completing the full close-reading structure.