Paper Summary Assistant

Problem Solved

Paper reading often wastes time before deciding whether an article is worth a close read. This skill takes a paper PDF, arXiv/aclanthology URL, DOI, title, or pasted text and compresses it into a concise Chinese summary that preserves the problem definition, method innovation, experimental setup, and core conclusions instead of giving a vague overview.

How It Works

The workflow follows fetch content → analyze structure → reconstruct summary:
- Input parsing: URLs use WebFetch to get the abstract and PDF link; local PDFs use Read; DOIs or titles use WebSearch; pasted text is processed directly.
- Structure extraction: It identifies title, authors, venue, problem, method, datasets, metrics, baselines, quantitative results, and contributions.
- Summary generation: It outputs a unified summary at 10%-20% of the original length, presents key data in tables, and marks missing sections as “not provided.”
- Quality checks: It stays objective, does not fabricate results, and keeps terms such as Transformer and BERT in their original form.

Boundaries

It fits literature screening, group-meeting outlines, and cross-field quick reading, but it should not replace close reading. Scanned PDFs without text layers, paywalled papers, and very long papers may require OCR, a local PDF, or section focus; the output is a reading aid.