dsh-paper-tutor

If your native language is Chinese and every English paper feels like decoding a foreign document—jargon, dense sentences, chained derivations—dsh-paper-tutor was built for you. It is not a PDF viewer; it is an agent preset that, once selected in the session picker, wraps every reading session in a Chinese-language explainer persona that walks you from having no idea what the text says to being able to present it in a group meeting.

The preset ships four tools that work together: paper_scan inventories the PDFs, LaTeX sources, figures, and companion code in your workspace; pdf_extract pulls full text from PDFs using a pure-JS engine that preserves two-column order and layout, and auto-switches to poppler pdftotext when available for better quality; tex_structure maps the LaTeX skeleton—sections, equations, citations, and input dependencies; paper_figure extracts embedded figures, heuristically matches their captions, and optionally calls an OpenAI-compatible vision endpoint to interpret chart content, gracefully degrading to a file path when no endpoint is configured. The default cadence is explain as if from zero, with four depth levels—skim, close-read, source-trace against code, and critical analysis—and the session ends with a paper card, a glossary, and a paper-to-code mapping table. Zero npm runtime dependencies; cross-platform out of the box.

It is aimed at Chinese-native readers tackling English academic literature for the first time or regularly—graduate students encountering field-leading papers, undergraduates writing literature reviews, engineers self-studying a new direction. Put the paper PDF and its companion code in the same workspace and the source-trace layer will cross-reference every claim with the actual implementation. The plugin works fully standalone; adding dsh-tool-vision later makes figure interpretation plug-and-play.