scholar-slides

Converts research papers into fidelity-first academic slide decks with vector equations and editable PPTX.

827|98|Updated Jan 13, 2026
One-click install
npx skills add https://github.com/luwill/research-skills --skill scholar-slides
Or copy as Structured Prompt for Agentβ–Ό
Please help me install this Agent Skill.
Skill: scholar-slides
Source: https://github.com/luwill/research-skills/tree/main/scholar-slides
Command: npx skills add https://github.com/luwill/research-skills --skill scholar-slides

SYSTEM DOCUMENTATION & REQUIREMENTS

πŸ’‘ This Skill requires pymupdf, pillow, requests, matplotlib, numpy, python-pptx, pytest, katex, playwright, pptxgenjs, reveal.js, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

Generic AI slide tools rasterize equations, garble numbers, and fabricate citations, making them unusable for academic talks. This Skill turns a research paper, arXiv/DOI link, or topic into a slide deck where every equation, table, figure, and citation stays true text/vector and traceable to the source.

Core Features & Use Cases

  • Fidelity-first pipeline: A 7-stage workflow (ingest β†’ digest β†’ outline β†’ per-slide spec β†’ render β†’ QA β†’ export) with 3 human checkpoints and an always-on integrity gate that flags missing or unverified content instead of inventing it.
  • Real academic content: KaTeX vector equations, real HTML/OOXML tables with exact source numbers, bbox-cropped paper figures with provenance citations, and Zotero/Crossref/arXiv-resolved references.
  • Multiple outputs and registers: Generates an interactive reveal.js deck, one-page-per-slide vector PDF, editable PPTX, bilingual (EN/δΈ­ζ–‡) speaker notes with timing estimates, and data-bound matplotlib charts; supports journal-club and conference themes.
  • Use Case: You need to present a paper at your lab meeting (η»„δΌš). Point the Skill at the PDF or arXiv ID, confirm the digest and outline, and receive a projector-ready deck whose BLEU scores, equations, and figures match the paper exactly.

Quick Start

Ask the assistant to make slides from a paper PDF or arXiv link, for example: turn arXiv:1706.03762 into a 15-minute journal-club deck.

Frequently Asked Questions about scholar-slides

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make slides from a research paper PDF?β–Ό

Run scripts/prepare_source.py with the PDF path or arXiv ID to build a digest bundle, confirm the digest and outline at the checkpoints, then build_deck.mjs renders the reveal.js deck. Export to vector PDF and editable PPTX with render_deck.mjs and export_pptx.mjs.

How are equations and tables handled in generated slides?β–Ό

Equations render as KaTeX vector/MathML text, never rasterized images, and tables are rebuilt as real HTML or OOXML tables with the exact source numbers. Figures are bbox-cropped from the paper PDF with provenance citations.

Can I export the deck to an editable PowerPoint file?β–Ό

Yes, export_pptx.mjs produces an editable PPTX with native text, real OOXML tables, and native speaker notes. Figures and equations become images, and verify_pptx_parity.py checks that all spec content survives natively.

Does it support Chinese or bilingual academic presentations?β–Ό

Yes, decks are bilingual with English as default and full CJK support including Chinese punctuation and mixed CN/EN math slides. On Linux you need a Noto CJK font installed; macOS and Windows already include CJK fonts.

What are the limitations of PDF figure extraction?β–Ό

Figure localization runs about 95-100% on single-column and arXiv-style papers but drops to roughly 75% on dense two-column IEEE layouts. Low-confidence crops are flagged for confirmation at checkpoint 1, and scanned PDFs require OCR which is not built in.

When should I not use this for slide generation?β–Ό

Avoid it for marketing or pitch decks with no scholarly source, and for stylized image-based slides where look matters more than editable precision. It also cannot write the paper itself or process scanned PDFs without external OCR.