paper2slides

paper2slides — Agent-Driven Journal Club Presentations

Overview

This skill transforms a scientific paper PDF into a show-ready, curated Reveal.js presentation with speaker notes. It combines two layers:

1. Mechanical layer — the paper2slides Python library extracts figures, segments panels, and scaffolds a raw Reveal.js deck.
2. Intelligence layer — you, the agent, read the paper, understand the science, curate the slide deck, write titles/captions, and compose speaker notes.

The mechanical extraction alone produces a figure dump. The agent's job is to turn that into a coherent, PhD-level journal club talk.

---

Prerequisites

The paper2slides library must be installed. If the library source is available locally:

cd <path_to_paper2slides_library> && pip install -e .

Otherwise install from the packaged distribution:

pip install paper2slides

Required Python packages (auto-installed): pymupdf, pillow, numpy, jinja2, click

---

Workflow (5 Phases)

Phase 1: Extract & Segment (Mechanical)

Run the paper2slides CLI to extract figures and segment panels:

paper2slides "<path/to/paper.pdf>" -o "<output_dir>"

This produces:

• <output_dir>/assets/img/panels/fig1a.png, fig1b.png, ... — individual panel images
• <output_dir>/index.html — a raw, uncurated Reveal.js deck (figure dump)
• <output_dir>/assets/css/style.css — slate + accented theme with expanded font choices, high-contrast body text, per-aim color tokens, feature-family badge colors, and ready-made component classes for aim-flow / info-row-3 / taxonomy-grid / compare-grid / criteria-grid / badges / schematics

Verify: Check the terminal output for the number of figures and panels extracted. If zero figures found, the PDF may use vector graphics — consider alternatives like screenshots.

Phase 2: Read & Comprehend the Paper

Use PyMuPDF to extract and read the paper text programmatically:

import fitz
doc = fitz.open("<path/to/paper.pdf>")
for i in range(len(doc)):
    page = doc[i]
    text = page.get_text('text')
    print(f'=== PAGE {i+1} ===')
    print(text)
doc.close()