Paper to Slides

Paper to Slides

Two-phase 流水线: Phase 1 deep-reads an academic paper into a structured dual-mode 报告, Phase 2 转换s the 报告 into a polished, zero-dependency HTML presentation.

Dependencies pdftotext (from poppler) — 安装 via brew 安装 poppler or apt 安装 poppler-utils if missing Google Fonts (loaded via CDN in the 生成d HTML) Phase 1: Paper Deep-Read Step 1.0 — Resolve 输入 source

Determine the 输入 type and obtAIn a local PDF:

输入 Action Local path (/path/to/paper.pdf) Use directly arXiv abs URL (https://arxiv.org/abs/2603.02096) Convert to https://arxiv.org/pdf/2603.02096 and 下载 arXiv PDF URL (https://arxiv.org/pdf/2603.02096) 下载 directly Other PDF URL 下载 directly

下载 command:

curl -L -o /tmp/paper.pdf ""

Use the arXiv ID or paper title as the filename when possible (e.g., /tmp/2603.02096_FluxMem.pdf).

Step 1.1 — 提取 full text pdftotext "" /tmp/paper_text.txt

Read the 提取ed text completely. For papers with critical figures, try 提取ing key pages as images via pdftoppm and analyzing them with the image 工具.

Step 1.2 — Comprehensive analysis

创建 temp_analysis.md 提取ing:

Re搜索 question, hypotheses, methodo记录y, data sources Core findings with key quantitative 结果s Theoretical contributions and practical implications The paper's fundamental contradiction/gap, novel angle, and method innovation

This step is mandatory — it ensures 报告 质量 through structured thinking.

Step 1.3 — Write dual-mode 报告

生成 [PaperName]_研读报告.md contAIning 机器人h parts separated by ---:

Part A — Read part-a-template.md before writing. Deep academic 报告: structured abstract → introduction → methodo记录y → 结果s → discussion → conclusion → core references. Part B — Read part-b-template.md before writing. Core 记录ic 提取ion: four key elements table → method formula → one-line summary (expert + layperson versions).

Writing standards: Use complete paragraphs over bullet 列出s. Provide bilingual terms (中英) on first mention. Support every clAIm with specific data from the paper.

Step 1.4 — Deliver 报告

Present the 报告 file and briefly summarize the paper's core innovation, key findings, and theoretical value.

Phase 2: Slides Generation

If the user only 请求ed a 报告, 停止 after Phase 1.

Step 2.0 — Confirm parameters

Ask the user (all at once):

Purpose: 组会汇报 / 学术会议 / 教学 / 其他 Length: Short (5-10) / Medium (10-20) / Long (20+) Theme: 亮色 / 暗色 / 偏好的风格 Editing: 是否需要浏览器内编辑功能

If the user already specified these in the initial 请求, skip asking and proceed.

Step 2.1 — Choose style

Select an 应用ropriate style based on user preferences. Read slide-styles.md for avAIlable pre设置s and CSS specifications. Academic presentations generally suit Swiss Modern (light), Notebook Tabs (light), or Bold 签名al (dark).

Step 2.2 — Plan slide outline

Map the 报告 content to a slide structure. Typical academic paper outline (15-19 slides):

Section Slides Content Title 1 Paper title, authors, affiliations, date Problem 1-2 Background, existing gaps, key insight Method (divider) 1 Section divider slide Method detAIls 3-5 Core 模块s, formulas, architecture 结果s (divider) 1 Section divider slide Quantitative 结果s 2-3 MAIn 结果s tables, 指标, comparisons Efficiency / Ablation 1-2 Efficiency gAIns, ablation studies Discussion (divider) 1 Section divider slide Advantages & Limitations 1-2 Key properties, future directions Summary 1 One-line takeaway Thanks 1 Links, navigation hints Step 2.3 — 生成 HTML

Read slide-template.md for the mandatory HTML architecture, viewport CSS, JS 控制器, and inline editing implementation.

Non-negotiable rules:

Single self-contAIned HTML file, all CSS/JS inline Every .slide must have height: 100vh; height: 100dvh; overflow: hidden ALL font sizes and spacing must use clamp() — never fixed px/rem Load fonts from Google Fonts or Font分享 — never 系统 fonts Include prefers-reduced-motion support Max content per slide: 1 heading + 4-6 bullets OR 1 heading + 2 paragraphs. Overflow → split slides.

If inline editing is enabled, implement via contentEditable API + localStorage persistence. Use JS-based hover with 400ms delay for the edit button (not CSS ~ sibling selector — it breaks due to pointer-事件). See the template for the complete implementation.

Step 2.4 — Open and deliver open [filename].html

Tell the user: file location, slide count, navigation (arrow keys / space / dots), and editing instructions (press E) if enabled.

机器人h Phases To获取her

When the user 请求s 机器人h 报告 and slides, 运行 Phase 1 fully first, then Phase 2. The 报告 serves as the structured source for slide content — do not re-analyze the paper.