paper-to-beamer

name	paper-to-beamer
description	Convert a research paper PDF into a Beamer presentation. Extracts tables, analyzes structure, generates .tex with NankaiBeamer theme, compiles, and fixes overflow.
argument-hint	[PDF filename in Papers/ directory] [optional: number of slides, e.g. 10]
trigger-phrases	["paper to beamer","论文转PPT","制作课堂展示","从论文生成PPT","paper to slides"]
allowed-tools	["Read","Write","Edit","Bash","Glob","Grep","Agent"]

Paper-to-Beamer Workflow

Convert a research paper PDF into a complete Beamer presentation using the NankaiBeamer theme.

Input

$ARGUMENTS = [PDF filename] [optional slide count]

Examples:

• /paper-to-beamer paper.pdf → ask user for slide count
• /paper-to-beamer paper.pdf 15 → generate exactly 15 slides
• /paper-to-beamer paper.pdf 8-12 → generate 8 to 12 slides

Steps

1. Locate the PDF

ls Papers/

Confirm the file exists. If not found, list available PDFs and ask the user to clarify.

2. Ask for Slide Count (if not provided)

If no slide count is given in $ARGUMENTS, ask the user:

您希望生成多少页幻灯片？

• 精简版（12-15页）：适合20-30分钟报告
• 标准版（23-25页）：适合45分钟报告
• 详细版（33-35页）：适合60分钟以上报告
• 自定义：请输入具体页数

Use the answer to determine the slide count target (TARGET_SLIDES).

2.5. Ask for Generation Mode

Always ask the user to choose generation mode:

请选择生成模式：

1. 全自动生成 (auto) — 我根据论文内容和最佳实践自动生成（推荐）
2. 个性化定制 (custom) — 通过多轮对话确定重点和风格

Store the user's choice as GEN_MODE (default: auto).

2.6. Ask for Table Extraction Mode

Always ask the user to choose a table extraction mode:

请选择表格处理方式：

1. 全自动截图 (auto) — 使用 pdfplumber + pdftoppm 自动提取表格图片（默认）
2. 手动截图 (manual) — 您自己截图，我会告诉您需要截哪些表格
3. 纯文本提取 (text) — 我用 pdfplumber 提取表格数据，用 LaTeX 重新排版（无图片）

Store the user's choice as TABLE_MODE (default: auto).

Slide count guidelines:

Target	Slides	Coverage
精简版	12-15	动机、数据、主要结果、结论
标准版	23-25	上述 + 文献、识别策略、稳健性、机制
详细版	33-35	上述 + 异质性、政策含义、附录

3. Extract Tables (mode-dependent)

If TABLE_MODE == "auto":

python scripts/extract_tables.py "Papers/[FILENAME]"
python scripts/render_pages.py "Papers/[FILENAME]" 200

If TABLE_MODE == "manual":

1. First, scan the PDF to identify tables:

import pdfplumber
with pdfplumber.open("Papers/[FILENAME]") as pdf:
    for i, page in enumerate(pdf.pages, 1):
        tables = page.find_tables()
        if tables:
            print(f"Page {i}: {len(tables)} table(s)")

2. Tell the user:

我检测到以下页面有表格：

• 第 X 页：Y 个表格
• 第 Z 页：W 个表格

请您手动截图并保存到 Figures/[BASENAME]_tables/ 文件夹，命名规范：

• table-1.png — 第一个表格
• table-2.png — 第二个表格
• ...

截图建议：使用高 DPI（200+），PNG 格式，裁剪掉多余空白。

完成后请回复"完成"或"done"。

3. Wait for user confirmation before proceeding.

If TABLE_MODE == "text":

Extract table data as text and store in JSON:

python scripts/extract_tables.py "Papers/[FILENAME]" --text-only

This will create Figures/[BASENAME]_tables/tables_text.json with raw table data.

4. Analyze Paper Structure

Read the PDF text to identify:

• Title, authors, journal/year
• Abstract (1-2 key sentences)
• Research question / hypothesis
• Data and sample
• Identification strategy
• Main results
• Robustness checks
• Conclusion / policy implications

import pdfplumber
with pdfplumber.open("Papers/file.pdf") as pdf:
    text = "\n".join(p.extract_text() or "" for p in pdf.pages[:5])
print(text[:3000])

4.5. Customization Planning (if GEN_MODE == "custom")

If user chose custom mode, conduct multi-round planning dialogue:

Round 1: Identify Focus Areas

Ask the user:

我已分析论文结构。请告诉我您希望在报告中：

1. 重点强调哪些方面？（可多选）
   • 理论贡献
   • 识别策略的创新性
   • 数据的独特性
   • 实证结果的稳健性
   • 政策含义
   • 机制分析
2. 可以略谈哪些部分？
3. 目标听众是谁？（学术会议/课堂教学/政策制定者）

Store responses as FOCUS_AREAS, SKIP_AREAS, AUDIENCE.

Round 2: Content Prioritization

Based on TARGET_SLIDES and user's focus, propose a slide allocation:

根据您的需求，我建议以下幻灯片分配：

• 动机与研究问题: X 页
• 文献定位: Y 页
• 数据: Z 页
• 识别策略: W 页
• 主要结果: V 页
• [用户强调的重点]: U 页
• 结论: T 页

是否需要调整？

Allow user to adjust allocation.

Round 3: Style Preferences

Ask:

展示风格偏好：

1. 表格展示：图片 / LaTeX重排 / 混合
2. 方程展示：完整数学公式 / 简化直觉解释 / 两者结合
3. 结果强调：用彩色框 / 用加粗 / 最小化格式
4. 是否需要动画效果？（不推荐学术报告）

Store as STYLE_PREFS.

Round 4: Final Outline Confirmation

Generate and present detailed outline:

最终大纲 (共 N 页):

1. 标题页 2-3. 研究问题与动机 (强调: [用户指定]) 4-5. [根据用户需求定制] ...

确认后我将开始生成 LaTeX 代码。

Wait for user confirmation before proceeding.

5. Plan the Slide Deck

Based on TARGET_SLIDES, select which sections to include:

Always include (core):

1. Title slide
2. Research question / motivation
3. Data & sample
4. Main results (1-3 slides depending on target)
5. Conclusion

Add for standard/detailed:

• Literature position
• Identification strategy
• Robustness checks

Add for detailed only:

• Mechanism / heterogeneity
• Policy implications
• Additional results

6. Generate Beamer .tex File (mode-dependent)

IMPORTANT: Before generating, read the best practices prompt:

cat .claude/prompts/beamer-generation-prompt.md

Apply all principles from the prompt file, especially:

• Use assertion-evidence approach (conclusion-based titles)
• Follow cognitive load principles (max 5 bullets per slide)
• Add visual hierarchy (colored boxes for key results)
• Provide economic interpretation, not just statistical significance

If GEN_MODE == "custom", strictly follow the user's:

• FOCUS_AREAS - allocate more slides and detail
• SKIP_AREAS - minimize or omit
• AUDIENCE - adjust technical depth
• STYLE_PREFS - apply formatting preferences

Create Slides/[BASENAME].tex using the NankaiBeamer theme:

\documentclass[aspectratio=169]{beamer}
\usepackage{xeCJK}
\setCJKmainfont{SimSun}
\setCJKsansfont{SimHei}
\setCJKmonofont{FangSong}
\usetheme{NankaiBeamer}

% Custom box environments
\usepackage{tcolorbox}
\tcbuselibrary{skins}
\newenvironment{resultbox}{\begin{tcolorbox}[colframe=nankai,colback=nankai!5]}{\end{tcolorbox}}
\newenvironment{methodbox}{\begin{tcolorbox}[colframe=orange!80!black,colback=orange!5]}{\end{tcolorbox}}
\newenvironment{highlightbox}{\begin{tcolorbox}[colframe=nankai!40,colback=nankai!8]}{\end{tcolorbox}}

\title{[Paper Title]}
\subtitle{[Journal, Year]}
\author{[Authors]}
\date{\today}

\begin{document}
\maketitle
% ... slides ...
\end{document}

Rules for table slides (mode-dependent):

If TABLE_MODE == "auto" or TABLE_MODE == "manual":

• Use \includegraphics[width=0.85\textwidth,height=0.55\textheight,keepaspectratio]{...} for table images
• Add 1-2 bullet points explaining the key finding

If TABLE_MODE == "text":

• Generate LaTeX tabular environment from extracted JSON data
• Use \small or \footnotesize for dense tables
• Add caption and label
• Example:

\begin{table}[h]
\centering
\small
\begin{tabular}{lcc}
\toprule
Variable & Coefficient & SE \\
\midrule
Treatment & 0.45*** & (0.12) \\
Control & -0.23* & (0.10) \\
\bottomrule
\end{tabular}
\caption{Main Results}
\end{table}

Rules for text slides:

• Max 5 bullet points per slide
• Use \small for dense content

7. Compile (automatically via /compile-latex)

Invoke /compile-latex [BASENAME] which handles:

• 3-pass XeLaTeX compilation
• Windows/Mac path separator detection
• Error reporting

8. Check and Fix Overflow (automatically)

After compilation, automatically invoke /beamer-overflow-fix [BASENAME].tex if overflow warnings exist.

9. Quality Scoring (automatically)

Run quality check:

python scripts/quality_score.py Slides/[BASENAME].tex --summary

Report the score. If score < 80, list critical issues and offer to fix.

10. Proofread (automatically if score < 90)

If quality score is below 90, automatically invoke /proofread [BASENAME].tex to check:

• Grammar and typos
• Academic writing quality
• Notation consistency

Apply fixes if user approves.

11. Present Results

Report:

• Number of slides generated (vs. target)
• Table extraction mode used: [auto/manual/text]
• Tables extracted and included
• Quality score: [N]/100 ([GATE])
• Any issues found and fixed
• Path to compiled PDF: Slides/[BASENAME].pdf

Quality Gates:

• 95-100: EXCELLENCE ⭐
• 90-94: PR-ready ✓
• 80-89: Commit-ready
• <80: Needs improvement (list issues)

Error Handling

Error	Action
pdfplumber not installed	pip install pdfplumber
pdf2image not installed	pip install pdf2image
\kaishu undefined	Add \providecommand{\kaishu}{\rmfamily} to theme file
NankaiBeamer.sty missing	Fall back to \usetheme{Madrid}
Compilation fails	Read first 50 lines of .log, fix the error, retry

Output

• Slides/[BASENAME].tex — Beamer source
• Slides/[BASENAME].pdf — Compiled presentation
• Figures/[BASENAME]_tables/ — Cropped table images