slide-generation（幻灯片生成）

Slide Generation Skill

Generate Marp slides from source content following the AM Blue Course theme and workflow rules.

Quick Start

1. Identify sources: Read all source .md files (or all .md files in a folder)
2. Read references: Load references/rules.md and references/marp-guide.md before generating
3. Design structure: Plan slide sections from source content
4. Generate: Output .md file with Marp format

Input Modes

Single File

User: 生成关于"记忆与检索"的幻灯片，基于第八章.md
→ Read 第八章.md, generate slides

Multiple Files / Folder

User: 把这个文件夹里的内容整合成一组教学幻灯片
→ Read all .md files in the folder, merge by topic/chapter, generate unified slides

When multiple sources are provided:

• Merge content by topic or chapter sequence
• Avoid duplicate content across sources
• If sources conflict, prefer the more recent or comprehensive one

Output Target

When user specifies an output path (e.g. chapter8/memory-slide.md):

1. Create the output folder (e.g. chapter8/)
2. Copy all used images to chapter8/images/
3. Write the slide markdown to chapter8/memory-slide.md

Image path mapping:

• Source path: ../images/xxx.png → Output path: ./images/xxx.png
• Copy source images to the output images/ folder alongside the slide file

Marp YAML Header (Required)

Always use this exact header:

---
marp: true
size: 16:9
theme: am_blue_course
paginate: true
headingDivider: [2,3]
footer: '*敬业乐群*'
---

Required Slide Structure

1. Cover Page (First Slide)

<!-- _class: cover_c -->
<!-- _paginate: "" -->
<!-- _footer: 敬业乐群 -->
<!-- _header: ![](images/logo.png) -->

###### [Subtitle]

# [Main Title]

@PresenterName
<Email>

Subtitle format: use ###### (level 6 heading) Main title: use # (level 1 heading)

2. Table of Contents (Second Slide)

## 本节内容

<!-- _class: toc_a -->
<!-- _header: "CONTENTS" -->
<!-- _footer: 敬业乐群 -->
<!-- _paginate: "" -->

- [Topic 1](#3)
- [Topic 2](#N)
- ...

3. Content Slides

Use ## (level 2 heading) for main section titles. The headingDivider: [2,3] setting automatically splits pages at ## and ### headings.

For content-heavy pages, apply text-size classes:

• <!-- _class: smalltext --> — 0.9x font size
• <!-- _class: tinytext --> — 0.8x font size

For narrow/wide images, avoid left-right layouts; use full-width instead.

Core Rules

Must Follow

• ✅ Content must come ONLY from source files — do not add external knowledge
• ✅ Keep content rich but within source bounds
• ✅ Preserve original images from source; copy to output images/ folder
• ✅ Use logo.png as header image (path: images/logo.png in output)
• ✅ Presenter name and email as specified by user
• ✅ Use <!-- *class: smalltext* --> or <!-- *class: tinytext* --> for dense content
• ✅ Inline math: wrap in $...$ (e.g. $f(x) = y$)
• ✅ Convert code to pseudocode if it aids comprehension
• ✅ Section numbering: use 2.2 not 10.2.2; omit chapter prefix

Must NOT

• ❌ Use classes not defined in theme files
• ❌ Use transition slides (<!-- _class: trans -->)
• ❌ Use --- for manual page breaks (headingDivider already handles this)
• ❌ Title encoding like 10.2.2 — use 2.2 instead
• ❌ Generate images — only use images from source content

Layout Decision Rules

Based on image/table shape:

Image scaling: use ![#c w:700](path) or ![#c h:300](path) to adjust size dynamically.

Column Layout Examples

Two columns (五五分):

<!-- _class: cols-2 -->
<div class=ldiv>

Left column content
</div>

<div class=rdiv>

Right column content
</div>

Two columns (六四分, image on left):

<!-- _class: cols-2-64 -->
<div class=limg>

![#c](image.png)
</div>

<div class=rdiv>

Right column text
</div>

fig-top layout (image on top):

<!-- _class: fig-top -->
<div class=fig-container>

![width:850px](image.png)
</div>

<div class=text-container>

Content text
</div>

Available Classes

Reference: assets/example_file/AwesomeMarp_blue.md and assets/example_file/slide.md

Page layout: cols-2, cols-2-64, cols-2-73, cols-2-46, cols-2-37, cols-3, rows-2, pin-3

Text size: tinytext, smalltext, largetext, hugetext

Special: toc_a, cover_c, fig-top, fglass, caption, fixedtitleA, fixedtitleB

Callouts: bq-purple, bq-blue, bq-green, bq-red, bq-black

Image Handling

1. Scan source for ![]() image syntax
2. Copy each image file to output images/ folder
3. In slide markdown, reference images relative to output location: ![](images/filename.png)
4. Map original paths (e.g. ../images/8-figures/8-1.png) to new paths (images/8-1.png)

For logos and theme images, use: images/logo.png (copied from assets/images/logo.png)

Multi-Source Synthesis

When synthesizing from multiple sources:

1. Read all sources first — get full picture of content
2. Identify overlap — note repeated topics across files
3. Deduplicate — consolidate overlapping content
4. Merge by structure:

- If sources are chapters of the same book → follow chapter order - If sources are different perspectives → create unified section structure - If sources have conflicting info → prefer user's stated preference or most recent

1. Generate one coherent slide deck — unified structure, consistent style

Quality Checklist

Before finishing, verify:

• [ ] YAML header matches required format exactly
• [ ] Cover page has all: main title, subtitle, presenter name, email, logo, footer, no page number
• [ ] TOC page links to correct slide numbers
• [ ] All images from source are included and copied to images/ folder
• [ ] No Marp classes used beyond what's in theme files
• [ ] No --- page breaks (headingDivider handles it)
• [ ] Inline math uses $...$ syntax
• [ ] No title encodings like 10.2.2 — use 2.2
• [ ] Content stays within source material bounds
• [ ] Logo path is images/logo.png