ppt-from-template（模板中的 ppt）

PPT from Template

Generate presentations by extracting style from a reference template, then building fresh slides with PptxGenJS.

Core principle: Never modify existing PPT XML. Look at the style, then draw from scratch.

Template Management

Template Directory

Default template directory: {workspaceDir}/template/

On every invocation:

1. Scan {workspaceDir}/template/ for *.pptx files.
2. If one template found → use it automatically.
3. If multiple templates found → list them and ask user to choose.
4. If none found → tell user to upload a .pptx file or provide a path.

User Upload

Accept .pptx files uploaded by user. Before processing:

• Size check: reject files > 50 MB with message: "模板文件超过 50MB 限制，请压缩后重试"
• Save to {workspaceDir}/template/ for reuse.

Workflow

Phase 1: Extract Style

# If .pptx, convert to PDF for visual analysis:
python3 {baseDir}/scripts/pptx_to_pdf.py template.pptx /tmp/ppt_style/template.pdf

# Extract page images:
bash {baseDir}/scripts/extract_pages.sh /tmp/ppt_style/template.pdf /tmp/ppt_style/ 150

# Precision extraction from PPTX XML:
python3 {baseDir}/scripts/extract_style.py template.pptx -o /tmp/ppt_style/style_raw.yaml

Read style_raw.yaml for exact data (hex colors, font names, font sizes in pt, positions in inches, fill types, line styles). Then read 3-5 page images for semantic understanding (element roles, layout classification).

Extract two levels:

Element types: text, image, video, shape, line, numbered_list, step_list, tag_group, chart, table.

For images/videos: set placeholder: true with description for user replacement.

Write results to style.yaml. Schema: references/style-schema.md.

Phase 2: Generate PPT

Use the pptx skill (PptxGenJS) to create slides:

1. Read style.yaml for visual parameters.
2. Combine with user content (topic, page count, outline).
3. Generate PptxGenJS JavaScript applying extracted style.
4. Run JS to produce .pptx.
5. QA: convert to images, verify style fidelity.

Large Deck Strategy (>15 pages)

PptxGenJS generation is fast (seconds), but writing the JS code for many slides can hit context/time limits. Mitigate:

Slide factory pattern for large decks:

// Define layout factories from style.yaml
function makeCover(pres, content) { /* ... */ }
function makeContent(pres, content) { /* ... */ }
function makeSection(pres, content) { /* ... */ }

// Content array — easy to extend
const slides = [
  { layout: "cover", title: "...", subtitle: "..." },
  { layout: "content", title: "...", items: [...] },
  // ... 50+ entries
];

slides.forEach(s => layoutFactories[s.layout](pres, s));

This keeps code size O(layouts) not O(pages).

Output Size Control

Target output < 20 MB. PptxGenJS output is typically small (100-500 KB for text-only decks). Large files come from embedded images. If output approaches 20 MB:

1. Reduce image resolution/quality in PptxGenJS options.
2. Use placeholders instead of embedded images.
3. Split into multiple files if unavoidable.
4. Warn user: "PPT 接近 20MB 限制，已使用图片占位符，请手动替换"

Placeholder Convention

• Image: Dark rectangle + dashed gray border + 🖼️ + label + suggestion
• Video: Dark rectangle + dashed red border + ▶️ + label + suggestion

Key Rules

• Never modify existing PPT XML — always generate from scratch.
• style.yaml is reusable — once extracted, generate unlimited PPTs in the same style.
• Use shrinkText: true in PptxGenJS to auto-fit long text.
• Match slide dimensions exactly from style_raw.yaml (not hardcoded 16:9).

File Convention

{workspaceDir}/
├── template/              ← .pptx templates (auto-discovered)
│   └── *.pptx             ← max 50 MB each
├── output/                ← generated PPTs
│   └── *.pptx             ← max 20 MB each
/tmp/ppt_style/            ← working directory (ephemeral)
├── template.pdf
├── page-*.jpg
├── style_raw.yaml         ← from extract_style.py
└── style.yaml             ← merged (exact + semantic)

Troubleshooting