PowerPoint Automation

AI-powered PPTX generation pipeline using Orchestrator-Workers pattern. Automatically extracts content, translates, applies templates, and performs quality reviews.

When to Use This Skill

Use this skill when the user wants to:

• Convert web articles or blog posts to PowerPoint presentations
• Translate English PPTX to Japanese (or other languages)
• Create presentations using custom templates
• Generate technical presentations with code blocks and diagrams
• Automate presentation creation with quality assurance

Quick Start

Prerequisites

Install dependencies:

# Python dependencies
pip install -r scripts/requirements.txt

# Node.js dependencies (for diagram generation)
npm install --prefix scripts

Basic Usage

From Web Article:

# Simply describe what you want with a URL
"Create a 15-slide presentation about Azure Copilot from this Zenn article:
https://zenn.dev/example/articles/pptx-guide"

# The system will automatically:
# 1. Extract content and images
# 2. Generate content.json
# 3. Create PPTX with template
# 4. Review quality

From Existing PPTX:

# Provide PPTX file path in your request
"Translate this English presentation to Japanese: input/presentation.pptx"

# System will:
# 1. Analyze structure
# 2. Extract content
# 3. Translate (if needed)
# 4. Apply new template

Architecture

This skill uses the Orchestrator-Workers pattern with 6 specialized agents. See Agents section for details.

Workflow Phases

TRIAGE → PLAN → PREPARE_TEMPLATE → EXTRACT → TRANSLATE → BUILD → REVIEW → DONE

Phase Details

diagnose_template.py

clean_template.py

extract_images.py

reconstruct_analyzer.py

create_from_template.py

Key Scripts

See

references/SCRIPTS.md

Most Common:

• create_from_template.py

  - Generate PPTX from content.json (⭐ Recommended)

• reconstruct_analyzer.py

  - Convert PPTX → content.json

• extract_images.py

  - Extract images from PPTX or web

• validate_content.py

  - Validate content.json schema

• validate_pptx.py

  - Detect text overflow, missing notes

Specialized:

• create_pptx.js

  - Generate diagrams with pptxgenjs

• merge_slides.py

  - Combine diagram slides into template

• summarize_content.py

  - Reduce slide count with AI

Intermediate Representation (IR)

All agents communicate via content.json (SSOT):

{
  "slides": [
    {
      "type": "title",
      "title": "Presentation Title",
      "subtitle": "Subtitle"
    },
    {
      "type": "content",
      "title": "Main Topic",
      "items": ["Point 1", "Point 2"],
      "image": {
        "path": "images/diagram.png",
        "position": "right",
        "width_percent": 45
      }
    }
  ]
}

See

references/schemas/content.schema.json

Design Principles

1. SSOT (Single Source of Truth): content.json is the canonical representation
2. SRP (Single Responsibility): Each agent/script has one clear purpose
3. Fail Fast: Errors detected early, automatic retry (max 3×)
4. Human in the Loop: User confirmation at PLAN phase
5. Dynamic Context: Adapt to any template size, not hardcoded

Templates

Two templates are included:

assets/base_template.pptx

assets/template.pptx

When to Use Each Template

• base_template.pptx

  (Recommended): Full-featured template with 11 standard English layouts. Use for most presentations including tech blogs, reports, and translations. Works best with

  analyze_template.py

  auto-detection.

• template.pptx

  : Legacy template with Japanese layout names. Use if you need specific Japanese layout naming or compatibility with older workflows.

To use custom templates:

python scripts/analyze_template.py path/to/custom.pptx

Error Handling

• Retry Policy: Max 3 attempts per phase
• Escalation: After 3 failures → human intervention
• Recovery:

  scripts/resume_workflow.py

  to resume from specific phase

Advanced Features

Diagram Generation (pptxgenjs)

// Create architecture diagrams with shapes and arrows
node scripts/create_pptx.js

Slide Merging

# Merge diagram slides into template
python scripts/merge_slides.py template.pptx diagrams.pptx output.pptx

Content Summarization

# Reduce slide count with AI summarization
python scripts/summarize_content.py content.json --target-slides 15

Agents

Detailed agent definitions in

references/agents/

orchestrator.agent.md

brainstormer.agent.md

localizer.agent.md

summarizer.agent.md

json-reviewer.agent.md

pptx-reviewer.agent.md

References

Detailed documentation in

references/

• SCRIPTS.md: Complete script reference and usage examples
• USE_CASES.md: Detailed workflows for common scenarios
• Agents:

  references/agents/*.agent.md

  - Agent definitions
• Instructions:

  references/instructions/*.instructions.md

  - Workflow rules
• Schemas:

  references/schemas/*.schema.json

  - JSON schemas
• AGENTS.md: Architecture overview and design principles

Common Use Cases

• Tech Blog → PPTX: Convert Zenn/Qiita articles to presentations (15-20 slides, ~5 min)
• English → Japanese: Translate presentations maintaining design (~15 min for 120 slides)
• Report Generation: Create business reports from notes + charts (~10 min)
• Technical Training: Convert API docs to training material with code examples (~8 min)
• Architecture Diagrams: Generate custom diagrams and merge into templates (~5 min)

See

references/USE_CASES.md

Limitations

• Language Support: Primarily English ↔ Japanese (extendable)
• Image Formats: PNG, JPG, JPEG (no SVG/GIF animation)
• Code Blocks: Best with pptxgenjs method
• Template Compatibility: Works with standard PowerPoint templates (16:9 recommended)

Troubleshooting

Common issues documented in repository's

docs/TROUBLESHOOTING.md

Quick checks:

# Validate content.json
python scripts/validate_content.py output_manifest/xxx_content.json

# Diagnose template
python scripts/diagnose_template.py templates/xxx.pptx

# Check PPTX quality
python scripts/validate_pptx.py output_ppt/xxx.pptx

License

This skill is licensed under CC BY-NC-SA 4.0. See LICENSE.txt for full terms.

Note: This skill orchestrates a complex multi-agent pipeline. For best results, follow the PLAN phase confirmation process and use the Orchestrator agent as the entry point.