SkillHub

doc2slides

v2.7.2

Convert PDF, Word, and Markdown documents into professional PowerPoint slides with 18+ layout types and smart layout matching. Use when: user wants to create slides from a document or convert content to PPT.

Sourced from ClawHub, Authored by lifei68801

Installation

Please help me install the skill `doc2slides` from SkillHub official store. npx skills add lifei68801/doc2slides

Doc2Slides

Convert documents into professional PowerPoint slides with smart layout matching.

What It Does

  • 18+ Layouts: Dashboard, Timeline, Flow Chart, Pyramid, Comparison, Matrix, etc.
  • Smart Matching: Picks the best layout per section (template-based, works offline)
  • Optional LLM Boost: Enable AI-powered layout analysis via environment variables
  • High Quality: 3x resolution with charts and data visualizations
  • Batch Mode: Handles long documents (10+ pages) via split generation

Environment Variables (All Optional)

Template-based layout works without any remote service. These enable LLM-enhanced analysis:

Variable Purpose
LLM_MODEL Override default model selection
OPENAI_API_KEY OpenAI provider
OPENAI_BASE_URL Custom OpenAI-compatible endpoint
ZHIPU_API_KEY Zhipu GLM provider
DEEPSEEK_API_KEY DeepSeek provider
OLLAMA_BASE_URL Local Ollama endpoint (default: localhost:11434)

Provider credentials can also come from ~/.openclaw/agents/main/agent/models.json (OpenClaw's model configuration).

Install

# Core dependencies (required)
pip3 install python-pptx playwright
playwright install chromium

# Optional — for LLM-enhanced layout intelligence
pip3 install openai requests

Quick Start

cd ~/.openclaw/workspace/skills/doc2slides/scripts
python3 workflow.py --input /path/to/document.pdf --output slides.pptx

Pre-Flight Checklist (MANDATORY)

Before running any command, the agent MUST ask the user for preferences. Do NOT start generation immediately after receiving a file.

Ask the user: "有什么特殊要求吗？比如风格、页数、重点内容、配色偏好等。直接说需求就行，也可以说'按默认来'。"

The user can provide any free-form instruction — it is passed to both the content analysis LLM and the HTML generation LLM as highest-priority override.

Examples of valid instructions: - "商务风格，重点突出财务数据，控制在10页以内" - "学术风格，每页要有数据引用，15页左右" - "科技感，暗色背景，每页必须有图表" - "简约风，只保留核心结论，不要太多文字"

The instruction is applied via --instruction and takes precedence over all default rules in both analysis and rendering stages.

If the user says "按默认来" / "直接生成" / "没有要求", proceed with defaults (no --instruction flag).

How It Works

  1. Read — Load document content (PDF, DOCX, Markdown)
  2. Analyze — Identify structure and key points
  3. Match — Select layout for each section (template fallback or LLM-assisted)
  4. Build — Create styled slide images
  5. Export — Assemble into PPTX file

Tips

  • Documents with clear headings and bullet points work best
  • Shorter documents (1-20 pages) produce optimal results
  • All source code is included in the skill bundle for review

License

MIT-0