ningzimu/codex-ppt-skill

What it solves

Creating visually appealing presentations often requires significant manual effort in layout and design. This project provides a structured workflow for AI agents to transform long-form content—such as articles, reports, papers, and course notes—into high-quality, image-based presentation slides that maintain visual consistency and professional aesthetics without requiring the user to manually design each page.

How it works

The tool operates as a "skill" for AI agents (like Codex, Claude Code, or OpenClaw). It follows a phased generation process:

1. Planning: The agent analyzes the input material to create an outline (outline.md) and determines the number of pages and key points.
2. Styling: The user selects from 12 built-in visual styles (e.g., McKinsey style, hand-drawn technical, scientific defense) or provides a custom reference image for the agent to replicate.
3. Sampling: A single sample slide is generated to confirm the layout and quality before proceeding.
4. Production: The agent generates each slide as a full-page 16:9 image using an image generation backend (such as gpt-image-2.5-flare or Codex's built-in tools).
5. Assembly: A local Python script (assemble_ppt.py) compiles these images into a final .pptx file and integrates a generated speech script (speech.md) into the slide notes.

Who it’s for

• Researchers and Students: For converting academic papers or thesis defenses into presentation decks.
• Technical Writers: For turning technical articles or product notes into sharing slides.
• Business Professionals: For creating commercial reports, product introductions, or survey summaries.
• Anyone needing a visually unified, image-based presentation quickly from existing text documents.

Highlights

• Multi-Agent Compatibility: Works across various environments supporting SKILL.md, including Codex, Claude Code, OpenClaw, and Hermes Agent.
• Style Library: Includes 12 preset professional styles and allows users to build a personal style library in ~/.codex-ppt-skill/references/ for future reuse.
• Custom Visuals: Supports inserting specific original images (e.g., architecture diagrams or experimental results) and generating the rest of the slide around them.
• Quality Control: Uses a phased approach (outline $\rightarrow$ style $\rightarrow$ sample $\rightarrow$ full generation) to reduce rework and ensure consistency.
• Automated Notes: Automatically generates a corresponding speech script and embeds it into the PowerPoint notes section.