Codex PPT

Codex PPT is a slide-deck Skill open-sourced by GitHub user ningzimu. The README says it is built mainly for Codex and also works in Claude Code, OpenClaw, Hermes Agent and other agents that read SKILL.md. Give it an article, report, paper, course notes, an outline, or a PDF or Word file. It first writes an outline.md and asks you to confirm the slide count and key points, offers two or three visual styles, tells you which image backend it plans to use and waits for your OK, then makes a single sample slide; only after you approve does it produce the full deck. Every slide is one complete 16:9 image, and a local Python script assembles them into a .pptx, along with a speech.md script that goes into each slide's speaker notes. There are 12 built-in style references, including clean professional, research defense, courseware and hand-drawn technical explainer. You can also upload screenshots, a PDF or a deck you like and have it match that style, save a style you're happy with to a personal library, and require a given slide to include an original paper figure, results chart or screenshot. What sets it apart is the approval gate at every stage: outline, style, image backend and sample slide all need your sign-off before it moves on. SKILL.md also requires every slide to come from the chosen image backend, and treats pages faked with local drawing, web screenshots or manual overlays as failures.

Who it's for:people who want to turn written material into a visually consistent deck and are fine with slides that are full-page images, such as creators of tech talks, course slides, thesis defenses and work reports. You'll need to be comfortable with an AI coding tool such as Codex or Claude Code; readers who never use these tools will find it hard going.

Before you start: you can ask your agent to install it from the README, or use npx skills add for Codex, Claude Code or Hermes Agent; OpenClaw users can install it from ClawHub. You need python3. The output is full-page images, so the text, charts and shapes on a slide can't be edited individually in PowerPoint, and SKILL.md says not to use it when every element must stay editable. The README points to the author's separate repo, image-to-editable-ppt-skill, for trying to convert image slides into an editable deck; that is a different project and AI Inspiration World hasn't reviewed it. In Codex, the README says you usually don't need an API key when the built-in image tool is available, but the actual model and resolution depend on your environment, and a subscription doesn't guarantee any particular model. Other agents usually need a third-party image API, either an OpenAI-compatible endpoint or AtlasCloud, which sponsors the project. In that case your prompts, which contain your slide content, and any reference images are sent to that service and billed per use at the provider's prices, so check before sending client material or unpublished papers. The API key is stored locally in ~/.codex-ppt-skill/.env; the Skill's runtime also creates a Python virtual environment and installs dependencies in that folder, where your personal style library lives too. The Skill instructions are in English, and the README comes in Chinese, English and Korean. The license is MIT.

Based on the project's public documentation. Not tested by AI Inspiration World.