# codex-ppt-skill: image-based PowerPoint generation inside Codex and other SKILL.md agents

> A Python skill that turns articles, papers and notes into full-page image slides, then assembles them into a .pptx with speaker notes. Strong visual consistency, but the pages themselves are pictures, not editable shapes.

**ningzimu/codex-ppt-skill** — GPT-Image-2 PPT Generator Skill for Creating Image-Based PowerPoint Presentations in Codex and Other Skill-Compatible Agents

- Repository: https://github.com/ningzimu/codex-ppt-skill
- Website: https://ppt-skill.ningzimu.vip
- Stars: 6,290 · Forks: 309
- Language: Python
- License: MIT
- Published: 2026-09-22 · Updated: 2026-09-22 · Language: en
- Canonical page: https://hysenlabs.com/projects/ningzimu-codex-ppt-skill

## What codex-ppt-skill solves, and who it is actually for

Most presentation tools start from a template and ask you to fill in placeholders. This skill inverts that. It takes a source document (an article, a report, a paper, course notes, Markdown, an outline, even a PDF or Word file) and produces a deck where every slide is a single generated 16:9 image, then wraps those images into a .pptx with a local Python script.

The audience is narrow and fairly clear. It is for people who already run Codex, Claude Code, OpenClaw or Hermes Agent and want a deck with a strong, uniform visual identity rather than a corporate template. The README is explicit that Codex is the recommended host, where the built-in image generation and editing path is preferred. In the other agents, the documentation says you usually need to configure gpt-image-2.5-flare, a third-party image API, or an OpenAI-compatible image endpoint.

That constraint matters. If you are not already inside one of those agents, this is not a standalone CLI that you point at a file. The skill is the workflow; the agent is the runtime.

## The staged pipeline: outline, style, sample, then batch generation

The mechanism is deliberately gated. According to the README, the skill first confirms the outline, the page count, the visual style, the image backend and a sample slide before it generates the full set. That sample step is the interesting design choice: instead of committing to twenty images and discovering the style is wrong, you approve one page first.

Each deck becomes its own project directory. The README documents this layout: an origin_image/ folder holding only the final adopted pages as slide_01.png, slide_02.png and so on, an outline.md with the confirmed structure, a speech.md script, and the assembled {name}.pptx. The speech file is not decorative; the README states it is written into the notes area of each slide during assembly.

After the sample is approved, the skill supports multi-agent concurrency, with one sub-agent per page, plus self-checks on text legibility, style consistency and content completeness. The README is honest that this generality has a cost: the flow is designed to be broadly compatible, and that complexity can introduce instability or redundancy. It notes that most users will settle on one route (built-in generation or a specific API, with or without sub-agents) and suggests asking the AI to edit the skill to pin those preferences.

## Installing the skill into Codex and producing a first deck

The README's recommended install is to hand the repository URL to your agent and ask it to install the skill. For a manual install into Codex, it gives the skills CLI command below, which places the skill in Codex's global skills directory. The --skill, --agent and --global flags are exactly as documented.

```bash
npx -y skills@latest add ningzimu/codex-ppt-skill \
  --skill codex-ppt \
  --agent codex \
  --global
```

After it finishes, restart Codex so the new skill is picked up. The README also documents an offline route: download codex-ppt-skill-v*.zip from GitHub Releases, unzip it, and place the codex-ppt folder at ~/.codex/skills/codex-ppt, then restart Codex.

With the skill installed, the first real use is a prompt rather than a command. The README gives a worked example where the user asked for a 20-page deck covering two skills, ten pages each, and the resulting PDF is committed to the repository as assets/skill_duo_intro.pdf. Expect the agent to come back with an outline and a style question before it generates anything; the README states the skill guides you to confirm outline.md, per-page points, style direction and a sample before continuing.

## The limitation the README states outright: the pages are images

The tip box near the top of the README is unambiguous. The skill produces image-based slides suited to strong visual expression, and the page elements themselves are not directly editable. This is not a bug to be worked around; it is the architecture. You get pixel-level control over how a page looks and no control over the text inside it after generation.

That rules out a whole class of work. If a reviewer will ask you to change one number in a chart, or if your organisation requires text to remain selectable and searchable for accessibility, an image deck is the wrong artifact. The README points to a separate project, image-to-editable-ppt-skill, for converting the finished images back into editable PowerPoint, which is effectively an admission that the two needs are distinct.

The second limitation is backend dependence. Outside Codex, the documentation says you typically need to configure gpt-image-2.5-flare or a compatible image endpoint with a base URL and model name. If your environment cannot reach such an endpoint, the generation step has nothing to call. The README does not document a local, offline image model as a fallback.

## How it compares to template-driven generators such as python-pptx

A library like python-pptx takes the opposite approach. You build slides from shapes, text frames and placeholders, and the output is fully editable because it is made of native PowerPoint objects. The visual ceiling is set by your layout code and the fonts available on the machine that renders it.

codex-ppt-skill trades that editability for appearance. Every page is generated as a complete image by a model, so the layout can be as expressive as the image model allows, and consistency comes from reusing a confirmed style rather than from a template file. The cost is that the .pptx is essentially a container for pictures plus notes.

The two are not mutually exclusive in a workflow. The README's own suggestion is to generate with this skill and then convert with image-to-editable-ppt-skill if you need editable pages, which is a two-stage pipeline rather than a single tool doing both jobs.

## Styles, the personal style library, and what survives a reinstall

The skill ships with twelve built-in style references, including clean professional, scientific defense, party-government red, teaching courseware, e-ink magazine, hand-drawn technical explanation, dashboard and McKinsey. The README recommends the hand-drawn technical explanation style as a starting point for people who do not want to write prompts.

Beyond the built-ins, you can supply a reference image, PDF or PPT/PPTX and have the agent analyse its colours, layout, typography and visual elements before generating in that style. When you land on something you like, the README says you can save it to a personal style library at ~/.codex-ppt-skill/references/. That path sits outside the skill installation directory, so the README states it survives updates and reinstalls, and a personal style with the same name takes precedence over a built-in one.

This is the part of the project with the longest shelf life. The generation pipeline will change between releases, but a folder of style references you built up is yours, and it is the one piece of configuration the README explicitly guarantees will not be lost.

## Licence, maintenance and the real upgrade cost

The repository is MIT licensed, which permits commercial use and modification provided the copyright notice and permission text are retained. That is the standard permissive position; it is not legal advice, and if you redistribute the skill inside a product you should read the LICENSE file at the repository root rather than this summary.

The last push was on 2026-09-13, and the most recent release listed is v0.6.0 on 2026-09-12, following v0.5.5 in July 2026. The project is not archived. The upgrade cost is not the install, which is a single command, but the local edits the README actively encourages. It suggests asking the AI to modify the skill so your preferred backend, sub-agent usage, output directory and page rhythm are fixed. Those edits live in the installed skill directory, which is exactly the directory an update replaces. The personal style library at ~/.codex-ppt-skill/references/ is outside it, but your pinned preferences are not, so a reinstall can silently drop them unless you keep your own copy.

## Conclusion

Adopt codex-ppt-skill if you want a visually consistent, image-first deck from an article, paper or course notes and you work inside Codex, Claude Code or another SKILL.md agent. Do not adopt it if you need to edit text boxes, charts or tables in PowerPoint afterwards, since the README states the slide elements are not directly editable. Before committing, verify which image backend your environment actually reaches (Codex built-in generation versus an OpenAI-compatible or AtlasCloud endpoint), confirm the skill folder lands in ~/.codex/skills/codex-ppt, and check that ~/.codex-ppt-skill/references/ is writable if you intend to keep a personal style library across reinstalls.

## FAQ

### What skills should I add to Codex, and is codex-ppt-skill one of them?

The README presents codex-ppt-skill as a Codex skill installable with the skills CLI into the global skills directory, and notes it also works in agents that support SKILL.md. What other skills you add is not covered by the repository.

### Can Codex work on PPT with codex-ppt-skill?

Yes. The README states the skill is most recommended in Codex, where it prefers the built-in image generation and editing path, and that it turns articles, reports, papers and course notes into image-based presentations assembled into a .pptx.

### How do I tell Codex to use the codex-ppt-skill?

The README's workflow is prompt-driven: after installing and restarting Codex, you describe the deck you want, and the skill guides you to confirm outline.md, per-page points, style direction and a sample before generating the full set.

## Sources

- [License: MIT](https://github.com/ningzimu/codex-ppt-skill/blob/main/LICENSE)
- [ningzimu/codex-ppt-skill on GitHub](https://github.com/ningzimu/codex-ppt-skill)
- [Project website](https://ppt-skill.ningzimu.vip)
- [README](https://github.com/ningzimu/codex-ppt-skill/blob/main/README.md)
- [Releases](https://github.com/ningzimu/codex-ppt-skill/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/ningzimu-codex-ppt-skill