Guide · Agent skill
How to use the Codex PPT Deck Builder agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
codex-ppt is an agent skill that turns an article, report, paper, set of notes or outline into a PowerPoint deck. Every slide is a full 16:9 picture produced by an image model, and the pictures are assembled into a .pptx file. This guide explains what the skill does, how to install it, how to ask for a deck in plain language, and what to check before you rely on the output.
Quick start
- 1Download and unzip. Download the skill ZIP from the directory page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder.
- 2Or upload in the Claude apps. In the Claude apps, open Settings → Capabilities → Skills and upload the ZIP.
- 3Confirm python3. Make sure python3 is available, since the skill requires it.
- 4Ask for a deck. Describe your source material and the presentation you want, with a page count if you have one.
- 5Approve the outline and style. Review the proposed slide roles and choose one of the offered visual styles.
- 6Confirm the image backend. Accept the backend the agent reports, or decide on the CLI/API fallback if no built-in tool is available.
- 7Approve the sample and receive the .pptx. Review the single sample slide, then let the agent generate the full deck and assemble the file.
What an agent skill is and what codex-ppt makes the agent do
An agent skill is a folder containing a SKILL.md file and supporting files. When the skill is installed, the agent reads its instructions and follows them for matching requests. The codex-ppt skill is licensed under MIT.
Once installed, the agent works through a fixed sequence: it understands the source material, drafts a deck outline, proposes a visual style, confirms an image backend, and generates one sample slide for your approval. Only after you approve the sample does it generate the full deck. Each slide is then produced by an image model and stitched into a .pptx with the bundled assembly script.
- Output is a .pptx made of full-slide images, one per slide.
- Text is baked into the images, so text boxes, charts and shapes are not separately editable.
- The agent checks in with you at each approval gate before generating images.
- Best for visually consistent decks from long source material. Not suited to decks that must stay fully editable.
Install the codex-ppt skill
Download the ZIP for the skill from the directory page and unzip it. Place the resulting skill folder in your personal skills directory, ~/.claude/skills/, or in a project's .claude/skills/ folder if you want it only for that project.
In the Claude apps you can instead upload the ZIP under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check what you need before you start
The skill requires python3. The image backend is the part that needs care: the skill prefers a built-in image generation tool in your agent environment. If that tool is not available, it can fall back to a local script that calls an OpenAI-compatible image API, which needs an API key.
The workflow stops if no image backend is available, so confirm you have one of the two before starting.
- python3 (required)
- OPENAI_API_KEY (optional, only for the CLI/API fallback)
- OPENAI_BASE_URL (optional, API base URL)
- CODEX_PPT_IMAGE_MODEL (optional, defaults to gpt-image-2.5-flare)
- CODEX_PPT_HOME (optional, overrides the runtime home)
Trigger the skill with a natural request
You do not need special syntax. Describe the source and the result you want, and say that you want a presentation. Include the page count if you have one; otherwise the agent picks a practical count, typically 8 to 12 slides.
Expect the agent to stop and ask for confirmation at several points. Answer those questions rather than telling it to skip them, because the gates exist to stop it spending effort on a deck you have not approved.
- Example: "Make a 10-slide deck from this report, for a non-technical board audience."
- Example: "Turn these research notes into slides. Show me the outline first."
- Example: "Build a presentation from this paper. I want to approve a sample slide before the full deck."
Walk through the approval stages
The agent proposes an outline with slide roles and any source images it needs, then waits for you. It offers two or three style directions, recommends one, and keeps that visual identity across the deck. It states which image backend it found and asks you to confirm it.
After you approve the outline, style and backend, it generates exactly one sample slide. Review it for text, style and layout. If you approve, the agent dispatches the remaining slides to slide subagents when the runtime supports them, records each result with the bundled scripts, then writes speaker notes and assembles the .pptx.
- Outline approval
- Style choice (for example from the bundled style references)
- Backend confirmation
- Sample slide approval
- Full deck generation, QA and assembly
Review the output and handle problems
Before assembly the agent inspects every slide for text accuracy, outline match, truncation, style, stray page numbers, overlaps and required assets. Severe failures are regenerated with a tighter prompt. Localised fixes use backend editing when it is available.
If a required subagent or image backend is unavailable, the agent is instructed to stop and report a blocker with the slide id rather than produce a lower-quality replacement. Check the final report for the backend used and any limitations.
- Because text is part of the image, typo fixes may need regeneration rather than a quick edit.
- Confirm any required figures or logos appear correctly on the slides that use them.
- Read the speaker notes against the source; they are generated text.
What people use it for
- Turning a long industry report into a visually consistent board presentation.
- Converting research paper notes into a teaching deck with full-slide illustrated pages.
- Drafting a team briefing from a set of meeting notes or an internal article.
- Producing a quick deck from an outline when editable slides are not required.
Frequently asked questions
How do I install the codex-ppt skill?
Download the skill ZIP from the directory page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps you can upload the ZIP under Settings → Capabilities → Skills instead.
Does codex-ppt need an OpenAI API key?
Not always. The skill prefers the built-in image generation tool in your environment. OPENAI_API_KEY is only needed for the CLI/API fallback, which is used when the built-in backend is unavailable or you ask for API mode.
Can I edit the text in the generated slides?
No. Each slide is a flat image with the text baked in, so text boxes, charts and shapes are not separately editable. If you need editable slides, this skill is not the right fit.
Is codex-ppt free and what is its licence?
The skill is listed under the MIT licence. Any costs depend on your agent environment and, if you use the fallback, on your image API account; the source does not state pricing.
Is it safe to use confidential documents with codex-ppt?
Use caution. When the CLI/API fallback is used, the content may be sent to an image API and leave your machine. The skill runs scripts and uses the network, so review it before installing and avoid sensitive material unless you accept that risk.
What if the skill stops with a backend error?
The workflow stops when no image backend is available. Check that a built-in image tool is callable in your environment, or configure OPENAI_API_KEY and optionally OPENAI_BASE_URL and CODEX_PPT_IMAGE_MODEL for the fallback.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Choose codex-ppt if you want a polished, consistent deck quickly from long source material and you are comfortable with full-slide images that cannot be edited as separate boxes. Pick a different tool if every chart, shape or text box must stay editable, or if your source material is confidential and must not go to an image API.
Download the skill