Ideate skill for Codex.
The Ideate skill has Codex turn a product brief into three independent UI design images, each with a different layout or product framing, then wait for you to pick one. It ships in OpenAI’s Product Design plugin.
Updated November 12, 2026 · 3 min read
At a glance
| Skill name | ideate |
| Comes with | The Product Design plugin, by OpenAI |
| Install | codex plugin add product-design@openai-curated, or /plugins in Codex |
| Codex uses it | When you ask for design variants, remixes, visual exploration or new directions for a product or screen |
| Source | openai/plugins on GitHub |
What it does
The skill won’t generate anything until the plugin’s get-context step has a minimum design brief. It then looks for real context: your screenshots, Figma frames, Storybook, design tokens and similar flows in the project. If a named reference can’t be opened, it stops and asks rather than ignoring it.
Then it plans and generates:
- Variation: with an existing design system it keeps that style and varies structure and hierarchy. With none, it varies both concept and visual style.
- Size: it picks dimensions up front, such as
390 x 844for mobile,1440 x 1024for a desktop app, or1440wide for a landing page. - Prompt: it sends a fixed Image Gen prompt that asks for one focused screen, no cards inside cards, body text around 14 to 16px, and at most two fonts.
- Three options: each is its own Image Gen call, and current dates are used for any mock data.
It numbers the options only in the order the images appear in the chat, then asks “Which option should I build: 1, 2, or 3?” Your feedback leads to new rounds, and the option you pick becomes the visual target for the Image to Code skill.
When to use it
Use it at the start of a feature or product, when you want to see a few real directions before any code is written.
Try it
Use the ideate skill. I’m building a habit tracker for runners: log a run, see weekly progress, and keep a streak. Give me three mobile design directions with different layouts.
Does it work in Claude Code?
No, not as written. The skill is built around Codex’s Image Gen tool, and it relies on the Product Design plugin’s router and get-context skill. Claude Code has no built-in image generation, so copying this folder into ~/.claude/skills won’t produce the three images. The brief-gathering rules and the prompt text are still useful as reference.