design-shotgun

Run /design-shotgun and describe the screen you want to explore. It generates distinct design concepts, confirms them with you, then launches parallel subagents to produce mockup images in about 60 seconds total.

The Skill builds an HTML comparison board from the generated variant images and serves it over HTTP in your browser. You click to submit a final choice or request regenerations, remixes, or more variants like a selected one.

Yes. If a local site is running and you reference it, the Skill screenshots the current page and uses the design binary's evolve mode to generate improvement variants from your existing design instead of starting from scratch.

Yes. It reads approved.json files from up to 10 prior sessions for the project and extracts patterns from your approved variants, biasing new generations toward your demonstrated taste unless you request a different direction.

The Skill checks for the design binary before any mockup command. If unavailable, it falls back to an HTML wireframe approach, and if the browse binary is missing it opens boards with the system default browser instead.

Failures are usually rate limiting from the image generation API. Each agent retries up to 3 times with backoff, and if all parallel attempts fail the Skill falls back to sequential generation one variant at a time.