This is a browser automation CLI wrapper that lets Claude drive Chrome through the DevTools Protocol. You point it at local preview URLs, click buttons, fill forms, take screenshots, and extract page state without writing Puppeteer scripts. It's built for QA loops and dogfooding Open Design prototypes, with explicit guardrails against submitting forms or bypassing auth without confirmation. The CDP startup sequence is particular about attachment order to avoid Chrome crash loops, and it keeps all upstream docs out of context by redirecting them to temp files. Useful when you need Claude to actually see and interact with rendered UI instead of guessing from HTML.
npx skills add https://github.com/nexu-io/open-design --skill agent-browser