Downloaded complete HTML may execute external or untrusted scripts when opened
Source references: 3The Skill directs the agent to save the service-returned complete HTML unchanged and states that default HTML loads an Iconify script. The content is generated by an external AI service, yet the instructions do not require inspection, sanitization, or isolation before opening or integration.
When a prototype is opened, scripts or remote resources in the HTML could run with browser privileges, make network requests, track access, or read same-origin-accessible data. The risk depends on the returned HTML; the evidence does not show that malicious code is actually present.
This is a plausible conditional risk. The Skill retrieves externally AI-generated component code and saves the complete HTML unchanged; it also states that the default HTML loads the Iconify script. If the user later opens that file in a browser, scripts can run in the page context and make third-party requests. No review, sanitization, or isolation requirement is visible. Users can require previews to run in an unprivileged sandbox and disable or inspect remote scripts before opening them.
### HTML prototypesThe component `code` is a complete HTML document. Save it directly to a `.html` file. No build step needed.Show 2 other places
Both list and get accept an optional `inlineIcons` query param (default `false`). When omitted, icons render as `<iconify-icon>` web components and the HTML pulls in the Iconify script, so leave it off by default. Pass `?inlineIcons=true` only when the consumer needs self-contained SVGs in the HTML (for example, importing into tools that don't run scripts).When the user wants to implement the designs in code (not just preview them), **always fetch the component HTML code**. Do not rely on screenshots alone.Use `GET /api/v1/projects/:id/components/:componentId` to fetch each screen's code. The `componentId` comes from the chat run's `result.operations`.