Please report security issues privately through GitHub Security Advisories rather than opening a public issue.

• Your prompts and images go to Google. Every prompt, and every image passed to edit_image or analyze_design, is uploaded to the Gemini API. Your .gemini-design-profile.json contents (colors, fonts, framework) are injected into prompts and sent too.
• One outbound host. generativelanguage.googleapis.com, via the google-genai SDK. There is no telemetry, no analytics, and no first-party backend.
• Your API key is read from the GEMINI_API_KEY environment variable. It is never logged — the startup line records only [set] or [missing].
• Local files. Generated assets and metadata sidecars are written to ~/.cache/gemini-visual-design/preview/ and pruned after 7 days. save_asset copies into a directory you name; the destination is constrained to your project root and the filename must be a bare filename with no path separators.

The plugin executes no user-supplied code and spawns no subprocesses. The main risk surface is filesystem writes via save_asset and init_style_profile, and prompt content reaching a third party.