This is a reverse-engineered Gemini Web API client that generates both text and images without going through official channels. It handles multi-turn conversations, accepts reference images for vision tasks, and supports models like gemini-3-pro and gemini-2.5-flash. The implementation requires explicit user consent before first use, which is appropriate given the nature of reverse-engineered APIs. Cookie management happens automatically through browser authentication, and sessions persist across conversations. It's useful when you need Gemini's image generation capabilities or want to prototype against Google's models without dealing with official API keys and billing, though you're accepting the usual risks that come with unofficial endpoints.
npx skills add https://github.com/freestylefly/canghe-skills --skill canghe-danger-gemini-web