The Agent Skills Directory

[SAFE] (SAFE): No security issues detected.\n- [PROMPT_INJECTION] (SAFE): The skill processes user instructions for image generation. This is the primary intended functionality, and potential prompt injection risks are limited to the generated image content. Analysis of Indirect Prompt Injection (Category 8): Ingestion points include user prompts and input images; capabilities are limited to file-write (images) and Gemini API calls; no specific boundary markers are used, but these capabilities are essential and pose no system-level risk.\n- [DATA_EXFILTRATION] (SAFE): GEMINI_API_KEY is handled securely via environment variables. File operations are strictly limited to reading and writing images at user-specified paths provided through command-line arguments.\n- [EXTERNAL_DOWNLOADS] (SAFE): Dependencies are limited to the official google-genai SDK and the established Pillow library, which are trusted and necessary for the skill's functionality.

nano-banana