The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill contains a vulnerability surface for indirect prompt injection through the search_grounded_image.py script, which incorporates external search results into the prompt context.
Ingestion points: External web search data (via grounded search) and user-supplied image/text inputs.
Boundary markers: None identified in the provided documentation or code snippets to separate external data from instructions.
Capability inventory: Network access (Google API), file system write (saving generated images), and shell script execution.
Sanitization: No sanitization or escaping of external search content is mentioned.
[EXTERNAL_DOWNLOADS] (SAFE): The skill requires google-genai and Pillow. google-genai is a package from the trusted organization google, and Pillow is a standard, reputable library. No unknown or untrusted dependencies were detected.
[CREDENTIALS_UNSAFE] (SAFE): API keys are managed through the GEMINI_API_KEY environment variable. No hardcoded credentials or secrets were found in the skill content.

nano-banana