The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied prompts and image files, which are then passed to the Gemini API for generation or editing. The absence of specific boundary markers or sanitization for these inputs creates a surface for indirect prompt injection where instructions embedded in the data could influence the model's behavior.
Ingestion points: The --prompt and --input command-line arguments in nanobanana.py are used to collect data for processing.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when passing the prompt to the API.
Capability inventory: The skill includes the ability to read local files via PIL.Image.open, write to the local filesystem via image.save, and perform network operations using the google-genai library.
Sanitization: The script does not perform validation or escaping on the prompt string before it is sent to the generative model.
[EXTERNAL_DOWNLOADS]: The skill documentation specifies prerequisites that involve downloading standard, well-known Python packages from official registries, including google-genai, Pillow, and python-dotenv.

nanobanana