The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Python script (nanobanana.py) using bash commands, interpolating user-provided inputs such as prompts and filenames directly into the command arguments. This pattern presents a risk of command injection if the agent does not properly escape characters or validate the input before execution.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user content to generate images, which could be used to influence the behavior of the underlying image model or the agent itself.
Ingestion points: User-provided text strings for the --prompt argument and image files for the --input argument in nanobanana.py.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the script or instructions.
Capability inventory: The skill possesses the ability to write files to the local filesystem (image.save) and make external network requests to the Gemini API.
Sanitization: The skill lacks explicit sanitization or validation logic for the content of the prompts or the format of input images before they are passed to the API.
[EXTERNAL_DOWNLOADS]: The skill documentation recommends installing several Python packages (python-dotenv, httpx[socks], google-genai, Pillow) via pip. These packages are from well-known registries and include official SDKs from Google.
[DATA_EXPOSURE]: The skill script reads configuration data from ~/.nanobanana.env to retrieve the GEMINI_API_KEY. While this is a sensitive file path, it is a standard practice for local secret management within this skill's architecture.

nanobanana-skill