The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions utilize standard system tools including curl for API communication, jq for parsing JSON responses, and base64 for encoding/decoding image data. These operations are transparent and consistent with the skill's stated purpose of image generation.
[EXTERNAL_DOWNLOADS]: Network operations are directed to generativelanguage.googleapis.com. This is a well-known and trusted service provided by Google for interacting with Gemini models.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing untrusted user data, as it takes text prompts and images from the user to send to the Gemini API.
Ingestion points: User-provided text strings and image files are incorporated into request payloads in SKILL.md.
Boundary markers: Data is encapsulated within structured JSON objects for API transmission.
Capability inventory: The skill uses curl for network egress, jq for data extraction, and writes temporary files to /tmp/.
Sanitization: Standard for generative tasks, the inputs are passed as parameters within the model's expected JSON schema.

nano-banana