The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/test_banana.py utilizes the subprocess module to execute the skill's own CLI tool. This execution is limited to automated unit testing of argument parsing and model capability validation and does not involve untrusted input or shell execution.\n- [INDIRECT_PROMPT_INJECTION]: The skill provides a surface for data ingestion via user-supplied text prompts and reference images which are processed by third-party generative models.\n
Ingestion points: Text prompts provided via -p / --prompt and image files provided via -i / --input-image in scripts/banana.py.\n
Boundary markers: None implemented; prompts are passed directly to the GenAI API.\n
Capability inventory: Capabilities are restricted to making authenticated API calls to Google's GenAI services and writing the resulting image data to a user-specified local file path.\n
Sanitization: No sanitization or delimiter wrapping is applied to the input text before model ingestion.\n- [EXTERNAL_DOWNLOADS]: The skill defines dependencies on well-known Python packages including google-genai, pillow, and google-auth within its script metadata. These libraries are necessary for the tool's core functions and are sourced from standard package registries.

nano-banana