The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation describes an attack surface for indirect prompt injection where untrusted user prompts could potentially influence sensitive tool operations.
Ingestion points: Natural language parameters prompt in the generate_image tool and instruction in the edit_image tool.
Boundary markers: No delimiters or isolation strategies are defined to separate user-provided input from the agent's internal logic.
Capability inventory: The skill possesses significant file system capabilities, including reading local files via upload_file and writing to arbitrary locations via the output_path parameter in generate_image.
Sanitization: No sanitization, validation, or filtering of user-provided paths or prompt content is specified in the documentation.

nanobanana