The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied image URLs and text content through several AI models, which creates a potential surface for indirect prompt injection.
Ingestion points: Image URLs and text inputs are processed by phota/train, inworld/text-to-speech-2, and pruna/p-video-avatar.
Boundary markers: The skill does not define specific delimiters to isolate untrusted external content.
Capability inventory: The skill utilizes Bash for command execution and Write for saving output data to local files.
Sanitization: No sanitization or validation logic is specified for the external inputs.
[COMMAND_EXECUTION]: The skill uses belt app run to execute various AI applications, which is a legitimate use of the permitted Bash capability for media generation tasks.
[EXTERNAL_DOWNLOADS]: The skill references AI services and models from Google and the vendor's own infrastructure to process media assets, which is consistent with its stated purpose.

ai-podcast