The Agent Skills Directory

[COMMAND_EXECUTION]: The transcribe.py script executes the gcloud CLI to manage authentication tokens and configuration. These subprocess calls are implemented securely using fixed argument lists, effectively preventing command injection risks.
[EXTERNAL_DOWNLOADS]: The skill performs HTTPS requests to transmit audio data to Google's generative language and AI platform endpoints. These operations are limited to well-known, trusted domains required for the service.
[PROMPT_INJECTION]: The skill processes external audio data for transcription, which constitutes an indirect prompt injection surface.
Ingestion points: Local audio files provided by the user are read and encoded in transcribe.py for model processing.
Boundary markers: Instructions are provided to the model to strictly perform transcription, although the binary audio data itself does not use specific delimiters.
Capability inventory: The skill is limited to making network requests to official APIs and retrieving local authentication; it does not possess capabilities for arbitrary code execution, file-system modification, or privileged operations based on the model's output.
Sanitization: The audio data is treated as a raw binary source for transcription, which is standard for this type of utility.

gemini-stt