watch
Pass
Audited by Gen Agent Trust Hub on Jul 3, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill accepts an optional free-text
focusparameter used to guide the agent's attention during observation, creating a surface for indirect prompt injection. * Ingestion point: Thefocusparameter defined in the inputs is interpolated into the workflow during the composition of the observation report. * Boundary markers: No delimiters or isolation markers are specified to separate thefocustext from the agent's instructions. * Capability inventory: The skill utilizesffmpegfor local file processing and interacts with external Whisper APIs for transcription. * Sanitization: No sanitization or validation logic is defined for the content of thefocusparameter. - [COMMAND_EXECUTION]: The skill invokes
ffmpegandffprobeto process local video files, extract frames, and sample audio according to user-defined parameters. - [EXTERNAL_DOWNLOADS]: The skill relies on well-known external transcription services, specifically Whisper backends provided by Groq or OpenAI, to handle audio processing.
Audit Metadata