tella-audio-sync
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads a MediaPipe face landmark model from Google's official cloud storage. This download includes a SHA-256 integrity check to verify the file before use. It also downloads video segments from the Tella platform for analysis.
- [COMMAND_EXECUTION]: The skill uses shell commands to manage its environment and process media. This includes using
pipto install specific versions of scientific libraries (numpy, scipy, mediapipe), and usingffmpegandffprobefor video processing and metadata extraction. Commands are executed using structured arguments to minimize injection risks. - [DYNAMIC_EXECUTION]: The skill extracts a Python processing script from its own reference documents at runtime and executes it to perform the lip-sync measurement. This script relies on the installed scientific libraries and the verified MediaPipe model.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as it processes external video files and URLs.
- Ingestion points: Video source URLs and binary video content from the Tella platform are downloaded for local processing.
- Boundary markers: No explicit natural language boundary markers are used during data processing, as the data is binary media rather than text.
- Capability inventory: The skill uses subprocess calls to execute
ffmpeg,ffprobe, and a local Python script; it also performs network downloads viacurlandurllib. - Sanitization: The skill uses structured list-based arguments for subprocess calls, which prevents standard command injection. It also performs integrity verification on its primary external dependency (the face landmarker model).
Audit Metadata