google-gemini-omni-video
Pass
Audited by Gen Agent Trust Hub on Jul 11, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches video artifacts and metadata from official Google Gemini API endpoints (generativelanguage.googleapis.com). These operations are restricted to well-known, trusted domains and involve no executable code downloads.
- [COMMAND_EXECUTION]: Utilizes ffprobe and ffmpeg for automated integrity and safety checks on generated video files. The implementation employs shell=False, bounded execution times, and strict output size limits to prevent exploitation.
- [CREDENTIALS_UNSAFE]: Correctly implements secure API key handling by instructing users to manage credentials via environment variables and server-side storage, effectively preventing hardcoded secret exposure.
- [DATA_EXFILTRATION]: Includes multi-layered validation logic, including SHA-256 integrity hashing and atomic file operations, to ensure secure handling of generated content and prevent unauthorized data access.
- [PROMPT_INJECTION]: While the skill processes user-supplied prompts, it incorporates comprehensive risk mitigation strategies including human review requirements, rights verification steps, and strict technical validation of all model-generated outputs.
- Ingestion points: User-provided prompts and media metadata via CLI arguments and manifest files.
- Boundary markers: Encourages natural language timing and explicit reference tags to structure inputs.
- Capability inventory: Requires network access for API communication, file system access for artifact storage, and subprocess execution for media processing.
- Sanitization: Implements rigorous post-generation checks including MP4 signature validation, duration/resolution/FPS verification, and full stream decoding.
Audit Metadata