vertex-ai-media-master
Pass
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection attacks because it instructs the agent to analyze external, untrusted media (video, audio, and images) while having access to sensitive tools.\n
- Ingestion points: Analysis of video files (up to 6 hours), competitor videos, audio tracks, and images via Gemini 2.5 Pro and Imagen 4 (SKILL.md).\n
- Boundary markers: No explicit boundary markers (e.g., XML delimiters) or instructions to ignore embedded commands are included in the prompt templates provided.\n
- Capability inventory: The skill is configured with broad tool access, including
Bash(general:*),Write,Edit, andRead(Frontmatter).\n - Sanitization: The instructions do not define sanitization or validation logic for the content processed or the resulting model outputs before they are used in automated workflows.\n- [SAFE]: The skill utilizes official Google Cloud SDKs (
google-cloud-aiplatform,google-generativeai) and provides links to authoritative Google Cloud documentation and a trusted repository (github.com/GoogleCloudPlatform).\n- [SAFE]: Security-conscious instructions are provided, specifically recommending Google Cloud Secret Manager for API key storage and the use of VPC Service Controls for data residency.
Audit Metadata