vertex-ai-media-master
Pass
Audited by Gen Agent Trust Hub on May 14, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill enables processing of external media files (video, audio, images) from local or Cloud Storage paths via Gemini and Imagen models. This creates a surface where adversarial content embedded within processed media could influence the agent's behavior, although this is a standard risk for multimodal processing workflows.
- Ingestion points: media files are ingested from Google Cloud Storage (
gs://) URIs and local filesystem paths as described inSKILL.mdandreferences/examples.md. - Boundary markers: Prompting examples in
references/examples.mddo not currently utilize explicit markers or system instructions to ignore embedded commands within source media. - Capability inventory: The skill is granted
Bash,Write,Edit, andReadtools, which allow for the management and modification of generated assets. - Sanitization: There are no specific instructions for sanitizing or filtering external media content for embedded instructions prior to model processing.
Audit Metadata