vss-search-archive

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses standard shell utilities such as curl, jq, and uuidgen to interact with VSS agent APIs, Elasticsearch, and Video Storage Toolkit (VST) services. These operations are restricted to the user-provided $HOST_IP and standard VSS service ports (e.g., 8000, 9200).
  • [EXTERNAL_DOWNLOADS]: The skill retrieves configuration data and container images from trusted NVIDIA-owned resources, including the NGC registry (nvcr.io) and official GitHub organizations.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface as it processes untrusted metadata from video files and RTSP streams.
  • Ingestion points: Data enters the system via file uploads and RTSP stream registration endpoints in the VSS backend.
  • Boundary markers: There are no explicit delimiters or instruction isolation markers described for when the agent summarizes search results or video content.
  • Capability inventory: The agent is capable of executing network requests via curl, writing files to /tmp, and performing visual inspection of screenshots.
  • Sanitization: The skill does not document specific sanitization or validation logic for external metadata before it is processed by the agent's language model.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 05:22 PM
Security Audit — agent-trust-hub — vss-search-archive