The Agent Skills Directory

[SAFE]: The skill performs legitimate operational tasks, such as checking the status of Docker containers (rtvi-vlm, perception-alerts) and interacting with local API services on ports 8000, 30888, and 9901. These operations are required for managing the video analytics pipeline and do not involve unauthorized command execution.
[SAFE]: All external references, including the GitHub repository for the blueprint and the container images for sample data, originate from official NVIDIA sources, which are trusted vendors for this skill.
[SAFE]: The skill identifies an indirect prompt injection surface as it passes natural language user requests to an internal agent endpoint (/generate). This is a functional requirement for the natural language interface. While explicit boundary markers are not mentioned in the instructions, the actions triggered are limited to the local VSS environment and do not pose a high security risk.
Ingestion points: The input_message parameter sent to the $AGENT/generate endpoint in SKILL.md.
Boundary markers: Absent in the provided skill instructions.
Capability inventory: Starting/stopping alerts, adding RTSP sensors to VIOS, and querying incident history from Elasticsearch.
Sanitization: No explicit sanitization or validation of the natural language input is described in the skill body.

alerts