The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands for service discovery, data retrieval, and text processing.
Evidence: Employs curl for API interactions, jq for JSON parsing, and python3 for regex-based response cleaning in SKILL.md.
[EXTERNAL_DOWNLOADS]: Performs network operations to interact with the VSS agent and VST (Video Storage Toolkit) APIs.
Evidence: Communication with http://${HOST_IP}:8000 and http://${HOST_IP}:30888 documented in SKILL.md. These endpoints correspond to the NVIDIA VSS architecture services.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by interpolating untrusted user queries into the payload of an API call to a downstream AI agent.
Ingestion points: User queries are embedded into the input_message field of the /generate endpoint call in SKILL.md.
Boundary markers: Absent; user input is directly concatenated into the instructional string.
Capability inventory: The agent can execute shell commands (curl, jq, python3) and perform network requests based on the skill's instructions.
Sanitization: No explicit validation or escaping of the user-supplied query is performed before interpolation.

vss-ask-video