vss-deploy-video-embedding

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill wires up the RT-Embed microservice which ingests outsider-authored free text at runtime via its REST API (e.g., POST /v1/generate_text_embeddings uses the caller-provided text_input).

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill pulls and runs a remote container image (nvcr.io/nvidia/vss-core/vss-rt-embed) at runtime and also fetches the model repo via MODEL_PATH (git:https://huggingface.co/nvidia/Cosmos-Embed1-448p) on first-boot to download model weights / build the Triton repo, which constitutes execution of externally fetched code/assets required for operation.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly instructs the agent to run privileged host operations (e.g., sudo -n docker, sudo -n chown -R 1001:1001) and to change host file ownership, which requests use of elevated privileges and modification of host filesystem state.