vss-build-vision-ai

SUSPICIOUS. The skill’s core purpose and capabilities are broadly aligned with building and deploying NVIDIA VSS stacks, and external endpoints/installers are mostly official same-org/vendor sources. However, it has a high operational footprint: host setup, credential discovery, artifact generation, sandbox onboarding, teardown, and autonomous deployment. This looks like a legitimate but powerful infrastructure skill with medium security risk from supply-chain hygiene, credential handling, and autonomous real-world actions rather than confirmed malware.

No direct malware or data-theft behavior is evident in the supplied configuration. However, it introduces a runtime remote configuration fetch that is imported into Kibana with overwrite privileges and runs the main service with extensive host-level privileges and unverified image provenance. Pin and verify image and artifact digests, prefer locally reviewed artifacts, minimize container privileges and mounts, and restrict network/API access before production use.

The script is a readable endpoint-probing utility and shows no clear malicious payload or sabotage behavior. Its main security risk is that REMOTE_API_KEY is forwarded to any user-supplied base URL without host or scheme restrictions, allowing accidental or intentional API-key disclosure if the URL is untrusted. Restricting allowed schemes and hosts, or requiring explicit consent before sending the key, would reduce this risk.