physicsnemo-discover

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s REQUIRED runtime workflow can shallow-clone an external Git repository (https://github.com/NVIDIA/physicsnemo) into a temp dir for repo discovery, and then reads text files from that outsider-authored repo into the agent’s LLM context (indirect prompt-injection risk via arbitrary repo content).

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

• Potentially malicious external URL detected (high risk: 0.80). The skill's runtime discovery step instructs agents to shallow-clone the canonical repo "https://github.com/NVIDIA/physicsnemo" into a temp directory when no local clone exists, and the fetched repository files are then read and used as authoritative input to control the agent's outputs, so this external URL is a runtime dependency.