physical-ai-neural-reconstruction

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

• Third-party content exposure detected (high risk: 0.75). The router’s runtime path clones/fetches the outsider-authored upstream repo https://github.com/NVIDIA/nurec-skills and then reads .../SKILL.md from that clone into the agent’s LLM context (indirect prompt injection risk via upstream free-text).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent at runtime to git clone and read the upstream skill repository at https://github.com/NVIDIA/nurec-skills.git, and those fetched SKILL.md files are used as authoritative instructions that directly control agent behavior.