physicsnemo-discover

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

• Third-party content exposure detected (high risk: 0.85). The required runtime workflow includes shallow-cloning the external PhysicsNeMo GitHub repository (https://github.com/NVIDIA/physicsnemo) and then reading its files (e.g., SKILL.md’s “shallow-clone … for path discovery only” and subsequent Read/Glob steps), so outsider-authored repo text can be ingested into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.80). The SKILL.md explicitly directs the agent at runtime to shallow-clone and read from the canonical repo URL (https://github.com/NVIDIA/physicsnemo) when a local clone is absent, and those fetched files are then read to directly control the agent's discovery/output, so this is a runtime external dependency: https://github.com/NVIDIA/physicsnemo