nel-assistant

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

• Third-party content exposure detected (high risk: 1.00). The skill's Step 3 explicitly instructs the agent to use WebSearch to fetch and "read the FULL text" of model cards from public sites like HuggingFace (and build.nvidia.com) and to extract configuration parameters that directly change deployment/run behavior, so it ingests untrusted third-party web content that can influence actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs using a deployment.pre_cmd to curl a remote Python parser from Hugging Face (e.g., https://huggingface.co/.../reasoning_parser.py or the expected nano_v3_reasoning_parser.py) which would be fetched at runtime and executed/used to control prompting/behavior, so this is a runtime remote-code dependency.