pyvene-interventions

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly loads external intervention/model artifacts at runtime from the Hugging Face hub (e.g., IntervenableModel.load("zhengxuanzenwu/intervenable_honest_llama2_chat_7B" — https://huggingface.co/zhengxuanzenwu/intervenable_honest_llama2_chat_7B), which injects externally provided intervention parameters that directly alter model generation/behavior.