pyvene-interventions

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

• Third-party content exposure detected (high risk: 0.80). The skill loads models and interventions from public hubs (e.g., transformers.from_pretrained("gpt2"/"meta-llama/...") and IntervenableModel.load("username/my-intervention" or "zhengxuanzenwu/...") / HuggingFace URLs), which are untrusted, user-provided third-party artifacts that the agent ingests at runtime.