adapting-transfer-learning-models

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

• Third-party content exposure detected (high risk: 0.70). The skill explicitly downloads pre-trained models and datasets from public hubs (e.g., SKILL.md examples: "Download the BERT/ResNet model and load a ... dataset" and assets/example_config.json comments: "Choose from Hugging Face model hub" / "Choose from Hugging Face datasets"), meaning it ingests untrusted, user-contributed third-party content that can influence model-training and subsequent actions.