lora

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

• Third-party content exposure detected (high risk: 0.90). Yes — the SKILL.md training/workflow examples explicitly fetch public models and datasets from Hugging Face (e.g., AutoModelForCausalLM.from_pretrained("meta-llama/...") and load_dataset("tatsu-lab/alpaca") in the "Basic Setup" and "Training Patterns" sections), meaning the agent ingests untrusted, user-contributed third-party content that can influence training and subsequent actions, enabling indirect prompt injection.