training-patterns

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

• Third-party content exposure detected (high risk: 0.80). This skill explicitly loads public, user-contributed datasets (e.g., the "HuggingFace Datasets Integration" examples and multiple load_dataset(...) calls like load_dataset('glue','sst2'), 'squad', 'cnn_dailymail') and accepts arbitrary dataset files for training, so the agent ingests untrusted third‑party content as part of its workflow.