ai-model-training

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

• Third-party content exposure detected (medium risk: 0.65). The skill’s runtime workflow ingests user-provided training data (e.g., JSONL/Parquet/DB “Raw Sources” → cleaning/formatting/tokenization → Trainer(... train_dataset=...)), and that data is free-form text authored by outsiders unless the operating user explicitly controls it; this text becomes LLM-readable training context via tokenization and batching.