simpo-training

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly ingests public, user-generated datasets (e.g., dataset_mixer entries like HuggingFaceH4/ultrafeedback_binarized, argilla/distilabel-math-preference-dpo, Anthropic/hh-rlhf and examples for username/my-preferences or json data_files) from Hugging Face/Argilla which the agent will read and use for training, exposing it to untrusted third-party content that could carry indirect prompt injection.