simpo-training

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly instructs loading public, user-generated datasets via dataset_mixer (e.g., HuggingFaceH4/ultrafeedback_binarized, argilla/..., Anthropic/hh-rlhf) in SKILL.md and references/datasets.md, which are untrusted third‑party sources whose prompt/response content is ingested and used to train/drive model behavior, enabling indirect prompt injection.