simpo

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

• Third-party content exposure detected (medium risk: 0.30). Training-time preference text is ingested from configured datasets (e.g., HuggingFaceH4/ultrafeedback_binarized, argilla/...cleaned, nvidia/HelpSteer, Anthropic/hh-rlhf) via the dataset_mixer workflow, so outsider-authored free text can be supplied through those dataset sources.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill's Quick start installation explicitly runs "git clone https://github.com/huggingface/alignment-handbook.git" followed by "python -m pip install .", which fetches and executes remote repository code at runtime (https://github.com/huggingface/alignment-handbook.git), so it is a required external dependency that executes remote code.