gptq

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

• Third-party content exposure detected (high risk: 1.00). The skill's workflow explicitly loads models and files from public HuggingFace URLs (e.g., from_quantized("TheBloke/..."), https://huggingface.co) and ingests datasets via datasets.load_dataset (e.g., "c4", "anon8231489123/ShareGPT_Vicuna_unfiltered") in SKILL.md and references/calibration.md, so it consumes untrusted, user-generated third-party content that can influence quantization and model behavior.