ai-for-science-proteinbert

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly instructs downloading a pretrained .pkl weight file from Zenodo (https://zenodo.org/records/10371965/files/epoch_92400_sample_23500000.pkl) (SKILL.md) which is then loaded via pickle.load in multiple runtime scripts (e.g., scripts/deploy_toolkit/convert_weights.py and scripts/proteinbert_pytorch/convert_weights.py/inference), meaning arbitrary, untrusted third‑party content is ingested and can materially change model behavior (and even execute code during unpickling).