ai-for-science-proteinbert

Warn

Audited by Gen Agent Trust Hub on Mar 28, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill implements weight conversion and inference logic in scripts/proteinbert_pytorch/convert_weights.py, scripts/deploy_toolkit/convert_weights.py, and scripts/proteinbert_pytorch/inference.py using pickle.load(). This method of deserialization is insecure; loading a maliciously crafted pickle file can result in arbitrary code execution on the user's system. While this is a standard practice for handling model weights, it necessitates caution regarding the source of the input files.\n- [EXTERNAL_DOWNLOADS]: The documentation in SKILL.md references model weights hosted on Zenodo (zenodo.org). Zenodo is a recognized research data repository. This resource is downloaded for the intended purpose of initializing the pre-trained model.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 28, 2026, 02:17 AM
Security Audit — agent-trust-hub — ai-for-science-proteinbert