openfold3

Pass

Audited by Gen Agent Trust Hub on Jul 2, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill downloads model weights from HuggingFace ('OpenFold/OpenFold3'). This is a standard procedure for using gated or large machine learning models and follows established security practices by requiring authentication via 'huggingface-cli' or 'HF_TOKEN'.
  • [EXTERNAL_DOWNLOADS]: Dependencies such as 'openfold3', 'nvidia-cutlass', 'boto3', and 'awscrt' are installed via pip. These are well-known, versioned packages relevant to the skill's biomodeling and GPU-acceleration tasks.
  • [COMMAND_EXECUTION]: The skill uses the 'run_openfold' command line tool to perform predictions. This is the primary intended function of the skill and operates on user-provided JSON query files.
  • [DATA_EXFILTRATION]: The documentation transparently notes that sequence data is sent to an external service ('api.colabfold.com') for Multiple Sequence Alignment (MSA) generation. This is a standard, performance-enhancing feature in the protein folding community and the skill explicitly provides instructions on how to disable this ('--use-msa-server false') for privacy-sensitive work.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 2, 2026, 01:21 AM
Security Audit — agent-trust-hub — openfold3