protein-subcellular-localization-prediction-biot5

Pass

Audited by Gen Agent Trust Hub on Mar 25, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides biological sequence analysis capabilities using the BioT5 model within the OpenBioMed framework. Its behavior aligns with the stated purpose of predicting cellular localization from amino acid sequences.- [COMMAND_EXECUTION]: No unauthorized or dangerous command execution was found. The Python scripts are limited to environment configuration and running standard inference pipelines for machine learning.- [DATA_EXFILTRATION]: The skill does not access sensitive local configuration files (e.g., SSH keys, cloud credentials) or perform network operations to non-whitelisted or suspicious domains.- [PROMPT_INJECTION]: The skill ingests untrusted protein sequences through the Protein.from_fasta method. While this establishes a theoretical surface for indirect prompt injection given the multi-modal nature of BioT5, the risk is negligible as the task question is static and the model output is not used to drive sensitive agent actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 25, 2026, 05:59 AM