esm

Pass

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: LOWEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS] (INFO): The skill depends on the esm Python library and PyTorch. Meta (facebookresearch) is a trusted organization, and these are standard industry-standard dependencies for bioinformatics.
  • [PROMPT_INJECTION] (LOW): The skill processes external FASTA files (--input-faa). This presents an indirect prompt injection surface where malicious sequence headers could attempt to influence agent behavior. However, because the primary outputs are numerical embeddings and log-likelihood scores, the capability for an attacker to influence downstream logic is minimal.
  • Ingestion points: reads protein sequences from user-provided .fasta files.
  • Boundary markers: None specified in the prompt template.
  • Capability inventory: Executes modal run for cloud-based inference and writes numerical data to local CSV/NPY files.
  • Sanitization: Not explicitly mentioned, though standard FASTA parsers typically enforce strict character sets.
Audit Metadata
Risk Level
LOW
Analyzed
Feb 16, 2026, 01:28 AM