esm-protein-language-model
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill implements standard biological modeling workflows using the legitimate
esmpackage and EvolutionaryScale's Forge API. API token management is handled via environment variables, following best practices. - [EXTERNAL_DOWNLOADS]: The skill automatically downloads model weights (approx. 1-4 GB) from EvolutionaryScale's official infrastructure during the first initialization. This is standard behavior for pretrained machine learning models.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external protein sequence strings and PDB structure files as input.
- Ingestion points: Untrusted data enters the context through protein sequence strings and external PDB files loaded via
ProteinChain.from_pdbinSKILL.md. - Boundary markers: No explicit boundary markers or 'ignore' instructions for malformed input are present.
- Capability inventory: The skill can write predicted structures to the local file system using the
.to_pdb()method and performs network requests to fetch model weights. - Sanitization: The skill relies on the underlying
esmlibrary to parse biological data; no additional sanitization of input strings is implemented.
Audit Metadata