fair-esm2
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation describes fetching pre-trained model weights from
dl.fbaipublicfiles.com. This is the established distribution domain for Meta AI (Facebook Research) models and is standard for thefair-esmlibrary's operation. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external protein sequence data, which represents a potential ingestion surface for untrusted data.
- Ingestion points: Protein sequence strings (e.g., in
bc()calls) and FASTA files referenced in the remote compute examples. - Boundary markers: None explicitly defined in the provided snippets; the agent is expected to handle raw biological sequence strings.
- Capability inventory: The skill uses the
fair-esmlibrary to perform transformer-based ML inference (embeddings, logits, and contact prediction) on the provided sequences. - Sanitization: The provided code snippets do not include explicit sequence validation or sanitization, which is typical for research-oriented biological modeling tools.
Audit Metadata