alterlab-boltz
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
Bash(python:*)andBash(uv:*)to execute theboltzCLI tool for molecular structure prediction. - [EXTERNAL_DOWNLOADS]: Upon first run, the skill downloads and caches Boltz-2 model weights from a remote source.
- [DATA_EXFILTRATION]: When the
--use_msa_serverflag is enabled, protein sequence data is transmitted to a remote hosted service to fetch Multiple Sequence Alignments (MSAs). The skill documentation includes an explicit privacy notice regarding this behavior. - [INDIRECT_PROMPT_INJECTION]: The skill processes biological data which serves as a potential attack surface. \n
- Ingestion points: FASTA files and YAML specifications containing protein/nucleic acid sequences and SMILES strings (
SKILL.md). \n - Boundary markers: None explicitly specified for sequence inputs. \n
- Capability inventory: Shell command execution of the
boltzCLI. \n - Sanitization: No specific sequence validation or chemical descriptor sanitization is mentioned.
Audit Metadata