boltzgen
Warn
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS] (MEDIUM): The skill clones untrusted repositories from
hgbrian/biomodalsandHannesStark/boltzgen. These sources are not on the trusted list. - [REMOTE_CODE_EXECUTION] (MEDIUM): The skill executes downloaded scripts via
modal runandpython. This pattern of 'download then execute' from untrusted sources is a high-risk vector, downgraded to MEDIUM as it is the primary intended use case for the design tools. - [COMMAND_EXECUTION] (LOW): Uses
modalandpythonCLI for various pipeline operations. - [INDIRECT_PROMPT_INJECTION] (LOW): Processes untrusted YAML and CIF/PDB files. Evidence: 1. Ingestion: binder_config.yaml, target.cif. 2. Boundary markers: Absent. 3. Capability inventory: modal run (remote execution), local python execution. 4. Sanitization: Includes 'boltzgen check' command for schema verification.
Audit Metadata