skills/aipoch/open-science/boltz/Gen Agent Trust Hub

boltz

Pass

Audited by Gen Agent Trust Hub on Sep 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the boltz predict command to process YAML-defined complexes and generate structure files (mmCIF/PDB) and confidence scores.
  • [EXTERNAL_DOWNLOADS]: The skill queries the ColabFold MSA server at api.colabfold.com to retrieve Multiple Sequence Alignments. The documentation explicitly states that the user's biological sequences are transmitted to this service for processing.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied protein sequences and SMILES/CCD ligand identifiers through YAML configuration files.
  • Ingestion points: The complex.yaml file, specifically the sequence and smiles fields.
  • Boundary markers: Data is parsed as structural YAML, which limits the potential for instructions to be misinterpreted as code.
  • Capability inventory: CLI command execution for prediction and local file system writes for results.
  • Sanitization: Input sequences and ligands are validated by the scientific parser in the Boltz-2 utility.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 22, 2026, 12:56 PM