binder-design

Pass

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [Indirect Prompt Injection] (LOW): The skill ingests untrusted protein structure data (PDB/CIF) as input for scientific orchestration.
  • Ingestion points: The skill reads target.pdb and target.cif files as inputs for BoltzGen, RFdiffusion, and validation protocols in SKILL.md.
  • Boundary markers: No delimiters or instructions to ignore embedded content within data files are present.
  • Capability inventory: Orchestrates external execution via the modal run command for multiple scripts (e.g., modal_boltzgen.py, modal_rfdiffusion.py, modal_ligandmpnn.py).
  • Sanitization: No file validation or sanitization of input structures is documented in the skill instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 17, 2026, 06:08 PM