alterlab-ligandmpnn

Warn

Audited by Gen Agent Trust Hub on Jul 6, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references an external GitHub repository (dauparas/LigandMPNN) to clone source code and download model checkpoints for operation.
  • [REMOTE_CODE_EXECUTION]: Instructions describe cloning external code and executing it via python run.py, which involves running unverified code from a remote source.
  • [COMMAND_EXECUTION]: Utilizes python and the uv tool to manage and run the structural design pipeline.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by processing user-provided PDB and mmCIF structure files (ingestion point). It lacks boundary markers or explicit sanitization of these structural inputs while maintaining capabilities like shell execution and file system access.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 6, 2026, 11:26 PM
Security Audit — agent-trust-hub — alterlab-ligandmpnn