diffdock
Pass
Audited by Gen Agent Trust Hub on Aug 1, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches ESMFold checkpoints (e.g., 'esm2_t33_650M_UR50D') from Meta's FAIR research repository via Torch Hub.
- [COMMAND_EXECUTION]: Runs the DiffDock inference script via 'python3 -m inference' and suggests using 'sed' to modify script constants to meet sandbox resource limits.
- [PROMPT_INJECTION]: The skill processes external scientific data files, creating a surface for indirect prompt injection.\n- Ingestion points: 'target.pdb', 'batch.csv', and ligand files (.sdf, .mol2) described in 'SKILL.md' and 'references/workflows.md'.\n- Boundary markers: Absent; no delimiters or specific instructions are provided to the agent to ignore instructions embedded in data files.\n- Capability inventory: The skill executes shell commands (python3) and writes results to the local filesystem.\n- Sanitization: Absent; the skill relies on the underlying scientific libraries to parse and validate input files.
Audit Metadata