biomed-skill-creator
Pass
Audited by Gen Agent Trust Hub on Mar 25, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted biomedical data that is subsequently passed to various analytical tools and inference models, creating a potential surface for indirect prompt injection. (1) Ingestion points: The skill handles user-supplied molecule names, SMILES strings, protein sequences (FASTA), and protein IDs across multiple files, including SKILL.md and examples/create_skill_example.py. (2) Boundary markers: There are no explicit delimiters or specific instructions provided to the agent to ignore or isolate potentially malicious instructions embedded within the user-provided data strings. (3) Capability inventory: The skill has extensive capabilities, including access to a broad tool registry for property calculation, structural analysis, and data retrieval, as well as an InferencePipeline for model-based tasks like structure prediction and docking, detailed in references/tools_reference.md. (4) Sanitization: The provided scripts and documentation do not implement explicit validation or sanitization of input strings before they are utilized by the underlying tools and pipelines.
- [SAFE]: No evidence of credential harvesting, unauthorized network exfiltration, persistence mechanisms, or obfuscated malicious code was found. The interaction with vendor-specific tools like PharmolixFM and MolCraft is consistent with the skill author's identity and primary purpose.
Audit Metadata