alterlab-rowan
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied data which could be used to influence agent behavior through data interpolation into API calls.
- Ingestion points: User-provided SMILES strings, PDB IDs, and molecular coordinate data (XYZ/EXTXYZ) are processed by the agent across all reference files and the main SKILL.md.
- Boundary markers: Absent. There are no instructions for the agent to use specific delimiters or to disregard potential instructions embedded within molecular data inputs.
- Capability inventory: The skill possesses the capability to execute Python scripts, access the local file system (Read/Write/Edit), and perform network operations to communicate with the Rowan API.
- Sanitization: The
protein.sanitize()method is mentioned for structural cleanup of PDB files, but no security-oriented sanitization of user-provided strings is provided. - [SAFE]: The skill uses a legitimate third-party service (Rowan) for its core functionality. All network communications and data submissions are documented and transparent to the user.
- [SAFE]: Secret management instructions are correct, advising users to utilize environment variables (
ROWAN_API_KEY) rather than hardcoding credentials.
Audit Metadata