glycoengineering
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill provides installation instructions for the GlycoSHIELD modeling tool from the Max Planck Institute's GitLab repository (gitlab.mpcdf.mpg.de). While originating from a reputable research organization, this involves downloading and installing code from a non-standard external source.
- [COMMAND_EXECUTION]: The skill includes Python code to interface with GROMACS (gmx) via the subprocess module to perform Solvent Accessible Surface Area (SASA) calculations.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external scientific APIs (GlyTouCan and GlyConnect), creating a vulnerability surface where malicious content in those databases could potentially influence the agent's behavior. 1. Ingestion points: Vendor-provided scripts 'scripts/glytoucan_lookup.py' and 'references/glycan_databases.md' fetch external SPARQL results. 2. Boundary markers: Absent in the data ingestion logic. 3. Capability inventory: 'subprocess.run' (executing GROMACS) and 'requests' (network communication). 4. Sanitization: Includes basic validation of returned data (e.g., checking for the 'WURCS=' prefix) and amino acid sequence normalization.
- [DATA_EXFILTRATION]: Uses the 'requests' library to communicate with public scientific databases at ts.glytoucan.org and glyconnect.expasy.org. These network operations are intended for the skill's primary research purpose but target non-whitelisted domains.
Audit Metadata