materials-screening
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill establishes a workflow that ingests data from external, third-party sources, creating a surface for indirect prompt injection attacks where malicious instructions could be embedded in scientific data.
- Ingestion points: Data is retrieved from the Materials Project API and external structural data files (CIF format) as described in Step 1.
- Boundary markers: The skill lacks explicit instructions or delimiters to ensure the agent ignores or sanitizes potential natural language instructions embedded within the metadata or comment fields of the retrieved materials data.
- Capability inventory: The skill coordinates structural analysis and optimization tasks through specialized tools (pymatgen and scipy). While these are computational in nature, the results are used to generate a final ranking and report, meaning injected instructions could influence the selection rationale or final output presented to the user.
- Sanitization: There is no mention of validation, schema enforcement, or filtering to prevent the agent from interpreting data fields as instructions.
Audit Metadata