ecommerce-patent-claims
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes external patent claim data, which could potentially contain malicious instructions intended to influence the agent's behavior.
- Ingestion points: The
scripts/patent_claims.pyscript fetches patent data from the vendor's API endpoint. - Boundary markers: The skill instructions do not specify explicit delimiters or "ignore" instructions for the retrieved claim text when presenting it to the user.
- Capability inventory: The skill performs file writes to a local directory and network requests to the vendor's proxy, but does not use dangerous execution functions (e.g.,
eval,exec,subprocess) on the retrieved data. - Sanitization: No specific sanitization or filtering of the retrieved patent claims is implemented before the data is presented to the agent context.
Audit Metadata