ecommerce-patent-references

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 displays citation data from the external Zhihuiya database, which introduces a potential surface for indirect prompt injection if the source data contains instructions designed to override agent behavior. \n
  • Ingestion points: External data is fetched from the patent database via the scripts/patent_references.py script using a POST request. \n
  • Boundary markers: The skill instructions lack explicit boundary markers or directions for the agent to ignore instructions that might be embedded in the retrieved patent titles or literature descriptions. \n
  • Capability inventory: The skill is capable of performing network requests to vendor-managed endpoints and writing JSON data to the local filesystem for session tracking and caching. \n
  • Sanitization: While the script handles JSON structure, it does not include specific sanitization or filtering of the text content within the patent references before presentation to the agent context.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:13 AM
Security Audit — agent-trust-hub — ecommerce-patent-references