epo-patent-search
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources (EPO OPS API and Google BigQuery), including patent titles, abstracts, and full-text descriptions. This data is then processed and presented by the agent, creating a surface for indirect prompt injection where malicious instructions could be embedded in patent metadata to influence agent behavior.
- Ingestion points: Data is fetched dynamically from the EPO Open Patent Services API and Google BigQuery's patent dataset as described in
SKILL.md. - Boundary markers: The instructions lack explicit delimitation (such as XML tags or unique delimiters) or instructions to the model to ignore embedded instructions within the fetched content.
- Capability inventory: The skill is granted the
BashandWritetools, which provide a powerful execution environment that could be targeted if the agent follows malicious instructions found within patent data. - Sanitization: No sanitization, escaping, or filtering of the external patent data is mentioned or implemented in the skill instructions.
Audit Metadata