linkfox-zhihuiya-patent-image-search
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data returned from patent databases, including patent titles, abstracts, descriptions, and specifications. This content is interpolated into the agent context and printed or summarized for the user, establishing an inherent indirect prompt injection risk surface if a patent contains adversarial text instructions.
- Ingestion points: Patent search response data received via the api endpoint in
scripts/zhihuiya_patent_image_search.py. - Boundary markers: Absent; data is handled as structured JSON and rendered into standard text reports.
- Capability inventory: File system writing and network requests via standard Python modules (
urllib.request) across the utility scripts. - Sanitization: Standard JSON serialization and field extraction are performed before displaying data, but text fields are not filtered for natural language instructions.
Audit Metadata