ecommerce-patent-detailed-bibliography

Warn

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted patent metadata which may contain malicious instructions.\n
  • Ingestion points: API responses processed in scripts/patent_detailed_bibliography.py.\n
  • Boundary markers: Absent. The script outputs raw or summarized patent data without delimiters or warnings.\n
  • Capability inventory: File-write operations and network requests are performed in scripts/patent_detailed_bibliography.py.\n
  • Sanitization: Absent. The script does not filter or escape the content of the patent abstracts, titles, or other metadata fields before they are displayed to the user or saved to disk.\n- [COMMAND_EXECUTION]: The script scripts/patent_detailed_bibliography.py constructs local file paths using the SESSION_ID environment variable without sanitization.\n
  • The sid variable, derived from the environment, is used in os.path.join to determine the save location for JSON responses. An unsanitized session ID containing traversal sequences (e.g., ../) could allow writing files to unintended locations on the host system.\n- [EXTERNAL_DOWNLOADS]: The skill performs network requests to an external API proxy.\n
  • Fetches patent information from the official Zhihuiya API proxy endpoint managed by the vendor nexscope-ai.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 14, 2026, 04:14 AM
Security Audit — agent-trust-hub — ecommerce-patent-detailed-bibliography