patent-search

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The file tools/stdio_utf8.py contains a utility function that wraps subprocess.run to ensure UTF-8 encoding and environment variables for child processes. While this capability exists within the toolset, it is not utilized for executing untrusted user input within the provided search logic.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests content from an external source (the CNIPA patent database). If patent metadata (such as titles or abstracts) were to contain malicious instructions, they could potentially influence the agent when it processes the resulting search report.
  • Ingestion points: Data is fetched via Playwright in tools/cnipa_crawler.py from epub.cnipa.gov.cn.
  • Boundary markers: The skill uses a structured Markdown template for reports, which separates metadata from content.
  • Capability inventory: The skill has capabilities for network access (via Playwright), file writing (to outputs/), and subprocess execution.
  • Sanitization: tools/cnipa_parse.py implements regex-based sanitization that strips all HTML tags (including <script> and <style> blocks) from the scraped content before processing. Additionally, tools/emit_search_report.py escapes Markdown table delimiters to prevent report formatting breakage.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 08:53 AM
Security Audit — agent-trust-hub — patent-search