design-around-multiple-patents-ip
Pass
Audited by Gen Agent Trust Hub on Aug 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data including patent claim text and user-provided technical specifications which are potential vectors for indirect prompt injection.
- Ingestion points: Ingests target product/process data (SKILL.md, Step 1) and external patent claim data via MCP tools (SKILL.md, Step 2).
- Boundary markers: The skill establishes clear operating boundaries, uses structured templates for all outputs, and mandates specific text states (e.g., 'Potential literal overlap') to prevent the AI from interpreting data as instructions.
- Capability inventory: Limited to read-only information retrieval from patent databases using authorized MCP servers (patsnap_patent_research, advanced_patent_search, etc.); no arbitrary command execution or file-system write access is specified.
- Sanitization: The instructions explicitly direct the agent to 'Escape all external content in HTML' and 'Reject unsafe URLs' during the generation of the final report.
- [EXTERNAL_DOWNLOADS]: The skill references several external resources for its MCP services.
- Evidence: References official marketplace pages and connector keys for Patsnap services (e.g., patsnap.com/marketplace/mcp-servers/patent-search).
- Context: All referenced domains and tools belong to the skill's author (patsnap), representing legitimate vendor infrastructure for patent research functionality.
Audit Metadata