patent-architect

Pass

Audited by Gen Agent Trust Hub on Mar 13, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the Bash tool with a restricted scope, specifically allowing curl to interact with search APIs. This is a functional requirement for the prior art search phase.
  • [EXTERNAL_DOWNLOADS]: The skill retrieves patent data and technical documentation from well-known services, including serpapi.com and api.exa.ai. These connections are documented and target reputable technology services.
  • [PROMPT_INJECTION]: The skill ingests untrusted data from external search results and web content, which represents an indirect prompt injection surface. This risk is managed by the skill's instructions to extract only specific technical metadata and its use of a rigid structural template for the final output.
  • Ingestion points: Search results from SerpAPI, Exa.ai, and the WebSearch tool (referenced in SKILL.md).
  • Boundary markers: The agent is instructed to follow the exact structure defined in template.md.
  • Capability inventory: The skill uses Bash(curl), Write, Edit, and WebSearch capabilities.
  • Sanitization: Instructions explicitly mandate the extraction of discrete technical elements (such as publication numbers, claims, and technical solutions) rather than verbatim processing of external content.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 13, 2026, 09:42 AM