research

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved via web search tools, which represents a surface for indirect prompt injection attacks.\n
  • Ingestion points: Web content fetched during the research fallback workflow.\n
  • Capability inventory: Tool use (web search/fetch), local script execution, and document generation.\n
  • Boundary markers: The skill enforces strict citation rules and explicitly differentiates between retrieved sources and background knowledge.\n
  • Sanitization: Content synthesis is handled by the agent's reasoning process without explicit technical filtering described.\n- [EXTERNAL_DOWNLOADS]: The skill documentation notes a dependency on the standard Node.js docx package for generating document-based reports, which is a legitimate dependency for its stated output capability.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 06:49 PM
Security Audit — agent-trust-hub — research