web-search-strategy
Pass
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to perform multi-round web searches and process content from various external online sources. This establishes a potential surface for indirect prompt injection, where malicious instructions could be embedded in retrieved search results, snippets, or documents.
- Ingestion points: Data enters the agent's context through search tool results (titles, snippets, and page text) and potentially through large documents like PDFs or annual reports referenced in search results.
- Boundary markers: The skill does not explicitly define hard delimiters or boundary markers to wrap untrusted content within its internal reasoning process, though it mandates a structured output template for reporting findings.
- Capability inventory: The skill invokes web search tools and analyzes search results to generate subsequent queries and summary reports, creating a multi-step chain where retrieved data influences future agent actions.
- Sanitization: The skill includes extensive guidelines for source triage and quality verification (e.g., prioritizing official sources and standard domain types) to mitigate the impact of low-quality or deceptive content, but it does not specify technical sanitization to prevent the interpretation of instructions hidden within the retrieved data.
Audit Metadata