deep-research
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes untrusted external data from the web using native search and fetch tools, creating a surface for potential indirect prompt injection attacks.
- Ingestion points: External content enters the agent's context through the
WebSearchandWebFetchtools used during the research phases. - Boundary markers: The skill explicitly instructs the agent to use a "skeptical evidence evaluation" framework, adversarial critique via a "Devil's Advocate" agent role, and a structured "Evidence Ledger" to isolate claims from their sources.
- Capability inventory: The skill utilizes web browsing capabilities and report generation; it does not contain local script execution, file system writes (other than report output), or persistence mechanisms.
- Sanitization: The skill mandates rigorous validation protocols, including a source hierarchy (prioritizing primary/scholarly sources), requirement for independent triangulation of material claims, and explicit labeling of claim status (e.g., confirmed, contested, or inferred) to mitigate misinformation or malicious instructions hidden in search results.
Audit Metadata