deep-research
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill utilizes the official
@google/genaiSDK to interact with well-known Google API services. All external communication is routed through trusted infrastructure. - [SAFE]: Research reports are stored locally in the
/tmpdirectory using unique interaction identifiers, minimizing data exposure risks while maintaining persistence for the current session. - [INDIRECT_PROMPT_INJECTION]: The skill accepts research queries and processes external web content as part of its primary research function.
- Ingestion points: Input queries passed to
deep-research.tsand external research data fetched via the Gemini Interactions API. - Boundary markers: None explicitly defined in the local script; relies on the remote API's safety and grounding filters.
- Capability inventory: Local file writing to
/tmpand standard output reporting indeep-research.ts. - Sanitization: Responses are filtered for text types and joined as plain markdown strings.
Audit Metadata