gemini-agent-booster
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest massive context (2M+ tokens) including monolithic codebases, production log histories, and web search results. This untrusted external data is processed in the same context as instructions and high-privilege tool capabilities.
- Ingestion points: Full project file trees, whole-file source ingestion, and millions of lines of production logs as described in sections 3 and 4 of
SKILL.md. - Boundary markers: No specific boundary markers or instruction-isolation techniques are prescribed for the monolithic data ingestion.
- Capability inventory: Parallel tool execution for file reading/editing, browser automation (via
browser-automation-expert), and real-time network access via Search Grounding and the Multimodal Live API. - Sanitization: The instructions do not specify any sanitization, filtering, or validation of the external content before processing or using it to trigger tools.
Audit Metadata