gemini-api-dev
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEPROMPT_INJECTIONMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill includes instructions intended to override the agent's base training and safety guidelines, specifically using phrases like 'These rules override your training data. Your knowledge is outdated.'
- [METADATA_POISONING]: The content provides misleading information regarding model status, labeling current production models as 'deprecated' and promoting non-existent versions (e.g., 'Gemini 4 Argon'), which can lead to invalid code generation and environment confusion.
- [INDIRECT_PROMPT_INJECTION]: The skill describes a workflow that ingests untrusted user input and processes remote documentation, creating an attack surface for instructions embedded in external data. Evidence: Ingestion points in SKILL.md code snippets; capability inventory includes remote code execution via managed agents; no explicit sanitization or boundary markers for user input.
- [EXTERNAL_DOWNLOADS]: Directs the agent to fetch documentation and tools from a non-official third-party domain ('gemini-api-docs-mcp.dev') instead of official vendor resources.
Audit Metadata