gemini-search
Pass
Audited by Gen Agent Trust Hub on May 4, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform web searches by executing the
geminiCLI tool through the system's Bash environment.\n - Evidence: Usage example
gemini --prompt "WebSearch: <query>"inSKILL.md.\n - Context: The
geminiCLI is an official utility for Google's Gemini models.\n- [COMMAND_EXECUTION]: An indirect command injection surface is present where user-controlled strings are interpolated into shell commands.\n - Ingestion points: The
<query>placeholder inSKILL.mdis populated with arbitrary text from user search requests.\n - Boundary markers: The query is enclosed in double quotes in the bash command, providing minimal protection against shell metacharacters.\n
- Capability inventory: The skill relies on the Bash tool for its primary functionality, providing an execution path for potential injection.\n
- Sanitization: There are no instructions to sanitize, validate, or escape the
<query>content before it is processed by the shell.
Audit Metadata