gemini-search
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION] (HIGH): The skill directs the agent to execute shell commands incorporating user-provided input (e.g.,
gemini --prompt "WebSearch: <query>"). This pattern is highly susceptible to shell command injection if the input is not strictly sanitized or escaped by the agent platform. - [PROMPT_INJECTION] (HIGH): The skill possesses a significant Indirect Prompt Injection surface (Category 8). It ingests untrusted data from the web via the
geminiCLI. - Ingestion points: Output from the
geminiCLI command. - Boundary markers: None. There are no instructions to the agent to treat the CLI output as untrusted or to ignore embedded instructions.
- Capability inventory: The skill has command execution capabilities, making it dangerous if search results contain malicious instructions that the agent might follow.
- Sanitization: None. The skill lacks any mechanism to filter or escape content retrieved from the web before the agent processes it.
Recommendations
- AI detected serious security threats
Audit Metadata