gemini-system
Pass
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill relies on the
geminiCLI tool to execute analysis and research tasks. The instructions include piping file contents and directory paths to this external command.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it automatically processes untrusted data from various sources.\n - Ingestion points: Content is ingested from external files (PDF, video, audio, images) and live Google Search results (GitHub, blogs, documentation).\n
- Boundary markers: The prompt templates provided in
SKILL.mdandreferences/lib-research-task.mddo not utilize delimiters or specific instructions to isolate untrusted content from the agent's core logic.\n - Capability inventory: The tool possesses significant capabilities, including the ability to read the entire project repository (
--include-directories) and access the network for research.\n - Sanitization: There is no evidence of sanitization or validation of the data extracted from multimodal files or web sources before it is interpreted by the agent.
Audit Metadata