gemini-cli
Warn
Audited by Gen Agent Trust Hub on Jun 24, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill documentation encourages and provides commands for installing external extensions directly from GitHub URLs, specifically
https://github.com/gemini-cli-extensions/conductorandhttps://github.com/endorlabs/gemini-extension, which introduces unvetted third-party code into the execution environment. - [COMMAND_EXECUTION]: The skill makes extensive use of the
--yolo(or-y) flag, which is designed to auto-approve all tool calls—such as file writes, network requests, and shell commands—bypassing user confirmation and increasing the impact of potential malicious instructions. - [PROMPT_INJECTION]: The skill instructions advise the agent to use 'forceful language' like 'Apply now' or 'Do this without asking for confirmation' to override the CLI's internal planning prompts and confirmation requirements, effectively teaching the model to bypass its own safety checkpoints.
Audit Metadata