The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs and executes gemini shell commands. It specifically instructs the use of printf and heredocs to handle prompts containing special characters, which effectively prevents shell command injection vulnerabilities when passing user-controlled content to the CLI.
[SAFE]: Potentially risky execution modes, such as --approval-mode yolo (which enables automatic task approval), are explicitly protected by a requirement to ask the user for permission via AskUserQuestion before use.
[SAFE]: The skill promotes secure execution environments by documenting the -s or --sandbox flag, noting its use of isolation technologies like gVisor on Linux and Seatbelt on macOS for tool execution.
[SAFE]: Error handling and retry logic (e.g., for MODEL_CAPACITY_EXHAUSTED) are implemented to manage runtime issues without compromising security or integrity.

gemini