The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute the gemini CLI tool via shell commands within the agent's environment.
[PROMPT_INJECTION]: The skill creates a significant indirect prompt injection surface by interpolating user-provided prompts directly into shell commands without sanitization or boundary markers.
Ingestion points: User instructions are passed to the gemini command via the $prompt variable in SKILL.md (lines 14, 23, 27, 28, 47).
Boundary markers: No delimiters or instructions are used to distinguish user data from agent instructions.
Capability inventory: The skill allows for autonomous tool operations and workspace modifications through the gemini CLI.
Sanitization: The skill suggests shell variable storage but provides no actual escaping or validation for malicious shell sequences or sub-agent instructions.
[PROMPT_INJECTION]: The skill provides an --approval-mode yolo option which explicitly bypasses human-in-the-loop confirmations for tool operations performed by the subordinate agent, significantly increasing the risk of automated damage from malicious prompts.
[PROMPT_INJECTION]: Example commands include the --skip-trust flag, which is used to bypass internal trust and safety verification mechanisms within the CLI tool.

gemini-cli