The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of the gemini CLI tool for tasks like code review and architectural analysis. It recommends specific flags such as --approval-mode plan to ensure the agent only proposes changes rather than executing them.
[DATA_EXPOSURE]: Includes proactive security measures to prevent credential leaks by instructing the agent to unset GEMINI_API_KEY and GOOGLE_API_KEY environment variables using the env -u command before invocation.
[INDIRECT_PROMPT_INJECTION]: The skill processes external repositories and artifacts, creating a surface for indirect prompt injection.
Ingestion points: External repositories, artifacts, and URLs passed to the gemini CLI (SKILL.md).
Boundary markers: The skill provides a prompt template that explicitly lists "Forbidden actions" such as accessing credentials or performing destructive commands.
Capability inventory: Shell command execution via the gemini CLI tool (SKILL.md).
Sanitization: The skill mandates local inspection and validation of all Gemini-generated reports and prohibits the use of API keys unless explicitly authorized.

gemini-cli-proper