If you don’t want the agent to probe the CLI when it needs it, a skill can describe the commands, arguments and flags so the agent can use them as needed.