A termination condition is the predicate that an agent loop evaluates after each step to decide whether to continue or stop. Common termination conditions include: the goal has been achieved, a maximum number of steps or tokens has been reached, the allocated budget or time has been exhausted, or the agent's confidence in its output exceeds a threshold. Without well-defined termination conditions, agents risk running indefinitely, wasting resources, or producing diminishing-quality outputs through unnecessary iterations.