A situated agent is a computer-based system that is embedded in a real-time world or environment, is ascribed with some mental states, and may enjoy a disconcerting variety of properties such as proactivity, reactivity, etc. In this paper we present the design of a class of knowledge-based situated agents and an agent specification language, called SICLE. After the SICLE interpreter receives a SICLE specification of an agent's initial knowledge base, atomic actions (their conditional effects and executability conditions), initial intentions and some predefined optimal plans, the agent for this specification is born and becomes active. The agent will always try to realize all its intentions at its best and may produce new intentions in the real-time. The SICLE interpreter employs a non-monotonic reasoning system, namely an extended logic programming system, to model the real-world, to record the past history of the environment, and to reify the underlying action theory. For those situations which may frequently arise, the user can provide optimal plans to achieve timely response. The SICLE language and its associated architecture and interpreter was designed based on an extended logic programming system and action theory. A simple example of commuter between two cities is used to present our approach.