Vivid agents (Wagner, 1996) are software-controlled systems whose state com
prises the mental components of knowledge, perceptions, tasks, and intentio
ns, and whose behavior is represented by means of action and reaction rules
. An execution model for vivid agents is presented, which is based on an ar
chitecture for concurrent action and planning (CAP). The concept of vivid a
gents and the CAP architecture in distributed diagnosis is evaluated, inclu
ding fault-tolerant diagnosis and the diagnosis of an unreliable communicat
ion protocol.