A systematic approach for detecting faults in agent designs

This thesis proposes a mechanism, including automated tool support, for early-phase defect
detection by comparing the plan structures of a belief-desire-intention (BDI) agent design against
the following: (1) requirement models, specified in terms of scenarios and goals; and (2) agent
communication models. The intuition of our approach is to extract sets of possible behaviour runs
from the agent-behaviour models and to verify whether these runs conform to the specifications
of the system-to-be. The proposed approach in this thesis is applicable at design time and does
not require source code. Our approach is based on the Prometheus agent-design methodology but
is applicable to other methodologies that support the same notions.

We evaluate the proposed verification framework on designs, ranging from student projects
to case studies of industry-level projects. Our evaluation demonstrates that even a simple specification
developed by relatively experienced developers is prone to defects, and our approach is successful in
uncovering most of these defects. In addition, we conduct a scalability analysis of our methods,
and the outcomes reveal that our approach can scale when designs grow in size.