We describe a model-based construction of an online tester for black-box testing of the implementation under test(IUT). The external behaviour of the IUT is modelled as an output observable nondeterministic EFSM with the assumption that all transition paths are feasible. A test purpose is attributed to the IUT model by a set of Boolean variables called traps that are used to measure the progress of the test run. These variables are associated with the transitions of the IUT model. The situation where all traps have been reached means that the test purpose has been achieved. We present a way to construct a tester that at runtime selects a suboptimal test path from trap to trap by finding the shortest path to the next unvisited trap. The principles of reactive planning are implemented in the form of the decision rules of selecting the shortest paths at runtime. The decision rules are constructed in advance from the IUT model and the test purpose. Preliminary experimental results confirm that this method clearly out performs random choice and is better than the anti-ant algorithm in terms of resultant test sequence length to achieve the test purpose