This article shows how advice can be enhanced in Intelligent Tutoring Systems (ITSs). Our approach is based on the systematic exploitation of two relations in task graphs : part-whole and temporal precedence. The first relation describes the decomposition of tasks into sub-tasks, and distinguishes concrete actions from abstract tasks. The second relation describes temporal constraints between tasks. The underlying reasoning mechanism can easily communicate any given task to the human learner. Moreover it performs the analysis of his/her actions during the solution of a problem. This analysis is used to generate relevant advice in a tutoring context. The mechanism is general and can be applied to arbitrary task graphs, thereby endowing the ITS with a non trivial advice-generating capacity.