In planning based on hierarchical task networks (HTN), plans are generated by refining high-level actions (‘compound tasks’) into lower-level actions, until primitive actions are obtained that can be sent to execution. While a primitive action is defined by its precondition and effects, a high-level action is defined by zero, one or several methods: sets of (high-level or primitive) actions decomposing it together with a constraint. We give a semantics of HTNs in terms of dynamic logic with program inclusion. We propose postulates guaranteeing soundness and completeness of action refinement. We also show that hybrid planning can be analysed in the same dynamic logic framework.