TP-VML is meant as a graphical equivalent of the Task Planning model (or you might think of it as a meta-model, which is also correct). The TP model is defined here.
At the moment, the TP-VML document uses a set of draw.io elements which are not semantic, i.e. they don’t know about the meta-model. However the Better ArchetypeDesigner does, and it has a visual mode, based on the TP (meta-)model.
Things are still moving both at the TP level and also at the level of graphical representation. If you are thinking of supporting it in a tool, please let us know your ideas and needs.