In the original WW paper, we distinguish between "crude" and "topological" boundary conditions. (I think "algebraic" would be a better name than "topological" here, but perhaps it is too late to change the naming conventions. Or perhaps not.)
As you write, we associate a linear category $A(Y)$ to a 2-manifold $Y$. The crude boundary conditions are the objects of $A(Y)$. The topological/algebraic boundary conditions are the representations of $A(Y)$. An elementary excitation is an irreducible representation of $A(Y)$. (In this case $Y$ is the boundary of the excitation -- a sphere in the case of a point excitation, a torus in the case of a loop excitation.)
The boundary of a loop excitation is a torus $T$, so the possible elementary loop excitations correspond to irreducible representations of $A(T)$.
Defects are another sort of boundary condition, and correspond to higher categorical representations. For example, a domain wall type defect between a WW model and the vacuum corresponds to a (higher) representation of the 3-category $A(p)$ associated to a point $p$.
Codimension-2 defects (e.g. loop defects) correspond to representations of the 2-category $A(S)$, where $S$ is a circle. (You should think of $S$ as a small circle which links the loop defect, not the large circle (loop) which describes the location of the loop defect.) For a WW model, the 0-morphisms of $A(S)$ are trivial (only one 0-morphism), the 1-morphisms are ribbon end points in $S\times I$, and the 2-morphisms are linear combinations of string nets in $S\times D$, modulo the usual relations. (Here $D$ is a disk.) Since the 0-morphisms are trivial, this 2-category can be thought of as a tensor category.
In this context, a representation of $A(S)$ is just a module category for $A(S)$.
Every loop defect gives rise to a loop excitation, but not vice-versa.
