By design, the VectorCalculus package computes the differential operators Divergence, Curl, Gradient, and Laplacian only in orthogonal coordinates. It uses scale factors, and that works only with orthogonal systems. However, Maple's plot package can draw graphs for non-orthogonal systems, so there is this option in VectorCalculus to add a non-orthogonal system for the purpose of graphing, but not for computing.
To obtain the gradient in non-orthogonal systems, you would need the Tensor subpackage of the DifferentialGeometry package. In essence, such a gradient must be computed via the covariant derivative, which is the generalization of the directional derivative. In multivariate calculus, one meets the directional derivative of a scalar. When this derivative exists, it can be computed by taking the dot product of the gradient vector with the direction vector. Well, when you ask for a directional derivative of a vector field, you first need to generalize the gradient vector. This generalization is the "covariant derivative." Dotting the covariant derivative with the direction vector then produces the directional deriavtive of the vector field.
So, what does a covariant derivative look like? In this case it would be a rank-2 tensor. This object can be represented as a matrix - its components can be organized into a matrix format. Then, matrix multiplication with a vector can be used to compute the equivalent of the summation process that's lurking in the tensor calculus.
If you really need to learn this stuff, I suggest you read the Reporter Article "Classroom Tips and Techniques: Tensor Calculus with the Differential Geometry Package" that's now in the Maple Application Center at the end of the following link.
Alternatively, go to the Application Center, and search for articles by Robert Lopez. There are at least 50, but the one I'm suggesting is in the list.