Automatic conversion of derived classes from Eigen vector/matrix types to numpy array

Consider the following class inherited from Eigen::VectorXd that prevents client code to resize it:

class VectorXdConstSize : public VectorXd
{
public:
    using VectorXd::VectorXd;

    VectorXdConstSize() = delete;

    template<typename Derived>
    auto operator=(const Eigen::MatrixBase<Derived>& other) -> VectorXdConstSize&
    {
        Ref<VectorXd> aux(*this);
        aux = other;
        return *this;
    }
    
    auto resize(Index) -> void = delete;
    
    operator Ref<VectorXd>() { return *this; }
    operator Ref<const VectorXd>() const { return *this; }
};

How can I get pybind11 to automatically cast this type to numpy array, the same way it already works for Eigen types (e.g., VectorXd, Ref, etc.)?

No files or tests are named. Start by locating pybind11's existing Eigen-to-NumPy conversion path and compare how it handles VectorXd and Ref with the shown derived class. Done means the derived Eigen vector or matrix class is automatically cast to a NumPy array without requiring a separate binding.