Ginkgo is the testing framework used for Go on Codewars. It's rather the contrary: don't use reflect when the framework can do the job.

In the solution, var output []int would be my pick. We don't need to allocate if there's nothing to do. See https://github.com/golang/go/wiki/CodeReviewComments#declaring-empty-slices.

Test 4 should test len(output) == 0 since either []int{} or nil would fit the criteria (nil would be more idiomatic). Failing the hidden test case for returning nil instead of []int{} would be very annoying.

Go comes with testing built-in, the use of an external testing library seems unnecessary. reflect.DeepEqual(want, got) would be enough for Tests 1-3, and len(got) == 0 for Test 4, or the negation of these to signal test failure.

Two alternate ways to write the loop:

	for _, val := range l {
		if n, ok := val.(int); ok {
			output = append(output, n)
		}
	}

	for _, val := range l {
		switch val := val.(type) {
		case int:
			output = append(output, val)
		}
	}