Feldman basically says that im situations of high epistemic uncertainty, it is often better to stay agnostic .
In order to do so, we need to judge that the matter at hand features so much epistemic uncertainty that we cannot judge whether it is true or false. Hence, it is a meta- judgement to suspend judgement regarding this particular question, nothing more, nothing less.
While you are correct that in a way, this means that this meta-judgement needs to have more epistemic certainty or would have to be suspended itself, this is no inherent problem:
It may both be possible to say that at least it is certain that it is uncertain what really is the case, and that we just add another layer and say that the particular matter and its epistemic certainty are uncertain, yet at least that is sufficiently certain.