I think the problems associated with defining "imperative" non-circularly are sufficiently large that they preclude answering the question. It is presumably clear that meaning and usage can't distinguish imperatives from other constructions such as subjunctives. Imperatives are not definable morphologically (there are not always bare verbs -- see for instance Sanskrit with imperatives inflected for all persons, Finnish with imperatives for all persons besides 1s). They are also not definable as "the verb forms that you get just in case there is no overt subject" -- there are plenty of pro-drop languages which don't require overt subjects (pronouns or NPs).
We often take the reflexivization facts to be sufficient evidence that the subject of an imperative must be 2nd person in English. This argument is weakened a bit by famous Dong 1967 examples like "Fuck/screw/forget/damn you" (where you can dispose of the problem by saying they aren't imperatives), and by "Everyone make yourself a sandwich" where we have an overt subject of an imperative that doesn't control reflexivization. Since subjects generally control verb agreement in English, Spanish, Greek, Sanskrit, Finnish and so on, we ought to conclude that if there is any subject agreement at all, we have proof that there is a covert subject.
Here is an imaginary scenario that could detect a subjectless imperative. Gwambomambo has a single bare-verb form "imperative" that doesn't indicate subject. It has obligatory person-and-number agreeing reflexive object pronouns. But in Gwambomambo, no reflexive can be the object of an imperative, instead there is also a special imperative object reflexive pronoun -- a form that doesn't agree with anything, which is selected because there isn't anything to agree with (no subject). If you could find this language, it could establish that the claim about there being covert subjects of imperatives is not universally true.