I think a point you may be missing here is that the indices $a, b, \dots$ are not mere labels, they actually conveniently label the representations involved and can be used to construct manifest invariants of the local Lorentz group. This point is sometimes assumed without emphasizing or explaining, so I thought I'd give you the explanation.
The local Lorentz algebra $\mathfrak{so}_{3,1}$ is a subgroup of $\mathfrak{gl}_4$, the Lie algebra of $4 \times 4$ matrices with the Lie bracket given by the commutator of the matrices. Therefore, representations of $\mathfrak{gl}_4$ are by extension also representations of $\mathfrak{so}_{3,1}$.
As a special case, if something is $\mathfrak{gl}_4$-invariant (that is, belongs to the trivial representation), it must also be $\mathfrak{so}_{3,1}$-invariant. The converse is not true, however. There are objects such as $\eta_{ab}$ and $\varepsilon_{abcd}$ for which $\mathfrak{so}_{3,1}$-invariance must be carefully demonstrated, as they are not $\mathfrak{gl}_4$-invariant.
This formalism is convenient, because we can use the tensor product algebra of the representations of $\mathfrak{gl}_4$ instead of working with the representations of $\mathfrak{so}_{3,1}$. It is easier to work with, because we can use the index notation that is so familiar to physicists. Whenever you see the up/down latin index such as e.g. in ${F^a}_b$, the notation signals which representation of $\mathfrak{gl}_4$ the object belongs to, and therefore it also fixes its $\mathfrak{so}_{3,1}$ representation.
As for $\mathfrak{so}_{3,1}$-invariant objects that are not also $\mathfrak{gl}_4$-invariant, we can still make use of them under this formalism. We can use them in index contraction, such as e.g.
$$ \eta_{ab} e^a_{\mu} e^b_{\nu} $$
the result would be $\mathfrak{gl}_4$-invariant if $\eta_{ab}$ was a (2,0)-tensor under $\mathfrak{gl}_4$, but it isn't, so the contraction is not invariant under $\mathfrak{gl}_4$. However, it is trivial to show that it is invariant under $\mathfrak{so}_{3,1}$. To do that, define an auxiliary quantity $g_{ab}$ which is defined to be equal to $\eta_{ab}$ in some basis, and transforms under $\mathfrak{gl}_4$ as a (2,0)-tensor by definition. Then observe that
$$ g_{ab} e^a_{\mu} a^b_{\nu} $$
is now a $\mathfrak{gl}_4$-invariant, and so also a $\mathfrak{so}_{3,1}$-invariant. But all actions of the $\mathfrak{so}_{3,1}$ subalgebra preserve the (non-tensorial) relation $g_{ab} = \eta_{ab}$, therefore under the action of the whole subgroup generated by the Lorentz subalgebra, the quantities $g_{ab} e^a_{\mu} e^b_{\nu}$ and $\eta_{ab} e^a_{\mu} e^b_{\nu}$ must coincide. We conclude that $\eta_{ab} e^a_{\mu} e^b_{\nu}$ is a $\mathfrak{so}_{3,1}$-invariant.
By the same logic,
$$ (\star F)_{ab} \wedge e^a \wedge e^b = \frac{1}{2} \varepsilon_{abcd} \eta^{c e} \eta^{d f} F_{ef} \wedge e^a \wedge e^b $$
is a full contraction of $\mathfrak{gl}_4$ tensors and objects that are invariant under $\mathfrak{so}_{3,1}$ that we can replace with $\mathfrak{gl}_4$ tensors and then show that they coincide under the action of the local Lorentz group. Hence, it is manifestly Lorentz-invariant.