In the book An Introduction to Computational Stochastic PDEs, you can find the following assumption:
Doesn't this already imply that $A$ is bounded and hence $\mathcal D(A)=H$? By assumptio, we can write $$A=\sum_{j\in\mathbb N}\lambda_j\phi_j\otimes\phi_j\tag1$$ and the right-hand side should be bounded; at least if $\sum_{j\in\mathbb N}\lambda_j<\infty$; or is this the crucial point?

If $$ A=\sum_{j\in\mathbb N}\lambda_j\phi_j\otimes\phi_j $$ with the $\{\phi_j\}$ orthonormal, then $A$ is bounded if and only if the sequence $\{\lambda_j\}$ is bounded. In fact, $$\tag1 \|A\|=\sup\{\lambda_j:\ j\}. $$ This is easily seen. Indeed, any $\xi\in H$ can be written as $\xi=\sum_jx_j\,\phi_j$. Then, if $c$ is the supremum in $(1)$,
$$ \|A\xi\|^2=\Big\|\sum_j x_j\lambda_j\,\phi_j\Big\|^2=\sum_j|x_j|^2\,\lambda_j^2\leq c^2\,\sum_j\|x_j|^2=c^2\,\|x\|^2. $$ So $\|A\|\leq c$. Conversely, if $\{\lambda_{j_k}\}$ is a subsequence such that $\lambda_{j_k}\to c$, then $$ \|A\phi_{j_k}\|=\lambda_{j_k}\to c, $$ which shows that $\|A\|\geq c$.
The condition $\sum_j\lambda_j<\infty$ is what one calls for $A$ to be trace-class. It does imply that $A$ is bounded, but there are many bounded operators which are not trace-class.
It is also important to know that a bounded operator that has an orthonormal basis of eigenvectors is normal, and selfadjoint if the eigenvalues are real.