[guided]The goal is to turn local frame sections into global sections without losing smoothness. The obstruction is that $\sigma_{j,a}: U_j \to E|_{U_j}$ is only defined on $U_j$, so we first multiply it by a smooth function that vanishes before reaching the boundary of $U_j$.
By the smooth partition of unity theorem subordinate to a finite open cover (citing a result not yet in the wiki: Smooth Partition of Unity Subordinate to an Open Cover), applied to the finite open cover $X = \bigcup_{j=1}^m U_j$, there exist smooth functions
\begin{align*}
\psi_j: X &\to [0,1],
\qquad j \in \{1, \dots, m\},
\end{align*}
with $\operatorname{supp}\psi_j \subset U_j$ and
\begin{align*}
\sum_{j=1}^m \psi_j(x) = 1
\end{align*}
for every $x \in X$.
For each $x \in X$, let $0_{E_x}$ denote the zero vector in the fibre $E_x$. For each $j$ and $a$, define
\begin{align*}
s_{j,a}: X &\to E
\end{align*}
as follows. If $x \in U_j$, set
\begin{align*}
s_{j,a}(x) = \psi_j(x)\sigma_{j,a}(x).
\end{align*}
If $x \in X \setminus \operatorname{supp}\psi_j$, set
\begin{align*}
s_{j,a}(x) = 0_{E_x}.
\end{align*}
Why is this a legitimate global definition? Since $\operatorname{supp}\psi_j \subset U_j$, the function $\psi_j$ vanishes on the open set $X \setminus \operatorname{supp}\psi_j$, and every point outside $U_j$ is certainly outside $\operatorname{supp}\psi_j$. Thus there is no point where we need to evaluate $\sigma_{j,a}$ outside its domain while $\psi_j$ is nonzero.
On $U_j$, the expression $x \mapsto \psi_j(x)\sigma_{j,a}(x)$ is smooth because scalar multiplication in a smooth vector bundle is smooth, $\psi_j$ is a smooth real-valued function, and $\sigma_{j,a}$ is a smooth local section. On $X \setminus \operatorname{supp}\psi_j$, the section is the smooth zero section. These two smooth local descriptions agree on the overlap, because there $\psi_j = 0$. Hence the gluing property for smooth sections gives $s_{j,a} \in \Gamma(E)$.[/guided]