[guided]The main issue in the converse direction is not fibrewise bijectivity; the frame hypothesis already gives that. The point is to verify smoothness of both the map and its inverse as maps between manifolds.
Fix $p \in M$ and choose an open neighbourhood $U \subset M$ of $p$ together with a smooth bundle trivialization
\begin{align*}
\tau:\pi^{-1}(U) \to U \times \mathbb{R}^k
\end{align*}
over $U$. This means that $\tau$ identifies every fibre $E_x$ with $\{x\}\times\mathbb{R}^k$ by a linear isomorphism, smoothly in $x$.
Let $\operatorname{pr}_2:U\times\mathbb{R}^k\to\mathbb{R}^k$ denote the second projection. For each frame section $e_i:M \to E$, define its coordinate vector in this trivialization by
\begin{align*}
v_i:U \to \mathbb{R}^k, \qquad v_i(x):=\operatorname{pr}_2(\tau(e_i(x))).
\end{align*}
The map $v_i$ is smooth because $e_i$ is smooth, $\tau$ is smooth, and $\operatorname{pr}_2$ is smooth. Now define
\begin{align*}
A:U &\to \mathbb{R}^{k\times k}
\end{align*}
by declaring that the $i$-th column of $A(x)$ is $v_i(x)$. Thus $A(x)$ records the frame vectors in the local trivialization.
In these coordinates, the map $\Phi$ becomes
\begin{align*}
U \times \mathbb{R}^k \to U \times \mathbb{R}^k, \qquad (x,a) \mapsto (x,A(x)a).
\end{align*}
Indeed, if $a=(a_1,\dots,a_k)$, then linearity of the fibre map induced by $\tau$ gives
\begin{align*}
\tau\left(\sum_{i=1}^k a_i e_i(x)\right)
=
\left(x,\sum_{i=1}^k a_i v_i(x)\right)
=
(x,A(x)a).
\end{align*}
This coordinate expression is smooth because each component is a finite sum of products of smooth functions.
We must also show that the inverse is smooth. For each $x \in U$, the vectors $e_1(x),\dots,e_k(x)$ form a basis of $E_x$. Since $\tau$ is fibrewise linear and invertible, the coordinate vectors $v_1(x),\dots,v_k(x)$ form a basis of $\mathbb{R}^k$. Therefore $A(x)$ is an invertible matrix, so $A(x)\in GL(k,\mathbb{R})$.
The inverse coordinate map must solve $b=A(x)a$ for $a$, so it is
\begin{align*}
U \times \mathbb{R}^k \to U \times \mathbb{R}^k, \qquad (x,b) \mapsto (x,A(x)^{-1}b).
\end{align*}
The entries of $A(x)^{-1}$ are smooth functions of $x$: by the cofactor formula, each entry is a polynomial in the entries of $A(x)$ divided by $\det A(x)$, and $\det A(x)$ never vanishes on $U$. Hence the inverse coordinate map is smooth. Since this argument works around every point $p \in M$, $\Phi$ is a smooth map with smooth inverse globally.[/guided]