$\def\bbR{\mathbb{R}} \def\frX{\mathfrak{X}} \def\hb{\mathcal{H}om} \def\hom{\operatorname{Hom}}$Let $E$ be a smooth vector bundle over a smooth manifold $M$. A metric in $E$ is a bundle homomorphism $E\otimes E\to M\times\bbR$ which is fiberwise positive-definite. If $E$ has a metric, a connection $\nabla$ on $E$ is said to be metric or compatible with the metric if $$ \nabla_X\langle Y,Z\rangle=\langle\nabla_XY,Z\rangle+\langle Y,\nabla_XZ\rangle $$ for all $Y,Z\in\Gamma(E)$ and $X\in\frX(M):=\Gamma(TM)$. Now, suppose $F$ is another vector bundle over $M$. Denote $\hb(E,F)$ to the hom-bundle of $E$ and $F$, i.e., it is the vector bundle over $M$ whose fiber over $p\in M$ is $\hb(E,F)_p=\hom(E_p,F_p)$. The connections on $E$ and on $F$ induce a connection on the hom-bundle in the following way: for $\varphi\in\Gamma(\hb(E,F))=\hom(E,F)$, we have $$ (\nabla_X\varphi)Y=\nabla_X(\varphi(Y))-\varphi(\nabla_XY), $$ for $X\in\frX(M),Y\in\Gamma(E)$ (details here).
Now suppose that $F$ has a metric. The metrics of $E$ and $F$ induce then a metric on $\hb(E,F)$ which fiberwise is given as here.
I was wondering if the connections on $E$ and $F$ being metric implies that the induced connection on $\hb(E,F)$ is metric as well. Here are my thoughts on this: Pick an orthonormal frame $(E_i)$ of $E$ (we can achieve this by picking first a local frame and then applying Gram-Schmidt. Since the GS formulas are smooth, so is the resulting frame).
Let $\varphi,\psi\in\Gamma(\hb(E,F))$, $X\in\frX(M)$, and compute \begin{align*} \nabla_X\langle\varphi,\psi\rangle &=\sum_i\nabla_X\langle\varphi(E_i),\psi(E_i)\rangle\\ &=\sum_i\langle\nabla_X(\varphi E_i),\psi E_i\rangle +\langle\varphi E_i,\nabla_X(\psi E_i)\rangle\\ &=\sum_i\langle(\nabla_X\varphi)E_i+\varphi(\nabla_X E_i),\psi E_i\rangle +\langle \varphi E_i,(\nabla_X\psi)E_i+\psi(\nabla_X E_i)\rangle\\ &=\sum_i\langle(\nabla_X\varphi)E_i,\psi E_i\rangle+\langle\varphi E_i,(\nabla_X\psi)E_i\rangle\\ &\quad +\sum_i\langle\varphi(\nabla_X E_i),\psi E_i\rangle+\langle\varphi E_i,\psi(\nabla_X E_i)\rangle. \end{align*} The second-to-last line equals $\langle\nabla_X\varphi,\psi\rangle+\langle\varphi,\nabla_X\psi\rangle$, but what about the last line? Does this expression vanish? In case it did, it would suffice to show it for $X=E_j$, by $C^\infty(M)$-linearity in $X$. In this case the last line becomes $$ \sum_i\langle\varphi(\nabla_X E_i),\psi E_i\rangle+\langle\varphi E_i,\psi(\nabla_X E_i)\rangle =\sum_i\Gamma_{ji}^k\left(\langle\varphi E_k,\psi E_i\rangle+\langle\varphi E_i,\psi E_k\rangle\right). $$ But I still regard this a little intractable.
The crucial observation is that there is a natural isomorphism $$\operatorname{Hom}(E,F) \simeq F\otimes E^*.$$ Also, it suffices to define and verify everything for simple tensors $f\otimes \epsilon \in F\otimes E^*$.
If $E$ has an inner product, then there is a naturally induced metric on $E^*$. If $F$ also has an metric, then there is e naturally induced metric on $F\otimes E^*$, where for any $f \in F$ and $\epsilon\in E^*$, $$ \langle f_1\otimes \epsilon_1,f_2\otimes\epsilon_2\rangle = \langle f_1,f_2\rangle\langle \epsilon_1,\epsilon_2\rangle. $$
If $E$ has a connection, then there is a naturally induced connection on $E^*$. If $F$ also has a connection, then there is a naturally induced connection on $F\otimes E^*$, where if $f$ is a section of $F$, $\epsilon$ is a section of $E^*$, and $X \in TM$, $$ \nabla_X(f\otimes\epsilon) = (\nabla_X f)\otimes\epsilon + f\otimes \nabla_X\epsilon. $$
It is now a straightforward calculation to show that if the connections on $E$ and $F$ are compatible with the respective metrics on $E$ and $F$, then, given sections $f_1, f_2$ of $F$ and $\epsilon_1,\epsilon_2$ of $E^*$, and $X \in TM$, \begin{align*} X\langle f_1\otimes \epsilon_1, f_2\otimes \epsilon_2\rangle &= X(\langle f_1,f_2\rangle\langle \epsilon_1,\epsilon_2\rangle)\\ &= \cdots\\ &= \langle\nabla_X(f_1\otimes\epsilon_1),f_2\otimes\epsilon_2\rangle + \langle f_1\otimes\epsilon_1,\nabla_X(f_2\otimes\epsilon_2)\rangle. \end{align*} In other words, the naturally induced connection on $\operatorname{Hom}(E,F)$ is compatible with the naturally induced metric on $\operatorname{Hom}(E,F)$.