The proof is based on the following observation: If $A \in \mathrm{M}_{n,m}(R)$ is an $n \times m$ matrix with coefficients $a_{ij}$ such that there exists indices $i,j$ where $a_{ij}$ is a unit, then there exist $B \in \mathrm{GL}_n(R)$ and $C \in \mathrm{GL}_m(R)$ such that
$$
B A C = \begin{bmatrix}
1 & 0 & \dots & 0 \\
0 & * & \dots & * \\
\vdots & \vdots & \ddots & \vdots \\
0 & * & \dots & *
\end{bmatrix}.
$$
To find $B$ and $C$ you simply reorder your basis and then perform Gaussian row and column elimination. Now we have a commutative diagram
$\require{AMScd}$
\begin{CD}
R^m @>A>> R^n\\
@V C^{-1} V V @VV B V\\
R \oplus R^{m - 1} @>>1 \oplus A'> R \oplus R^{n - 1}
\end{CD}
where $A'$ is the lower right block in $B A C$.
Recall that a free resolution $$F_i \xrightarrow{A_i} F_{i-1} \to \dots \xrightarrow{A_1} F_0 \xrightarrow{} M$$ is minimal if and only if all the coefficients of the matrices $A_i$ are contained in $\mathfrak{m}$. If the resolution is not minimal, there exists some matrix coefficient which is not in $\mathfrak{m}$. Since $R$ is local, this coefficient will be a unit and we can apply the above to write
$$
F_\bullet \cong F_\bullet' \oplus (R[l] \xrightarrow{1} R[l - 1])
$$
for some shift $l$.
If $F_\bullet'$ is not minimal we inductively continue until we reach a minimal complex $G_\bullet$. Now $F_\bullet \cong G_\bullet \oplus Q_\bullet$, where $Q_\bullet$ is a direct sum of shifts of $R \xrightarrow{1} R$.