[guided]The reverse direction is the substantial content of the theorem. We need to show that if the winding number of $\gamma$ around every point outside $\Omega$ vanishes, then $\gamma$ integrates every closed holomorphic $1$-form to zero.
The strategy uses de Rham cohomology: we show that the winding number forms $dz/(z - z_j)$ — one for each "hole" in $\Omega$ — form a basis for the space of closed $1$-forms modulo exact $1$-forms. Once we know this, any closed $1$-form can be decomposed into a linear combination of winding number forms plus an exact form. The exact form integrates to zero over any closed curve, and each winding number form integrates to $2\pi i \cdot n(\gamma, z_j) = 0$ by hypothesis. So the total integral is zero.
Let $\{C_j\}_{j \in J}$ be the bounded connected components of $\mathbb{C} \setminus \Omega$. (The unbounded component does not contribute a generator because it is "at infinity.") Choose representative points $z_j \in C_j$ and define
\begin{align*}
\omega_j := \frac{1}{2\pi i} \frac{dz}{z - z_j}, \quad j \in J.
\end{align*}
Why do these form a basis for $H^1_{\mathrm{dR}}(\Omega)$? The topological reason is that $\Omega$, being an open subset of $\mathbb{C}$, deformation retracts onto a graph. Its first homology $H_1(\Omega; \mathbb{Z})$ is free abelian with one generator for each bounded complementary component — geometrically, each "hole" contributes one independent cycle. De Rham's theorem gives $\dim H^1_{\mathrm{dR}}(\Omega) = \operatorname{rank} H_1(\Omega; \mathbb{Z}) = |J|$.
To verify linear independence, choose for each $k \in J$ a small circle $\sigma_k$ in $\Omega$ encircling $z_k$ once. The pairing gives
\begin{align*}
\oint_{\sigma_k} \omega_j = n(\sigma_k, z_j) = \delta_{jk}.
\end{align*}
This is a non-degenerate pairing (an identity matrix), which forces the $[\omega_j]$ to be linearly independent. The point $z_j$ for $j \neq k$ lies in a connected component of $\mathbb{C} \setminus \Omega$ different from $C_k$. Since $\sigma_k$ is a small circle around $z_k$, all of $C_j$ (for $j \neq k$) lies in the unbounded component of $\mathbb{C} \setminus \operatorname{im}(\sigma_k)$, so by the [Local Constancy of the Winding Number](/theorems/3360), $n(\sigma_k, z_j) = 0$.
To verify they span: take any closed holomorphic $1$-form $\omega$ on $\Omega$. Compute its periods $a_k := \oint_{\sigma_k} \omega$. Then $\omega - \sum_j 2\pi i\, a_j\, \omega_j$ has zero period around every $\sigma_k$, hence zero integral around every cycle in $\Omega$ (since the $\sigma_k$ generate $H_1$). A closed $1$-form with zero periods on all cycles is exact (by de Rham's theorem). So $[\omega] = \sum_j 2\pi i\, a_j\, [\omega_j]$ in $H^1_{\mathrm{dR}}(\Omega)$.
Now we can decompose any closed holomorphic $1$-form $\omega$ on $\Omega$ as
\begin{align*}
\omega = \sum_{j \in J} c_j \cdot \frac{dz}{z - z_j} + df
\end{align*}
for some constants $c_j \in \mathbb{C}$ and some holomorphic function $f: \Omega \to \mathbb{C}$ (the primitive of the exact remainder). Integrating over $\gamma$:
\begin{align*}
\oint_\gamma \omega = \sum_{j \in J} c_j \oint_\gamma \frac{dz}{z - z_j} + \oint_\gamma df.
\end{align*}
The second term vanishes because $\gamma$ is a closed curve: $\oint_\gamma df = f(\gamma(b)) - f(\gamma(a)) = 0$ since $\gamma(a) = \gamma(b)$.
For the first sum, each integral gives $\oint_\gamma \frac{dz}{z - z_j} = 2\pi i \cdot n(\gamma, z_j)$. By hypothesis, $n(\gamma, z_j) = 0$ for each $j$ (since $z_j \in C_j \subset \mathbb{C} \setminus \Omega$). Therefore
\begin{align*}
\oint_\gamma \omega = \sum_{j \in J} c_j \cdot 2\pi i \cdot 0 + 0 = 0.
\end{align*}
Since $\omega$ was arbitrary, $\gamma$ is null-homologous in $\Omega$.[/guided]