I've been given different answers to this question in different courses. Some professors say it is (using the example alternative hypothesis of $\mu > 3$):
$$H_0: \mu = 3$$ $$H_1: \mu > 3$$
Others say it is:
$$H_0: \mu \le 3$$ $$H_1: \mu > 3$$
How should I write my null hypothesis for 1-tailed tests?
In general, a hypothesis is a statement that a restriction is true, where a restriction takes the form $\theta\in \Theta_0$ with $\Theta_0$ is a strict subset of a parameter space $\Theta$. If the null hypothesis is defined as $$ H_0:\theta\in \Theta_0 $$ the alternative hypothesis is its complement $$ H_1:\theta\in \Theta_0^c. $$ By this formal definition, the second one is correct and the first one does not constitute a hypothesis test unless $\Theta=[3,+\infty)$, which is highly unlikely in application.
PS: The definition comes from Econometrics by Bruce E. Hansen, Chapter 8 .