For a one-sided significance test comparing two means with small samples from highly skewed population distributions, it’s safer to use a t test than a Wilcoxon test. This is because the Wilcoxon test assumes normal population distributions and is not robust if that assumption is violated.