Suppose that $\limsup(x_n+y_n)>\limsup(x_n)+\limsup(y_n)$. So there exists $\varepsilon>0$ such that $\limsup(x_n+y_n)>\limsup(x_n)+\limsup(y_n)+\varepsilon$, and by definition, there exists a subsequence $(x_{\varphi(n)}+y_{\varphi(n)})_n$ of $(x_n+y_n)$ which converges to some $\ell>\limsup(x)+\limsup(y)+\varepsilon$. This implies that for $n$ large enough, $x_{\varphi(n)}+y_{\varphi(n)}>\limsup(x)+\limsup(y)+\varepsilon$. But the latter implies that for $n$ large enough, either $x_{\varphi(n)}>\limsup(x)+\frac12\varepsilon$ or $y_{\varphi(n)}>\limsup(y)+\frac12\varepsilon$ (if none of those two inequalities held, we would not have the former one).
We deduce that either there exist infinitely many integers $n$ such that $x_{\varphi(n)}>\limsup(x)+\frac12\varepsilon$, or there exist infinitely many integers $n$ such that $y_{\varphi(n)}>\limsup(y)+\frac12\varepsilon$. Without loss of generality we can suppose that $x_{\varphi(n)}>\limsup(x)+\frac12\varepsilon$ holds for infinitely many integers $n$. But that would mean that $\limsup((x_{\varphi(n)})_n)\ge\limsup(x_n)_n+\frac12\varepsilon$, which is nonsense since we obviously have $\limsup((x_{\varphi(n)})_n)\le\limsup(x_n)$ (any subsequence of $(x_{\varphi(n)})_n$ is a subsequence of $(x_n)_n$).