$A$ is similar (via a permutation matrix) to $A_1\oplus A_2\oplus\cdots\oplus A_m$, where $m=\lceil\frac n2\rceil$, the first $m-1$ diagonal blocks are $2\times2$ and the last one is $2\times2$ when $n$ is even, or $1\times 1$ when $n$ is odd.
As a block diagonal matrix is diagonalisable if and only if its diagonal subblocks are all diagonalisable, it suffices to consider the problem clockwise and we may ignore the last $1\times1$ block (which is always diagonalisable) if any. So, consider $A_k=\pmatrix{0&\lambda_k\\ \lambda_{n+1-k}&0}$. It is annihilated by $p(x)=x^2-\lambda_k\lambda_{n+1-k}$.
When $\lambda_k\lambda_{n+1-k}\ne0$, $p$ is always splittable into two distinct linear factors over $\mathbb C$, and it splits over $\mathbb R$ if and only if $\lambda_k\lambda_{n+1-k}>0$. Therefore $A_k$ is diagonalisable over $\mathbb R$ if and only if $\lambda_k\lambda_{n+1-k}>0$.
If $\lambda_k\lambda_{n+1-k}=0$, $A_k$ is nilpotent. Hence $A_k$ is diagonalisable over $\mathbb C$ or $\mathbb R$ if and only if it is the zero matrix, i.e. iff $\lambda_k=\lambda_{n+1-k}=0$. In short:
A real anti-diagonal matrix $A$ is diagonalisable over $\mathbb R$ if and only if for every symmetric pair of entries $\{\lambda_k,\lambda_{n+1-k}\}$ on the anti-diagonal, either $\lambda_k\lambda_{n+1-k}>0$ or $\lambda_k=\lambda_{n+1-k}=0$.
Similarly:
A complex anti-diagonal matrix $A$ is diagonalisable over $\mathbb C$ if and only if for every symmetric pair of entries $\{\lambda_k,\lambda_{n+1-k}\}$ on the anti-diagonal, either $\lambda_k\lambda_{n+1-k}\ne0$ or $\lambda_k=\lambda_{n+1-k}=0$.