Let $x_1, \dots, x_N$ be a finite collection of distinct points on the unit sphere $\mathbb{S}^{n-1} \subseteq \mathbb{R}^n$. Consider the convex hull $C$ of these points, defined to be $$ C = \{ \sum^N_{j = 1} \lambda_j x_j : \lambda_j \ge 0, \, \sum^N_{j = 1} \lambda_j = 1 \} \subseteq \{x \in \mathbb{R}^n : \|x\| \le 1\}. $$
I would like to prove that $C \cap \mathbb{S}^{n-1} = \{x_1, \dots, x_N \}.$
When $N = 2$, there is a nice proof of this by finding the maxima the function $$ [0, 1] \ni t \mapsto \|tx_1 + (1-t)x_2 \|^2, $$ where $\| \cdot \|$ is the usual norm on $\mathbb{R}^n$. But I am unsure how to generalize to several points. Hints or solutions are greatly appreciated.
Let me slightly improve your case $N=2$. We will show that if $\lVert x\rVert=1$ and $\lVert y \rVert\leq 1$, then $\lVert tx+(1-t)y\rVert<1$ for $t \in (0, 1)$. In fact, $$ f(t)=\lVert tx+(1-t)y\rVert^2 =t^2+(1-t)^2\lVert y\rVert^2 + 2t(1-t)\langle x, y\rangle. $$ Just as in the post you quoted, the leading term is $t^2(1+\lVert y\rVert^2-2\langle x, y\rangle)$, which we can bound by Cauchy-Schwarz: $$ 1+\lVert y\rVert^2-2\langle x, y\rangle \geq 1-\lVert y\rVert^2\geq 0. $$ Therefore, if $\lVert y\rVert<1$, the map $f(t)$ is a quadratic function with positive leading term, and so its maximum is attained at the boundary. If we have equality, then it follows by the discussion in your linked answer.
This way, we can show inductively on $n\geq 2$ that $\lVert\sum_{j=1}^n \lambda_j x_j\rVert<1$ for $\lambda_j> 0, \sum_j \lambda_j=1$ (note the strict inequality!). The base case is already covered. For the induction step, write $$ \lambda_1 x_1+\dots+\lambda_{n+1}x_{n+1} = (1-\lambda_{n+1}) \left( \mu_1x_1+\dots+\mu_n x_n \right) + \lambda_{n+1}x_{n+1}, $$ where $\mu_i=\lambda_i/(1-\lambda_{n+1})$. The induction step shows that $\lVert \mu_1x_1+\dots+\mu_n x_n\rVert<1$, and so we conclude. This shows that the only elements in $C$ with norm 1 are the ones with $n=1$, as required