The first-order necessary condition for the constrained problem $\underset{x \in \Omega}{\min} f(x)$ is $$\nabla f(x^*)^Tp \geq 0$$ for all feasible directions $p$ at the solution $x^*$. Intuitively it makes sense that there are no descent directions at $x^*$.
It's not clear to me why the above condition reduces to $\nabla f(x^*)^Tp = 0$ for linear constraints $\Omega = \{x :Ax=b \}$. Could you clarify the intermediate steps? Why should the gradient be orthogonal to all feasible $p$?