The reduced cost of a nonbasic variable $x$ is the net rate of change to the objective if you increase the value of $x$ by a small amount and optimally adjust the other variables to preserve feasibility. Regardless of whether you are maximizing or minimizing, a positive reduced cost means increasing $x$ will increase the objective value and a negative reduced cost means increasing $x$ will decrease the objective value. The reduced cost of a basic variable is always 0.
Assuming that the problem has been solved to optimality, and assuming that the domain of $x$ is $[0, \infty)$ (typical for most LP formulations), then if $x>0$ in the optimal solution it will be a basic variable, and if $x$ is basic the reduced cost will be 0.
If $x$ has a nonzero lower bound $\ell > 0$ that is expressed as a bound on the variable rather than as a constraint (meaning that the constraint matrix does not include $x\ge \ell$), and if $x=\ell$ in the solution, then it may be nonbasic even though it is positive. The same is true if $x$ has a finite upper bound $u$ expressed as a bound rather than a constraint and $x=u$ in the optimal solution. This is because there are versions of the simplex algorithm that handle a lower bound $\ell$ by substituting $\hat{x}=x - \ell$ for $x$ in the model, and handle an upper bound $u$ by substituting $\hat{x} = u - x$ for $x$ when $x$ hits its upper bound.