You've done everything correct, except the final bit.
When you sub the eigenvalues into the matrix
A-lambda*I
the matrix is singular, so the eigenvectors cannot be determined uniquely. However, you can assume their magnitude is also unity (for instance, which gives you an extra degree of freedom to determine them.
So for the final step, you get (from either equation)
x_1 = 2/3 * x_2
so the eigenvector is something like
v = a*[2/3, 1]
where a is a scalar.
Requiring the eigenvector to have unity magnituce allows you to get a value for a.
If you think its a bit weird, go back to the definition of an eigenvector / eigenvalue, .i.e
A*v = lambda*v
i.e the matrix product simply scales the eigenvector. If you double the eigenvector, this still holds as
A*2*v = 2*lambda*v
so the eigenvector cannot be determined uniquely. Its direction can, but not its magnitude.