Running oci ce cluster create-kubeconfig with a non-default --profile produced a kubeconfig whose embedded exec command for generate-token did not include that profile, so kubectl failed with You must be logged in to the server because the token was minted under the wrong credentials. For years the only fix was to hand-edit the kubeconfig and add --profile (and --auth) to the exec args, which broke again every time the file was regenerated. Oracle finally shipped a proper fix in oci-cli v3.52.0: the new --with-auth-context flag makes the CLI carry the profile and auth context from the create-kubeconfig invocation into the generated exec args. Two users on the issue confirmed that upgrading past 3.52.0 and using the flag resolved the problem.