I use the R functions lassoCV and ridgeCV to train models as described in the documentation and vignette, but I am still unsure how to evaluate it.

Both functions use cross-validation and calculate MSE/RMSE, SEP etc. Can I use these MSE/RMSE, SEP etc. values directly for evaluation or should I use e.g. just 90% of the data for lassoCV and ridgeCV, train the models and than in a second step: extract the best model and evaluate it on the remaining 10% of the data again?

To me, the description looks like the use of just lassoCV and ridgeCV is sufficient. Therefore it's not necessary separating the data into training and test data, since this is already done by lassoCV and ridgeCV. But, am I right/is this really the truth? Do the resulting MSE/RMSE etc. values really come from just training data? Respectively is the training strictly/sufficient separated from the evaluation?