rugarch VaR calculation "manually"

I want to use all observations for my
fitting and then do one step ahead in sample predictions.

There is no such thing as a one step ahead in-sample  *prediction*.

Please stop and think a moment about how absurd that is before you waste 
more of everyone's time.

At each time t, I may make a prediction for time t+1

ugarchroll will give you a set of rolling one or n step ahead 
*predictions*, as Alexios has also already told you.

If you want the in-sample *estimate* from the fitted model values (it is 
not a prediction), then you want the 'fitted' and 'sigma' values, as 
Alexios has told you at least twice now.

...and I still don't see you signing your name.

Brian

Brian G. Peterson
http://braverock.com/brian/
Ph: 773-459-4973
IM: bgpbraverock