> I want to download documents from a website, where
> references are dynamicaly created by htlml forms like
> <form action=... method=...
> ....
> <input name=... value=

You can parse multiple forms generated by the "capture 
URL" feature, but forms can not be parsed 
automatically.
The main reason is that is is very difficult to ask 
the user to fill out all fields. Many fields are hard 
to parse, too, like image submit, file input types and 
so on.. but this feature may be added in the future!