I'm a tad bit confused what you are asking, but I'll answer what I think you might be asking.
I believe the original ejbdoclet application, on which the current XDoclet is based, was based on javadoc doclets. For those that don't know, doclets are plugins to javadoc that let you do things other than standard html generation from javadoc. However, XDoclet is not based on doclets now. The javadoc framework was too limited, so xjavadoc was created as a more suitable front end for extracting the javadoc-like attributes from source files.
However, even as a module developer you won't see much much of xjavadoc. Modules are written by developing XDoclet tasks and subtasks which are invoked from
ant to perform the code generation. Code generation happens by means of templates. Think of templates a
JSP files with lots of tag libraries that let you inspect classe and their tags and do your code generation.
Templates are generated in a context. They can be used in the context of a single class, which we call transform generation. This is when you take one class and it's metadata and generate one other class. (like generting a home interface from an EJB) Or, the template can be run in the context of a set of
Java classes. (like generating a web.xml from all the
servlets, filters, taglibs, etc... in your project)