
- playback
  - compare reponse content.  use Algorithm-Diff
  - profiling... where are all the cycles going?

- log evaluation
  - write it to match user-centric
  - response code counts per url, global

- documentation
  - finish _script_hacks_ file
  - write something on how to use dlg_proxy
  - write something on how to assemble an attack

- allow for more than one text regexp for a single URL regexp in
  the V_pos_*/V_neg_* hashes (duh, don't use hashes).

- recording should probably lock the script file while writing.  if
  any of the events start plowing through each other, it'll have to
  be done.

- User.pm is big and messy.  there's too much happening in it; it
  should probably be broken into several different classes.  Maybe
  a userdef class for the stuff from the config file, maybe a proxy
  class.  anything else?

- an "include" directive in the configuration file would be nice, so
  that you could put all your user definitions in different files, and
  just "include" them into the top level dcf.

- check for the existance of the log file, and bail out if it already
  exists.  i've overwritten too many useful log files this way.

- get Parallel::UserAgent to do secure requests, like it says it does.

- needs Makefile for installation.  this means you'll have to fix
  the -I stuff at the top of all the files for now.

- verify off-site links, but don't parse.

- parallel lists are so LISP.  be more perl5.

- make config file parser more elegant
  - tags should have data flags with them to lose the multiple quasi-switches

- make a second proxy server that acts as a live dlg_eval.  calling up a
  url in the log file creates a page where the requested url is in a
  subframe, and surrounding it are the stats from the logfile for that
  page.

