diff --git a/commit-patch b/commit-patch
index cf88be2..09dba3f 100755
--- a/commit-patch
+++ b/commit-patch
@@ -98,3 +98,116 @@ if ($@) {
     die "Failed: $@";
 }
 foreach my $k (keys %clean) { unlink $k }
+
+
+=head1 NAME
+
+commit-patch - commit patches to I<CVS>, I<Darcs> or I<Mercurial> repositories
+
+=head1 SYNOPSIS
+
+commit-patch [B<-m> I<message>] I<patch-file>
+
+=head1 DESCRIPTION
+
+Normally version control systems don't allow fine grained
+commits. B<commit-patch> allows the user to control I<exactly> what
+gets committed (or "recorded", in I<Darcs> parlance) by letting the user
+supply a patch to be committed rather than using the files in the
+current working directory.
+
+B<commit-patch> currently supports the following version control systems:
+B<I<CVS>>, B<I<Darcs>>, and B<I<Mercurial>>.
+
+=head1 OPTIONS
+
+B<-m>, B<--message>=I<message> - An optional I<message> to use as the commit
+text. If the message is multiple lines then I<Mercurial> and I<Darcs>
+will use the first line as the patch name and the rest as commit
+details. If the C<-m> option is not specified then the result will be
+the same as whatever the underlying version control system would do if
+you didn't specify a message name on the command line. That is,
+B<commit-patch> does not interfere with the patch naming process of
+the underlying version control system; I<Darcs> will still ask you
+interactively and I<CVS> will still launch your editor.
+
+=head1 DIAGNOSTICS
+
+B<commit-patch> works by manipulating the working directory using
+C<patch>, C<interdiff>, and the underlying version control system's
+C<diff>.  If any part of the process fails, B<commit-patch> will
+attempt to restore the working directory to the state it was before
+the command was run. Any errors from the underlying version control
+system or from patch will be printed.
+
+=head1 CAVEATS
+
+The patch specified on the command line must originate from the same
+place as the current directory. That is, the following will not work:
+
+  cvs diff -u > ../a.patch
+  cd ..
+  commit-patch a.patch
+
+You B<must> run B<commit-patch> from the same directory that the
+original patch was based from.
+
+I<Darcs> and I<Mercurial> put C<a/> and C<b/> in front of all the paths
+in the diff output. Don't worry about this; B<commit-patch> takes it into
+account.
+
+=head1 EXAMPLES
+
+Typical I<CVS> usage:
+
+  cvs diff -u > a.patch
+  emacs a.patch
+  commit-patch a.patch
+
+I<Mercurial> usage with a message specified:
+
+  hg diff > a.patch
+  emacs a.patch
+  commit-patch -m "This is a commit message" a.patch
+
+I<Darcs> usage with a multi-line message specified:
+
+  darcs diff -u > a.patch
+  emacs a.patch
+  commit-patch -m 'This is the patch name
+  Here are the patch details' a.patch
+
+=head1 AUTHORS
+
+=over
+
+=item *
+
+David Caldwell <david@porkrind.org>
+
+=item *
+
+Jim Radford <radford@blackbean.org>
+
+=back
+
+=head1 COPYRIGHT AND LICENSE
+
+Copyright 2003-2006 by David Caldwell and Jim Radford.
+
+B<commit-patch> is distributed under the GNU General Public
+License. See the COPYING file in the distribution for more details.
+
+=head1 HISTORY
+
+B<commit-patch> was originally called C<cvs-commit-patch> and was a
+bash script written in 2003 by Jim Radford (with David Caldwell in the
+room drawing the procedure on a white board). David later converted it
+do C<darcs-commit-patch>, then integrated them back together into
+B<commit-patch>. I<Mercurial> support was then added. At some point
+David translated from bash into perl because funky bash quoting issues
+were causing problems with a repository that had a space in one of the
+directory names.
+
+=cut
+