diff --git a/tools/compare_codereview.py b/tools/compare_codereview.py
new file mode 100755
index 0000000..16f3a01
--- /dev/null
+++ b/tools/compare_codereview.py
@@ -0,0 +1,387 @@
+#!/usr/bin/python2
+
+# Copyright 2014 Google Inc.
+#
+# Use of this source code is governed by a BSD-style license that can be
+# found in the LICENSE file.
+
+"""Skia's Chromium Codereview Comparison Script.
+
+This script takes two Codereview URLs, looks at the trybot results for
+the two codereviews and compares the results.
+
+Usage:
+  compare_codereview.py CONTROL_URL ROLL_URL
+"""
+
+import collections
+import os
+import re
+import sys
+import urllib2
+import HTMLParser
+
+
+class CodeReviewHTMLParser(HTMLParser.HTMLParser):
+    """Parses CodeReview web page.
+
+    Use the CodeReviewHTMLParser.parse static function to make use of
+    this class.
+
+    This uses the HTMLParser class because it's the best thing in
+    Python's standard library.  We need a little more power than a
+    regex.  [Search for "You can't parse [X]HTML with regex." for more
+    information.
+    """
+    # pylint: disable=I0011,R0904
+    @staticmethod
+    def parse(url):
+        """Parses a CodeReview web pages.
+
+        Args:
+            url (string), a codereview URL like this:
+                'https://codereview.chromium.org/?????????'.
+
+        Returns:
+            A dictionary; the keys are bot_name strings, the values
+            are CodeReviewHTMLParser.Status objects
+        """
+        parser = CodeReviewHTMLParser()
+        try:
+            parser.feed(urllib2.urlopen(url).read())
+        except (urllib2.URLError,):
+            print >> sys.stderr, 'Error getting', url
+            return None
+        parser.close()
+        return parser.statuses
+
+    # namedtuples are like lightweight structs in Python.  The low
+    # overhead of a tuple, but the ease of use of an object.
+    Status = collections.namedtuple('Status', ['status', 'url'])
+
+    def __init__(self):
+        HTMLParser.HTMLParser.__init__(self)
+        self._id = None
+        self._status = None
+        self._href = None
+        self._anchor_data = ''
+        self._currently_parsing_trybotdiv = False
+        # statuses is a dictionary of CodeReviewHTMLParser.Status
+        self.statuses = {}
+
+    def handle_starttag(self, tag, attrs):
+        """Overrides the HTMLParser method to implement functionality.
+
+        [[begin standard library documentation]]
+        This method is called to handle the start of a tag
+        (e.g. <div id="main">).
+
+        The tag argument is the name of the tag converted to lower
+        case. The attrs argument is a list of (name, value) pairs
+        containing the attributes found inside the tag's <>
+        brackets. The name will be translated to lower case, and
+        quotes in the value have been removed, and character and
+        entity references have been replaced.
+
+        For instance, for the tag <A HREF="http://www.cwi.nl/">, this
+        method would be called as handle_starttag('a', [('href',
+        'http://www.cwi.nl/')]).
+        [[end standard library documentation]]
+        """
+        attrs = dict(attrs)
+        if tag == 'div':
+            # We are looking for <div id="tryjobdiv*">.
+            id_attr = attrs.get('id','')
+            if id_attr.startswith('tryjobdiv'):
+                self._id = id_attr
+        if (self._id and tag == 'a'
+            and 'build-result' in attrs.get('class', '').split()):
+            # If we are already inside a <div id="tryjobdiv*">, we
+            # look for a link if the form
+            # <a class="build-result" href="*">.  Then we save the
+            # (non-standard) status attribute and the URL.
+            self._status = attrs.get('status')
+            self._href = attrs.get('href')
+            self._currently_parsing_trybotdiv = True
+            # Start saving anchor data.
+
+    def handle_data(self, data):
+        """Overrides the HTMLParser method to implement functionality.
+
+        [[begin standard library documentation]]
+        This method is called to process arbitrary data (e.g. text
+        nodes and the content of <script>...</script> and
+        <style>...</style>).
+        [[end standard library documentation]]
+        """
+        # Save the text inside the <a></a> tags.  Assume <a> tags
+        # aren't nested.
+        if self._currently_parsing_trybotdiv:
+            self._anchor_data += data
+
+    def handle_endtag(self, tag):
+        """Overrides the HTMLParser method to implement functionality.
+
+        [[begin standard library documentation]]
+        This method is called to handle the end tag of an element
+        (e.g. </div>).  The tag argument is the name of the tag
+        converted to lower case.
+        [[end standard library documentation]]
+        """
+        if tag == 'a' and self._status:
+            # We take the accumulated self._anchor_data and save it as
+            # the bot name.
+            bot = self._anchor_data.strip()
+            stat = CodeReviewHTMLParser.Status(status=self._status,
+                                               url=self._href)
+            if bot:
+                # Add to accumulating dictionary.
+                self.statuses[bot] = stat
+            # Reset state to search for the next bot.
+            self._currently_parsing_trybotdiv = False
+            self._anchor_data = ''
+            self._status = None
+            self._href = None
+
+
+class BuilderHTMLParser(HTMLParser.HTMLParser):
+    """parses Trybot web pages.
+
+    Use the BuilderHTMLParser.parse static function to make use of
+    this class.
+
+    This uses the HTMLParser class because it's the best thing in
+    Python's standard library.  We need a little more power than a
+    regex.  [Search for "You can't parse [X]HTML with regex." for more
+    information.
+    """
+    # pylint: disable=I0011,R0904
+    @staticmethod
+    def parse(url):
+        """Parses a Trybot web page.
+
+        Args:
+            url (string), a trybot result URL.
+
+        Returns:
+            An array of BuilderHTMLParser.Results, each a description
+            of failure results, along with an optional url
+        """
+        parser = BuilderHTMLParser()
+        try:
+            parser.feed(urllib2.urlopen(url).read())
+        except (urllib2.URLError,):
+            print >> sys.stderr, 'Error getting', url
+            return []
+        parser.close()
+        return parser.failure_results
+
+    Result = collections.namedtuple('Result', ['text', 'url'])
+
+    def __init__(self):
+        HTMLParser.HTMLParser.__init__(self)
+        self.failure_results = []
+        self._current_failure_result = None
+        self._divlevel = None
+        self._li_level = 0
+        self._li_data = ''
+        self._current_failure = False
+        self._failure_results_url = ''
+
+    def handle_starttag(self, tag, attrs):
+        """Overrides the HTMLParser method to implement functionality.
+
+        [[begin standard library documentation]]
+        This method is called to handle the start of a tag
+        (e.g. <div id="main">).
+
+        The tag argument is the name of the tag converted to lower
+        case. The attrs argument is a list of (name, value) pairs
+        containing the attributes found inside the tag's <>
+        brackets. The name will be translated to lower case, and
+        quotes in the value have been removed, and character and
+        entity references have been replaced.
+
+        For instance, for the tag <A HREF="http://www.cwi.nl/">, this
+        method would be called as handle_starttag('a', [('href',
+        'http://www.cwi.nl/')]).
+        [[end standard library documentation]]
+        """
+        attrs = dict(attrs)
+        if tag == 'li':
+            # <li> tags can be nested.  So we have to count the
+            # nest-level for backing out.
+            self._li_level += 1
+            return
+        if tag == 'div' and attrs.get('class') == 'failure result':
+            # We care about this sort of thing:
+            # <li>
+            #   <li>
+            #     <li>
+            #       <div class="failure result">...</div>
+            #     </li>
+            #   </li>
+            #   We want this text here.
+            # </li>
+            if self._li_level > 0:
+                self._current_failure = True  # Tells us to keep text.
+            return
+
+        if tag == 'a' and self._current_failure:
+            href = attrs.get('href')
+            # Sometimes we want to keep the stdio url.  We always
+            # return it, just in case.
+            if href.endswith('/logs/stdio'):
+                self._failure_results_url = href
+
+    def handle_data(self, data):
+        """Overrides the HTMLParser method to implement functionality.
+
+        [[begin standard library documentation]]
+        This method is called to process arbitrary data (e.g. text
+        nodes and the content of <script>...</script> and
+        <style>...</style>).
+        [[end standard library documentation]]
+        """
+        if self._current_failure:
+            self._li_data += data
+
+    def handle_endtag(self, tag):
+        """Overrides the HTMLParser method to implement functionality.
+
+        [[begin standard library documentation]]
+        This method is called to handle the end tag of an element
+        (e.g. </div>).  The tag argument is the name of the tag
+        converted to lower case.
+        [[end standard library documentation]]
+        """
+        if tag == 'li':
+            self._li_level -= 1
+            if 0 == self._li_level:
+                if self._current_failure:
+                    result = self._li_data.strip()
+                    first = result.split()[0]
+                    if first:
+                        result = re.sub(
+                            r'^%s(\s+%s)+' % (first, first), first, result)
+                        # Sometimes, it repeats the same thing
+                        # multiple times.
+                    result = re.sub(r'unexpected flaky.*', '', result)
+                    # Remove some extra unnecessary text.
+                    result = re.sub(r'\bpreamble\b', '', result)
+                    result = re.sub(r'\bstdio\b', '', result)
+                    url = self._failure_results_url
+                    self.failure_results.append(
+                        BuilderHTMLParser.Result(result, url))
+                    self._current_failure_result = None
+                # Reset the state.
+                self._current_failure = False
+                self._li_data = ''
+                self._failure_results_url = ''
+
+
+def printer(indent, string):
+    """Print indented, wrapped text.
+    """
+    def wrap_to(line, columns):
+        """Wrap a line to the given number of columns, return a list
+        of strings.
+        """
+        ret = []
+        nextline = ''
+        for word in line.split():
+            if nextline:
+                if len(nextline) + 1 + len(word) > columns:
+                    ret.append(nextline)
+                    nextline = word
+                else:
+                    nextline += (' ' + word)
+            else:
+                nextline = word
+        if nextline:
+            ret.append(nextline)
+        return ret
+    out = sys.stdout
+    spacer = '  '
+    for line in string.split('\n'):
+        for i, wrapped_line in enumerate(wrap_to(line, 68 - (2 * indent))):
+            out.write(spacer * indent)
+            if i > 0:
+                out.write(spacer)
+            out.write(wrapped_line)
+            out.write('\n')
+    out.flush()
+
+
+def main(control_url, roll_url, verbosity=1):
+    """Compare two Codereview URLs
+
+    Args:
+        control_url, roll_url: (strings) URL of the format
+            https://codereview.chromium.org/?????????
+
+        verbosity: (int) verbose level.  0, 1, or 2.
+    """
+    # pylint: disable=I0011,R0914,R0912
+    control = CodeReviewHTMLParser.parse(control_url)
+    roll = CodeReviewHTMLParser.parse(roll_url)
+    if not (control and roll):
+        return
+
+    control_name = '[control %s]' % control_url.split('/')[-1]
+    roll_name = '[roll %s]' % roll_url.split('/')[-1]
+    all_bots = set(control) & set(roll)  # Set intersection.
+
+    out = sys.stdout
+    if verbosity > 0:
+        # Print out summary of all of the bots.
+        out.write('%11s %11s %4s %s\n\n' %
+                  ('CONTROL', 'ROLL', 'DIFF', 'BOT'))
+        for bot in sorted(all_bots):
+            if control[bot].status != roll[bot].status:
+                diff = '****'
+            elif (control[bot].status != 'success' or
+                roll[bot].status != 'success'):
+                diff = '....'
+            else:
+                diff = ''
+            out.write('%11s %11s %4s %s\n' % (
+                    control[bot].status, roll[bot].status, diff, bot))
+        out.write('\n')
+        out.flush()
+
+    for bot in sorted(all_bots):
+        if (roll[bot].status == 'success'):
+            if verbosity > 1:
+                printer(0, '==%s==' % bot)
+                printer(1, 'OK')
+            continue
+        printer(0, '==%s==' % bot)
+
+        for (status, name, url) in (
+            (control[bot].status, control_name, control[bot].url),
+            (roll[bot].status, roll_name, roll[bot].url)):
+
+            if status == 'failure':
+                printer(1, name)
+                results = BuilderHTMLParser.parse(url)
+                for result in results:
+                    formatted_result = re.sub(
+                        r'(\S*\.html) ', '\n__\g<1>\n', result.text)
+                    printer(2, formatted_result)
+                    if ('compile' in result.text
+                        or '...and more' in result.text):
+                        printer(3, re.sub('/[^/]*$', '/', url) + result.url)
+            else:
+                printer(1, name)
+                printer(2, status)
+        out.write('\n')
+
+
+if __name__ == '__main__':
+    if len(sys.argv) < 3:
+        print >> sys.stderr, __doc__
+        exit(1)
+    main(sys.argv[1], sys.argv[2],
+         int(os.environ.get('COMPARE_CODEREVIEW_VERBOSITY', 1)))
+