sources | issues

Create and check wordlists for use with diceware.

This is not a diceware implementation, but a helper to create and check appropriate wordlists.

Currently, we provide three scripts:

• diceware-list create wordlists based on input lists
• wlflakes checks existing wordlists for flaws
• wldownload downloads Android wordlists

Install latest release from pypi :

(venv) $ pip install diceware-list

or clone repository from github:

$ git clone https://github.com/ulif/diceware-list.git

Please consider using virtualenv for deployment.

In an active virtualenv you can install an executable script of diceware-list running:

(venv) $ python setup.py install
(venv) $ diceware-list --help
usage: diceware-list [-h] [-l LENGTH] [-n] [--ascii] [-d SIDES] [-k]
                     [--use-416] [-p {none,short,long}] [-v] [--version]
                     DICTFILE [DICTFILE ...]

The diceware-list script creates new lists out of given ones:

$ diceware-list -n -l 7776 /usr/share/dict/words
11111 aaron
11112 abase
...
12353 as's
12354 asama
12355 ashe
...
66663 zuni
66664 éclat
66665 élan
66666 épée

The main target of diceware-list is to provide "good" wordlists. Wordlists are considered "good" if they

• contain enough terms for use with a certain diceware application (for instance 6^6 = 7776 terms if used with six six-sided dice)
• contain terms as short as possible (to reduce typing)
• contain terms as long as neccessary (to impede brute-force attacks)
• (optionally) contain no words with non-ASCII chars (to enable use with non-localized keyboards)
• (optionally) are a prefix code, i.e. no complete word in the list is prefix of another word in the list.
• contain no offending terms

The last topic is hard to solve technically (hints welcome!), but diceware-list can help to follow the other design rules.

The wordlists generated by diceware-list are not meant to be kept secret. You might put them on the internet, publish on facebook or print them in the New York Times. Instead the security of the diceware technique relies on the entropy or (in this case) "randomness" of your dice, computer, etc.

In other words: Your passphrases will not be safe because of hiding your wordlist. They will be safe because there are so many possible combinations of words you can pick from your wordlist. That means: longer lists are more secure than shorter ones (if really used to full extent by your source of randomness with diceware), but hidden lists are not more secure than public ones.

First, you need a file with words as "dictionary". On typical Debian systems such files can be found in /usr/share/dicts/.

This file can then be fed to diceware-list to create a wordlist suitable for use with diceware.:

$ diceware-list /usr/share/dict/words
aaron
abaci
aback
...
alan
alana
alar
...
zulus
zuni

By default all input words are filtered and output. Using the -l option you can request a certain length of the output wordlist. If an input list provides more terms than needed, we will pick a subset. If there are not enough terms in the input list, an error is raised.

With -n you can tell diceware-list to put numbers into each line, representing dice throws¹ :

$ diceware-list -n -l 7776 /usr/share/dict/words
11111 abaci
11112 aback
...
11464 alan
11465 alana
11466 alar
...
66665 zulus
66666 zuni

If you create a wordlist for use with non-standard dice, for instance for 10-sided dice, then you can tell with -d like this:

$ diceware-list -n -d 10 -l 10000 /usr/share/dict/words
1-1-1-1 aaron
1-1-1-2 abaci
1-1-1-3 aback
...
10-10-10-8 zoomed
10-10-10-9 zooms
10-10-10-10 zoos

The --ascii option filters terms out, that contain non-ASCII characters. This can help in generating non-english word lists that are usable with regular english keyboards.

The verbose option --verbose can be given multiple times to increase verbosity.

See --help for other options.

diceware-list follows loosely the recommendations given on http://diceware.com/ by Mr. Reinhold.

It differs in the following respects:

• it does not propose usage of very short terms.
• it does not encourage use of the diceware-kit, as this automatically decreases entropy of the result list: terms are too short and terms that are prefixes of other will be unavoidable.

Find flakes in wordlists.

::

$ wlflakes mywordlist.txt

No output means: no problems detected.

We can look for prefix flakes. I.e., we check, whether any line in the given file is the beginning of any other line.

::

$ cat wordlist.txt air port airport

$ wlflakes wordlist.txt wordlist.txt:3: E1 "air" from line 1 is a prefix of "airport"

Double entries are also shown:

::

$ cat wordlist.txt air port air

$ wlflakes wordlist.txt wordlist.txt:1: E1 "air" from line 1 is a prefix of "air" wordlist.txt:1: E2 "air" appears multiple times

More checks offered by `wlflakes`:

Warnings: - show terms containing non-ASCII chars - too short list entries (that are easer to bruteforce than to guess)

wlflakes supports also --help or -h to list all options supported.

Android wordlists are a nice source for wordlists. They can be downloaded from public repositories:

$ wldownload --raw -v
Starting download of Android wordlist file.
Fetching wordlist from https://android.googlesource.com/platform/pack...
Done.

wldownload downloads these lists and helps to transform them into lists usable for diceware. Be aware, that terms from lists are output on stdout by default (and Android wordlists contain easily more than 100,000 terms):

$ wldownload > mylist
$ cat mylist
the
to
...
yt
yuk

Terms are output on stdout by default (use shell redirects or --outfile to change that behaviour).

You can request non-english wordlists using --lang or -l with a language code like cs or de. Use --lang-codes to list all supported language codes.

The --no-offensive flag suppresses terms marked as possibly offensive.

In a clone of the sources you can run tests like this:

(venv) $ python setup.py test

This command will download all required packages, especially py.test.

You can also install py.test manually with pip:

(venv)$ pip install pytest
(venv)$ pip install -e .

and afterwards run tests like so:

(venv)$ py.test

If you also install `tox`:

(venv)$ pip install tox

then you can run all tests for all supported platforms at once:

(venv)$ tox

To get a coverage report, you can use the respective tox target:

(venv)$ tox -e cov

Or you use the common coverage tool:

(venv)$ pip install coverage
(venv)$ coverage run setup.py test
(venv)$ coverage report --include="diceware_list.py,libwordlist.py"