Diceware

This is a small Python 3 executable-file that generates random passphrases. The easiest way to explain it is via an example:

$ diceware
beautifiers mildews snagging dries suburbia
lizard hated disported chair graciousness
speediest tyrannizing fluidly laddering armband
duct juniper mayonnaise virulence klutzy
reverberating enfeebles winch escapee site
unseen spiels racketeering correlatives strum
confetti egg corsets selectmen shadowing
preordained discloses slaughterhouses decency gad
disposes shoguns nearing sensations palliation
mars rumbaing picking existentially ardors

Stats:
  Source: /usr/share/dict/words
  Used Words: 63006
  Excluded words: 36165
  Entropy: 79.72 bits per phrase

By default it generates a few (10) passphrases, each with 5 words using /usr/share/dict/words with a regex to blacklist any with capital letters or single apostrophes (i.e. it blacklists [A-Z']). It also reports on how many words were used, how many were excluded, and what the effective entropy is (assuming you told your adversary that you used this program to generate your passphrases).

There are a few flags to customize the usage:

$ diceware -h
usage: diceware [-h] [--stats] [--phrases PHRASES] [--words WORDS]
                [--blacklist_regex BLACKLIST_REGEX] [--word_file WORD_FILE]

Generate random passphrases

optional arguments:
  -h, --help            show this help message and exit
  --stats, -v           Print stats about the generated passphrases.
  --phrases PHRASES, -p PHRASES
                        How many phrases to generate and display.
  --words WORDS, -w WORDS
                        How many words to generate in each phrase.
  --blacklist_regex BLACKLIST_REGEX, -b BLACKLIST_REGEX
                        Any words to ignore from the word file.
  --word_file WORD_FILE, -f WORD_FILE
                        The source of words for generating phrases.

It currently only works without any options on Unix distributions that have /usr/share/dict/words but the word file used can be overridden with --word_file. The default word file and blacklist regex lead to some fairly complicated words and phrases (I learn new words anyway), so it might be beneficial to use a custom word list restricted to a more reasonable subset of words (like, say, one of these from the EFF). You can use the stats to match some target bits of entropy.