This was asked a while ago so you've probably all organised yourselves by now, but this was also asked over on the API.ai forums recently and there is still no GIT way of doing this.
There are however youtube videos (see api forum for post, don't have it to hand) showing large projects being built in which the devs set up a sequence of agents and linked them by using the export-import feature to move intents and entities along the process and into the production version. They insist that the agent trains itself very quickly so nothing is lost during this transfer.
So I have a couple of conversationalist creating User Says and Text Responses based on the bot personality. Each one has their own agent to play with and are assigned intents to complete.
Then these are imported into Dev Bot where things are tested with the backend, other media responses and integration responses are added. Dev Bot exports itself as versions, and also into Test Bot where everyone gets to play with it. Test Bot does the same, exports to versions and
then when everyone is happy Test Bot gets exported into Prod Bot.
Its not ideal. Its not the way you would imagine a product owned by google to be forcing us to work, its a manual process, but it does sort of work.