Idea: find a way for running actions whenever needed.

Currently there is no way to interweave manual testing with the automated actions. It would make the actions be used in a more flexible way if we can initiate them whenever needed, e.g. filling the checkout form by running `checkout` action.

I think it's possible to implement an interactive commandline prompt or another floating GUI dialog communicating with the underlying node process, which in turn communicates to the browser instance.

No files or tests are named. Start by tracing how actions currently reach the browser instance and how the underlying Node process is launched; done should allow a tester to initiate an action such as `checkout` during manual testing.

Written by the indexing model from the issue text.