Someone else asked the question, so I can just guess from what they said.

My guess is that the interest is not so much in coding the guts of the client in shell script (not exactly a high performance engine for computationally intensive tasks), but rather in exposing the data flows and stores to shell script manipulation.

In other words, as I have often done myself in other tasks, write several tools that each have specific capabilities, and which have interfaces that work well when combined in shell scripts with each other and other *nix utilities.  One shot commands are handled in that environment by forking/execing a command that is passed command line arguments, environment variables and files or file descriptors to read or write.  Higher performance for repetitive tasks is obtained by using commands in a series of filters, operating on a stream of line oriented text.  The new, individual tools would each be written in C or Python or some other such language that provided adequate performance and library support for their particular purpose.