Bret et al - Ignoring CACAO for a second, how one would do this in a programming language is pass the output from the 1st command (e.g. HTTP in your example) to the 2nd command as input using a data processing step that manipulates the information into the correct representation required for the 2nd command.
For me, this is what we had discussed as part of the variable assignment infrastructure we previously discussed and suggested where you could pass one variable content (containing the raw data) via a data processing step (sed, grep....etc) into another data format.
This is also akin to command chaining that often can be seen in Unix command shells.
With all this said, don't we already have the ability to pass data via a variable to a command such as sed or grep or similar and then get the resultant output back from that command?
I would also argue this was exactly why we wanted to show an example of using a LLM to get results and then return the output in a subsequent desired output. We agreed on the call that LLM integration could be achieved by just sequencing the prompts and results to the LLM via HTTP use of the LLM API.
So I'm not sure why you think we are missing a command type given this?
I think it would be good to show examples of this in the spec but I don't think we need a new command.