Develop automated testing harness using mocked LLM responses

For continuous integration and delivery, we will need to have some way to test that the agent reacts well to a number of scenarios in terms of possible outputs from the LLM.

We need to first check the available literature on this matter. We can complement the literature with two fallback options. One option could be to try something like MockLLM:

https://github.com/StacklokLabs/mockllm

Another option could be to put LLM use behind some kind of component that we can swap out with a mock implementation at any point.