Having dabbled in an ontology + first-order logic approach to a general intelligence agent, which seems perhaps in the same ballpark as this approach, I think one of the hardest aspects is automatically getting an actually useful, extensive KB. It essentially represents an agent gaining "real world experience", maybe the most important thing for a general intelligence bot. The documentation doesn't appear to describe obtaining the concepts for the graph, so I'm wondering, from where or through what methods do you get the concepts for your CGKB graph?
It's still a very early project, trying to figure that out myself too! A very first working implementation can take the intersection between the functions the bot can perform (since otherwise it'd be meaningless to list it as an executable function) and the nodes in the kb. So those marked nodes would be the "trigger" that once called, will traverse the graph to collect the relevant functional arguments (or ask the human for missing info otherwise). Easier said than done of course, still quite a lot of details to flush out.