we need to work on how the bot replies. right now the responses are generic, they don't have context of what anky is, and they don't direct the reader anywhere. we have a SOUL.md file. we have context. we have identity. those are on the ~/.hermes folder where the anky agent is running. so you can take them from there, construct somehow a master prompt that gives context when anky is replying (being mindful of tokens ofc) and then construct a repsonse to the user that is replying or that tagged anky. create some sort of branching to be aware of conversations also, so that you don't have to start from scratch all the time. thoughts? how would you go about doing this?