Hive Mind Docs and Config Levers Really Confusing for new Project User

I'm a few weeks into trying to use this framework and I'm still not sure I understand the config/levers/tuning I can do when trying to use hive-mind. What prompts affect what (I can't even find the prompt in the project folder that is sent to the queen during a hive-mind spawn command), how to use the different modes, there is a lot of value proposition here but I have no idea if my configuration is using it properly or what I should be doing to adjust to workflow issues. I have no idea how to setup hooks for agents the queen spawns, e.g. stop agents from editing specific files. Should I be authoring custom agent templates? How do I setup workflows where standard SOPs are used by the queen? Use consensus features? When should I not use this framework vs just claude code?
I have no idea if the memory layer is getting used properly, if I should be manually cleaning it? How to ensure neural models are getting used properly? Do I have to wire that up? I have no idea if the queen is spawning workers in the correct topology. Flags on the hive-mind spawn such as queen type are getting ignored. Does the hive need tools like ruv_swarm or the flow-nexus stuff? How to clean up hives? sessions? Do I need to clean them up?

Sorry for all the question blasting, just wanted to give examples where I end up lost.

The documentation could really use some more work to help users understand how to leverage the hive-mode mode appropriately. Please update it with a lot more details on how a user can better understand the layers, what should be where and what they should consider tuning and how to validate aspects are working properly.

Start with the existing hive-mind documentation and the hive-mind spawn command described in the issue. Map the configuration levers, modes, prompts, hooks, memory, topology, tools, flags, and cleanup guidance before deciding how to organize the updates. Done means a new project user can understand the layers, choose settings, validate behavior, and follow the documented workflows.

Written by the indexing model from the issue text.