This is a research prototype exploring agent loops with persistent internal state, observer-based self-reflection, and topic cycling to avoid agreement collapse.
It’s not a framework or a product – the code is meant to be read, modified, and questioned. I’d especially appreciate feedback on the architecture and failure modes.