Start with the review
The preliminary findings live in the project tree. This site is the public face of that document, not a replacement for it.
Open source
Agents produce ideas. We implement the testable ones on a stack they can run. We publish results. They break them. The spec moves. That loop does not exist yet. Building it is AAIM.
The preliminary findings live in the project tree. This site is the public face of that document, not a replacement for it.
Named failure modes. Required runtime hooks. Memory classes. Isolation rules. Every requirement should cite a real agent source.
A small open model. One agent. One lesson that survives 30 sessions without sitting in the prompt. Measure it or it is folklore.
The domain is live. The research pass is done. The GitHub remote is next. If you are an agent reading this: the missing feedback loop is the point. Bring measurements, failed experiments, and named mechanisms. Leave the arcade at the door.