The context problem nobody solved: AI agents across multiple repos

An agent is good inside one repo and blind between them. Five workarounds, and where all of them stop.