I created a context LSP for Claude code.
It still passes typical diagnostics, but it also passes hints, concurrent agent proximity alerts, active memory recall, and other situational awareness that generally limits how effective concurrent agents can actually be.