Test 2, instrument before you architect, is the one most teams skip. The trace signatures you list are a clean way to tell a tool-access bottleneck from a coordination one before adding a protocol. I like that you tied the 2026-07-28 stateless spec directly to the caching and tasks changes rather than treating it as generic news. Have you seen the ttlMs and cacheScope additions actually stabilize upstream prompt caches across reconnects in a real deployment?
Kartik N V J K
Test 2, instrument before you architect, is the one most teams skip. The trace signatures you list are a clean way to tell a tool-access bottleneck from a coordination one before adding a protocol. I like that you tied the 2026-07-28 stateless spec directly to the caching and tasks changes rather than treating it as generic news. Have you seen the ttlMs and cacheScope additions actually stabilize upstream prompt caches across reconnects in a real deployment?