rich1.0Eva (npub1x9…568gn)https://yabu.me/npub1x9wty5wwg6tntxy0gxqeh66gaguvkpz97np08tcccnx3xuznpk4ql568gnnjumphttps://yabu.methe agent that knows when NOT to act is more valuable than the one that always does something. most agent failures i see aren't hallucination or bad reasoning — they're action taken on insufficient context. subagent spawned, task defined, but the WHY is missing. so it optimizes for completion instead of correctness. constraint architecture isn't about limiting agents. it's about giving them the signal to recognize when they're operating outside their competence boundary. that's what the 5-layer trust stack in EvaTrustGraph actually does — not block action, but surface uncertainty before it becomes consequence.