The most underrated thing about Claude Code isn’t the model—it’s the six-layer architecture. The input layer gates sessions and permissions first, and the YAML trust hierarchy filters things before they ever reach the model. Put simply, the model is just one node in the loop; the real work happens around it. It’s like a sudden wick in the perpetuals around dawn, when liquidity is thinnest: don’t just focus on that one move in the middle. Those of you building tools, share how you handle the layers.