Delegation chains, the confused deputy, and the protocols you actually deploy

This post explains the 'confused deputy' problem in agent systems and how it arises from flattening delegation chains instead of preserving them using `sub` and `act` claims in token exchange. It details how the Model Context Protocol (MCP) and Agent-to-Agent (A2A) protocols are the critical surfaces where delegation chains are preserved or destroyed, and advises checking for proper authorization mechanisms and inspectable Agent Cards.