◇Execution Protocol

Models can generate text and make decisions. None of them control what an agent is actually allowed to do. Execution Protocol is the layer that does — a deterministic boundary between an agent's intent and the action that commits.

One structured message goes in; one terminal outcome comes out — executed, instructed (when the request was incomplete), or blocked. Credentials never reach the agent. The boundary decision happens outside the model's control. And every outcome, including the ones that didn't happen, produces a hash-chained, signed receipt you can verify offline.

It is the difference between an agent that can act and an agent you can trust to act.