Capability Boundaries

Capability boundaries are the enforcement layer on top of role definitions. A role tells an agent what to do; a capability boundary tells it what tools it is allowed to use. The research agent gets read access. The generator gets write access. The critic gets evaluation access. No agent gets everything.

This is the principle of least privilege applied to agent systems. When SCOPE researches a competitor, he reads public data and produces intelligence briefs. He does not draft outreach emails — that is HUNTER's job, with HUNTER's tools. When you give an agent tools beyond its role, you have not empowered it. You have created a scope-creep vector that compounds at 3.47% per week if unmonitored.