The requested result enters with the identity, project context and limits that define the operating boundary.
One control authority around every governed execution.
Core connects the requested outcome to permissions, bounded execution, verification, evidence and recovery without handing system authority to a model, agent or provider.
From intent to an accepted result.
Policy, permissions and approvals decide what the work is allowed to do before sensitive execution.
The work is decomposed and routed to bounded capabilities; a model or agent name is never authority by itself.
Acceptance checks, evidence and bounded recovery decide whether the result is ready to deliver.
Execution resources can change. Authority does not.
Identity, policy, approvals and permitted actions remain explicit and centralized.
Agents, skills, tools and providers operate only inside the scope granted to the job.
Validation and required approvals determine whether produced work can advance.
Material state, provenance and failure handling stay reviewable when work succeeds or fails.

