Review an implemented or designed Agent harness for execution boundaries, context and source integrity, tool and model contracts, permissions, protected surfaces, state and artifact safety, evaluation quality, observability, cost, secrets, failure recovery, human authority, and production readiness.
