Author
Mariano Garralda-Barrio
This paper introduces a governance-oriented framework for the controlled evolution of agent runtimes through Executable Operational Cognition.
Rather than treating agent-generated artifacts as transient outputs, it formalizes their evolution into persistent runtime capabilities governed through explicit validation, lifecycle management, mutation, provenance, observability and rollback mechanisms. The paper proposes HarnessMutation as a governed adaptation mechanism and introduces a runtime model in which operational capabilities evolve under explicit governance constraints instead of unrestricted self-modification.