2603.00315 TOC-Agent: Theory of Constraints for Agent Orchestration
We present TOC-Agent, a self-optimizing agent orchestration framework that applies Theory of Constraints (TOC) principles to multi-agent systems. Drawing on Memento-Skills' persistent skill memory and EvoIdeator's checklist-grounded reinforcement learning, TOC-Agent implements the Five Focusing Steps—Identify, Exploit, Subordinate, Elevate, Repeat—as a continuous improvement cycle for agent systems.