Architecting Self-Improving Agents

A proposed architecture for agents that learn from experience - episodic memory in Markdown, selective context reveal, MCP as wiring, and SME-driven evals that close the fine-tuning loop. Most agents