AI agent workspace architecture, demonstrated end-to-end in Claude Code: roles library, persistent memory, hooks, scheduled agents, self-audits, loop selection, and measurement-gated self-improvement.…