cft0808/edict ? reverse-engineered prompt

I want you to build me a multi-agent AI orchestration system inspired by the structure of ancient Chinese imperial bureaucracy. The core idea is to create a system with strong checks and balances, not just a free-for-all chat between agents.

The workflow should be hierarchical. When I submit a task, a "planning" agent breaks it down. Then, a crucial "auditor" agent must review and approve this plan. It can even reject a bad plan and send it back for revision. Only once approved does the plan get passed to a team of specialized "execution" agents to do the work.

This whole process must be visible on a real-time web dashboard. The dashboard should use a Kanban-style view to track tasks as they move through the different agent "departments." From there, I also need to be able to see full audit trails of completed tasks, monitor agent health, and configure which LLM each agent uses. Please build the backend in Python.