Edict

Edict is a multi-agent orchestration system for OpenClaw that structures agent collaboration on the model of a historical Chinese imperial bureaucracy: a task is triaged, planned, reviewed, dispatched, and then executed in parallel by a set of specialized ministry agents before reporting back. Twelve agents fill these roles, each with an independent workspace, skills and model.

Its central difference from frameworks such as CrewAI or AutoGen is a dedicated review stage that checks the quality of a plan before it reaches execution and can reject and force a rework of substandard output, rather than only an optional human-in-the-loop step. A strict permission matrix controls which agents may message each other, and state transitions are validated so an agent cannot jump to an invalid status.

A real-time dashboard shows a Kanban board of tasks with department filtering and search, department-level monitoring of task volume and agent health, an archive of completed tasks as a timeline, task templates, per-agent token usage statistics, and per-agent model switching. It can run standalone in Docker with sample data, or connect to a real OpenClaw gateway. It is written in Python under the MIT license.