What it does

Edict runs 12 LLM agents through a rigid pipeline: a “Crown Prince” triages incoming messages, three “Imperial Secretariats” plan, review, and dispatch work, and seven “Ministries” execute in parallel. A React dashboard tracks the whole flow in real time. It is built on top of OpenClaw and ships as a Docker image with a one-line installer.

The interesting bit

The author did not just borrow metaphors. The “Chancellery” (门下省) acts as a mandatory QA gate: it can reject and send back plans until they meet quality standards, with enforced state-machine transitions preventing illegal jumps. That is a genuine architectural constraint, not an optional plugin.

Key highlights

• Mandatory review loop: every task must pass the Chancellery; no exceptions, no silent failures.
• Real-time kanban + audit trail: heartbeat badges, intervention controls (stop / cancel / resume), and archived “memorials” with full lineage.
• Per-agent model hot-swapping: change an individual agent’s LLM from the dashboard; gateway restarts in ~5 seconds.
• Zero backend dependencies: the Python server uses only the standard library; React 18 handles the frontend.
• Docker demo available: docker run -p 7891:7891 cft0808/sansheng-demo serves pre-loaded mock data if you lack an OpenClaw setup.

Caveats

• Requires OpenClaw to be installed first for full use; the Docker image only shows simulated data.
• macOS / Linux only; Windows support is not mentioned.
• The README claims 12 agents but the description says 9; the breakdown is 11 business roles plus 1 compatibility role, so the exact count depends on how you classify them.

Verdict

Worth a look if you are running CrewAI or AutoGen and tired of opaque agent chatter with no oversight. Skip it if you are not willing to adopt OpenClaw or if your stack is Windows-first.