cft0808/edict in Claude Code: three departments, six ministries
By Flowy · Updated 2026-08-28
cft0808/edict is a Claude Code listing that models multi-agent orchestration on China's three departments and six ministries bureaucracy. Flowy indexes 4 agent components here: a global rules file, shared review rules for the three coordinator roles, shared execution rules for the six ministry roles, and one example persona file. It needs OpenClaw, a separate agent runtime, installed to actually run.
What Flowy indexes for cft0808/edict
Flowy indexes 4 agent components for cft0808/edict, filed under automation, from a repository carrying 16,415 GitHub stars: GLOBAL, sansheng, liubu, and SOUL. All four are configuration and persona files, not task skills. The project's own materials describe something bigger than four pieces. Its GitHub description mentions nine specialized agents. Its README opens by counting twelve: eleven working roles plus one role kept for compatibility with an older setup. The two numbers do not agree with each other, and neither is what gets indexed here. What Flowy carries is the rule structure underneath that larger system: a global baseline, two group-level rulesets, and one worked example of a role's own file, not a name for every role the project claims to run.
Why sansheng and liubu are named after a dynasty
The project's own description, 三省六部制, names the Tang dynasty's imperial bureaucracy: three coordinating departments (三省) that plan, review, and dispatch work, and six ministries (六部) that execute it, each covering one domain such as revenue, justice, or public works. cft0808/edict maps that structure onto AI agents. One intake role sorts incoming requests. Three coordinator roles plan a task, review the plan, and hand off approved work. A further set of ministry roles carry out the work itself, one per domain: data, documentation, engineering, compliance, infrastructure, and personnel, in the version its README describes. sansheng and liubu are the pinyin names for exactly those two groups, and their descriptions match that split: sansheng holds the review rules the three coordinator roles share, liubu holds the task execution rules the ministry roles share.
What GLOBAL, sansheng, liubu, and SOUL actually do
GLOBAL sits above both groups. Its description says it holds the rules every role must follow, and that an individual role's own file can override them. SOUL is that per-role file: one copy per role in the source project, each written for a specific job. The single copy Flowy indexed and described is for the ministry the project maps to engineering work. It opens by identifying itself as a role invoked as a subagent to handle engineering implementation, architecture design, and feature development. Read together, the four indexed components are a small but complete slice of one rule hierarchy: a floor every role follows, one shared ruleset for the reviewers, one shared ruleset for the doers, and one worked example of how a single role's file overrides the floor.
What running this actually requires
None of this runs on its own. The project's own setup instructions list OpenClaw, a separate multi-agent runtime, as something you need installed and configured before its installer does anything, alongside Python and, if you want its dashboard rebuilt locally, Node.js. Its own materials also describe a browser dashboard for watching tasks move through the pipeline, per-role model switching, and an audit log of every status change. None of that shows up as an indexed component here, because a dashboard and an audit log are application code, not agent or skill definitions. You interact with the finished system by messaging it through Feishu, Telegram, or Signal, rather than by prompting Claude Code directly.
Who this plugin fits
This fits a team that wants a heavier, more accountable pipeline than a single assistant answering directly: a forced review step before work starts, a separate role per domain, and a record of who did what. It fits less well as a lightweight add-on to an existing Claude Code setup. There is no skill, command, MCP server, or hook here to drop into a project on its own, and getting any use out of the four indexed components means also standing up OpenClaw and the rest of the system they belong to.
How do I install it
Install from the listing page linked at the top of this guide. Flowy indexes what cft0808/edict ships here; it does not own, bundle, or host it. The fuller setup lives entirely in the source repository, not in Flowy: cloning it, installing OpenClaw, and running the project's own installer script.
Common questions
- What is cft0808/edict?
- It is a Claude Code listing built around China's historical three departments and six ministries bureaucracy, mapped onto a multi-agent pipeline: an intake role sorts requests, three coordinator roles plan and review a task, and several ministry roles execute it in a specific domain. Flowy indexes 4 agent components from it. Running it needs OpenClaw, a separate multi-agent runtime, not just Claude Code.
- How many agents does this plugin actually have?
- That depends which of the project's own numbers you read: its GitHub description says nine, and its README opens by counting twelve. Flowy does not try to referee that difference. What is actually indexed and described is 4 agent components: a global rules file, one shared ruleset for the coordinator roles, one shared ruleset for the ministry roles, and one example persona file.
- What is the difference between GLOBAL, sansheng, and liubu?
- GLOBAL holds the baseline rules every role in the system must follow, and its own description notes that an individual role's persona file can override it. sansheng and liubu are named for the two role groups in the underlying three departments and six ministries structure: sansheng carries the shared review rules for the three coordinator roles, and liubu carries the shared task execution rules for the ministry roles that do the actual work.
- Do I need OpenClaw to use cft0808/edict?
- Yes, according to its own setup instructions. The project lists OpenClaw, a separate multi-agent runtime, as a prerequisite before its installer does anything, alongside Python, and Node.js if you want its dashboard's frontend rebuilt locally. It is not a plugin you add directly to a plain Claude Code setup with nothing else installed.
- What does the SOUL component actually describe?
- The one Flowy indexed and described is for the role the project maps to engineering work: its description identifies it as invoked as a subagent to handle engineering implementation, architecture design, and feature development. Every role in the source project has its own version of this file under the same generic name, so this is one worked example rather than the full set.