Model or dataset
wanikua/danghuangshang avatar
wanikua/danghuangshang

danghuangshang ships three versions, a lint that cannot fail, and OpenClaw at latest

Open-source multi-agent collaboration system inspired by Chinese governance — deploy and coordinate specialized AI agents with OpenClaw.

2,718 stars260 forksTypeScriptMIT

At a glance

What is it?
danghuangshang wraps OpenClaw in a Ming dynasty ministry metaphor, installs with a one-line shell script, and carries a Chinese-first README. Its package.json says 3.7.1, its image label says 3.6.0, and its only tagged release is v3.0.
Who is it for?
danghuangshang is worth a look if you want a multi-agent arrangement laid out in roles you can point colleagues at, and you are willing to read Chinese, since README.md leads the project and README_EN.md and README_JA.md sit beside it. Install it with the local clone route rather than the piped one-liner, so you can read scripts/full-install.sh before it runs, and keep a copy of MEMORY.md yourself, because the automatic backup covers only the config file.
Can I use it commercially?
Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 177 days ago.
What is it written in?
Mainly TypeScript, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • wanikua/danghuangshang2,718 ★ · +17
Star history
Repository2026-09-102026-11-15
wanikua/danghuangshang2.7k2.7k

17 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

The manifest says 3.7.1, the image label says 3.6.0, the release tag says v3.0

Three numbers describe the same project and none of them match. The root package.json carries version 3.7.1 and the name clawd. The Dockerfile sets org.opencontainers.image.version to 3.6.0, a hardcoded label rather than a build argument. And the release history holds exactly one entry, v3.0, published 2026-03-12, described in Chinese as a documentation restructure plus a Docker image plus full platform coverage. The last push to the repository is dated 2026-05-22, so the working tree moved more than two months after the only tag and after the label was set. Nothing reconciles the three. A reader who sees 3.7.1 on npm and 3.6.0 on a pulled image has no way to tell from either artifact whether one is simply behind.

The package is named clawd and its main entry is not in the tree

The manifest names the package clawd, not danghuangshang, and declares main as index.js. There is no index.js at the repository root and no file by that name anywhere in the top level listing. Two bin entries do resolve: task-store points at ./scripts/task-store.js and context-compressor at ./scripts/context-compressor.js, both inside the scripts directory that is present. Those two are the only things in the manifest a consumer can actually reach, and they are exposed as command line programs rather than as a library surface. The dependency list is short, @notionhq/client at 5.9.0 and winston at 3.19.0, with jest and @types/jest as the only development entries. The engine floor is Node 22.19.0 or later, which is also the base image the Dockerfile uses, node:22-alpine.

The lint script ends in || true, so it cannot fail

The lint entry in package.json is one line: bash -n scripts/*.sh && node -c scripts/*.js 2>/dev/null || true. Two things happen inside it. The shell glob expands to many paths, and node is handed every one of them at once rather than one file at a time, so the syntax check errors out, and the redirect sends the error to /dev/null where nobody reads it. Then the trailing || true converts any remaining non-zero status into success. The net effect is that the command always exits zero, so it cannot fail a pre-commit hook, a CI job, or a pull request check. Meanwhile the project is described as TypeScript and the only tsc invocation in the repository sits inside the Docker GUI build stage, where gui/package.json and gui/package-lock.json are copied and built with npx tsc -b followed by npx vite build. There is no typecheck script in the manifest at all.

OpenClaw is installed at latest, so two builds are never the same image

The second Dockerfile stage installs the framework globally with npm install -g openclaw@latest and a loglevel of error. Nothing pins it. There is no ARG for the framework anywhere in the build, and the only two exact versions in the repository live in package.json, which the Dockerfile never reads. So the layer that supplies the entire runtime is resolved from the registry at build time and recorded nowhere in the image. The label does not help either, since it carries 3.6.0 rather than a framework version. Rebuilding the same commit on two different days can produce two different runtimes under an identical version label, which is a specific problem for a project whose stated purpose is agents that stay online around the clock holding shared state.

Compose binds both ports to localhost and still sets read_only: false

The compose file maps two ports and binds both to the loopback interface, 127.0.0.1:18789 for the gateway dashboard and 127.0.0.1:18795 for the pineapple GUI, with a Chinese comment stating that the default binds localhost and that production should change it. The service also sets security_opt to no-new-privileges:true, which is the one clearly defensive line in the file, and immediately sets read_only: false in the same block, so the container filesystem stays writable. Resource limits are 4G of memory and 2.0 CPUs with reservations of 1G and 0.5, logging is the json-file driver capped at 50 MB across three files, and the healthcheck curls http://localhost:18789/health every 30 seconds with a 10 second timeout, 3 retries, and a 60 second start period. The service is named court, the container is named ai-court, and the image is boluobobo/ai-court:latest while the compose file also carries a build section.

OpenViking is installed into /opt but mounted from /home/court/.openviking

An optional Python component is installed with a shell idiom that makes the build survive its own failure. The recipe creates a virtual environment at /opt/openviking, attempts a pip install of openviking with its output discarded, symlinks the result into /usr/local/bin/openviking, and ends the whole group with an or that prints a Chinese message saying the OpenViking install was skipped and the core is unaffected. The image then sets PATH to include /opt/openviking/bin regardless. So a build can complete with no OpenViking present and with a PATH entry pointing at a directory that holds nothing. The compose file makes it worse by declaring a third path: a named volume court-openviking mounted at /home/court/.openviking. Install location, PATH entry, and mount target are three different directories for the same optional piece of software.

Webhook signing and allowBots both landed after the only tagged release

Two safety items in the README are dated, and both dates sit after v3.0 on 2026-03-12. The first is a security note marked 2026-03-22 announcing webhook signature verification so that forged requests are rejected, with a separate configuration guide for it. The second is a Discord multi-bot warning whose fix is a specific configuration value: allowBots set to mentions, so a bot only answers another bot when mentioned. The README calls the newer installer default already safe, tells users of older configurations to add that block by hand, and explicitly forbids allowBots set to true because the bots then trigger each other into a message storm, linking issue 107. In other words, the loop that produces a message storm is a configuration value, not a bug that was fixed, so an old config file carried into a new install can bring the problem back with it.

The documented update is stash, pull, stash pop, and MEMORY.md is yours

The update section offers two routes. The one it recommends is a single script that backs up and checks as it goes:

bash
bash scripts/safe-update.sh

The manual route is four commands run from the home directory, with an explicit warning that conflicts are possible and have to be resolved by hand:

bash
cd ~/clawd
git stash
git pull
git stash pop

Underneath both, the README makes a distinction worth pausing on. The system backs up the configuration file at ~/.openclaw/openclaw.json automatically, but workspace files such as MEMORY.md have to be backed up manually. So the file that is auto protected is the one you can regenerate, and the file holding accumulated agent memory is the one you can lose. Related to that, the older install.sh is documented as unable to run remotely and failing with a /dev/fd path error.

Editorial conclusion

danghuangshang is worth a look if you want a multi-agent arrangement laid out in roles you can point colleagues at, and you are willing to read Chinese, since README.md leads the project and README_EN.md and README_JA.md sit beside it. Install it with the local clone route rather than the piped one-liner, so you can read scripts/full-install.sh before it runs, and keep a copy of MEMORY.md yourself, because the automatic backup covers only the config file. Check which version you actually have before trusting a number: the manifest, the image label, and the release tag all disagree. And if you build the image, expect openclaw@latest to move under you, since nothing in the build pins it.

Frequently asked questions

Is danghuangshang an OpenClaw plugin or its own runtime?

Its own runtime built around OpenClaw. The project is described as built on the OpenClaw framework, and its installer puts OpenClaw in place globally with npm install -g openclaw@latest. A parallel route exists for Nous Research's Hermes Agent through install-hermes.sh, and the two runtimes are documented as able to coexist.

How many agents does danghuangshang create?

It depends on the system you choose at install time. The Ming cabinet default is 18 agents, while the Tang three departments arrangement and the modern CEO setup are 14 each. You can switch later with scripts/switch-regime.sh and the arguments tang-sansheng, modern-ceo, or ming-neige.

Which ports does the danghuangshang container expose?

Two, both bound to 127.0.0.1 by default: 18789 for the gateway dashboard and 18795 for the pineapple GUI, with a note in the compose file that production deployments should change the binding. The healthcheck curls http://localhost:18789/health on a 30 second interval.

What does danghuangshang install onto a server?

The local route is git clone, cd danghuangshang, then bash scripts/full-install.sh, described as offline capable and modifiable. A remote one-liner pipes the same script into bash, a lite script covers the case where OpenClaw already exists, and install.ps1 covers Windows PowerShell. The old install.sh is documented as not supporting remote execution and failing with a /dev/fd path error.

Is danghuangshang available in English?

The repository root holds README_EN.md and README_JA.md next to the Chinese README.md, and the text in hand is the Chinese version, which stops partway through a diagram of the Tang dynasty arrangement. The Chinese README states that the project is MIT licensed and invites pull requests and forks on the condition that attribution is given and the license is kept.

Official sources

  1. License: MIT
  2. Project website
  3. README
  4. Releases
  5. wanikua/danghuangshang on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/wanikua-danghuangshang.svg)](https://hysenlabs.com/projects/wanikua-danghuangshang)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/wanikua-danghuangshang.svg)](https://hysenlabs.com/projects/wanikua-danghuangshang)

Was this useful?