AI Court: Multi-Agent Collaboration with Ming Dynasty Six Ministries
One command to start your dynasty, all AI ministers at the ready. Built on OpenClaw, this is a 24/7 AI management team modeled after the Ming court.
Evidence shows: README emphasizes security configurations (e.g., allowBots: 'mentions'), SECURITY.md provides security best practices including API key protection, Docker non-privileged running, webhook signature verification. However, lacks user confirmation mechanism (e.g., sensitive operations require manual confirmation), data flow transparency is limited (no detailed explanation of data flow), sensitive data handling only mentions API key protection, dependency security only mentions npm audit and Docker scout, external effects (e.g., Discord message loops) are warned but not fully controlled, rollback has backup scripts but no clear recovery process, source attribution has originality claim but not verified. Deductions: missing user confirmation, opaque data flow, incomplete sensitive data handling, no specific vulnerability info for dependencies, incomplete external effects control, imperfect rollback mechanism, unverified source attribution.
Evidence shows: README and documentation are consistent, test files exist (e.g., test-context-compressor.js), but no test results provided. Dependency availability: few dependencies (@notionhq/client, winston), but no version locking or integrity checks. Failure messages: error handling module (scripts/error.js) and tests exist, but no user-friendly error prompts. Deductions: tests not executed, dependencies not locked, failure messages not fully demonstrated.
Evidence shows: README clearly defines target users (beginners, developers), provides multiple deployment methods (local, remote, Docker), supports multiple platforms (Discord, Feishu, etc.), provides three regimes (Ming, Tang, Modern Enterprise) to suit different scenarios. Capability boundaries: core capabilities and skill list clearly listed, but limitations not detailed. Trigger precision: @mention mechanism exists, but trigger rules not detailed. Environment fit: supports multiple platforms and architectures, but detailed environment requirements not provided. Deductions: insufficient trigger precision, unclear environment requirements.
Evidence shows: README has clear structure with TOC, quick start, architecture, FAQ, good information architecture. Installation notes detailed with multiple methods. Naming stability: version numbers exist (v3.5.3, 3.7.1), but no naming conventions. Examples and FAQ abundant. Known limitations: not explicitly listed. License: MIT license file present. Versioning changelog: no CHANGELOG provided. Maintenance responsibility: maintainer (工部) mentioned, but responsibilities not clear. Deductions: insufficient naming stability, missing known limitations, missing changelog, unclear maintenance responsibility.
Evidence shows: Output usability: GUI, Discord interaction, diverse output forms. Marginal value: advantages over ChatGPT etc. (multi-agent, memory, tools), but no actual performance data. Cost-benefit: cost estimates provided ($10-30/month), but no detailed cost analysis. Deductions: marginal value lacks data support, cost-benefit analysis not in-depth.
Evidence shows: Claims in README (e.g., 18+ Agents, 60+ Skills) lack specific evidence, test files exist but no test results, architecture diagrams exist but not verified. Deductions: claims lack traceability, insufficient cross-validation, facts and inferences not clearly separated.
- Publisher identity is unverified; treat provenance claims with caution.
- Install scripts execute remote code; review them before running.
- Sensitive information like API keys and Discord bot tokens must be handled carefully.
- Multi-agent systems may cause message loops; follow security configuration guidelines strictly.
What does this agent do, and when should you use it?
AI Court is a multi-agent collaboration system built on the OpenClaw framework, mirroring the Ming Dynasty's cabinet system. It organizes 18+ AI agents into roles like Grand Secretariat, Six Ministries, and Investigation Bureau. Three governance templates are available: Ming Cabinet, Tang Three Departments, and Modern Enterprise. Interaction is via Discord, Feishu, etc., with agents collaborating through workflows (e.g., Grand Secretariat receives orders, Inner Cabinet optimizes, Six Ministries execute, Investigation Bureau reviews). The system includes 60+ built-in skills (GitHub, Notion, browser automation), a one-click installation script, a web dashboard, memory backups, cron jobs, and sandbox isolation. It is MIT licensed and an original architecture.
The system runs multiple agents on OpenClaw, each representing a department or role (e.g., Grand Secretariat, Inner Cabinet, Ministry of War). Users @mention an agent on Discord or Feishu to trigger workflows: the Grand Secretariat receives orders, passes them to the Inner Cabinet for prompt optimization, dispatches to the Six Ministries for execution, and the Review Board automatically checks code when pushed to GitHub. Each agent has independent memory (SQLite database) and can invoke 60+ skills (GitHub, Notion, browser automation, cron jobs). It provides one-click install scripts (full-install.sh), regime-switching scripts (switch-regime.sh), memory backup (memory-backup.sh), and diagnostics (doctor.sh).
- Solo developers quickly set up a multi-agent environment for project management and code review via Discord.
- Teams simulate task distribution, approval, and execution using the Ming Cabinet system for better collaboration.
- Users who want automated code review, leveraging the Review Board triggered on GitHub push.
- Notion workflow automation with agents writing daily summaries to Notion.
- Enthusiasts of historical governance exploring AI simulations of the Three Departments or Modern Enterprise systems.
- Ops scenarios needing 24/7 monitoring and scheduled tasks (health checks, backups).
What are this agent's strengths and limitations?
- Zero-code multi-agent orchestration via configuration, lowering the entry barrier.
- Original architecture inspired by historical Ming governance structures.
- Multi-platform support (Discord, Feishu, Slack, Telegram) with rich integrations.
- 60+ built-in skills covering development, documentation, and automation.
- One-click install with 5-minute deployment and diagnostic tools for low ops cost.
- Memory backup and cron scheduling for long-running operations.
- Depends on the OpenClaw framework, requiring prior setup and a learning curve.
- Multiple bots require precise
allowBotsconfiguration to avoid message loops. - Operational cost depends on LLM API usage, ~$10-30/month.
- Complex architecture with many agents and skills may overwhelm novices.
- Documentation is primarily in Chinese, posing a language barrier for non-Chinese users.
How do you install or deploy this agent?
Before installation, prepare a cloud server (recommended), an LLM API key (Anthropic/OpenAI, etc.), and a Discord Bot token. Steps: 1) Local install: git clone https://github.com/wanikua/danghuangshang.git && cd danghuangshang && bash scripts/full-install.sh; remote install: bash <(curl -fsSL https://raw.githubusercontent.com/wanikua/danghuangshang/main/scripts/full-install.sh). 2) Follow the prompts to enter your LLM API key and Discord Bot token. 3) Mention your bot on Discord to start.
How do you use this agent?
After installation, the Ming Cabinet system is enabled by default. On Discord, @mention the Grand Secretariat with a task (e.g., "write a login API"). It will delegate to the Inner Cabinet for optimization, then to relevant ministries (e.g., War), and the Review Board will review after code push. Switch regimes with bash scripts/switch-regime.sh tang-sansheng or bash scripts/switch-regime.sh modern-ceo. Use memory backup with bash scripts/memory-backup.sh. Access the web GUI by starting node index.js.
How does this agent compare with similar options?
Compared to ChatGPT, AutoGPT, or CrewAI, AI Court emphasizes "out-of-the-box" readiness: configuration without coding for multi-agent setups, whereas AutoGPT/CrewAI require Python programming. AI Court offers independent memory and 60+ skills, while ChatGPT is a single model and AutoGPT needs vector DBs. AI Court natively supports chat interfaces like Discord, making deployment easy, though it sacrifices flexibility compared to coding frameworks.
FAQ
What is the monthly cost?
Which LLM models are supported?
How do I prevent bot message loops?
allowBots: "mentions"; avoid true or false.