agency-agents-zh is a Chinese-language registry of 277 plug-and-play AI expert roles, built for individual developers, small teams, and China-market operators who want named specialists with their own personas and deliverables driving tools such as Claude Code, Cursor, and Copilot.

What it is

The project is the Chinese community edition of the upstream repository msitarzewski/agency-agents. It ships 277 agent definitions spread across 20 departments, covering company operations roles such as CEO, CTO, CMO, COO, CPO, and CFO, alongside engineering, design, marketing, product, game development, security, GIS, and finance. Of those 277 agents, 213 are translations of the upstream English set, and 64 are original agents written for the Chinese market. It lives in the AI development tooling ecosystem, distributed as the npm package agency-agents-zh and as a Shell-based repository under the MIT licence, with localized documentation in README.md and README.zh-TW.md.

The concrete problem it solves is the generic prompt template. Rather than handing a model a one-line instruction and hoping for structure, each agent carries a distinct persona, a professional process, and defined deliverables. The registry also replaces the manual work of wiring multiple specialists together: paired with the companion orchestrator agency-orchestrator, a single sentence can dispatch several experts that collaborate automatically along a DAG. The topics list confirms the packaging intent, naming agent-definitions, cursor-rules, copilot-agent, and claude-code, which is how the definitions reach the 20 supported tools.

Key capabilities

• Ships 277 agent definitions organized into 20 departments, including company operations roles for CEO, CTO, CMO, COO, CPO, and CFO.
• Covers 20 AI coding tools, with Claude Code, Cursor, and Copilot named explicitly and deepseek and hermes-agent listed among the topics.
• Includes 64 original China-market agents alongside 213 translations of the upstream English set.
• Provides platform-operations agents for 小红书, 抖音, 微信, B站, 飞书, and 钉钉.
• Adds vertical agents for 跨境电商, 政务ToG, 医疗合规, Qt 工业上位机, 机械设计, and 畜禽养殖档案核对.
• Works with agency-orchestrator so one sentence can trigger multi-expert collaboration along a DAG.
• Gives every agent its own persona, professional workflow, and deliverable, rather than a shared generic prompt template.

Who uses it and how

• Individual developers and one-person companies, following the 35-lesson 实战 course that covers 单兵点名, 自动组队, and the full 一人公司 workflow.
• Teams that want a unified agent workbench, using the native desktop client for macOS, Windows, and Linux, which avoids installing Node.
• China-market content and platform operators, who staff the 小红书, 抖音, 微信, B站, 飞书, and 钉钉 agent roles.
• Industrial and regulated-sector users, applying the Qt 上位机, 机械设计, 医疗合规, and 政务ToG agents.
• A large community: the repository records 20,752 stars and 3,346 forks, with one open issue.

Getting started

Install the npm package agency-agents-zh, or download the native desktop client from the jnMetaCode/agency-orchestrator releases page for macOS, Windows, or Linux, which requires no Node installation. A hosted option is available for trial at ao.aiolaola.com/experts.

How it compares

The closest named tool is the upstream msitarzewski/agency-agents, an English-language project that this repository translates in full before adding 64 China-market agents of its own. The companion agency-orchestrator is what turns single-expert invocation into DAG-based multi-expert collaboration, while host tools such as Claude Code, Cursor, and Copilot supply the underlying model runtime rather than competing with the registry.

When to use it — and when not to

There is nothing to operate beyond the definitions themselves: no database, storage layer, or SMTP service is involved, since the repository ships agent definitions and an npm package rather than a server. Skip it if the work has no Chinese-language or China-platform component, because the upstream English project already covers 213 of these agents without the added specialization. One documented inconsistency is worth noting: the introduction states 63 original China-market agents while the project scale table and the registry summary both state 64, and the README devotes significant space to sponsor promotions for third-party API providers, which makes the promotional sections longer than the technical ones.

npm install -g agency-orchestrator
ao compose "帮我写一篇关于 AI Agent 的深度分析文章" --run