Overview
agency-agents-zh is the Chinese community edition of msitarzewski/agency-agents, an MIT-licensed collection of AI agent role definitions. The README describes 277 plug-and-play expert roles across 20 departments, from a C-suite "company" department (CEO, CTO, CPO, CMO, COO, CFO) through engineering, design, marketing, paid media, sales, finance, HR, legal, supply chain, testing, game development and academic research. By the README's own count, 213 of them are translations of the English upstream and 64 are originals written for the Chinese market, marked with a star in the role tables.
Each role is a Markdown file with an identity, key rules, a working process and the deliverables it should produce, rather than a one-line "you are an expert" prompt. The README's examples are a security engineer that reviews code item by item against the OWASP Top 10 and a Xiaohongshu operator that outputs a full seeding-note and influencer-collaboration plan. The originals cover platform operations for Xiaohongshu, Douyin, WeChat, Bilibili, Kuaishou, Weibo and Zhihu, Feishu and DingTalk integration development, and verticals such as cross-border e-commerce, government procurement, medical compliance, Qt industrial host software and mechanical design.
What makes it installable rather than a reading list is the `scripts/` folder: `convert.sh` rewrites every role into the format each tool expects (Cursor `.mdc` rules, Codex CLI TOML agents, OpenClaw's SOUL.md / AGENTS.md / IDENTITY.md split, SKILL.md files for Hermes Agent, WorkBuddy, CodeWhale, DeerFlow and QwenPaw, a single CONVENTIONS.md for Aider), and `install.sh` copies them to the right global or project folder. The same author's Agency Orchestrator uses this library as its Chinese role set to run several roles together as a YAML workflow.
What it does
- 277 role definitions across 20 departments, per the README, each with an identity, rules, workflow and deliverables
- 64 China-market originals alongside translations of the English agency-agents upstream, flagged with a star in the README's role tables
- `./scripts/install.sh` auto-detects installed tools, or targets one with `--tool` for 20 integrations including Claude Code, GitHub Copilot, Cursor, Kiro, Trae, OpenCode, Aider, Windsurf, Gemini CLI, Qwen Code, Codex CLI, Qoder and OpenClaw
- `./scripts/convert.sh` regenerates the per-tool formats after you add or edit a role; Claude Code and GitHub Copilot take the Markdown files unconverted
- `--category` installs only chosen departments, which the README recommends for Hermes Agent's Discord mode and its slash-command size limit
- `./scripts/lint-agents.sh` checks role files for format problems
Getting started
Clone the repository (https://github.com/jnMetaCode/agency-agents-zh) and run the scripts from its root. Roles are then activated in your coding tool with a plain-language request.
Install into every detected tool, or pick one
Claude Code and GitHub Copilot install directly; other tools need `./scripts/convert.sh` first.
# auto-detect installed tools and install
./scripts/install.sh
# or target a specific tool
./scripts/install.sh --tool claude-code
./scripts/install.sh --tool copilotConvert, then install for tools that need their own format
Project-level tools such as Cursor are installed from inside your project. The README advises keeping only 10–20 Cursor rules, because installing all of them dilutes description-based matching.
# in the agency-agents-zh folder
./scripts/convert.sh --tool cursor
# in your project folder
cd /your/project
/path/to/agency-agents-zh/scripts/install.sh --tool cursorOr copy role files by hand
For Claude Code, the Markdown files go straight into `~/.claude/agents/`; then ask for a role in plain language, for example "activate frontend developer mode and help me build a React component" (the README's example is in Chinese).
cp -r marketing/*.md ~/.claude/agents/Install only some departments
./scripts/install.sh --tool hermes --category engineering --category designRegenerate after editing roles
./scripts/convert.sh # all tools
./scripts/convert.sh --tool cursor # one tool
./scripts/lint-agents.shCommands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Give a Chinese-speaking team the same set of specialist subagents in Claude Code, Cursor, Copilot or a domestic tool such as Trae, Qwen Code or Qoder
- Get platform-specific marketing help for Xiaohongshu, Douyin, WeChat or Bilibili from roles written for the Chinese market rather than translated
- Hand role files to Agency Orchestrator to run several experts as one multi-step workflow
- Use the role files as a reference when writing your own structured agent prompts
How agency-agents-zh compares
agency-agents-zh alongside other open-source agent skills & plugins tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| Superpowers | ★ 297k | A composable skills plugin that installs a spec-first, TDD, subagent-driven development methodology into Claude Code, Codex, Cursor, Gemini CLI and other coding harnesses. |
| Skills for Real Engineers | ★ 282k | Matt Pocock's everyday agent skills for coding agents, covering alignment grilling, planning, code review and research — small, composable and meant to be edited. |
| ECC | ★ 276k | An installable plugin that adds 68 agents, 286 skills, hooks, rules, memory and an agent-config security scanner to Claude Code, Codex and other coding harnesses. |
| Karpathy Coding Guidelines | ★ 218k | A Claude Code plugin and CLAUDE.md rule set built from Andrej Karpathy's observations on LLM coding pitfalls: think before coding, keep it simple, make surgical changes, and work to verifiable success criteria. |
| Ponytail | ★ 159k | A skill and plugin that makes a coding agent take the laziest correct route — a seven-rung ladder from "does this need to exist?" to "the minimum that works" — while keeping validation, security and accessibility off the chopping block. |
| Caveman | ★ 111k | A skill plus local proxy that shortens an AI coding agent's prose while leaving code, commands and paths untouched — the rule file cuts what the agent writes, the proxy compresses what it reads. |
| Addy’s Agent Skills | ★ 104k | A pack of production engineering skills for AI coding agents, with nine lifecycle slash commands — /spec, /plan, /build, /test, /review, /ship — installable into 70+ agents. |
| agency-agents-zh | ★ 21.1k | A Chinese-language library of AI expert role definitions, converted and installed into 20 coding tools by one script |