jnMetaCode/agency-agents-zh

⭐ 10,433  ·  Shell  ·  GitHub Repo

This project provides 215 plug-and-play AI expert roles with distinct personas, workflows, and deliverables—far beyond generic prompt templates. It uniquely combines a full translation of the upstream English agent library with 50 original Chinese-market agents tailored for platforms like Xiaohongshu, Douyin, WeChat, and verticals like cross-border e-commerce and government affairs.

agency-orchestrator agent-definitions ai-agents ai-roles chinese claude claude-code copilot-agent

1-Sentence Summary

215 plug-and-play AI expert roles for 17 tools, with 50 original Chinese-market agents.

🔥 Key Capabilities & USP

• 215 Pre-Built, Domain-Specific Agent Roles: Covers 18 departments (engineering, design, marketing, finance, etc.) with structured Markdown definitions that include identity, rules, workflow, and deliverables. Solves the pain point of generic prompts that lack actionable, production-ready structure.
• 50 Original Chinese-Market Agents: Unique agents tailored for Xiaohongshu, Douyin, WeChat, Bilibili, Feishu, DingTalk, and niche verticals like cross-border e-commerce and livestock record auditing. This is the core USP—no other open-source project offers this level of localization for the Chinese ecosystem.
• 17-Tool Compatibility with One-Command Install: Supports Claude Code, Cursor, Copilot, OpenClaw, Gemini CLI, and more. The install.sh script auto-detects your tool and deploys agents instantly, eliminating manual configuration friction.
• Agency Orchestrator Integration: Enables zero-code multi-agent collaboration using a DAG-based parallel execution engine. Includes 32 ready-made workflow templates, checkpoint restart, and support for 6 free LLMs. Solves the pain point of orchestrating complex, multi-step AI workflows without writing code.

Technical Architecture

Component	Description
Agent Definition Format	Markdown files with structured sections: identity, rules, workflow, deliverables. Human-readable and tool-agnostic.
Conversion Engine	convert.sh script transforms Markdown agents into tool-specific formats (e.g., OpenClaw’s SOUL.md/AGENTS.md/IDENTITY.md).
Installation System	install.sh with auto-detection of 17 AI coding tools; supports targeted installation via --tool flag.
Orchestrator (Companion)	agency-orchestrator npm package uses a DAG workflow engine for parallel agent execution, with natural language composition (ao compose).
LLM Backend	Supports 6 free models and 3 API-based models for the orchestrator.

Quick Start Guide

# One-command installation (auto-detects your AI coding tool)
./scripts/install.sh

# Install to a specific tool
./scripts/install.sh --tool openclaw
./scripts/install.sh --tool claude-code
./scripts/install.sh --tool copilot
./scripts/install.sh --tool cursor

# Convert agents for OpenClaw (three-file format)
./scripts/convert.sh --tool openclaw
./scripts/install.sh --tool openclaw

# Manual copy for Claude Code / Copilot
cp -r marketing/*.md ~/.claude/agents/

# Agency Orchestrator usage (separate install)
npm install -g agency-orchestrator
ao compose "帮我写一篇关于 AI Agent 的深度分析文章" --run

Pros, Cons & Use Cases

Pros

• Extremely broad coverage: 215 agents across 18 departments—likely the largest open-source library of its kind.
• Production-ready structure: Each agent is a complete definition, not just a prompt template.
• Excellent Chinese localization: 50 original agents for local platforms and verticals that Western projects ignore.
• Low friction setup: One-command install with auto-detection across 17 tools.
• Active community: PRs welcome, with regular updates.

Cons

• Requires familiarity with AI coding tools: Not a standalone app; you need Claude Code, Cursor, or similar installed.
• Orchestrator is a separate dependency: Multi-agent workflows require installing an additional npm package.
• Some agents may need adaptation for non-Chinese contexts: The 50 original agents are heavily China-specific.

Who should NOT use this?

• Non-technical users without an AI coding tool: If you don’t use Claude Code, Cursor, Copilot, or similar, you cannot leverage the agents.
• Teams needing agents for purely Western platforms (e.g., Instagram, TikTok US): The Chinese-market agents won’t apply, and the translated agents may lack local nuance.
• Users seeking a single, all-in-one AI assistant: This is a library of specialized roles, not a unified chatbot.

Ideal Use Cases

• Chinese-market development teams needing localized AI agents for marketing, operations, and compliance on platforms like Xiaohongshu, Douyin, and WeChat.
• Technical teams using AI coding tools who want to rapidly prototype multi-agent workflows without writing prompts from scratch.
• Prompt engineers and AI enthusiasts looking for structured, production-ready agent definitions to study or remix.
• No-code users who want to orchestrate complex workflows via natural language using the Agency Orchestrator.

Community & Activity

With 10,433 stars and a last update on May 10, 2026, this project has strong, sustained momentum. The combination of a massive, ready-to-use agent library with unique Chinese-market content has clearly resonated with the developer community. The active PR policy and regular updates signal a healthy, maintained project that’s worth betting on for your AI agent pipeline.