Turning Claude and Cursor into a 215-person Chinese agency

What it does

This is the Chinese community fork of agency-agents. It ships 215 markdown role definitions—165 translated from upstream, 50 created for the Chinese market—that turn generic AI coding assistants into specialists with specific workflows and deliverables. Shell scripts copy or convert these files into the config directories of 17 supported tools. It is fundamentally a curated prompt library with installer glue, but the rigor of each role—identity, workflow, deliverables—is what separates it from a generic template dump.

The interesting bit

The 50 China-original roles cover delightfully vertical territory: Feishu chatbot integration, DingTalk apps, livestock farming record verification, Qt-based industrial HMI development, and Xiaohongshu influencer strategy. Each agent is structured so the LLM knows not just who it is, but what to do—the security engineer runs an OWASP Top 10 checklist, the mechanical designer checks against GB/ISO standards.

Key highlights

• 215 agents across 18 departments, each with identity, rules, workflow, and deliverables.
• 50 China-market originals covering WeChat mini-programs, Douyin ops, cross-border e-commerce, medical compliance, and mechanical design.
• Shell installers for 17 tools; Claude Code and Copilot install directly, while others need format conversion first.
• Companion agency-orchestrator project strings agents into DAG workflows for multi-agent collaboration.
• MIT licensed.

Caveats

• Most tools require running convert.sh before installation, so the “one command” experience is currently limited to Claude Code and Copilot.
• The heavy lifting is still done by the underlying LLM; this repo provides structure, not intelligence.

Verdict

Grab this if you use AI coding assistants for Chinese-market work and want structured, reusable personas. Skip it if you already maintain a private prompt library or have no use for domain-specific roles like livestock farming record verification.