What it does

This repo is a curated library of 215 AI agent definitions—persona files you drop into Claude Code, Cursor, Copilot, or 14 other tools. Each file defines not just “you are an expert,” but how the expert thinks, works, and what they hand back. The Chinese fork adds 50 locally original roles (Xiaohongshu operator, Feishu integrator, Qt industrial HMI engineer, livestock compliance auditor) on top of 165 translated from upstream.

The interesting bit

The install script auto-detects your tooling and copies the right format; for OpenClaw it even splits each persona into SOUL.md + AGENTS.md + IDENTITY.md for native multi-agent orchestration. The companion Agency Orchestrator project can then compose these roles into DAG-executed teams. It’s glue, but it’s thoughtful glue.

Key highlights

• 215 roles across 18 departments: engineering, design, marketing, finance, security, game dev, etc.
• One-command install for 17 tools (Claude Code, Cursor, Copilot, Trae, Windsurf, Qwen, DeerFlow, WorkBuddy, Hermes Agent, etc.)
• 50 China-market originals covering WeChat mini-programs, Douyin/Xiaohongshu ops, ToG compliance, mechanical design per GB/ISO standards
• MIT licensed; PRs welcome
• Companion orchestrator supports 6 free LLM backends and 3 API providers with breakpoint-resumable DAG execution

Caveats

• The “215” count and “17 tools” claim appear in the README but the actual file tree isn’t fully shown in the truncated source; verify before relying on specific role availability
• Some integrations (Claude Code, Copilot) install directly; others require a convert.sh pass—check your tool’s path
• Sponsor block is prominent and includes referral links

Verdict

Grab this if you’re a Chinese-speaking team trying to get consistent, repeatable output from AI coding assistants without writing prompts from scratch. Skip if you already maintain your own system-prompt library or don’t work in the covered domains.