Agency Agents Chinese Expert Library
Installable Chinese expert-role definitions for AI coding tools.
- Source repo
- jnMetaCode/agency-agents-zh
- Stars
- ★ 22k
- Last updated
- 3d ago
- License
- MIT
- Primary language
- Shell
- FA score
- 28/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesCodex · Claude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A Claude Code developer who wants local, predefined workflows for a frontend developer or code-reviewer role.
- Not a fit if
- Users who want zero conversion steps before installing
- Teams wanting one generic prompt instead of full role definitions
- Users needing a model runtime or DAG execution engine
- Source review
- 28/100 · Major gaps 7 safety controls not found
What does this agent do, and when should you use it?
Agency Agents Chinese Expert Library is a collection of 268 ready-to-use expert-role definitions spanning 20 departments. Each Markdown definition specifies an identity, rules, workflow, and deliverables rather than supplying a one-line persona prompt. Its scripts/convert.sh and scripts/install.sh convert and place those definitions into the documented locations for 18 tools, while Claude Code and GitHub Copilot can use direct copies. The repository is a role library rather than a model runtime; DAG-based multi-expert orchestration is presented as a capability of the separate agency-orchestrator project.
From the repository directory, users run ./scripts/convert.sh or ./scripts/convert.sh --tool <tool> to transform role definitions, then run ./scripts/install.sh or ./scripts/install.sh --tool <tool> to install them. Generated formats include Cursor .mdc rules, Codex CLI TOML agent files, and OpenClaw's SOUL.md, AGENTS.md, and IDENTITY.md files. Once installed, roles can be activated with natural-language requests in tools such as Claude Code and GitHub Copilot; Cursor, Trae, and Qoder also document description-based matching or explicit selection.
- A Claude Code developer who wants local, predefined workflows for a frontend developer or code-reviewer role.
- A Cursor project team that selects 10–20 relevant roles and uses .cursor/rules for frontend implementation, security review, or Git-workflow tasks.
- A China-market team working on Xiaohongshu, Douyin, WeChat, Feishu, or DingTalk that needs the repository's marked original localized role definitions.
- A Codex CLI project that installs converted TOML files in .codex/agents and invokes expert roles as subagents.
- An OpenClaw user who wants each expert role split into SOUL.md, AGENTS.md, and IDENTITY.md for role-based collaboration.
How do you install or deploy this agent?
From the repository directory, run:
./scripts/install.sh --tool claude-codeFor tools that require conversion, run:
./scripts/convert.sh --tool codex
./scripts/install.sh --tool codexNo credential step is documented for these local conversion and installation commands.
How do you use this agent?
After Claude Code installation, a documented invocation is:
激活前端开发者模式，帮我审查这个组件。
For Codex CLI, the documented flow is conversion followed by project-level installation in .codex/agents; the roles are then called as subagents. Cursor supports automatic rule matching and explicit @rule-name references.
What are this agent's strengths and limitations?
- The library spans 20 departments, including engineering, design, marketing, finance, games, and spatial computing, rather than limiting itself to coding roles.
- It includes 53 roles marked as original for Chinese-market platforms and workflows, including Xiaohongshu, Douyin, WeChat, Feishu, DingTalk, cross-border commerce, and specialist domains.
- One source library is converted by scripts for 18 tools, with direct installation paths for Claude Code and GitHub Copilot.
- It documents concrete delivery formats and install locations for distinct targets such as Codex CLI, Cursor, and OpenClaw.
- This is a prompt and role-definition library, not a model runtime; DAG execution across multiple experts requires the separate agency-orchestrator project.
- Most supported tools other than Claude Code and GitHub Copilot require a conversion step before installation.
- The Cursor guidance warns that installing a large rule set can reduce matching accuracy and recommends retaining only 10–20 frequently used rules.
- Hermes Discord integration cannot install every skill at once because Discord has an 8,000-character limit on total command JSON serialization; categories must be installed in batches.
How does this agent compare with similar options?
Compared with the upstream msitarzewski/agency-agents repository, this project is positioned as a Chinese community edition: it translates upstream roles and adds original roles for Chinese platforms and business scenarios.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Agency Agents Chinese Expert Library This agent | 28 · Major gaps | CLIFree + model costs | ★ 22k | 3d ago | Shell | Codex · Claude Code |
| Agentic Plugin Marketplace | 66 · Some gaps | Agent plugin / skillFree + model costs | ★ 42k | 2d ago | Python | Codex · Claude Code |
| Agent Skills Library | 67 · Some gaps | Agent plugin / skillFree + model costs | ★ 240 | 2mo ago | Python | Codex · Claude Code |
| Agent Toolkit | 74 · Some gaps | CLIFree | ★ 19 | today | V | ChatGPT · Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: The repository is a collection of prompts with no code execution, so no permission, confirmation, data flow, sensitive data handling, dependency security, external effects, or rollback mechanisms exist. MIT license and clear copyright attribution (upstream and localization) are present, but publisher is unverified. Deductions: These criteria are zero due to lack of relevant implementation; source attribution scores 1 because license and README clearly state upstream and localization sources.
Evidence: Inconsistency in agent count between README (268) and package.json (267) indicates weak self-consistency. Dependency availability: Only Node.js scripts and GitHub Actions, but no dependency locking or verification. Failure messages: None provided. Deductions: Count inconsistency, no failure handling documentation.
Evidence: Clear target audience (Chinese market) and scenarios (Xiaohongshu, Douyin, etc.), and support for 18 tools. Capability boundaries: Each agent has a description but no explicit boundaries. Trigger precision: Natural language activation but no precise trigger words. Environment fit: Supports 18 tools but installation scripts require conversion. Deductions: Boundaries and trigger precision are insufficient.
Evidence: Clear directory structure (by department), detailed installation instructions, clear license, version number exists. Naming stability: File names are stable, but version history not provided. Examples and FAQ: Examples present but no FAQ. Known limitations: Not mentioned. Versioning changelog: Not provided. Maintenance responsibility: Author and contact present but unverified. Deductions: Missing known limitations, changelog, and FAQ.
Evidence: Output is prompts that can be copied and used directly, but quality not verified. Marginal value: Provides many localized agents with unique value. Cost-benefit: Free and open source, but requires user configuration. Deductions: Output usability not verified, cost-benefit not quantified.
Evidence: README claims 268 agents but package.json says 267, and no verification method provided. Cross-source corroboration: None. Fact-inference separation: Not distinguished. Deductions: Count inconsistency, no verification mechanism.
- Not found in source: least-privilege scopingGrant only what the task needs: a dedicated account or read-only token, scoped to specific directories and repos.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Agent count inconsistent between README and package.json (268 vs 267), needs verification.
- Publisher identity unverified, maintenance responsibility and update path unclear.
- No known limitations, changelog, or FAQ provided; users must assess risks themselves.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v1.2.5 v1.2.5 — README 在线展示区 + 精简
Release notes excerpt
变更 - docs: README 新增「在线浏览全部专家」展示区(截图 + [ao.aiolaola.com/experts](https://ao.aiolaola.com/experts) 链接) - docs: 精简 Agency Orchestrator 区块——已有桌面客户端,去掉冗长能力表/ASCII 块,保留核心(npm / 客户端下载 / 在线体验) > 角色文件无变化,仅文档更新。View on GitHub ↗ -
v1.2.4 v1.2.4 — 计数补齐 216 + 计数守卫
Release notes excerpt
变更 - docs: AGENT-LIST.md / CATALOG.md 补齐到 216 个角色(加入「服装工厂规划工程师」,修正部门/总计计数) - tooling: 新增 npm run check:counts(scripts/check-counts.mjs)——校验文档总数与实际角色文件一致,防止计数再悄悄滞后 - package 描述 215 → 216 > 角色文件本身无变化(与 1.2.3 相同),仅文档与工具更新。View on GitHub ↗