专家工具箱

v1.1.1

集成178+ AI专家角色,支持中文调用、分类浏览、关键词搜索及智能匹配,助你快速找到并使用各领域专家解决问题。

1· 87·0 current·0 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for catkid010520/expert-toolkit.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "专家工具箱" (catkid010520/expert-toolkit) from ClawHub.
Skill page: https://clawhub.ai/catkid010520/expert-toolkit
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install expert-toolkit

ClawHub CLI

Package manager switcher

npx clawhub@latest install expert-toolkit
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
medium confidence
Purpose & Capability
Name/description (Expert Toolkit for browsing and invoking local expert-role prompts) matches what the package contains: config mapping files, search/matching code, and routines to load markdown role files. There are no unexpected credentials, binaries, or install steps.
Instruction Scope
SKILL.md instructs only local usage (read role markdowns, search/match, return text). The code reads config/ and a roles directory and returns formatted results. Note: the library will read any markdown files under the configured roles root (default ~/.openclaw/workspace/knowledge/agency-orchestrator/roles), and an environment variable (EXPERT_TOOLKIT_ROLES_ROOT) can override that path — so the skill will read whatever is placed in that directory. The skill claims it does not execute user code or make network requests; the visible code implements read-only loading/formatting of role markdowns and does not show execution or outbound network calls in the inspected portions.
Install Mechanism
No install spec — instruction-only with packaged Python scripts and config files. Nothing is downloaded or executed during install; no archives or remote installers are used.
Credentials
The skill requests no credentials and has no required environment variables. It does honor EXPERT_TOOLKIT_ROLES_ROOT to override the roles directory — this is reasonable for flexibility but means a user or process could point it at an arbitrary path, allowing the skill to read files outside the default workspace. That behavior is explainable by the skill's purpose but worth noting.
Persistence & Privilege
No elevated privileges requested. always is false; the skill does not claim to persist or modify other skills or system settings. It reads local files and config only within its own scope.
Assessment
This skill is coherent with its description: it loads local 'expert' markdown files, indexes and matches them, and returns text. Before installing or using it: 1) Verify the expert-role files you place under the roles directory come from trusted sources (prompts can contain instructions or sensitive content). 2) Avoid setting EXPERT_TOOLKIT_ROLES_ROOT to system or sensitive directories (that would let the skill read unrelated files). 3) If you expect the skill to call external services, inspect the remaining parts of scripts/expert_toolkit.py for network calls — the inspected portions show only file reads/formatting. 4) If you will use third-party role packs (e.g., from GitHub), scan them for prompts that ask for secrets or to contact external endpoints before adding them to your roles folder.

Like a lobster shell, security has layers — review code before you run it.

agencyvk972q9bfr794q09jdqtjtzyv3x852rbqagentsvk972q9bfr794q09jdqtjtzyv3x852rbqaivk972q9bfr794q09jdqtjtzyv3x852rbqchinesevk972q9bfr794q09jdqtjtzyv3x852rbqexpertvk972q9bfr794q09jdqtjtzyv3x852rbqlatestvk972q9bfr794q09jdqtjtzyv3x852rbqrolevk972q9bfr794q09jdqtjtzyv3x852rbqtoolsvk972q9bfr794q09jdqtjtzyv3x852rbq
87downloads
1stars
2versions
Updated 1w ago
v1.1.1
MIT-0

专家工具箱 (Expert Toolkit)

集成 agency-agents 项目的 178+ AI专家角色(来自 msitarzewski/agency-agents,原 agency-orchestrator 官方迁移),让你随时召唤不同领域的AI专家帮你解决问题。

📋 功能概览

功能说明使用示例
直接调用通过专家名直接调用专家/expert 产品经理 帮我分析这个需求
列出分类查看所有专家分类/expert categories
列出分类下专家查看某个分类下所有专家/expert list 技术
搜索专家按关键词搜索专家/expert search 代码
智能匹配根据你的需求自动匹配最合适的专家/expert @expert 帮我做代码审查

🚀 快速开始

1. 安装

  1. 解压 expert-toolkit.zip 到你的 skills 目录
  2. 确保已经导入了专家角色文件到 knowledge/agency-orchestrator/roles/ 目录
  3. 完成,直接使用!

2. 使用方式

专家工具箱支持三种调用方式,满足不同场景:


方式一:直接调用已知专家

如果你知道专家名字,直接调用:

/expert {专家名} {你的问题}

支持:

  • 中文名称(需要配置映射,常用专家已内置)
  • 英文ID(原始ID,一定可用)
  • 中文模糊匹配(输入"代码"能匹配"代码审查员")

示例:

/expert 产品经理 帮我分析这个用户需求,给出产品规划建议
/expert engineering-code-reviewer 帮我检查这段Python代码,找出问题

方式二:搜索/浏览专家

如果你不确定有哪些专家,可以先搜索浏览:

1️⃣ 查看所有分类:

/expert categories

返回所有分类,显示中文名称 + 专家数量。

2️⃣ 查看分类下所有专家:

/expert list {分类名}

支持中文分类名:

/expert list 技术
/expert list 产品
/expert list 测试

3️⃣ 搜索专家:

/expert search {关键词}

支持中文关键词、英文关键词,会搜索:

  • 专家ID
  • 专家中文名
  • 分类名称
  • 中文映射表(支持简写匹配)

示例:

/expert search 代码
/expert search manager
/expert search 财务

方式三:智能自动匹配(推荐!)

只需要说清楚你的需求,专家工具箱会自动匹配最合适的专家:

/expert @expert {你的需求描述}

示例:

/expert @expert 帮我做代码审查,看看这段代码有什么问题
/expert @expert 帮我做财务分析,评估这个项目投资回报率
/expert @expert 帮我做产品需求分析,画出用户故事地图

原理: 根据内置关键词映射,匹配最相关的专家,返回所有匹配结果。


🌍 中文支持说明

专家工具箱对中文用户做了完整优化:

中文分类映射

所有常用分类都有中文映射,你可以直接用中文:

中文英文专家数量
产品product5
设计design8
技术engineering29
财务finance5
游戏game-development5
市场marketing30
营销paid-media7
项目project-management6
销售sales8
战略strategy3
客服support6
测试testing8
数据data-

中文专家名称调用

常用专家已经配置了中文→英文映射,可以直接用中文调用:

示例:

  • 产品经理product-product-manager
  • 代码审查员engineering-code-reviewer
  • 架构师engineering-software-architect

逐步完善中,常用专家会陆续添加映射。

⚙️ 配置文件说明

配置文件都在 config/ 目录下:

文件作用
chinese_translate.json中文名称 → 专家ID映射,方便直接中文调用
keyword_mapping.json关键词 → 专家ID列表映射,用于智能自动匹配

添加自定义中文映射

编辑 config/chinese_translate.json

{
  "产品经理": "product-product-manager",
  "代码审查员": "engineering-code-reviewer",
  "你的自定义名称": "expert-id"
}

添加自定义关键词映射

编辑 config/keyword_mapping.json

{
  "代码审查": ["engineering-code-reviewer"],
  "需求分析": ["product-product-manager", "product-product-trend-researcher"],
  "你的关键词": ["expert-id-1", "expert-id-2"]
}

📊 专家分类统计

当前版本:

  • 总专家数: 178
  • 总分分类: 18

详细:

分类中文专家数
academic-5
design设计8
engineering技术29
examples-6
finance财务5
game-development游戏5
integrations-1
marketing市场30
paid-media营销7
product产品5
project-management项目6
sales销售8
scripts-0
spatial-computing-6
specialized-41
strategy战略3
support客服6
testing测试8

🔒 安全说明

专家工具箱完全安全

  • ❌ 不执行任何用户输入代码
  • ❌ 不写入任何文件(仅读取专家角色markdown)
  • ❌ 不发起任何网络请求
  • ✅ 所有专家角色文件都存储在本地,不向外传输

🎯 设计原则

  1. 不重复存储:专家角色存放在 knowledge/agency-orchestrator/ 复用原项目,skill只提供调用层
  2. 多种使用方式并存:直接调用/搜索/自动匹配,满足不同场景
  3. 中文优先:完整的中文分类、中文搜索、中文调用体验
  4. 高性能:启动建立O(1)索引,所有查询都是瞬间完成
  5. 优雅降级:配置文件损坏/文件读取失败都有警告,不崩溃

📝 更新日志

2026-04-17 v1.1.0

  • ✅ 同步最新 msitarzewski/agency-agents 仓库
  • ✅ 新增 marketing 分类,新增 30 位市场营销领域专家(抖音/小红书/知乎/直播电商等专家都有了)
  • ✅ 添加中文分类映射:市场 / 市场营销 → marketing
  • ✅ 现在总计 18 个分类,178 位专家

2026-04-17 v1.0.0

  • ✅ 完成基础功能开发
  • ✅ 支持列出分类、列出专家、搜索专家、直接调用、自动匹配
  • ✅ 完整中文支持
  • ✅ 所有测试通过,零已知bug
  • ✅ 正式发布

🙏 致谢

  • 专家角色来自 agency-agents 项目(原 agency-orchestrator 官方迁移),感谢原作者
  • 适配到 OpenClaw by 萌面大侠🦐

📄 License

Apache-2.0 (和原项目一致)

Comments

Loading comments...