万里牛RPA官方Skill

v1.0.0

湖畔 RPA 机器人管理。通过 MCP 协议连接本地 RPA 客户端,控制机器人执行、取消任务、查看状态等操作。当用户提到"机器人"、"RPA"、"执行机器人"、"查询任务"等时自动激活。

0· 80·0 current·0 all-time
byBenny Burton@bennybao

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for bennybao/hupun-rpa.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "万里牛RPA官方Skill" (bennybao/hupun-rpa) from ClawHub.
Skill page: https://clawhub.ai/bennybao/hupun-rpa
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 hupun-rpa

ClawHub CLI

Package manager switcher

npx clawhub@latest install hupun-rpa
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
medium confidence
Purpose & Capability
Name/description match the behavior: the SKILL.md describes controlling a local RPA client via MCP at http://127.0.0.1:13765/mcp. No unrelated environment variables, binaries, or cloud credentials are requested. The listed operations (list, execute, cancel, relaunch, etc.) are coherent with RPA management.
Instruction Scope
The instructions operate only against a local MCP HTTP endpoint and do not ask the agent to read arbitrary user files or environment variables. They do include potentially disruptive operations (relaunch, reinstall_browser, logout, cancel_task) which will affect the local system if invoked. The template for daily_briefing contains JavaScript-like expressions (e.g., details.succeed.map(...)) — this implies the agent or host will evaluate/interpret templates; confirm how the agent executes templates to avoid unintended code execution.
Install Mechanism
Instruction-only skill with no install spec and no code files; lowest install risk. README suggests installing via npx from a GitHub URL, which is standard, but the packaged skill itself contains only docs. No downloads/executables are defined in the skill.
Credentials
No environment variables, credentials, or config paths are requested. The skill only expects the local MCP service to be reachable at 127.0.0.1:13765, which is appropriate for a local RPA controller.
Persistence & Privilege
always is false and disable-model-invocation is false (normal). The skill does not request permanent presence or elevated agent-wide privileges in its metadata. It documents automatic activation on certain intents, which is a UX decision but not an incoherence by itself.
Assessment
This skill appears to do what it says: control a local Hupun RPA client over the MCP API on localhost and it requests no secrets. Before installing, verify the skill source (the package owner and repository) since the registry source is 'unknown' / no homepage is provided. Be aware the skill can perform disruptive local actions (relaunch, reinstall_browser, cancel_task, logout) — test it first with safe read-only commands like 'list_robots' and 'list_tasks'. Confirm how your agent evaluates the provided templates (they include JS-like map expressions) so you know whether code will be executed to format reports. If you have security policies, restrict the agent's network access to localhost only and monitor actions/logs produced by the skill. If you are not comfortable allowing automatic activation on keywords like “机器人”/“RPA”, disable auto-activation or require explicit invocation.

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

latestvk97ctd3hncvpzammtdk5jhnpb585exww
80downloads
0stars
1versions
Updated 4d ago
v1.0.0
MIT-0

湖畔 RPA 管理技能

本技能通过 MCP 协议与本地 RPA 客户端通信,提供机器人管理、任务控制等功能。

连接信息

  • MCP 服务地址:http://127.0.0.1:13765/mcp
  • 默认端口:13765(可在 RPA 设置中修改)

可用操作

1. list_robots - 列出所有机器人

列出当前系统中所有机器人及其状态。

参数:无

返回:机器人列表,包含 ID、定义ID、启用状态、冻结状态、最近任务信息


2. list_robot_defs - 列出所有机器人类型

列出所有支持的机器人类型基本信息。

参数:无

返回:机器人类型列表,包含 ID、名称、描述、图标、账号类型、平台标识

账号类型 (accountType)

  • ali - 阿里系(淘宝、天猫)
  • pdd - 拼多多

平台标识 (platformKey)

  • 0 - 淘宝
  • 28 - 拼多多
  • 1000 - 天猫

使用示例:说"查看有哪些机器人类型"或"列出机器人类型"


3. list_tasks - 查询任务列表

查询指定日期的任务执行记录。

参数

  • date(可选):日期,格式 YYYY-MM-DD,不传则查询当天

4. execute_robot - 执行机器人

触发机器人执行,支持多种指定方式。

参数

  • robotId:机器人ID,支持以下格式:
    • 字符串:单个机器人ID,如 "robot-abc123"
    • 数组:多个ID,如 ["robot-abc", "robot-def"]
    • 过滤条件对象:根据条件筛选
      • lastExecuteStatus:最后执行状态
      • lastExecuteTime:最后执行时间,可设为 "today"
      • robotType:机器人类型ID
      • enabled:是否启用

示例

触发所有启用的机器人:enabled: true
触发今天未执行过的机器人:lastExecuteTime: "today"
触发特定类型的机器人:robotType: "ali Mama"

5. cancel_task - 取消任务

取消正在等待或运行中的任务。

参数

  • taskId:任务ID,支持字符串、数组或过滤条件
    • 过滤条件:robotType - 按机器人类型筛选

6. relaunch - 重启应用

重启 RPA 客户端应用。正在执行的任务会被取消。

参数:无


7. check_update - 检查更新

检查应用是否有新版本。

参数:无


8. logout - 登出账号

登出当前登录的账号。

参数:无


9. send_notification - 发送通知

通过 RPA 客户端发送系统通知。

参数

  • title(必填):通知标题
  • content(可选):通知内容

10. set_top_notification - 设置顶部通知

在 RPA 客户端顶部显示通知栏消息。

参数

  • message(必填):通知消息,传空字符串可清除
  • duration(可选):显示时长(秒),不传则持续显示

11. reinstall_browser - 重装 Chrome

重新下载安装 Chrome 浏览器。正在执行的任务会被取消。

参数:无

12. daily_briefing - 今日简要

生成当日 RPA 任务执行情况摘要。返回 JSON 格式数据,本 Skill 负责格式化为 Markdown 展示。

参数:无

返回 JSON 格式

{
  "date": "2026-04-24",
  "stats": {
    "succeed": 8,
    "failed": 2,
    "running": 1
  },
  "details": {
    "succeed": [{ "name": "阿里妈妈账单", "count": 3 }],
    "failed": [{ "name": "拼多多账单", "count": 1, "reason": "Cookie失效" }],
    "running": [{ "name": "淘宝订单同步", "startTime": "15:30" }]
  },
  "comparison": {
    "totalChange": 2,
    "successRateChange": 5
  },
  "mainFailureReason": { "reason": "Cookie失效", "percent": 50 }
}

字段说明

  • stats - 执行统计(成功/失败/运行中数量)
  • details - 按机器人类型分组的执行详情
  • comparison - 环比昨日(仅当机器人类型数 >=10 且昨日任务数 >=10 时存在)
  • mainFailureReason - 主要失败类型及占比(仅当有失败任务时存在)

JSON 转 Markdown 格式规则

当用户问"今日简要"时,调用 daily_briefing 获取 JSON 数据,格式化为 Markdown:

# 📊 今日 RPA 简要 (${date})

## 📈 执行统计
| 状态 | 数量 |
|------|------|
| ✅ 成功 | ${stats.succeed} |
| ❌ 失败 | ${stats.failed} |
| 🔄 运行中 | ${stats.running} |

## 🤖 执行详情

### ✅ 成功
${details.succeed.map(x => `- ${x.name} x${x.count}`).join('\n')}

### ❌ 失败
${details.failed.map(x => `- ${x.name} x${x.count} → ${x.reason}`).join('\n')}

## 📉 环比昨日
| 对比 | 变化 |
|------|------|
| 总任务 | ${totalChange >= 0 ? '+' : ''}${totalChange} (${todayTotal}→${yesterdayTotal}) |
| 成功率 | ${successRateChange >= 0 ? '+' : ''}${successRateChange}% |

> ⚠️ 主要失败类型:${mainFailureReason.reason} (${mainFailureReason.percent}%)

失败归类规则

归类关键词
Cookie失效Cookie、登录失效、登录失败
网络超时超时、timeout、网络错误、请求失败
页面异常页面加载、元素未找到、DOM
脚本错误execute、script、脚本错误
系统错误系统错误、内存不足、磁盘空间
用户取消取消、cancelled

使用示例

查询所有机器人状态

直接说"查看机器人状态"或"列出所有机器人"

查看支持的机器人类型

说"查看有哪些机器人类型"或"列出机器人类型"

执行特定机器人

说"执行机器人 xxx"或"触发机器人 robot-abc123"

按条件执行

  • "执行所有启用的机器人" → execute_robot with { enabled: true }
  • "执行今天未执行的机器人" → execute_robot with { lastExecuteTime: "today" }
  • "执行阿里妈妈机器人" → execute_robot with { robotType: "ali Mama" }

查询任务

  • "查看今天的任务" → list_tasks(无参数)
  • "查看昨天的任务" → list_tasks with { date: "2026-04-23" }

取消任务

  • "取消任务 xxx" → cancel_task with { taskId: "xxx" }
  • "取消所有运行中的任务" → cancel_task with { taskId: { robotType: "xxx" } }

应用控制

  • "重启 RPA" → relaunch
  • "检查更新" → check_update
  • "退出登录" → logout

通知功能

  • "发送通知测试" → send_notification with { title: "测试通知", content: "这是一条测试消息" }
  • "显示顶部提示" → set_top_notification with { message: "你好", duration: 5 }

注意事项

  1. 执行机器人前:建议先查看机器人状态,确认机器人已启用且未在执行中
  2. 取消任务:只能取消等待中或运行中的任务,已完成的任务无法取消
  3. 批量操作:支持同时对多个机器人/任务进行操作
  4. 连接失败:如果 MCP 连接失败,请检查 RPA 客户端是否正在运行

Comments

Loading comments...