Develop and use a SKILL in QoderWork

Background, challenges, and solution

Enterprises often face the following challenges with agents:

• Low accuracy: Agents often fail to meet the high accuracy standards required by enterprises. When an agent operates business-critical software, it cannot guarantee precise outcomes. Errors can cause significant financial losses, such as during supply chain customs declarations or e-commerce product listings.

• Slow execution: Agents are often too slow for production environments. The agent's perception-planning-action cycle can result in lengthy trial and error or slow operation of complex enterprise applications.

• High token consumption: Agents consume many tokens, making them costly. Even for repetitive tasks, the AI model must be involved throughout the process.

To address these challenges, this guide proposes the following approach:

• Use Robotic Process Automation (RPA) to turn standard enterprise procedures into repeatable automation flows. The agent can then use these flows, which are integrated as SKILLs.

• When users interact with the agent, it can use these SKILLs to deliver tangible business value.

Use case

Business requirement

• Operations staff need to analyze product sales data daily. They face two main challenges: First, products are sold across multiple platforms, requiring manual download of sales data from each store before analysis. Second, some analysis is routine, but manual repetition is tedious and inefficient.

Implementation approach

• Develop a SKILL that includes two processes: data collection and analysis. Data collection is handled by an automation flow developed with RPA, and the analysis requirements are explicitly defined within the SKILL.

• Operations staff use this SKILL in QoderWork.

• Runtime architecture

  

Procedure:

1. Develop an automation flow using RPA and publish it as an MCP Tool.

2. Use an RPA service-based robot as the runtime for the SKILL.

3. Add the MCP Server to QoderWork.

4. Write the SKILL and add it to QoderWork.

Procedure

Publish an automation flow as an MCP Tool

1. For detailed instructions on developing an automation flow with RPA, see Develop an automation flow.

2. The flow input and flow output for the automation flow in this example are as follows:

   Automation flow	Input	Output
   Automatically retrieve store product sales data	statis_date: Query product sales data for the specified date	The retrieved product data is in JSON format and includes the product ID, product name, style tags, units sold, unit price, favorites count, and more for each store. The automation flow saves the retrieved data using task_result.

   Important

   The specific implementation of the automation flow in this example is not universally applicable because each enterprise uses different e-commerce platforms, requires different data, and operates different software. You can refer to Develop an automation flow to create an automation flow tailored to your business processes.

3. Publish the automation flow as an RPA application. For details, see Publish and manage applications.

   

4. Publish the application as an MCP Tool. For the procedure, see Publish as an MCP Tool. The information for the published MCP Tool is shown below.

   

   Important

   If the RPA application takes a long time to run, we recommend that you specify the estimated duration and the agent's polling interval in the Tool Description.

Prepare the SKILL runtime

This example uses an RPA service-based robot as the runtime for the SKILL.

An RPA service-based robot requires Wuying Workspace. To configure it, see Run automated processes as a service. After you create the robot, it appears as shown below. You will use this robot to run the SKILL.

Add an MCP server in QoderWork

1. In the RPA console, go to the MCP Server menu to retrieve the MCP Server configuration. This configuration might list multiple MCP Servers.

   

2. In QoderWork, go to Settings > MCP Service and select Import from JSON. Once imported, at least two new MCP Servers will appear.

   

Add the SKILL to QoderWork

This section shows you how to add a SKILL in QoderWork that helps the agent understand user intent and accurately complete tasks.

1. Prepare the SKILL.md file.

   Note

   The SKILL content in this guide is for reference only. You must modify it to fit your specific business scenario.

   ---
   name: rpa-skill
   version: 1.0.0
   description: An operations assistant. Use this SKILL for product sales data analysis and for setting and using the RobotId.
   ---
   
   # Product sales data analysis
   
   ## Analysis process
   
   1. Requires the user to provide a date in YYYY-MM-DD format, such as 2025-10-02.
   2. Calls list_product_metrics via MCP to get the product sales data for the specified date.
   3. Generates a sales analysis report in Markdown format, including:
      - Overall sales summary
      - Sales performance by store
      - Top 10 products by sales revenue
      - Best-selling products by store
      - Conversion rate analysis
      - Price range analysis
   
   # Pre-use setup
   
   ## Set and use RobotId
   
   1. Before use, the user must set the RobotId. The RobotId is a string of letters and numbers that can be found in the Alibaba Cloud RPA console.
   2. The RobotId is read from the `ALIYUN_RPA_RobotId` variable in the config/.env file.
   3. If `ALIYUN_RPA_RobotId` is empty, prompt the user to provide it. Once provided, save it to the config/.env file.
   4. If `ALIYUN_RPA_RobotId` exists, it is used by default for all sessions.
   5. If the user requests to set the RobotId, update the `ALIYUN_RPA_RobotId` in the config/.env file.
   
   
   # Usage notes
   
   ## MCP calls
   
   - If an MCP call requires a RobotId, it preferentially reads the `ALIYUN_RPA_RobotId` from the config/.env file.

2. Add the SKILL in QoderWork, as shown below:

   

3. Configure the RobotId.

   To avoid providing the robot ID each time, set the robot that will run the SKILL. In a QoderWork session, set the RobotId to the ID of the RPA service-based robot you created earlier, as shown below:

   

This completes the configuration process.

Results

In a QoderWork session, enter a prompt to analyze the product sales data for a specific day, as shown below: