Overview

A Composite skill allows you to combine multiple individual skills into a single, cohesive workflow. This approach simplifies the execution of tasks that require a sequence of actions.

Key Benefits

• Incremental Skill Addition: Build composite skills by adding individual skills step-by-step for customized workflows.

• Sequential Execution: Execute selected skills in order, like calling an API, processing output, and then publishing results.

• Efficiency Improvements: Simplifies workflow by combining multiple steps into one composite skill, reducing manual sequencing.

Steps to Add a Composite Skill

Refer to this article for detailed steps on adding a skill to an AI Agent.

1. On the Skills window, click Add Skill in the top right corner of the window and select Composite from the dropdown.

   

2. On the Composite Skills window, enter the Name and Description.

   Guidelines for Skill Names:

   • Be Clear and Direct: Use action-oriented, unambiguous verbs.

   • Keep It Concise: Aim for 2-4 words that clearly describe the function's purpose.

   • Avoid Generic Names: Ensure the name is distinct from other functions.

   • Use a Noun or Output Indicator: Reflect the returned value (such as fetch_exchange_rate, not convert_money).

   Guidelines for Skill Descriptions:

   • Start with a Clear Action Verb: Use verbs like "Retrieves," "Fetches," or "Generates."

   • Clearly State Function Purpose: Describe what the function does without assuming prior knowledge.

   • Mention Key Inputs: Include relevant inputs, but avoid specifying types (handled separately).

   • Specify Expected Output: Clarify what the function returns.

   • Keep it Concise: Aim for a description under 20 words.

3. Select Skill Execution Behaviour as Standard or Parallel. Select the skill to be added from the Skill Flow dropdown. To add multiple skills, click the Plus icon and add.

   1. Standard: In this behaviour, the skills are executed one after the other as per the workflow.

   2. Parallel: In this behaviour, the skills are executed simultaneously.

      

4. After selecting the required skills, click Save to create a Composite Skill for your AI Agent.

Applications of Composite Skills

Making Multiple API Calls Efficiently

For example, a software application needs to retrieve data from ten different APIs to provide a quick response to the user. The developer creates a composite skill that includes all ten API calls. During this setup, the option for parallel execution is selected to allow simultaneous processing.

When the composite skill is triggered, it initiates all ten API calls at once rather than waiting for each one to complete sequentially. For example:

• API Call 1: Retrieve data from API A.

• API Call 2: Retrieve data from API B.

• API Call 3: Retrieve data from API C.

• API Call 4: Continue this for APIs D through J.

As each API responds, the outputs are collected and stored. This process is managed within the composite skill, which efficiently handles incoming data from multiple sources. By executing all API calls in parallel, the total response time is significantly reduced. Instead of taking approximately one minute to complete all calls (if each takes 5 seconds), the composite skill can return results in about 10 seconds.

Summarizing Outputs from Multiple Systems

For example, a data analyst needs to gather and summarize information from ten different systems to generate a comprehensive report. The analyst creates a composite skill that integrates ten individual skills, each corresponding to one of the systems. This composite skill organizes the workflow and allows for sequential execution.

The composite skill is designed to call each system one after the other. For instance:

• Step 1: Call System A and retrieve data.

• Step 2: Call System B and retrieve data.

• Step 3: Continue this process for Systems C through J.

After each system's output is retrieved, the composite skill summarizes the results in smaller, manageable chunks, allowing for better token management during processing. For example, each system's output is processed and temporarily stored before moving on to the next system.