AI Skills

​

When should you create a Skill?

Choose workflows where AI can meaningfully reduce manual work and where the expected process can be described clearly.

​

Good candidates

• Repetitive, high-frequency tasks currently handled by your team
• Tasks that require several questions or checks before completion
• Processes with clear business rules, required information, and exit conditions
• Tasks that need data from an external system through a data connector
• Processes that may need a human handoff under specific conditions

​

Before you begin

• Select one narrowly defined business task
• Document the correct process and important edge cases
• Confirm which customer information the task requires
• Prepare and activate any required data connectors
• Define when the AI must stop or hand the conversation to a human

---

​

Create and manage Skills

​

Open the Skills page

1. Open the AI Agent you want to configure.
2. Open Settings, then select the Skills tab.
3. Select New Skill.

Open the YCloud Skills page

​

Manage existing Skills

The Skills list shows each Skill’s name, Start trigger, last update time, and status.

• Search: Find a Skill by name.
• Used / Unused: Use the status switch to control whether the AI Agent can use the Skill.
• Edit: Select a Skill name to update its trigger or instructions.
• Delete: Open the Action menu for a Skill and select Delete.

---

​

Configure a Skill

​

1. Enter a clear title

Name the outcome or task, not a broad department or topic. Clear titles make Skills easier to find and maintain.

​

2. Define the Start trigger

The Start trigger tells the AI Agent when and when not to use the Skill. Include the customer’s intent, required context, and important exclusions.

Use this Skill when a customer clearly asks to cancel an eligible order and receive a refund. Do not use it for return-policy questions, payment failures, or orders that have already shipped.

​

Start trigger checklist

• Describe the customer’s intent using natural language.
• Include common variations of the request where helpful.
• State when the Skill must not run if a similar request requires a different process.
• Avoid vague triggers such as “when needed” or “for support questions.”

​

3. Write Step-by-Step instructions

Describe the complete procedure the AI Agent should follow. Each step should have one clear purpose and begin with an action.

1. Acknowledge the customer’s request.
2. Collect the information required to identify the order.
3. Check whether the request meets the refund policy.
4. Use the appropriate data connector to retrieve or update the order.
5. Confirm the result and explain what happens next.
6. Hand off or close the conversation when the relevant condition is met.

​

Writing best practices

• Number the steps in the order they should run.
• Keep each step focused on one action or decision.
• Specify required information instead of expecting the AI to infer it.
• Define what to do when data is missing, invalid, or contradictory.
• State completion and handoff conditions explicitly.
• Split long or conditional instructions into smaller steps.

---

​

Use commands in Skill instructions

In the Step-by-Step instructions editor, type / to insert an available command. The menu shows built-in actions and the active data connectors available to the AI Agent.

​

Built-in actions

To insert an action, type / in Step-by-Step instructions, then select it under Built in actions. The action appears as a command in the Skill instructions.

​

Data connectors

Data connectors allow a Skill to query or update external business data. Only configured and active connectors appear in the command menu. For each connector step, clearly describe:

• When the connector should be used
• Which input values should be collected or passed
• How the AI Agent should interpret the result
• What to do when the request fails or returns no matching data
• Whether the customer must confirm before data is changed

---

​

Preview, publish, and improve

​

Test in Preview

Use the Preview panel while editing the Skill. Test both requests that should trigger the Skill and similar requests that should not trigger it. Restart Preview when you need a clean conversation.

​

Minimum test checklist

• A typical customer request triggers the correct Skill
• A similar but excluded request does not trigger the Skill
• The AI Agent collects all required information before taking action
• Steps run in the intended order
• Data connector success and failure paths behave correctly
• The Skill hands off when human judgment or approval is required
• The conversation closes only after the task is complete
• The AI Agent does not invent data, eligibility, or action results

​

Enable and publish

1. Return to the Skills list and set the Skill status to Used.
2. Select Publish to make the Agent’s Skill changes live.
3. Test the published Agent in the customer channel where it will be used.
4. Review real conversations and refine the Start trigger or instructions when you find recurring failures.