Example: Build a skill

This article is for admins who set up a customer service agent in Customer Service Management.

These examples show how to turn a customer task into a skill — the description that tells the agent when to use it, the instructions that tell it what to do, and the tools it calls along the way. To create a skill in the product, see Create skills for your AI agents.

The tool labels below are illustrative. Replace them with the names of the tools available to your agent, and adapt the instructions to the information those tools accept and return.

Example 1: Investigate an unexpected charge

This skill uses two tools to compare a customer’s invoice with their payment records and explain an unexpected charge.

Skill name

Investigate an unexpected charge

Description

Use this skill when a customer asks about an unexpected charge or thinks they’ve been charged twice.

Tools

  • Get invoice details — retrieves itemized charges and the total amount due.

  • Get payment history — retrieves payments linked to the invoice, including their amounts and statuses.

Instructions

  1. Ask for the invoice number and clarify which charge the customer is questioning, if they haven’t already provided this.

  2. Use Get invoice details to retrieve the charges and total due. If the invoice can’t be found, ask the customer to check the invoice number before continuing.

  3. Use Get payment history with the invoice reference to check the associated payments. Compare their amounts and statuses with the invoice and the customer’s concern.

  4. Explain what the records show, including whether payments are completed, pending, or reversed. If a tool fails or the records don’t explain the charge, state what you couldn’t verify. Don’t guess or promise a refund.

For a request such as “I think I’ve been charged twice for invoice INV-123. Can you check?”, the agent retrieves the invoice, checks the associated payments, and explains whether the records show two completed payments or another situation, such as one completed payment and one pending.

Example 2: Raise a support request on a customer’s behalf

This skill identifies the customer, reuses what it already knows about them, creates a request, and hands off to a human agent when it can’t finish on its own.

Skill name

Raise a support request

Description

Use this skill when a customer asks to raise a support request. It identifies the customer, collects any missing details, creates the request, and hands off to a human agent if it can’t finish.

Tools

  • Customer lookup — retrieves a customer’s details from your customer accounts or Assets using an identifier such as an email address or customer ID.

  • Create request — creates a request on the customer’s behalf.

  • Human handoff — passes the conversation, with its history, to a human agent.

Instructions

  1. Check whether the customer has already given an identifier such as an email address. If they have, use it. If not, ask for one.

  2. Use Customer lookup to retrieve their details. Treat the stored details as the source of truth, and don’t ask for information you already have.

  3. Ask only for the details the request needs that aren’t already available, one question at a time.

  4. Make sure the customer’s issue is clear before continuing. Reuse what they’ve already described rather than asking them to repeat it.

  5. Use Create request to create the request on the customer’s behalf, then confirm it to the customer, including any reference the tool returns.

  6. If the customer can’t be found, a tool isn’t available, or the request can’t be created, use Human handoff to pass the conversation to a human agent, and include everything the customer has already shared.

Keep each instruction focused on a single action, and tell the agent to reuse information it already has. This keeps the conversation short for the customer and avoids asking the same thing twice.

Test your skill

While your agent version is a draft, test the skill by starting a conversation that matches the description and checking that the agent follows the steps and uses the right tools. Open the test conversation in Conversation review and select View steps to confirm the skill was invoked. When you’re happy with it, publish the agent version to make the skill available to customers. See Review your AI agent’s conversations.

Still need help?

The Atlassian Community is here for you.