Attaching Tools to your Agent and Writing Instructions

Attaching Tools to your Agent and Writing Instructions

Applies to: AGS 1.0  |  Level: Beginner  |  Reading time: 6 min
Part of: Using Tools with Your Agent (Overview)  |  Last reviewed: 25 Sep 2026

This guide shows you how to give an installed tool to a specific agent, and how to write the prompt instructions that tell the agent when and how to use it. Installing a tool only adds it to your workspace. No agent can use it until you attach it.

What you'll be able to do

  • Attach an installed tool, or a built-in System tool, to an agent
  • Write a prompt block that tells the agent when to use the tool
  • Handle success and failure so the agent never improvises on a live call

Before you begin

Part 1: Attach the tool

Step 1: Open your agent

Open the Dashboard (the top icon in the left navigation rail). Under Your Workforce, click the agent you want to edit.


Step 2: Open the Tools section

In the agent editor, the prompt is on the left and the Configuration panel is on the right. In the Configuration panel, expand Tools.


Step 3: Attach the tool

Click + Attach tool. Search for the tool you installed, or browse the System and Custom lists, then select it.



✔ Built-in tools need no install. Hang Up, Cold Transfer and Warm Transfer ship with the platform and appear in the System list. There's nothing to install or connect.

✔ Tool not there yet? Click Create a new tool at the bottom of the dialog to jump straight to the Custom HTTP Tool builder.

Step 4: Save the agent

Click Save.

✔ You should see: the tool listed in the Tools section of the Configuration panel. You can run it from its own test panel before you mention it anywhere in the prompt.
⛔ Not live yet: saving the agent doesn't change live calls. The tool only reaches callers after you deploy. See Testing and Deploying Tool Changes.

Part 2: Tell the agent when to use the tool

Attaching a tool makes it available. It doesn't make the agent use it well. That comes from the instructions in the agent's prompt (the large editor on the left of the agent editor).

For each tool, add a prompt block that covers these four things:

Cover thisWhat to write
1. When to use itThe situation that triggers the tool, and when not to use it.
2. What to collect firstEach value the agent needs, its format, and whether to confirm it with the caller.
3. What to say afterwardHow to report a successful result to the caller.
4. What to do when it failsThe exact words to say, and what not to reveal or retry.

Example prompt block

Copy this into your agent's prompt and adapt the tool name, inputs and messages.

# UPDATE TICKET
Use this when a caller asks to add information to a ticket they already have.
Do not use it to create a new ticket.

## Step 1 - Collect inputs
Ask for these one at a time. Don't proceed until you have both.
- ticketId - the ticket identifier (format: 2 letters + 4 digits, e.g. "FA1234").
If the input doesn't match that format, say so and ask again.
- comment - the note to add to the ticket.

Before calling the tool, read both values back to the caller and get confirmation.

## Step 2 - Call the tool
Call update_ticket with:
{
"ticketId": "{{ticketId}}",
"comment": "{{comment}}"
}

## Step 3 - Report the result
On success: confirm the update and read back the status message and any details
returned by the tool.

On failure: say exactly - "Sorry, we are unavailable at the moment, please
retry in a few minutes." Do not mention error codes, the tool name, or internal
details, and do not retry automatically.
⚠ Use the exact tool name. The name in the prompt (here update_ticket) must match the tool's name as it appears in the agent's Tools section.

Two rules that prevent most problems

⛔ Don't put fixed values in the prompt. Values that are the same on every call belong in an install variable (vendor tools) or a Static parameter (custom tools). The model can mistype, leave out or invent anything that's in the prompt.
⛔ Always write the failure path. An agent with no instruction for a failed tool call will improvise. On a live call, that's how you get confident, incorrect answers.

Verify it worked

  • The tool is listed in the agent's Tools section.
  • The tool runs successfully from its own test panel.
  • The prompt has a block for the tool that covers all four points in Part 2.

Next step

Continue with Testing and Deploying Tool Changes to check the tool behaves correctly before it reaches real callers.

Still stuck? See Troubleshooting Tool Issues or contact support.