---
title: Launch checklist
description: "The pre-ship checks an agent has to pass before it is published: what is checked, what Required means and how to re-run it."
---

The launch checklist is the list of checks the platform runs against the agent's
configuration before it is published. It sits in **Publish**, under **Release** in
the workspace sidebar, the same section the **Publish** button in the header opens.
Run it before you attach the agent to a phone number, a campaign or a deployment;
fix every item marked **Required** first. The sidebar entry shows a count of failing
and warning checks.

## How it works

The section has a status card, the **Publish agent** button and the list of checks.
Opening the section runs the checks; **Publish agent** reads **Checking…** until they
finish. Each check scores pass, warn or fail and the card sums them up: **N Pass**,
**N Warn**, **N Fail**. The card's title says where the agent stands:

| Card | Meaning |
|---|---|
| **Ready to publish** | Every required check passes. With no warnings it adds **All required checks pass. Publish to take your changes live.**; with warnings it asks you to review them, or publish anyway. |
| **Not ready to publish** | A required check fails. **Publish agent** is disabled. |
| **Up to date** | No required check fails, and there are no unpublished changes to publish. |
| **Run the launch checklist** | The checks have not run for this agent yet. The button reads **Run checklist** instead of **Publish agent**. |
| **Checks unavailable** | The checks could not be loaded. Publishing stays disabled until they can. |

![The status card with both required checks passing.](/media/test-and-improve/launch-checklist-ready.webp)

## The checks

| Check | Badge | What it looks for | How to fix a failure |
|---|---|---|---|
| **Prompt is configured** | Required | The system prompt is not empty. The app's own words: the system prompt defines how the agent behaves, so it can't be empty. | Fill in **Agent Identity** and **Tasks** under **Settings**, or write the flow for a flow agent, and save. |
| **At least one test scenario** | Required | The agent has 1 or more scenarios under **Evaluation**. | Create one; **AI Generate** takes a few seconds. See [Evaluations](/test-and-improve/evaluations). |

A **Required** check that fails disables **Publish agent** behind a **Not ready to
publish** banner. Saving is never blocked; publishing is.

![The two required checks.](/media/test-and-improve/launch-checklist-checks.webp)

## Run it

1. **Save the agent**

    The checks read the saved configuration. If **Save** is enabled, save first.

2. **Open the checklist**

    Under **Release**, choose **Publish**, or choose **Publish** in the header. The
    checks run as the section opens and the status card updates within a second or
    two.

3. **Fix and repeat**

    For each failed item, follow the fix in the table above, save, and open
    **Publish** again to re-run the checks.

## Verify

- The status card reads **Ready to publish**, or **Up to date** once published, and
  **Fail** is 0.

## Troubleshooting

- **At least one test scenario fails although I wrote one.** The scenario was not
  created; the **Create Scenario** button stays disabled until name, a step and
  success criteria are all filled. Check **Existing Scenarios** on the **Evaluation**
  page.
- **Prompt is configured fails on a flow agent.** The flow has no instruction on its
  start node. See [Nodes](/build/flow-agents/nodes).
- **The card says Run the launch checklist.** Choose **Run checklist**.

## Related

- [Evaluations](/test-and-improve/evaluations)
- [Publishing](/test-and-improve/publishing)
- [Picking a channel](/deploy/picking-a-channel)
- [Prompting guide](/build/prompting/prompting-guide)