---
title: "Schedule recurring checks"
description: "Have your AI agent check your app's health and your running A/B tests on a schedule with the growth-advisor skill, and send the report to Slack, email, or your agent's chat."
---

> **AI agents**: to search Adapty docs faster and with fewer tokens, install the Adapty skills: `npx adapty skills install --yes`

The [growth-advisor skill](export-analytics-with-ai) can check your app for you every week: whether revenue is at its usual level, and whether a running A/B test is ready to decide. The check runs as a scheduled task in your AI agent. Each run reads your Adapty data and reports what changed, either in your agent or in a place your team already reads, such as a Slack channel. A run changes nothing in your account until you answer its report.

:::note
A check is a task in your AI agent's scheduler, not a setting in Adapty. It runs only while your agent can run scheduled tasks, for example while your computer is on, or in the cloud if your agent supports that.
:::

## Checks you can schedule

| Check | Default schedule | What the report says |
|---|---|---|
| **Your app's health** | Every Monday at 9:00 | How the last 7 days compare with a usual week: revenue, new paid subscriptions, new trials, installs, and refunds, with a verdict on top. It also lists anything due in your [backlog](autopilot), such as a test that ended without a result. A monthly check compares the last 28 days with the 28 before. |
| **One A/B test** | Every Monday and Thursday at 9:00 | Whether the test is **Not ready yet**, **Ready to decide**, or whether **Something changed** since the check was set up, such as an arm, a weight, or its audience. For each arm, how far it is from about 200 purchases. A leader is named only once it has settled. |
| **Every running A/B test** | Every Monday at 9:00 | One block per live paywall or flow A/B test, the ones ready to decide first: its placement, days run, each arm's purchases, and whether a leader has settled. |

You choose the schedule. The defaults are what the skill suggests when you don't say.

A check of one A/B test ends with the test. Once the test is stopped and its hypothesis has a result, the report says the check has nothing left to do and names the task to delete.

*Example report from a weekly check of a sample app. Your numbers will differ.*

**Agent:**

**Revenue has been 22% under its usual level since March 3, about $9k so far, and all of it is Google Play.**

| | Last 7 days | Usual week | Change |
|---|---|---|---|
| Revenue | $31.2k | $40.0k | **−22%** |
| New paid subscriptions | 410 | 425 | −4% |
| New trials | 1,180 | 1,210 | −2% |
| Installs | 24.1k | 24.6k | −2% |
| Refunds | $610 | $590 | +3% |

Buyers are coming in as usual, so the money is lost on renewals or on one store's payments.

Due in your backlog: "Annual at $33.99 on onboarding" is ready to decide.

**Next steps**

1. You: did anything ship on Google Play around March 3?
2. Find why Google Play revenue fell?

## Set up a check

To set up a check, ask your agent, for example:

```
Check my app every Monday and tell me if something is off.

Tell me when my onboarding A/B test is ready to decide.

Send a weekly report on my running A/B tests to the #growth channel in Slack.
```

The skill also offers a check on its own: after a [health check](export-analytics-with-ai#health-check), and after it links a live A/B test to its hypothesis. If a check for the same app already exists and your agent can list its tasks, the skill offers to change that check instead of adding a second one.

Before it creates the task, the skill asks two things:

1. **How often**: Keep the default schedule or name your own.
2. **Where the reports go**: By default, a report shows where your agent shows a scheduled run, such as its chat or notifications. To post reports elsewhere, connect a tool that can post a message or write a page, such as Slack, email, Teams, Notion, or Linear, in your agent's connector or MCP settings. Then name the exact channel, address, or page. The skill never picks one for you.

The skill then creates the task with your agent's own scheduling tool and names it after the check, for example `Adapty: health for Snapshot` or `Adapty: tests for Snapshot`.

### Where the check runs

- **On your computer**: The check uses your Adapty sign-in and your connectors on this computer.
- **On another machine or in the cloud**: Set the `ADAPTY_TOKEN` environment variable there to your [Adapty CLI token](developer-cli-authentication), and connect the reporting tool there too. The skill never puts a token in the task's prompt.

If a run can't reach your Adapty account, its whole report is one line asking you to sign in again on that computer or to set `ADAPTY_TOKEN` where the check runs.

### If your agent has no scheduler

The skill gives you the check's prompt to save as a file, and a line for your system's scheduler, such as cron. The line runs your agent in its non-interactive mode with that prompt and appends each report to a file you pick. For Claude Code, the line looks like this:

```bash
0 9 * * 1 cd ~/my-app && claude -p "$(cat adapty-weekly-health.txt)" --allowedTools "Bash" >> ~/adapty-checks.md
```

Other agents have their own non-interactive mode, such as `codex exec` or `gemini -p`. The run needs permission to run shell commands without asking, and each agent names that permission differently: see your agent's docs.

## Act on a report

A run only reads. It suggests no new hypotheses and changes nothing until you answer. Each action the agent can take for you is in **Next steps** as a question you can answer yes to, such as "Stop the test and keep Annual 2?". Steps only you can take start with "You:".

To act on a report, answer it in your agent: in the scheduled run's own chat, or by pasting the report into a new chat. A report posted to Slack or email can't be answered there, so its last line says where to answer.

Once you answer, the agent works as in any conversation with the skill. It reads the test and placement again first, because the report may be days old. Every change waits for your confirmation: stopping a test, marking a hypothesis won, lost, or inconclusive, or linking a new test.

## Change or stop a check

To change how often a check runs or where its reports go, ask the agent that set it up. To stop a check, delete its task in your agent's scheduler, or remove the cron line.