Health alerts

Attribution breaks quietly: a Meta token expires, a webhook stops being accepted, a theme update drops the script from the checkout page. Wexlyn checks every account every hour and keeps one alert per problem open until it clears.

What is checked

• Integrations in error: a platform that refused Wexlyn's last call needs reconnecting (critical).
• Sign-ins about to expire: a token that Wexlyn cannot refresh itself, such as Meta's 60-day token, raises a warning seven days before it expires and a critical alert once it has (critical).
• Webhooks failing: three or more events from one provider that arrived but could not be processed in the last 24 hours (warning).
• Conversions not sending: three or more conversions to one ad platform that failed or were skipped in the last 24 hours, with the most common reason (warning).
• Conversions stuck: anything queued for more than two hours (info).
• Outbound webhooks failing: three or more deliveries to your own endpoints gave up (warning).
• Tracking stopped or dropped: no page views in 24 hours on a site that averaged 20 a day, or a fall of more than 75% (critical / warning).
• Tracking percentage fell: a drop of 20 points or more between two checks (warning).
• A webhook integration went quiet: no events in seven days from a provider that used to send them (info).

How you are told

• The bell in the top bar counts open alerts; Notifications lists them with the reason, when they started and a link to the page where you fix them. Snooze hides one for a day; it comes back if the problem is still there.
• Owners and admins of the workspace receive an email when a critical or warning alert opens. Alerts close on their own when the next check finds the problem gone, so there is nothing to tick off.
• Check now on the Notifications page runs the check immediately after you fix something.

Have your AI assistant report it

The checks run inside Wexlyn; an AI assistant cannot wake up on its own. What it can do is read the result through the Wexlyn connector tool get_health, which returns the open alerts and the latest snapshot. In Claude, create a scheduled task such as "Every weekday at 8 am, check my Wexlyn health and tell me what needs attention", and it will call the tool and summarise. Grok and ChatGPT can do the same through their own scheduling, and all three can run the check on demand with "refresh".