Health checks are read-only. The cart check verifies whether a capability is advertised; it does not add products to a cart or perform checkout actions.
The seven health checks
Shopify analytics access
Confirms that AgentWatch can submit an authenticated, read-only ShopifyQL query. An access warning commonly means report permissions or protected customer data approval are unavailable.
Global Catalog analytics
Confirms that Shopify accepts the Global Catalog analytics query. A successful query with no rows is valid and does not mean the catalog is unavailable.
AI commerce connection
Checks whether the storefront commerce endpoint returns a valid machine-readable response.
Product discovery tools
Checks whether the required read-only catalog search, lookup, and product tools are advertised.
Product visibility
Uses an active store product for a safe, read-only lookup and checks whether the returned product matches the merchant-owned product. If the store has no active product, the result is Unknown instead of a failure.
Cart capability
Checks whether the storefront advertises cart support for AI agents. No cart mutation is performed.
Policies and FAQs
Checks whether the policy and FAQ endpoint responds and advertises its required read-only discovery tool.
Health statuses
Healthy — Working as expected: the check passed.
Warning — Needs attention: the check returned an unexpected or unavailable result.
Critical — Currently unavailable: repeated failures have reached the critical incident threshold.
Unknown — Could not be verified: AgentWatch lacked a safe test condition, such as an active product.
Run checks manually
Select Run health checks at the top of the Health page. Manual checks have a one-minute cooldown. Running checks repeatedly does not make completed-day analytics arrive sooner.
The latest run shows:
The overall status.
When the run started.
Whether it was manual, scheduled, or installation-triggered.
Each check's response time and plain-language summary.
Expand View technical details when troubleshooting. These details are sanitized, but you should still avoid copying unrelated store or customer data into a support message.
When a health incident opens
AgentWatch avoids opening an incident for a single temporary failure:
Two consecutive failures open a Warning incident.
Three consecutive failures escalate it to Critical.
Two consecutive healthy runs resolve it automatically.
Some conditions are informational and do not create incidents. For example, blocked analytics access and the absence of a safe active product have their own setup or Unknown states.
Recent history
The Health page keeps the 10 most recent health runs visible. Use the history to distinguish a persistent problem from a one-time response or network issue.