All jobs

Which Checkout Errors Did Shoppers Hit This Week? An Online Store's Monday Report From Axiom Logs

An online store's agent queries its storefront logs in Axiom every Monday and lists last week's checkout errors with counts, pages and first-seen dates.

Run this on RushOpens the console with this prompt ready to send. Connect Axiom there first.

What the agent does

This is for an online store, say one selling outdoor gear, whose storefront sends its request and server logs to Axiom. The store has no engineer on staff: a freelance developer looks after the site a few hours a week, and the owner finds out about a broken checkout from a customer email or a dip in sales. The agent reads last week's logs every Monday and writes the owner and the developer a short list of what went wrong at checkout, with enough detail for the developer to start. Someone connects Axiom once with POST /me/connections {"type": "mcp", "name": "axiom", "url": "https://mcp.axiom.co/mcp", "auth": {"mode": "oauth"}} and signs in to Axiom in the browser.

The agent calls listDatasets and getDatasetFields on the dataset named in its input, so it uses the field names the storefront actually sends (status, path, message, error.type or whatever they are) rather than guessed ones, and stops with a clear note if a field it needs is gone. It calls getSavedQueries too: when the developer has already saved a checkout query, the agent starts from it.

It then runs APL with queryDataset over the seven days before Monday: requests whose path starts with /checkout or /cart that answered 5xx or 4xx other than 404, grouped by path, status and error message, with a daily count from bin(_time, 1d). For each group with more than a handful of hits it runs a second query for the earliest time it appears in the last 30 days, so a new error is told apart from one that has been there all month. It calls checkMonitors and getMonitorHistory to say whether one of the store's monitors already fired on it.

It writes checkout-errors.md, newest problems first: the path, the status, the message as logged, how many requests hit it each day, when it first appeared, whether a monitor fired, and the APL that found it so the developer can run it in Axiom and see the same rows. A message that includes a shopper's email or address is shortened to its error type. The agent creates, changes and deletes nothing in Axiom: no dashboard, monitor, notifier or dataset field.

Start it with one API call

POST https://api.phoenix.so/o/{org}/sessions with this body:

{
  "agent": "claude",
  "input": "Write last week's checkout error report from Axiom dataset storefront-logs, for 2026-11-09 00:00 to 2026-11-16 00:00 America/Denver. Use the Axiom connection. First call getDatasetFields on storefront-logs and use the real field names for path, status and error message; if one is missing, stop and say which. Call getSavedQueries and reuse any saved query about checkout. With queryDataset (APL), find requests whose path starts with /checkout or /cart and whose status is 500 or above, or 400 to 499 except 404; group by path, status and message, with a daily count using bin(_time, 1d). For each group with 5 or more hits, query the last 30 days for its first appearance. Call checkMonitors and getMonitorHistory and say whether a monitor fired on each group. Write checkout-errors.md: new groups first, then by count; for each, path, status, message, daily counts, first seen, monitor, and the APL you ran. Replace any email address, phone number or street address in a message with its error type. Do not create, update or delete any dashboard, monitor, notifier or dataset field.",
  "connections": [
    "axiom"
  ]
}
  1. Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
  2. Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
  3. Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.

What comes back

The session returns checkout-errors.md as an artifact and a two-line summary: how many error groups, and which are new this week. The owner reads the summary, and the developer opens the file and runs each APL query in Axiom before deciding what to fix first.

What they check: that a group is a real shopper-facing failure and not a bot hammering an old cart URL, that the dataset covered the whole week (a gap in logging reads as a quiet week), and the first-seen date of anything new, which usually lines up with a theme update, an app install or a payment-provider change. The report counts requests from the logs; how many orders it cost is not something the logs show, and the agent does not guess.

Run it on every Monday morning

In the console, a Project automation with a schedule trigger (0 7 * * 1 with the store's timezone) runs this prompt every Monday. A schedule trigger runs the saved prompt as written and fills in no dates, so the saved prompt asks for "the seven days ending at midnight before this run" in place of the dates above. Its notify_each_run set to true sends each Monday's result on its own.

During a sale the owner can start a one-off session from the console with the same prompt and its window cut to the last few hours. Each run is its own session, so this week's report can be read beside last week's to see whether a fix held.

The use case this job follows, with its automation and trigger: Produce a report or file, no repo.

Why run it on Rush?

Axiom's monitors fire on one threshold each; the owner's question is broader: what went wrong for shoppers last week, which of it is new, and is anyone already looking. The agent asks Axiom each part of that and writes it in the store's terms. The Axiom sign-in stays in Rush's connection, and because the agent acts as the signed-in person, the query cost limits set on that person's Axiom role cap what it can spend on queries.

Each run has its own computer, a transcript of every query it ran and the rows that came back, and the cost of its model calls on the session. When the developer asks why a group was or was not in the report, the run for that week shows the exact APL.

Frequently asked questions

Our Axiom organization is in the EU. Does this work?

Axiom also runs an EU server, https://mcp.eu.axiom.co/mcp, with the same sign-in, which keeps query results in EU data centres. Connect that address instead and name the connection on the request. Axiom's docs say non-query requests, such as listing datasets, still go to its US API on both servers.

Can the agent change our monitors or dashboards?

Axiom's server has tools that create, change and delete dashboards, monitors and notifiers, and one that deletes dataset fields. The prompt forbids all of them, and the transcript would show any such call. To rule it out entirely, connect Axiom as a person whose role cannot change them.

Will it see customers' personal data in the logs?

It reads whatever the storefront logs, which is why the prompt tells it to replace emails, phone numbers and addresses with the error type in the report. The better fix is at the source: keep personal data out of the storefront's log lines.

We do not know APL. Is that a problem?

No. The agent writes the queries, and the report includes each one so the developer can run it in Axiom. A query the developer saves there is picked up on the next Monday through getSavedQueries.

More Axiom jobs

More jobs like this: Produce a report or file, no repo

Keep reading