Onboarding

Agents / Onboarding is where the tests you already have arrive: from a document, a spreadsheet, a requirements tool or a folder full of Word files. Your assistant reads them and writes them here as procedures, test cases, sentences and data points — things that run.

An import is an event, not a state. Every import stays as its own record, and the page is where somebody looks in three months to answer the question that always comes back: where did this test case come from?

What the page shows

Per run: what was imported, with the procedure it landed in, how many steps and how many of them are checks, what happened to it — created, updated, left as it was, deleted — and the source it was read from, as a link where there is one.

Below that: what was adapted — a test case taken over, but not word for word: what differs, from what to what, why, and whether the meaning stayed the same. Where the meaning changed, the person who agreed is named; without that name the run is not accepted. The test case's name leads to it. Then what was not written, each with the reason. A line that could not become a step is worth more here than a test case that pretends to run.

Before an assistant hands the run in, it checks each procedure it imported into with check_procedure and runs one test case of each as a dry run; if it has not, the check before handing in says which are missing.

Every run names the source it read — with its name as a link and its scope, the set or folder it was taken from — and each test case carries its own Source line in its description. That line is never translated: the next import looks for it literally.

An import keeps the order and meaning of each test as the source has it. Where a source sets a condition after the event that should cause the reaction, or builds one condition out of two such events, the assistant brings the test in as it is and says so — a coverage run or you decide. It may propose the corrected order alongside; it does not rearrange anything on its own.

What you do here

• Schedule onboarding asks your assistant for an import: choose the source of the test cases — one that is recorded, or a new one with its link and the part of it that is meant. The request waits in the Agent queue; tell your assistant Work through the open requests in XPressRT. It is closed by the onboarding run that read from that source.
• Read what came in, and open a test case from the list.
• Check the refused ones — usually a sentence that has no step, or a value nobody could resolve.
• Export the run as HTML when somebody outside needs the record.

What to expect of an import

Test cases arrive tagged ai-generated: an assistant wrote them and nobody has read them back yet. Removing the tag is the review. What the import creates — sentences, data points, block files — is written in English, whatever language you speak with the assistant; only quotes from your source stay as they are.

An import never changes a result, and it never starts a real run.

A whole set at once

An assistant does not have to create a set one call at a time. It writes a plan — a JSON file in your workspace, usually under local/data/ — that holds what is to come in: data points, step sentences, procedures, test cases, preparations. Then one call brings it in.

Two things about it are worth knowing, because they decide what you can rely on:

• It may do exactly what one call at a time may do. Every piece goes through the same door as its single counterpart and through the same checks: a sentence that does not exist, a data point in the wrong direction, a placeholder left in a test case — everything that is refused singly is refused here. New is the number, not the permission.
• All of it or none of it. At the first no, your workspace stands as it did before, and the answer says which piece it was and why. A half-imported set is the state nobody can say anything about afterwards — not you, and not the assistant that wrote it.

The plan also has a place for what the assistant could not place, with a reason. That part is not written; it is the honest half, and it comes back in the answer.

The plan file is not the record. What somebody opens in three months to find out where a test case came from is this page — so an assistant that imports also hands in an onboarding run. XPressRT says so by itself: a workspace with test cases that carry a Source: line and no onboarding run recording them is named in the assistant's answers, at three, ten, twenty-five of them.