LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Python ai-agentsdata-engineeringdocument-aigenerative-aiidpjson-extractionllmmcp-serverocrpdf-extractionprompt-engineeringstructured-output
import { expect, test } from "./fixtures.js";
/**
* Workflows — authoring and the API-deployment surface.
*
* Covers the UI side of `workflow-author` and `api-deployment-provision`. The
* execution half (`workflow-create-execute`) is API-driven and already lives in
* tests/e2e/workflows; duplicating it through a browser would be slower and
* flakier for no extra signal.
*/
test.describe("workflows", () => {
test("the listing loads and offers workflow creation", async ({
page,
app,
}) => {
await app.goto("workflows");
await expect(
page.getByRole("button", { name: /New Workflow/i }),
"no way to create a workflow",
).toBeVisible();
});
/*
* Read-only check of the `workflow-author` entry point: the form opens and
* enforces its required fields, so a half-built workflow cannot be POSTed.
*/
test("the create-workflow form requires its mandatory fields", async ({
page,
app,
}) => {
await app.goto("workflows");
await page.getByRole("button", { name: /New Workflow/i }).click();
const dialog = page.getByTestId("new-workflow-modal");
await expect(dialog).toBeVisible();
/*
* Asserting the dialog STAYS OPEN, not that a particular message appears.
* Verified against a running app: submitting this form empty renders no
* error text at all — the guard is that it refuses to proceed, which is
* what actually protects against a half-built workflow being created.
*/
// "Create Workflow" here, "Edit Workflow" in the edit flow — same button.
await dialog.getByTestId("new-workflow-modal-ok").click();
await expect(dialog, "empty form was accepted").toBeVisible();
await expect(page).toHaveURL(/\/workflows(\/|$|\?)/);
});
/*
* Opening a workflow must reach its builder. The Actions menu is the entry
* point for running it and for viewing file history, so its presence is the
* cheapest proof the detail page wired up rather than half-rendering.
*/
test("a workflow opens and exposes its actions", async ({ page, app }) => {
await app.goto("workflows");
// Was `.list-view-row` — dead since the ResourceTable migration, which
// made this test skip instead of fail. See prompt-studio.spec.js.
const card = page.locator('[data-testid^="workflow-list-row-"]').first();
if ((await card.count()) === 0) {
test.skip(true, "no workflows in this org");
}
await card.click();
const actions = page.getByTestId("workflow-actions-btn");
await expect(actions, "workflow detail has no Actions menu").toBeVisible({
timeout: 20000,
});
// The menu is portalled; its entries derive their ids from the Dropdown's.
await actions.click();
await expect(
page.getByTestId("workflow-actions-item-run-workflow"),
).toBeVisible();
});
});
/**
* API deployments — `api-deployment-provision` from the UI side.
*
* The deployment's endpoint and key are what downstream callers depend on, so
* the listing surfacing them is the thing worth guarding here. Actually calling
* the endpoint is covered API-side in tests/e2e/api_deployment.
*/
test.describe("api deployments", () => {
test("the listing loads and offers a new deployment", async ({
page,
app,
}) => {
await app.goto("api");
await expect(
page.getByRole("button", { name: /API Deployment/i }).first(),
"no way to create an API deployment",
).toBeVisible();
});
test("a deployment exposes its endpoint and management controls", async ({
page,
app,
}) => {
await app.goto("api");
/*
* Scope to ONE deployment card rather than taking `.first()` of a
* page-wide `[role=switch]`: that locator would happily match a switch
* belonging to a different card, or to no card at all.
*/
const card = page
.locator('[data-testid^="api-deployment-list-card-"]')
.first();
if ((await card.count()) === 0) {
test.skip(true, "no API deployments in this org");
}
await expect(card.getByText(/API Endpoint/i).first()).toBeVisible();
// The enable/disable toggle is the control that takes a deployment in and
// out of service — the single most consequential switch on this page.
await expect(
card.locator('[data-testid^="api-deployment-toggle-"]'),
"deployment has no enable/disable control",
).toBeVisible();
});
});