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 "@playwright/test";
/**
* Scaffolding smoke test for the `ui` rig group.
*
* Deliberately minimal: it proves the harness works end to end — config
* resolves, a browser launches, the app is reachable, and JUnit lands where the
* rig expects it — without asserting anything about a specific screen. Real UI
* assertions go in sibling spec files.
*
* It skips rather than fails when nothing is serving, so running the specs
* directly on a machine with no platform reports "skipped", not a red build.
* The group is `optional` in groups.yaml for the same reason.
*
* That skip is gated on UNSTRACT_RIG_SESSION_ID, which `tests/rig/cli.py` sets
* unconditionally for every group that declares `requires_platform`. When it
* is present the rig booted the stack and `_wait_ready` already proved the
* frontend answers, so an unreachable app is a REAL failure — swallowing it
* there let the whole group report success without ever opening the UI.
*/
const RIG_MANAGED = Boolean(process.env.UNSTRACT_RIG_SESSION_ID);
test.describe("ui harness", () => {
test("the app serves a document", async ({ page, baseURL }) => {
let response;
try {
response = await page.goto("/", { waitUntil: "domcontentloaded" });
} catch (err) {
if (RIG_MANAGED) {
throw new Error(
`the rig reported the platform up, but nothing is serving at ` +
`${baseURL}: ${err.message}`,
);
}
test.skip(true, `no app reachable at ${baseURL}: ${err.message}`);
return;
}
// A redirect to the login page is a perfectly good "it's alive" — assert
// reachability, not a particular route.
expect(response?.status(), `unexpected status from ${baseURL}`).toBeLessThan(
400,
);
await expect(page).toHaveTitle(/.+/);
});
});