clawmates/frontend/tests/e2e/p1-chat.spec.ts

import { expect, test, type Page } from "@playwright/test";

// P1 exit criterion (spec §17): a multi-session conversation with
// replayable step traces, end-to-end against the real backend running the
// scripted provider (deploy/e2e/scenarios.toml).

const OWNER_EMAIL = "[email protected]";
const OWNER_PASSWORD = "e2e-password";

async function signIn(page: Page) {
  await page.goto("/login");
  await page.getByLabel("Email address").fill(OWNER_EMAIL);
  await page.getByLabel("Password").fill(OWNER_PASSWORD);
  await page.getByRole("button", { name: "Sign in" }).click();
  await page.waitForURL((url) => !url.pathname.endsWith("/login"), { timeout: 15000 });
}

async function openScout(page: Page) {
  await page.getByRole("link", { name: /Scout/ }).click();
  await expect(page).toHaveURL(/\/claws\/.+\/chat\//);
}

/** Clicks "New" and waits for the navigation: the previous session may show
 *  the same welcome heading, so the URL change is the reliable signal. */
async function newSession(page: Page) {
  const before = page.url();
  await page.getByRole("button", { name: "New" }).click();
  await page.waitForURL((url) => url.toString() !== before);
}

async function sendMessage(page: Page, text: string) {
  const box = page.getByLabel("Message Scout");
  await box.fill(text);
  await box.press("Enter");
}

test("welcome state appears for a fresh session", async ({ page }) => {
  await signIn(page);
  await openScout(page);
  await newSession(page);
  await expect(
    page.getByRole("heading", { name: /Hi, I'm Scout/ }),
  ).toBeVisible();
  await expect(
    page.getByRole("button", { name: "Create a daily briefing" }),
  ).toBeVisible();
});

test("a message streams back a scripted reply", async ({ page }) => {
  await signIn(page);
  await openScout(page);
  await sendMessage(page, "hello [[scenario:hello]]");
  await expect(page.getByText("hello [[scenario:hello]]")).toBeVisible();
  await expect(
    page.getByText("Hello! I'm Scout, your research analyst."),
  ).toBeVisible();
});

test("tool runs render a step trace that survives reload", async ({ page }) => {
  await signIn(page);
  await openScout(page);
  await newSession(page);
  await expect(
    page.getByRole("heading", { name: /Hi, I'm Scout/ }),
  ).toBeVisible();

  await sendMessage(page, "what time is it? [[scenario:tool-time]]");
  await expect(page.getByText(/I checked the current time/)).toBeVisible();

  const trace = page.getByRole("button", { name: /1 step/ });
  await expect(trace).toBeVisible();
  await trace.click();
  await expect(page.getByText(/clock\.now/)).toBeVisible();

  // Reload: history replays the identical transcript and trace (§13).
  await page.reload();
  await expect(page.getByText(/I checked the current time/)).toBeVisible();
  const replayedTrace = page.getByRole("button", { name: /1 step/ });
  await replayedTrace.click();
  await expect(page.getByText(/clock\.now/)).toBeVisible();
});

test("multiple sessions hold separate transcripts", async ({ page }) => {
  await signIn(page);
  await openScout(page);

  // First session.
  await newSession(page);
  await expect(
    page.getByRole("heading", { name: /Hi, I'm Scout/ }),
  ).toBeVisible();
  await sendMessage(page, "first session message");
  await expect(page.getByText("I received: first session message")).toBeVisible();

  // Second session.
  await newSession(page);
  await expect(
    page.getByRole("heading", { name: /Hi, I'm Scout/ }),
  ).toBeVisible();
  await sendMessage(page, "second session message");
  await expect(
    page.getByText("I received: second session message"),
  ).toBeVisible();
  await expect(page.getByText("first session message")).not.toBeVisible();

  // Switch back via the sessions column (?sessions=1, §6). Rows sort most
  // recently active first: [current, first-session, original]; the first
  // non-active row is the session we just left.
  await page.getByRole("button", { name: "Sessions" }).click();
  const column = page.getByRole("complementary", { name: "Sessions" });
  await expect(column.getByRole("listitem").first()).toBeVisible();
  await column
    .locator("button:not([aria-current])")
    .filter({ hasText: "Untitled" })
    .first()
    .click();
  await expect(page.getByText("I received: first session message")).toBeVisible();
});

test("the 3-step wizard creates a live claw and lands in its chat", async ({
  page,
}) => {
  await signIn(page);
  await page.getByRole("link", { name: "New claw" }).click();

  // Step 1: identity (accent swatch + name + title).
  await page.getByRole("radio", { name: "Accent #65a8f9" }).click();
  await page.getByLabel("Name", { exact: true }).fill("Drafter");
  await page.getByLabel("Job title").fill("Writer");
  await page.getByLabel("Job description").fill("You draft crisp documents.");
  await page.getByRole("button", { name: "Continue →" }).click();

  // Step 2: access.
  await expect(page).toHaveURL(/step=access/);
  await page.getByRole("radio", { name: /Any Claw on the team/ }).click();
  await page.getByRole("button", { name: "Continue →" }).click();

  // Step 3: slack (optional) → explicit confirm before the agent goes live.
  await expect(page).toHaveURL(/step=slack/);
  await page.getByRole("button", { name: "Review & create" }).click();
  await expect(page.getByText("Create Drafter?")).toBeVisible();
  await page.getByRole("button", { name: "Create claw" }).click();

  await expect(page).toHaveURL(/\/claws\/.+\/chat\//);
  await expect(
    page.getByRole("heading", { name: /Hi, I'm Drafter/ }),
  ).toBeVisible();

  const box = page.getByLabel("Message Drafter");
  await box.fill("ping");
  await box.press("Enter");
  await expect(page.getByText("I received: ping")).toBeVisible();
});