import { expect, test, type Page } from "@playwright/test"; // P2 BLOCKING exit criterion (spec §15/§17): a sensitive call is // intercepted → previewed → queued → blocks → executes ONLY on approval → // is audited. Runs against the real backend with the scripted provider. const OWNER_EMAIL = "owner@acme.test"; const OWNER_PASSWORD = "e2e-password"; const GATED_PROMPT = "email the CEO [[scenario:gated-email]]"; async function signIn(page: Page) { await page.goto("/login"); await page.getByLabel("Email").fill(OWNER_EMAIL); await page.getByRole("button", { name: /Continue with work email/ }).click(); await page.getByLabel("Password").fill(OWNER_PASSWORD); await page.getByRole("button", { name: "Sign in" }).click(); await expect(page.getByRole("heading", { name: "Clawmates" })).toBeVisible(); } async function openFreshScoutSession(page: Page) { await page.getByRole("link", { name: /Scout/ }).click(); await expect(page).toHaveURL(/\/claws\/.+\/chat\//); // Wait for the navigation to the NEW session: the previous (possibly // empty) session shows the same welcome heading, so the URL is the only // reliable signal that the fresh session is mounted. const before = page.url(); await page.getByRole("button", { name: "New" }).click(); await page.waitForURL((url) => url.toString() !== before); await expect( page.getByRole("heading", { name: /Hi, I'm Scout/ }), ).toBeVisible(); } async function triggerGatedEmail(page: Page) { const box = page.getByLabel("Message Scout"); await box.fill(GATED_PROMPT); await box.press("Enter"); // Intercepted + previewed: the approval card shows the EXACT payload. const card = page.getByRole("region", { name: "Review and approve" }); await expect(card).toBeVisible(); await expect(card.getByText(/wants to:/)).toContainText( "Send email to ceo@example.com", ); await expect(card).toContainText("Revenue is up 14% quarter over quarter."); return card; } test("gated email blocks, previews, and executes only after approval", async ({ page, }) => { await signIn(page); await openFreshScoutSession(page); const card = await triggerGatedEmail(page); await card.getByRole("button", { name: "Approve" }).click(); // The continuation streams the approved execution. await expect(page.getByText(/The email step is finished/)).toBeVisible(); const trace = page.getByRole("button", { name: /1 step/ }); await trace.click(); await expect(page.getByText(/✓ email\.send/)).toBeVisible(); await expect(page.getByText(/"queued": ?true/)).toBeVisible(); // The transcript and trace replay identically after reload (journal). await page.reload(); await expect(page.getByText(/The email step is finished/)).toBeVisible(); await page.getByRole("button", { name: /1 step/ }).click(); await expect(page.getByText(/✓ email\.send/)).toBeVisible(); }); test("rejection executes nothing and the agent continues in-band", async ({ page, }) => { await signIn(page); await openFreshScoutSession(page); const card = await triggerGatedEmail(page); await card.getByRole("button", { name: "Reject" }).click(); // The run completes; the step trace records the refusal, not an execution. await expect(page.getByText(/The email step is finished/)).toBeVisible(); await page.getByRole("button", { name: /1 step/ }).click(); await expect(page.getByText(/✗ email\.send/)).toBeVisible(); await expect(page.getByText(/rejected/)).toBeVisible(); }); test("the approvals queue lists pending actions and decides them", async ({ page, }) => { await signIn(page); await openFreshScoutSession(page); await triggerGatedEmail(page); // Queued: the global approvals page shows the same exact preview. // Earlier sessions may have left pending approvals; decide them all. await page.getByRole("link", { name: "Approvals" }).click(); await expect(page.getByText(/awaits? your review/)).toBeVisible(); const cards = page.getByRole("region", { name: "Review and approve" }); await expect(cards.first()).toContainText("Send email to ceo@example.com"); const total = await cards.count(); for (let remaining = total; remaining > 0; remaining--) { await cards.first().getByRole("button", { name: "Approve" }).click(); // Wait for the refresh to apply before touching the next card. await expect(cards).toHaveCount(remaining - 1); } await expect(page.getByText(/All clear/)).toBeVisible(); });