fix: isolate e2e daemon and onboarding state (#1606)

src/main/daemon/daemon-health.test.ts

@@ -120,22 +120,48 @@ describe('parseDaemonPidFile', () => {
     const serialized = serializeDaemonPidFile({ pid: 12345, startedAtMs: 1_700_000_000_000 })
     expect(parseDaemonPidFile(serialized)).toEqual({
       pid: 12345,
-      startedAtMs: 1_700_000_000_000
+      startedAtMs: 1_700_000_000_000,
+      entryPath: null
+    })
+  })
+
+  it('parses JSON pid files with entryPath', () => {
+    const serialized = serializeDaemonPidFile({
+      pid: 12345,
+      startedAtMs: 1_700_000_000_000,
+      entryPath: '/repo/out/main/daemon-entry.js'
+    })
+    expect(parseDaemonPidFile(serialized)).toEqual({
+      pid: 12345,
+      startedAtMs: 1_700_000_000_000,
+      entryPath: '/repo/out/main/daemon-entry.js'
     })
   })
 
   it('accepts JSON with startedAtMs missing and returns null for it', () => {
     // Why: forward-compatible with hypothetical future daemons that might write
     // pid without startedAtMs (platform where getProcessStartedAtMs returns null).
-    expect(parseDaemonPidFile('{"pid":9999}')).toEqual({ pid: 9999, startedAtMs: null })
+    expect(parseDaemonPidFile('{"pid":9999}')).toEqual({
+      pid: 9999,
+      startedAtMs: null,
+      entryPath: null
+    })
   })
 
   it('falls back to bare-integer parsing for legacy pid files', () => {
     // Why: pre-Phase-0 daemons wrote the pid file as a bare integer.
     // parseDaemonPidFile must still accept those to avoid leaking a stale
     // daemon across a single upgrade boundary.
-    expect(parseDaemonPidFile('12345')).toEqual({ pid: 12345, startedAtMs: null })
-    expect(parseDaemonPidFile('  12345\n')).toEqual({ pid: 12345, startedAtMs: null })
+    expect(parseDaemonPidFile('12345')).toEqual({
+      pid: 12345,
+      startedAtMs: null,
+      entryPath: null
+    })
+    expect(parseDaemonPidFile('  12345\n')).toEqual({
+      pid: 12345,
+      startedAtMs: null,
+      entryPath: null
+    })
   })
 
   it('returns null for malformed input', () => {

src/main/daemon/daemon-health.ts

@@ -15,6 +15,7 @@ const START_TIME_TOLERANCE_MS = 1_500
 type ParsedDaemonPid = {
   pid: number
   startedAtMs: number | null
+  entryPath: string | null
 }
 
 function canConnectSocket(socketPath: string): Promise<boolean> {
@@ -139,14 +140,19 @@ function commandLineMatchesDaemon(
 export function parseDaemonPidFile(contents: string): ParsedDaemonPid | null {
   const trimmed = contents.trim()
   try {
-    const parsed = JSON.parse(trimmed) as { pid?: unknown; startedAtMs?: unknown }
+    const parsed = JSON.parse(trimmed) as {
+      pid?: unknown
+      startedAtMs?: unknown
+      entryPath?: unknown
+    }
     if (typeof parsed.pid === 'number' && Number.isFinite(parsed.pid)) {
       return {
         pid: parsed.pid,
         startedAtMs:
           typeof parsed.startedAtMs === 'number' && Number.isFinite(parsed.startedAtMs)
             ? parsed.startedAtMs
-            : null
+            : null,
+        entryPath: typeof parsed.entryPath === 'string' ? parsed.entryPath : null
       }
     }
   } catch {
@@ -154,7 +160,7 @@ export function parseDaemonPidFile(contents: string): ParsedDaemonPid | null {
   }
 
   const pid = Number(trimmed)
-  return Number.isFinite(pid) ? { pid, startedAtMs: null } : null
+  return Number.isFinite(pid) ? { pid, startedAtMs: null, entryPath: null } : null
 }
 
 function getLinuxProcessStartedAtMs(pid: number): number | null {
@@ -278,6 +284,78 @@ function isDaemonProcess(
   }
 }
 
+function getDaemonCommandLine(pid: number): string | null {
+  if (process.platform === 'win32') {
+    try {
+      return execFileSync(
+        'powershell.exe',
+        [
+          '-NoProfile',
+          '-NonInteractive',
+          '-Command',
+          `(Get-CimInstance Win32_Process -Filter "ProcessId = ${pid}").CommandLine`
+        ],
+        {
+          encoding: 'utf8',
+          timeout: 3_000
+        }
+      )
+    } catch {
+      return null
+    }
+  }
+
+  try {
+    return readFileSync(`/proc/${pid}/cmdline`, 'utf8')
+  } catch {
+    try {
+      return execFileSync('ps', ['-p', String(pid), '-o', 'command='], {
+        encoding: 'utf8',
+        timeout: 2_000
+      })
+    } catch {
+      return null
+    }
+  }
+}
+
+export type DaemonLaunchIdentity = 'match' | 'mismatch' | 'unknown'
+
+export function getDaemonLaunchIdentity(
+  runtimeDir: string,
+  socketPath: string,
+  tokenPath: string,
+  expectedEntryPath: string,
+  protocolVersion = PROTOCOL_VERSION
+): DaemonLaunchIdentity {
+  let parsedPid: ParsedDaemonPid | null
+  try {
+    parsedPid = parseDaemonPidFile(
+      readFileSync(getDaemonPidPath(runtimeDir, protocolVersion), 'utf8')
+    )
+  } catch {
+    return 'unknown'
+  }
+
+  if (!parsedPid || !isDaemonProcess(parsedPid.pid, socketPath, tokenPath, parsedPid.startedAtMs)) {
+    return 'unknown'
+  }
+
+  if (parsedPid.entryPath) {
+    return parsedPid.entryPath === expectedEntryPath ? 'match' : 'mismatch'
+  }
+
+  // Why: older pid files did not persist entryPath. The command line still
+  // carries daemon-entry.js, so use it to stop dev worktrees from reusing a
+  // daemon forked from a deleted sibling checkout. If command-line probing is
+  // unavailable, fail open so we don't kill live sessions unnecessarily.
+  const commandLine = getDaemonCommandLine(parsedPid.pid)
+  if (!commandLine) {
+    return 'unknown'
+  }
+  return commandLine.includes(expectedEntryPath) ? 'match' : 'mismatch'
+}
+
 export async function killStaleDaemon(
   runtimeDir: string,
   socketPath: string,

src/main/daemon/daemon-init.test.ts

@@ -14,9 +14,12 @@ import { PROTOCOL_VERSION } from './types'
 // non-daemon-init dependency is replaced by a minimal stub that records calls.
 const {
   getPathMock,
+  isPackagedMock,
   probeSocketExistsMock,
   netConnectMock,
+  forkMock,
   healthCheckDaemonMock,
+  getDaemonLaunchIdentityMock,
   killStaleDaemonMock,
   getProcessStartedAtMsMock,
   daemonClientMock,
@@ -27,8 +30,10 @@ const {
   rebindLocalProviderListenersMock
 } = vi.hoisted(() => {
   const getPathMock = vi.fn(() => '/fake/userData')
+  const isPackagedMock = vi.fn(() => false)
 
   const probeSocketExistsMock = vi.fn(() => false)
+  const forkMock = vi.fn()
   const netConnectMock = vi.fn(() => {
     // Why: the real probeSocket() in daemon-init connects to the socket and
     // resolves true on 'connect', false on 'error'. Our launcher never runs
@@ -51,6 +56,7 @@ const {
   })
 
   const healthCheckDaemonMock = vi.fn(async () => true)
+  const getDaemonLaunchIdentityMock = vi.fn(() => 'match')
   const killStaleDaemonMock = vi.fn(async () => true)
   const getProcessStartedAtMsMock = vi.fn(() => 1_000_000)
 
@@ -73,9 +79,12 @@ const {
 
   return {
     getPathMock,
+    isPackagedMock,
     probeSocketExistsMock,
     netConnectMock,
+    forkMock,
     healthCheckDaemonMock,
+    getDaemonLaunchIdentityMock,
     killStaleDaemonMock,
     getProcessStartedAtMsMock,
     daemonClientMock,
@@ -120,7 +129,13 @@ type MockAdapter = {
 }
 
 vi.mock('electron', () => ({
-  app: { getPath: getPathMock, getAppPath: () => '/fake/app' }
+  app: {
+    get isPackaged() {
+      return isPackagedMock()
+    },
+    getPath: getPathMock,
+    getAppPath: () => '/fake/app'
+  }
 }))
 
 vi.mock('fs', () => ({
@@ -130,11 +145,12 @@ vi.mock('fs', () => ({
   writeFileSync: vi.fn()
 }))
 
-vi.mock('child_process', () => ({ fork: vi.fn() }))
+vi.mock('child_process', () => ({ fork: forkMock }))
 
 vi.mock('net', () => ({ connect: netConnectMock }))
 
 vi.mock('./daemon-health', () => ({
+  getDaemonLaunchIdentity: getDaemonLaunchIdentityMock,
   healthCheckDaemon: healthCheckDaemonMock,
   killStaleDaemon: killStaleDaemonMock,
   getProcessStartedAtMs: getProcessStartedAtMsMock
@@ -224,7 +240,11 @@ async function importFresh() {
   unbindLocalProviderListenersMock.mockClear()
   rebindLocalProviderListenersMock.mockClear()
   healthCheckDaemonMock.mockClear()
+  getDaemonLaunchIdentityMock.mockClear()
   killStaleDaemonMock.mockClear()
+  forkMock.mockReset()
+  isPackagedMock.mockReset()
+  isPackagedMock.mockReturnValue(false)
   daemonClientMock.mockClear()
   probeSocketExistsMock.mockClear()
   // Why: importing daemon-init *after* resetModules means the module-level
@@ -600,4 +620,74 @@ describe('daemon-init: runRestartDaemon (7-step sequence)', () => {
       'restartDaemon called before initDaemonPtyProvider'
     )
   })
+
+  it('respawns instead of reusing a healthy daemon launched from another app path', async () => {
+    const mod = await importFresh()
+    await mod.initDaemonPtyProvider()
+
+    const launcher = spawnerInstances[0].launcher as (
+      socketPath: string,
+      tokenPath: string
+    ) => Promise<{ shutdown(): Promise<void> }>
+    getDaemonLaunchIdentityMock.mockReturnValueOnce('mismatch')
+    forkMock.mockImplementationOnce(() => {
+      const handlers: Record<string, ((arg?: unknown) => void)[]> = {
+        message: [],
+        error: [],
+        exit: []
+      }
+      return {
+        pid: 12345,
+        on(event: string, cb: (arg?: unknown) => void) {
+          handlers[event]?.push(cb)
+          if (event === 'message') {
+            queueMicrotask(() => cb({ type: 'ready' }))
+          }
+          return this
+        },
+        disconnect: vi.fn(),
+        unref: vi.fn()
+      }
+    })
+
+    await launcher('/fake/socket', '/fake/token')
+
+    expect(getDaemonLaunchIdentityMock).toHaveBeenCalledWith(
+      '/fake/userData/daemon',
+      '/fake/socket',
+      '/fake/token',
+      '/fake/app/out/main/daemon-entry.js'
+    )
+    expect(killStaleDaemonMock).toHaveBeenCalledWith(
+      '/fake/userData/daemon',
+      '/fake/socket',
+      '/fake/token'
+    )
+    expect(forkMock).toHaveBeenCalledWith(
+      '/fake/app/out/main/daemon-entry.js',
+      ['--socket', '/fake/socket', '--token', '/fake/token'],
+      expect.objectContaining({ detached: true })
+    )
+  })
+
+  it('keeps packaged healthy-daemon reuse independent of dev app-path identity', async () => {
+    const mod = await importFresh()
+    await mod.initDaemonPtyProvider()
+
+    const launcher = spawnerInstances[0].launcher as (
+      socketPath: string,
+      tokenPath: string
+    ) => Promise<{ shutdown(): Promise<void> }>
+    getDaemonLaunchIdentityMock.mockClear()
+    killStaleDaemonMock.mockClear()
+    forkMock.mockClear()
+    isPackagedMock.mockReturnValue(true)
+    getDaemonLaunchIdentityMock.mockReturnValueOnce('mismatch')
+
+    await launcher('/fake/socket', '/fake/token')
+
+    expect(getDaemonLaunchIdentityMock).not.toHaveBeenCalled()
+    expect(killStaleDaemonMock).not.toHaveBeenCalled()
+    expect(forkMock).not.toHaveBeenCalled()
+  })
 })

src/main/daemon/daemon-init.ts

@@ -27,7 +27,12 @@ import {
   PROTOCOL_VERSION,
   type ListSessionsResult
 } from './types'
-import { getProcessStartedAtMs, healthCheckDaemon, killStaleDaemon } from './daemon-health'
+import {
+  getDaemonLaunchIdentity,
+  getProcessStartedAtMs,
+  healthCheckDaemon,
+  killStaleDaemon
+} from './daemon-health'
 import {
   setLocalPtyProvider,
   unbindLocalProviderListeners,
@@ -91,13 +96,26 @@ function probeSocket(socketPath: string): Promise<boolean> {
 
 function createOutOfProcessLauncher(runtimeDir: string): DaemonLauncher {
   return async (socketPath, tokenPath) => {
+    const entryPath = getDaemonEntryPath()
     const healthy = await healthCheckDaemon(socketPath, tokenPath)
     if (healthy) {
-      // Why: daemon is already running from a previous app session and
-      // responded to a protocol-level ping. Safe to reuse.
-      return {
-        shutdown: async () => {
-          await cleanupDaemonForProtocol(runtimeDir, PROTOCOL_VERSION)
+      // Why: dev worktrees share the same orca-dev userData, so a daemon from
+      // a deleted sibling checkout can pass protocol health checks while still
+      // pointing at missing native modules. Packaged app paths are stable and
+      // should preserve existing warm daemon reuse semantics.
+      const identity = app.isPackaged
+        ? 'match'
+        : getDaemonLaunchIdentity(runtimeDir, socketPath, tokenPath, entryPath)
+      if (identity === 'mismatch') {
+        console.warn('[daemon] Replacing daemon launched from a different app path')
+        await cleanupDaemonForProtocol(runtimeDir, PROTOCOL_VERSION)
+      } else {
+        // Why: daemon is already running from a previous app session and
+        // responded to a protocol-level ping. Safe to reuse.
+        return {
+          shutdown: async () => {
+            await cleanupDaemonForProtocol(runtimeDir, PROTOCOL_VERSION)
+          }
         }
       }
     }
@@ -106,7 +124,6 @@ function createOutOfProcessLauncher(runtimeDir: string): DaemonLauncher {
     // before respawn so the new daemon does not race the stale process.
     await killStaleDaemon(runtimeDir, socketPath, tokenPath)
 
-    const entryPath = getDaemonEntryPath()
     const userDataPath = app.getPath('userData')
     const child = fork(entryPath, ['--socket', socketPath, '--token', tokenPath], {
       // Why: detached + unref lets the daemon outlive the Electron process.
@@ -156,7 +173,8 @@ function createOutOfProcessLauncher(runtimeDir: string): DaemonLauncher {
               getDaemonPidPath(runtimeDir),
               serializeDaemonPidFile({
                 pid: child.pid,
-                startedAtMs: getProcessStartedAtMs(child.pid)
+                startedAtMs: getProcessStartedAtMs(child.pid),
+                entryPath
               }),
               { mode: 0o600 }
             )

src/main/daemon/daemon-spawner.ts

@@ -10,6 +10,7 @@ export type DaemonConnectionInfo = {
 export type DaemonPidFile = {
   pid: number
   startedAtMs: number | null
+  entryPath?: string
 }
 
 export type DaemonProcessHandle = {

tests/e2e/helpers/orca-app.ts

@@ -30,6 +30,11 @@ type OrcaTestFixtures = {
   electronApp: ElectronApplication
   sharedPage: Page
   orcaPage: Page
+  // Why: every fresh userData dir paints the first-launch onboarding overlay
+  // (closedAt=null), which is `fixed inset-0 z-[100]` and intercepts pointer
+  // events for every other test. Dismiss it by default; onboarding.spec.ts
+  // opts out via `test.use({ dismissOnboarding: false })`.
+  dismissOnboarding: boolean
 }
 
 type OrcaWorkerFixtures = {
@@ -140,10 +145,38 @@ export const test = base.extend<OrcaTestFixtures, OrcaWorkerFixtures>({
   ],
 
   // Test-scoped: one Electron app per test
-  // oxlint-disable-next-line no-empty-pattern -- Playwright fixture callbacks require object destructuring here.
-  electronApp: async ({}, provideFixture, testInfo) => {
+  electronApp: async ({ dismissOnboarding }, provideFixture, testInfo) => {
     const mainPath = path.join(process.cwd(), 'out', 'main', 'index.js')
     const userDataDir = mkdtempSync(path.join(os.tmpdir(), 'orca-e2e-userdata-'))
+
+    if (dismissOnboarding) {
+      // Why: onboarding renders a fullscreen `fixed inset-0 z-[100]` overlay
+      // when persisted `closedAt` is null, which intercepts pointer events for
+      // every other test. Seed explicit fresh-user state: an empty file would
+      // make persistence treat the profile as an existing-user upgrade cohort
+      // and mount the telemetry notice overlay instead.
+      writeFileSync(
+        path.join(userDataDir, 'orca-data.json'),
+        `${JSON.stringify(
+          {
+            settings: {
+              telemetry: {
+                optedIn: true,
+                installId: '00000000-0000-4000-8000-000000000000',
+                existedBeforeTelemetryRelease: false
+              }
+            },
+            onboarding: {
+              closedAt: 1,
+              outcome: 'completed',
+              lastCompletedStep: 4
+            }
+          },
+          null,
+          2
+        )}\n`
+      )
+    }
     const headful = shouldLaunchHeadful(testInfo)
     // Why: strip ELECTRON_RUN_AS_NODE before spawning. Some host shells (e.g.
     // Orca's own agent runtime) set it so Electron behaves as a plain Node
@@ -161,9 +194,7 @@ export const test = base.extend<OrcaTestFixtures, OrcaWorkerFixtures>({
     // Why: testInfo.outputDir is created lazily by Playwright; on Windows the
     // dir may not exist when the fixture initializes, and Electron silently
     // drops the recording. mkdir up-front so the recorder always has a home.
-    const recordVideoDir = process.env.ORCA_E2E_RECORD_VIDEO === '1'
-      ? testInfo.outputDir
-      : null
+    const recordVideoDir = process.env.ORCA_E2E_RECORD_VIDEO === '1' ? testInfo.outputDir : null
     if (recordVideoDir) {
       mkdirSync(recordVideoDir, { recursive: true })
     }
@@ -212,6 +243,9 @@ export const test = base.extend<OrcaTestFixtures, OrcaWorkerFixtures>({
     rmSync(userDataDir, { recursive: true, force: true })
   },
 
+  // Default: dismiss the onboarding overlay so it doesn't intercept clicks.
+  dismissOnboarding: [true, { option: true }],
+
   // Test-scoped: grab the first BrowserWindow, add the test repo, and wait
   // until the session is fully ready with a worktree active.
   sharedPage: async ({ electronApp, testRepoPath }, provideFixture) => {

tests/e2e/onboarding.spec.ts

@@ -34,6 +34,11 @@ async function getDocumentThemeClass(page: Page): Promise<'dark' | 'light'> {
 }
 
 test.describe('Onboarding flow', () => {
+  // Why: the shared fixture pre-seeds onboarding as closed so non-onboarding
+  // tests don't get blocked by the fullscreen overlay. Opt out here so this
+  // spec actually exercises the first-launch flow.
+  test.use({ dismissOnboarding: false })
+
   test.beforeEach(async ({ orcaPage }) => {
     // Per-test userData is freshly minted by the orcaPage fixture, so persisted
     // onboarding state defaults to `closedAt: null, lastCompletedStep: -1` and
@@ -42,9 +47,9 @@ test.describe('Onboarding flow', () => {
   })
 
   test('renders on first launch with the agent step active', async ({ orcaPage }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
     await expect(orcaPage.getByText('1 of 4')).toBeVisible()
     await expect(orcaPage.getByRole('button', { name: 'Continue' })).toBeVisible()
     await expect(orcaPage.getByRole('button', { name: 'Skip' })).toBeVisible()
@@ -61,9 +66,9 @@ test.describe('Onboarding flow', () => {
   test('Continue advances steps, persists progress, and applies user-visible settings', async ({
     orcaPage
   }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
 
     // --- Step 1: agent ---
     // Force a deterministic, non-default selection so the assertion below
@@ -87,9 +92,7 @@ test.describe('Onboarding flow', () => {
     await codexButton.click()
 
     await orcaPage.getByRole('button', { name: 'Continue' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Make it feel like home/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Make it feel like home/i })).toBeVisible()
     await expect(orcaPage.getByText('2 of 4')).toBeVisible()
     await expect
       .poll(async () => (await getOnboardingState(orcaPage)).lastCompletedStep, {
@@ -134,9 +137,9 @@ test.describe('Onboarding flow', () => {
         message: 'lastCompletedStep did not advance to 2 after second Continue'
       })
       .toBe(2)
-    await expect.poll(async () => (await getSettings(orcaPage)).theme, { timeout: 5_000 }).toBe(
-      oppositeTheme
-    )
+    await expect
+      .poll(async () => (await getSettings(orcaPage)).theme, { timeout: 5_000 })
+      .toBe(oppositeTheme)
 
     // --- Step 3: notifications ---
     // Why: the wizard force-defaults every toggle ON (use-onboarding-flow.ts),
@@ -145,9 +148,7 @@ test.describe('Onboarding flow', () => {
     // the post-Continue assertion proves the wizard wrote its opt-in defaults
     // through the IPC boundary, including the inverted suppressWhenFocused.
     await orcaPage.getByRole('button', { name: 'Continue' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Point Orca at some code/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Point Orca at some code/i })).toBeVisible()
     await expect(orcaPage.getByText('4 of 4')).toBeVisible()
     await expect(orcaPage.getByRole('button', { name: 'Continue' })).toHaveCount(0)
     await expect(orcaPage.getByRole('button', { name: /I'll add one later/ })).toBeVisible()
@@ -182,9 +183,9 @@ test.describe('Onboarding flow', () => {
   })
 
   test('Cmd/Ctrl+Enter advances steps like Continue', async ({ orcaPage }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
 
     // Why: the OS the renderer reports drives whether Cmd or Ctrl is the
     // accelerator (OnboardingFlow.tsx checks navigator.userAgent).
@@ -196,9 +197,7 @@ test.describe('Onboarding flow', () => {
     // inert area inside the overlay first to anchor focus, then press.
     await orcaPage.locator('footer').click({ position: { x: 1, y: 1 } })
     await orcaPage.keyboard.press(accelerator)
-    await expect(
-      orcaPage.getByRole('heading', { name: /Make it feel like home/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Make it feel like home/i })).toBeVisible()
     await expect
       .poll(async () => (await getOnboardingState(orcaPage)).lastCompletedStep, {
         timeout: 5_000
@@ -207,9 +206,9 @@ test.describe('Onboarding flow', () => {
   })
 
   test('selected agent button reports aria-pressed=true', async ({ orcaPage }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
 
     const codexButton = orcaPage.getByRole('button', { name: /^Codex\s/ })
     const codexVisible = await codexButton
@@ -227,13 +226,11 @@ test.describe('Onboarding flow', () => {
   })
 
   test('notification toggles flip independently and persist on Continue', async ({ orcaPage }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Make it feel like home/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Make it feel like home/i })).toBeVisible()
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
     await expect(
       orcaPage.getByRole('heading', { name: /Know when an agent needs you/i })
@@ -248,9 +245,7 @@ test.describe('Onboarding flow', () => {
     await expect(bellSwitch).toHaveAttribute('aria-checked', 'false')
 
     await orcaPage.getByRole('button', { name: 'Continue' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Point Orca at some code/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Point Orca at some code/i })).toBeVisible()
     await expect
       .poll(
         async () => {
@@ -268,16 +263,14 @@ test.describe('Onboarding flow', () => {
   test('typing in the clone-url input does not hijack Enter as a global shortcut', async ({
     orcaPage
   }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
     // Skip to the repo step.
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Point Orca at some code/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Point Orca at some code/i })).toBeVisible()
 
     // Why: focus the clone-url input and press Cmd/Ctrl+Enter. The capture-
     // phase keydown handler should bail via isEditableTarget, so the folder
@@ -291,22 +284,18 @@ test.describe('Onboarding flow', () => {
     await input.press(accelerator)
     // Brief wait so any (incorrect) handler firing would have already happened.
     await orcaPage.waitForTimeout(250)
-    await expect(
-      orcaPage.getByRole('heading', { name: /Point Orca at some code/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Point Orca at some code/i })).toBeVisible()
     // Onboarding must still be open (closedAt remains null).
     expect((await getOnboardingState(orcaPage)).closedAt).toBeNull()
   })
 
   test('Back returns to the previous step without losing progress', async ({ orcaPage }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
 
     await orcaPage.getByRole('button', { name: 'Continue' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Make it feel like home/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Make it feel like home/i })).toBeVisible()
     await expect
       .poll(async () => (await getOnboardingState(orcaPage)).lastCompletedStep, {
         timeout: 5_000
@@ -316,9 +305,7 @@ test.describe('Onboarding flow', () => {
     // Why: exact match — the app sidebar also exposes a "Go back" button that
     // would otherwise match this regex.
     await orcaPage.getByRole('button', { name: 'Back', exact: true }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible()
     await expect(orcaPage.getByText('1 of 4')).toBeVisible()
 
     // Why: "without losing progress" means persisted lastCompletedStep stays
@@ -332,32 +319,29 @@ test.describe('Onboarding flow', () => {
   })
 
   test('"I\'ll add one later" on the repo step dismisses onboarding', async ({ orcaPage }) => {
-    await expect(
-      orcaPage.getByRole('heading', { name: /Pick your default agent/i })
-    ).toBeVisible({ timeout: 15_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Pick your default agent/i })).toBeVisible({
+      timeout: 15_000
+    })
 
     // Skip through the first three steps. On steps 1–3 the affordance is
     // labelled "Skip"; on the repo step it is "I'll add one later".
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Make it feel like home/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Make it feel like home/i })).toBeVisible()
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
     await expect(
       orcaPage.getByRole('heading', { name: /Know when an agent needs you/i })
     ).toBeVisible()
     await orcaPage.getByRole('button', { name: 'Skip' }).click()
-    await expect(
-      orcaPage.getByRole('heading', { name: /Point Orca at some code/i })
-    ).toBeVisible()
+    await expect(orcaPage.getByRole('heading', { name: /Point Orca at some code/i })).toBeVisible()
 
     await orcaPage.getByRole('button', { name: /I'll add one later/ }).click()
 
     // The overlay is unmounted once `closedAt` is set, so the heading must
     // disappear from the DOM, not merely become invisible.
-    await expect(
-      orcaPage.getByRole('heading', { name: /Point Orca at some code/i })
-    ).toHaveCount(0, { timeout: 10_000 })
+    await expect(orcaPage.getByRole('heading', { name: /Point Orca at some code/i })).toHaveCount(
+      0,
+      { timeout: 10_000 }
+    )
 
     // Why: DOM unmount fires when closedAt flips in the renderer, but the
     // main-process write can lag by an IPC tick. Poll until the persisted