test: add golden e2e tests for agent TUI launch and shell recovery

Add test fixtures and E2E tests to verify agent TUI functionality:
- Stub agent implementation supports cross-platform execution (Unix/Windows)
- Test verifies multiline composer with Shift+Enter support in agent TUI
- Test verifies clean shell resumes after agent exit without state leakage

tests/e2e/fixtures/golden-stub-agent/codex

@@ -0,0 +1,2 @@
+#!/bin/sh
+exec "$(dirname "$0")/golden-stub-agent" "$@"

tests/e2e/fixtures/golden-stub-agent/codex.cmd

@@ -0,0 +1,2 @@
+@echo off
+call "%~dp0golden-stub-agent.cmd" %*

tests/e2e/fixtures/golden-stub-agent/golden-stub-agent

@@ -0,0 +1,2 @@
+#!/bin/sh
+exec node "$(dirname "$0")/golden-stub-agent.js" "$@"

tests/e2e/fixtures/golden-stub-agent/golden-stub-agent.cmd

@@ -0,0 +1,2 @@
+@echo off
+node "%~dp0golden-stub-agent.js" %*

tests/e2e/fixtures/golden-stub-agent/golden-stub-agent.js

@@ -0,0 +1,98 @@
+#!/usr/bin/env node
+
+const READY_MARKER = 'GOLDEN_STUB_AGENT_READY'
+const EXIT_MARKER = 'GOLDEN_STUB_AGENT_EXITED'
+
+let composer = ''
+let lastSubmission = ''
+let pendingInput = ''
+let exiting = false
+
+function render() {
+  const lines = composer.split('\n')
+  const renderedComposer = lines.map((line, index) => `${index === 0 ? '> ' : '  '}${line}`)
+  process.stdout.write(
+    `\x1b]0;Golden Stub Agent\x07${[
+      '\x1b[H\x1b[2JGolden Stub Agent',
+      `[${READY_MARKER}]`,
+      '',
+      ...renderedComposer,
+      '',
+      'Shift+Enter inserts a newline. Type exit then Enter to quit.',
+      ...(lastSubmission ? [`[GOLDEN_STUB_AGENT_SUBMITTED] ${lastSubmission}`] : [])
+    ].join('\r\n')}`
+  )
+}
+
+function exitCleanly() {
+  if (exiting) {
+    return
+  }
+  exiting = true
+  if (process.stdin.isTTY) {
+    process.stdin.setRawMode(false)
+  }
+  process.stdout.write(`\x1b[?1049l[${EXIT_MARKER}]\r\n`, () => process.exit(0))
+}
+
+function submit() {
+  if (composer.trim() === 'exit') {
+    exitCleanly()
+    return
+  }
+  lastSubmission = composer
+  composer = ''
+  render()
+}
+
+function consumeInput() {
+  while (pendingInput.length > 0 && !exiting) {
+    const shiftEnter = ['\x1b[13;2u', '\x1b[13;2~', '\x1b\r'].find((sequence) =>
+      pendingInput.startsWith(sequence)
+    )
+    if (shiftEnter) {
+      pendingInput = pendingInput.slice(shiftEnter.length)
+      composer += '\n'
+      render()
+      continue
+    }
+    if (
+      pendingInput === '\x1b' ||
+      ('\x1b[13;2u'.startsWith(pendingInput) && pendingInput.startsWith('\x1b[')) ||
+      ('\x1b[13;2~'.startsWith(pendingInput) && pendingInput.startsWith('\x1b['))
+    ) {
+      return
+    }
+
+    const char = pendingInput[0]
+    pendingInput = pendingInput.slice(1)
+    if (char === '\r' || char === '\n') {
+      submit()
+    } else if (char === '\x04') {
+      exitCleanly()
+    } else if (char === '\x7f' || char === '\b') {
+      composer = composer.slice(0, -1)
+      render()
+    } else if (char >= ' ') {
+      composer += char
+      render()
+    }
+  }
+}
+
+if (process.stdin.isTTY) {
+  process.stdin.setRawMode(true)
+}
+process.stdin.setEncoding('utf8')
+process.stdin.on('data', (chunk) => {
+  pendingInput += chunk
+  consumeInput()
+})
+process.stdin.resume()
+
+for (const signal of ['SIGINT', 'SIGTERM', 'SIGHUP']) {
+  process.on(signal, exitCleanly)
+}
+
+process.stdout.write('\x1b[?1049h')
+render()

tests/e2e/golden-agent-tui-launch.spec.ts

@@ -0,0 +1,32 @@
+import { expect, test } from './helpers/orca-app'
+import {
+  configureGoldenStubAgent,
+  getGoldenStubAgentLaunchEnv,
+  launchGoldenStubAgentFromNewTab
+} from './helpers/golden-stub-agent'
+import { ensureTerminalVisible, waitForActiveWorktree, waitForSessionReady } from './helpers/store'
+import { focusActiveTerminalInput, getTerminalContent } from './helpers/terminal'
+
+test.use({ launchEnv: getGoldenStubAgentLaunchEnv() })
+
+test('@golden launches an agent TUI with a live multiline composer', async ({ orcaPage }) => {
+  await waitForSessionReady(orcaPage)
+  await waitForActiveWorktree(orcaPage)
+  await ensureTerminalVisible(orcaPage)
+  await configureGoldenStubAgent(orcaPage)
+  await launchGoldenStubAgentFromNewTab(orcaPage)
+
+  const activeTab = orcaPage.locator('[data-testid="sortable-tab"][data-active="true"]')
+  await expect(activeTab).toHaveAttribute('data-tab-title', /Codex|Golden Stub Agent/i)
+  await expect(activeTab).not.toHaveAttribute('data-tab-title', /^Terminal \d+$/)
+
+  await focusActiveTerminalInput(orcaPage)
+  await orcaPage.keyboard.type('hello from e2e')
+  await orcaPage.keyboard.press('Shift+Enter')
+  await orcaPage.keyboard.type('second line')
+
+  await expect
+    .poll(() => getTerminalContent(orcaPage), { timeout: 10_000 })
+    .toContain('> hello from e2e\r\n  second line')
+  expect(await getTerminalContent(orcaPage)).not.toContain('GOLDEN_STUB_AGENT_SUBMITTED')
+})

tests/e2e/golden-shell-after-agent-exit.spec.ts

@@ -0,0 +1,47 @@
+import { expect, test } from './helpers/orca-app'
+import {
+  configureGoldenStubAgent,
+  getGoldenStubAgentLaunchEnv,
+  GOLDEN_STUB_EXIT_MARKER,
+  launchGoldenStubAgentFromNewTab
+} from './helpers/golden-stub-agent'
+import { ensureTerminalVisible, waitForActiveWorktree, waitForSessionReady } from './helpers/store'
+import {
+  focusActiveTerminalInput,
+  waitForActivePanePtyId,
+  waitForTerminalOutput
+} from './helpers/terminal'
+
+test.use({ launchEnv: getGoldenStubAgentLaunchEnv() })
+
+test('@golden opens a clean live shell after an agent exits', async ({ orcaPage }) => {
+  await waitForSessionReady(orcaPage)
+  await waitForActiveWorktree(orcaPage)
+  await ensureTerminalVisible(orcaPage)
+  await configureGoldenStubAgent(orcaPage)
+  await launchGoldenStubAgentFromNewTab(orcaPage)
+
+  await orcaPage.keyboard.type('exit')
+  await orcaPage.keyboard.press('Enter')
+  await waitForTerminalOutput(orcaPage, GOLDEN_STUB_EXIT_MARKER, 15_000)
+
+  const tabsBeforeShell = await orcaPage.locator('[data-testid="sortable-tab"]').count()
+  await orcaPage.getByRole('button', { name: 'New tab' }).click({ force: true })
+  await orcaPage
+    .getByRole('menuitem', { name: /New Terminal/i })
+    .first()
+    .click({ force: true })
+  await expect(orcaPage.locator('[data-testid="sortable-tab"]')).toHaveCount(tabsBeforeShell + 1)
+  await waitForActivePanePtyId(orcaPage)
+
+  await focusActiveTerminalInput(orcaPage)
+  await orcaPage.keyboard.type('echo after-agent')
+  await orcaPage.keyboard.press('Enter')
+  await waitForTerminalOutput(orcaPage, 'after-agent', 15_000)
+
+  await orcaPage.keyboard.press('Shift+Enter')
+  await orcaPage.keyboard.type('echo after-shift-enter')
+  await orcaPage.keyboard.press('Enter')
+  await waitForTerminalOutput(orcaPage, 'after-shift-enter', 15_000)
+  await expect(orcaPage.locator('[data-testid="sortable-tab"]')).toHaveCount(tabsBeforeShell + 1)
+})

tests/e2e/helpers/golden-stub-agent.ts

@@ -0,0 +1,39 @@
+import path from 'node:path'
+import type { Page } from '@stablyai/playwright-test'
+import { expect } from '@stablyai/playwright-test'
+import { focusActiveTerminalInput, waitForTerminalOutput } from './terminal'
+
+export const GOLDEN_STUB_READY_MARKER = 'GOLDEN_STUB_AGENT_READY'
+export const GOLDEN_STUB_EXIT_MARKER = 'GOLDEN_STUB_AGENT_EXITED'
+
+const fixtureDir = path.join(process.cwd(), 'tests', 'e2e', 'fixtures', 'golden-stub-agent')
+
+export function getGoldenStubAgentLaunchEnv(): NodeJS.ProcessEnv {
+  const pathKey = Object.keys(process.env).find((key) => key.toLowerCase() === 'path') ?? 'PATH'
+  return {
+    [pathKey]: [fixtureDir, process.env[pathKey] ?? ''].filter(Boolean).join(path.delimiter)
+  }
+}
+
+export async function configureGoldenStubAgent(page: Page): Promise<void> {
+  await page.evaluate(async () => {
+    const store = window.__store
+    if (!store) {
+      throw new Error('Orca store is unavailable')
+    }
+    await store.getState().updateSettings({
+      defaultTuiAgent: 'codex',
+      agentCmdOverrides: { codex: 'golden-stub-agent' },
+      agentDefaultArgs: { codex: '' }
+    })
+  })
+}
+
+export async function launchGoldenStubAgentFromNewTab(page: Page): Promise<void> {
+  await page.getByRole('button', { name: 'New tab' }).click({ force: true })
+  const launchOption = page.getByRole('menuitem', { name: /^Codex(?:\s|$)/i }).first()
+  await expect(launchOption).toBeVisible({ timeout: 15_000 })
+  await launchOption.click({ force: true })
+  await focusActiveTerminalInput(page)
+  await waitForTerminalOutput(page, GOLDEN_STUB_READY_MARKER, 20_000)
+}