test(desktop): live e2e for a teammate's @hermes handoff driving the primary bot

Real-Electron Playwright spec for #100406: a two-member room (primary profile + code-farmer). The user addresses only @code-farmer; its scripted reply @mentions hermes; the assertion is a `default`-authored "B" entry in the persisted room log. On origin/main the room settles after Code Farmer's line and the spec fails at that assertion; with the mention-alias fix it passes. The mock inference server gains a per-speaker script for group rooms: `E2E_SAY(<handle>)[<line>]` tokens in the user's send answer the member whose turn prompt opens with `You are @<handle>`; unscripted members reply "(pass)". `{at}` stands for `@` so the script itself never mentions anyone and round one only drives the member the user tagged.
2026-09-14 14:45:34 -07:00
parent d92ce52e71
commit 52972d15f3
2 changed files with 152 additions and 0 deletions
@@ -0,0 +1,109 @@
+import { type MockBackendFixture, setupMockBackend, waitForAppReady } from './fixtures'
+import { expect, test } from './test'
+
+// #100406: in a Bot Mode group room a teammate's `@hermes` handoff must give
+// the primary profile (internal name `default`) its turn, exactly like
+// `@hermes → @code-farmer` already does. The live roster stamps the primary
+// row's handle as the bare profile id ("default"), and the mention parser let
+// that stamped handle shadow the `@hermes` alias — so the room settled with
+// Hermes never driven. The mock inference server scripts each member's line
+// from the user's send (`E2E_SAY(<handle>)[…]`), so the assertion is on the
+// persisted room log: an entry authored by `default` saying "B".
+
+let fixture: MockBackendFixture | null = null
+
+type Page = MockBackendFixture['page']
+
+interface RoomLogEntry {
+  from?: { kind?: string; name?: string }
+  text?: string
+}
+
+async function openBots(page: Page): Promise<void> {
+  const tab = page.getByRole('button', { name: 'Bots', exact: true }).or(page.getByRole('tab', { name: 'Bots', exact: true })).first()
+  await tab.click()
+  await expect(page.getByRole('button', { name: 'New bot or group chat' })).toBeVisible()
+}
+
+async function createAgent(page: Page, name: string, title: string): Promise<void> {
+  await page.getByRole('button', { name: 'New bot or group chat' }).click()
+  await page.getByRole('menuitem', { name: 'New Bot' }).click()
+
+  const dialog = page.getByRole('dialog', { name: 'New Bot' })
+  await dialog.getByPlaceholder('inbox-triage').fill(name)
+  await dialog.getByPlaceholder('Inbox Triage').fill(title)
+  await dialog.getByRole('button', { name: 'Create Bot' }).click()
+  await expect(dialog).toBeHidden({ timeout: 30_000 })
+  await expect(page.getByRole('button', { name: new RegExp(`^${title}\\b`) }).first()).toBeVisible({ timeout: 30_000 })
+}
+
+/** The plugin's persisted room log (`hermes.plugin.hermes-bots.group-chats`). */
+async function roomLog(page: Page, group: string): Promise<RoomLogEntry[]> {
+  return page.evaluate(name => {
+    const raw = window.localStorage.getItem('hermes.plugin.hermes-bots.group-chats')
+    const rooms = raw ? (JSON.parse(raw) as Record<string, { log?: RoomLogEntry[] }>) : {}
+
+    return rooms[name]?.log ?? []
+  }, group)
+}
+
+test.beforeAll(async () => {
+  fixture = await setupMockBackend()
+  await waitForAppReady(fixture, 120_000)
+})
+
+test.afterAll(async () => {
+  await fixture?.cleanup()
+  fixture = null
+})
+
+test('a teammate handing off with @hermes drives the primary profile', async () => {
+  test.setTimeout(420_000)
+  const page = fixture!.page
+  const group = 'Hermes, Code Farmer'
+
+  await openBots(page)
+  await createAgent(page, 'code-farmer', 'Code Farmer')
+
+  await page.getByRole('button', { name: 'New bot or group chat' }).click()
+  await page.getByRole('menuitem', { name: 'New Group Chat' }).click()
+
+  const dialog = page.getByRole('dialog', { name: 'New Group Chat' })
+
+  for (const title of ['Hermes', 'Code Farmer']) {
+    await dialog.getByText(title, { exact: true }).locator('xpath=ancestor::label').getByRole('checkbox').click()
+  }
+
+  await dialog.getByRole('textbox', { name: 'Group name' }).fill(group)
+  await dialog.getByRole('button', { name: 'Create Group (2)' }).click()
+
+  const composer = page.getByRole('textbox', { name: `Message ${group}` }).filter({ visible: true })
+  await expect(composer).toBeVisible({ timeout: 20_000 })
+
+  // Only Code Farmer is addressed by the user. Its scripted reply hands off
+  // to @hermes; Hermes' scripted reply is "B". Neither script token carries
+  // a literal `@`, so the user send itself never mentions Hermes.
+  await composer.fill(
+    '@code-farmer Please reply with one line only. ' +
+      'E2E_SAY(code-farmer)[{at}hermes Please reply with the letter B.] E2E_SAY(hermes)[B]'
+  )
+  await composer.press('Enter')
+
+  // Code Farmer's handoff line lands first (the reverse direction is not in
+  // question); then the room must NOT settle without Hermes' turn.
+  await expect
+    .poll(async () => (await roomLog(page, group)).some(e => e.from?.name === 'code-farmer' && /@hermes/.test(e.text || '')), {
+      timeout: 180_000
+    })
+    .toBe(true)
+
+  await expect
+    .poll(async () => (await roomLog(page, group)).some(e => e.from?.name === 'default' && (e.text || '').trim() === 'B'), {
+      timeout: 180_000,
+      message: 'the primary profile (default / @hermes) never took its turn after being @mentioned by a teammate'
+    })
+    .toBe(true)
+
+  // And the transcript shows the handoff answered.
+  await expect(page.getByText('B', { exact: true }).filter({ visible: true }).first()).toBeVisible()
+})
@@ -382,6 +382,34 @@ function includesBatchClarifyTrigger(value: unknown): boolean {
  return false
 }

+/**
+ * Per-speaker scripted line for Bot Mode group rooms. A room turn prompt opens
+ * with `You are @<handle>` and quotes the user's message verbatim, so one user
+ * send can script every member's reply:
+ * `E2E_SAY(code-farmer)[{at}hermes Reply with B.] E2E_SAY(hermes)[B]`.
+ * The script deliberately carries no literal `@` (the room's mention parser
+ * would otherwise pull every scripted speaker into round one); `{at}` becomes
+ * `@` in the reply. The mock answers with the bracketed text whose handle
+ * matches the prompt's `You are @…` line; other prompts fall through.
+ */
+export function groupScriptedLine(userText: string, history: string[] = []): string | null {
+  const viewer = /You are @([a-z0-9][a-z0-9._-]*)/i.exec(userText)?.[1]?.toLowerCase()
+
+  if (!viewer || ![userText, ...history].some(text => text.includes('E2E_SAY('))) {
+    return null
+  }
+
+  for (const match of userText.matchAll(/E2E_SAY\(([a-z0-9][a-z0-9._-]*)\)\[([^\]]*)\]/gi)) {
+    if (match[1].toLowerCase() === viewer) {
+      return match[2].replace(/\{at\}/g, '@')
+    }
+  }
+
+  // A scripted room turn with nothing for this speaker stays silent, so the
+  // room settles instead of every member echoing the canned reply.
+  return '(pass)'
+}
+
 function includesBlockingClarifyTrigger(value: unknown): boolean {
  if (typeof value === 'string') {
    return value.includes(BLOCKING_CLARIFY_TRIGGER)
@@ -649,6 +677,21 @@ export function startMockServer(options: MockServerOptions = {}): Promise<MockSe
            return
          }

+          const groupLine = groupScriptedLine(
+            userText,
+            messages.flatMap(m => (m?.role === 'user' && typeof m?.content === 'string' ? [m.content] : [])),
+          )
+
+          if (groupLine !== null) {
+            if (stream) {
+              streamTextResponse(res, model, groupLine)
+            } else {
+              nonStreamingTextResponse(res, model, groupLine)
+            }
+
+            return
+          }
+
          if (stream) {
            const holdThisStream = Boolean(
              options.holdFirstStreamForPrompt && typeof lastUserMessage?.content === 'string' &&