fix(chat): group run output and hide empty messages

tests/client/chat-group-run-messages.test.ts

@@ -0,0 +1,22 @@
+import { describe, expect, it } from 'vitest'
+import { groupRunMessages } from '../../packages/client/src/components/hermes/chat/groupRunMessages'
+import type { Message } from '../../packages/client/src/stores/hermes/chat'
+
+const message = (id: string, role: Message['role'], content = ''): Message => ({ id, role, content, timestamp: 1 })
+
+describe('groupRunMessages', () => {
+  it('groups text across tool calls and marks completion only when the run ends', () => {
+    const input = [message('u', 'user', 'question'), message('a', 'assistant', 'first'), message('t', 'tool'), message('b', 'assistant', 'second')]
+    const active = groupRunMessages(input, true)
+    expect(active.map(m => m.role)).toEqual(['user', 'assistant', 'tool'])
+    expect(active[1]).toMatchObject({ id: 'b', content: 'first\n\nsecond' })
+    const finished = groupRunMessages(input, false)
+    expect(finished.at(-1)?.systemType).toBe('run-end')
+  })
+
+  it('keeps separate user turns and explicit run markers separate', () => {
+    const input = [message('u', 'user'), { ...message('a', 'assistant', 'one'), runMarker: '1' }, { ...message('b', 'assistant', 'two'), runMarker: '2' }, message('u2', 'user'), message('c', 'assistant', 'three')]
+    expect(groupRunMessages(input, false).filter(m => m.role === 'assistant').map(m => m.content)).toEqual(['one', 'two', 'three'])
+    expect(groupRunMessages(input, false).filter(m => m.systemType === 'run-end')).toHaveLength(3)
+  })
+})

tests/e2e/chat-rendering-smoke.spec.ts

@@ -82,6 +82,62 @@ test('streams a run and renders markdown, code, math and mermaid without leaking
   expect(api.unexpectedRequests).toEqual([])
 })
 
+test('does not render empty assistant bubbles between tool calls', async ({ page }) => {
+  await authenticate(page, TEST_ACCESS_KEY, 'research')
+  await mockHermesApi(page)
+  await mockChatSocket(page)
+  await page.goto('/#/hermes/chat')
+  await sendChatMessage(page, 'run tools')
+  const { sessionId } = await waitForRunSessionId(page)
+
+  await page.evaluate((sid) => {
+    const socket = (window as any).__PW_CHAT_SOCKET__.latest
+    const common = { session_id: sid, run_id: 'blank-bubbles' }
+    socket.__trigger('run.started', { ...common, event: 'run.started' })
+    for (let i = 0; i < 3; i++) {
+      socket.__trigger('message.delta', { ...common, event: 'message.delta', delta: '\n  ' })
+      socket.__trigger('tool.started', { ...common, event: 'tool.started', tool_call_id: `tool-${i}`, tool: 'terminal' })
+      socket.__trigger('tool.completed', { ...common, event: 'tool.completed', tool_call_id: `tool-${i}`, output: 'done' })
+    }
+  }, sessionId)
+  await expect(page.locator('.message.assistant')).toHaveCount(0)
+
+  await page.evaluate((sid) => {
+    const socket = (window as any).__PW_CHAT_SOCKET__.latest
+    socket.__trigger('message.delta', { event: 'message.delta', session_id: sid, run_id: 'blank-bubbles', delta: '\n' })
+    socket.__trigger('message.delta', { event: 'message.delta', session_id: sid, run_id: 'blank-bubbles', delta: 'Visible answer' })
+  }, sessionId)
+  await expect(page.locator('.message.assistant')).toHaveCount(1)
+  await expect(page.locator('.message.assistant')).toContainText('Visible answer')
+})
+
+test('groups assistant output across tool calls until the run ends', async ({ page }) => {
+  await authenticate(page, TEST_ACCESS_KEY, 'research')
+  await mockHermesApi(page)
+  await mockChatSocket(page)
+  await page.goto('/#/hermes/chat')
+  await sendChatMessage(page, 'inspect')
+  const { sessionId } = await waitForRunSessionId(page)
+  await page.evaluate((sid) => {
+    const socket = (window as any).__PW_CHAT_SOCKET__.latest
+    const common = { session_id: sid, run_id: 'grouped-run' }
+    socket.__trigger('run.started', { ...common, event: 'run.started' })
+    socket.__trigger('message.delta', { ...common, event: 'message.delta', delta: 'First update' })
+    socket.__trigger('tool.started', { ...common, event: 'tool.started', tool_call_id: 't1', tool: 'terminal' })
+    socket.__trigger('tool.completed', { ...common, event: 'tool.completed', tool_call_id: 't1', output: 'done' })
+    socket.__trigger('message.delta', { ...common, event: 'message.delta', delta: 'Second update' })
+  }, sessionId)
+  const assistant = page.locator('.message.assistant')
+  await expect(assistant).toHaveCount(1)
+  await expect(assistant).toContainText('First update')
+  await expect(assistant).toContainText('Second update')
+  await expect(page.locator('.run-end')).toHaveCount(0)
+  await page.evaluate((sid) => {
+    (window as any).__PW_CHAT_SOCKET__.latest.__trigger('run.completed', { event: 'run.completed', session_id: sid, run_id: 'grouped-run' })
+  }, sessionId)
+  await expect(page.locator('.run-end')).toHaveText('Run output ended')
+})
+
 test('opens the session model dialog and the session search from the chat header', async ({ page }) => {
   await authenticate(page, TEST_ACCESS_KEY, 'research')
   await mockHermesApi(page, { sessions: [sampleSession] })