Keep conversations ordered while background work updates independently

Why: Repeated progress and coupled scrolling obscured chat, and task choices lacked a reliable independent interaction path.

Evidence: 285 frontend unit tests and 23 browser scenarios pass locally before the shared event-loader extraction; final combination checks follow in Dev.

Impact: Coalesce cumulative progress, retain execution records, show timestamps and quiet notifications, support versioned work actions and explicit result references without automatic chat backfill.

Refs: feature/workbench-flow-completion

test/conversation-work.test.ts

@@ -69,6 +69,29 @@ describe('conversation work reading', () => {
     expect(html).not.toContain('data-testid="input-status"')
   })
 
+  it('renders stable semantic timestamps on user and assistant messages', () => {
+    snapshot.state!.messages = [{ ...reply([{ type: 'text', text: '请整理材料' }]), id: 'user', role: 'user' }, reply([{ type: 'text', text: '材料已收到' }])]
+    const html = render()
+    expect(html.match(/<time class="message-timestamp"/g)).toHaveLength(2)
+    expect(html.match(/dateTime="2026-10-12T10:00:00.000Z"/g)).toHaveLength(2)
+    expect(html).toContain('title="2026/10/12')
+  })
+
+  it('announces unread work without inserting its full result into the chat', () => {
+    snapshot.state!.ui.rightCollapsed = false
+    snapshot.state!.cards = [card({ status: 'done', work: { result: 'background-result-full-text', resultVersion: 2, readVersion: 1 } })]
+    const html = render()
+    expect(html).toContain('1 项工作有新结果')
+    expect(html).not.toContain('background-result-full-text')
+  })
+
+  it('shows an explicit removable result reference without copying its content into the draft', () => {
+    snapshot.state!.workReference = { cardId: 'work', resultVersion: 2, title: '材料整理' }
+    const html = render()
+    expect(html).toContain('基于「材料整理」的第 2 版结果')
+    expect(html).toContain('aria-label="移除工作结果引用"')
+  })
+
   it('shows the current real work with the auxiliary pane closed', () => {
     snapshot.state!.cards = [card()]
     const html = render()

test/live.test.ts

@@ -195,6 +195,27 @@ it('retains collaboration notes from a child after the parent has replied', () =
   expect(parts.every(part => part.replaySource === 'agent')).toBe(true)
 })
 
+it('coalesces cumulative child updates without mixing tasks or erasing tool phases', () => {
+  const notes = [
+    ['one', '我来检索'], ['two', '核对第二份材料'], ['one', '我来检索一下材料'],
+    ['one', '调用 knowledge_search'], ['one', '根据检索结果，开始整理'],
+    ...Array.from({ length: 80 }, () => ['one', '根据检索结果，开始整理完整报告']),
+  ]
+  const events: StreamEvent[] = notes.map(([agentRun, note], seq) => ({
+    seq: seq + 1, type: 'agent.progress', payload: { agentRun, note }, runId: agentRun,
+    conversationId: 'c', ts: '2026-10-15T10:00:00Z', v: 1,
+  }))
+  expect(runEventsToTimeline(events, 'parent').map(part => part.text)).toEqual([
+    '协作进展：我来检索一下材料', '协作进展：核对第二份材料',
+    '协作进展：调用 knowledge_search', '协作进展：根据检索结果，开始整理完整报告',
+  ])
+})
+
+it('uses the stable run start time for the temporary assistant message', () => {
+  expect(liveToMessage(live({ startedAt: Date.parse('2026-10-15T10:00:00Z') })).createdAt)
+    .toBe('2026-10-15T10:00:00.000Z')
+})
+
 describe('livePhaseLabel', () => {
   it('names the current activity in Chinese', () => {
     let items: LiveItem[] = applyThinkingDelta([], '想', 0)

test/scrollsync.test.ts

@@ -1,16 +0,0 @@
-import { describe, expect, it } from 'vitest'
-import { scrollRatio, scrollTopForRatio } from '../src/scrollsync'
-
-describe('scrollsync', () => {
-  it('没有溢出时不同步', () => {
-    expect(scrollRatio({ scrollTop: 0, scrollHeight: 100, clientHeight: 100 })).toBeNull()
-    expect(scrollTopForRatio({ scrollHeight: 80, clientHeight: 80 }, 0.5)).toBeNull()
-  })
-
-  it('按可滚动范围的比例映射，并夹在 0 到 1', () => {
-    expect(scrollRatio({ scrollTop: 50, scrollHeight: 200, clientHeight: 100 })).toBe(0.5)
-    expect(scrollTopForRatio({ scrollHeight: 300, clientHeight: 100 }, 0.5)).toBe(100)
-    expect(scrollTopForRatio({ scrollHeight: 300, clientHeight: 100 }, 2)).toBe(200)
-    expect(scrollTopForRatio({ scrollHeight: 300, clientHeight: 100 }, -1)).toBe(0)
-  })
-})

test/work-api.test.ts

@@ -0,0 +1,24 @@
+import { afterEach, describe, expect, it, vi } from 'vitest'
+import { api } from '../src/api'
+import { auth } from '../src/auth'
+
+afterEach(() => vi.restoreAllMocks())
+
+describe('versioned work request bodies', () => {
+  it('preserves multiple selections and free input on the card action endpoint', async () => {
+    const request = vi.spyOn(auth, 'fetch').mockResolvedValue(new Response(JSON.stringify({ card: { id: 'task' } })))
+    await api.cardAction('task', 'work.submit', { resultVersion: 3, selectedIds: ['summary', 'parameters'], freeText: '请说明依据' })
+    const [url, options] = request.mock.calls[0]
+    expect(url).toBe('/api/cards/task/actions/work.submit')
+    expect(JSON.parse(String(options?.body))).toEqual({ resultVersion: 3, selectedIds: ['summary', 'parameters'], freeText: '请说明依据' })
+  })
+
+  it('sends a result identity separately from the user message and discards client display metadata', async () => {
+    const request = vi.spyOn(auth, 'fetch').mockResolvedValue(new Response(JSON.stringify({ message: { id: 'reply' } })))
+    const reference = { cardId: 'task', resultVersion: 3, title: '说明书整理', result: 'must-not-be-client-context' }
+    await api.sendMessage('conversation', '比较这些重点', [], 'request-1', reference)
+    expect(JSON.parse(String(request.mock.calls[0][1]?.body))).toEqual({
+      text: '比较这些重点', attachmentIds: [], requestId: 'request-1', taskRef: { cardId: 'task', resultVersion: 3 },
+    })
+  })
+})

test/work-state.test.ts

@@ -248,6 +248,70 @@ describe('work projection ordering and context', () => {
 })
 
 describe('shared action feedback', () => {
+  it('sends a versioned optional choice once and keeps it out of the conversation', async () => {
+    const task = { ...card('work-choice', 'done'), work: { resultVersion: 2, readVersion: 0,
+      result: '完整材料结果', decision: { mode: 'multiple' as const, state: 'open' as const } } }
+    store.upsertCard(task)
+    const response = deferred<Awaited<ReturnType<typeof api.cardAction>>>()
+    const action = vi.spyOn(api, 'cardAction').mockReturnValue(response.promise)
+    const payload = { resultVersion: 2, selectedIds: ['summary', 'facts'], freeText: '请保留来源' }
+    const pending = store.cardAction(task, 'work.submit', payload)
+    await store.cardAction(task, 'work.submit', payload)
+    expect(action).toHaveBeenCalledOnce()
+    expect(action).toHaveBeenCalledWith(task.id, 'work.submit', payload)
+    response.resolve({ card: { ...task, work: { ...task.work, decision: { ...task.work.decision, state: 'submitted' } } },
+      receipt: { id: 'unexpected-work-receipt', conversationId: 'a', role: 'assistant', parts: [], createdAt: '' } })
+    await pending
+    expect(store.getState().messages).toEqual([])
+    expect(store.getState().cardActionStates[task.id]).toBeUndefined()
+    expect(store.getState().cards.find(c => c.id === task.id)?.work?.decision?.state).toBe('submitted')
+  })
+
+  it('does not unlock an uncertain optional submission merely because its result is done', async () => {
+    const task = { ...card('work-unknown', 'done'), work: { resultVersion: 1, readVersion: 0,
+      result: '结果', decision: { mode: 'single' as const, state: 'open' as const } } }
+    store.upsertCard(task)
+    vi.spyOn(api, 'cardAction').mockRejectedValue(new TypeError('connection lost'))
+    vi.mocked(api.listCards).mockResolvedValue([task])
+    await store.cardAction(task, 'work.submit', { resultVersion: 1, selectedIds: ['summary'] })
+    expect(store.getState().cardActionStates[task.id]).toBe('unknown')
+    store.upsertCard({ ...task, work: { ...task.work, decision: { ...task.work.decision, state: 'submitted' } } })
+    expect(store.getState().cardActionStates[task.id]).toBeUndefined()
+  })
+
+  it('settles successful metadata actions while an ordinary card still needs action', async () => {
+    const task = { ...card('work-read', 'action_required'), work: { resultVersion: 1, readVersion: 0 } }
+    store.upsertCard(task)
+    vi.spyOn(api, 'cardAction').mockResolvedValue({ card: { ...task, work: { ...task.work, readVersion: 1 } } })
+    await store.cardAction(task, 'work.read', { resultVersion: 1 })
+    expect(store.getState().cardActionStates[task.id]).toBeUndefined()
+  })
+
+  it('retains an explicit result reference across a queued send without adding it to chat text', async () => {
+    const task = { ...card('work-ref', 'done'), work: { result: '完整结果', resultVersion: 3, readVersion: 3 } }
+    store.upsertCard(task)
+    store.referenceWork(task)
+    const reference = store.getState().workReference!
+    store.enqueueOutbox('比较重点', [], [], reference)
+    store.clearWorkReference()
+    expect(store.getState().outbox[0].taskRef).toEqual({ cardId: 'work-ref', resultVersion: 3, title: '整理材料' })
+    const send = vi.spyOn(api, 'sendMessage').mockResolvedValue({ message: { id: 'queued-ref', conversationId: 'a', role: 'user', parts: [], createdAt: '' } })
+    await store.flushOutbox()
+    expect(send).toHaveBeenCalledWith('a', '比较重点', [], expect.any(String), reference)
+    expect(store.getState().workReference).toBeNull()
+  })
+
+  it('clears record selection and result reference when navigating to another conversation', async () => {
+    const task = { ...card('work-record', 'done'), sourceRunId: 'child', work: { result: '完整结果', resultVersion: 1, readVersion: 1 } }
+    store.upsertCard(task)
+    store.openWorkRecord(task)
+    store.referenceWork(task)
+    expect(store.getState().workRecord?.runId).toBe('child')
+    await store.openConversation('b')
+    expect(store.getState().workRecord).toBeNull()
+    expect(store.getState().workReference).toBeNull()
+  })
+
   it.each(['card', 'hitl'] as const)('clears its definite rejected %s action after A to B navigation', async kind => {
     const confirm = { ...card(`reject-${kind}`, 'action_required'), kind: 'confirm' as const,
       ...(kind === 'hitl' ? { body: [{ type: 'kv' as const, kv: [['来源', `hitl:reject-${kind}`]] as [string, string][] }] } : {}) }