diff --git a/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.test.tsx b/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.test.tsx
index b3326b4cb..724237e0a 100644
--- a/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.test.tsx
+++ b/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.test.tsx
@@ -10,6 +10,10 @@ import {
 } from '@testing-library/react';
 import { beforeEach, describe, expect, it, vi } from 'vitest';
 
+import type {
+  EditorAgentMessage,
+  EditorAgentMessageResponse,
+} from '@/packages/shared/src/contracts';
 import type { EditorAgentConversationClient } from '@/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.ts';
 import { useImageCanvasContextStore } from '@/src/components/image-editor/useImageCanvasContextStore.ts';
 
@@ -19,24 +23,24 @@ const createEditorProjectResourceMock = vi.hoisted(() => vi.fn());
 const uploadEditorMediaAssetFileMock = vi.hoisted(() => vi.fn());
 const probeImageFileDimensionsMock = vi.hoisted(() => vi.fn());
 
-vi.mock('../../services/image-editor/editorProjectClient', async () => {
+vi.mock('@/src/services/image-editor/editorProjectClient.ts', async () => {
   const actual = await vi.importActual<
     typeof import('../../../services/image-editor/editorProjectClient.ts')
-  >('../../services/image-editor/editorProjectClient');
+  >('@/src/services/image-editor/editorProjectClient.ts');
   return {
     ...actual,
     createEditorProjectResource: createEditorProjectResourceMock,
   };
 });
 
-vi.mock('../../services/image-editor/editorMediaAssetUploadClient', () => ({
+vi.mock('@/src/services/image-editor/editorMediaAssetUploadClient.ts', () => ({
   uploadEditorMediaAssetFile: uploadEditorMediaAssetFileMock,
 }));
 
-vi.mock('./ImageCanvasFileModel', async () => {
-  const actual = await vi.importActual<typeof import('../ImageCanvasFileModel.ts')>(
-    './ImageCanvasFileModel',
-  );
+vi.mock('@/src/components/image-editor/ImageCanvasFileModel.ts', async () => {
+  const actual = await vi.importActual<
+    typeof import('../ImageCanvasFileModel.ts')
+  >('@/src/components/image-editor/ImageCanvasFileModel.ts');
   return {
     ...actual,
     probeImageFileDimensions: probeImageFileDimensionsMock,
@@ -49,8 +53,6 @@ function createClient(): EditorAgentConversationClient {
       {
         conversationId: 'conversation-1',
         projectId: 'project-1',
-        title: '角色参考',
-        createdAt: '2026-07-03T00:00:00.000Z',
         updatedAt: '2026-07-03T00:00:00.000Z',
       },
     ]),
@@ -68,13 +70,11 @@ function createClient(): EditorAgentConversationClient {
       title: '角色参考',
       messages: [
         {
-          id: 'assistant-old',
+          id: 1,
           role: 'assistant',
-          kind: 'chat',
           text: '已经看到画布内容',
           attachments: [],
-          generations: [],
-          status: 'completed',
+          toolCall: null,
           createdAt: '2026-07-03T00:00:10.000Z',
         },
       ],
@@ -89,28 +89,43 @@ function createClient(): EditorAgentConversationClient {
       createdAt: '2026-07-03T00:00:00.000Z',
       updatedAt: '2026-07-03T00:00:00.000Z',
     }),
-    streamMessage: vi
-      .fn()
-      .mockImplementation(async (_conversationId, _payload, options) => {
-        options.onEvent?.({
-          event: 'stage',
-          data: { conversationId: _conversationId, stage: 'responding' },
-        });
-        options.onEvent?.({
-          event: 'message_delta',
-          data: {
-            conversationId: _conversationId,
-            messageId: 'assistant-stream',
-            role: 'assistant',
-            kind: 'chat',
-            textDelta: '收到，我会参考这张图。',
-          },
-        });
-        options.onEvent?.({
-          event: 'done',
-          data: { conversationId: _conversationId, title: null },
-        });
-      }),
+    sendMessage: vi.fn().mockResolvedValue({
+      deltaMessages: [
+        {
+          id: 2,
+          role: 'assistant',
+          text: '收到，我会参考这张图。',
+          attachments: [],
+          toolCall: null,
+          createdAt: '2026-07-03T00:00:00.000Z',
+        },
+      ],
+      errorMessage: null,
+    } as EditorAgentMessageResponse),
+    confirmToolCall: vi.fn(),
+    cancelToolCall: vi.fn(),
+  };
+}
+
+function createPendingToolCallMessage(): EditorAgentMessage {
+  return {
+    id: 2,
+    role: 'system',
+    text: 'internal system prompt that must stay hidden',
+    attachments: [],
+    toolCall: {
+      toolName: 'edit-image',
+      summary: '',
+      status: 'pending_confirmation',
+      args: {
+        object_image_id: 'source-image-1',
+        reference_image_ids: ['reference-image-1', 'reference-image-2'],
+        prompt: '把角色换成像素风',
+      },
+      images: [],
+      error: null,
+    },
+    createdAt: '2026-07-10T00:00:00.000Z',
   };
 }
 
@@ -120,7 +135,8 @@ describe('EditorAgentConversationPanelView', () => {
     uploadEditorMediaAssetFileMock.mockReset();
     uploadEditorMediaAssetFileMock.mockResolvedValue({
       src: '/generated/pasted.png',
-      objectKey: 'generated-character-drafts/editor/agent-paste/image/pasted.png',
+      objectKey:
+        'generated-character-drafts/editor/agent-paste/image/pasted.png',
       assetObjectId: 'asset-object-pasted',
       legacyPublicPath: '/generated/pasted.png',
     });
@@ -130,7 +146,8 @@ describe('EditorAgentConversationPanelView', () => {
     createEditorProjectResourceMock.mockResolvedValue({
       resourceId: 'resource-pasted',
       imageSrc: '/generated/pasted.png',
-      objectKey: 'generated-character-drafts/editor/agent-paste/image/pasted.png',
+      objectKey:
+        'generated-character-drafts/editor/agent-paste/image/pasted.png',
       label: '粘贴图片',
       width: 320,
       height: 240,
@@ -227,7 +244,7 @@ describe('EditorAgentConversationPanelView', () => {
     await waitFor(() => {
       expect(screen.getByText('收到，我会参考这张图。')).toBeTruthy();
     });
-    expect(client.streamMessage).toHaveBeenCalledWith(
+    expect(client.sendMessage).toHaveBeenCalledWith(
       'conversation-2',
       expect.objectContaining({
         text: '参考附件做像素风',
@@ -282,7 +299,8 @@ describe('EditorAgentConversationPanelView', () => {
       'project-1',
       expect.objectContaining({
         imageSrc: '/generated/pasted.png',
-        objectKey: 'generated-character-drafts/editor/agent-paste/image/pasted.png',
+        objectKey:
+          'generated-character-drafts/editor/agent-paste/image/pasted.png',
         assetObjectId: 'asset-object-pasted',
         width: 320,
         height: 240,
@@ -296,7 +314,7 @@ describe('EditorAgentConversationPanelView', () => {
     fireEvent.click(screen.getByRole('button', { name: '发送' }));
 
     await waitFor(() => {
-      expect(client.streamMessage).toHaveBeenCalledWith(
+      expect(client.sendMessage).toHaveBeenCalledWith(
         'conversation-1',
         expect.objectContaining({
           text: '',
@@ -361,7 +379,7 @@ describe('EditorAgentConversationPanelView', () => {
     fireEvent.click(screen.getByRole('button', { name: '发送' }));
 
     await waitFor(() => {
-      expect(client.streamMessage).toHaveBeenCalledWith(
+      expect(client.sendMessage).toHaveBeenCalledWith(
         'conversation-1',
         expect.objectContaining({
           text: '',
@@ -377,41 +395,21 @@ describe('EditorAgentConversationPanelView', () => {
     });
   });
 
-  it('keeps streamed reply and generating state when the panel is collapsed and reopened', async () => {
+  it('preserves messages when the panel is collapsed and reopened', async () => {
     const client = createClient();
-    let finishStream = () => {};
-    vi.mocked(client.streamMessage).mockImplementation(
-      async (conversationId, _payload, options) => {
-        options.onEvent?.({
-          event: 'message_delta',
-          data: {
-            conversationId,
-            messageId: 'assistant-stream',
-            role: 'assistant',
-            kind: 'chat',
-            textDelta: '我来生成图片。',
-          },
-        });
-        options.onEvent?.({
-          event: 'stage',
-          data: { conversationId, stage: 'generating' },
-        });
-        options.onEvent?.({
-          event: 'tool_started',
-          data: {
-            conversationId,
-            messageId: 'assistant-stream',
-            toolCallId: 'tool-call-generating',
-            toolName: 'generate_image',
-            taskId: 'task-generating',
-            model: 'gpt-image-2',
-          },
-        });
-        await new Promise<void>((resolve) => {
-          finishStream = resolve;
-        });
-      },
-    );
+    vi.mocked(client.sendMessage).mockResolvedValue({
+      deltaMessages: [
+        {
+          id: 3,
+          role: 'assistant',
+          text: '我来生成图片。',
+          attachments: [],
+          toolCall: null,
+          createdAt: '2026-07-03T00:00:00.000Z',
+        },
+      ],
+      errorMessage: null,
+    } as EditorAgentMessageResponse);
 
     const { rerender } = render(
       <EditorAgentConversationPanelView
@@ -433,8 +431,6 @@ describe('EditorAgentConversationPanelView', () => {
     await waitFor(() => {
       expect(screen.getByText('我来生成图片。')).toBeTruthy();
     });
-    expect(screen.getByText('生成中')).toBeTruthy();
-    expect(screen.getByRole('button', { name: '停止' })).toBeTruthy();
 
     rerender(
       <EditorAgentConversationPanelView
@@ -453,13 +449,127 @@ describe('EditorAgentConversationPanelView', () => {
       />,
     );
 
+    expect(screen.getByText('已经看到画布内容')).toBeTruthy();
     expect(screen.getByText('我来生成图片。')).toBeTruthy();
-    expect(screen.getByText('生成中')).toBeTruthy();
-    expect(screen.getByRole('button', { name: '停止' })).toBeTruthy();
+  });
+
+  it('shows structured pending details, hides system text and confirms once', async () => {
+    const client = createClient();
+    const pendingMessage = createPendingToolCallMessage();
+    vi.mocked(client.getConversation).mockResolvedValue({
+      conversationId: 'conversation-1',
+      projectId: 'project-1',
+      title: '角色参考',
+      messages: [pendingMessage],
+      createdAt: '2026-07-03T00:00:00.000Z',
+      updatedAt: '2026-07-03T00:00:10.000Z',
+    });
+    let resolveConfirmation: (message: EditorAgentMessage) => void = () => {};
+    vi.mocked(client.confirmToolCall).mockImplementation(
+      () =>
+        new Promise<EditorAgentMessage>((resolve) => {
+          resolveConfirmation = resolve;
+        }),
+    );
+    const onCanvasRefreshRequested = vi.fn();
+
+    render(
+      <EditorAgentConversationPanelView
+        open
+        onToggleOpen={vi.fn()}
+        client={client}
+        onCanvasRefreshRequested={onCanvasRefreshRequested}
+      />,
+    );
+
+    expect(await screen.findByText('把角色换成像素风')).toBeTruthy();
+    expect(screen.getByText('source-image-1')).toBeTruthy();
+    expect(screen.getByText('2 张')).toBeTruthy();
+    expect(
+      screen.queryByText('internal system prompt that must stay hidden'),
+    ).toBeNull();
+
+    fireEvent.click(screen.getByRole('button', { name: '确认' }));
+    await waitFor(() => {
+      expect(client.confirmToolCall).toHaveBeenCalledWith('conversation-1', 2);
+    });
+    expect(
+      (screen.getByRole('button', { name: '执行中' }) as HTMLButtonElement)
+        .disabled,
+    ).toBe(true);
+    expect(
+      (screen.getByRole('button', { name: '取消' }) as HTMLButtonElement)
+        .disabled,
+    ).toBe(true);
 
     await act(async () => {
-      finishStream();
+      resolveConfirmation({
+        ...pendingMessage,
+        text: 'internal completed tool output that must stay hidden',
+        toolCall: {
+          ...pendingMessage.toolCall!,
+          status: 'completed',
+          images: [
+            {
+              resourceId: null,
+              objectKey: 'generated/result.png',
+              imageSrc: '/result.png',
+              thumbnailSrc: null,
+              width: 512,
+              height: 512,
+            },
+          ],
+        },
+      });
     });
+
+    expect(await screen.findByText('已完成')).toBeTruthy();
+    expect(
+      screen.queryByText(
+        'internal completed tool output that must stay hidden',
+      ),
+    ).toBeNull();
+    expect(onCanvasRefreshRequested).toHaveBeenCalledTimes(1);
+  });
+
+  it('cancels a pending tool call without exposing its system text', async () => {
+    const client = createClient();
+    const pendingMessage = createPendingToolCallMessage();
+    vi.mocked(client.getConversation).mockResolvedValue({
+      conversationId: 'conversation-1',
+      projectId: 'project-1',
+      title: '角色参考',
+      messages: [pendingMessage],
+      createdAt: '2026-07-03T00:00:00.000Z',
+      updatedAt: '2026-07-03T00:00:10.000Z',
+    });
+    vi.mocked(client.cancelToolCall).mockResolvedValue({
+      ...pendingMessage,
+      text: 'internal cancelled tool output that must stay hidden',
+      toolCall: {
+        ...pendingMessage.toolCall!,
+        status: 'cancelled',
+      },
+    });
+
+    render(
+      <EditorAgentConversationPanelView
+        open
+        onToggleOpen={vi.fn()}
+        client={client}
+      />,
+    );
+
+    fireEvent.click(await screen.findByRole('button', { name: '取消' }));
+
+    expect(await screen.findByText('已取消')).toBeTruthy();
+    expect(client.cancelToolCall).toHaveBeenCalledWith('conversation-1', 2);
+    expect(
+      screen.queryByText(
+        'internal cancelled tool output that must stay hidden',
+      ),
+    ).toBeNull();
+    expect(screen.queryByRole('button', { name: '确认' })).toBeNull();
   });
 
   it('keeps wheel scrolling inside the message history and input', async () => {
diff --git a/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.tsx b/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.tsx
index 9c6f96516..1207c2449 100644
--- a/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.tsx
+++ b/src/components/image-editor/EditorAgentConversation/EditorAgentConversationPanelView.tsx
@@ -1,6 +1,5 @@
 import {
   Bot,
-  Image as ImageIcon,
   Loader2,
   MessageCircle,
   Paperclip,
@@ -22,8 +21,6 @@ import {
 import {
   EDITOR_AGENT_MAX_ATTACHMENTS,
   type EditorAgentAttachmentRef,
-  type EditorAgentGenerationResultEvent,
-  type EditorAgentStage,
 } from '@/packages/shared/src/contracts';
 import { PlatformActionButton } from '@/src/components/common/PlatformActionButton.tsx';
 import { PlatformDangerConfirmDialog } from '@/src/components/common/PlatformDangerConfirmDialog.tsx';
@@ -32,6 +29,7 @@ import { attachmentKey } from '@/src/components/image-editor/EditorAgentConversa
 import {
   AttachmentChip,
   MessageBubble,
+  ThinkingBubble,
 } from '@/src/components/image-editor/EditorAgentConversation/MessageBubble.tsx';
 import type {
   CanvasLayer,
@@ -61,7 +59,7 @@ type EditorAgentConversationPanelViewProps = {
   onToggleOpen: () => void;
   layers?: CanvasLayer[];
   assets?: EditorAsset[];
-  onGenerationResult?: (event: EditorAgentGenerationResultEvent) => void;
+  onCanvasRefreshRequested?: () => void;
   client?: EditorAgentConversationClient;
 };
 
@@ -125,183 +123,6 @@ function createLibraryAttachmentOptions(
   });
 }
 
-function stageLabel(stage: EditorAgentStage) {
-  if (stage === 'thinking') {
-    return '思考中';
-  }
-  if (stage === 'responding') {
-    return '回复中';
-  }
-  if (stage === 'generating') {
-    return '生成中';
-  }
-  if (stage === 'completed') {
-    return '完成';
-  }
-  if (stage === 'failed') {
-    return '失败';
-  }
-  return '';
-}
-
-function toolLabel(toolName: EditorAgentGenerationRecord['toolName']) {
-  if (toolName === 'edit_image') {
-    return '修改图片';
-  }
-  if (toolName === 'generate_character') {
-    return '生成角色';
-  }
-  if (toolName === 'generate_icon_spritesheet') {
-    return '生成图标';
-  }
-  if (toolName === 'generate_ui_design') {
-    return '生成 UI';
-  }
-  return '生成图片';
-}
-
-function messageRoleLabel(role: EditorAgentMessage['role']) {
-  return role === 'user' ? '你' : 'Agent';
-}
-
-function AttachmentChip({
-  attachment,
-  onRemove,
-}: {
-  attachment: EditorAgentAttachmentRef;
-  onRemove?: () => void;
-}) {
-  const label = attachment.label?.trim() || attachment.referenceId;
-  return (
-    <span className="group relative inline-flex max-w-full items-center gap-1.5 rounded-full border border-slate-200 bg-white px-2.5 py-1 text-xs text-slate-600 shadow-sm">
-      <ImageIcon className="h-3.5 w-3.5 shrink-0" aria-hidden="true" />
-      <span className="truncate">{label}</span>
-      {onRemove ? (
-        <button
-          type="button"
-          className="-mr-1 inline-flex h-5 w-5 items-center justify-center rounded-full text-slate-400 hover:bg-slate-100 hover:text-slate-700"
-          aria-label={`移除附件 ${label}`}
-          onClick={onRemove}
-        >
-          <X className="h-3 w-3" aria-hidden="true" />
-        </button>
-      ) : null}
-      {attachment.thumbnailSrc || attachment.imageSrc ? (
-        <span className="pointer-events-none absolute bottom-[calc(100%+0.4rem)] left-0 hidden rounded-2xl border border-white bg-white p-1 shadow-xl group-hover:block">
-          <ResolvedAssetImage
-            src={attachment.thumbnailSrc ?? attachment.imageSrc}
-            objectKey={attachment.objectKey}
-            refreshKey={attachment.referenceId}
-            alt=""
-            className="h-24 w-24 rounded-xl object-cover"
-          />
-        </span>
-      ) : null}
-    </span>
-  );
-}
-
-function GenerationRecordsView({
-  generations,
-}: {
-  generations: EditorAgentGenerationRecord[];
-}) {
-  if (!generations.length) {
-    return null;
-  }
-  return (
-    <div className="mt-2 space-y-2">
-      {generations.map((generation) => (
-        <div
-          key={generation.toolCallId}
-          className="rounded-2xl border border-slate-200 bg-white/80 p-2 text-xs text-slate-600"
-        >
-          {generation.status === 'generating' || generation.error ? (
-            <div className="flex items-center gap-2">
-              {generation.status === 'generating' ? (
-                <Loader2
-                  className="h-3.5 w-3.5 animate-spin"
-                  aria-hidden="true"
-                />
-              ) : (
-                <ImageIcon className="h-3.5 w-3.5" aria-hidden="true" />
-              )}
-              <span>{toolLabel(generation.toolName)}</span>
-              {generation.model ? <span>{generation.model}</span> : null}
-            </div>
-          ) : null}
-          {generation.error ? (
-            <div className="mt-1 text-red-600">{generation.error}</div>
-          ) : null}
-          {generation.images.length ? (
-            <div className="mt-2 grid grid-cols-3 gap-2">
-              {generation.images.map((image, index) => (
-                <div
-                  key={`${generation.toolCallId}-${image.resourceId ?? index}`}
-                  className="overflow-hidden rounded-xl border border-slate-200 bg-slate-100"
-                >
-                  <ResolvedAssetImage
-                    src={image.thumbnailSrc ?? image.imageSrc}
-                    objectKey={image.objectKey}
-                    refreshKey={
-                      image.resourceId ??
-                      generation.taskId ??
-                      generation.toolCallId
-                    }
-                    alt=""
-                    className="h-20 w-full object-cover"
-                  />
-                </div>
-              ))}
-            </div>
-          ) : null}
-        </div>
-      ))}
-    </div>
-  );
-}
-
-function MessageBubble({
-  message,
-}: {
-  message: EditorAgentMessage;
-}) {
-  const isUser = message.role === 'user';
-  return (
-    <article
-      className={`flex ${isUser ? 'justify-end' : 'justify-start'}`}
-      aria-label={`${messageRoleLabel(message.role)}消息`}
-    >
-      <div
-        className={`max-w-[86%] rounded-3xl px-3.5 py-3 text-sm leading-6 shadow-sm ${
-          isUser
-            ? 'bg-slate-900 text-white'
-            : message.kind === 'error'
-              ? 'border border-red-200 bg-red-50 text-red-700'
-              : 'border border-slate-200 bg-white text-slate-700'
-        }`}
-      >
-        <div className="whitespace-pre-wrap break-words">
-          {message.text || (message.status === 'streaming' ? '...' : '')}
-        </div>
-        {message.attachments.length ? (
-          <div className="mt-2 flex flex-wrap gap-1.5">
-            {message.attachments.map((attachment) => (
-              <AttachmentChip
-                key={attachmentKey(attachment)}
-                attachment={attachment}
-              />
-            ))}
-          </div>
-        ) : null}
-        <GenerationRecordsView
-          generations={message.generations}
-        />
-      </div>
-    </article>
-  );
-}
-
 function AttachmentPickerModal({
   open,
   tab,
@@ -415,7 +236,7 @@ export function EditorAgentConversationPanelView({
   onToggleOpen,
   layers = [],
   assets = [],
-  onGenerationResult,
+  onCanvasRefreshRequested,
   client,
 }: EditorAgentConversationPanelViewProps) {
   const [hasConversationMounted, setHasConversationMounted] = useState(open);
@@ -424,30 +245,31 @@ export function EditorAgentConversationPanelView({
       setHasConversationMounted(true);
     }
   }, [open]);
-  const projectId = useImageCanvasContextStore(
-    (state) => state.projectId,
-  );
+  const projectId = useImageCanvasContextStore((state) => state.projectId);
   const effectiveProjectId = hasConversationMounted ? projectId : null;
   const {
     conversations,
     activeConversationId,
     messages,
-    stage,
     isLoadingConversations,
     isLoadingMessages,
     isCreatingConversation,
     isDeletingConversation,
-    isStreaming,
+    isWaiting,
+    toolCallAction,
+    isToolCallActionPending,
     errorMessage,
     createConversation,
     selectConversation,
     sendMessage,
     stopCurrentTurn,
+    confirmToolCall,
+    cancelToolCall,
     deleteActiveConversation,
   } = useEditorAgentConversation({
     projectId: effectiveProjectId,
     client,
-    onGenerationResult,
+    onCanvasRefreshRequested,
   });
   const [draftText, setDraftText] = useState('');
   const [attachments, setAttachments] = useState<EditorAgentAttachmentRef[]>(
@@ -479,8 +301,8 @@ export function EditorAgentConversationPanelView({
     return optionMap;
   }, [canvasAttachmentOptions, libraryAttachmentOptions]);
 
-  const currentStageLabel = stageLabel(stage);
   const hasProject = Boolean(projectId?.trim());
+  const isConversationBusy = isWaiting || isToolCallActionPending;
 
   const openAttachmentPicker = () => {
     setAttachmentError(null);
@@ -516,7 +338,7 @@ export function EditorAgentConversationPanelView({
 
   const submitMessage = (event: FormEvent<HTMLFormElement>) => {
     event.preventDefault();
-    if (isStreaming) {
+    if (isWaiting) {
       stopCurrentTurn();
       return;
     }
@@ -612,7 +434,9 @@ export function EditorAgentConversationPanelView({
     return [...fileItems];
   }
 
-  const handleInputPaste = (event: ReactClipboardEvent<HTMLTextAreaElement>) => {
+  const handleInputPaste = (
+    event: ReactClipboardEvent<HTMLTextAreaElement>,
+  ) => {
     const imageFiles = extractClipboardImageFiles(event.clipboardData);
     if (!imageFiles.length) {
       return;
@@ -684,7 +508,9 @@ export function EditorAgentConversationPanelView({
             aria-label="当前对话"
             value={activeConversationId ?? ''}
             disabled={
-              !conversations.length || isLoadingConversations || isStreaming
+              !conversations.length ||
+              isLoadingConversations ||
+              isConversationBusy
             }
             onChange={(event) => {
               const nextConversationId = event.currentTarget.value;
@@ -699,7 +525,7 @@ export function EditorAgentConversationPanelView({
                   key={conversation.conversationId}
                   value={conversation.conversationId}
                 >
-                  {conversation.title}
+                  {conversation.conversationId}
                 </option>
               ))
             ) : (
@@ -710,7 +536,9 @@ export function EditorAgentConversationPanelView({
             type="button"
             className="inline-flex h-9 w-9 items-center justify-center rounded-full bg-slate-900 text-white disabled:opacity-45"
             aria-label="新建对话"
-            disabled={!hasProject || isCreatingConversation || isStreaming}
+            disabled={
+              !hasProject || isCreatingConversation || isConversationBusy
+            }
             onClick={() => void createConversation()}
           >
             <Plus className="h-4 w-4" aria-hidden="true" />
@@ -720,7 +548,9 @@ export function EditorAgentConversationPanelView({
             className="inline-flex h-9 w-9 items-center justify-center rounded-full bg-white text-slate-500 hover:bg-red-50 hover:text-red-600 disabled:opacity-45"
             aria-label="删除当前对话"
             disabled={
-              !activeConversationId || isDeletingConversation || isStreaming
+              !activeConversationId ||
+              isDeletingConversation ||
+              isConversationBusy
             }
             onClick={() => setDeleteConfirmOpen(true)}
           >
@@ -735,15 +565,16 @@ export function EditorAgentConversationPanelView({
             <X className="h-4 w-4" aria-hidden="true" />
           </button>
         </header>
-        {currentStageLabel ? (
+        {isConversationBusy ? (
           <div className="flex items-center gap-2 border-b border-slate-200 bg-white/60 px-4 py-2 text-xs text-slate-500">
-            {isStreaming ? (
-              <Loader2
-                className="h-3.5 w-3.5 animate-spin"
-                aria-hidden="true"
-              />
-            ) : null}
-            <span>{currentStageLabel}</span>
+            <Loader2 className="h-3.5 w-3.5 animate-spin" aria-hidden="true" />
+            <span>
+              {toolCallAction?.action === 'confirm'
+                ? '执行中'
+                : toolCallAction?.action === 'cancel'
+                  ? '取消中'
+                  : '思考中'}
+            </span>
           </div>
         ) : null}
         <div
@@ -757,12 +588,26 @@ export function EditorAgentConversationPanelView({
               加载中
             </div>
           ) : messages.length ? (
-            messages.map((message) => (
-              <MessageBubble
-                key={message.id}
-                message={message}
-              />
-            ))
+            <>
+              {messages.map((message) => (
+                <MessageBubble
+                  key={message.id}
+                  message={message}
+                  busyAction={
+                    toolCallAction?.messageId === message.id
+                      ? toolCallAction.action
+                      : null
+                  }
+                  onConfirmToolCall={(messageId) => {
+                    void confirmToolCall(messageId);
+                  }}
+                  onCancelToolCall={(messageId) => {
+                    void cancelToolCall(messageId);
+                  }}
+                />
+              ))}
+              {isWaiting ? <ThinkingBubble /> : null}
+            </>
           ) : (
             <div className="rounded-3xl border border-dashed border-slate-200 bg-white/70 px-4 py-8 text-center text-sm text-slate-400">
               暂无消息
@@ -821,11 +666,12 @@ export function EditorAgentConversationPanelView({
               type="submit"
               className="inline-flex h-10 min-w-16 shrink-0 items-center justify-center gap-1.5 rounded-full bg-slate-900 px-3 text-sm font-semibold text-white disabled:opacity-45"
               disabled={
-                !isStreaming &&
-                ((!draftText.trim() && !attachments.length) || !hasProject)
+                isToolCallActionPending ||
+                (!isWaiting &&
+                  ((!draftText.trim() && !attachments.length) || !hasProject))
               }
             >
-              {isStreaming ? (
+              {isWaiting ? (
                 <>
                   <Square className="h-3.5 w-3.5" aria-hidden="true" />
                   停止
diff --git a/src/components/image-editor/EditorAgentConversation/MessageBubble.tsx b/src/components/image-editor/EditorAgentConversation/MessageBubble.tsx
index 5d2bb3f5d..c17232fed 100644
--- a/src/components/image-editor/EditorAgentConversation/MessageBubble.tsx
+++ b/src/components/image-editor/EditorAgentConversation/MessageBubble.tsx
@@ -1,42 +1,20 @@
-import { FileAudio, Image as ImageIcon, Loader2, X } from 'lucide-react';
+import { Check, Image as ImageIcon, Loader2, X } from 'lucide-react';
 
 import type {
   EditorAgentAttachmentRef,
-  EditorAgentGenerationRecord,
   EditorAgentMessage,
+  EditorAgentToolCall,
 } from '@/packages/shared/src/contracts';
 import { attachmentKey } from '@/src/components/image-editor/EditorAgentConversation/common.ts';
+import { PendingToolCall } from '@/src/components/image-editor/EditorAgentConversation/PendingToolCall.tsx';
+import { editorAgentToolLabel } from '@/src/components/image-editor/EditorAgentConversation/toolCallPresentation.ts';
 import { ResolvedAssetImage } from '@/src/components/ResolvedAssetImage.tsx';
-import { ResolvedAssetVideo } from '@/src/components/ResolvedAssetVideo.tsx';
-import { useResolvedAssetReadUrl } from '@/src/hooks/useResolvedAssetReadUrl.ts';
-
-function toolLabel(toolName: EditorAgentGenerationRecord['toolName']) {
-  if (toolName === 'edit_image') {
-    return '修改图片';
-  }
-  if (toolName === 'generate_character') {
-    return '生成角色';
-  }
-  if (toolName === 'generate_icon_spritesheet') {
-    return '生成图标';
-  }
-  if (toolName === 'generate_ui_design') {
-    return '生成 UI';
-  }
-  if (toolName === 'generate_video') {
-    return '生成视频';
-  }
-  if (toolName === 'generate_sound_effect') {
-    return '生成音效';
-  }
-  if (toolName === 'generate_background_music') {
-    return '生成背景音乐';
-  }
-  return '生成图片';
-}
 
 function messageRoleLabel(role: EditorAgentMessage['role']) {
-  return role === 'user' ? '你' : 'Agent';
+  if (role === 'user') {
+    return '你';
+  }
+  return 'Agent';
 }
 
 export function AttachmentChip({
@@ -76,139 +54,140 @@ export function AttachmentChip({
   );
 }
 
-function GenerationRecordsView({
-  generations,
-}: {
-  generations: EditorAgentGenerationRecord[];
-}) {
-  if (!generations.length) {
-    return null;
-  }
+function ToolCallView({ toolCall }: { toolCall: EditorAgentToolCall }) {
+  const isExecuting = toolCall.status === 'executing';
+  const statusLabel =
+    toolCall.status === 'completed'
+      ? '已完成'
+      : toolCall.status === 'cancelled'
+        ? '已取消'
+        : toolCall.status === 'failed'
+          ? '失败'
+          : toolCall.status === 'pending_confirmation'
+            ? '待确认'
+            : '执行中';
   return (
-    <div className="mt-2 space-y-2">
-      {generations.map((generation) => (
-        <div
-          key={generation.toolCallId}
-          className="rounded-2xl border border-slate-200 bg-white/80 p-2 text-xs text-slate-600"
-        >
-          {generation.status === 'generating' || generation.error ? (
-            <div className="flex items-center gap-2">
-              {generation.status === 'generating' ? (
-                <Loader2
-                  className="h-3.5 w-3.5 animate-spin"
-                  aria-hidden="true"
-                />
-              ) : (
-                <ImageIcon className="h-3.5 w-3.5" aria-hidden="true" />
-              )}
-              <span>{toolLabel(generation.toolName)}</span>
-              {generation.model ? <span>{generation.model}</span> : null}
+    <div className="rounded-lg border border-slate-200 bg-white/80 p-2 text-xs text-slate-600">
+      <div className="flex items-center gap-2">
+        {isExecuting ? (
+          <Loader2 className="h-3.5 w-3.5 animate-spin" aria-hidden="true" />
+        ) : toolCall.status === 'completed' ? (
+          <Check className="h-3.5 w-3.5" aria-hidden="true" />
+        ) : toolCall.status === 'cancelled' ? (
+          <X className="h-3.5 w-3.5" aria-hidden="true" />
+        ) : (
+          <ImageIcon className="h-3.5 w-3.5" aria-hidden="true" />
+        )}
+        <span>{editorAgentToolLabel(toolCall.toolName)}</span>
+        <span className="ml-auto text-slate-400">{statusLabel}</span>
+      </div>
+      {toolCall.error ? (
+        <div className="mt-1 text-red-600">{toolCall.error}</div>
+      ) : null}
+      {toolCall.images.length ? (
+        <div className="mt-2 grid grid-cols-3 gap-2">
+          {toolCall.images.map((image, index) => (
+            <div
+              key={`${toolCall.toolName}-${image.resourceId ?? index}`}
+              className="overflow-hidden rounded-xl border border-slate-200 bg-slate-100"
+            >
+              <ResolvedAssetImage
+                src={image.thumbnailSrc ?? image.imageSrc}
+                objectKey={image.objectKey}
+                refreshKey={image.resourceId ?? toolCall.toolName}
+                alt=""
+                className="h-20 w-full object-cover"
+              />
             </div>
-          ) : null}
-          {generation.error ? (
-            <div className="mt-1 text-red-600">{generation.error}</div>
-          ) : null}
-          {generation.images.length ? (
-            <div className="mt-2 grid grid-cols-3 gap-2">
-              {generation.images.map((image, index) => (
-                <div
-                  key={`${generation.toolCallId}-${image.resourceId ?? index}`}
-                  className="overflow-hidden rounded-xl border border-slate-200 bg-slate-100"
-                >
-                  <ResolvedAssetImage
-                    src={image.thumbnailSrc ?? image.imageSrc}
-                    objectKey={image.objectKey}
-                    refreshKey={
-                      image.resourceId ??
-                      generation.taskId ??
-                      generation.toolCallId
-                    }
-                    alt=""
-                    className="h-20 w-full object-cover"
-                  />
-                </div>
-              ))}
-            </div>
-          ) : null}
-          {generation.videos.length ? (
-            <div className="mt-2 grid grid-cols-1 gap-2">
-              {generation.videos.map((video, index) => (
-                <div
-                  key={`${generation.toolCallId}-${video.resourceId ?? index}`}
-                  className="overflow-hidden rounded-xl border border-slate-200 bg-slate-100"
-                >
-                  <ResolvedAssetVideo
-                    src={video.videoSrc}
-                    objectKey={video.objectKey}
-                    poster={video.thumbnailSrc ?? undefined}
-                    preload="metadata"
-                    // the height is decided by width (already stable) and the video aspect radio
-                    className="h-full w-full object-cover"
-                    controls
-                    playsInline
-                  />
-                </div>
-              ))}
-            </div>
-          ) : null}
-          {generation.audios.length ? (
-            <div className="mt-2 grid grid-cols-1 gap-2">
-              {generation.audios.map((audio, index) => (
-                <div
-                  key={`${generation.toolCallId}-${audio.resourceId ?? index}`}
-                  className="overflow-hidden rounded-xl border border-slate-200 bg-slate-100"
-                >
-                  <ResolvedAssetAudio
-                    audio={audio}
-                  />
-                </div>
-              ))}
-            </div>
-          ) : null}
+          ))}
         </div>
-      ))}
+      ) : null}
     </div>
   );
 }
-function ResolvedAssetAudio({
-                              audio,
-                            }: {
-  audio: EditorAgentGenerationRecord['audios'][number];
-}) {
-  const { resolvedUrl } = useResolvedAssetReadUrl(audio.audioSrc, {
-    objectKey: audio.objectKey,
-  });
+
+export function ThinkingBubble() {
   return (
-    <audio
-      controls
-      preload="metadata"
-      src={resolvedUrl == '' ? null : resolvedUrl}
-      className="w-full"
-      aria-label="生成音频预览"
-    />
+    <article className="flex justify-start" aria-label="Agent思考中">
+      <div className="max-w-[86%] rounded-3xl border border-slate-200 bg-white px-3.5 py-3 text-sm leading-6 shadow-sm">
+        <div className="flex items-center gap-1.5">
+          <span className="flex gap-0.5">
+            <span
+              className="h-1.5 w-1.5 animate-bounce rounded-full bg-slate-400"
+              style={{ animationDelay: '0ms' }}
+            />
+            <span
+              className="h-1.5 w-1.5 animate-bounce rounded-full bg-slate-400"
+              style={{ animationDelay: '150ms' }}
+            />
+            <span
+              className="h-1.5 w-1.5 animate-bounce rounded-full bg-slate-400"
+              style={{ animationDelay: '300ms' }}
+            />
+          </span>
+        </div>
+      </div>
+    </article>
   );
 }
 
-export function MessageBubble({ message }: { message: EditorAgentMessage }) {
+type MessageBubbleProps = {
+  message: EditorAgentMessage;
+  busyAction: 'confirm' | 'cancel' | null;
+  onConfirmToolCall: (messageId: number) => void;
+  onCancelToolCall: (messageId: number) => void;
+};
+
+export function MessageBubble({
+  message,
+  busyAction,
+  onConfirmToolCall,
+  onCancelToolCall,
+}: MessageBubbleProps) {
+  if (message.role === 'system' && !message.toolCall) {
+    return null;
+  }
+  if (
+    message.role === 'system' &&
+    message.toolCall?.status === 'pending_confirmation'
+  ) {
+    return (
+      <PendingToolCall
+        messageId={message.id}
+        toolCall={message.toolCall}
+        busyAction={busyAction}
+        onConfirm={onConfirmToolCall}
+        onCancel={onCancelToolCall}
+      />
+    );
+  }
+
   const isUser = message.role === 'user';
+  const isSystem = message.role === 'system';
   return (
     <article
       className={`flex ${isUser ? 'justify-end' : 'justify-start'}`}
-      aria-label={`${messageRoleLabel(message.role)}消息`}
+      aria-label={
+        isSystem ? 'Agent操作' : `${messageRoleLabel(message.role)}消息`
+      }
     >
       <div
-        className={`max-w-[86%] rounded-3xl px-3.5 py-3 text-sm leading-6 shadow-sm ${
-          isUser
-            ? 'bg-slate-900 text-white'
-            : message.kind === 'error'
-              ? 'border border-red-200 bg-red-50 text-red-700'
-              : 'border border-slate-200 bg-white text-slate-700'
-        }`}
+        className={
+          isSystem
+            ? 'max-w-[86%]'
+            : `max-w-[86%] rounded-3xl px-3.5 py-3 text-sm leading-6 shadow-sm ${
+                isUser
+                  ? 'bg-slate-900 text-white'
+                  : message.toolCall?.status === 'failed'
+                    ? 'border border-red-200 bg-red-50 text-red-700'
+                    : 'border border-slate-200 bg-white text-slate-700'
+              }`
+        }
       >
-        <div className="whitespace-pre-wrap break-words">
-          {message.text || (message.status === 'streaming' ? '...' : '')}
-        </div>
-        {message.attachments.length ? (
+        {!isSystem && message.text ? (
+          <div className="whitespace-pre-wrap break-words">{message.text}</div>
+        ) : null}
+        {!isSystem && message.attachments.length ? (
           <div className="mt-2 flex flex-wrap gap-1.5">
             {message.attachments.map((attachment) => (
               <AttachmentChip
@@ -218,7 +197,7 @@ export function MessageBubble({ message }: { message: EditorAgentMessage }) {
             ))}
           </div>
         ) : null}
-        <GenerationRecordsView generations={message.generations} />
+        {message.toolCall ? <ToolCallView toolCall={message.toolCall} /> : null}
       </div>
     </article>
   );
diff --git a/src/components/image-editor/EditorAgentConversation/PendingToolCall.tsx b/src/components/image-editor/EditorAgentConversation/PendingToolCall.tsx
new file mode 100644
index 000000000..753270118
--- /dev/null
+++ b/src/components/image-editor/EditorAgentConversation/PendingToolCall.tsx
@@ -0,0 +1,133 @@
+import { Check, Loader2, Pencil, X } from 'lucide-react';
+
+import type { EditorAgentToolCall } from '@/packages/shared/src/contracts';
+import { editorAgentToolLabel } from '@/src/components/image-editor/EditorAgentConversation/toolCallPresentation.ts';
+
+type PendingToolCallAction = 'confirm' | 'cancel' | null;
+
+type PendingToolCallProps = {
+  messageId: number;
+  toolCall: EditorAgentToolCall;
+  busyAction: PendingToolCallAction;
+  onConfirm: (messageId: number) => void;
+  onCancel: (messageId: number) => void;
+};
+
+function isRecord(value: unknown): value is Record<string, unknown> {
+  return typeof value === 'object' && value !== null;
+}
+
+function readString(value: unknown) {
+  return typeof value === 'string' && value.trim() ? value.trim() : null;
+}
+
+function readStringArray(value: unknown) {
+  return Array.isArray(value)
+    ? value.map(readString).filter((item): item is string => item !== null)
+    : [];
+}
+
+function readPendingToolDetails(toolCall: EditorAgentToolCall) {
+  const args = isRecord(toolCall.args) ? toolCall.args : {};
+  const prompt = readString(args.prompt);
+  const targetImage =
+    readString(args.object_image_id) ?? readString(args.objectImageId);
+  const referenceImages = readStringArray(
+    args.reference_image_ids ?? args.referenceImageIds,
+  );
+  const summary = readString(toolCall.summary);
+
+  return {
+    prompt: prompt ?? (summary?.startsWith('{') ? null : summary),
+    targetImage,
+    referenceImageCount: referenceImages.length,
+  };
+}
+
+export function PendingToolCall({
+  messageId,
+  toolCall,
+  busyAction,
+  onConfirm,
+  onCancel,
+}: PendingToolCallProps) {
+  const details = readPendingToolDetails(toolCall);
+  const label = editorAgentToolLabel(toolCall.toolName);
+  const isBusy = busyAction !== null;
+
+  return (
+    <article className="flex justify-start" aria-label={`待确认的${label}操作`}>
+      <div className="w-full max-w-[86%] rounded-lg border border-slate-200 bg-white p-3 text-sm text-slate-700 shadow-sm">
+        <div className="flex items-center gap-2 font-medium text-slate-900">
+          <Pencil className="h-4 w-4 shrink-0" aria-hidden="true" />
+          <span>{label}</span>
+          <span className="ml-auto text-xs font-normal text-amber-700">
+            待确认
+          </span>
+        </div>
+
+        {details.prompt ? (
+          <p className="mt-2 whitespace-pre-wrap break-words text-sm leading-5">
+            {details.prompt}
+          </p>
+        ) : null}
+
+        {details.targetImage || details.referenceImageCount > 0 ? (
+          <dl className="mt-2 space-y-1 text-xs text-slate-500">
+            {details.targetImage ? (
+              <div className="flex min-w-0 gap-2">
+                <dt className="shrink-0">目标图片</dt>
+                <dd className="min-w-0 break-all text-slate-700">
+                  {details.targetImage}
+                </dd>
+              </div>
+            ) : null}
+            {details.referenceImageCount > 0 ? (
+              <div className="flex gap-2">
+                <dt>参考图片</dt>
+                <dd className="text-slate-700">
+                  {details.referenceImageCount} 张
+                </dd>
+              </div>
+            ) : null}
+          </dl>
+        ) : null}
+
+        <div className="mt-3 flex justify-end gap-2">
+          <button
+            type="button"
+            className="inline-flex h-9 items-center justify-center gap-1.5 rounded-md border border-slate-200 bg-white px-3 text-sm text-slate-600 hover:bg-slate-50 disabled:cursor-not-allowed disabled:opacity-50"
+            disabled={isBusy}
+            onClick={() => onCancel(messageId)}
+          >
+            {busyAction === 'cancel' ? (
+              <Loader2
+                className="h-3.5 w-3.5 animate-spin"
+                aria-hidden="true"
+              />
+            ) : (
+              <X className="h-3.5 w-3.5" aria-hidden="true" />
+            )}
+            {busyAction === 'cancel' ? '取消中' : '取消'}
+          </button>
+          <button
+            type="button"
+            className="inline-flex h-9 items-center justify-center gap-1.5 rounded-md bg-slate-900 px-3 text-sm font-medium text-white hover:bg-slate-800 disabled:cursor-not-allowed disabled:opacity-50"
+            disabled={isBusy}
+            onClick={() => onConfirm(messageId)}
+          >
+            {busyAction === 'confirm' ? (
+              <Loader2
+                className="h-3.5 w-3.5 animate-spin"
+                aria-hidden="true"
+              />
+            ) : (
+              <Check className="h-3.5 w-3.5" aria-hidden="true" />
+            )}
+            {busyAction === 'confirm' ? '执行中' : '确认'}
+          </button>
+        </div>
+      </div>
+    </article>
+  );
+}
diff --git a/src/components/image-editor/EditorAgentConversation/toolCallPresentation.ts b/src/components/image-editor/EditorAgentConversation/toolCallPresentation.ts
new file mode 100644
index 000000000..434a9cba9
--- /dev/null
+++ b/src/components/image-editor/EditorAgentConversation/toolCallPresentation.ts
@@ -0,0 +1,24 @@
+export function editorAgentToolLabel(toolName: string) {
+  if (toolName === 'edit-image' || toolName === 'edit_image') {
+    return '修改图片';
+  }
+  if (toolName === 'generate_character') {
+    return '生成角色';
+  }
+  if (toolName === 'generate_icon_spritesheet') {
+    return '生成图标';
+  }
+  if (toolName === 'generate_ui_design') {
+    return '生成 UI';
+  }
+  if (toolName === 'generate_video') {
+    return '生成视频';
+  }
+  if (toolName === 'generate_sound_effect') {
+    return '生成音效';
+  }
+  if (toolName === 'generate_background_music') {
+    return '生成背景音乐';
+  }
+  return '生成图片';
+}
diff --git a/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.test.tsx b/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.test.tsx
index 69a32a8a7..d26c58b84 100644
--- a/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.test.tsx
+++ b/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.test.tsx
@@ -3,6 +3,7 @@
 import { act, renderHook, waitFor } from '@testing-library/react';
 import { beforeEach, describe, expect, it, vi } from 'vitest';
 
+import type { EditorAgentMessageResponse } from '../../../../packages/shared/src/contracts/editorAgent.ts';
 import {
   type EditorAgentConversationClient,
   useEditorAgentConversation,
@@ -14,8 +15,6 @@ function createClient(): EditorAgentConversationClient {
       {
         conversationId: 'conversation-1',
         projectId: 'project-1',
-        title: '角色参考',
-        createdAt: '2026-07-03T00:00:00.000Z',
         updatedAt: '2026-07-03T00:00:00.000Z',
       },
     ]),
@@ -43,45 +42,18 @@ function createClient(): EditorAgentConversationClient {
       createdAt: '2026-07-03T00:00:00.000Z',
       updatedAt: '2026-07-03T00:00:00.000Z',
     }),
-    streamMessage: vi
-      .fn()
-      .mockImplementation(async (_conversationId, _payload, options) => {
-        options.onEvent?.({
-          event: 'stage',
-          data: {
-            conversationId: 'conversation-1',
-            stage: 'thinking',
-          },
-        });
-        options.onEvent?.({
-          event: 'message_delta',
-          data: {
-            conversationId: 'conversation-1',
-            messageId: 'assistant-1',
-            role: 'assistant',
-            kind: 'chat',
-            textDelta: '我来处理',
-          },
-        });
-        options.onEvent?.({
-          event: 'tool_started',
-          data: {
-            conversationId: 'conversation-1',
-            messageId: 'assistant-1',
-            toolCallId: 'tool-call-1',
+    sendMessage: vi.fn().mockResolvedValue({
+      deltaMessages: [
+        {
+          id: 1,
+          role: 'assistant',
+          text: '我来处理',
+          attachments: [],
+          toolCall: {
             toolName: 'generate_image',
-            taskId: 'task-1',
-            model: 'gpt-image-2',
-          },
-        });
-        options.onEvent?.({
-          event: 'generation_result',
-          data: {
-            conversationId: 'conversation-1',
-            messageId: 'assistant-1',
-            toolCallId: 'tool-call-1',
-            toolName: 'generate_image',
-            model: null,
+            summary: '生成像素风格角色',
+            status: 'completed',
+            args: {},
             images: [
               {
                 resourceId: 'resource-result-1',
@@ -91,16 +63,15 @@ function createClient(): EditorAgentConversationClient {
                 height: 1024,
               },
             ],
+            error: null,
           },
-        });
-        options.onEvent?.({
-          event: 'done',
-          data: {
-            conversationId: 'conversation-1',
-            title: '像素角色',
-          },
-        });
-      }),
+          createdAt: '2026-07-03T00:00:00.000Z',
+        },
+      ],
+      errorMessage: null,
+    } as EditorAgentMessageResponse),
+    confirmToolCall: vi.fn(),
+    cancelToolCall: vi.fn(),
   };
 }
 
@@ -109,14 +80,14 @@ describe('useEditorAgentConversation', () => {
     vi.clearAllMocks();
   });
 
-  it('loads conversations and applies message stream events', async () => {
+  it('loads conversations and applies delta messages', async () => {
     const client = createClient();
-    const onGenerationResult = vi.fn();
+    const onCanvasRefreshRequested = vi.fn();
     const { result } = renderHook(() =>
       useEditorAgentConversation({
         projectId: 'project-1',
         client,
-        onGenerationResult,
+        onCanvasRefreshRequested,
       }),
     );
 
@@ -130,38 +101,25 @@ describe('useEditorAgentConversation', () => {
       await result.current.sendMessage('把这个角色改成像素风');
     });
 
-    expect(client.streamMessage).toHaveBeenCalledWith(
+    expect(client.sendMessage).toHaveBeenCalledWith(
       'conversation-1',
       expect.objectContaining({
         text: '把这个角色改成像素风',
         attachments: [],
       }),
       expect.objectContaining({
-        onEvent: expect.any(Function),
         signal: expect.any(AbortSignal),
       }),
     );
-    expect(result.current.stage).toBe('completed');
-    expect(result.current.isStreaming).toBe(false);
-    expect(onGenerationResult).toHaveBeenCalledWith(
-      expect.objectContaining({
-        toolCallId: 'tool-call-1',
-        images: [
-          expect.objectContaining({
-            resourceId: 'resource-result-1',
-          }),
-        ],
-      }),
-    );
+    expect(result.current.isWaiting).toBe(false);
+    expect(onCanvasRefreshRequested).toHaveBeenCalledTimes(1);
     expect(result.current.messages.map((message) => message.text)).toEqual([
       '把这个角色改成像素风',
       '我来处理',
     ]);
-    expect(result.current.messages[1]?.generations).toEqual([
+    expect(result.current.messages[1]?.toolCall).toEqual(
       expect.objectContaining({
-        toolCallId: 'tool-call-1',
-        taskId: 'task-1',
-        model: 'gpt-image-2',
+        toolName: 'generate_image',
         status: 'completed',
         images: [
           expect.objectContaining({
@@ -170,8 +128,7 @@ describe('useEditorAgentConversation', () => {
           }),
         ],
       }),
-    ]);
-    expect(result.current.conversations[0]?.title).toBe('像素角色');
+    );
   });
 
   it('creates a conversation before sending when the project has no history', async () => {
@@ -190,7 +147,7 @@ describe('useEditorAgentConversation', () => {
     });
 
     expect(client.createConversation).toHaveBeenCalledWith('project-1', {});
-    expect(client.streamMessage).toHaveBeenCalledWith(
+    expect(client.sendMessage).toHaveBeenCalledWith(
       'conversation-2',
       expect.objectContaining({ text: '新建后发送' }),
       expect.any(Object),
@@ -221,7 +178,7 @@ describe('useEditorAgentConversation', () => {
       ]);
     });
 
-    expect(client.streamMessage).toHaveBeenCalledWith(
+    expect(client.sendMessage).toHaveBeenCalledWith(
       'conversation-1',
       expect.objectContaining({
         text: '',
@@ -236,36 +193,12 @@ describe('useEditorAgentConversation', () => {
     );
   });
 
-  it('keeps LLM planning failures as one failed assistant message', async () => {
+  it('handles backend error responses', async () => {
     const client = createClient();
-    vi.mocked(client.streamMessage).mockImplementation(
-      async (_conversationId, _payload, options) => {
-        options.onEvent?.({
-          event: 'message_delta',
-          data: {
-            conversationId: 'conversation-1',
-            messageId: 'assistant-planning-error',
-            role: 'assistant',
-            kind: 'error',
-            textDelta: '画布 Agent 的 LLM 未配置，无法处理这句话。',
-          },
-        });
-        options.onEvent?.({
-          event: 'stage',
-          data: {
-            conversationId: 'conversation-1',
-            stage: 'failed',
-          },
-        });
-        options.onEvent?.({
-          event: 'done',
-          data: {
-            conversationId: 'conversation-1',
-            title: null,
-          },
-        });
-      },
-    );
+    vi.mocked(client.sendMessage).mockResolvedValue({
+      deltaMessages: [],
+      errorMessage: 'LLM 未配置，无法处理这句话。',
+    } as EditorAgentMessageResponse);
     const { result } = renderHook(() =>
       useEditorAgentConversation({ projectId: 'project-1', client }),
     );
@@ -280,71 +213,158 @@ describe('useEditorAgentConversation', () => {
       await result.current.sendMessage('这是美术素材');
     });
 
-    expect(result.current.stage).toBe('failed');
-    expect(result.current.errorMessage).toBe(
-      '画布 Agent 的 LLM 未配置，无法处理这句话。',
-    );
-    expect(
-      result.current.messages.filter((message) => message.kind === 'error'),
-    ).toEqual([
-      expect.objectContaining({
-        id: 'assistant-planning-error',
-        text: '画布 Agent 的 LLM 未配置，无法处理这句话。',
-        status: 'failed',
-      }),
-    ]);
+    expect(result.current.errorMessage).toBe('LLM 未配置，无法处理这句话。');
   });
 
-  it('shows a failed tool completion as a generation record', async () => {
+  it('confirms a pending tool call, replaces its message and requests a canvas refresh', async () => {
     const client = createClient();
-    vi.mocked(client.streamMessage).mockImplementation(
-      async (_conversationId, _payload, options) => {
-        options.onEvent?.({
-          event: 'message_delta',
-          data: {
-            conversationId: 'conversation-1',
-            messageId: 'assistant-failed',
-            role: 'assistant',
-            kind: 'chat',
-            textDelta: '我先尝试生成。',
+    vi.mocked(client.sendMessage).mockResolvedValue({
+      deltaMessages: [
+        {
+          id: 1,
+          role: 'system',
+          text: '需要生成一张图',
+          attachments: [],
+          toolCall: {
+            toolName: 'edit-image',
+            summary: '',
+            status: 'pending_confirmation',
+            args: {
+              object_image_id: 'source-image',
+              prompt: '换成像素风',
+            },
+            images: [],
+            error: null,
           },
-        });
-        options.onEvent?.({
-          event: 'tool_started',
-          data: {
-            conversationId: 'conversation-1',
-            messageId: 'assistant-failed',
-            toolCallId: 'tool-call-failed',
-            toolName: 'generate_image',
-            taskId: null,
-            model: 'gpt-image-2',
-            status: 'generating',
+          createdAt: '2026-07-03T00:00:00.000Z',
+        },
+      ],
+      errorMessage: null,
+    } as EditorAgentMessageResponse);
+    vi.mocked(client.confirmToolCall).mockResolvedValue({
+      id: 1,
+      role: 'system',
+      text: 'internal completed tool output',
+      attachments: [],
+      toolCall: {
+        toolName: 'edit-image',
+        summary: '',
+        status: 'completed',
+        args: {
+          object_image_id: 'source-image',
+          prompt: '换成像素风',
+        },
+        images: [
+          {
+            resourceId: null,
+            objectKey: 'generated/result.png',
+            imageSrc: '/result.png',
+            thumbnailSrc: null,
+            width: 512,
+            height: 512,
           },
-        });
-        options.onEvent?.({
-          event: 'tool_completed',
-          data: {
-            conversationId: 'conversation-1',
-            messageId: 'assistant-failed',
-            toolCallId: 'tool-call-failed',
-            toolName: 'generate_image',
-            taskId: null,
-            model: 'gpt-image-2',
-            status: 'failed',
-            error: '泥点余额不足',
-          },
-        });
-        options.onEvent?.({
-          event: 'error',
-          data: {
-            conversationId: 'conversation-1',
-            code: 'INSUFFICIENT_BALANCE',
-            message: '泥点余额不足',
-            recoverable: true,
-          },
-        });
+        ],
+        error: null,
       },
+      createdAt: '2026-07-03T00:00:00.000Z',
+    });
+    const onCanvasRefreshRequested = vi.fn();
+    const { result } = renderHook(() =>
+      useEditorAgentConversation({
+        projectId: 'project-1',
+        client,
+        onCanvasRefreshRequested,
+      }),
     );
+
+    await waitFor(() => {
+      expect(result.current.activeConversation?.conversationId).toBe(
+        'conversation-1',
+      );
+    });
+
+    await act(async () => {
+      await result.current.sendMessage('生成一张图');
+    });
+
+    const systemMsg = result.current.messages[1];
+    expect(systemMsg?.role).toBe('system');
+    expect(systemMsg?.toolCall?.status).toBe('pending_confirmation');
+
+    const messageCount = result.current.messages.length;
+    await act(async () => {
+      await result.current.confirmToolCall(1);
+    });
+
+    expect(client.confirmToolCall).toHaveBeenCalledWith('conversation-1', 1);
+    expect(result.current.messages).toHaveLength(messageCount);
+    expect(result.current.messages[1]?.toolCall?.status).toBe('completed');
+    expect(result.current.messages[1]?.toolCall?.images[0]?.objectKey).toBe(
+      'generated/result.png',
+    );
+    expect(onCanvasRefreshRequested).toHaveBeenCalledTimes(1);
+  });
+
+  it('cancels a pending tool call and keeps the replacement in the same position', async () => {
+    const client = createClient();
+    const pendingMessage = {
+      id: 0,
+      role: 'system' as const,
+      text: 'internal pending tool prompt',
+      attachments: [],
+      toolCall: {
+        toolName: 'edit-image',
+        summary: '',
+        status: 'pending_confirmation' as const,
+        args: { object_image_id: 'source-image', prompt: '换成像素风' },
+        images: [],
+        error: null,
+      },
+      createdAt: '2026-07-03T00:00:00.000Z',
+    };
+    vi.mocked(client.getConversation).mockResolvedValue({
+      conversationId: 'conversation-1',
+      projectId: 'project-1',
+      title: '角色参考',
+      messages: [pendingMessage],
+      createdAt: '2026-07-03T00:00:00.000Z',
+      updatedAt: '2026-07-03T00:00:00.000Z',
+    });
+    vi.mocked(client.cancelToolCall).mockResolvedValue({
+      ...pendingMessage,
+      text: 'internal cancelled tool output',
+      toolCall: {
+        ...pendingMessage.toolCall,
+        status: 'cancelled',
+      },
+    });
+    const onCanvasRefreshRequested = vi.fn();
+    const { result } = renderHook(() =>
+      useEditorAgentConversation({
+        projectId: 'project-1',
+        client,
+        onCanvasRefreshRequested,
+      }),
+    );
+
+    await waitFor(() => {
+      expect(result.current.messages[0]?.toolCall?.status).toBe(
+        'pending_confirmation',
+      );
+    });
+    await act(async () => {
+      await result.current.cancelToolCall(0);
+    });
+
+    expect(client.cancelToolCall).toHaveBeenCalledWith('conversation-1', 0);
+    expect(result.current.messages).toHaveLength(1);
+    expect(result.current.messages[0]?.toolCall?.status).toBe('cancelled');
+    expect(onCanvasRefreshRequested).not.toHaveBeenCalled();
+  });
+
+  it('handles fetch error as failed stage', async () => {
+    const client = createClient();
+    vi.mocked(client.sendMessage).mockRejectedValue(new Error('Network error'));
     const { result } = renderHook(() =>
       useEditorAgentConversation({ projectId: 'project-1', client }),
     );
@@ -356,45 +376,21 @@ describe('useEditorAgentConversation', () => {
     });
 
     await act(async () => {
-      await expect(
-        result.current.sendMessage('生成一张图'),
-      ).resolves.toBeUndefined();
+      await expect(result.current.sendMessage('test')).resolves.toBeUndefined();
     });
 
-    const failedMessage = result.current.messages.find(
-      (message) => message.id === 'assistant-failed',
-    );
-    expect(failedMessage?.status).toBe('failed');
-    expect(failedMessage?.generations).toEqual([
-      expect.objectContaining({
-        toolCallId: 'tool-call-failed',
-        status: 'failed',
-        model: 'gpt-image-2',
-        error: '泥点余额不足',
-      }),
-    ]);
-    expect(result.current.errorMessage).toBe('泥点余额不足');
+    expect(result.current.errorMessage).toBe('Network error');
   });
 
-  it('aborts the active stream and marks streaming messages as stopped', async () => {
+  it('aborts the active request and resets state on stopCurrentTurn', async () => {
     const client = createClient();
-    let streamSignal: AbortSignal | null = null;
-    vi.mocked(client.streamMessage).mockImplementation(
+    let capturedSignal: AbortSignal | null = null;
+    vi.mocked(client.sendMessage).mockImplementation(
       (_conversationId, _payload, options) =>
-        new Promise<void>((resolve) => {
-          streamSignal = options.signal ?? null;
-          options.onEvent?.({
-            event: 'message_delta',
-            data: {
-              conversationId: 'conversation-1',
-              messageId: 'assistant-streaming',
-              role: 'assistant',
-              kind: 'chat',
-              textDelta: '处理中',
-            },
-          });
-          streamSignal?.addEventListener('abort', () => resolve(), {
-            once: true,
+        new Promise<EditorAgentMessageResponse>((resolve, reject) => {
+          capturedSignal = options.signal ?? null;
+          capturedSignal?.addEventListener('abort', () => {
+            reject(new DOMException('Aborted', 'AbortError'));
           });
         }),
     );
@@ -412,7 +408,7 @@ describe('useEditorAgentConversation', () => {
       void result.current.sendMessage('请继续');
     });
     await waitFor(() => {
-      expect(result.current.isStreaming).toBe(true);
+      expect(result.current.isWaiting).toBe(true);
     });
 
     act(() => {
@@ -420,9 +416,8 @@ describe('useEditorAgentConversation', () => {
     });
 
     await waitFor(() => {
-      expect(streamSignal?.aborted).toBe(true);
+      expect(capturedSignal?.aborted).toBe(true);
     });
-    expect(result.current.messages.at(-1)?.status).toBe('stopped');
-    expect(result.current.isStreaming).toBe(false);
+    expect(result.current.isWaiting).toBe(false);
   });
 });
diff --git a/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.ts b/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.ts
index 7cc68bcb3..1d85a0f2a 100644
--- a/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.ts
+++ b/src/components/image-editor/EditorAgentConversation/useEditorAgentConversation.ts
@@ -5,20 +5,19 @@ import type {
   EditorAgentAttachmentRef,
   EditorAgentConversationDetail,
   EditorAgentConversationSummary,
-  EditorAgentGenerationResultEvent,
-  EditorAgentGenerationRecord,
   EditorAgentMessage,
-  EditorAgentSseEvent,
-  EditorAgentStage,
-  StreamEditorAgentMessageRequest,
+  EditorAgentMessageRequest,
+  EditorAgentMessageResponse,
 } from '../../../../packages/shared/src/contracts/editorAgent.ts';
 import {
+  cancelEditorAgentToolCall,
+  confirmEditorAgentToolCall,
   createEditorAgentConversation,
   deleteEditorAgentConversation,
   getEditorAgentConversation,
   listEditorAgentConversations,
-  streamEditorAgentMessage,
-  type StreamEditorAgentMessageOptions,
+  sendEditorAgentMessage,
+  type SendEditorAgentMessageOptions,
 } from '../../../services/image-editor/editorAgentClient.ts';
 
 export type EditorAgentConversationClient = {
@@ -35,39 +34,44 @@ export type EditorAgentConversationClient = {
   deleteConversation: (
     conversationId: string,
   ) => Promise<EditorAgentConversationSummary>;
-  streamMessage: (
+  sendMessage: (
     conversationId: string,
-    payload: StreamEditorAgentMessageRequest,
-    options: StreamEditorAgentMessageOptions,
-  ) => Promise<void>;
+    payload: EditorAgentMessageRequest,
+    options: SendEditorAgentMessageOptions,
+  ) => Promise<EditorAgentMessageResponse>;
+  confirmToolCall: (
+    conversationId: string,
+    messageId: number,
+  ) => Promise<EditorAgentMessage>;
+  cancelToolCall: (
+    conversationId: string,
+    messageId: number,
+  ) => Promise<EditorAgentMessage>;
 };
 
 type UseEditorAgentConversationOptions = {
   projectId?: string | null;
   client?: EditorAgentConversationClient;
-  onGenerationResult?: (event: EditorAgentGenerationResultEvent) => void;
+  onCanvasRefreshRequested?: () => void;
 };
 
+export type EditorAgentToolCallAction = 'confirm' | 'cancel';
+
+export type EditorAgentToolCallActionState = {
+  messageId: number;
+  action: EditorAgentToolCallAction;
+} | null;
+
 const defaultEditorAgentConversationClient: EditorAgentConversationClient = {
   listConversations: listEditorAgentConversations,
   createConversation: createEditorAgentConversation,
   getConversation: getEditorAgentConversation,
   deleteConversation: deleteEditorAgentConversation,
-  streamMessage: streamEditorAgentMessage,
+  sendMessage: sendEditorAgentMessage,
+  confirmToolCall: confirmEditorAgentToolCall,
+  cancelToolCall: cancelEditorAgentToolCall,
 };
 
-function createClientId(prefix: string) {
-  if (
-    typeof crypto !== 'undefined' &&
-    typeof crypto.randomUUID === 'function'
-  ) {
-    return `${prefix}-${crypto.randomUUID()}`;
-  }
-  return `${prefix}-${Date.now().toString(36)}-${Math.random()
-    .toString(36)
-    .slice(2)}`;
-}
-
 function isAbortError(error: unknown) {
   return (
     error instanceof Error &&
@@ -79,97 +83,20 @@ function isAbortError(error: unknown) {
 }
 
 function createLocalUserMessage(params: {
-  id: string;
+  id: number;
   text: string;
   attachments: EditorAgentAttachmentRef[];
 }): EditorAgentMessage {
   return {
     id: params.id,
     role: 'user',
-    kind: 'chat',
     text: params.text,
     attachments: params.attachments,
-    generations: [],
-    status: 'completed',
+    toolCall: null,
     createdAt: new Date().toISOString(),
   };
 }
 
-function createAssistantMessage(params: {
-  id: string;
-  role?: EditorAgentMessage['role'];
-  kind?: EditorAgentMessage['kind'];
-  text?: string;
-  status?: EditorAgentMessage['status'];
-  generations?: EditorAgentGenerationRecord[];
-}): EditorAgentMessage {
-  return {
-    id: params.id,
-    role: params.role ?? 'assistant',
-    kind: params.kind ?? 'chat',
-    text: params.text ?? '',
-    attachments: [],
-    generations: params.generations ?? [],
-    status: params.status ?? 'streaming',
-    createdAt: new Date().toISOString(),
-  };
-}
-
-function upsertMessage(
-  messages: EditorAgentMessage[],
-  messageId: string,
-  createMessage: () => EditorAgentMessage,
-  updateMessage: (message: EditorAgentMessage) => EditorAgentMessage,
-) {
-  let updated = false;
-  const nextMessages = messages.map((message) => {
-    if (message.id !== messageId) {
-      return message;
-    }
-    updated = true;
-    return updateMessage(message);
-  });
-  if (updated) {
-    return nextMessages;
-  }
-  return [...messages, updateMessage(createMessage())];
-}
-
-function upsertGenerationRecord(
-  records: EditorAgentGenerationRecord[],
-  nextRecord: EditorAgentGenerationRecord,
-) {
-  const existingIndex = records.findIndex(
-    (record) => record.toolCallId === nextRecord.toolCallId,
-  );
-  if (existingIndex === -1) {
-    return [...records, nextRecord];
-  }
-  return records.map((record, index) =>
-    index === existingIndex
-      ? {
-          ...record,
-          ...nextRecord,
-          summary: nextRecord.summary ?? record.summary,
-          taskId: nextRecord.taskId ?? record.taskId,
-          model: nextRecord.model ?? record.model,
-          images: nextRecord.images.length ? nextRecord.images : record.images,
-        }
-      : record,
-  );
-}
-
-function markStreamingMessages(
-  messages: EditorAgentMessage[],
-  status: EditorAgentMessage['status'],
-) {
-  return messages.map((message) =>
-    message.status === 'streaming' || message.status === 'generating'
-      ? { ...message, status }
-      : message,
-  );
-}
-
 function sortConversationsByUpdatedAt(
   conversations: EditorAgentConversationSummary[],
 ) {
@@ -186,8 +113,6 @@ function summaryFromDetail(
   return {
     conversationId: detail.conversationId,
     projectId: detail.projectId,
-    title: detail.title,
-    createdAt: detail.createdAt,
     updatedAt: detail.updatedAt,
   };
 }
@@ -211,7 +136,7 @@ function upsertConversationSummary(
 export function useEditorAgentConversation({
   projectId,
   client = defaultEditorAgentConversationClient,
-  onGenerationResult,
+  onCanvasRefreshRequested,
 }: UseEditorAgentConversationOptions) {
   const normalizedProjectId = projectId?.trim() ?? '';
   const [conversations, setConversations] = useState<
@@ -221,15 +146,18 @@ export function useEditorAgentConversation({
     string | null
   >(null);
   const [messages, setMessages] = useState<EditorAgentMessage[]>([]);
-  const [stage, setStage] = useState<EditorAgentStage>('idle');
   const [isLoadingConversations, setIsLoadingConversations] = useState(false);
   const [isLoadingMessages, setIsLoadingMessages] = useState(false);
   const [isCreatingConversation, setIsCreatingConversation] = useState(false);
   const [isDeletingConversation, setIsDeletingConversation] = useState(false);
-  const [isStreaming, setIsStreaming] = useState(false);
+  const [isWaiting, setIsWaiting] = useState(false);
+  const [toolCallAction, setToolCallAction] =
+    useState<EditorAgentToolCallActionState>(null);
   const [errorMessage, setErrorMessage] = useState<string | null>(null);
-  const activeStreamAbortControllerRef = useRef<AbortController | null>(null);
+  const activeRequestAbortControllerRef = useRef<AbortController | null>(null);
   const activeConversationIdRef = useRef<string | null>(null);
+  const activeToolCallActionRef = useRef<EditorAgentToolCallActionState>(null);
+  const nextLocalMessageIdRef = useRef(-1);
 
   useEffect(() => {
     activeConversationIdRef.current = activeConversationId;
@@ -256,7 +184,6 @@ export function useEditorAgentConversation({
       );
       setActiveConversationId(detail.conversationId);
       setMessages(detail.messages);
-      setStage('idle');
     },
     [],
   );
@@ -286,7 +213,6 @@ export function useEditorAgentConversation({
       setConversations([]);
       setActiveConversationId(null);
       setMessages([]);
-      setStage('idle');
       setErrorMessage(null);
       return undefined;
     }
@@ -365,156 +291,28 @@ export function useEditorAgentConversation({
     [activeConversationId, loadConversation],
   );
 
-  const handleSseEvent = useCallback((event: EditorAgentSseEvent) => {
-    const eventConversationId = event.data.conversationId;
-    if (
-      eventConversationId &&
-      eventConversationId !== activeConversationIdRef.current
-    ) {
-      return;
-    }
-
-    if (event.event === 'stage') {
-      setStage(event.data.stage);
-      return;
-    }
-
-    if (event.event === 'message_delta') {
-      if (event.data.kind === 'error') {
-        const nextErrorMessage = event.data.textDelta.trim();
-        if (nextErrorMessage) {
-          setErrorMessage(nextErrorMessage);
-        }
+  const requestCanvasRefreshForMessages = useCallback(
+    (nextMessages: EditorAgentMessage[]) => {
+      if (
+        nextMessages.some(
+          (message) =>
+            message.toolCall?.status === 'completed' &&
+            message.toolCall.images.length > 0,
+        )
+      ) {
+        onCanvasRefreshRequested?.();
       }
-      setMessages((currentMessages) =>
-        upsertMessage(
-          currentMessages,
-          event.data.messageId,
-          () =>
-            createAssistantMessage({
-              id: event.data.messageId,
-              role: event.data.role,
-              kind: event.data.kind,
-            }),
-          (message) => ({
-            ...message,
-            role: event.data.role,
-            kind: event.data.kind,
-            text: `${message.text}${event.data.textDelta}`,
-            status: event.data.kind === 'error' ? 'failed' : 'streaming',
-          }),
-        ),
-      );
-      return;
-    }
+    },
+    [onCanvasRefreshRequested],
+  );
 
-    if (event.event === 'tool_started' || event.event === 'tool_completed') {
-      const status =
-        event.event === 'tool_started'
-          ? 'generating'
-          : (event.data.status ?? 'completed');
-      const nextRecord: EditorAgentGenerationRecord = {
-        toolCallId: event.data.toolCallId,
-        toolName: event.data.toolName,
-        summary: event.data.summary ?? null,
-        taskId: event.data.taskId ?? null,
-        status,
-        model: event.data.model ?? null,
-        images: [],
-        error: event.data.error ?? null,
-      };
-      setMessages((currentMessages) =>
-        upsertMessage(
-          currentMessages,
-          event.data.messageId,
-          () =>
-            createAssistantMessage({
-              id: event.data.messageId,
-              status,
-              generations: [nextRecord],
-            }),
-          (message) => ({
-            ...message,
-            status,
-            generations: upsertGenerationRecord(
-              message.generations,
-              nextRecord,
-            ),
-          }),
-        ),
-      );
-      return;
-    }
-
-    if (event.event === 'generation_result') {
-      onGenerationResult?.(event.data);
-      const nextRecord: EditorAgentGenerationRecord = {
-        toolCallId: event.data.toolCallId,
-        toolName: event.data.toolName,
-        summary: null,
-        taskId: null,
-        status: 'completed',
-        model: event.data.model,
-        images: event.data.images,
-      };
-      setMessages((currentMessages) =>
-        upsertMessage(
-          currentMessages,
-          event.data.messageId,
-          () =>
-            createAssistantMessage({
-              id: event.data.messageId,
-              status: 'completed',
-              generations: [nextRecord],
-            }),
-          (message) => ({
-            ...message,
-            status: 'completed',
-            generations: upsertGenerationRecord(
-              message.generations,
-              nextRecord,
-            ),
-          }),
-        ),
-      );
-      return;
-    }
-
-    if (event.event === 'error') {
-      const messageId = createClientId('agent-error');
-      setStage('failed');
-      setErrorMessage(event.data.message);
-      setMessages((currentMessages) => [
-        ...currentMessages,
-        createAssistantMessage({
-          id: messageId,
-          kind: 'error',
-          text: event.data.message,
-          status: 'failed',
-        }),
-      ]);
-      return;
-    }
-
-    if (event.event === 'done') {
-      setStage((currentStage) =>
-        currentStage === 'failed' ? 'failed' : 'completed',
-      );
-      setMessages((currentMessages) =>
-        markStreamingMessages(currentMessages, 'completed'),
-      );
-      const nextTitle = event.data.title?.trim();
-      if (nextTitle) {
-        setConversations((currentConversations) =>
-          currentConversations.map((conversation) =>
-            conversation.conversationId === event.data.conversationId
-              ? { ...conversation, title: nextTitle }
-              : conversation,
-          ),
-        );
-      }
-    }
-  }, [onGenerationResult]);
+  const applyDeltaMessages = useCallback(
+    (deltaMessages: EditorAgentMessage[]) => {
+      setMessages((currentMessages) => [...currentMessages, ...deltaMessages]);
+      requestCanvasRefreshForMessages(deltaMessages);
+    },
+    [requestCanvasRefreshForMessages],
+  );
 
   const ensureConversationForSend = useCallback(async () => {
     if (activeConversationId) {
@@ -529,44 +327,46 @@ export function useEditorAgentConversation({
       const text = rawText.trim();
       if (
         (!text && !attachments.length) ||
-        isStreaming ||
+        isWaiting ||
+        activeToolCallActionRef.current !== null ||
         isLoadingConversations ||
         isLoadingMessages
       ) {
         return;
       }
       const conversationId = await ensureConversationForSend();
-      const clientMessageId = createClientId('client-message');
       const abortController = new AbortController();
-      activeStreamAbortControllerRef.current = abortController;
+      activeRequestAbortControllerRef.current = abortController;
       setErrorMessage(null);
-      setStage('thinking');
-      setIsStreaming(true);
+      setIsWaiting(true);
+      const localMessageId = nextLocalMessageIdRef.current;
+      nextLocalMessageIdRef.current -= 1;
       setMessages((currentMessages) => [
         ...currentMessages,
         createLocalUserMessage({
-          id: clientMessageId,
+          id: localMessageId,
           text,
           attachments,
         }),
       ]);
 
       try {
-        await client.streamMessage(
+        const response = await client.sendMessage(
           conversationId,
           {
-            clientMessageId,
             text,
             attachments,
           },
           {
             signal: abortController.signal,
-            onEvent: handleSseEvent,
           },
         );
-        setMessages((currentMessages) =>
-          markStreamingMessages(currentMessages, 'completed'),
-        );
+
+        if (response.errorMessage) {
+          setErrorMessage(response.errorMessage);
+        } else {
+          applyDeltaMessages(response.deltaMessages);
+        }
       } catch (error) {
         if (isAbortError(error) || abortController.signal.aborted) {
           return;
@@ -574,44 +374,90 @@ export function useEditorAgentConversation({
         const message =
           error instanceof Error ? error.message : '发送画布 Agent 消息失败';
         setErrorMessage(message);
-        setStage('failed');
-        setMessages((currentMessages) => [
-          ...markStreamingMessages(currentMessages, 'failed'),
-          createAssistantMessage({
-            id: createClientId('agent-error'),
-            kind: 'error',
-            text: message,
-            status: 'failed',
-          }),
-        ]);
-        throw error;
       } finally {
-        if (activeStreamAbortControllerRef.current === abortController) {
-          activeStreamAbortControllerRef.current = null;
+        if (activeRequestAbortControllerRef.current === abortController) {
+          activeRequestAbortControllerRef.current = null;
         }
-        setIsStreaming(false);
+        setIsWaiting(false);
       }
     },
     [
       client,
       ensureConversationForSend,
-      handleSseEvent,
+      applyDeltaMessages,
       isLoadingConversations,
       isLoadingMessages,
-      isStreaming,
+      isWaiting,
     ],
   );
 
   const stopCurrentTurn = useCallback(() => {
-    activeStreamAbortControllerRef.current?.abort();
-    activeStreamAbortControllerRef.current = null;
-    setMessages((currentMessages) =>
-      markStreamingMessages(currentMessages, 'stopped'),
-    );
-    setStage('idle');
-    setIsStreaming(false);
+    activeRequestAbortControllerRef.current?.abort();
+    activeRequestAbortControllerRef.current = null;
+    setIsWaiting(false);
   }, []);
 
+  const resolveToolCall = useCallback(
+    async (messageId: number, action: EditorAgentToolCallAction) => {
+      const conversationId = activeConversationIdRef.current;
+      if (!conversationId || activeToolCallActionRef.current) {
+        return null;
+      }
+
+      const nextAction = { messageId, action } as const;
+      activeToolCallActionRef.current = nextAction;
+      setToolCallAction(nextAction);
+      setErrorMessage(null);
+
+      try {
+        const updatedMessage = await (action === 'confirm'
+          ? client.confirmToolCall(conversationId, messageId)
+          : client.cancelToolCall(conversationId, messageId));
+
+        if (activeConversationIdRef.current !== conversationId) {
+          return updatedMessage;
+        }
+
+        setMessages((currentMessages) =>
+          currentMessages.map((message) =>
+            message.id === updatedMessage.id ? updatedMessage : message,
+          ),
+        );
+        if (action === 'confirm') {
+          requestCanvasRefreshForMessages([updatedMessage]);
+        }
+        return updatedMessage;
+      } catch (error) {
+        if (activeConversationIdRef.current === conversationId) {
+          setErrorMessage(
+            error instanceof Error
+              ? error.message
+              : action === 'confirm'
+                ? '确认画布 Agent 操作失败'
+                : '取消画布 Agent 操作失败',
+          );
+        }
+        return null;
+      } finally {
+        if (activeToolCallActionRef.current === nextAction) {
+          activeToolCallActionRef.current = null;
+          setToolCallAction(null);
+        }
+      }
+    },
+    [client, requestCanvasRefreshForMessages],
+  );
+
+  const confirmToolCall = useCallback(
+    (messageId: number) => resolveToolCall(messageId, 'confirm'),
+    [resolveToolCall],
+  );
+
+  const cancelToolCall = useCallback(
+    (messageId: number) => resolveToolCall(messageId, 'cancel'),
+    [resolveToolCall],
+  );
+
   const deleteActiveConversation = useCallback(async () => {
     if (!activeConversationId) {
       return;
@@ -633,7 +479,6 @@ export function useEditorAgentConversation({
       }
       setActiveConversationId(null);
       setMessages([]);
-      setStage('idle');
     } catch (error) {
       setErrorMessage(
         error instanceof Error ? error.message : '删除画布 Agent 会话失败',
@@ -649,17 +494,20 @@ export function useEditorAgentConversation({
     activeConversation,
     activeConversationId,
     messages,
-    stage,
     isLoadingConversations,
     isLoadingMessages,
     isCreatingConversation,
     isDeletingConversation,
-    isStreaming,
+    isWaiting,
+    toolCallAction,
+    isToolCallActionPending: toolCallAction !== null,
     errorMessage,
     createConversation,
     selectConversation,
     sendMessage,
     stopCurrentTurn,
+    confirmToolCall,
+    cancelToolCall,
     deleteActiveConversation,
   };
 }