Fix image generate (#1068)

2026-05-26 18:24:31 +08:00
parent 39c8400421
commit be7f8fc768
32 changed files with 1534 additions and 165 deletions
--- a/electron/utils/openclaw-auth.ts
+++ b/electron/utils/openclaw-auth.ts
@@ -1773,6 +1773,7 @@ export async function syncOpenAiCompatibleImageRelay(params: {
      }
      removePluginRegistrations(config, [CLAWX_OPENAI_IMAGE_PROVIDER_KEY]);
      await writeOpenClawJson(config);
+      await removeProviderKeyFromOpenClaw(CLAWX_OPENAI_IMAGE_PROVIDER_KEY);
      if (params.apiKey?.trim()) {
        await saveProviderKeyToOpenClaw(CLAWX_OPENAI_IMAGE_PROVIDER_KEY, params.apiKey.trim());
      }
--- a/electron/utils/openclaw-image-generation-runtime.ts
+++ b/electron/utils/openclaw-image-generation-runtime.ts
@@ -47,25 +47,30 @@ type ModelInputModule = {
  resolveAgentModelPrimaryValue: (model?: unknown) => string | undefined;
 };

+const OPENCLAW_IMAGE_GENERATION_RUNTIME = 'openclaw/plugin-sdk/image-generation-runtime';
+const OPENCLAW_MEDIA_STORE = 'openclaw/plugin-sdk/media-store';
+const OPENCLAW_MEDIA_RUNTIME = 'openclaw/plugin-sdk/media-runtime';
+const OPENCLAW_IMAGE_GENERATION_CORE = 'openclaw/plugin-sdk/image-generation-core';
+
 let imageRuntimeModule: ImageGenerationRuntimeModule | null = null;
 let mediaStoreModule: MediaStoreModule | null = null;
 let imageOpsModule: ImageOpsModule | null = null;
 let modelInputModule: ModelInputModule | null = null;

-async function importRuntimeModule<T>(relativePath: string): Promise<T> {
-  const modulePath = resolveOpenClawRuntimeModulePath(relativePath);
+async function importOpenClawSdkModule<T>(specifier: string): Promise<T> {
+  const modulePath = resolveOpenClawRuntimeModulePath(specifier);
  return import(pathToFileURL(modulePath).href) as Promise<T>;
 }

 async function getImageGenerationRuntime(): Promise<ImageGenerationRuntimeModule> {
  if (!imageRuntimeModule) {
-    const mod = await importRuntimeModule<{
-      t: ImageGenerationRuntimeModule['generateImage'];
-      n: ImageGenerationRuntimeModule['listRuntimeImageGenerationProviders'];
-    }>('./dist/runtime-Oyct10ZH.js');
+    const mod = await importOpenClawSdkModule<{
+      generateImage: ImageGenerationRuntimeModule['generateImage'];
+      listRuntimeImageGenerationProviders: ImageGenerationRuntimeModule['listRuntimeImageGenerationProviders'];
+    }>(OPENCLAW_IMAGE_GENERATION_RUNTIME);
    imageRuntimeModule = {
-      generateImage: mod.t,
-      listRuntimeImageGenerationProviders: mod.n,
+      generateImage: mod.generateImage,
+      listRuntimeImageGenerationProviders: mod.listRuntimeImageGenerationProviders,
    };
  }
  return imageRuntimeModule;
@@ -73,20 +78,20 @@ async function getImageGenerationRuntime(): Promise<ImageGenerationRuntimeModule

 async function getMediaStore(): Promise<MediaStoreModule> {
  if (!mediaStoreModule) {
-    const mod = await importRuntimeModule<{ u: MediaStoreModule['saveMediaBuffer'] }>(
-      './dist/store-b792nN7l.js',
+    const mod = await importOpenClawSdkModule<{ saveMediaBuffer: MediaStoreModule['saveMediaBuffer'] }>(
+      OPENCLAW_MEDIA_STORE,
    );
-    mediaStoreModule = { saveMediaBuffer: mod.u };
+    mediaStoreModule = { saveMediaBuffer: mod.saveMediaBuffer };
  }
  return mediaStoreModule;
 }

 async function getImageOps(): Promise<ImageOpsModule> {
  if (!imageOpsModule) {
-    const mod = await importRuntimeModule<{ a: ImageOpsModule['getImageMetadata'] }>(
-      './dist/image-ops-CL9ZQP7k.js',
+    const mod = await importOpenClawSdkModule<{ getImageMetadata: ImageOpsModule['getImageMetadata'] }>(
+      OPENCLAW_MEDIA_RUNTIME,
    );
-    imageOpsModule = { getImageMetadata: mod.a };
+    imageOpsModule = { getImageMetadata: mod.getImageMetadata };
  }
  return imageOpsModule;
 }
@@ -100,10 +105,10 @@ export async function resolveImageGenerationPrimaryFromConfig(

 async function getModelInputHelpers(): Promise<ModelInputModule> {
  if (!modelInputModule) {
-    const mod = await importRuntimeModule<{ i: ModelInputModule['resolveAgentModelPrimaryValue'] }>(
-      './dist/model-input-B9p-bobB.js',
+    const mod = await importOpenClawSdkModule<{ resolveAgentModelPrimaryValue: ModelInputModule['resolveAgentModelPrimaryValue'] }>(
+      OPENCLAW_IMAGE_GENERATION_CORE,
    );
-    modelInputModule = { resolveAgentModelPrimaryValue: mod.i };
+    modelInputModule = { resolveAgentModelPrimaryValue: mod.resolveAgentModelPrimaryValue };
  }
  return modelInputModule;
 }
--- a/src/components/settings/ImageGenerationSettings.tsx
+++ b/src/components/settings/ImageGenerationSettings.tsx
@@ -2,14 +2,16 @@
 * Global image generation settings (agents.defaults.imageGenerationModel).
 */
 import { useCallback, useEffect, useMemo, useState } from 'react';
-import { Eye, EyeOff, ImagePlus, Loader2, Play, RefreshCw } from 'lucide-react';
+import { Eye, EyeOff, ImagePlus, Loader2, Play, RefreshCw, Trash2 } from 'lucide-react';
 import { useTranslation } from 'react-i18next';
 import { toast } from 'sonner';
 import { Button } from '@/components/ui/button';
 import { Input } from '@/components/ui/input';
 import { Label } from '@/components/ui/label';
 import { Badge } from '@/components/ui/badge';
+import { ConfirmDialog } from '@/components/ui/confirm-dialog';
 import {
+  clearImageGenerationSettings,
  fetchImageGenerationSettings,
  runImageGenerationTest,
  saveImageGenerationSettings,
@@ -32,10 +34,12 @@ function extractTestOutputPath(result: unknown): string | null {
 }

 export function ImageGenerationSettings() {
-  const { t } = useTranslation(['dashboard', 'settings']);
+  const { t } = useTranslation(['dashboard', 'settings', 'common']);
  const [loading, setLoading] = useState(true);
  const [saving, setSaving] = useState(false);
+  const [clearing, setClearing] = useState(false);
  const [testing, setTesting] = useState(false);
+  const [clearConfirmOpen, setClearConfirmOpen] = useState(false);
  const [snapshot, setSnapshot] = useState<ImageGenerationSettingsSnapshot | null>(null);

  const [timeoutMs, setTimeoutMs] = useState('180000');
@@ -79,6 +83,15 @@ export function ImageGenerationSettings() {
    );
  }, [snapshot, timeoutMs, relayBaseUrl, relayModel, relayApiKey]);

+  const hasConfiguredRelay = useMemo(() => {
+    if (!snapshot) return false;
+    return Boolean(
+      snapshot.openAiRelay?.enabled
+      || snapshot.openAiRelay?.baseUrl?.trim()
+      || snapshot.config.primary?.trim(),
+    );
+  }, [snapshot]);
+
  const handleSave = async () => {
    setSaving(true);
    try {
@@ -120,11 +133,32 @@ export function ImageGenerationSettings() {
    }
  };

+  const handleClear = async () => {
+    setClearing(true);
+    try {
+      const next = await clearImageGenerationSettings();
+      setSnapshot(next);
+      setRelayBaseUrl('');
+      setRelayModel('gpt-image-2');
+      setRelayApiKey('');
+      setShowRelayApiKey(false);
+      setClearConfirmOpen(false);
+      toast.success(t('imageGeneration.toast.cleared'));
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : String(error));
+    } finally {
+      setClearing(false);
+    }
+  };
+
  const handleTest = async () => {
    if (dirty) {
      toast.message(t('imageGeneration.toast.saveBeforeTest'));
      return;
    }
+    if (!hasConfiguredRelay) {
+      return;
+    }
    setTesting(true);
    try {
      const result = await runImageGenerationTest({
@@ -368,7 +402,7 @@ export function ImageGenerationSettings() {
              variant="outline"
              className="rounded-full h-10"
              onClick={() => void handleTest()}
-              disabled={testing || !relayModel.trim()}
+              disabled={testing || !hasConfiguredRelay || dirty}
              data-testid="image-generation-test-button"
            >
              {testing ? (
@@ -387,9 +421,34 @@ export function ImageGenerationSettings() {
              {saving ? <Loader2 className="h-4 w-4 mr-2 animate-spin" /> : null}
              {saving ? t('imageGeneration.saving') : t('imageGeneration.save')}
            </Button>
+            <Button
+              variant="outline"
+              className="rounded-full h-10 text-destructive hover:text-destructive"
+              onClick={() => setClearConfirmOpen(true)}
+              disabled={clearing || !hasConfiguredRelay}
+              data-testid="image-generation-clear"
+            >
+              {clearing ? (
+                <Loader2 className="h-4 w-4 mr-2 animate-spin" />
+              ) : (
+                <Trash2 className="h-4 w-4 mr-2" />
+              )}
+              {clearing ? t('imageGeneration.clearing') : t('imageGeneration.clear')}
+            </Button>
          </div>
        </div>
      )}
+
+      <ConfirmDialog
+        open={clearConfirmOpen}
+        title={t('imageGeneration.clearConfirmTitle')}
+        message={t('imageGeneration.clearConfirmMessage')}
+        confirmLabel={t('imageGeneration.clearConfirmAction')}
+        cancelLabel={t('common:actions.cancel')}
+        variant="destructive"
+        onConfirm={handleClear}
+        onCancel={() => setClearConfirmOpen(false)}
+      />
    </div>
  );
 }
--- a/src/i18n/locales/en/chat.json
+++ b/src/i18n/locales/en/chat.json
@@ -159,10 +159,14 @@
        "branchLabel": "branch",
        "thinkingLabel": "Thinking",
        "errorLabel": "Error",
+        "imageGenerateLabel": "Image generation",
        "agentRun": "{{agent}} execution",
        "collapsedSummary": "{{toolCount}} tool calls · {{processCount}} process messages",
        "collapseAction": "Collapse execution graph"
    },
+    "imageGeneration": {
+        "generating": "Generating image, please wait…"
+    },
    "composer": {
        "attachFiles": "Attach files",
        "pickSkill": "Choose skill",
--- a/src/i18n/locales/en/dashboard.json
+++ b/src/i18n/locales/en/dashboard.json
@@ -78,9 +78,15 @@
        "testing": "Generating…",
        "save": "Save",
        "saving": "Saving…",
+        "clear": "Clear configuration",
+        "clearing": "Clearing…",
+        "clearConfirmTitle": "Clear image generation configuration?",
+        "clearConfirmMessage": "This removes the custom image endpoint, model selection, and stored API key. Chat providers are not affected.",
+        "clearConfirmAction": "Clear",
        "testPrompt": "A small red circle on a white background, minimal flat illustration.",
        "toast": {
            "saved": "Image generation settings saved",
+            "cleared": "Image generation configuration cleared",
            "saveBeforeTest": "Save settings before running a test",
            "testSuccess": "Test image generated ({{ms}} ms)",
            "testSuccessWithPath": "Test image generated ({{ms}} ms): {{path}}",
@@ -90,7 +96,7 @@
            "title": "Custom image endpoint",
            "description": "Send image generation to a dedicated OpenAI Images-compatible endpoint (/v1/images/generations). This does not change any chat provider.",
            "baseUrl": "Image Base URL",
-            "baseUrlHint": "e.g. https://taolat.com/v1 — /v1 is appended automatically when omitted.",
+            "baseUrlHint": "e.g. https://api.example.com/v1 — /v1 is appended automatically when omitted.",
            "apiKey": "API key",
            "apiKeyPlaceholder": "sk-...",
            "apiKeyPlaceholderConfigured": "Already configured — leave blank to keep",
--- a/src/i18n/locales/ja/chat.json
+++ b/src/i18n/locales/ja/chat.json
@@ -159,10 +159,14 @@
        "branchLabel": "branch",
        "thinkingLabel": "考え中",
        "errorLabel": "エラー",
+        "imageGenerateLabel": "画像生成",
        "agentRun": "{{agent}} の実行",
        "collapsedSummary": "ツール呼び出し {{toolCount}} 件 · プロセスメッセージ {{processCount}} 件",
        "collapseAction": "実行グラフを折りたたむ"
    },
+    "imageGeneration": {
+        "generating": "画像を生成しています。少々お待ちください…"
+    },
    "composer": {
        "attachFiles": "ファイルを添付",
        "pickSkill": "Skill を選択",
--- a/src/i18n/locales/ja/dashboard.json
+++ b/src/i18n/locales/ja/dashboard.json
@@ -78,9 +78,15 @@
        "testing": "生成中…",
        "save": "保存",
        "saving": "保存中…",
+        "clear": "設定をクリア",
+        "clearing": "クリア中…",
+        "clearConfirmTitle": "画像生成設定をクリアしますか？",
+        "clearConfirmMessage": "カスタム画像エンドポイント、モデル設定、保存済み API キーを削除します。チャットプロバイダーには影響しません。",
+        "clearConfirmAction": "クリア",
        "testPrompt": "白背景に小さな赤い円、ミニマルなフラットイラスト。",
        "toast": {
            "saved": "画像生成設定を保存しました",
+            "cleared": "画像生成設定をクリアしました",
            "saveBeforeTest": "テスト前に設定を保存してください",
            "testSuccess": "テスト画像を生成しました（{{ms}} ms）",
            "testSuccessWithPath": "テスト画像を生成しました（{{ms}} ms）：{{path}}",
@@ -90,7 +96,7 @@
            "title": "カスタム画像エンドポイント",
            "description": "画像生成を専用の OpenAI Images 互換エンドポイント（/v1/images/generations）へ送ります。チャットプロバイダーは変更しません。",
            "baseUrl": "画像 Base URL",
-            "baseUrlHint": "例: https://taolat.com/v1 — /v1 が無い場合は自動付与します。",
+            "baseUrlHint": "例: https://api.example.com/v1 — /v1 が無い場合は自動付与します。",
            "apiKey": "API キー",
            "apiKeyPlaceholder": "sk-...",
            "apiKeyPlaceholderConfigured": "設定済み — 空欄のままなら変更しません",
--- a/src/i18n/locales/ru/chat.json
+++ b/src/i18n/locales/ru/chat.json
@@ -159,10 +159,14 @@
        "branchLabel": "ветвь",
        "thinkingLabel": "Думаю",
        "runErrorLabel": "Ошибка модели",
+        "imageGenerateLabel": "Генерация изображения",
        "agentRun": "Выполнение {{agent}}",
        "collapsedSummary": "Вызовов инструментов: {{toolCount}} · Промежуточных сообщений: {{processCount}}",
        "collapseAction": "Свернуть граф выполнения"
    },
+    "imageGeneration": {
+        "generating": "Изображение генерируется, подождите…"
+    },
    "composer": {
        "attachFiles": "Прикрепить файлы",
        "pickSkill": "Выбрать Skill",
--- a/src/i18n/locales/ru/dashboard.json
+++ b/src/i18n/locales/ru/dashboard.json
@@ -90,7 +90,7 @@
            "title": "Пользовательский endpoint изображений",
            "description": "Отправлять генерацию изображений на отдельный OpenAI Images-совместимый endpoint (/v1/images/generations), не меняя чат-провайдеры.",
            "baseUrl": "Base URL изображений",
-            "baseUrlHint": "например https://taolat.com/v1 — /v1 добавится автоматически.",
+            "baseUrlHint": "например https://api.example.com/v1 — /v1 добавится автоматически.",
            "apiKey": "API key",
            "apiKeyPlaceholder": "sk-...",
            "apiKeyPlaceholderConfigured": "Уже настроен — оставьте пустым, чтобы не менять",
--- a/src/i18n/locales/zh/chat.json
+++ b/src/i18n/locales/zh/chat.json
@@ -159,10 +159,14 @@
        "branchLabel": "分支",
        "thinkingLabel": "思考中",
        "modelErrorTitle": "模型调用失败",
+        "imageGenerateLabel": "图片生成",
        "agentRun": "{{agent}} 执行",
        "collapsedSummary": "{{toolCount}} 个工具调用，{{processCount}} 条过程消息",
        "collapseAction": "收起执行关系图"
    },
+    "imageGeneration": {
+        "generating": "图片生成中，请稍候…"
+    },
    "composer": {
        "attachFiles": "添加文件",
        "pickSkill": "选择技能",
--- a/src/i18n/locales/zh/dashboard.json
+++ b/src/i18n/locales/zh/dashboard.json
@@ -78,9 +78,15 @@
        "testing": "生成中…",
        "save": "保存",
        "saving": "保存中…",
+        "clear": "清除配置",
+        "clearing": "清除中…",
+        "clearConfirmTitle": "清除图像生成配置？",
+        "clearConfirmMessage": "将移除自定义生图端点、模型设置和已保存的 API Key，不会影响聊天 Provider。",
+        "clearConfirmAction": "清除",
        "testPrompt": "白色背景上的小红圆，极简扁平插画。",
        "toast": {
            "saved": "图像生成设置已保存",
+            "cleared": "图像生成配置已清除",
            "saveBeforeTest": "请先保存设置再运行测试",
            "testSuccess": "测试图像已生成（{{ms}} ms）",
            "testSuccessWithPath": "测试图像已生成（{{ms}} ms）：{{path}}",
@@ -90,7 +96,7 @@
            "title": "自定义生图端点",
            "description": "将生图请求发往独立的 OpenAI Images 兼容端点（/v1/images/generations），不会改变任何聊天 Provider。",
            "baseUrl": "生图 Base URL",
-            "baseUrlHint": "例如 https://taolat.com/v1；未带 /v1 时会自动补全。",
+            "baseUrlHint": "例如 https://api.example.com/v1；未带 /v1 时会自动补全。",
            "apiKey": "API Key",
            "apiKeyPlaceholder": "sk-...",
            "apiKeyPlaceholderConfigured": "已配置，留空则不修改",
--- a/src/lib/image-generation.ts
+++ b/src/lib/image-generation.ts
@@ -61,6 +61,10 @@ export async function fetchImageGenerationSettings(): Promise<ImageGenerationSet
  return response;
 }

+export async function clearImageGenerationSettings(): Promise<ImageGenerationSettingsSnapshot> {
+  return saveImageGenerationSettings({ openAiRelayEnabled: false });
+}
+
 export async function saveImageGenerationSettings(payload: {
  primary?: string | null;
  fallbacks?: string[];
--- a/src/pages/Chat/ChatMessage.tsx
+++ b/src/pages/Chat/ChatMessage.tsx
@@ -16,7 +16,8 @@ import { Button } from '@/components/ui/button';
 import { cn } from '@/lib/utils';
 import { invokeIpc, statFile } from '@/lib/api-client';
 import type { RawMessage, AttachedFileMeta } from '@/stores/chat';
-import { extractText, extractImages, extractToolUse, formatTimestamp } from './message-utils';
+import { extractText, extractImages, extractToolUse, formatTimestamp, isUnresolvableImageUrl } from './message-utils';
+import { copyImageToClipboard, type ImageCopyTarget } from './copy-image';

 interface ChatMessageProps {
  message: RawMessage;
@@ -222,7 +223,7 @@ function normalizeLatexDelimiters(input: string): string {

 /** Resolve an ExtractedImage to a displayable src string, or null if not possible. */
 function imageSrc(img: ExtractedImage): string | null {
-  if (img.url) return img.url;
+  if (img.url) return isUnresolvableImageUrl(img.url) ? null : img.url;
  if (img.data) return `data:${img.mimeType};base64,${img.data}`;
  return null;
 }
@@ -248,6 +249,7 @@ export const ChatMessage = memo(function ChatMessage({
  const hideAssistantText = suppressAssistantText && !isUser;
  const hasText = !hideAssistantText && text.trim().length > 0;
  const images = extractImages(message);
+  const resolvableContentImages = images.filter((img) => imageSrc(img) != null);
  const tools = extractToolUse(message);
  const visibleTools = suppressToolCards ? [] : tools;
  const [validatedPaths, setValidatedPaths] = useState<Record<string, boolean>>({});
@@ -321,16 +323,18 @@ export const ChatMessage = memo(function ChatMessage({
    if (!kind || !file.filePath) return true;
    return validatedPaths[file.filePath] === true;
  });
-  const attachedFiles = suppressProcessAttachments && (hasText || images.length > 0 || visibleTools.length > 0)
+  const attachedFiles = suppressProcessAttachments && (hasText || resolvableContentImages.length > 0 || visibleTools.length > 0)
    ? processVisibleAttachments
    : existingDerivedAttachedFiles;
+  const imageCopyTarget = resolvePrimaryImageCopyTarget(resolvableContentImages, attachedFiles);
+  const showAssistantHoverBar = !isUser && (hasText || imageCopyTarget != null);
  const [lightboxImg, setLightboxImg] = useState<{ src: string; fileName: string; filePath?: string; base64?: string; mimeType?: string } | null>(null);

  // Never render tool result messages in chat UI
  if (isToolResult) return null;

  const hasStreamingToolStatus = isStreaming && streamingTools.length > 0;
-  if (!hasText && images.length === 0 && visibleTools.length === 0 && attachedFiles.length === 0 && !hasStreamingToolStatus) return null;
+  if (!hasText && resolvableContentImages.length === 0 && visibleTools.length === 0 && attachedFiles.length === 0 && !hasStreamingToolStatus) return null;

  return (
    <div
@@ -359,9 +363,9 @@ export const ChatMessage = memo(function ChatMessage({

        {/* Images — rendered ABOVE text bubble for user messages */}
        {/* Images from content blocks (Gateway session data / channel push photos) */}
-        {isUser && images.length > 0 && (
+        {isUser && resolvableContentImages.length > 0 && (
          <div className="flex flex-wrap gap-2">
-            {images.map((img, i) => {
+            {resolvableContentImages.map((img, i) => {
              const src = imageSrc(img);
              if (!src) return null;
              return (
@@ -384,7 +388,7 @@ export const ChatMessage = memo(function ChatMessage({
            {attachedFiles.map((file, i) => {
              const isImage = file.mimeType.startsWith('image/');
              // Skip image attachments if we already have images from content blocks
-              if (isImage && images.length > 0) return null;
+              if (isImage && resolvableContentImages.length > 0) return null;
              if (isImage) {
                return file.preview ? (
                  <ImageThumbnail
@@ -420,9 +424,9 @@ export const ChatMessage = memo(function ChatMessage({
        )}

        {/* Images from content blocks — assistant messages (below text) */}
-        {!isUser && images.length > 0 && (
+        {!isUser && resolvableContentImages.length > 0 && (
          <div className="flex flex-wrap gap-2">
-            {images.map((img, i) => {
+            {resolvableContentImages.map((img, i) => {
              const src = imageSrc(img);
              if (!src) return null;
              return (
@@ -444,7 +448,7 @@ export const ChatMessage = memo(function ChatMessage({
          <div className="flex flex-wrap gap-2">
            {attachedFiles.map((file, i) => {
              const isImage = file.mimeType.startsWith('image/');
-              if (isImage && images.length > 0) return null;
+              if (isImage && resolvableContentImages.length > 0) return null;
              if (isImage && file.preview) {
                return (
                  <ImagePreviewCard
@@ -476,9 +480,9 @@ export const ChatMessage = memo(function ChatMessage({
          </span>
        )}

-        {/* Hover row for assistant messages — only when there is real text content */}
-        {!isUser && hasText && (
-          <AssistantHoverBar text={text} timestamp={message.timestamp} />
+        {/* Hover row for assistant messages */}
+        {showAssistantHoverBar && (
+          <AssistantHoverBar text={text} timestamp={message.timestamp} imageCopyTarget={imageCopyTarget} />
        )}
      </div>

@@ -547,16 +551,60 @@ function ToolStatusBar({
  );
 }

+function resolvePrimaryImageCopyTarget(
+  images: Array<{ mimeType: string; data?: string; url?: string }>,
+  attachedFiles: AttachedFileMeta[],
+): ImageCopyTarget | null {
+  for (const file of attachedFiles) {
+    if (!file.mimeType.startsWith('image/')) continue;
+    if (file.filePath || file.preview?.startsWith('data:')) {
+      return {
+        filePath: file.filePath,
+        preview: file.preview,
+        mimeType: file.mimeType,
+      };
+    }
+  }
+
+  for (const image of images) {
+    if (image.data) {
+      return { base64: image.data, mimeType: image.mimeType };
+    }
+    if (image.url?.startsWith('data:')) {
+      return { preview: image.url, mimeType: image.mimeType };
+    }
+  }
+
+  return null;
+}
+
 // ── Assistant hover bar (timestamp + copy, shown on group hover) ─

-function AssistantHoverBar({ text, timestamp }: { text: string; timestamp?: number }) {
+function AssistantHoverBar({
+  text,
+  timestamp,
+  imageCopyTarget,
+}: {
+  text: string;
+  timestamp?: number;
+  imageCopyTarget?: ImageCopyTarget | null;
+}) {
  const [copied, setCopied] = useState(false);

-  const copyContent = useCallback(() => {
-    navigator.clipboard.writeText(text);
+  const copyContent = useCallback(async () => {
+    if (imageCopyTarget) {
+      const copiedImage = await copyImageToClipboard(imageCopyTarget);
+      if (copiedImage) {
+        setCopied(true);
+        setTimeout(() => setCopied(false), 2000);
+        return;
+      }
+    }
+    if (!text.trim()) return;
+    await navigator.clipboard.writeText(text);
    setCopied(true);
    setTimeout(() => setCopied(false), 2000);
-  }, [text]);
+  }, [text, imageCopyTarget]);

  return (
    <div className="flex items-center justify-between w-full opacity-0 group-hover:opacity-100 transition-opacity duration-200 select-none px-1">
@@ -629,6 +677,12 @@ function AssistantMarkdown({
              </a>
            );
          },
+          img({ src, alt }) {
+            if (!src || isUnresolvableImageUrl(String(src))) return null;
+            return (
+              <img src={String(src)} alt={typeof alt === 'string' ? alt : 'image'} className="max-w-full rounded-lg" />
+            );
+          },
        }}
      >
        {normalizeLatexDelimiters(text)}
--- a/src/pages/Chat/ExecutionGraphCard.tsx
+++ b/src/pages/Chat/ExecutionGraphCard.tsx
@@ -47,6 +47,9 @@ function StepDetailCard({ step }: { step: TaskStep }) {
  const isNarration = step.kind === 'message';
  const isTool = step.kind === 'tool';
  const isThinking = step.kind === 'thinking';
+  const displayToolLabel = isTool && step.label === 'image_generate'
+    ? t('executionGraph.imageGenerateLabel')
+    : step.label;
  // System steps (subagent branch roots etc.) share the tool row layout:
  // bold label + truncated single-line detail preview + click-to-expand,
  // i.e. no rounded card / no separate detail line below the title.
@@ -56,7 +59,7 @@ function StepDetailCard({ step }: { step: TaskStep }) {
  const hideStatusText = (isTool || isSystem) && step.status === 'completed';
  const detailPreview = step.detail?.replace(/\s+/g, ' ').trim();
  const canExpand = hasDetail;
-    const displayLabel = isThinking ? t('executionGraph.thinkingLabel') : step.label;
+    const displayLabel = isThinking ? t('executionGraph.thinkingLabel') : (isTool ? displayToolLabel : step.label);

  return (
    <div
--- a/src/pages/Chat/copy-image.ts
+++ b/src/pages/Chat/copy-image.ts
@@ -0,0 +1,52 @@
+import { readBinaryFile } from '@/lib/api-client';
+
+export interface ImageCopyTarget {
+  filePath?: string;
+  preview?: string | null;
+  mimeType?: string;
+  base64?: string;
+}
+
+function blobFromBytes(data: Uint8Array, mimeType: string): Blob {
+  const copy = new Uint8Array(data);
+  return new Blob([copy], { type: mimeType });
+}
+
+async function blobFromDataUrl(dataUrl: string): Promise<Blob | null> {
+  try {
+    const response = await fetch(dataUrl);
+    if (!response.ok) return null;
+    return await response.blob();
+  } catch {
+    return null;
+  }
+}
+
+async function writeImageBlob(blob: Blob, fallbackMimeType: string): Promise<boolean> {
+  if (!navigator.clipboard?.write || typeof ClipboardItem === 'undefined') return false;
+  const type = blob.type || fallbackMimeType;
+  await navigator.clipboard.write([new ClipboardItem({ [type]: blob })]);
+  return true;
+}
+
+export async function copyImageToClipboard(target: ImageCopyTarget): Promise<boolean> {
+  const mimeType = target.mimeType?.startsWith('image/') ? target.mimeType : 'image/png';
+
+  if (target.base64) {
+    const blob = await blobFromDataUrl(`data:${mimeType};base64,${target.base64}`);
+    return blob ? writeImageBlob(blob, mimeType) : false;
+  }
+
+  if (target.preview?.startsWith('data:')) {
+    const blob = await blobFromDataUrl(target.preview);
+    return blob ? writeImageBlob(blob, mimeType) : false;
+  }
+
+  if (target.filePath) {
+    const result = await readBinaryFile(target.filePath);
+    if (!result.ok || !result.data) return false;
+    return writeImageBlob(blobFromBytes(result.data, result.mimeType || mimeType), mimeType);
+  }
+
+  return false;
+}
--- a/src/pages/Chat/image-generation-status.ts
+++ b/src/pages/Chat/image-generation-status.ts
@@ -0,0 +1,162 @@
+import type { RawMessage, ToolStatus } from '@/stores/chat';
+import {
+  extractImages,
+  extractToolUse,
+  isGeneratingStatusNarration,
+  isInternalAssistantReplyText,
+} from './message-utils';
+
+const IMAGE_GENERATE_TOOL = 'image_generate';
+const ASYNC_IMAGE_TASK_START_RE = /Background task started for image generation \(([0-9a-f-]{36})\)/i;
+const INTER_SESSION_IMAGE_TASK_RE = /sourceSession=image_generate:([0-9a-f-]{36})/i;
+/** Match OpenClaw agents.defaults.imageGenerationModel.timeoutMs default. */
+export const IMAGE_GENERATION_TIMEOUT_MS = 180_000;
+const IMAGE_GENERATION_TIMEOUT_BUFFER_MS = 15_000;
+
+function toMs(timestamp: number | undefined): number | null {
+  if (timestamp == null || !Number.isFinite(timestamp)) return null;
+  return timestamp < 1e12 ? timestamp * 1000 : timestamp;
+}
+
+function getMessagePlainText(message: RawMessage): string {
+  const { content } = message;
+  if (typeof content === 'string') return content;
+  if (!Array.isArray(content)) return '';
+  const parts: string[] = [];
+  for (const block of content as Array<{ type?: string; text?: string }>) {
+    if (block.type === 'text' && typeof block.text === 'string') {
+      parts.push(block.text);
+    }
+  }
+  return parts.join('\n');
+}
+
+function messageHasDeliveredImage(message: RawMessage): boolean {
+  if ((message._attachedFiles ?? []).some((file) => file.mimeType.startsWith('image/'))) {
+    return true;
+  }
+  return extractImages(message).length > 0;
+}
+
+function findLastImageGenerateToolCallIndex(segmentMessages: RawMessage[]): number {
+  for (let index = segmentMessages.length - 1; index >= 0; index -= 1) {
+    const message = segmentMessages[index];
+    if (message.role !== 'assistant') continue;
+    if (extractToolUse(message).some((tool) => tool.name === IMAGE_GENERATE_TOOL)) {
+      return index;
+    }
+  }
+  return -1;
+}
+
+function collectAsyncImageTaskStarts(
+  segmentMessages: RawMessage[],
+): Array<{ taskId: string; startedAtMs: number | null }> {
+  const starts: Array<{ taskId: string; startedAtMs: number | null }> = [];
+  for (const message of segmentMessages) {
+    const role = String(message.role ?? '').toLowerCase();
+    if (role !== 'toolresult' && role !== 'tool_result') continue;
+    if (message.toolName !== IMAGE_GENERATE_TOOL) continue;
+    const match = getMessagePlainText(message).match(ASYNC_IMAGE_TASK_START_RE);
+    if (!match?.[1]) continue;
+    starts.push({
+      taskId: match[1],
+      startedAtMs: toMs(message.timestamp),
+    });
+  }
+  return starts;
+}
+
+function collectCompletedAsyncImageTaskIds(segmentMessages: RawMessage[]): Set<string> {
+  const completed = new Set<string>();
+  for (const message of segmentMessages) {
+    const role = String(message.role ?? '').toLowerCase();
+    if (role !== 'user') continue;
+    const text = getMessagePlainText(message).trim();
+    if (!/^\[Inter-session message\]/i.test(text)) continue;
+    const match = text.match(INTER_SESSION_IMAGE_TASK_RE);
+    if (match?.[1]) completed.add(match[1]);
+  }
+  return completed;
+}
+
+function hasTimedOut(startedAtMs: number | null, now: number): boolean {
+  if (startedAtMs == null) return false;
+  return now - startedAtMs > IMAGE_GENERATION_TIMEOUT_MS + IMAGE_GENERATION_TIMEOUT_BUFFER_MS;
+}
+
+function hasDeliveredImageAfter(segmentMessages: RawMessage[], fromIndex: number): boolean {
+  for (let index = fromIndex + 1; index < segmentMessages.length; index += 1) {
+    const message = segmentMessages[index];
+    if (message.role !== 'assistant') continue;
+    if (messageHasDeliveredImage(message)) return true;
+  }
+  return false;
+}
+
+function hasTerminalReplyAfterToolCall(segmentMessages: RawMessage[], fromIndex: number): boolean {
+  for (let index = fromIndex + 1; index < segmentMessages.length; index += 1) {
+    const message = segmentMessages[index];
+    if (message.role !== 'assistant') continue;
+    if (extractToolUse(message).length > 0) continue;
+
+    const text = getMessagePlainText(message).trim();
+    if (!text || isInternalAssistantReplyText(text)) continue;
+    if (isGeneratingStatusNarration(text)) continue;
+    return true;
+  }
+  return false;
+}
+
+function hasFreshRunningStreamingTool(streamingTools: ToolStatus[], now: number): boolean {
+  return streamingTools.some((tool) => {
+    if (tool.name !== IMAGE_GENERATE_TOOL || tool.status !== 'running') return false;
+    return now - tool.updatedAt <= IMAGE_GENERATION_TIMEOUT_MS + IMAGE_GENERATION_TIMEOUT_BUFFER_MS;
+  });
+}
+
+/** True while an async `image_generate` task is in flight for this run segment. */
+export function isImageGenerationPending(
+  segmentMessages: RawMessage[],
+  streamingTools: ToolStatus[] = [],
+  now = Date.now(),
+): boolean {
+  if (hasFreshRunningStreamingTool(streamingTools, now)) {
+    return true;
+  }
+
+  const toolCallIndex = findLastImageGenerateToolCallIndex(segmentMessages);
+  const asyncStarts = collectAsyncImageTaskStarts(segmentMessages);
+  const completedTaskIds = collectCompletedAsyncImageTaskIds(segmentMessages);
+
+  if (toolCallIndex < 0 && asyncStarts.length === 0) {
+    return false;
+  }
+
+  const toolCallStartedAt = toolCallIndex >= 0
+    ? toMs(segmentMessages[toolCallIndex]?.timestamp)
+    : null;
+  const latestAsyncStartedAt = asyncStarts.reduce<number | null>((latest, start) => {
+    if (start.startedAtMs == null) return latest;
+    return latest == null ? start.startedAtMs : Math.max(latest, start.startedAtMs);
+  }, null);
+  const anchorStartedAt = [toolCallStartedAt, latestAsyncStartedAt]
+    .filter((value): value is number => value != null)
+    .reduce<number | null>((latest, value) => (latest == null ? value : Math.max(latest, value)), null);
+
+  if (hasTimedOut(anchorStartedAt, now)) {
+    return false;
+  }
+
+  if (toolCallIndex >= 0) {
+    if (hasDeliveredImageAfter(segmentMessages, toolCallIndex)) return false;
+    if (hasTerminalReplyAfterToolCall(segmentMessages, toolCallIndex)) return false;
+  }
+
+  if (asyncStarts.length > 0) {
+    const hasOpenAsyncTask = asyncStarts.some((start) => !completedTaskIds.has(start.taskId));
+    if (!hasOpenAsyncTask) return false;
+  }
+
+  return toolCallIndex >= 0 || asyncStarts.some((start) => !completedTaskIds.has(start.taskId));
+}
--- a/src/pages/Chat/index.tsx
+++ b/src/pages/Chat/index.tsx
@@ -7,6 +7,7 @@
 import { Suspense, lazy, useCallback, useEffect, useMemo, useRef, useState } from 'react';
 import { AlertCircle, ArrowDownToLine, Loader2, Sparkles } from 'lucide-react';
 import { useChatStore, type RawMessage } from '@/stores/chat';
+import { isInternalMessage } from '@/stores/chat/helpers';
 import { buildBaselineRunKey, getBaseline } from '@/stores/baseline-cache';
 import { useGatewayStore } from '@/stores/gateway';
 import { useAgentsStore } from '@/stores/agents';
@@ -18,7 +19,7 @@ import { ChatMessage } from './ChatMessage';
 import { ChatInput } from './ChatInput';
 import { ExecutionGraphCard } from './ExecutionGraphCard';
 import { ChatToolbar } from './ChatToolbar';
-import { extractImages, extractText, extractThinking, extractToolUse, normalizeMessageRole, stripProcessMessagePrefix } from './message-utils';
+import { extractImages, extractText, extractThinking, extractToolUse, isInternalAssistantReplyText, isInternalProcessNarration, normalizeMessageRole, stripProcessMessagePrefix } from './message-utils';
 import {
  buildRunSegmentMessageIndices,
  deriveTaskSteps,
@@ -30,6 +31,7 @@ import {
  segmentHasFinalReply,
  type TaskStep,
 } from './task-visualization';
+import { isImageGenerationPending } from './image-generation-status';
 import { useTranslation } from 'react-i18next';
 import { cn } from '@/lib/utils';
 import { useStickToBottomInstant } from '@/hooks/use-stick-to-bottom-instant';
@@ -95,6 +97,14 @@ function getPrimaryMessageStepTexts(steps: TaskStep[]): string[] {
    .map((step) => step.detail!);
 }

+function sanitizeGraphSteps(steps: TaskStep[]): TaskStep[] {
+  return steps.filter((step) => {
+    if (step.kind === 'thinking') return false;
+    if (step.kind === 'message' && step.detail && isInternalProcessNarration(step.detail)) return false;
+    return true;
+  });
+}
+
 function buildQuestionDirectoryTitle(message: RawMessage, fallback: string): string {
  const normalized = extractText(message).replace(/\s+/g, ' ').trim();
  if (!normalized) return fallback;
@@ -103,6 +113,7 @@ function buildQuestionDirectoryTitle(message: RawMessage, fallback: string): str

 function isRealUserMessage(msg: RawMessage): boolean {
  if (normalizeMessageRole(msg.role) !== 'user') return false;
+  if (isInternalMessage(msg)) return false;
  const content = msg.content;
  if (!Array.isArray(content)) return true;
  // If every block in the content is a tool_result, this is a Gateway
@@ -111,6 +122,10 @@ function isRealUserMessage(msg: RawMessage): boolean {
  return blocks.length === 0 || !blocks.every((b) => b.type === 'tool_result' || b.type === 'toolResult');
 }

+function hasUserFacingImageAttachments(msg: RawMessage): boolean {
+  return (msg._attachedFiles ?? []).some((file) => file.mimeType.startsWith('image/'));
+}
+
 function generatedFileToTarget(file: GeneratedFile): FilePreviewTarget {
  return {
    filePath: file.filePath,
@@ -493,15 +508,16 @@ export function Chat() {
      && streamTools.length === 0
      && !hasRunningStreamToolStatus;

-    let steps = buildSteps(rawStreamingReplyCandidate);
+    let steps = sanitizeGraphSteps(buildSteps(rawStreamingReplyCandidate));
    let streamingReplyText: string | null = null;
    if (rawStreamingReplyCandidate) {
      const trimmedReplyText = stripProcessMessagePrefix(streamText, getPrimaryMessageStepTexts(steps));
-      const hasReplyText = trimmedReplyText.trim().length > 0;
+      const hasReplyText = trimmedReplyText.trim().length > 0
+        && !isInternalAssistantReplyText(trimmedReplyText);
      if (hasReplyText || hasStreamImages) {
-        streamingReplyText = trimmedReplyText;
+        streamingReplyText = hasReplyText ? trimmedReplyText : '';
      } else {
-        steps = buildSteps(false);
+        steps = sanitizeGraphSteps(buildSteps(false));
      }
    }

@@ -545,9 +561,9 @@ export function Chat() {
      // generated message steps that include accumulated narration + reply
      // text.  Strip these out — historical message steps (from messages[])
      // will be properly recomputed on the next render with fresh data.
-      const cleanedSteps = cached.steps.filter(
+      const cleanedSteps = sanitizeGraphSteps(cached.steps.filter(
        (s) => !(s.kind === 'message' && s.id.startsWith('stream-message')),
-      );
+      ));
      return [{
        triggerIndex: idx,
        replyIndex: cached.replyIndex,
@@ -622,6 +638,7 @@ export function Chat() {
  }, [messages, subagentCompletionInfos, currentSessionKey, streamingMessage, streamingTools, pendingFinal, sending, hasAnyStreamContent, hasStreamText, hasStreamImages, streamText, streamTools.length, hasRunningStreamToolStatus, childTranscripts, currentAgentId, agents, sessionLabels, graphStepCache, runError, isRunTrigger]);
  const hasActiveExecutionGraph = userRunCards.some((card) => card.active);
  let latestRunSegmentCompletion = { hasFinalReply: false, hasToolActivity: false };
+  let pendingImageGeneration = false;
  for (let idx = messages.length - 1; idx >= 0; idx -= 1) {
    if (!isRealUserMessage(messages[idx]) || subagentCompletionInfos[idx]) continue;
    const nextUserIndex = nextUserMessageIndexes[idx];
@@ -632,6 +649,10 @@ export function Chat() {
        m.role === 'assistant' && extractToolUse(m).length > 0,
      ),
    };
+    pendingImageGeneration = isImageGenerationPending(
+      postTrigger,
+      nextUserIndex === -1 ? streamingTools : [],
+    );
    break;
  }
  const runSettledInHistory = latestRunSegmentCompletion.hasFinalReply
@@ -824,7 +845,9 @@ export function Chat() {
                    </div>
                  )}
                  {messages.map((msg, idx) => {
-                    if (foldedNarrationIndices.has(idx)) return null;
+                    if (isInternalMessage(msg) && !hasUserFacingImageAttachments(msg)) return null;
+                    const isFoldedNarration = foldedNarrationIndices.has(idx);
+                    if (isFoldedNarration && !hasUserFacingImageAttachments(msg)) return null;
                    const suppressToolCards = runSegmentMessageIndices.has(idx);
                    const isToolOnlyAssistant = normalizeMessageRole(msg.role) === 'assistant'
                      && extractToolUse(msg).length > 0
@@ -843,6 +866,7 @@ export function Chat() {
                      <ChatMessage
                        message={msg}
                        textOverride={replyTextOverrides.get(idx)}
+                        suppressAssistantText={isFoldedNarration}
                        suppressToolCards={suppressToolCards}
                        suppressProcessAttachments={suppressToolCards}
                        onOpenFile={handleOpenAttachedFile}
@@ -946,8 +970,12 @@ export function Chat() {
                    <ActivityIndicator phase="tool_processing" />
                  )}

+                  {pendingImageGeneration && (
+                    <ImageGeneratingIndicator />
+                  )}
+
                  {/* Typing indicator when sending but no stream content yet */}
-                  {inputRunActive && !pendingFinal && !hasAnyStreamContent && !hasActiveExecutionGraph && (
+                  {inputRunActive && !pendingFinal && !hasAnyStreamContent && !hasActiveExecutionGraph && !pendingImageGeneration && (
                    <TypingIndicator />
                  )}
                </>
@@ -1195,4 +1223,21 @@ function ActivityIndicator({ phase }: { phase: 'tool_processing' }) {
  );
 }

+function ImageGeneratingIndicator() {
+  const { t } = useTranslation('chat');
+  return (
+    <div className="flex gap-3" data-testid="chat-image-generating-indicator">
+      <div className="flex h-8 w-8 shrink-0 items-center justify-center rounded-full mt-1 bg-black/5 dark:bg-white/5 text-foreground">
+        <Sparkles className="h-4 w-4" />
+      </div>
+      <div className="bg-black/5 dark:bg-white/5 text-foreground rounded-2xl px-4 py-3">
+        <div className="flex items-center gap-2 text-sm text-muted-foreground">
+          <Loader2 className="h-3.5 w-3.5 animate-spin text-primary" />
+          <span>{t('imageGeneration.generating')}</span>
+        </div>
+      </div>
+    </div>
+  );
+}
+
 export default Chat;
--- a/src/pages/Chat/message-utils.ts
+++ b/src/pages/Chat/message-utils.ts
@@ -5,13 +5,39 @@
 */
 import type { RawMessage, ContentBlock } from '@/stores/chat';

+/**
+ * Strip OpenClaw's inbound-image vision envelope from user message text.
+ * The runtime expands uploaded images into:
+ *   [Image]
+ *   User text:
+ *   <message or "Process the attached file(s).">
+ *   Description:
+ *   <auto-generated caption>
+ */
+function stripInboundMediaVisionEnvelope(text: string): string {
+  if (!/\[Image\]/i.test(text) && !/^User text:/im.test(text) && !/\nDescription:\s*\n/i.test(text)) {
+    return text;
+  }
+
+  let result = text.replace(/^\s*\[Image\]\s*\n?/i, '');
+
+  const userTextBlock = result.match(/^User text:\s*\n([\s\S]*?)(?:\n\s*Description:\s*\n[\s\S]*)?\s*$/i);
+  if (userTextBlock) {
+    const userText = userTextBlock[1].trim();
+    return /^Process the attached file\(s\)\.\s*$/i.test(userText) ? '' : userText;
+  }
+
+  return result.replace(/\n\s*Description:\s*\n[\s\S]*$/i, '').trim();
+}
+
 /**
 * Clean Gateway metadata from user message text for display.
 * Strips: [media attached: ... | ...], [message_id: ...],
 * and the timestamp prefix [Day Date Time Timezone].
 */
 function cleanUserText(text: string): string {
-  return text
+  return stripInboundMediaVisionEnvelope(
+    text
    // Remove [media attached: path (mime) | path] references
    .replace(/\s*\[media attached:[^\]]*\]/g, '')
    // Remove [message_id: uuid]
@@ -30,7 +56,8 @@ function cleanUserText(text: string): string {
    .replace(/^Conversation info\s*\([^)]*\):\s*\{[\s\S]*?\}\s*/i, '')
    // Remove Gateway timestamp prefix like [Fri 2026-02-13 22:39 GMT+8]
    .replace(/^\[(?:Mon|Tue|Wed|Thu|Fri|Sat|Sun)\s+\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+[^\]]+\]\s*/i, '')
-    .trim();
+    .trim(),
+  );
 }

 /**
@@ -51,8 +78,13 @@ function stripAssistantMediaTags(text: string): string {
  // would still leak the literal `MEDIA:/.../截屏 2026-05-06 17.46.51.png`
  // to the user when the underlying path detection succeeds.
  const tagged = new RegExp(`(^|[\\s(\\[{>])(?:MEDIA|media):(?:\\/|~\\/)[^\\n"'()\\[\\],<>]*?\\.(?:${exts})(?=$|[\\s\\n"'()\\[\\],<>]|[，。；;,.!?])`, 'g');
+  // Bare OpenClaw artifact paths emitted alongside `_attachedFiles` cards.
+  // Scope to `.openclaw/media/` so normal absolute paths in prose stay visible.
+  const bareOpenClawMedia = new RegExp(`(^|[\\s(\\[{>])(?:(?:\\/|~\\/)[^\\n"'()\\[\\],<>]*?\\.openclaw\\/media\\/[^\\n"'()\\[\\],<>]*?\\.(?:${exts}))(?=$|[\\s\\n"'()\\[\\],<>]|[，。；;,.!?])`, 'g');
  return text
    .replace(tagged, (_, lead: string) => lead)
+    .replace(bareOpenClawMedia, (_, lead: string) => lead)
+    .replace(/!\[[^\]]*\]\([^)]+\)/g, '')
    // Collapse the empty lines / orphan whitespace the strip leaves behind.
    .replace(/[ \t]+\n/g, '\n')
    .replace(/\n{3,}/g, '\n\n')
@@ -143,6 +175,57 @@ function consumeLeadingSegment(text: string, segment: string): number {
  return match ? match[0].length : 0;
 }

+/** True for OpenClaw internal assistant tokens that must never appear in the chat UI. */
+export function isInternalAssistantReplyText(text: string): boolean {
+  return /^(HEARTBEAT_OK|NO_REPLY)\s*$/i.test(text.trim());
+}
+
+/** Interim image-generation status the user should not see as a final answer. */
+export function isGeneratingStatusNarration(text: string): boolean {
+  const trimmed = text.trim();
+  if (!trimmed) return false;
+  if (/^(?:图片(?:正在)?生成中|正在生成(?:图片|图像)|生成中)/i.test(trimmed)) return true;
+  if (/稍等(片刻|一下)?/.test(trimmed) && trimmed.length <= 80 && /生成|generat/i.test(trimmed)) return true;
+  return false;
+}
+
+/** OpenClaw runtime continuation prompt injected between internal async events. */
+export function isOpenClawRuntimeEventPrompt(text: string): boolean {
+  const trimmed = text.trim();
+  if (!trimmed) return false;
+  if (/^Continue the OpenClaw runtime event\.?\s*$/i.test(trimmed)) return true;
+  // Some transcripts store the prompt as the only line in a longer envelope.
+  return trimmed.split(/\n+/).some((line) => /^Continue the OpenClaw runtime event\.?\s*$/i.test(line.trim()));
+}
+
+/** Process narration that should never appear in the execution graph or chat stream. */
+export function isInternalProcessNarration(text: string): boolean {
+  const trimmed = text.trim();
+  if (!trimmed) return true;
+  if (isInternalAssistantReplyText(trimmed)) return true;
+  if (isGeneratingStatusNarration(trimmed)) return true;
+  if (isOpenClawRuntimeEventPrompt(trimmed)) return true;
+  if (/^\[Inter-session message\]/i.test(trimmed)) return true;
+  if (/OpenClaw runtime event/i.test(trimmed)) return true;
+  if (/Handle the result internally/i.test(trimmed) && /Do not relay it to the user/i.test(trimmed)) {
+    return true;
+  }
+  return false;
+}
+
+/** True when a renderer `src` cannot be fetched directly (Gateway proxy / local paths). */
+export function isUnresolvableImageUrl(url: string): boolean {
+  const trimmed = url.trim();
+  if (!trimmed) return true;
+  if (trimmed.startsWith('data:')) return false;
+  if (/^https?:\/\//i.test(trimmed)) return false;
+  if (trimmed.startsWith('/api/chat/media/')) return true;
+  if (trimmed.startsWith('file://')) return true;
+  if (trimmed.includes('.openclaw/media/')) return true;
+  if (trimmed.startsWith('/') || trimmed.startsWith('~/')) return true;
+  return false;
+}
+
 /**
 * Extract displayable text from a message's content field.
 * Handles both string content and array-of-blocks content.
@@ -178,6 +261,8 @@ export function extractText(message: RawMessage | unknown): string {
  if (isUser && result) {
    result = cleanUserText(result);
  } else if (!isUser && result) {
+    if (isInternalAssistantReplyText(result)) return '';
+    if (isGeneratingStatusNarration(result)) return '';
    // Assistant-side cleanup: keep the bubble free of `MEDIA:/path` tags
    // that the runtime emits to point at produced artifacts.  The same
    // path is surfaced as a clickable file card via `_attachedFiles`,
@@ -348,7 +433,7 @@ export function extractImages(
      const src = block.source;
      if (src.type === 'base64' && src.media_type && src.data) {
        images.push({ mimeType: src.media_type, data: src.data });
-      } else if (src.type === 'url' && src.url) {
+      } else if (src.type === 'url' && src.url && !isUnresolvableImageUrl(src.url)) {
        images.push({ mimeType: src.media_type || 'image/jpeg', url: src.url });
      }
      continue;
--- a/src/pages/Chat/task-visualization.ts
+++ b/src/pages/Chat/task-visualization.ts
@@ -1,4 +1,12 @@
-import { extractText, extractTextSegments, extractThinkingSegments, extractToolUse } from './message-utils';
+import {
+  extractText,
+  extractTextSegments,
+  extractToolUse,
+  isGeneratingStatusNarration,
+  isInternalAssistantReplyText,
+  isInternalProcessNarration,
+} from './message-utils';
+import { isInternalMessage } from '@/stores/chat/helpers';
 import type { RawMessage, ToolStatus } from '@/stores/chat';

 export type TaskStepStatus = 'running' | 'completed' | 'error';
@@ -37,7 +45,9 @@ export function findReplyMessageIndex(messages: RawMessage[], hasStreamingReply:
  for (let idx = messages.length - 1; idx >= 0; idx -= 1) {
    const message = messages[idx];
    if (!message || message.role !== 'assistant') continue;
-    if (extractText(message).trim().length === 0) continue;
+    const replyText = extractText(message).trim();
+    if (replyText.length === 0 || isInternalAssistantReplyText(replyText)) continue;
+    if (isGeneratingStatusNarration(replyText)) continue;
    return idx;
  }
  return -1;
@@ -149,7 +159,9 @@ export function segmentHasFinalReply(segmentMessages: RawMessage[]): boolean {
  return segmentMessages.some((message, index) => {
    if (index <= lastToolUseOffset) return false;
    if (message.role !== 'assistant') return false;
-    if (extractText(message).trim().length === 0) return false;
+    const replyText = extractText(message).trim();
+    if (replyText.length === 0 || isInternalAssistantReplyText(replyText)) return false;
+    if (isGeneratingStatusNarration(replyText)) return false;
    const content = message.content;
    if (!Array.isArray(content)) return true;
    return !(content as Array<{ type?: string }>).some(
@@ -310,7 +322,8 @@ function appendDetailSegments(
 ): void {
  const normalizedSegments = segments
    .map((segment) => normalizeText(segment))
-    .filter((segment): segment is string => !!segment);
+    .filter((segment): segment is string => !!segment)
+    .filter((segment) => !isInternalProcessNarration(segment));

  normalizedSegments.forEach((detail, index) => {
    options.upsertStep({
@@ -361,31 +374,27 @@ export function deriveTaskSteps({
  for (const [messageIndex, message] of messages.entries()) {
    if (!message || message.role !== 'assistant') continue;

-    appendDetailSegments(extractThinkingSegments(message), {
-      idPrefix: `history-thinking-${message.id || messageIndex}`,
-      label: 'Thinking',
-      kind: 'thinking',
-      running: false,
-      upsertStep,
-    });
-
    const toolUses = extractToolUse(message);
-    // Fold any intermediate assistant text into the graph as a narration
-    // step — including text that lives on a mixed `text + toolCall` message.
-    // The narration step is emitted BEFORE the tool steps so the graph
-    // preserves the original ordering (the assistant "thinks out loud" and
-    // then invokes the tool).
-    const narrationSegments = extractTextSegments(message);
-    const graphNarrationSegments = messageIndex === replyIndex
-      ? narrationSegments.slice(0, -1)
-      : narrationSegments;
-    appendDetailSegments(graphNarrationSegments, {
-      idPrefix: `history-message-${message.id || messageIndex}`,
-      label: 'Message',
-      kind: 'message',
-      running: false,
-      upsertStep,
-    });
+    if (!isInternalMessage(message)) {
+      // Fold any intermediate assistant text into the graph as a narration
+      // step — including text that lives on a mixed `text + toolCall` message.
+      // The narration step is emitted BEFORE the tool steps so the graph
+      // preserves the original ordering (the assistant "thinks out loud" and
+      // then invokes the tool).
+      const narrationSegments = extractTextSegments(message);
+      const graphNarrationSegments = messageIndex === replyIndex
+        ? narrationSegments.slice(0, -1)
+        : narrationSegments;
+      appendDetailSegments(graphNarrationSegments, {
+        idPrefix: `history-message-${message.id || messageIndex}`,
+        label: 'Message',
+        kind: 'message',
+        running: false,
+        upsertStep,
+      });
+    } else if (toolUses.length === 0) {
+      continue;
+    }

    toolUses.forEach((tool, index) => {
      const input = tool.input as Record<string, unknown>;
@@ -403,19 +412,6 @@ export function deriveTaskSteps({
  }

  if (streamMessage) {
-    // When the reply is being rendered as a separate bubble
-    // (omitLastStreamingMessageSegment), thinking that accompanies
-    // the reply belongs to the bubble — omit it from the graph.
-    if (!omitLastStreamingMessageSegment) {
-      appendDetailSegments(extractThinkingSegments(streamMessage), {
-        idPrefix: 'stream-thinking',
-        label: 'Thinking',
-        kind: 'thinking',
-        running: true,
-        upsertStep,
-      });
-    }
-
    // Stream-time narration should also appear in the execution graph so that
    // intermediate process output stays in P1 instead of leaking into the
    // assistant reply area.
--- a/src/stores/chat.ts
+++ b/src/stores/chat.ts
@@ -45,6 +45,12 @@ import {
  type ToolStatus,
 } from './chat/types';
 import type { ChatGet, ChatSet } from './chat/store-api';
+import { enrichWithToolCallAttachments, shouldDropMessageFromHistory } from './chat/helpers';
+import {
+  isGeneratingStatusNarration,
+  isInternalAssistantReplyText,
+  isOpenClawRuntimeEventPrompt,
+} from '@/pages/Chat/message-utils';

 export type {
  AttachedFileMeta,
@@ -567,8 +573,25 @@ function normalizeStreamingMessage(message: unknown): unknown {
 * `[media attached:` instead — leaving the timestamp in the normalized
 * comparison text and breaking optimistic-vs-echo dedupe.
 */
+function stripInboundMediaVisionEnvelope(text: string): string {
+  if (!/\[Image\]/i.test(text) && !/^User text:/im.test(text) && !/\nDescription:\s*\n/i.test(text)) {
+    return text;
+  }
+
+  let result = text.replace(/^\s*\[Image\]\s*\n?/i, '');
+
+  const userTextBlock = result.match(/^User text:\s*\n([\s\S]*?)(?:\n\s*Description:\s*\n[\s\S]*)?\s*$/i);
+  if (userTextBlock) {
+    const userText = userTextBlock[1].trim();
+    return /^Process the attached file\(s\)\.\s*$/i.test(userText) ? '' : userText;
+  }
+
+  return result.replace(/\n\s*Description:\s*\n[\s\S]*$/i, '').trim();
+}
+
 function stripGatewayUserMetadata(text: string): string {
-  return text
+  return stripInboundMediaVisionEnvelope(
+    text
    .replace(/\s*\[media attached:[^\]]*\]/g, '')
    .replace(/\s*\[message_id:\s*[^\]]+\]/g, '')
    .replace(/^Sender\s*\([^)]*\)\s*:\s*```[a-z]*\n[\s\S]*?```\s*/i, '')
@@ -579,13 +602,16 @@ function stripGatewayUserMetadata(text: string): string {
    .replace(/^Sender\s*:\s*[^\n]*(?:\n\s*)*/i, '')
    .replace(/^Conversation info\s*\([^)]*\):\s*```[a-z]*\n[\s\S]*?```\s*/i, '')
    .replace(/^Conversation info\s*\([^)]*\):\s*\{[\s\S]*?\}\s*/i, '')
-    .replace(/^\s*\[(?:Mon|Tue|Wed|Thu|Fri|Sat|Sun)\s+\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+[^\]]+\]\s*/i, '');
+    .replace(/^\s*\[(?:Mon|Tue|Wed|Thu|Fri|Sat|Sun)\s+\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+[^\]]+\]\s*/i, ''),
+  );
 }

 function normalizeComparableUserText(content: unknown): string {
-  return stripGatewayUserMetadata(getMessageText(content))
+  const text = stripGatewayUserMetadata(getMessageText(content))
    .replace(/\s+/g, ' ')
    .trim();
+  if (/^\(file attached\)$/i.test(text)) return '';
+  return text;
 }

 function getComparableAttachmentSignature(message: Pick<RawMessage, '_attachedFiles'>): string {
@@ -620,6 +646,13 @@ function matchesOptimisticUserMessage(
  if (sameText && sameAttachments) return true;
  if (sameText && (!optimisticAttachments || !candidateAttachments) && (timestampMatches || !hasCandidateTimestamp)) return true;
  if (sameAttachments && (!optimisticText || !candidateText) && (timestampMatches || !hasCandidateTimestamp)) return true;
+
+  const optimisticHadAttachmentsOnly = optimisticAttachments.length > 0 && !optimisticText;
+  const candidateIsAttachmentEcho = !candidateText
+    && /\[(?:media attached:|\s*Image\s*\])/i.test(getMessageText(candidate.content));
+  if (optimisticHadAttachmentsOnly && candidateIsAttachmentEcho && (timestampMatches || !hasCandidateTimestamp)) {
+    return true;
+  }
  return false;
 }

@@ -768,6 +801,13 @@ function getMessageText(content: unknown): string {
  return '';
 }

+function getMessageTextForFilter(msg: { content?: unknown; text?: unknown }): string {
+  const fromContent = getMessageText(msg.content);
+  if (fromContent.trim()) return fromContent;
+  if (typeof msg.text === 'string') return msg.text;
+  return '';
+}
+
 function getMessageStopReason(message: RawMessage | unknown): string | null {
  if (!message || typeof message !== 'object') return null;
  const msg = message as Record<string, unknown>;
@@ -853,6 +893,26 @@ function mimeFromExtension(filePath: string): string {
  return map[ext] || 'application/octet-stream';
 }

+function extractFilePathsFromToolArgs(args: Record<string, unknown>): string[] {
+  const paths: string[] = [];
+  const direct = args.file_path ?? args.filePath ?? args.path ?? args.file;
+  if (typeof direct === 'string' && direct.trim()) paths.push(direct.trim());
+
+  const attachments = args.attachments;
+  if (Array.isArray(attachments)) {
+    for (const item of attachments) {
+      if (!item || typeof item !== 'object') continue;
+      const att = item as Record<string, unknown>;
+      const filePath = att.filePath ?? att.file_path ?? att.path ?? att.file;
+      if (typeof filePath === 'string' && filePath.trim()) {
+        paths.push(filePath.trim());
+      }
+    }
+  }
+
+  return paths;
+}
+
 const DIRECTORY_MIME_TYPE = 'application/x-directory';

 function trimPathTerminators(filePath: string): string {
@@ -1018,8 +1078,8 @@ function getToolCallFilePath(msg: RawMessage, toolCallId: string): string | unde
      if ((block.type === 'tool_use' || block.type === 'toolCall') && block.id === toolCallId) {
        const args = (block.input ?? block.arguments) as Record<string, unknown> | undefined;
        if (args) {
-          const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-          if (typeof fp === 'string') return fp;
+          const paths = extractFilePathsFromToolArgs(args);
+          if (paths[0]) return paths[0];
        }
      }
    }
@@ -1037,8 +1097,8 @@ function getToolCallFilePath(msg: RawMessage, toolCallId: string): string | unde
        args = typeof fn.arguments === 'string' ? JSON.parse(fn.arguments) : (fn.arguments ?? fn.input) as Record<string, unknown>;
      } catch { /* ignore */ }
      if (args) {
-        const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-        if (typeof fp === 'string') return fp;
+        const paths = extractFilePathsFromToolArgs(args);
+        if (paths[0]) return paths[0];
      }
    }
  }
@@ -1056,8 +1116,8 @@ function collectToolCallPaths(msg: RawMessage, paths: Map<string, string>): void
      if ((block.type === 'tool_use' || block.type === 'toolCall') && block.id) {
        const args = (block.input ?? block.arguments) as Record<string, unknown> | undefined;
        if (args) {
-          const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-          if (typeof fp === 'string') paths.set(block.id, fp);
+          const filePaths = extractFilePathsFromToolArgs(args);
+          if (filePaths[0]) paths.set(block.id, filePaths[0]);
        }
      }
    }
@@ -1074,8 +1134,8 @@ function collectToolCallPaths(msg: RawMessage, paths: Map<string, string>): void
        args = typeof fn.arguments === 'string' ? JSON.parse(fn.arguments) : (fn.arguments ?? fn.input) as Record<string, unknown>;
      } catch { /* ignore */ }
      if (args) {
-        const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-        if (typeof fp === 'string') paths.set(id, fp);
+        const filePaths = extractFilePathsFromToolArgs(args);
+        if (filePaths[0]) paths.set(id, filePaths[0]);
      }
    }
  }
@@ -1613,11 +1673,12 @@ function isToolResultRole(role: unknown): boolean {
 }

 /** True for internal plumbing messages that should never be shown in the UI. */
-function isInternalMessage(msg: { role?: unknown; content?: unknown; idempotencyKey?: unknown; model?: unknown }): boolean {
+function isInternalMessage(msg: { role?: unknown; content?: unknown; idempotencyKey?: unknown; model?: unknown; text?: unknown }): boolean {
  if (msg.role === 'system') return true;
-  const text = getMessageText(msg.content);
+  const text = getMessageTextForFilter(msg);
  if (msg.role === 'assistant') {
-    if (/^(HEARTBEAT_OK|NO_REPLY)\s*$/.test(text)) return true;
+    if (isInternalAssistantReplyText(text)) return true;
+    if (isGeneratingStatusNarration(text)) return true;
    // OpenClaw's gateway writes a fallback `assistant-media` transcript
    // message when its `createManagedOutgoingImageBlocks` pipeline fails
    // ("could not be prepared" warning in stderr). The fallback has:
@@ -1642,6 +1703,12 @@ function isInternalMessage(msg: { role?: unknown; content?: unknown; idempotency
      // canonical render — keep them. Only hide the text-only fallback.
      if (!hasImageUrlBlock) return true;
    }
+    if (!text.trim() && Array.isArray(msg.content)) {
+      const blocks = msg.content as ContentBlock[];
+      const hasThinking = blocks.some((block) => block.type === 'thinking' && block.thinking?.trim());
+      const hasVisibleText = blocks.some((block) => block.type === 'text' && block.text?.trim());
+      if (hasThinking && !hasVisibleText) return true;
+    }
  }
  if (msg.role === 'user' && /^\[OpenClaw heartbeat poll\]\s*$/i.test(text.trim())) return true;
  // Runtime system injections: these arrive as user or assistant-role messages
@@ -1668,6 +1735,9 @@ function isRuntimeSystemInjection(text: string): boolean {
  ) {
    return true;
  }
+  if (/^\[Inter-session message\]/i.test(normalized)) return true;
+  if (isOpenClawRuntimeEventPrompt(normalized)) return true;
+
  if (
    /^\s*Current time\s*:/i.test(normalized)
    && /^\s*Current time\s*:[^\n]*\/\s*\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+UTC\s*$/i.test(normalized)
@@ -2631,7 +2701,8 @@ export const useChatStore = create<ChatState>((set, get) => ({

      // Before filtering: attach images/files from tool_result messages to the next assistant message
      const messagesWithToolImages = enrichWithToolResultFiles(rawMessages);
-      const filteredMessages = messagesWithToolImages.filter((msg) => !isToolResultRole(msg.role) && !isInternalMessage(msg));
+      const messagesWithToolAttachments = enrichWithToolCallAttachments(messagesWithToolImages);
+      const filteredMessages = messagesWithToolAttachments.filter((msg) => !shouldDropMessageFromHistory(msg));
      // Restore file attachments for user/assistant messages (from cache + text patterns)
      const enrichedMessages = enrichWithCachedImages(filteredMessages);

@@ -3024,7 +3095,8 @@ export const useChatStore = create<ChatState>((set, get) => ({
      // window with a larger suffix from the transcript.  This keeps render
      // cost bounded while allowing long conversations to page backwards.
      const messagesWithToolImages = enrichWithToolResultFiles(rawMessages);
-      const filteredMessages = messagesWithToolImages.filter((msg) => !isToolResultRole(msg.role) && !isInternalMessage(msg));
+      const messagesWithToolAttachments = enrichWithToolCallAttachments(messagesWithToolImages);
+      const filteredMessages = messagesWithToolAttachments.filter((msg) => !shouldDropMessageFromHistory(msg));
      const enrichedMessages = enrichWithCachedImages(filteredMessages);
      set({
        messages: enrichedMessages,
--- a/src/stores/chat/helpers.ts
+++ b/src/stores/chat/helpers.ts
@@ -1,4 +1,9 @@
 import { invokeIpc } from '@/lib/api-client';
+import {
+  isGeneratingStatusNarration,
+  isInternalAssistantReplyText,
+  isOpenClawRuntimeEventPrompt,
+} from '@/pages/Chat/message-utils';
 import type { AttachedFileMeta, ChatSession, ContentBlock, RawMessage, ToolStatus } from './types';

 // Module-level timestamp tracking the last chat event received.
@@ -186,8 +191,25 @@ function normalizeStreamingMessage(message: unknown): unknown {
 * `[media attached:` instead — leaving the timestamp in the normalized
 * comparison text and breaking optimistic-vs-echo dedupe.
 */
+function stripInboundMediaVisionEnvelope(text: string): string {
+  if (!/\[Image\]/i.test(text) && !/^User text:/im.test(text) && !/\nDescription:\s*\n/i.test(text)) {
+    return text;
+  }
+
+  let result = text.replace(/^\s*\[Image\]\s*\n?/i, '');
+
+  const userTextBlock = result.match(/^User text:\s*\n([\s\S]*?)(?:\n\s*Description:\s*\n[\s\S]*)?\s*$/i);
+  if (userTextBlock) {
+    const userText = userTextBlock[1].trim();
+    return /^Process the attached file\(s\)\.\s*$/i.test(userText) ? '' : userText;
+  }
+
+  return result.replace(/\n\s*Description:\s*\n[\s\S]*$/i, '').trim();
+}
+
 function stripGatewayUserMetadata(text: string): string {
-  return text
+  return stripInboundMediaVisionEnvelope(
+    text
    .replace(/\s*\[media attached:[^\]]*\]/g, '')
    .replace(/\s*\[message_id:\s*[^\]]+\]/g, '')
    .replace(/^Sender\s*\([^)]*\)\s*:\s*```[a-z]*\n[\s\S]*?```\s*/i, '')
@@ -198,13 +220,16 @@ function stripGatewayUserMetadata(text: string): string {
    .replace(/^Sender\s*:\s*[^\n]*(?:\n\s*)*/i, '')
    .replace(/^Conversation info\s*\([^)]*\):\s*```[a-z]*\n[\s\S]*?```\s*/i, '')
    .replace(/^Conversation info\s*\([^)]*\):\s*\{[\s\S]*?\}\s*/i, '')
-    .replace(/^\s*\[(?:Mon|Tue|Wed|Thu|Fri|Sat|Sun)\s+\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+[^\]]+\]\s*/i, '');
+    .replace(/^\s*\[(?:Mon|Tue|Wed|Thu|Fri|Sat|Sun)\s+\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+[^\]]+\]\s*/i, ''),
+  );
 }

 function normalizeComparableUserText(content: unknown): string {
-  return stripGatewayUserMetadata(getMessageText(content))
+  const text = stripGatewayUserMetadata(getMessageText(content))
    .replace(/\s+/g, ' ')
    .trim();
+  if (/^\(file attached\)$/i.test(text)) return '';
+  return text;
 }

 function getComparableAttachmentSignature(message: Pick<RawMessage, '_attachedFiles'>): string {
@@ -239,6 +264,13 @@ function matchesOptimisticUserMessage(
  if (sameText && sameAttachments) return true;
  if (sameText && (!optimisticAttachments || !candidateAttachments) && (timestampMatches || !hasCandidateTimestamp)) return true;
  if (sameAttachments && (!optimisticText || !candidateText) && (timestampMatches || !hasCandidateTimestamp)) return true;
+
+  const optimisticHadAttachmentsOnly = optimisticAttachments.length > 0 && !optimisticText;
+  const candidateIsAttachmentEcho = !candidateText
+    && /\[(?:media attached:|\s*Image\s*\])/i.test(getMessageText(candidate.content));
+  if (optimisticHadAttachmentsOnly && candidateIsAttachmentEcho && (timestampMatches || !hasCandidateTimestamp)) {
+    return true;
+  }
  return false;
 }

@@ -399,6 +431,13 @@ function getMessageText(content: unknown): string {
  return '';
 }

+function getMessageTextForFilter(msg: { content?: unknown; text?: unknown }): string {
+  const fromContent = getMessageText(msg.content);
+  if (fromContent.trim()) return fromContent;
+  if (typeof msg.text === 'string') return msg.text;
+  return '';
+}
+
 function getMessageStopReason(message: RawMessage | unknown): string | null {
  if (!message || typeof message !== 'object') return null;
  const msg = message as Record<string, unknown>;
@@ -486,12 +525,131 @@ function mimeFromExtension(filePath: string): string {
  return map[ext] || 'application/octet-stream';
 }

+/** Extract local file paths declared in tool call arguments. */
+function extractFilePathsFromToolArgs(args: Record<string, unknown>): string[] {
+  const paths: string[] = [];
+  const direct = args.file_path ?? args.filePath ?? args.path ?? args.file;
+  if (typeof direct === 'string' && direct.trim()) paths.push(direct.trim());
+
+  const attachments = args.attachments;
+  if (Array.isArray(attachments)) {
+    for (const item of attachments) {
+      if (!item || typeof item !== 'object') continue;
+      const att = item as Record<string, unknown>;
+      const filePath = att.filePath ?? att.file_path ?? att.path ?? att.file;
+      if (typeof filePath === 'string' && filePath.trim()) {
+        paths.push(filePath.trim());
+      }
+    }
+  }
+
+  return paths;
+}
+
+/**
+ * Surface user-facing attachments declared in assistant tool calls (e.g.
+ * `message` tool `attachments: [{ filePath }]`) on the calling turn itself.
+ */
+function enrichWithToolCallAttachments(messages: RawMessage[]): RawMessage[] {
+  return messages.map((msg) => {
+    if (msg.role !== 'assistant') return msg;
+
+    const attachmentPaths = new Set<string>();
+    const content = msg.content;
+    if (Array.isArray(content)) {
+      for (const block of content as ContentBlock[]) {
+        if (block.type !== 'tool_use' && block.type !== 'toolCall') continue;
+        const args = (block.input ?? block.arguments) as Record<string, unknown> | undefined;
+        if (!args) continue;
+        for (const filePath of extractFilePathsFromToolArgs(args)) {
+          attachmentPaths.add(filePath);
+        }
+      }
+    }
+
+    const msgAny = msg as unknown as Record<string, unknown>;
+    const toolCalls = msgAny.tool_calls ?? msgAny.toolCalls;
+    if (Array.isArray(toolCalls)) {
+      for (const tc of toolCalls as Array<Record<string, unknown>>) {
+        const fn = (tc.function ?? tc) as Record<string, unknown>;
+        let args: Record<string, unknown> | undefined;
+        try {
+          args = typeof fn.arguments === 'string'
+            ? JSON.parse(fn.arguments)
+            : (fn.arguments ?? fn.input) as Record<string, unknown>;
+        } catch { /* ignore */ }
+        if (!args) continue;
+        for (const filePath of extractFilePathsFromToolArgs(args)) {
+          attachmentPaths.add(filePath);
+        }
+      }
+    }
+
+    if (attachmentPaths.size === 0) return msg;
+
+    const existingPaths = new Set(
+      (msg._attachedFiles || []).map((file) => file.filePath).filter(Boolean),
+    );
+    const newFiles = [...attachmentPaths]
+      .filter((filePath) => !existingPaths.has(filePath))
+      .map((filePath) => makeAttachedFile({ filePath, mimeType: mimeFromExtension(filePath) }, 'tool-result'));
+
+    if (newFiles.length === 0) return msg;
+    return {
+      ...msg,
+      _attachedFiles: [...(msg._attachedFiles || []), ...newFiles],
+    };
+  });
+}
+
 const DIRECTORY_MIME_TYPE = 'application/x-directory';

 function trimPathTerminators(filePath: string): string {
  return filePath.replace(/[，。；;,.!?]+$/u, '');
 }

+type MarkdownImageRef =
+  | { filePath: string; mimeType: string; fileName: string }
+  | { gatewayUrl: string; mimeType: string; fileName: string; source: 'gateway-media' };
+
+/** Extract image targets from markdown `![alt](target)` in assistant text. */
+function extractMarkdownImageRefs(text: string): MarkdownImageRef[] {
+  if (!text) return [];
+  const refs: MarkdownImageRef[] = [];
+  const seen = new Set<string>();
+  const markdownImageRegex = /!\[([^\]]*)\]\(([^)]+)\)/g;
+  let match: RegExpExecArray | null;
+  while ((match = markdownImageRegex.exec(text)) !== null) {
+    const alt = match[1]?.trim() || 'image';
+    let target = match[2]?.trim() ?? '';
+    if (!target) continue;
+    if (target.startsWith('file://')) {
+      target = decodeURIComponent(target.replace(/^file:\/\//, ''));
+    }
+    if (target.startsWith('/api/chat/media/')) {
+      if (seen.has(target)) continue;
+      seen.add(target);
+      refs.push({
+        gatewayUrl: target,
+        mimeType: 'image/png',
+        fileName: alt,
+        source: 'gateway-media',
+      });
+      continue;
+    }
+    const normalizedPath = trimPathTerminators(target);
+    if (!normalizedPath.startsWith('/') && !normalizedPath.startsWith('~/')) continue;
+    if (seen.has(normalizedPath)) continue;
+    seen.add(normalizedPath);
+    refs.push({
+      filePath: normalizedPath,
+      mimeType: mimeFromExtension(normalizedPath),
+      fileName: alt,
+    });
+  }
+  return refs;
+}
+
 /**
 * Extract raw file paths from message text.
 * Detects absolute paths (Unix: / or ~/, Windows: C:\ etc.) ending with common file extensions.
@@ -661,8 +819,8 @@ function getToolCallFilePath(msg: RawMessage, toolCallId: string): string | unde
      if ((block.type === 'tool_use' || block.type === 'toolCall') && block.id === toolCallId) {
        const args = (block.input ?? block.arguments) as Record<string, unknown> | undefined;
        if (args) {
-          const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-          if (typeof fp === 'string') return fp;
+          const paths = extractFilePathsFromToolArgs(args);
+          if (paths[0]) return paths[0];
        }
      }
    }
@@ -680,8 +838,8 @@ function getToolCallFilePath(msg: RawMessage, toolCallId: string): string | unde
        args = typeof fn.arguments === 'string' ? JSON.parse(fn.arguments) : (fn.arguments ?? fn.input) as Record<string, unknown>;
      } catch { /* ignore */ }
      if (args) {
-        const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-        if (typeof fp === 'string') return fp;
+        const paths = extractFilePathsFromToolArgs(args);
+        if (paths[0]) return paths[0];
      }
    }
  }
@@ -699,8 +857,8 @@ function collectToolCallPaths(msg: RawMessage, paths: Map<string, string>): void
      if ((block.type === 'tool_use' || block.type === 'toolCall') && block.id) {
        const args = (block.input ?? block.arguments) as Record<string, unknown> | undefined;
        if (args) {
-          const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-          if (typeof fp === 'string') paths.set(block.id, fp);
+          const filePaths = extractFilePathsFromToolArgs(args);
+          if (filePaths[0]) paths.set(block.id, filePaths[0]);
        }
      }
    }
@@ -717,8 +875,8 @@ function collectToolCallPaths(msg: RawMessage, paths: Map<string, string>): void
        args = typeof fn.arguments === 'string' ? JSON.parse(fn.arguments) : (fn.arguments ?? fn.input) as Record<string, unknown>;
      } catch { /* ignore */ }
      if (args) {
-        const fp = args.file_path ?? args.filePath ?? args.path ?? args.file;
-        if (typeof fp === 'string') paths.set(id, fp);
+        const filePaths = extractFilePathsFromToolArgs(args);
+        if (filePaths[0]) paths.set(id, filePaths[0]);
      }
    }
  }
@@ -744,7 +902,6 @@ function enrichWithToolResultFiles(messages: RawMessage[]): RawMessage[] {
    }

    if (isToolResultRole(msg.role)) {
-      // Resolve file path from the matching tool call
      const matchedPath = msg.toolCallId ? toolCallPaths.get(msg.toolCallId) : undefined;

      // 1. Image/file content blocks in the structured content array.
@@ -788,6 +945,11 @@ function enrichWithToolResultFiles(messages: RawMessage[]): RawMessage[] {
    }

    if (msg.role === 'assistant' && pending.length > 0) {
+      // Internal-only turns (NO_REPLY, interim narration, ...) must not consume
+      // pending attachments — the next visible assistant reply should get them.
+      if (isInternalMessage(msg) && !messageHasToolUse(msg)) {
+        return msg;
+      }
      const toAttach = pending.splice(0);
      // Deduplicate against files already on the assistant message
      const existingPaths = new Set(
@@ -857,6 +1019,13 @@ function enrichWithCachedImages(messages: RawMessage[]): RawMessage[] {
    if (msg.role === 'assistant' && !isToolOnlyMessage(msg)) {
      // Own text
      rawRefs = extractRawFilePaths(text).filter(r => !mediaRefPaths.has(r.filePath));
+      const rawPathSet = new Set(rawRefs.map((ref) => ref.filePath));
+      for (const ref of extractMarkdownImageRefs(text)) {
+        if ('filePath' in ref && !mediaRefPaths.has(ref.filePath) && !rawPathSet.has(ref.filePath)) {
+          rawPathSet.add(ref.filePath);
+          rawRefs.push({ filePath: ref.filePath, mimeType: ref.mimeType });
+        }
+      }

      // Nearest preceding user message text (look back up to 5 messages)
      const seenPaths = new Set(rawRefs.map(r => r.filePath));
@@ -884,7 +1053,16 @@ function enrichWithCachedImages(messages: RawMessage[]): RawMessage[] {
    }

    const allRefs = [...mediaRefs, ...rawRefs];
-    if (allRefs.length === 0 && gatewayMediaFiles.length === 0) return msg;
+    const markdownImageRefs = msg.role === 'assistant' && !isToolOnlyMessage(msg)
+      ? extractMarkdownImageRefs(text)
+      : [];
+    if (
+      allRefs.length === 0
+      && gatewayMediaFiles.length === 0
+      && markdownImageRefs.length === 0
+    ) {
+      return msg;
+    }

    const existingFiles = msg._attachedFiles || [];
    const existingPaths = new Set(existingFiles.map(file => file.filePath).filter(Boolean));
@@ -902,8 +1080,19 @@ function enrichWithCachedImages(messages: RawMessage[]): RawMessage[] {
    const dedupedGatewayMedia = gatewayMediaFiles.filter(
      file => file.gatewayUrl && !existingGatewayUrls.has(file.gatewayUrl),
    );
-    if (files.length === 0 && dedupedGatewayMedia.length === 0) return msg;
-    return { ...msg, _attachedFiles: [...existingFiles, ...files, ...dedupedGatewayMedia] };
+    const markdownGatewayMedia: AttachedFileMeta[] = markdownImageRefs
+      .filter((ref): ref is Extract<MarkdownImageRef, { gatewayUrl: string }> => 'gatewayUrl' in ref)
+      .filter((ref) => ref.gatewayUrl && !existingGatewayUrls.has(ref.gatewayUrl))
+      .map((ref) => ({
+        fileName: ref.fileName,
+        mimeType: ref.mimeType,
+        fileSize: 0,
+        preview: null,
+        gatewayUrl: ref.gatewayUrl,
+        source: 'gateway-media' as const,
+      }));
+    if (files.length === 0 && dedupedGatewayMedia.length === 0 && markdownGatewayMedia.length === 0) return msg;
+    return { ...msg, _attachedFiles: [...existingFiles, ...files, ...dedupedGatewayMedia, ...markdownGatewayMedia] };
  });
 }

@@ -1077,12 +1266,31 @@ function isToolResultRole(role: unknown): boolean {
  return normalized === 'toolresult' || normalized === 'tool_result';
 }

+function messageHasToolUse(msg: { role?: unknown; content?: unknown; tool_calls?: unknown; toolCalls?: unknown }): boolean {
+  if (msg.role !== 'assistant') return false;
+  if (Array.isArray(msg.content)) {
+    const blocks = msg.content as ContentBlock[];
+    if (blocks.some((block) => block.type === 'tool_use' || block.type === 'toolCall')) {
+      return true;
+    }
+  }
+  const toolCalls = msg.tool_calls ?? msg.toolCalls;
+  return Array.isArray(toolCalls) && toolCalls.length > 0;
+}
+
 /** True for internal plumbing messages that should never be shown in the UI. */
-function isInternalMessage(msg: { role?: unknown; content?: unknown }): boolean {
+function isInternalMessage(msg: { role?: unknown; content?: unknown; text?: unknown }): boolean {
  if (msg.role === 'system') return true;
-  const text = getMessageText(msg.content);
+  const text = getMessageTextForFilter(msg);
  if (msg.role === 'assistant') {
-    if (/^(HEARTBEAT_OK|NO_REPLY)\s*$/.test(text)) return true;
+    if (isInternalAssistantReplyText(text)) return true;
+    if (isGeneratingStatusNarration(text)) return true;
+    if (!text.trim() && Array.isArray(msg.content)) {
+      const blocks = msg.content as ContentBlock[];
+      const hasThinking = blocks.some((block) => block.type === 'thinking' && block.thinking?.trim());
+      const hasVisibleText = blocks.some((block) => block.type === 'text' && block.text?.trim());
+      if (hasThinking && !hasVisibleText) return true;
+    }
  }
  if (msg.role === 'user' && /^\[OpenClaw heartbeat poll\]\s*$/i.test(text.trim())) return true;
  // Runtime system injections: these arrive as user or assistant-role messages
@@ -1091,6 +1299,17 @@ function isInternalMessage(msg: { role?: unknown; content?: unknown }): boolean
  return false;
 }

+/**
+ * History filtering must keep assistant tool-call turns even when their visible
+ * text is internal narration (e.g. "生成中，稍等" + `image_generate`). Those
+ * turns power the execution graph and run lifecycle detection.
+ */
+function shouldDropMessageFromHistory(msg: { role?: unknown; content?: unknown; text?: unknown; tool_calls?: unknown; toolCalls?: unknown }): boolean {
+  if (isToolResultRole(msg.role)) return true;
+  if (messageHasToolUse(msg)) return false;
+  return isInternalMessage(msg);
+}
+
 /**
 * Detect runtime-injected system messages that should be hidden from the chat UI.
 * These are injected by the OpenClaw runtime as user-role messages and include:
@@ -1114,8 +1333,11 @@ function isRuntimeSystemInjection(text: string): boolean {
    return true;
  }

-  // Standalone time injection (e.g. "Current time: Wednesday, April 22nd, 2026 - 10:06 (Asia/Shanghai) / 2026-04-22 02:06 UTC")
-  // Only match when the full message is the time announcement.
+  if (/^\[Inter-session message\]/i.test(normalized)) return true;
+
+  if (isOpenClawRuntimeEventPrompt(normalized)) return true;
+
+  // Standalone time injection
  if (
    /^\s*Current time\s*:/i.test(normalized)
    && /^\s*Current time\s*:[^\n]*\/\s*\d{4}-\d{2}-\d{2}\s+\d{2}:\d{2}\s+UTC\s*$/i.test(normalized)
@@ -1439,7 +1661,9 @@ export {
  extractRawFilePaths,
  makeAttachedFile,
  enrichWithToolResultFiles,
+  enrichWithToolCallAttachments,
  isInternalMessage,
+  shouldDropMessageFromHistory,
  isToolResultRole,
  enrichWithCachedImages,
  loadMissingPreviews,
--- a/src/stores/chat/history-actions.ts
+++ b/src/stores/chat/history-actions.ts
@@ -3,14 +3,14 @@ import { hostApiFetch } from '@/lib/host-api';
 import { useGatewayStore } from '@/stores/gateway';
 import {
  clearHistoryPoll,
+  enrichWithToolCallAttachments,
  enrichWithCachedImages,
  enrichWithToolResultFiles,
  getLatestOptimisticUserMessage,
  getMessageErrorMessage,
  getMessageStopReason,
  getMessageText,
-  isInternalMessage,
-  isToolResultRole,
+  shouldDropMessageFromHistory,
  loadMissingPreviews,
  mergePendingOptimisticUserMessages,
  dropRedundantOptimisticUserMessages,
@@ -112,7 +112,8 @@ export function createHistoryActions(
        if (!isCurrentSession()) return false;
        // Before filtering: attach images/files from tool_result messages to the next assistant message
        const messagesWithToolImages = enrichWithToolResultFiles(rawMessages);
-        const filteredMessages = messagesWithToolImages.filter((msg) => !isToolResultRole(msg.role) && !isInternalMessage(msg));
+        const messagesWithToolAttachments = enrichWithToolCallAttachments(messagesWithToolImages);
+        const filteredMessages = messagesWithToolAttachments.filter((msg) => !shouldDropMessageFromHistory(msg));
        // Restore file attachments for user/assistant messages (from cache + text patterns)
        const enrichedMessages = enrichWithCachedImages(filteredMessages);

--- a/tests/e2e/image-generation-settings.spec.ts
+++ b/tests/e2e/image-generation-settings.spec.ts
@@ -33,6 +33,7 @@ test.describe('Image generation settings page', () => {
    await expect(page.getByTestId('image-generation-primary')).toHaveCount(0);
    await expect(page.getByTestId('image-generation-fallbacks')).toHaveCount(0);
    await expect(page.getByTestId('image-generation-save')).toBeVisible();
+    await expect(page.getByTestId('image-generation-clear')).toBeDisabled();
  });

  test('configures an independent OpenAI-compatible image endpoint', async ({ page }) => {
@@ -45,7 +46,7 @@ test.describe('Image generation settings page', () => {

    await expect(page.getByTestId('image-generation-settings')).toBeVisible();
    await expect(page.getByTestId('image-generation-relay-base-url')).toBeVisible();
-    await page.getByTestId('image-generation-relay-base-url').fill('https://taolat.com/v1');
+    await page.getByTestId('image-generation-relay-base-url').fill('https://api.example.com/v1');
    await page.getByTestId('image-generation-relay-model').fill('gpt-image-2');
    await page.getByTestId('image-generation-relay-api-key').fill('sk-test-image');

@@ -81,7 +82,7 @@ test.describe('Image generation settings page', () => {
              ],
              openAiRelay: {
                enabled: true,
-                baseUrl: 'https://taolat.com/v1',
+                baseUrl: 'https://api.example.com/v1',
                model: 'gpt-image-2',
                providerKey: 'clawx-openai-image',
                apiKeyConfigured: true,
@@ -103,4 +104,79 @@ test.describe('Image generation settings page', () => {
    await expect(page.getByTestId('image-generation-api-key-status')).not.toBeEmpty();
    await expect(page.getByTestId('image-generation-relay-api-key')).toHaveAttribute('placeholder', /.+/);
  });
+
+  test('clears configured image generation settings after confirmation', async ({ electronApp, page }) => {
+    const configuredResponse = {
+      success: true,
+      config: {
+        primary: 'clawx-openai-image/gpt-image-2',
+        fallbacks: [],
+        timeoutMs: 180000,
+      },
+      autoProviderFallback: false,
+      defaultAgentId: 'default',
+      agents: [
+        {
+          id: 'default',
+          name: 'Default',
+          isDefault: true,
+          provider: null,
+          configured: false,
+        },
+      ],
+      openAiRelay: {
+        enabled: true,
+        baseUrl: 'https://api.example.com/v1',
+        model: 'gpt-image-2',
+        providerKey: 'clawx-openai-image',
+        apiKeyConfigured: true,
+      },
+    };
+    const clearedResponse = {
+      ...configuredResponse,
+      config: {
+        primary: null,
+        fallbacks: [],
+        timeoutMs: 180000,
+      },
+      openAiRelay: {
+        enabled: false,
+        baseUrl: '',
+        model: 'gpt-image-2',
+        providerKey: undefined,
+        apiKeyConfigured: false,
+      },
+    };
+
+    await installIpcMocks(electronApp, {
+      hostApi: {
+        '["/api/media/image-generation","GET"]': {
+          ok: true,
+          data: { status: 200, ok: true, json: configuredResponse },
+        },
+        '["/api/media/image-generation","PUT"]': {
+          ok: true,
+          data: { status: 200, ok: true, json: { success: true, ...clearedResponse } },
+        },
+      },
+    });
+
+    await expect(page.getByTestId('setup-page')).toBeVisible();
+    await page.getByTestId('setup-skip-button').click();
+
+    await expect(page.getByTestId('main-layout')).toBeVisible();
+    await unlockDeveloperMode(page);
+    await page.getByTestId('sidebar-nav-image-generation').click();
+
+    await expect(page.getByTestId('image-generation-relay-base-url')).toHaveValue('https://api.example.com/v1');
+    await expect(page.getByTestId('image-generation-clear')).toBeEnabled();
+
+    await page.getByTestId('image-generation-clear').click();
+    const confirmDialog = page.getByRole('dialog');
+    await expect(confirmDialog).toBeVisible();
+    await confirmDialog.getByRole('button', { name: 'Clear', exact: true }).click();
+
+    await expect(page.getByTestId('image-generation-relay-base-url')).toHaveValue('');
+    await expect(page.getByTestId('image-generation-clear')).toBeDisabled();
+  });
 });
--- a/tests/unit/chat-assistant-media-display.test.ts
+++ b/tests/unit/chat-assistant-media-display.test.ts
@@ -0,0 +1,25 @@
+import { describe, expect, it } from 'vitest';
+import { extractText } from '@/pages/Chat/message-utils';
+
+describe('assistant media path display cleanup', () => {
+  it('strips bare OpenClaw media paths when the image is shown as an attachment card', () => {
+    const text = [
+      '宇航员图片生成完成啦 🧑‍🚀✨',
+      '/Users/zhonghaolu/.openclaw/media/tool-image-generation/clawx-image-1---82d6c7e6-ea44-4850-a24b-9e88e1660683.png',
+    ].join('\n');
+
+    expect(extractText({ role: 'assistant', content: text })).toBe('宇航员图片生成完成啦 🧑‍🚀✨');
+  });
+
+  it('still strips MEDIA: tagged OpenClaw artifact paths', () => {
+    const text = 'Done:\n\nMEDIA:/Users/alice/.openclaw/media/outbound/cat---abc.png';
+
+    expect(extractText({ role: 'assistant', content: text })).toBe('Done:');
+  });
+
+  it('strips markdown image syntax that cannot be rendered directly', () => {
+    const text = '宇航员图片完成啦 🧑‍🚀✨\n\n![Astronaut with Milky Way in helmet visor](/api/chat/media/outgoing/agent%3Amain%3As-1/abc/full)';
+
+    expect(extractText({ role: 'assistant', content: text })).toBe('宇航员图片完成啦 🧑‍🚀✨');
+  });
+});
--- a/tests/unit/chat-copy-image.test.ts
+++ b/tests/unit/chat-copy-image.test.ts
@@ -0,0 +1,56 @@
+import { beforeEach, describe, expect, it, vi } from 'vitest';
+import { copyImageToClipboard } from '@/pages/Chat/copy-image';
+
+const readBinaryFileMock = vi.fn();
+
+vi.mock('@/lib/api-client', () => ({
+  readBinaryFile: (...args: unknown[]) => readBinaryFileMock(...args),
+}));
+
+describe('copyImageToClipboard', () => {
+  beforeEach(() => {
+    readBinaryFileMock.mockReset();
+    class MockClipboardItem {
+      constructor(public items: Record<string, Blob>) {}
+    }
+    Object.assign(globalThis, { ClipboardItem: MockClipboardItem });
+    Object.assign(navigator, {
+      clipboard: {
+        write: vi.fn(async () => undefined),
+        writeText: vi.fn(async () => undefined),
+      },
+    });
+  });
+
+  it('copies image bytes from a data URL preview', async () => {
+    const pngBytes = Uint8Array.from([137, 80, 78, 71]);
+    const base64 = btoa(String.fromCharCode(...pngBytes));
+    const preview = `data:image/png;base64,${base64}`;
+
+    const ok = await copyImageToClipboard({
+      preview,
+      mimeType: 'image/png',
+    });
+
+    expect(ok).toBe(true);
+    expect(navigator.clipboard.write).toHaveBeenCalledTimes(1);
+    expect(readBinaryFileMock).not.toHaveBeenCalled();
+  });
+
+  it('copies image bytes from a local file path', async () => {
+    readBinaryFileMock.mockResolvedValueOnce({
+      ok: true,
+      data: Uint8Array.from([1, 2, 3]),
+      mimeType: 'image/png',
+    });
+
+    const ok = await copyImageToClipboard({
+      filePath: '/tmp/cat.png',
+      mimeType: 'image/png',
+    });
+
+    expect(ok).toBe(true);
+    expect(readBinaryFileMock).toHaveBeenCalledWith('/tmp/cat.png');
+    expect(navigator.clipboard.write).toHaveBeenCalledTimes(1);
+  });
+});
--- a/tests/unit/chat-helpers-enrichment.test.ts
+++ b/tests/unit/chat-helpers-enrichment.test.ts
@@ -1,6 +1,7 @@
 import { describe, expect, it, vi } from 'vitest';
 import {
  enrichWithToolResultFiles,
+  enrichWithToolCallAttachments,
  enrichWithCachedImages,
 } from '@/stores/chat/helpers';
 import type { RawMessage } from '@/stores/chat';
@@ -137,6 +138,38 @@ describe('enrichWithToolResultFiles', () => {
    const paths = (reply._attachedFiles ?? []).map((f) => f.filePath);
    expect(paths).toContain('/tmp/foo.pdf');
  });
+
+  it('skips internal NO_REPLY turns when attaching pending tool-result files', () => {
+    const messages: RawMessage[] = [
+      {
+        role: 'assistant',
+        id: 'a1',
+        content: [{ type: 'toolCall', id: 'tc1', name: 'exec', input: { command: 'echo report > /tmp/report.pdf' } }],
+      },
+      {
+        role: 'toolresult',
+        id: 't1',
+        toolCallId: 'tc1',
+        toolName: 'exec',
+        content: [{ type: 'text', text: 'Wrote /tmp/report.pdf' }],
+      },
+      {
+        role: 'assistant',
+        id: 'no-reply',
+        content: [{ type: 'text', text: 'NO_REPLY' }],
+      },
+      {
+        role: 'assistant',
+        id: 'final',
+        content: [{ type: 'text', text: 'Report is ready.' }],
+      },
+    ];
+
+    const enriched = enrichWithToolResultFiles(messages);
+    const final = enriched.find((m) => m.id === 'final')!;
+    expect(final._attachedFiles?.map((file) => file.filePath)).toEqual(['/tmp/report.pdf']);
+    expect(enriched.find((m) => m.id === 'no-reply')?._attachedFiles ?? []).toEqual([]);
+  });
 });

 describe('enrichWithCachedImages — Gateway media bubble dedup', () => {
@@ -273,4 +306,68 @@ describe('enrichWithCachedImages — Gateway media bubble dedup', () => {
    const replyPaths = (reply._attachedFiles ?? []).map((f) => f.filePath);
    expect(replyPaths).toEqual(['/tmp/foo.png']);
  });
+
+  it('promotes markdown local image paths to attached files', () => {
+    const messages: RawMessage[] = [
+      {
+        role: 'assistant',
+        id: 'reply',
+        content: [{
+          type: 'text',
+          text: '宇航员图片完成啦 🧑‍🚀✨\n\n![Astronaut with Milky Way in helmet visor](/Users/me/.openclaw/media/tool-image-generation/cat.png)',
+        }],
+      },
+    ];
+
+    const enriched = enrichWithCachedImages(messages);
+    expect(enriched[0]?._attachedFiles?.map((file) => file.filePath)).toEqual([
+      '/Users/me/.openclaw/media/tool-image-generation/cat.png',
+    ]);
+  });
+
+  it('promotes markdown gateway image URLs to gateway-media attachments', () => {
+    const messages: RawMessage[] = [
+      {
+        role: 'assistant',
+        id: 'reply',
+        content: [{
+          type: 'text',
+          text: 'Done\n\n![Astronaut with Milky Way in helmet visor](/api/chat/media/outgoing/agent%3Amain%3As-1/abc/full)',
+        }],
+      },
+    ];
+
+    const enriched = enrichWithCachedImages(messages);
+    expect(enriched[0]?._attachedFiles?.[0]).toMatchObject({
+      gatewayUrl: '/api/chat/media/outgoing/agent%3Amain%3As-1/abc/full',
+      source: 'gateway-media',
+      fileName: 'Astronaut with Milky Way in helmet visor',
+    });
+  });
+});
+
+describe('enrichWithToolCallAttachments', () => {
+  it('attaches image paths from message tool attachments array', () => {
+    const messages: RawMessage[] = [
+      {
+        role: 'assistant',
+        id: 'send-image',
+        content: [{
+          type: 'tool_use',
+          id: 'tool-1',
+          name: 'message',
+          input: {
+            action: 'send',
+            attachments: [{ filePath: '/Users/me/.openclaw/media/tool-image-generation/cat.png' }],
+          },
+        }],
+      },
+    ];
+
+    const enriched = enrichWithToolCallAttachments(messages);
+    expect(enriched[0]?._attachedFiles?.map((file) => file.filePath)).toEqual([
+      '/Users/me/.openclaw/media/tool-image-generation/cat.png',
+    ]);
+    expect(enriched[0]?._attachedFiles?.[0]?.mimeType).toBe('image/png');
+  });
 });
--- a/tests/unit/chat-history-actions.test.ts
+++ b/tests/unit/chat-history-actions.test.ts
@@ -7,6 +7,7 @@ const gatewayStoreGetStateMock = vi.fn();
 const clearHistoryPoll = vi.fn();
 const enrichWithCachedImages = vi.fn((messages) => messages);
 const enrichWithToolResultFiles = vi.fn((messages) => messages);
+const enrichWithToolCallAttachments = vi.fn((messages) => messages);
 const getMessageErrorMessage = vi.fn((message: { errorMessage?: string; error_message?: string } | undefined) => {
  if (!message) return null;
  return message.errorMessage ?? message.error_message ?? null;
@@ -29,6 +30,23 @@ const isInternalMessage = vi.fn((msg: { role?: unknown; content?: unknown }) =>
  }
  return false;
 });
+const messageHasToolUse = (msg: { role?: unknown; content?: unknown; tool_calls?: unknown; toolCalls?: unknown }) => {
+  if (msg.role !== 'assistant') return false;
+  if (Array.isArray(msg.content)) {
+    return (msg.content as Array<{ type?: string }>).some(
+      (block) => block.type === 'tool_use' || block.type === 'toolCall',
+    );
+  }
+  const toolCalls = (msg as { tool_calls?: unknown; toolCalls?: unknown }).tool_calls
+    ?? (msg as { toolCalls?: unknown }).toolCalls;
+  return Array.isArray(toolCalls) && toolCalls.length > 0;
+};
+const shouldDropMessageFromHistory = vi.fn((msg: { role?: unknown; content?: unknown; tool_calls?: unknown; toolCalls?: unknown }) => {
+  if (isToolResultRole(msg.role)) return true;
+  if (messageHasToolUse(msg)) return false;
+  return isInternalMessage(msg);
+});
+const setLastChatEventAt = vi.fn();
 const loadMissingPreviews = vi.fn(async () => false);
 const toMs = vi.fn((ts: number) => ts < 1e12 ? ts * 1000 : ts);

@@ -50,6 +68,7 @@ vi.mock('@/stores/chat/helpers', () => ({
  clearHistoryPoll: (...args: unknown[]) => clearHistoryPoll(...args),
  enrichWithCachedImages: (...args: unknown[]) => enrichWithCachedImages(...args),
  enrichWithToolResultFiles: (...args: unknown[]) => enrichWithToolResultFiles(...args),
+  enrichWithToolCallAttachments: (...args: unknown[]) => enrichWithToolCallAttachments(...args),
  getLatestOptimisticUserMessage: (messages: Array<{ role: string; timestamp?: number }>, userTimestampMs: number) =>
    [...messages].reverse().find(
      (message) => message.role === 'user'
@@ -57,9 +76,13 @@ vi.mock('@/stores/chat/helpers', () => ({
    ),
  getMessageText: (...args: unknown[]) => getMessageText(...args),
  hasNonToolAssistantContent: (...args: unknown[]) => hasNonToolAssistantContent(...args),
+  hasAssistantAfterLastRealUser: (messages: Array<{ role?: string }>) =>
+    messages.some((message) => message.role === 'assistant'),
  isInternalMessage: (...args: unknown[]) => isInternalMessage(...args),
+  shouldDropMessageFromHistory: (...args: unknown[]) => shouldDropMessageFromHistory(...args),
  isToolResultRole: (...args: unknown[]) => isToolResultRole(...args),
  loadMissingPreviews: (...args: unknown[]) => loadMissingPreviews(...args),
+  setLastChatEventAt: (...args: unknown[]) => setLastChatEventAt(...args),
  mergePendingOptimisticUserMessages: (_sessionKey: string, messages: unknown[]) => messages,
  hasOptimisticServerEcho: (
    loadedMessages: Array<{ role: string; timestamp?: number; content?: unknown; _attachedFiles?: Array<{ filePath?: string; fileName?: string; mimeType?: string; fileSize?: number }> }>,
--- a/tests/unit/chat-internal-message-filter.test.ts
+++ b/tests/unit/chat-internal-message-filter.test.ts
@@ -1,5 +1,5 @@
 import { describe, expect, it } from 'vitest';
-import { isInternalMessage } from '@/stores/chat/helpers';
+import { isInternalMessage, shouldDropMessageFromHistory } from '@/stores/chat/helpers';

 describe('chat internal message filter', () => {
  it('filters runtime system injection bundle like async exec completion payload', () => {
@@ -33,4 +33,54 @@ describe('chat internal message filter', () => {
  it('filters OpenClaw heartbeat poll user turns', () => {
    expect(isInternalMessage({ role: 'user', content: '[OpenClaw heartbeat poll]' })).toBe(true);
  });
+
+  it('filters inter-session routing payloads', () => {
+    expect(isInternalMessage({
+      role: 'user',
+      content: '[Inter-session message] sourceSession=image_generate:abc sourceTool=image_generate',
+    })).toBe(true);
+  });
+
+  it('filters OpenClaw runtime continuation prompts', () => {
+    expect(isInternalMessage({ role: 'user', content: 'Continue the OpenClaw runtime event.' })).toBe(true);
+  });
+
+  it('filters runtime continuation prompts stored on msg.text', () => {
+    expect(isInternalMessage({
+      role: 'user',
+      content: [],
+      text: 'Continue the OpenClaw runtime event.',
+    })).toBe(true);
+  });
+
+  it('filters interim image-generation status narration', () => {
+    expect(isInternalMessage({
+      role: 'assistant',
+      content: '生成中，稍等 👨‍🚀',
+    })).toBe(true);
+  });
+
+  it('keeps assistant tool-call turns in history even when narration is internal', () => {
+    expect(shouldDropMessageFromHistory({
+      role: 'assistant',
+      content: [
+        { type: 'text', text: '生成中，稍等 👨‍🚀' },
+        { type: 'tool_use', id: 'tool-image', name: 'image_generate', input: { prompt: 'astronaut' } },
+      ],
+    })).toBe(false);
+    expect(isInternalMessage({
+      role: 'assistant',
+      content: [
+        { type: 'text', text: '生成中，稍等 👨‍🚀' },
+        { type: 'tool_use', id: 'tool-image', name: 'image_generate', input: { prompt: 'astronaut' } },
+      ],
+    })).toBe(true);
+  });
+
+  it('filters assistant turns that only contain chain-of-thought', () => {
+    expect(isInternalMessage({
+      role: 'assistant',
+      content: [{ type: 'thinking', thinking: 'The user is asking me to continue with an OpenClaw runtime event.' }],
+    })).toBe(true);
+  });
 });
--- a/tests/unit/chat-message.test.tsx
+++ b/tests/unit/chat-message.test.tsx
@@ -5,6 +5,7 @@ import type { RawMessage } from '@/stores/chat';

 vi.mock('@/lib/api-client', () => ({
  invokeIpc: vi.fn(),
+  readBinaryFile: vi.fn(),
  statFile: vi.fn(async (path: string) => {
    if (path.includes('missing') || path.includes('不存在')) {
      return { ok: false, error: 'notFound' };
@@ -405,3 +406,50 @@ describe('ChatMessage reply styling', () => {
    expect(bubble).toHaveTextContent('Keep the prompt bubble.');
  });
 });
+
+describe('ChatMessage image copy', () => {
+  beforeEach(() => {
+    class MockClipboardItem {
+      constructor(public items: Record<string, Blob>) {}
+    }
+    Object.assign(globalThis, { ClipboardItem: MockClipboardItem });
+    Object.assign(navigator, {
+      clipboard: {
+        write: vi.fn(async () => undefined),
+        writeText: vi.fn(async () => undefined),
+      },
+    });
+  });
+
+  it('copies image bytes instead of the media URL text when an image attachment is present', async () => {
+    const { readBinaryFile } = await import('@/lib/api-client');
+    vi.mocked(readBinaryFile).mockResolvedValueOnce({
+      ok: true,
+      data: Uint8Array.from([137, 80, 78, 71]),
+      mimeType: 'image/png',
+    });
+
+    const message: RawMessage = {
+      role: 'assistant',
+      content: 'http://127.0.0.1:18789/api/chat/media/outgoing/agent/main/full',
+      _attachedFiles: [
+        {
+          fileName: 'cat.png',
+          mimeType: 'image/png',
+          fileSize: 1234,
+          preview: null,
+          filePath: '/tmp/cat.png',
+          source: 'tool-result',
+        },
+      ],
+    };
+
+    render(<ChatMessage message={message} />);
+
+    fireEvent.click(screen.getByRole('button'));
+    await vi.waitFor(() => {
+      expect(navigator.clipboard.write).toHaveBeenCalledTimes(1);
+    });
+    expect(navigator.clipboard.writeText).not.toHaveBeenCalled();
+  });
+});
--- a/tests/unit/chat-user-message-display.test.ts
+++ b/tests/unit/chat-user-message-display.test.ts
@@ -0,0 +1,58 @@
+import { describe, expect, it } from 'vitest';
+import { extractText } from '@/pages/Chat/message-utils';
+import { matchesOptimisticUserMessage } from '@/stores/chat/helpers';
+
+const gatewayImageEcho = [
+  'Sender (untrusted metadata):',
+  '```json',
+  '{',
+  '  "label": "ClawX (gateway-client)",',
+  '  "id": "gateway-client",',
+  '  "name": "ClawX",',
+  '  "username": "ClawX"',
+  '}',
+  '```',
+  '',
+  '[media attached: media://inbound/image---abc.png (image/png)]',
+  '[Image]',
+  'User text:',
+  'Process the attached file(s).',
+  '[media attached: /Users/test/.openclaw/media/outbound/out.png (image/png) | /Users/test/.openclaw/media/outbound/out.png]',
+  'Description:',
+  'An astronaut in a white space suit floats in space, reaching a gloved hand toward the viewer.',
+].join('\n');
+
+describe('user message display cleanup', () => {
+  it('hides the inbound-image vision envelope for attachment-only uploads', () => {
+    expect(extractText({ role: 'user', content: gatewayImageEcho })).toBe('');
+  });
+
+  it('keeps the user caption while stripping auto-generated description', () => {
+    const content = gatewayImageEcho
+      .replace('Process the attached file(s).', '改成西装加领带');
+
+    expect(extractText({ role: 'user', content })).toBe('改成西装加领带');
+  });
+
+  it('matches optimistic attachment-only bubbles against the gateway vision echo', () => {
+    const optimistic = {
+      role: 'user' as const,
+      content: '(file attached)',
+      timestamp: 1_700_000_000,
+      _attachedFiles: [{
+        fileName: 'out.png',
+        mimeType: 'image/png',
+        fileSize: 123,
+        preview: null,
+        filePath: '/Users/test/.openclaw/media/outbound/out.png',
+      }],
+    };
+    const candidate = {
+      role: 'user' as const,
+      content: gatewayImageEcho,
+      timestamp: 1_700_000_000,
+    };
+
+    expect(matchesOptimisticUserMessage(candidate, optimistic, 1_700_000_000_000)).toBe(true);
+  });
+});
--- a/tests/unit/image-generation-status.test.ts
+++ b/tests/unit/image-generation-status.test.ts
@@ -0,0 +1,136 @@
+import { describe, expect, it } from 'vitest';
+import type { RawMessage } from '@/stores/chat';
+import {
+  IMAGE_GENERATION_TIMEOUT_MS,
+  isImageGenerationPending,
+} from '@/pages/Chat/image-generation-status';
+
+const TASK_ID = '27443fdb-6cca-48e6-a3a7-ee34b0491aee';
+const NOW = 1_700_000_000_000;
+
+describe('isImageGenerationPending', () => {
+  it('returns true while waiting after image_generate without a final reply', () => {
+    const segmentMessages: RawMessage[] = [
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000,
+        content: [{ type: 'toolCall', id: 'call_1', name: 'image_generate', arguments: { prompt: 'astronaut' } }],
+      },
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000 + 5,
+        content: [{ type: 'text', text: '图片生成中，稍等片刻 🧑‍🚀✨' }],
+      },
+    ];
+
+    expect(isImageGenerationPending(segmentMessages, [], NOW)).toBe(true);
+  });
+
+  it('returns false after the inter-session completion event arrives', () => {
+    const segmentMessages: RawMessage[] = [
+      {
+        role: 'toolresult',
+        toolName: 'image_generate',
+        timestamp: NOW / 1000,
+        content: [{
+          type: 'text',
+          text: `Background task started for image generation (${TASK_ID}).`,
+        }],
+      },
+      {
+        role: 'user',
+        content: [{
+          type: 'text',
+          text: `[Inter-session message] sourceSession=image_generate:${TASK_ID} sourceTool=image_generate`,
+        }],
+      },
+    ];
+
+    expect(isImageGenerationPending(segmentMessages, [], NOW)).toBe(false);
+  });
+
+  it('returns false once a non-generating assistant reply arrives', () => {
+    const segmentMessages: RawMessage[] = [
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000,
+        content: [{ type: 'toolCall', id: 'call_1', name: 'image_generate', arguments: { prompt: 'astronaut' } }],
+      },
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000 + 5,
+        content: [{ type: 'text', text: '图片生成中，稍等片刻 🧑‍🚀✨' }],
+      },
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000 + 120,
+        content: [{ type: 'text', text: '抱歉，图片生成超时了，稍后重试一下吧 🧑‍🚀' }],
+      },
+    ];
+
+    expect(isImageGenerationPending(segmentMessages, [], NOW + 120_000)).toBe(false);
+  });
+
+  it('returns false after the configured timeout even if only generating narration exists', () => {
+    const segmentMessages: RawMessage[] = [
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000,
+        content: [{ type: 'toolCall', id: 'call_1', name: 'image_generate', arguments: { prompt: 'astronaut' } }],
+      },
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000 + 5,
+        content: [{ type: 'text', text: '图片生成中，稍等片刻 🧑‍🚀✨' }],
+      },
+    ];
+
+    expect(
+      isImageGenerationPending(
+        segmentMessages,
+        [],
+        NOW + IMAGE_GENERATION_TIMEOUT_MS + 20_000,
+      ),
+    ).toBe(false);
+  });
+
+  it('returns false when a stale streaming tool status is still marked running', () => {
+    expect(isImageGenerationPending([], [{
+      name: 'image_generate',
+      status: 'running',
+      updatedAt: NOW - IMAGE_GENERATION_TIMEOUT_MS - 60_000,
+    }], NOW)).toBe(false);
+  });
+
+  it('returns true while image_generate is actively running in streaming tools', () => {
+    expect(isImageGenerationPending([], [{
+      name: 'image_generate',
+      status: 'running',
+      updatedAt: NOW,
+    }], NOW)).toBe(true);
+  });
+
+  it('returns false once an image attachment is delivered', () => {
+    const segmentMessages: RawMessage[] = [
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000,
+        content: [{ type: 'toolCall', id: 'call_1', name: 'image_generate', arguments: { prompt: 'astronaut' } }],
+      },
+      {
+        role: 'assistant',
+        timestamp: NOW / 1000 + 90,
+        content: [{ type: 'text', text: '宇航员图片完成啦 🧑‍🚀✨' }],
+        _attachedFiles: [{
+          fileName: 'astronaut.png',
+          mimeType: 'image/png',
+          fileSize: 123,
+          preview: null,
+          filePath: '/tmp/astronaut.png',
+        }],
+      },
+    ];
+
+    expect(isImageGenerationPending(segmentMessages, [], NOW + 90_000)).toBe(false);
+  });
+});
--- a/tests/unit/task-visualization.test.ts
+++ b/tests/unit/task-visualization.test.ts
@@ -4,7 +4,7 @@ import { stripProcessMessagePrefix } from '@/pages/Chat/message-utils';
 import type { RawMessage, ToolStatus } from '@/stores/chat';

 describe('deriveTaskSteps', () => {
-  it('builds running steps from streaming thinking and tool status', () => {
+  it('builds running steps from streaming tool status without exposing chain-of-thought', () => {
    const streamingTools: ToolStatus[] = [
      {
        name: 'web_search',
@@ -27,12 +27,6 @@ describe('deriveTaskSteps', () => {
    });

    expect(steps).toEqual([
-      expect.objectContaining({
-        id: 'stream-thinking-0',
-        label: 'Thinking',
-        status: 'running',
-        kind: 'thinking',
-      }),
      expect.objectContaining({
        label: 'web_search',
        status: 'running',
@@ -160,7 +154,7 @@ describe('deriveTaskSteps', () => {
    }));
  });

-  it('keeps recent completed steps from assistant history', () => {
+  it('keeps recent completed tool steps from assistant history', () => {
    const messages: RawMessage[] = [
      {
        role: 'assistant',
@@ -179,12 +173,6 @@ describe('deriveTaskSteps', () => {
    });

    expect(steps).toEqual([
-      expect.objectContaining({
-        id: 'history-thinking-assistant-1-0',
-        label: 'Thinking',
-        status: 'completed',
-        kind: 'thinking',
-      }),
      expect.objectContaining({
        id: 'tool-2',
        label: 'read_file',
@@ -194,7 +182,7 @@ describe('deriveTaskSteps', () => {
    ]);
  });

-  it('splits cumulative streaming thinking into separate execution steps', () => {
+  it('does not expose streaming chain-of-thought in the execution graph', () => {
    const steps = deriveTaskSteps({
      messages: [],
      streamingMessage: {
@@ -208,21 +196,36 @@ describe('deriveTaskSteps', () => {
      streamingTools: [],
    });

+    expect(steps).toEqual([]);
+  });
+
+  it('skips internal assistant turns and hides NO_REPLY from the execution graph', () => {
+    const steps = deriveTaskSteps({
+      messages: [
+        {
+          role: 'assistant',
+          content: [{ type: 'thinking', thinking: 'Continue the OpenClaw runtime event internally.' }],
+        },
+        {
+          role: 'assistant',
+          content: [{ type: 'text', text: 'NO_REPLY' }],
+        },
+        {
+          role: 'assistant',
+          content: [
+            { type: 'tool_use', id: 'tool-image', name: 'image_generate', input: { prompt: 'astronaut' } },
+          ],
+        },
+      ],
+      streamingMessage: null,
+      streamingTools: [],
+    });
+
    expect(steps).toEqual([
      expect.objectContaining({
-        id: 'stream-thinking-0',
-        detail: 'Reviewing X.',
-        status: 'completed',
-      }),
-      expect.objectContaining({
-        id: 'stream-thinking-1',
-        detail: 'Comparing Y.',
-        status: 'completed',
-      }),
-      expect.objectContaining({
-        id: 'stream-thinking-2',
-        detail: 'Drafting answer.',
-        status: 'running',
+        id: 'tool-image',
+        label: 'image_generate',
+        kind: 'tool',
      }),
    ]);
  });