Add canvas agent mode and management

app/(admin)/settings/admin-settings-client.tsx

@@ -229,6 +229,7 @@ type SiteSettings = {
   tutorialEnabled: boolean;
   rankingEnabled: boolean;
   canvasEnabled: boolean;
+  agentEnabled: boolean;
   realtimeCanvasEnabled: boolean;
   aiDetailGenerationEnabled: boolean;
   canvasDefaultModelId: string | null;
@@ -618,6 +619,7 @@ export function AdminSettingsClient() {
     tutorialEnabled: false,
     rankingEnabled: true,
     canvasEnabled: true,
+    agentEnabled: true,
     realtimeCanvasEnabled: true,
     aiDetailGenerationEnabled: false,
     canvasDefaultModelId: null,
@@ -689,6 +691,7 @@ export function AdminSettingsClient() {
         tutorialEnabled: siteSettingsData.tutorialEnabled ?? false,
         rankingEnabled: siteSettingsData.rankingEnabled ?? true,
         canvasEnabled: siteSettingsData.canvasEnabled ?? true,
+        agentEnabled: (siteSettingsData as any).agentEnabled ?? true,
         realtimeCanvasEnabled: (siteSettingsData as any).realtimeCanvasEnabled ?? true,
         aiDetailGenerationEnabled: siteSettingsData.aiDetailGenerationEnabled ?? false,
         canvasDefaultModelId: siteSettingsData.canvasDefaultModelId || null,
@@ -5237,6 +5240,22 @@ export function AdminSettingsClient() {
                   </div>
                   {siteSettings.canvasEnabled && (
                     <>
+                      <div className="flex items-center justify-between rounded-lg border p-3">
+                        <div className="space-y-0.5">
+                          <Label htmlFor="agentEnabled">智能体功能</Label>
+                          <div className="text-xs text-muted-foreground">
+                            开启后，首页侧边栏显示“智能体”入口，用户可进入画布智能体模式；关闭后隐藏入口。
+                          </div>
+                        </div>
+                        <Switch
+                          id="agentEnabled"
+                          checked={siteSettings.agentEnabled}
+                          onCheckedChange={(checked) =>
+                            setSiteSettings({ ...siteSettings, agentEnabled: checked })
+                          }
+                        />
+                      </div>
+
                       <div className="flex items-center justify-between rounded-lg border p-3">
                         <div className="space-y-0.5">
                           <Label htmlFor="realtimeCanvasEnabled">实时画布功能</Label>

app/(chat)/api/chat/route.ts

@@ -8,6 +8,7 @@ import {
   stepCountIs,
   streamText,
 } from "ai";
+import { eq, sql } from "drizzle-orm";
 import { unstable_cache as cache } from "next/cache";
 import { after, NextResponse } from "next/server";
 import {
@@ -71,8 +72,10 @@ import {
   getAiProviderById,
   checkUserImageQuota,
   consumeUserPoints,
+  createAgentPresetExecutionSnapshot,
   createMessageTargetResolution,
   createMediaAssetRelation,
+  getAccessibleAgentPresetById,
   getConversationWorkingStateByChatId,
   getMediaAssetsByUserAndUrls,
   upsertConversationWorkingState,
@@ -80,7 +83,11 @@ import {
 } from "@/lib/db/queries";
 import { db } from "@/lib/db";
 import type { DBMessage } from "@/lib/db/schema";
-import { generatedImage, message as messageTable } from "@/lib/db/schema";
+import {
+  agentPreset as agentPresetTable,
+  generatedImage,
+  message as messageTable,
+} from "@/lib/db/schema";
 import { generateCanvasImage } from "@/lib/canvas/generate-image";
 import { resolveChatMediaInput } from "@/lib/ai/context-memory/media-input";
 import { resolveTargetAssets } from "@/lib/ai/context-memory/target-resolution";
@@ -415,6 +422,11 @@ function writeCanvasMediaChunk(
     videoUrls?: string[];
     prompt?: string | null;
     sourceImageUrls?: string[] | null;
+    agentPresetId?: string | null;
+    agentPresetName?: string | null;
+    agentCategoryId?: string | null;
+    agentCategoryName?: string | null;
+    agentExecutionSnapshotId?: string | null;
   },
   label: string,
 ) {
@@ -454,6 +466,12 @@ function writeCanvasMediaChunk(
             .map((url) => String(url || "").trim())
             .filter((url) => /^https?:\/\//i.test(url))
         : [],
+      agentPresetId: String(params.agentPresetId || "").trim() || undefined,
+      agentPresetName: String(params.agentPresetName || "").trim() || undefined,
+      agentCategoryId: String(params.agentCategoryId || "").trim() || undefined,
+      agentCategoryName: String(params.agentCategoryName || "").trim() || undefined,
+      agentExecutionSnapshotId:
+        String(params.agentExecutionSnapshotId || "").trim() || undefined,
     },
     label,
   );
@@ -2578,6 +2596,70 @@ async function saveGeneratedImages(
   }
 }
 
+type AgentPresetMessageImagePart = {
+  type: "file";
+  url: string;
+  name: string;
+  mediaType: string;
+  assetId?: string;
+};
+
+function extractCurrentMessageImageParts(
+  parts: Array<Record<string, any>>,
+): AgentPresetMessageImagePart[] {
+  return parts
+    .filter(
+      (part): part is AgentPresetMessageImagePart =>
+        part?.type === "file" &&
+        typeof part?.url === "string" &&
+        /^https?:\/\//i.test(part.url) &&
+        typeof part?.mediaType === "string" &&
+        part.mediaType.startsWith("image/")
+    )
+    .map((part) => ({
+      type: "file",
+      url: String(part.url).trim(),
+      name: String(part.name || "图片").trim() || "图片",
+      mediaType: String(part.mediaType || "image/jpeg").trim() || "image/jpeg",
+      assetId:
+        typeof part.assetId === "string" && part.assetId.trim() ? part.assetId.trim() : undefined,
+    }));
+}
+
+function normalizeAgentSupplementPrompt(text: string, presetName?: string | null) {
+  const normalized = String(text || "").trim();
+  if (!normalized) return "";
+  const escapedPresetName = String(presetName || "").replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+  const genericPattern = escapedPresetName
+    ? new RegExp(`^请按「${escapedPresetName}」智能体配置顺序处理这\\d+张图片。?$`)
+    : /^请按「.+」智能体配置顺序处理这\d+张图片。?$/;
+  if (genericPattern.test(normalized)) {
+    return "";
+  }
+  return normalized;
+}
+
+function buildAgentSlotPrompt(params: {
+  slotPrompt: string;
+  slotTitle?: string | null;
+  userSupplementPrompt?: string | null;
+}) {
+  const lines: string[] = [];
+  const slotTitle = String(params.slotTitle || "").trim();
+  if (slotTitle) {
+    lines.push(`【槽位】${slotTitle}`);
+  }
+  const slotPrompt = String(params.slotPrompt || "").trim();
+  if (slotPrompt) {
+    lines.push(slotPrompt);
+  }
+  const supplement = String(params.userSupplementPrompt || "").trim();
+  if (supplement) {
+    lines.push(`【补充要求】${supplement}`);
+  }
+  return lines.join("\n\n").trim();
+}
+
 const MODEL_UUID_REGEX =
   /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
 
@@ -2781,6 +2863,7 @@ export async function POST(request: Request) {
     const {
       id,
       message,
+      agentPresetId,
       selectedChatModel,
       selectedVisibilityType,
       hiddenPrompt,
@@ -2790,6 +2873,7 @@ export async function POST(request: Request) {
     }: {
       id: string;
       message: ChatMessage;
+      agentPresetId?: string;
       selectedChatModel: ChatModel["id"];
       selectedVisibilityType: VisibilityType;
       hiddenPrompt?: string;
@@ -4403,6 +4487,505 @@ export async function POST(request: Request) {
       return createSseResponse(clarificationStream);
     }
 
+    if (agentPresetId) {
+      const selectedAgentPreset = await getAccessibleAgentPresetById({
+        id: agentPresetId,
+        userId: session.user.id,
+      });
+
+      if (!selectedAgentPreset) {
+        return NextResponse.json(
+          {
+            code: "not_found:chat",
+            message: "智能体不存在或无权使用",
+          },
+          { status: 404 },
+        );
+      }
+
+      const orderedSlots = [...(selectedAgentPreset.slots || [])].sort(
+        (left, right) => left.slotIndex - right.slotIndex,
+      );
+      if (orderedSlots.length === 0) {
+        return NextResponse.json(
+          {
+            code: "bad_request:api",
+            message: "当前智能体未配置可执行槽位",
+          },
+          { status: 400 },
+        );
+      }
+
+      const uploadedImageParts = extractCurrentMessageImageParts(message.parts as any[]);
+      const requiredUploadCount = 3;
+
+      if (uploadedImageParts.length < requiredUploadCount) {
+        return NextResponse.json(
+          {
+            code: "bad_request:api",
+            message: `当前智能体需要上传 ${requiredUploadCount} 张图片`,
+          },
+          { status: 400 },
+        );
+      }
+
+      const presetModelId = String(selectedAgentPreset.defaultModelId || "").trim();
+      if (!presetModelId) {
+        return NextResponse.json(
+          {
+            code: "bad_request:api",
+            message: "当前智能体未配置默认模型",
+          },
+          { status: 400 },
+        );
+      }
+
+      const sanitizedUserSupplementPrompt = normalizeAgentSupplementPrompt(
+        requestTextContent,
+        selectedAgentPreset.name,
+      );
+
+      const presetSnapshot = await createAgentPresetExecutionSnapshot({
+        presetId: selectedAgentPreset.id,
+        userId: session.user.id,
+        projectId: contextPayload?.projectId || null,
+        chatId: id,
+        messageId: message.id,
+        presetName: selectedAgentPreset.name,
+        presetType: selectedAgentPreset.type,
+        visibility: selectedAgentPreset.visibility,
+        categoryId: selectedAgentPreset.category?.id || selectedAgentPreset.categoryId || null,
+        categoryName: selectedAgentPreset.category?.name || null,
+        promptTemplate: selectedAgentPreset.promptTemplate,
+        defaultModelId: selectedAgentPreset.defaultModelId || null,
+        resolvedModelId: presetModelId,
+        defaultParams: selectedAgentPreset.defaultParams || null,
+        slotCount: orderedSlots.length,
+        requiredUploadCount,
+        slots: orderedSlots.map((slot) => ({
+          slotIndex: slot.slotIndex,
+          title: slot.title || null,
+          promptTemplate: slot.promptTemplate,
+          defaultParams: slot.defaultParams || null,
+          requireUpload: slot.requireUpload !== false,
+          uploadLabel: slot.uploadLabel || null,
+          referenceAssetIds: slot.referenceAssetIds || [],
+          referenceImageUrls: (slot.referenceAssets || [])
+            .map((asset) => String(asset.originalUrl || "").trim())
+            .filter(Boolean),
+        })),
+        referenceAssetIds: orderedSlots.flatMap((slot) => slot.referenceAssetIds || []),
+        referenceImageUrls: orderedSlots.flatMap((slot) =>
+          (slot.referenceAssets || [])
+            .map((asset) => String(asset.originalUrl || "").trim())
+            .filter(Boolean),
+        ),
+        userInputText: sanitizedUserSupplementPrompt || null,
+        requestPayload: {
+          selectedChatModel,
+          agentPresetId,
+          requiredUploadCount,
+          uploadedImageUrls: uploadedImageParts.map((part) => part.url),
+        },
+      });
+
+      try {
+        await db
+          .update(agentPresetTable)
+          .set({
+            usageCount: sql`${agentPresetTable.usageCount} + 1`,
+            lastUsedAt: new Date(),
+            updatedAt: new Date(),
+          })
+          .where(eq(agentPresetTable.id, selectedAgentPreset.id));
+      } catch (agentUsageError) {
+        console.warn("[Chat API] Failed to update agent preset usage stats:", agentUsageError);
+      }
+
+      const agentSequenceStream = createUIMessageStream({
+        execute: async ({ writer }) => {
+          if (selectedAgentPreset.type === "image") {
+            writeAssistantTextChunk(
+              writer,
+              "<!-- IMAGE_GENERATION_START -->",
+              "agent-preset:image:start",
+            );
+            if (contextPayload?.source === "canvas") {
+              writeImageGenerationStatusChunk(
+                writer,
+                {
+                  phase: "processing",
+                  modelName:
+                    selectedAgentPreset.defaultModel?.displayName ||
+                    selectedAgentPreset.name,
+                  imageCount: orderedSlots.length,
+                },
+                "agent-preset:image:status-processing",
+              );
+            }
+          }
+
+          let successCount = 0;
+          const sharedUploadedImageUrls = uploadedImageParts
+            .map((part) => part.url)
+            .filter((url) => /^https?:\/\//i.test(url));
+          const sharedUploadedAssetIds = uploadedImageParts
+            .map((part) => part.assetId)
+            .filter((assetId): assetId is string => Boolean(assetId));
+
+          for (const [slotOrder, slot] of orderedSlots.entries()) {
+            const slotLabel =
+              String(slot.title || slot.uploadLabel || "").trim() || `第${slotOrder + 1}张`;
+            const slotReferenceImageUrls = Array.from(
+              new Set(
+                (slot.referenceAssets || [])
+                  .map((asset) => String(asset.originalUrl || "").trim())
+                  .filter(Boolean),
+              ),
+            );
+            const sourceImageUrls = Array.from(
+              new Set(
+                [
+                  ...sharedUploadedImageUrls,
+                  ...slotReferenceImageUrls,
+                ].filter(Boolean),
+              ),
+            );
+            const sourceAssetIds = Array.from(
+              new Set(
+                [
+                  ...sharedUploadedAssetIds,
+                  ...(slot.referenceAssetIds || []),
+                ].filter(Boolean),
+              ),
+            );
+            const slotPrompt = buildAgentSlotPrompt({
+              slotPrompt: slot.promptTemplate,
+              slotTitle: slot.title || slot.uploadLabel || null,
+              userSupplementPrompt: sanitizedUserSupplementPrompt || null,
+            });
+            const assistantMessageId = generateUUID();
+
+            try {
+              if (selectedAgentPreset.type === "image") {
+                const mergedImageParams = {
+                  ...(selectedAgentPreset.defaultParams || {}),
+                  ...(slot.defaultParams || {}),
+                } as Record<string, any>;
+                const result = await generateCanvasImage({
+                  prompt: slotPrompt || `生成${slotLabel}`,
+                  negativePrompt:
+                    typeof mergedImageParams.negativePrompt === "string"
+                      ? mergedImageParams.negativePrompt
+                      : undefined,
+                  model: presetModelId,
+                  aspectRatio:
+                    typeof mergedImageParams.aspectRatio === "string"
+                      ? mergedImageParams.aspectRatio
+                      : effectiveImageParams?.aspectRatio || defaultImageQualityParams.aspectRatio,
+                  imageSize:
+                    typeof mergedImageParams.imageSize === "string"
+                      ? mergedImageParams.imageSize
+                      : effectiveImageParams?.imageSize || defaultImageQualityParams.imageSize,
+                  imageCount: 1,
+                  referenceImages: sourceImageUrls,
+                });
+
+                const imageMarkdown = buildAgentToolImagesMarkdown(
+                  result.images.map((image) => ({
+                    originalUrl: image.originalUrl,
+                    thumbnailUrl: image.thumbnailUrl,
+                    imageSize:
+                      typeof mergedImageParams.imageSize === "string"
+                        ? mergedImageParams.imageSize
+                        : effectiveImageParams?.imageSize || defaultImageQualityParams.imageSize,
+                  })),
+                );
+                const assistantText = prependImageModelBadge(
+                  [`### ${slotLabel}`, imageMarkdown].filter(Boolean).join("\n\n"),
+                  result.model,
+                );
+
+                await saveMessages({
+                  messages: [
+                    {
+                      id: assistantMessageId,
+                      chatId: id,
+                      role: "assistant",
+                      parts: [{ type: "text", text: assistantText }],
+                      attachments: [],
+                      createdAt: new Date(),
+                      videoStorageStatus: null,
+                      temporaryVideoUrl: null,
+                      permanentVideoUrl: null,
+                      archivedAt: null,
+                      videoMetadata: null,
+                    },
+                  ],
+                });
+
+                await saveGeneratedImages(
+                  assistantText,
+                  session.user.id,
+                  id,
+                  assistantMessageId,
+                  slotPrompt || slotLabel,
+                  result.modelRef || presetModelId,
+                  {
+                    sourceAssetIds,
+                    sourceImageUrls,
+                    relationType: "generated_from",
+                    sourceType:
+                      sourceImageUrls.length > 0 ? "reference_generate" : "generated",
+                    projectId: contextPayload?.projectId || null,
+                  },
+                );
+
+                writeAssistantTextChunk(
+                  writer,
+                  assistantText,
+                  `agent-preset:image:slot-${slotOrder + 1}`,
+                  assistantMessageId,
+                );
+                if (contextPayload?.source === "canvas") {
+                  writeCanvasMediaChunk(
+                    writer,
+                    {
+                      messageId: assistantMessageId,
+                      imageUrls: result.images.map((image) => image.originalUrl),
+                      prompt: slotPrompt || slotLabel,
+                      sourceImageUrls,
+                      agentPresetId: selectedAgentPreset.id,
+                      agentPresetName: selectedAgentPreset.name,
+                      agentCategoryId:
+                        selectedAgentPreset.category?.id ||
+                        selectedAgentPreset.categoryId ||
+                        null,
+                      agentCategoryName: selectedAgentPreset.category?.name || null,
+                      agentExecutionSnapshotId: presetSnapshot.id,
+                    },
+                    `agent-preset:image:slot-${slotOrder + 1}:canvas-media`,
+                  );
+                }
+              } else {
+                const mergedVideoParams = {
+                  ...(selectedAgentPreset.defaultParams || {}),
+                  ...(slot.defaultParams || {}),
+                } as Record<string, any>;
+                const requestedMode = String(mergedVideoParams.mode || "").trim() as
+                  | "text-to-video"
+                  | "first-frame"
+                  | "first-last-frame"
+                  | "reference-images";
+                const mode =
+                  requestedMode ||
+                  (sourceImageUrls.length >= 2
+                    ? "first-last-frame"
+                    : sourceImageUrls.length >= 1
+                      ? "first-frame"
+                      : "text-to-video");
+                const directVideoRatio =
+                  mergedVideoParams.ratio === "16:9" ||
+                  mergedVideoParams.ratio === "4:3" ||
+                  mergedVideoParams.ratio === "1:1" ||
+                  mergedVideoParams.ratio === "3:4" ||
+                  mergedVideoParams.ratio === "9:16" ||
+                  mergedVideoParams.ratio === "21:9" ||
+                  mergedVideoParams.ratio === "adaptive"
+                    ? mergedVideoParams.ratio
+                    : undefined;
+                const images =
+                  mode === "first-last-frame"
+                    ? sourceImageUrls.slice(0, 2)
+                    : mode === "first-frame"
+                      ? sourceImageUrls.slice(0, 1)
+                      : mode === "reference-images"
+                        ? sourceImageUrls.slice(0, 4)
+                        : [];
+
+                const result = await executeDirectVideoGeneration({
+                  prompt: slotPrompt || `生成${slotLabel}`,
+                  mode,
+                  images,
+                  resolution: mergedVideoParams.resolution,
+                  ratio: directVideoRatio,
+                  duration: mergedVideoParams.duration,
+                  preferredModelRef: presetModelId,
+                });
+
+                const videoUrl = String(result.videoUrl || "").trim();
+                if (!videoUrl) {
+                  throw new Error("视频生成完成但未返回视频地址");
+                }
+
+                const assistantText = prependImageModelBadge(
+                  [
+                    `### ${slotLabel}`,
+                    buildAgentToolVideoMarkdown(videoUrl),
+                  ]
+                    .filter(Boolean)
+                    .join("\n\n"),
+                  result.modelDisplayName,
+                );
+
+                await saveMessages({
+                  messages: [
+                    {
+                      id: assistantMessageId,
+                      chatId: id,
+                      role: "assistant",
+                      parts: [{ type: "text", text: assistantText }],
+                      attachments: [],
+                      createdAt: new Date(),
+                      videoStorageStatus: "temporary",
+                      temporaryVideoUrl: videoUrl,
+                      permanentVideoUrl: null,
+                      archivedAt: null,
+                      videoMetadata: (result.metadata || {}) as any,
+                    },
+                  ],
+                });
+
+                try {
+                  await upsertConversationWorkingState({
+                    userId: session.user.id,
+                    chatId: id,
+                    projectId: contextPayload?.projectId || null,
+                    currentIntent: "video_gen",
+                    lastMessageId: assistantMessageId,
+                  });
+                } catch (memoryError) {
+                  console.warn(
+                    "[Chat API] Failed to update working state after agent preset video generation:",
+                    memoryError,
+                  );
+                }
+
+                await consumeGeneratedVideoPoints({
+                  userId: session.user.id,
+                  chatId: id,
+                  messageId: assistantMessageId,
+                  modelRefs: [result.modelRef || presetModelId],
+                });
+
+                scheduleVideoArchive({
+                  chatId: id,
+                  messageId: assistantMessageId,
+                  text: assistantText,
+                  temporaryVideoUrl: videoUrl,
+                  taskId: assistantMessageId,
+                  metadata: result.metadata,
+                });
+
+                writeAssistantTextChunk(
+                  writer,
+                  assistantText,
+                  `agent-preset:video:slot-${slotOrder + 1}`,
+                  assistantMessageId,
+                );
+                if (contextPayload?.source === "canvas") {
+                  writeCanvasMediaChunk(
+                    writer,
+                    {
+                      messageId: assistantMessageId,
+                      videoUrls: [videoUrl],
+                      prompt: slotPrompt || slotLabel,
+                      sourceImageUrls,
+                      agentPresetId: selectedAgentPreset.id,
+                      agentPresetName: selectedAgentPreset.name,
+                      agentCategoryId:
+                        selectedAgentPreset.category?.id ||
+                        selectedAgentPreset.categoryId ||
+                        null,
+                      agentCategoryName: selectedAgentPreset.category?.name || null,
+                      agentExecutionSnapshotId: presetSnapshot.id,
+                    },
+                    `agent-preset:video:slot-${slotOrder + 1}:canvas-media`,
+                  );
+                }
+              }
+
+              successCount += 1;
+            } catch (slotError) {
+              console.error("[Chat API] Agent preset slot execution failed:", {
+                slotIndex: slot.slotIndex,
+                slotLabel,
+                error: slotError,
+              });
+              const errorMessage = `### ${slotLabel}\n\n执行失败：${
+                slotError instanceof Error ? slotError.message : "请稍后重试"
+              }`;
+              await saveMessages({
+                messages: [
+                  {
+                    id: assistantMessageId,
+                    chatId: id,
+                    role: "assistant",
+                    parts: [{ type: "text", text: errorMessage }],
+                    attachments: [],
+                    createdAt: new Date(),
+                    videoStorageStatus: null,
+                    temporaryVideoUrl: null,
+                    permanentVideoUrl: null,
+                    archivedAt: null,
+                    videoMetadata: null,
+                  },
+                ],
+              });
+              writeAssistantTextChunk(
+                writer,
+                errorMessage,
+                `agent-preset:error:slot-${slotOrder + 1}`,
+                assistantMessageId,
+              );
+            }
+          }
+
+          try {
+            await upsertConversationWorkingState({
+              userId: session.user.id,
+              chatId: id,
+              projectId: contextPayload?.projectId || null,
+              currentIntent: selectedAgentPreset.type === "video" ? "video_gen" : "image_gen",
+              lastMessageId: message.id,
+              lastTurnSummary: {
+                text: requestTextContent.slice(0, 500),
+                agentPresetId: selectedAgentPreset.id,
+                agentPresetName: selectedAgentPreset.name,
+                slotCount: orderedSlots.length,
+                successCount,
+              },
+            });
+          } catch (agentPresetMemoryError) {
+            console.warn("[Chat API] Failed to persist agent preset working state:", agentPresetMemoryError);
+          }
+
+          if (selectedAgentPreset.type === "image") {
+            if (contextPayload?.source === "canvas") {
+              writeImageGenerationStatusChunk(
+                writer,
+                {
+                  phase: successCount > 0 ? "completed" : "failed",
+                  modelName:
+                    selectedAgentPreset.defaultModel?.displayName ||
+                    selectedAgentPreset.name,
+                  imageCount: successCount,
+                },
+                "agent-preset:image:status-finished",
+              );
+            }
+            writeAssistantTextChunk(
+              writer,
+              "<!-- IMAGE_GENERATION_END -->",
+              "agent-preset:image:end",
+            );
+          }
+        },
+      });
+
+      return createSseResponse(agentSequenceStream);
+    }
+
     const directAgentExecution = shouldRunAgentConversation
       ? resolveDirectAgentToolExecution(smartAutoExecution, smartAutoPlan)
       : null;

app/(chat)/api/chat/schema.ts

@@ -80,6 +80,7 @@ export const postRequestBodySchema = z.object({
   }),
   hiddenPrompt: z.string().max(4000).optional(),
   autoContext: z.string().max(4000).optional(),
+  agentPresetId: z.string().uuid().optional(),
   // 支持预定义模型名称或数据库中的模型 UUID
   selectedChatModel: z.string().min(1).max(128),
   selectedVisibilityType: z.enum(["public", "private"]),

app/(chat)/canvas/page.tsx

@@ -11,11 +11,16 @@ export default function CanvasRedirectPage() {
 
   useEffect(() => {
     const shouldCreateNew = searchParams.get("new") === "1";
+    const passthroughParams = new URLSearchParams(searchParams.toString());
+    passthroughParams.delete("new");
+    const passthroughQuery = passthroughParams.toString();
+    const buildCanvasUrl = (canvasId: string) =>
+      `/canvas/${canvasId}${passthroughQuery ? `?${passthroughQuery}` : ""}`;
 
     const openCanvas = async () => {
       if (shouldCreateNew) {
         const newId = crypto.randomUUID();
-        router.replace(`/canvas/${newId}`);
+        router.replace(buildCanvasUrl(newId));
         return;
       }
 
@@ -32,7 +37,7 @@ export default function CanvasRedirectPage() {
             });
             const latest = sorted[0];
             if (latest?.id) {
-              router.replace(`/canvas/${latest.id}`);
+              router.replace(buildCanvasUrl(latest.id));
               return;
             }
           }
@@ -42,7 +47,7 @@ export default function CanvasRedirectPage() {
       }
 
       const newId = crypto.randomUUID();
-      router.replace(`/canvas/${newId}`);
+      router.replace(buildCanvasUrl(newId));
     };
 
     void openCanvas();

app/(chat)/layout.tsx

@@ -71,6 +71,7 @@ async function SidebarWrapper({ children }: { children: React.ReactNode }) {
       }))}
       tutorialEnabled={siteSettings.tutorialEnabled ?? false}
       rankingEnabled={siteSettings.rankingEnabled ?? true}
+      agentEnabled={(siteSettings as any).agentEnabled ?? true}
     >
       {children}
     </ChatShell>

app/api/agent-categories/route.ts

@@ -0,0 +1,81 @@
+import { NextRequest, NextResponse } from "next/server";
+import { z } from "zod";
+import { auth } from "@/app/(auth)/auth";
+import { ChatSDKError } from "@/lib/errors";
+import {
+  createAgentCategory,
+  getVisibleAgentCategories,
+} from "@/lib/db/queries";
+
+const scopeSchema = z.enum(["all", "public", "private"]).default("public");
+
+const createCategorySchema = z.object({
+  name: z.string().trim().min(1).max(64),
+  scope: z.enum(["public", "private"]).default("public"),
+});
+
+export async function GET(request: NextRequest) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const { searchParams } = new URL(request.url);
+    const scope = scopeSchema.parse(searchParams.get("scope") || "public");
+    const categories = await getVisibleAgentCategories({
+      userId: session.user.id,
+      scope,
+    });
+
+    return NextResponse.json({ success: true, data: categories });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    if (error instanceof z.ZodError) {
+      return NextResponse.json(
+        { success: false, error: error.message },
+        { status: 400 }
+      );
+    }
+    return NextResponse.json(
+      { success: false, error: "获取智能体分类失败" },
+      { status: 500 }
+    );
+  }
+}
+
+export async function POST(request: NextRequest) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const body = await request.json();
+    const data = createCategorySchema.parse(body);
+
+    const category = await createAgentCategory({
+      userId: session.user.id,
+      name: data.name,
+      scope: data.scope,
+    });
+
+    return NextResponse.json({ success: true, data: category });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    if (error instanceof z.ZodError) {
+      return NextResponse.json(
+        { success: false, error: error.message },
+        { status: 400 }
+      );
+    }
+    return NextResponse.json(
+      { success: false, error: "创建智能体分类失败" },
+      { status: 500 }
+    );
+  }
+}

app/api/agent-execution-snapshots/[id]/route.ts

@@ -0,0 +1,36 @@
+import { NextRequest, NextResponse } from "next/server";
+import { auth } from "@/app/(auth)/auth";
+import { ChatSDKError } from "@/lib/errors";
+import { getAccessibleAgentPresetExecutionSnapshotById } from "@/lib/db/queries";
+
+export async function GET(
+  request: NextRequest,
+  { params }: { params: Promise<{ id: string }> }
+) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const { id } = await params;
+    const snapshot = await getAccessibleAgentPresetExecutionSnapshotById({
+      id,
+      userId: session.user.id,
+    });
+
+    if (!snapshot) {
+      return NextResponse.json({ success: false, error: "执行快照不存在" }, { status: 404 });
+    }
+
+    return NextResponse.json({ success: true, data: snapshot });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    return NextResponse.json(
+      { success: false, error: "获取执行快照失败" },
+      { status: 500 }
+    );
+  }
+}

app/api/agents/[id]/fork/route.ts

@@ -0,0 +1,47 @@
+import { NextRequest, NextResponse } from "next/server";
+import { z } from "zod";
+import { auth } from "@/app/(auth)/auth";
+import { ChatSDKError } from "@/lib/errors";
+import { forkAgentPreset } from "@/lib/db/queries";
+
+const forkSchema = z.object({
+  name: z.string().trim().min(1).max(128).optional(),
+});
+
+export async function POST(
+  request: NextRequest,
+  { params }: { params: Promise<{ id: string }> }
+) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const { id } = await params;
+    const body = await request.json().catch(() => ({}));
+    const data = forkSchema.parse(body);
+
+    const preset = await forkAgentPreset({
+      id,
+      userId: session.user.id,
+      name: data.name,
+    });
+
+    return NextResponse.json({ success: true, data: preset });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    if (error instanceof z.ZodError) {
+      return NextResponse.json(
+        { success: false, error: error.message },
+        { status: 400 }
+      );
+    }
+    return NextResponse.json(
+      { success: false, error: "复制智能体失败" },
+      { status: 500 }
+    );
+  }
+}

app/api/agents/[id]/route.ts

@@ -0,0 +1,154 @@
+import { NextRequest, NextResponse } from "next/server";
+import { z } from "zod";
+import { auth } from "@/app/(auth)/auth";
+import { ChatSDKError } from "@/lib/errors";
+import {
+  deleteAgentPresetById,
+  getAccessibleAgentPresetById,
+  updateAgentPreset,
+} from "@/lib/db/queries";
+
+const slotSchema = z.object({
+  title: z.string().trim().max(128).optional(),
+  promptTemplate: z.string().trim().min(1).max(8000),
+  defaultParams: z.record(z.string(), z.any()).nullable().optional(),
+  requireUpload: z.boolean().optional(),
+  uploadLabel: z.string().trim().max(128).optional(),
+  referenceAssetIds: z.array(z.string().uuid()).max(12).optional(),
+});
+
+const updateAgentSchema = z
+  .object({
+    name: z.string().trim().min(1).max(128).optional(),
+    type: z.enum(["image", "video"]).optional(),
+    visibility: z.enum(["public", "private"]).optional(),
+    categoryId: z.string().uuid().nullable().optional(),
+    newCategoryName: z.string().trim().min(1).max(64).optional(),
+    description: z.string().trim().max(2000).nullable().optional(),
+    defaultModelId: z.string().uuid().optional(),
+    defaultParams: z.record(z.string(), z.any()).nullable().optional(),
+    avatarAssetId: z.string().uuid().nullable().optional(),
+    slots: z.array(slotSchema).min(1).max(20).optional(),
+    status: z.enum(["active", "disabled", "hidden", "blocked"]).optional(),
+  })
+  .superRefine((value, ctx) => {
+    if (value.categoryId && value.newCategoryName) {
+      ctx.addIssue({
+        code: z.ZodIssueCode.custom,
+        message: "categoryId 和 newCategoryName 不能同时传入",
+        path: ["newCategoryName"],
+      });
+    }
+  });
+
+export async function GET(
+  request: NextRequest,
+  { params }: { params: Promise<{ id: string }> }
+) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const { id } = await params;
+    const preset = await getAccessibleAgentPresetById({
+      id,
+      userId: session.user.id,
+    });
+
+    if (!preset) {
+      return NextResponse.json({ success: false, error: "智能体不存在" }, { status: 404 });
+    }
+
+    return NextResponse.json({ success: true, data: preset });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    return NextResponse.json(
+      { success: false, error: "获取智能体失败" },
+      { status: 500 }
+    );
+  }
+}
+
+export async function PUT(
+  request: NextRequest,
+  { params }: { params: Promise<{ id: string }> }
+) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const { id } = await params;
+    const body = await request.json();
+    const data = updateAgentSchema.parse(body);
+
+    const preset = await updateAgentPreset({
+      id,
+      userId: session.user.id,
+      name: data.name,
+      type: data.type,
+      visibility: data.visibility,
+      categoryId: data.categoryId ?? undefined,
+      newCategoryName: data.newCategoryName,
+      description: data.description ?? undefined,
+      defaultModelId: data.defaultModelId,
+      defaultParams: data.defaultParams ?? undefined,
+      avatarAssetId: data.avatarAssetId ?? undefined,
+      slots: data.slots,
+      status: data.status,
+    });
+
+    return NextResponse.json({ success: true, data: preset });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    if (error instanceof z.ZodError) {
+      return NextResponse.json(
+        { success: false, error: error.message },
+        { status: 400 }
+      );
+    }
+    return NextResponse.json(
+      { success: false, error: "更新智能体失败" },
+      { status: 500 }
+    );
+  }
+}
+
+export async function DELETE(
+  request: NextRequest,
+  { params }: { params: Promise<{ id: string }> }
+) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const { id } = await params;
+    const deleted = await deleteAgentPresetById({
+      id,
+      userId: session.user.id,
+    });
+
+    if (!deleted) {
+      return NextResponse.json({ success: false, error: "智能体不存在" }, { status: 404 });
+    }
+
+    return NextResponse.json({ success: true, data: deleted });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    return NextResponse.json(
+      { success: false, error: "删除智能体失败" },
+      { status: 500 }
+    );
+  }
+}

app/api/agents/route.ts

@@ -0,0 +1,154 @@
+import { NextRequest, NextResponse } from "next/server";
+import { z } from "zod";
+import { auth } from "@/app/(auth)/auth";
+import { ChatSDKError } from "@/lib/errors";
+import {
+  createAgentPreset,
+  getVisibleAgentCategories,
+  getVisibleAgentPresets,
+} from "@/lib/db/queries";
+
+const listQuerySchema = z.object({
+  scope: z.enum(["all", "public", "private"]).default("all"),
+  type: z.enum(["image", "video"]).optional(),
+  categoryId: z.string().uuid().optional(),
+  keyword: z.string().trim().max(128).optional(),
+  sort: z.enum(["latest", "hottest", "most_used"]).default("latest"),
+});
+
+const slotSchema = z.object({
+  title: z.string().trim().max(128).optional(),
+  promptTemplate: z.string().trim().min(1).max(8000),
+  defaultParams: z.record(z.string(), z.any()).optional(),
+  requireUpload: z.boolean().optional(),
+  uploadLabel: z.string().trim().max(128).optional(),
+  referenceAssetIds: z.array(z.string().uuid()).max(12).optional(),
+});
+
+const createAgentSchema = z
+  .object({
+    name: z.string().trim().min(1).max(128),
+    type: z.enum(["image", "video"]),
+    visibility: z.enum(["public", "private"]).default("private"),
+    categoryId: z.string().uuid().optional(),
+    newCategoryName: z.string().trim().min(1).max(64).optional(),
+    description: z.string().trim().max(2000).optional(),
+    defaultModelId: z.string().uuid(),
+    defaultParams: z.record(z.string(), z.any()).optional(),
+    avatarAssetId: z.string().uuid().optional(),
+    slots: z.array(slotSchema).min(1).max(20),
+  })
+  .superRefine((value, ctx) => {
+    if (value.categoryId && value.newCategoryName) {
+      ctx.addIssue({
+        code: z.ZodIssueCode.custom,
+        message: "categoryId 和 newCategoryName 不能同时传入",
+        path: ["newCategoryName"],
+      });
+    }
+    if (value.visibility === "public" && !value.categoryId && !value.newCategoryName) {
+      ctx.addIssue({
+        code: z.ZodIssueCode.custom,
+        message: "公开智能体必须选择分类或新建分类",
+        path: ["categoryId"],
+      });
+    }
+  });
+
+export async function GET(request: NextRequest) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const { searchParams } = new URL(request.url);
+    const query = listQuerySchema.parse({
+      scope: searchParams.get("scope") || "all",
+      type: searchParams.get("type") || undefined,
+      categoryId: searchParams.get("categoryId") || undefined,
+      keyword: searchParams.get("keyword") || undefined,
+      sort: searchParams.get("sort") || "latest",
+    });
+
+    const [presets, categories] = await Promise.all([
+      getVisibleAgentPresets({
+        userId: session.user.id,
+        scope: query.scope,
+        type: query.type,
+        categoryId: query.categoryId,
+        keyword: query.keyword,
+        sort: query.sort,
+      }),
+      getVisibleAgentCategories({
+        userId: session.user.id,
+        scope: query.scope === "private" ? "private" : "all",
+      }),
+    ]);
+
+    return NextResponse.json({
+      success: true,
+      data: {
+        categories,
+        public: presets.filter((item) => item.visibility === "public"),
+        private: presets.filter((item) => item.visibility === "private"),
+      },
+    });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    if (error instanceof z.ZodError) {
+      return NextResponse.json(
+        { success: false, error: error.message },
+        { status: 400 }
+      );
+    }
+    return NextResponse.json(
+      { success: false, error: "获取智能体列表失败" },
+      { status: 500 }
+    );
+  }
+}
+
+export async function POST(request: NextRequest) {
+  try {
+    const session = await auth();
+    if (!session?.user?.id) {
+      return NextResponse.json({ success: false, error: "未登录" }, { status: 401 });
+    }
+
+    const body = await request.json();
+    const data = createAgentSchema.parse(body);
+
+    const preset = await createAgentPreset({
+      userId: session.user.id,
+      name: data.name,
+      type: data.type,
+      visibility: data.visibility,
+      categoryId: data.categoryId,
+      newCategoryName: data.newCategoryName,
+      description: data.description,
+      defaultModelId: data.defaultModelId,
+      defaultParams: data.defaultParams,
+      avatarAssetId: data.avatarAssetId,
+      slots: data.slots,
+    });
+
+    return NextResponse.json({ success: true, data: preset });
+  } catch (error) {
+    if (error instanceof ChatSDKError) {
+      return error.toResponse();
+    }
+    if (error instanceof z.ZodError) {
+      return NextResponse.json(
+        { success: false, error: error.message },
+        { status: 400 }
+      );
+    }
+    return NextResponse.json(
+      { success: false, error: "创建智能体失败" },
+      { status: 500 }
+    );
+  }
+}

app/api/site-settings/route.ts

@@ -336,6 +336,7 @@ export async function GET() {
         tutorialEnabled: false,
         rankingEnabled: true,
         canvasEnabled: true,
+        agentEnabled: true,
         realtimeCanvasEnabled: true,
         realtimeCanvasServiceEnabled: true,
         storageFailoverConfig,
@@ -424,6 +425,7 @@ const updateSchema = z.object({
   tutorialEnabled: z.boolean().optional(),
   rankingEnabled: z.boolean().optional(),
   canvasEnabled: z.boolean().optional(),
+  agentEnabled: z.boolean().optional(),
   realtimeCanvasEnabled: z.boolean().optional(),
   aiDetailGenerationEnabled: z.boolean().optional(),
   canvasDefaultModelId: z.string().nullable().optional(),
@@ -518,6 +520,7 @@ export async function PUT(request: NextRequest) {
         .update(siteSettings)
         .set({
           ...siteSettingsPayload,
+          agentEnabled: siteSettingsPayload.agentEnabled ?? existing.agentEnabled ?? true,
           updatedAt: new Date(),
           updatedBy: session.user.id,
         })
@@ -881,6 +884,7 @@ export async function PUT(request: NextRequest) {
           tutorialEnabled: siteSettingsPayload.tutorialEnabled ?? false,
           rankingEnabled: siteSettingsPayload.rankingEnabled ?? true,
           canvasEnabled: siteSettingsPayload.canvasEnabled ?? true,
+          agentEnabled: siteSettingsPayload.agentEnabled ?? true,
           realtimeCanvasEnabled: siteSettingsPayload.realtimeCanvasEnabled ?? true,
           canvasDefaultModelId: siteSettingsPayload.canvasDefaultModelId,
           canvasAutoModeDefaultModelId: siteSettingsPayload.canvasAutoModeDefaultModelId,

app/canvas/[id]/page.tsx

@@ -1,9 +1,10 @@
 "use client";
 
 import { useEffect, useState, useCallback, use, useRef } from "react";
-import { useRouter, useParams } from "next/navigation";
+import { useRouter, useParams, useSearchParams } from "next/navigation";
 import useSWR from "swr";
 import { 
+  AlertCircle,
   Save, 
   Undo2, 
   Redo2, 
@@ -56,6 +57,7 @@ import { ScrollArea } from "@/components/ui/scroll-area";
 import { toast } from "sonner";
 import { useCanvasStore, canvasSelectors } from "@/store/canvas";
 import { canvasService, imageGenerationService } from "@/services/canvas";
+import { AgentPresetPanel } from "@/components/agents/AgentPresetPanel";
 import { CanvasWorkspace } from "@/components/canvas";
 import { CanvasChatPanel } from "@/components/canvas/CanvasChatPanel";
 import { CanvasFileListPanel } from "@/components/canvas/CanvasFileListPanel";
@@ -88,8 +90,10 @@ const FLOATING_BOTTOM_INSET = 20;
 export default function CanvasEditorPage({ params }: PageProps) {
   // 使用 useParams 确保客户端导航时能正确获取更新的 id
   const routeParams = useParams();
+  const searchParams = useSearchParams();
   const id = (routeParams?.id as string) || use(params).id;
   const router = useRouter();
+  const mode = searchParams.get("mode");
   const goHome = useCallback(() => {
     if (typeof window !== "undefined") {
       window.location.assign("/");
@@ -138,6 +142,7 @@ export default function CanvasEditorPage({ params }: PageProps) {
     siteName?: string;
     siteLogo?: string | null;
     siteLogoDark?: string | null;
+    agentEnabled?: boolean;
     realtimeCanvasEnabled?: boolean;
     realtimeCanvasServiceEnabled?: boolean;
   }>("/api/site-settings", fetcher);
@@ -148,6 +153,11 @@ export default function CanvasEditorPage({ params }: PageProps) {
   const [aspectRatio, setAspectRatio] = useState("1:1");
   const [imageCount, setImageCount] = useState(1);
   const previousNodeIdsRef = useRef<Set<string>>(new Set());
+  const selectedAgentPresetId = searchParams.get("agentPresetId");
+  const requestedAgentMode = mode === "agent" || Boolean(selectedAgentPresetId);
+  const isAgentFeatureEnabled = siteSettingsData?.agentEnabled !== false;
+  const isAgentMode =
+    isAgentFeatureEnabled && requestedAgentMode;
 
   // 获取选中的图片节点
   const selectedImageNode = selectedNodeIds.length === 1 
@@ -155,12 +165,63 @@ export default function CanvasEditorPage({ params }: PageProps) {
     : null;
 
   const resolvedPanelWidth = viewportWidth > 0
-    ? Math.min(FLOATING_PANEL_WIDTH, Math.max(viewportWidth - DOCKED_PANEL_INSET, 0))
+    ? isAgentMode
+      ? Math.min(
+          520,
+          Math.max(Math.floor((viewportWidth - DOCKED_PANEL_INSET * 2) / 2), 340)
+        )
+      : Math.min(FLOATING_PANEL_WIDTH, Math.max(viewportWidth - DOCKED_PANEL_INSET, 0))
     : FLOATING_PANEL_WIDTH;
+  const agentPanelWidth = isAgentMode ? resolvedPanelWidth : 0;
   const panelOcclusionRight = isPanelOpen ? resolvedPanelWidth + DOCKED_PANEL_INSET : 0;
   const floatingToolbarOffset = isPanelOpen ? -(panelOcclusionRight / 2) : 0;
   const compactControlsLeft = FLOATING_UI_GAP;
 
+  const updateCanvasSearchParams = useCallback(
+    (updates: Record<string, string | null>) => {
+      const nextParams = new URLSearchParams(searchParams.toString());
+      Object.entries(updates).forEach(([key, value]) => {
+        if (value === null || value === "") {
+          nextParams.delete(key);
+        } else {
+          nextParams.set(key, value);
+        }
+      });
+
+      const nextQuery = nextParams.toString();
+      router.replace(nextQuery ? `/canvas/${id}?${nextQuery}` : `/canvas/${id}`);
+    },
+    [id, router, searchParams]
+  );
+
+  const handleSelectAgentPreset = useCallback(
+    (presetId: string | null) => {
+      updateCanvasSearchParams({
+        mode: presetId ? "agent" : "agent",
+        agentPresetId: presetId,
+      });
+      setIsPanelOpen(true);
+      setPanelMode("chat");
+    },
+    [updateCanvasSearchParams]
+  );
+
+  const handleToggleAgentMode = useCallback(() => {
+    if (isAgentMode) {
+      updateCanvasSearchParams({
+        mode: null,
+        agentPresetId: null,
+      });
+      return;
+    }
+
+    updateCanvasSearchParams({
+      mode: "agent",
+    });
+    setIsPanelOpen(true);
+    setPanelMode("chat");
+  }, [isAgentMode, updateCanvasSearchParams]);
+
   // 加载画布数据
   useEffect(() => {
     console.log('[Canvas] Loading project:', id);
@@ -202,6 +263,13 @@ export default function CanvasEditorPage({ params }: PageProps) {
     return () => window.removeEventListener("resize", updateViewportWidth);
   }, []);
 
+  useEffect(() => {
+    if (isAgentMode) {
+      setIsPanelOpen(true);
+      setPanelMode("chat");
+    }
+  }, [isAgentMode]);
+
   const persistCanvasState = useCallback(
     async (reason: "autosave" | "delete" | "beforeunload") => {
       const canvasData = getCanvasState();
@@ -669,6 +737,16 @@ export default function CanvasEditorPage({ params }: PageProps) {
           </Button>
         </div>
 
+        {/* Agent Disabled Notice */}
+        {requestedAgentMode && !isAgentFeatureEnabled ? (
+          <div className="absolute top-4 left-1/2 z-20 -translate-x-1/2">
+            <div className="inline-flex items-center gap-2 rounded-full border border-amber-200 bg-amber-50 px-4 py-2 text-sm text-amber-800 shadow-sm dark:border-amber-900/60 dark:bg-amber-950/80 dark:text-amber-200">
+              <AlertCircle className="h-4 w-4" />
+              <span>智能体功能未开启，请联系管理员在系统设置中开启。</span>
+            </div>
+          </div>
+        ) : null}
+
         {/* 2. Floating Actions (Top Right) */}
         {!isPanelOpen && (
           <div className="absolute z-20" style={{ top: FLOATING_PANEL_INSET, right: FLOATING_PANEL_INSET }}>
@@ -684,6 +762,26 @@ export default function CanvasEditorPage({ params }: PageProps) {
         )}
 
         {/* 4. Floating Sidebar (Right) */}
+        {isAgentMode ? (
+          <div
+            className="absolute z-10 transition-all duration-300 ease-in-out"
+            style={{
+              top: DOCKED_PANEL_INSET,
+              left: DOCKED_PANEL_INSET,
+              bottom: DOCKED_PANEL_INSET,
+              width: agentPanelWidth,
+              maxWidth: `calc(100vw - ${DOCKED_PANEL_INSET}px)`,
+            }}
+          >
+            <div className="h-full w-full overflow-hidden border-r bg-white dark:bg-zinc-950">
+              <AgentPresetPanel
+                selectedPresetId={selectedAgentPresetId}
+                onSelectPreset={handleSelectAgentPreset}
+              />
+            </div>
+          </div>
+        ) : null}
+
         <div className={cn(
             "absolute z-10 transition-all duration-300 ease-in-out",
             isPanelOpen ? "translate-x-0 opacity-100" : "translate-x-[120%] opacity-0 pointer-events-none"
@@ -704,6 +802,7 @@ export default function CanvasEditorPage({ params }: PageProps) {
                     chatId={`canvas-${id}`} 
                     onClose={() => setIsPanelOpen(false)} 
                     onOpenFileList={() => setPanelMode('fileList')}
+                    selectedAgentPresetId={isAgentFeatureEnabled ? selectedAgentPresetId : null}
                   />
                 </div>
               ) : (
@@ -754,6 +853,23 @@ export default function CanvasEditorPage({ params }: PageProps) {
               <TooltipContent side="top">平移 (H)</TooltipContent>
             </Tooltip>
 
+            <Tooltip>
+              <TooltipTrigger asChild>
+                <Button
+                  variant={isAgentMode ? "secondary" : "ghost"}
+                  size="icon"
+                  onClick={handleToggleAgentMode}
+                  className={cn(
+                    "h-9 w-9 rounded-xl",
+                    isAgentMode && "bg-primary text-primary-foreground hover:bg-primary/90"
+                  )}
+                >
+                  <Wand2 className="h-4 w-4" />
+                </Button>
+              </TooltipTrigger>
+              <TooltipContent side="top">智能体模式</TooltipContent>
+            </Tooltip>
+
             <Tooltip>
               <TooltipTrigger asChild>
                 <Button

components/agents/AgentCategoryTabs.tsx

@@ -0,0 +1,54 @@
+"use client";
+
+import { Badge } from "@/components/ui/badge";
+import { cn } from "@/lib/utils";
+import type { AgentCategory } from "@/types/agent";
+
+interface AgentCategoryTabsProps {
+  categories: AgentCategory[];
+  selectedCategoryId?: string | null;
+  onSelectCategory: (categoryId: string | null) => void;
+}
+
+export function AgentCategoryTabs({
+  categories,
+  selectedCategoryId,
+  onSelectCategory,
+}: AgentCategoryTabsProps) {
+  return (
+    <div className="flex flex-wrap gap-2">
+      <button
+        type="button"
+        onClick={() => onSelectCategory(null)}
+        className={cn(
+          "rounded-full border px-3 py-1 text-xs transition-colors",
+          !selectedCategoryId
+            ? "border-primary bg-primary/10 text-primary"
+            : "border-border bg-background text-muted-foreground hover:bg-muted/50"
+        )}
+      >
+        全部
+      </button>
+      {categories.map((category) => (
+        <button
+          key={category.id}
+          type="button"
+          onClick={() => onSelectCategory(category.id)}
+          className={cn(
+            "rounded-full border px-3 py-1 text-xs transition-colors",
+            selectedCategoryId === category.id
+              ? "border-primary bg-primary/10 text-primary"
+              : "border-border bg-background text-muted-foreground hover:bg-muted/50"
+          )}
+        >
+          <span>{category.name}</span>
+          {category.scope === "global" ? (
+            <Badge variant="outline" className="ml-2 px-1.5 py-0 text-[10px]">
+              全局
+            </Badge>
+          ) : null}
+        </button>
+      ))}
+    </div>
+  );
+}

components/agents/AgentExecutionSnapshotDialog.tsx

@@ -0,0 +1,155 @@
+"use client";
+
+import useSWR from "swr";
+import { Loader2 } from "lucide-react";
+import {
+  Dialog,
+  DialogContent,
+  DialogDescription,
+  DialogHeader,
+  DialogTitle,
+} from "@/components/ui/dialog";
+import { ScrollArea } from "@/components/ui/scroll-area";
+import { fetcher } from "@/lib/utils";
+import type { AgentExecutionSnapshot } from "@/types/agent";
+
+interface AgentExecutionSnapshotDialogProps {
+  snapshotId?: string | null;
+  open: boolean;
+  onOpenChange: (open: boolean) => void;
+}
+
+export function AgentExecutionSnapshotDialog({
+  snapshotId,
+  open,
+  onOpenChange,
+}: AgentExecutionSnapshotDialogProps) {
+  const { data, isLoading } = useSWR<{ success?: boolean; data?: AgentExecutionSnapshot }>(
+    open && snapshotId ? `/api/agent-execution-snapshots/${snapshotId}` : null,
+    fetcher
+  );
+
+  const snapshot = data?.success ? data.data : null;
+  const uploadedImageUrls = Array.isArray(snapshot?.requestPayload?.uploadedImageUrls)
+    ? snapshot?.requestPayload?.uploadedImageUrls
+        .map((item) => String(item || "").trim())
+        .filter(Boolean)
+    : [];
+
+  return (
+    <Dialog open={open} onOpenChange={onOpenChange}>
+      <DialogContent className="max-w-4xl bg-background">
+        <DialogHeader>
+          <DialogTitle>执行快照</DialogTitle>
+          <DialogDescription>
+            查看本次智能体执行时使用的素材图、槽位提示词、参考图与模型。
+          </DialogDescription>
+        </DialogHeader>
+
+        <ScrollArea className="max-h-[70vh] pr-3">
+          {isLoading ? (
+            <div className="flex items-center justify-center py-12">
+              <Loader2 className="h-5 w-5 animate-spin text-muted-foreground" />
+            </div>
+          ) : !snapshot ? (
+            <div className="rounded-xl border border-dashed p-6 text-sm text-muted-foreground">
+              暂无快照详情
+            </div>
+          ) : (
+            <div className="space-y-5">
+              <div className="grid grid-cols-1 gap-3 md:grid-cols-2">
+                <div className="rounded-xl border p-3">
+                  <div className="text-xs text-muted-foreground">智能体</div>
+                  <div className="mt-1 font-medium">{snapshot.presetName}</div>
+                  <div className="mt-1 text-sm text-muted-foreground">
+                    类型：{snapshot.presetType === "video" ? "视频" : "生图"}
+                  </div>
+                  <div className="text-sm text-muted-foreground">
+                    分类：{snapshot.categoryName || "未分类"}
+                  </div>
+                </div>
+                <div className="rounded-xl border p-3">
+                  <div className="text-xs text-muted-foreground">模型</div>
+                  <div className="mt-1 font-medium">
+                    {snapshot.resolvedModel?.displayName ||
+                      snapshot.defaultModel?.displayName ||
+                      "未记录"}
+                  </div>
+                  <div className="mt-1 text-sm text-muted-foreground">
+                    槽位：{snapshot.slotCount} 个
+                  </div>
+                  <div className="text-sm text-muted-foreground">
+                    素材图：{snapshot.requiredUploadCount} 张
+                  </div>
+                </div>
+              </div>
+
+              <div className="rounded-xl border p-3">
+                <div className="mb-2 text-sm font-medium">本次使用的 3 张素材图</div>
+                {uploadedImageUrls.length > 0 ? (
+                  <div className="flex flex-wrap gap-3">
+                    {uploadedImageUrls.map((url, index) => (
+                      <div key={`${url}-${index}`} className="space-y-1">
+                        <img
+                          src={url}
+                          alt={`素材图 ${index + 1}`}
+                          className="h-24 w-24 rounded-lg border object-cover"
+                        />
+                        <div className="text-center text-xs text-muted-foreground">
+                          {index === 0 ? "产品图1" : index === 1 ? "产品图2" : "模特图"}
+                        </div>
+                      </div>
+                    ))}
+                  </div>
+                ) : (
+                  <div className="text-sm text-muted-foreground">未记录素材图</div>
+                )}
+              </div>
+
+              {snapshot.userInputText ? (
+                <div className="rounded-xl border p-3">
+                  <div className="mb-2 text-sm font-medium">用户补充提示</div>
+                  <div className="whitespace-pre-wrap text-sm text-muted-foreground">
+                    {snapshot.userInputText}
+                  </div>
+                </div>
+              ) : null}
+
+              <div className="space-y-3">
+                <div className="text-sm font-medium">槽位详情</div>
+                {snapshot.slots.map((slot, index) => (
+                  <div key={`snapshot-slot-${slot.slotIndex}-${index}`} className="rounded-xl border p-3">
+                    <div className="mb-2 flex items-center justify-between">
+                      <div className="font-medium">
+                        {slot.title || slot.uploadLabel || `槽位 ${slot.slotIndex + 1}`}
+                      </div>
+                      <div className="text-xs text-muted-foreground">序号 {slot.slotIndex + 1}</div>
+                    </div>
+                    <div className="rounded-lg bg-muted/40 p-3 text-sm whitespace-pre-wrap">
+                      {slot.promptTemplate}
+                    </div>
+                    {slot.referenceImageUrls.length > 0 ? (
+                      <div className="mt-3">
+                        <div className="mb-2 text-xs text-muted-foreground">参考图</div>
+                        <div className="flex flex-wrap gap-2">
+                          {slot.referenceImageUrls.map((url, imageIndex) => (
+                            <img
+                              key={`${url}-${imageIndex}`}
+                              src={url}
+                              alt={`参考图 ${imageIndex + 1}`}
+                              className="h-16 w-16 rounded-lg border object-cover"
+                            />
+                          ))}
+                        </div>
+                      </div>
+                    ) : null}
+                  </div>
+                ))}
+              </div>
+            </div>
+          )}
+        </ScrollArea>
+      </DialogContent>
+    </Dialog>
+  );
+}

components/agents/AgentPresetCard.tsx

@@ -0,0 +1,111 @@
+"use client";
+
+import { Copy, Pencil, Sparkles, Trash2, Video } from "lucide-react";
+import { Button } from "@/components/ui/button";
+import { Badge } from "@/components/ui/badge";
+import { cn } from "@/lib/utils";
+import type { AgentPreset } from "@/types/agent";
+
+interface AgentPresetCardProps {
+  preset: AgentPreset;
+  selected?: boolean;
+  onSelect: (presetId: string) => void;
+  onFork?: (preset: AgentPreset) => void;
+  onEdit?: (preset: AgentPreset) => void;
+  onDelete?: (preset: AgentPreset) => void;
+}
+
+export function AgentPresetCard({
+  preset,
+  selected = false,
+  onSelect,
+  onFork,
+  onEdit,
+  onDelete,
+}: AgentPresetCardProps) {
+  return (
+    <div
+      className={cn(
+        "rounded-2xl border p-3 transition-colors",
+        selected ? "border-primary bg-primary/5" : "border-border bg-background hover:bg-muted/30"
+      )}
+    >
+      <div className="flex items-start justify-between gap-3">
+        <div className="min-w-0">
+          <div className="flex items-center gap-2">
+            {preset.type === "video" ? (
+              <Video className="h-4 w-4 text-violet-500" />
+            ) : (
+              <Sparkles className="h-4 w-4 text-amber-500" />
+            )}
+            <h4 className="truncate text-sm font-semibold">{preset.name}</h4>
+          </div>
+          <div className="mt-2 flex flex-wrap gap-1.5">
+            <Badge variant="outline">{preset.type === "video" ? "视频" : "生图"}</Badge>
+            <Badge variant="outline">{preset.visibility === "public" ? "公共" : "私有"}</Badge>
+            {preset.category?.name ? (
+              <Badge variant="outline">{preset.category.name}</Badge>
+            ) : null}
+          </div>
+        </div>
+        {preset.avatarUrl || preset.referenceAssets?.[0]?.thumbnailUrl || preset.referenceAssets?.[0]?.originalUrl ? (
+          <img
+            src={
+              preset.avatarUrl ||
+              preset.referenceAssets?.[0]?.thumbnailUrl ||
+              preset.referenceAssets?.[0]?.originalUrl ||
+              ""
+            }
+            alt={preset.name}
+            className="h-12 w-12 rounded-xl border object-cover"
+          />
+        ) : null}
+      </div>
+
+      <div className="mt-3 space-y-1 text-xs text-muted-foreground">
+        <div>模型：{preset.defaultModel?.displayName || "未设置"}</div>
+        <div>槽位：{preset.slotCount} 个</div>
+        <div>需上传：{preset.requiredUploadCount} 张</div>
+      </div>
+
+      <div className="mt-3 flex items-center gap-2">
+        <Button size="sm" className="h-8 flex-1" onClick={() => onSelect(preset.id)}>
+          {selected ? "已选中" : "使用"}
+        </Button>
+        {preset.visibility === "private" && onEdit ? (
+          <Button
+            size="sm"
+            variant="outline"
+            className="h-8"
+            onClick={() => onEdit(preset)}
+          >
+            <Pencil className="mr-1 h-3.5 w-3.5" />
+            编辑
+          </Button>
+        ) : null}
+        {preset.visibility === "private" && onDelete ? (
+          <Button
+            size="sm"
+            variant="outline"
+            className="h-8 text-destructive hover:text-destructive"
+            onClick={() => onDelete(preset)}
+          >
+            <Trash2 className="mr-1 h-3.5 w-3.5" />
+            删除
+          </Button>
+        ) : null}
+        {preset.visibility === "public" && onFork ? (
+          <Button
+            size="sm"
+            variant="outline"
+            className="h-8"
+            onClick={() => onFork(preset)}
+          >
+            <Copy className="mr-1 h-3.5 w-3.5" />
+            另存
+          </Button>
+        ) : null}
+      </div>
+    </div>
+  );
+}

components/agents/AgentPresetDialog.tsx

@@ -0,0 +1,632 @@
+"use client";
+
+import { useEffect, useMemo, useRef, useState } from "react";
+import useSWR from "swr";
+import { Loader2, Plus, Trash2, Upload } from "lucide-react";
+import { toast } from "sonner";
+import {
+  Dialog,
+  DialogContent,
+  DialogDescription,
+  DialogFooter,
+  DialogHeader,
+  DialogTitle,
+} from "@/components/ui/dialog";
+import { Button } from "@/components/ui/button";
+import { Input } from "@/components/ui/input";
+import { Textarea } from "@/components/ui/textarea";
+import {
+  Select,
+  SelectContent,
+  SelectItem,
+  SelectTrigger,
+  SelectValue,
+} from "@/components/ui/select";
+import { ScrollArea } from "@/components/ui/scroll-area";
+import { fetcher } from "@/lib/utils";
+import { imageGenerationService } from "@/services/canvas";
+import { agentService } from "@/services/agents";
+import type { AgentCategory, AgentPreset, AgentPresetSlot } from "@/types/agent";
+
+type DialogSlotState = AgentPresetSlot & {
+  localReferenceAssets: Array<{
+    assetId: string;
+    url: string;
+    name?: string | null;
+  }>;
+};
+
+interface AgentPresetDialogProps {
+  open: boolean;
+  onOpenChange: (open: boolean) => void;
+  categories: AgentCategory[];
+  initialPreset?: AgentPreset | null;
+  onSaved?: (preset: AgentPreset) => void;
+}
+
+const NEW_CATEGORY_VALUE = "__new__";
+
+export function AgentPresetDialog({
+  open,
+  onOpenChange,
+  categories,
+  initialPreset,
+  onSaved,
+}: AgentPresetDialogProps) {
+  const { data: modelsData } = useSWR<{
+    models?: Array<{
+      id: string;
+      displayName: string;
+      supportsImageGen?: boolean;
+      providerType?: string;
+    }>;
+  }>("/api/models", fetcher);
+
+  const [name, setName] = useState("");
+  const [type, setType] = useState<"image" | "video">("image");
+  const [visibility, setVisibility] = useState<"public" | "private">("private");
+  const [categoryMode, setCategoryMode] = useState<string>("");
+  const [newCategoryName, setNewCategoryName] = useState("");
+  const [defaultModelId, setDefaultModelId] = useState("");
+  const [description, setDescription] = useState("");
+  const [avatarAssetId, setAvatarAssetId] = useState<string>("");
+  const [avatarUrl, setAvatarUrl] = useState<string>("");
+  const [isSubmitting, setIsSubmitting] = useState(false);
+  const [uploadingSlotIndex, setUploadingSlotIndex] = useState<number | null>(null);
+  const [avatarUploading, setAvatarUploading] = useState(false);
+  const [slots, setSlots] = useState<DialogSlotState[]>([
+    {
+      slotIndex: 0,
+      title: "第1张",
+      promptTemplate: "",
+      requireUpload: true,
+      uploadLabel: "第1张",
+      defaultParams: null,
+      referenceAssetIds: [],
+      localReferenceAssets: [],
+    },
+  ]);
+  const fileInputRefs = useRef<Array<HTMLInputElement | null>>([]);
+  const avatarInputRef = useRef<HTMLInputElement | null>(null);
+
+  const videoProviderTypes = useMemo(
+    () => new Set(["volcano-video", "doubao-video", "qianwen-video", "fal-video", "ruxa-video"]),
+    []
+  );
+  const filteredModels = useMemo(() => {
+    const models = modelsData?.models || [];
+    if (type === "video") {
+      return models.filter((model) => videoProviderTypes.has(String(model.providerType || "")));
+    }
+    return models.filter(
+      (model) =>
+        model.supportsImageGen && !videoProviderTypes.has(String(model.providerType || ""))
+    );
+  }, [modelsData?.models, type, videoProviderTypes]);
+
+  useEffect(() => {
+    if (!open) return;
+    if (filteredModels.length === 0) {
+      setDefaultModelId("");
+      return;
+    }
+    if (!filteredModels.some((model) => model.id === defaultModelId)) {
+      setDefaultModelId(filteredModels[0]?.id || "");
+    }
+  }, [defaultModelId, filteredModels, open]);
+
+  useEffect(() => {
+    if (!open) return;
+    setSlots((current) =>
+      current.map((slot, index) => ({
+        ...slot,
+        slotIndex: index,
+        title: slot.title || `第${index + 1}张`,
+        uploadLabel: slot.uploadLabel || `第${index + 1}张`,
+      }))
+    );
+  }, [open]);
+
+  const resetState = () => {
+    setName("");
+    setType("image");
+    setVisibility("private");
+    setCategoryMode("");
+    setNewCategoryName("");
+    setDefaultModelId("");
+    setDescription("");
+    setAvatarAssetId("");
+    setAvatarUrl("");
+    setSlots([
+      {
+        slotIndex: 0,
+        title: "第1张",
+        promptTemplate: "",
+        requireUpload: true,
+        uploadLabel: "第1张",
+        defaultParams: null,
+        referenceAssetIds: [],
+        localReferenceAssets: [],
+      },
+    ]);
+    setUploadingSlotIndex(null);
+  };
+
+  useEffect(() => {
+    if (!open) return;
+    if (!initialPreset) {
+      resetState();
+      return;
+    }
+
+    setName(initialPreset.name || "");
+    setType(initialPreset.type);
+    setVisibility(initialPreset.visibility);
+    setCategoryMode(initialPreset.category?.id || initialPreset.categoryId || "");
+    setNewCategoryName("");
+    setDefaultModelId(initialPreset.defaultModelId || "");
+    setDescription(initialPreset.description || "");
+    setAvatarAssetId(initialPreset.avatarAssetId || "");
+    setAvatarUrl(initialPreset.avatarUrl || "");
+    setSlots(
+      (initialPreset.slots || []).map((slot, index) => ({
+        slotIndex: index,
+        title: slot.title || `第${index + 1}张`,
+        promptTemplate: slot.promptTemplate || "",
+        requireUpload: slot.requireUpload !== false,
+        uploadLabel: slot.uploadLabel || `第${index + 1}张`,
+        defaultParams: slot.defaultParams || null,
+        referenceAssetIds: slot.referenceAssetIds || [],
+        localReferenceAssets: (slot.referenceAssets || []).map((asset) => ({
+          assetId: asset.assetId,
+          url: asset.thumbnailUrl || asset.previewUrl || asset.originalUrl,
+          name: asset.displayName,
+        })),
+      }))
+    );
+  }, [initialPreset, open]);
+
+  const handleSlotChange = (slotIndex: number, updates: Partial<DialogSlotState>) => {
+    setSlots((current) =>
+      current.map((slot, index) =>
+        index === slotIndex ? { ...slot, ...updates, slotIndex: index } : slot
+      )
+    );
+  };
+
+  const handleAddSlot = () => {
+    setSlots((current) => [
+      ...current,
+      {
+        slotIndex: current.length,
+        title: `第${current.length + 1}张`,
+        promptTemplate: "",
+        requireUpload: true,
+        uploadLabel: `第${current.length + 1}张`,
+        defaultParams: null,
+        referenceAssetIds: [],
+        localReferenceAssets: [],
+      },
+    ]);
+  };
+
+  const handleRemoveSlot = (slotIndex: number) => {
+    setSlots((current) =>
+      current
+        .filter((_, index) => index !== slotIndex)
+        .map((slot, index) => ({
+          ...slot,
+          slotIndex: index,
+          title: slot.title || `第${index + 1}张`,
+          uploadLabel: slot.uploadLabel || `第${index + 1}张`,
+        }))
+    );
+  };
+
+  const handleAvatarUpload = async (event: React.ChangeEvent<HTMLInputElement>) => {
+    const file = event.target.files?.[0];
+    if (!file) return;
+
+    setAvatarUploading(true);
+    try {
+      const result = await imageGenerationService.uploadImage(file);
+      if (!result.success || !result.data?.assetId || !result.data.url) {
+        throw new Error(result.error || "头像上传失败");
+      }
+      setAvatarAssetId(result.data.assetId);
+      setAvatarUrl(result.data.url);
+      toast.success("头像上传成功");
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : "头像上传失败");
+    } finally {
+      setAvatarUploading(false);
+      event.target.value = "";
+    }
+  };
+
+  const handleUploadReferenceImages = async (
+    slotIndex: number,
+    event: React.ChangeEvent<HTMLInputElement>
+  ) => {
+    const files = Array.from(event.target.files || []);
+    if (files.length === 0) return;
+
+    setUploadingSlotIndex(slotIndex);
+    try {
+      const uploadedAssets: Array<{ assetId: string; url: string; name?: string | null }> = [];
+
+      for (const file of files) {
+        const result = await imageGenerationService.uploadImage(file);
+        if (!result.success || !result.data?.assetId || !result.data.url) {
+          throw new Error(result.error || "参考图上传失败");
+        }
+        uploadedAssets.push({
+          assetId: result.data.assetId,
+          url: result.data.url,
+          name: file.name,
+        });
+      }
+
+      setSlots((current) =>
+        current.map((slot, index) => {
+          if (index !== slotIndex) return slot;
+          const nextAssets = [...slot.localReferenceAssets, ...uploadedAssets];
+          return {
+            ...slot,
+            referenceAssetIds: nextAssets.map((item) => item.assetId),
+            localReferenceAssets: nextAssets,
+          };
+        })
+      );
+
+      toast.success(
+        uploadedAssets.length === 1
+          ? "参考图上传成功"
+          : `已上传 ${uploadedAssets.length} 张参考图`
+      );
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : "参考图上传失败");
+    } finally {
+      setUploadingSlotIndex(null);
+      event.target.value = "";
+    }
+  };
+
+  const handleCreate = async () => {
+    if (!name.trim()) {
+      toast.error("请输入智能体名称");
+      return;
+    }
+    if (!defaultModelId) {
+      toast.error("请选择对应模型");
+      return;
+    }
+    if (visibility === "public" && !categoryMode && !newCategoryName.trim()) {
+      toast.error("公开智能体必须选择分类");
+      return;
+    }
+    if (slots.some((slot) => !slot.promptTemplate.trim())) {
+      toast.error("每个槽位都需要填写提示词");
+      return;
+    }
+
+    setIsSubmitting(true);
+    try {
+      const payload = {
+        name: name.trim(),
+        type,
+        visibility,
+        categoryId:
+          categoryMode && categoryMode !== NEW_CATEGORY_VALUE ? categoryMode : undefined,
+        newCategoryName:
+          categoryMode === NEW_CATEGORY_VALUE ? newCategoryName.trim() : undefined,
+        description: description.trim() || undefined,
+        defaultModelId,
+        slots: slots.map((slot, index) => ({
+          slotIndex: index,
+          title: slot.title || null,
+          promptTemplate: slot.promptTemplate.trim(),
+          requireUpload: slot.requireUpload !== false,
+          uploadLabel: slot.uploadLabel || null,
+          defaultParams: slot.defaultParams || undefined,
+          referenceAssetIds: slot.referenceAssetIds || [],
+        })),
+        avatarAssetId: avatarAssetId || undefined,
+      } as const;
+
+      const result = initialPreset
+        ? await agentService.updatePreset(initialPreset.id, payload)
+        : await agentService.createPreset(payload);
+
+      if (!result.success || !result.data) {
+        throw new Error(result.error || (initialPreset ? "更新智能体失败" : "创建智能体失败"));
+      }
+
+      toast.success(initialPreset ? "智能体更新成功" : "智能体创建成功");
+      onSaved?.(result.data);
+      onOpenChange(false);
+      resetState();
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : "创建智能体失败");
+    } finally {
+      setIsSubmitting(false);
+    }
+  };
+
+  return (
+    <Dialog
+      open={open}
+      onOpenChange={(nextOpen) => {
+        onOpenChange(nextOpen);
+        if (!nextOpen) resetState();
+      }}
+    >
+      <DialogContent className="max-w-3xl bg-background">
+        <DialogHeader>
+          <DialogTitle>{initialPreset ? "编辑智能体" : "新建智能体"}</DialogTitle>
+          <DialogDescription>配置模型、槽位提示词和参考图。</DialogDescription>
+        </DialogHeader>
+
+        <ScrollArea className="max-h-[70vh] pr-3">
+          <div className="space-y-4">
+            <div className="grid grid-cols-1 gap-3 md:grid-cols-2">
+              <div className="space-y-2">
+                <div className="text-sm font-medium">头像</div>
+                <div className="flex items-center gap-3">
+                  {avatarUrl ? (
+                    <img
+                      src={avatarUrl}
+                      alt="智能体头像"
+                      className="h-14 w-14 rounded-2xl border object-cover"
+                    />
+                  ) : (
+                    <div className="flex h-14 w-14 items-center justify-center rounded-2xl border bg-muted text-xs text-muted-foreground">
+                      未上传
+                    </div>
+                  )}
+                  <label className="inline-flex">
+                    <input
+                      ref={avatarInputRef}
+                      type="file"
+                      accept="image/*"
+                      className="hidden"
+                      onChange={(event) => {
+                        void handleAvatarUpload(event);
+                      }}
+                    />
+                    <Button
+                      type="button"
+                      variant="outline"
+                      size="sm"
+                      onClick={() => avatarInputRef.current?.click()}
+                      disabled={avatarUploading}
+                    >
+                      {avatarUploading ? (
+                        <Loader2 className="mr-1 h-4 w-4 animate-spin" />
+                      ) : (
+                        <Upload className="mr-1 h-4 w-4" />
+                      )}
+                      上传头像
+                    </Button>
+                  </label>
+                </div>
+              </div>
+              <div className="space-y-2">
+                <div className="text-sm font-medium">智能体名称</div>
+                <Input value={name} onChange={(event) => setName(event.target.value)} />
+              </div>
+            </div>
+
+            <div className="grid grid-cols-1 gap-3 md:grid-cols-3">
+              <div className="space-y-2">
+                <div className="text-sm font-medium">可见性</div>
+                <Select
+                  value={visibility}
+                  onValueChange={(value: "public" | "private") => setVisibility(value)}
+                >
+                  <SelectTrigger>
+                    <SelectValue />
+                  </SelectTrigger>
+                  <SelectContent>
+                    <SelectItem value="private">私有</SelectItem>
+                    <SelectItem value="public">公开</SelectItem>
+                  </SelectContent>
+                </Select>
+              </div>
+              <div className="space-y-2">
+                <div className="text-sm font-medium">类型</div>
+                <Select
+                  value={type}
+                  onValueChange={(value: "image" | "video") => {
+                    setType(value);
+                    setDefaultModelId("");
+                  }}
+                >
+                  <SelectTrigger>
+                    <SelectValue />
+                  </SelectTrigger>
+                  <SelectContent>
+                    <SelectItem value="image">生图</SelectItem>
+                    <SelectItem value="video">视频</SelectItem>
+                  </SelectContent>
+                </Select>
+              </div>
+              <div className="space-y-2">
+                <div className="text-sm font-medium">分类</div>
+                <Select value={categoryMode} onValueChange={setCategoryMode}>
+                  <SelectTrigger>
+                    <SelectValue placeholder="选择分类" />
+                  </SelectTrigger>
+                  <SelectContent>
+                    {categories.map((category) => (
+                      <SelectItem key={category.id} value={category.id}>
+                        {category.name}
+                      </SelectItem>
+                    ))}
+                    <SelectItem value={NEW_CATEGORY_VALUE}>新建分类</SelectItem>
+                  </SelectContent>
+                </Select>
+              </div>
+              <div className="space-y-2">
+                <div className="text-sm font-medium">对应模型</div>
+                <Select value={defaultModelId} onValueChange={setDefaultModelId}>
+                  <SelectTrigger>
+                    <SelectValue placeholder="选择模型" />
+                  </SelectTrigger>
+                  <SelectContent>
+                    {filteredModels.map((model) => (
+                      <SelectItem key={model.id} value={model.id}>
+                        {model.displayName}
+                      </SelectItem>
+                    ))}
+                  </SelectContent>
+                </Select>
+              </div>
+            </div>
+
+            {categoryMode === NEW_CATEGORY_VALUE ? (
+              <div className="space-y-2">
+                <div className="text-sm font-medium">新分类名称</div>
+                <Input
+                  value={newCategoryName}
+                  onChange={(event) => setNewCategoryName(event.target.value)}
+                />
+              </div>
+            ) : null}
+
+            <div className="space-y-2">
+              <div className="text-sm font-medium">描述</div>
+              <Textarea
+                value={description}
+                onChange={(event) => setDescription(event.target.value)}
+                rows={3}
+              />
+            </div>
+
+            <div className="space-y-3">
+              <div className="flex items-center justify-between">
+                <div>
+                  <div className="text-sm font-medium">槽位配置</div>
+                  <div className="text-xs text-muted-foreground">
+                    当前共 {slots.length} 个槽位，使用时固定上传 3 张素材图：产品图1、产品图2、模特图
+                  </div>
+                </div>
+                <Button type="button" variant="outline" size="sm" onClick={handleAddSlot}>
+                  <Plus className="mr-1 h-4 w-4" />
+                  新增槽位
+                </Button>
+              </div>
+
+              <div className="space-y-3">
+                {slots.map((slot, index) => (
+                  <div key={`slot-${index}`} className="rounded-2xl border p-3">
+                    <div className="mb-3 flex items-center justify-between">
+                      <div className="font-medium">槽位 {index + 1}</div>
+                      {slots.length > 1 ? (
+                        <Button
+                          type="button"
+                          variant="ghost"
+                          size="icon"
+                          className="h-8 w-8"
+                          onClick={() => handleRemoveSlot(index)}
+                        >
+                          <Trash2 className="h-4 w-4" />
+                        </Button>
+                      ) : null}
+                    </div>
+
+                    <div className="grid grid-cols-1 gap-3 md:grid-cols-2">
+                      <div className="space-y-2">
+                        <div className="text-sm font-medium">槽位名称</div>
+                        <Input
+                          value={slot.title || ""}
+                          onChange={(event) =>
+                            handleSlotChange(index, { title: event.target.value })
+                          }
+                        />
+                      </div>
+                      <div className="space-y-2">
+                        <div className="text-sm font-medium">上传提示语</div>
+                        <Input
+                          value={slot.uploadLabel || ""}
+                          onChange={(event) =>
+                            handleSlotChange(index, { uploadLabel: event.target.value })
+                          }
+                        />
+                      </div>
+                    </div>
+
+                    <div className="mt-3 space-y-2">
+                      <div className="text-sm font-medium">提示词</div>
+                      <Textarea
+                        rows={4}
+                        value={slot.promptTemplate}
+                        onChange={(event) =>
+                          handleSlotChange(index, { promptTemplate: event.target.value })
+                        }
+                      />
+                    </div>
+
+                    <div className="mt-3">
+                      <input
+                        ref={(element) => {
+                          fileInputRefs.current[index] = element;
+                        }}
+                        type="file"
+                        accept="image/*"
+                        multiple
+                        className="hidden"
+                        onChange={(event) => {
+                          void handleUploadReferenceImages(index, event);
+                        }}
+                      />
+                      <Button
+                        type="button"
+                        variant="outline"
+                        size="sm"
+                        onClick={() => fileInputRefs.current[index]?.click()}
+                        disabled={uploadingSlotIndex === index}
+                      >
+                        {uploadingSlotIndex === index ? (
+                          <Loader2 className="mr-1 h-4 w-4 animate-spin" />
+                        ) : (
+                          <Upload className="mr-1 h-4 w-4" />
+                        )}
+                        上传参考图
+                      </Button>
+                      {slot.localReferenceAssets.length > 0 ? (
+                        <div className="mt-2 flex flex-wrap gap-2">
+                          {slot.localReferenceAssets.map((asset, assetIndex) => (
+                            <div key={`${asset.assetId}-${assetIndex}`} className="relative">
+                              <img
+                                src={asset.url}
+                                alt={asset.name || `参考图 ${assetIndex + 1}`}
+                                className="h-16 w-16 rounded-lg border object-cover"
+                              />
+                            </div>
+                          ))}
+                        </div>
+                      ) : (
+                        <div className="mt-2 text-xs text-muted-foreground">未上传参考图</div>
+                      )}
+                    </div>
+                  </div>
+                ))}
+              </div>
+            </div>
+          </div>
+        </ScrollArea>
+
+        <DialogFooter>
+          <Button variant="outline" onClick={() => onOpenChange(false)}>
+            取消
+          </Button>
+          <Button onClick={handleCreate} disabled={isSubmitting}>
+            {isSubmitting ? <Loader2 className="mr-2 h-4 w-4 animate-spin" /> : null}
+            {initialPreset ? "保存修改" : "创建智能体"}
+          </Button>
+        </DialogFooter>
+      </DialogContent>
+    </Dialog>
+  );
+}

components/agents/AgentPresetPanel.tsx

@@ -0,0 +1,208 @@
+"use client";
+
+import { useMemo, useState } from "react";
+import useSWR from "swr";
+import { Loader2, Plus, Sparkles } from "lucide-react";
+import { toast } from "sonner";
+import { Button } from "@/components/ui/button";
+import { Input } from "@/components/ui/input";
+import { ScrollArea } from "@/components/ui/scroll-area";
+import { agentService } from "@/services/agents";
+import type { AgentPreset } from "@/types/agent";
+import { AgentCategoryTabs } from "@/components/agents/AgentCategoryTabs";
+import { AgentPresetCard } from "@/components/agents/AgentPresetCard";
+import { AgentPresetDialog } from "@/components/agents/AgentPresetDialog";
+
+interface AgentPresetPanelProps {
+  selectedPresetId?: string | null;
+  onSelectPreset: (presetId: string | null) => void;
+}
+
+export function AgentPresetPanel({
+  selectedPresetId,
+  onSelectPreset,
+}: AgentPresetPanelProps) {
+  const [keyword, setKeyword] = useState("");
+  const [selectedCategoryId, setSelectedCategoryId] = useState<string | null>(null);
+  const [showCreateDialog, setShowCreateDialog] = useState(false);
+  const [editingPreset, setEditingPreset] = useState<AgentPreset | null>(null);
+  const { data, isLoading, mutate } = useSWR(
+    ["agent-presets", selectedCategoryId],
+    async () => {
+      const result = await agentService.getPresets({
+        scope: "all",
+        sort: "latest",
+      });
+      if (!result.success || !result.data) {
+        throw new Error(result.error || "获取智能体失败");
+      }
+      return result.data;
+    }
+  );
+
+  const publicCategories = useMemo(
+    () => (data?.categories || []).filter((category) => category.scope !== "private"),
+    [data?.categories]
+  );
+  const filteredPublicPresets = useMemo(() => {
+    return (data?.public || []).filter((preset) => {
+      const matchedCategory = !selectedCategoryId || preset.category?.id === selectedCategoryId;
+      const matchedKeyword =
+        !keyword.trim() ||
+        preset.name.toLowerCase().includes(keyword.trim().toLowerCase()) ||
+        String(preset.category?.name || "")
+          .toLowerCase()
+          .includes(keyword.trim().toLowerCase());
+      return matchedCategory && matchedKeyword;
+    });
+  }, [data?.public, keyword, selectedCategoryId]);
+  const filteredPrivatePresets = useMemo(() => {
+    return (data?.private || []).filter((preset) => {
+      const matchedKeyword =
+        !keyword.trim() ||
+        preset.name.toLowerCase().includes(keyword.trim().toLowerCase()) ||
+        String(preset.category?.name || "")
+          .toLowerCase()
+          .includes(keyword.trim().toLowerCase());
+      return matchedKeyword;
+    });
+  }, [data?.private, keyword]);
+
+  const handleForkPreset = async (preset: AgentPreset) => {
+    const result = await agentService.forkPreset(preset.id, `${preset.name}（副本）`);
+    if (!result.success || !result.data) {
+      toast.error(result.error || "复制智能体失败");
+      return;
+    }
+    toast.success("已另存为我的智能体");
+    await mutate();
+    onSelectPreset(result.data.id);
+  };
+
+  const handleDeletePreset = async (preset: AgentPreset) => {
+    if (!window.confirm(`确定删除智能体「${preset.name}」吗？`)) {
+      return;
+    }
+
+    const result = await agentService.deletePreset(preset.id);
+    if (!result.success) {
+      toast.error(result.error || "删除智能体失败");
+      return;
+    }
+
+    toast.success("智能体已删除");
+    if (selectedPresetId === preset.id) {
+      onSelectPreset(null);
+    }
+    await mutate();
+  };
+
+  return (
+    <>
+      <div className="flex h-full flex-col border-r bg-white dark:bg-zinc-950">
+        <div className="border-b p-4">
+          <div className="mb-3 flex items-center justify-between">
+            <div>
+              <div className="flex items-center gap-2 text-sm font-semibold">
+                <Sparkles className="h-4 w-4 text-amber-500" />
+                智能体
+              </div>
+              <div className="mt-1 text-xs text-muted-foreground">
+                选择智能体后，右侧对话框会按其要求显示上传槽位
+              </div>
+            </div>
+            <Button size="sm" onClick={() => setShowCreateDialog(true)}>
+              <Plus className="mr-1 h-4 w-4" />
+              新建
+            </Button>
+          </div>
+          <Input
+            placeholder="搜索智能体"
+            value={keyword}
+            onChange={(event) => setKeyword(event.target.value)}
+          />
+        </div>
+
+        <ScrollArea className="flex-1">
+          <div className="space-y-5 p-4">
+            <section className="space-y-3">
+              <div className="space-y-2">
+                <div className="text-sm font-medium">公共智能体</div>
+                <AgentCategoryTabs
+                  categories={publicCategories}
+                  selectedCategoryId={selectedCategoryId}
+                  onSelectCategory={setSelectedCategoryId}
+                />
+              </div>
+              {isLoading ? (
+                <div className="flex items-center justify-center py-8">
+                  <Loader2 className="h-5 w-5 animate-spin text-muted-foreground" />
+                </div>
+              ) : filteredPublicPresets.length > 0 ? (
+                <div className="space-y-3">
+                  {filteredPublicPresets.map((preset) => (
+                    <AgentPresetCard
+                      key={preset.id}
+                      preset={preset}
+                      selected={selectedPresetId === preset.id}
+                      onSelect={onSelectPreset}
+                      onFork={handleForkPreset}
+                      onEdit={setEditingPreset}
+                      onDelete={handleDeletePreset}
+                    />
+                  ))}
+                </div>
+              ) : (
+                <div className="rounded-xl border border-dashed p-4 text-sm text-muted-foreground">
+                  暂无公共智能体
+                </div>
+              )}
+            </section>
+
+            <section className="space-y-3">
+              <div className="text-sm font-medium">我的智能体</div>
+              {isLoading ? null : filteredPrivatePresets.length > 0 ? (
+                <div className="space-y-3">
+                  {filteredPrivatePresets.map((preset) => (
+                    <AgentPresetCard
+                      key={preset.id}
+                      preset={preset}
+                      selected={selectedPresetId === preset.id}
+                      onSelect={onSelectPreset}
+                      onEdit={setEditingPreset}
+                      onDelete={handleDeletePreset}
+                    />
+                  ))}
+                </div>
+              ) : (
+                <div className="rounded-xl border border-dashed p-4 text-sm text-muted-foreground">
+                  还没有私有智能体，先新建一个吧
+                </div>
+              )}
+            </section>
+          </div>
+        </ScrollArea>
+      </div>
+
+      <AgentPresetDialog
+        open={showCreateDialog || Boolean(editingPreset)}
+        onOpenChange={(nextOpen) => {
+          if (!nextOpen) {
+            setShowCreateDialog(false);
+            setEditingPreset(null);
+            return;
+          }
+          setShowCreateDialog(true);
+        }}
+        categories={data?.categories || []}
+        initialPreset={editingPreset}
+        onSaved={async (preset) => {
+          await mutate();
+          onSelectPreset(preset.id);
+          setEditingPreset(null);
+          setShowCreateDialog(false);
+        }}
+      />
+    </>
+  );
+}

components/canvas/CanvasChatPanel.tsx

@@ -48,6 +48,7 @@ import {
 } from "@/lib/image-url";
 import { buildImageProxyUrl } from "@/lib/image-delivery";
 import type { Attachment, ChatMessage } from "@/lib/types";
+import type { AgentPreset } from "@/types/agent";
 import { type CanvasLaunchPayload, getCanvasLaunchStorageKey } from "@/lib/canvas-launch";
 import {
   DEFAULT_CANVAS_SKILL_TAGS_TITLE,
@@ -88,6 +89,11 @@ type CanvasMediaStreamPayload = {
   videoUrls?: string[];
   prompt?: string;
   sourceImageUrls?: string[];
+  agentPresetId?: string;
+  agentPresetName?: string;
+  agentCategoryId?: string;
+  agentCategoryName?: string;
+  agentExecutionSnapshotId?: string;
 };
 
 type ImageGenerationStatusStreamPayload = {
@@ -102,6 +108,7 @@ interface CanvasChatPanelProps {
   initialMessages?: ChatMessage[];
   onClose?: () => void;
   onOpenFileList?: () => void;
+  selectedAgentPresetId?: string | null;
 }
 
 type VideoStoryboardStreamData =
@@ -1579,7 +1586,13 @@ function buildDetailReviewAppendMarkdown(params: {
   return lines.join("\n").trim();
 }
 
-export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenFileList }: CanvasChatPanelProps) {
+export function CanvasChatPanel({
+  chatId,
+  initialMessages = [],
+  onClose,
+  onOpenFileList,
+  selectedAgentPresetId,
+}: CanvasChatPanelProps) {
   const { data: session } = useSession();
   const [input, setInput] = useState("");
   const [attachments, setAttachments] = useState<PickedObjectAttachment[]>([]);
@@ -1641,6 +1654,7 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
   const messagesEndRef = useRef<HTMLDivElement>(null);
   const fileInputRef = useRef<HTMLInputElement>(null);
   const videoSlotInputRefs = useRef<Array<HTMLInputElement | null>>([]);
+  const agentSlotInputRefs = useRef<Array<HTMLInputElement | null>>([]);
   const detailSlotInputRefs = useRef<Array<HTMLInputElement | null>>([]);
   const textareaRef = useRef<HTMLTextAreaElement>(null);
   const detailModeSwitchRef = useRef(false);
@@ -1669,6 +1683,7 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
   });
   const [videoParams, setVideoParams] = useState<VideoGenerationParams | null>(null);
   const [videoSlotUploading, setVideoSlotUploading] = useState<number | null>(null);
+  const [agentSlotUploading, setAgentSlotUploading] = useState<number | null>(null);
   const [lastImageModelId, setLastImageModelId] = useState<string | null>(null);
   const [lastVideoModelId, setLastVideoModelId] = useState<string | null>(null);
   const [lastConversationModelId, setLastConversationModelId] = useState<string | null>(null);
@@ -2544,6 +2559,10 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
       return response.json();
     }
   );
+  const { data: agentPresetResponse } = useSWR<{ success?: boolean; data?: AgentPreset }>(
+    selectedAgentPresetId ? `/api/agents/${selectedAgentPresetId}` : null,
+    fetcher
+  );
   const greetingUserName = useMemo(() => {
     const userName = String(session?.user?.name || "").trim();
     if (userName) return userName;
@@ -2678,6 +2697,10 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     () => canvasSkillTags.find((tag) => tag.id === selectedSkillTagId) || null,
     [canvasSkillTags, selectedSkillTagId]
   );
+  const selectedAgentPreset = useMemo(
+    () => (agentPresetResponse?.success && agentPresetResponse.data ? agentPresetResponse.data : null),
+    [agentPresetResponse]
+  );
   const SelectedSkillTagIcon = useMemo(
     () => getCanvasSkillIconComponent(selectedSkillTag?.iconName),
     [selectedSkillTag?.iconName]
@@ -2835,6 +2858,29 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     if (videoParams.mode === "first-frame") return ["首帧"] as string[];
     return [] as string[];
   }, [isVideoModel, videoParams]);
+  const activeComposerMode = useMemo<"agent" | "image" | "video" | "detail">(() => {
+    if (isDetailGenerationMode) return "detail";
+    if (isVideoModel) return "video";
+    if (isAgentModel) return "agent";
+    return "image";
+  }, [isAgentModel, isDetailGenerationMode, isVideoModel]);
+  const agentUploadSlots = useMemo(() => {
+    if (activeComposerMode !== "agent" || !selectedAgentPreset) {
+      return [] as Array<{
+        slotIndex: number;
+        label: string;
+      }>;
+    }
+
+    return [
+      { slotIndex: 0, label: "产品图1" },
+      { slotIndex: 1, label: "产品图2" },
+      { slotIndex: 2, label: "模特图" },
+    ];
+  }, [activeComposerMode, selectedAgentPreset]);
+  const hasAgentUploadSlots =
+    activeComposerMode === "agent" && agentUploadSlots.length > 0;
+  const agentRequiredUploadCount = hasAgentUploadSlots ? 3 : 0;
 
   // 监听 pendingChatAttachments：视频模式时填充首尾帧槽位，非视频模式插入对话标签
   useEffect(() => {
@@ -2899,6 +2945,56 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
       return;
     }
 
+    if (hasAgentUploadSlots) {
+      const slotCount = agentUploadSlots.length;
+      const currentSlots: Array<PickedObjectAttachment | null> = Array.from(
+        { length: slotCount },
+        (_, index) => attachments[index] ?? null
+      );
+      const existingUrls = new Set(currentSlots.filter(Boolean).map((item) => item!.url));
+      const toAdd = pendingChatAttachments
+        .filter((item) => item?.url && !existingUrls.has(item.url))
+        .map((item) => ({
+          url: item.url,
+          name: item.name || "参考图",
+          contentType: item.contentType || "image/jpeg",
+        })) as PickedObjectAttachment[];
+
+      let cursor = 0;
+      let inserted = 0;
+      for (let index = 0; index < slotCount && cursor < toAdd.length; index += 1) {
+        if (currentSlots[index]) continue;
+        currentSlots[index] = toAdd[cursor]!;
+        cursor += 1;
+        inserted += 1;
+      }
+
+      if (inserted > 0) {
+        const next = currentSlots
+          .slice(0, slotCount)
+          .filter((item): item is PickedObjectAttachment => Boolean(item));
+        setAttachments(next);
+        toast.success(
+          slotCount === 1
+            ? "已添加到智能体图片槽位"
+            : `已添加到智能体槽位（${inserted}/${slotCount}）`
+        );
+      } else if (toAdd.length > 0) {
+        toast.info("智能体图片槽位已满，请先替换或移除");
+      }
+
+      if (toAdd.length - inserted > 0) {
+        toast.info(`还有 ${toAdd.length - inserted} 张未添加，可继续替换智能体图片槽位`);
+      }
+
+      setTimeout(() => {
+        (textareaRef.current as HTMLDivElement | null)?.focus();
+      }, 0);
+
+      clearPendingChatAttachments();
+      return;
+    }
+
     if (isVideoModel && videoFrameSlots.length > 0) {
       const slotCount = videoFrameSlots.length;
       const currentSlots: Array<PickedObjectAttachment | null> = Array.from(
@@ -2966,6 +3062,8 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     pendingChatAttachments,
     isDetailGenerationMode,
     detailSlotPickerIndex,
+    hasAgentUploadSlots,
+    agentUploadSlots.length,
     isVideoModel,
     videoFrameSlots.length,
     attachments,
@@ -2981,12 +3079,6 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     () => videoOnlyModels[0]?.id || null,
     [videoOnlyModels]
   );
-  const activeComposerMode = useMemo<"agent" | "image" | "video" | "detail">(() => {
-    if (isDetailGenerationMode) return "detail";
-    if (isVideoModel) return "video";
-    if (isAgentModel) return "agent";
-    return "image";
-  }, [isAgentModel, isDetailGenerationMode, isVideoModel]);
   const activeModeLabel =
     activeComposerMode === "video"
       ? "视频"
@@ -3028,6 +3120,31 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     setSelectedModelId(modelSelectValue);
   }, [modelSelectValue, selectedModelId]);
 
+  useEffect(() => {
+    if (!selectedAgentPresetId) return;
+    if (activeComposerMode === "agent") return;
+
+    const preferredAgentModelId =
+      allModels.find(
+        (model) =>
+          model.id === lastConversationModelId &&
+          model.isMultimodal &&
+          !VIDEO_PROVIDER_TYPES.has(model.providerType)
+      )?.id || "auto";
+
+    currentModelIdRef.current = preferredAgentModelId;
+    setSelectedModelId(preferredAgentModelId);
+    if (isDetailGenerationMode) {
+      setIsDetailGenerationMode(false);
+    }
+  }, [
+    activeComposerMode,
+    allModels,
+    isDetailGenerationMode,
+    lastConversationModelId,
+    selectedAgentPresetId,
+  ]);
+
   useEffect(() => {
     if (isDetailGenerationMode) {
       setAttachments([]);
@@ -3085,6 +3202,15 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     });
   }, [isVideoModel, videoFrameSlots.length]);
 
+  useEffect(() => {
+    if (!hasAgentUploadSlots) return;
+    const maxAttachments = agentUploadSlots.length;
+    setAttachments((current) => {
+      if (current.length <= maxAttachments) return current;
+      return current.slice(0, maxAttachments);
+    });
+  }, [agentUploadSlots.length, hasAgentUploadSlots]);
+
   useEffect(() => {
     if (!currentModel) return;
     if (currentModel.id !== "auto" && isDetailGenerationMode) {
@@ -3377,6 +3503,12 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
         : [];
       const messageId = String(payload.messageId || "").trim();
       const generationPrompt = String(payload.prompt || "").trim() || "对话生成";
+      const agentPresetId = String(payload.agentPresetId || "").trim() || undefined;
+      const agentPresetName = String(payload.agentPresetName || "").trim() || undefined;
+      const agentCategoryId = String(payload.agentCategoryId || "").trim() || undefined;
+      const agentCategoryName = String(payload.agentCategoryName || "").trim() || undefined;
+      const agentExecutionSnapshotId =
+        String(payload.agentExecutionSnapshotId || "").trim() || undefined;
       let addedImageCount = 0;
       let addedVideoCount = 0;
 
@@ -3438,6 +3570,11 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
               sourceMessageId: messageId || undefined,
               sourceImageKey,
               compareSourceUrl: canUseCompareSource ? compareSourceUrl : undefined,
+              agentPresetId,
+              agentPresetName,
+              agentCategoryId,
+              agentCategoryName,
+              agentExecutionSnapshotId,
             },
           },
           connections: sourceNodeId ? [sourceNodeId] : [],
@@ -3465,6 +3602,13 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
             src: url,
             status: "completed",
             prompt: generationPrompt,
+            metadata: {
+              agentPresetId,
+              agentPresetName,
+              agentCategoryId,
+              agentCategoryName,
+              agentExecutionSnapshotId,
+            },
           },
           connections: [],
         });
@@ -3529,6 +3673,10 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
           body: {
             id: rawId,
             message: request.messages.at(-1),
+            agentPresetId:
+              activeComposerMode === "agent" && selectedAgentPreset?.id
+                ? selectedAgentPreset.id
+                : undefined,
             selectedChatModel: currentModelIdRef.current,
             selectedVisibilityType: "private",
             deepThinkingEnabled: requestIsConversationModel
@@ -4480,6 +4628,24 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     [videoFrameSlots.length]
   );
 
+  const updateAttachmentAtAgentSlot = useCallback(
+    (slotIndex: number, attachment: PickedObjectAttachment | null) => {
+      const slotCount = Math.max(agentUploadSlots.length, 1);
+      setAttachments((current) => {
+        const slots: Array<PickedObjectAttachment | null> = Array.from(
+          { length: slotCount },
+          (_, index) => current[index] ?? null
+        );
+        slots[slotIndex] = attachment;
+        while (slots.length > 0 && !slots[slots.length - 1]) {
+          slots.pop();
+        }
+        return slots.filter((item): item is PickedObjectAttachment => Boolean(item));
+      });
+    },
+    [agentUploadSlots.length]
+  );
+
   const fillVideoSlotFromCanvasSelection = useCallback(
     (slotIndex: number) => {
       const preferredNode = selectedImageNodes[slotIndex] || selectedImageNodes[0];
@@ -4494,6 +4660,20 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     [getCanvasNodeAttachment, selectedImageNodes, updateAttachmentAtVideoSlot]
   );
 
+  const fillAgentSlotFromCanvasSelection = useCallback(
+    (slotIndex: number) => {
+      const preferredNode = selectedImageNodes[slotIndex] || selectedImageNodes[0];
+      const attachment = preferredNode
+        ? getCanvasNodeAttachment(preferredNode, `智能体图片 ${slotIndex + 1}`)
+        : null;
+      if (!attachment) return false;
+
+      updateAttachmentAtAgentSlot(slotIndex, attachment);
+      return true;
+    },
+    [getCanvasNodeAttachment, selectedImageNodes, updateAttachmentAtAgentSlot]
+  );
+
   const handleVideoSlotFileChange = useCallback(
     async (slotIndex: number, event: React.ChangeEvent<HTMLInputElement>) => {
       const file = event.target.files?.[0];
@@ -4566,6 +4746,80 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     [addNode, nodes.length, updateAttachmentAtVideoSlot]
   );
 
+  const handleAgentSlotFileChange = useCallback(
+    async (slotIndex: number, event: React.ChangeEvent<HTMLInputElement>) => {
+      const file = event.target.files?.[0];
+      if (!file) return;
+
+      if (!file.type.startsWith("image/")) {
+        toast.error("请上传图片文件");
+        event.currentTarget.value = "";
+        return;
+      }
+
+      if (file.size > 50 * 1024 * 1024) {
+        toast.error("图片大小不能超过 50MB");
+        event.currentTarget.value = "";
+        return;
+      }
+
+      setAgentSlotUploading(slotIndex);
+      try {
+        const formData = new FormData();
+        formData.append("file", file);
+
+        const response = await fetchUploadWithRetry(formData);
+
+        if (!response.ok) {
+          throw new Error("上传失败");
+        }
+
+        const data = await response.json();
+        const imageUrl = data.url as string;
+        const assetId =
+          typeof data.assetId === "string" && data.assetId.trim() ? data.assetId.trim() : undefined;
+
+        updateAttachmentAtAgentSlot(slotIndex, {
+          url: imageUrl,
+          name: file.name,
+          contentType: file.type || "image/jpeg",
+          assetId,
+        });
+
+        const dimensions = await getImageDimensions(imageUrl);
+        const nodeIndex = nodes.length + slotIndex;
+        const spacing = 20;
+        const x = 100 + (nodeIndex % 3) * (dimensions.width + spacing);
+        const y = 100 + Math.floor(nodeIndex / 3) * (dimensions.height + spacing);
+        addNode({
+          type: "image",
+          position: { x, y },
+          size: { width: dimensions.width, height: dimensions.height },
+          data: {
+            src: imageUrl,
+            status: "completed",
+            title: file.name.replace(/\.[^.]+$/, ""),
+            prompt: file.name.replace(/\.[^.]+$/, ""),
+            metadata: {
+              originalWidth: dimensions.originalWidth,
+              originalHeight: dimensions.originalHeight,
+            },
+          },
+          connections: [],
+        });
+
+        const slotLabel = agentUploadSlots[slotIndex]?.label || `第${slotIndex + 1}张`;
+        toast.success(`${slotLabel}上传成功`);
+      } catch (_error) {
+        toast.error("图片上传失败");
+      } finally {
+        setAgentSlotUploading(null);
+        event.currentTarget.value = "";
+      }
+    },
+    [addNode, agentUploadSlots, nodes.length, updateAttachmentAtAgentSlot]
+  );
+
   const updateDetailSlotAttachment = useCallback(
     (slotIndex: number, attachment: PickedObjectAttachment | null) => {
       setDetailSlots((current) => {
@@ -6354,6 +6608,16 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     const parts: any[] = [];
     
     let effectiveAttachments = [...attachments];
+    if (hasAgentUploadSlots) {
+      if (effectiveAttachments.length > agentRequiredUploadCount) {
+        effectiveAttachments = effectiveAttachments.slice(0, agentRequiredUploadCount);
+      }
+
+      if (effectiveAttachments.length < agentRequiredUploadCount) {
+        toast.error("当前智能体需要上传 3 张图片：产品图1、产品图2、模特图");
+        return;
+      }
+    }
     const videoImageLimit = isVideoModel
       ? getVideoImageLimit(videoParams, currentModel?.providerType, currentModel?.videoToolCapability)
       : null;
@@ -6428,6 +6692,9 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     if (!textContent && effectiveAttachments.length > 0) {
       textContent = "请处理这张图片";
     }
+    if (!textContent && hasAgentUploadSlots && selectedAgentPreset) {
+      textContent = `请按「${selectedAgentPreset.name}」智能体配置顺序处理产品图1、产品图2和模特图。`;
+    }
     if (!textContent && selectedSkillTag) {
       textContent = `请按「${selectedSkillTag.name}」标签的要求处理。`;
     }
@@ -6529,9 +6796,12 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
   }, [
     activeChatId,
     activeComposerMode,
+    agentRequiredUploadCount,
     input,
     attachments,
+    hasAgentUploadSlots,
     selectedSkillTag,
+    selectedAgentPreset,
     selectedModelId,
     sendMessage,
     clearSelection,
@@ -6674,6 +6944,47 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     const baseNodeCount = nodes.length;
     let uploadIndex = 0;
 
+    if (hasAgentUploadSlots) {
+      const slotCount = agentUploadSlots.length;
+      const currentSlots = Array.from({ length: slotCount }, (_, index) => attachments[index] ?? null);
+      let handledCount = 0;
+
+      for (const file of files) {
+        const targetSlotIndex = currentSlots.findIndex((slot) => !slot);
+        if (targetSlotIndex < 0) break;
+
+        const uploaded = await uploadImageFileAndAddToCanvas(file, baseNodeCount + uploadIndex);
+        if (!uploaded) continue;
+
+        const attachment: PickedObjectAttachment = {
+          url: uploaded.imageUrl,
+          name: uploaded.name,
+          contentType: uploaded.contentType,
+          assetId: uploaded.assetId,
+        };
+        updateAttachmentAtAgentSlot(targetSlotIndex, attachment);
+        currentSlots[targetSlotIndex] = attachment;
+        handledCount += 1;
+        uploadIndex += 1;
+        if (!firstImageUrl) {
+          firstImageUrl = uploaded.imageUrl;
+        }
+      }
+
+      if (handledCount > 0) {
+        toast.success(
+          handledCount === 1
+            ? "已粘贴图片到智能体槽位"
+            : `已粘贴 ${handledCount} 张图片到智能体槽位`
+        );
+      }
+      if (handledCount < files.length) {
+        toast.info("智能体图片槽位已满，剩余图片未添加");
+      }
+      void maybeAutoSetCanvasTitle(firstImageUrl);
+      return;
+    }
+
     if (isVideoModel && videoFrameSlots.length > 0) {
       const slotCount = videoFrameSlots.length;
       const currentSlots = Array.from({ length: slotCount }, (_, index) => attachments[index] ?? null);
@@ -6745,11 +7056,14 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
 
     void maybeAutoSetCanvasTitle(firstImageUrl);
   }, [
+    agentUploadSlots.length,
     attachments,
+    hasAgentUploadSlots,
     insertImageTagInEditor,
     isVideoModel,
     maybeAutoSetCanvasTitle,
     nodes.length,
+    updateAttachmentAtAgentSlot,
     updateAttachmentAtVideoSlot,
     uploadImageFileAndAddToCanvas,
     videoFrameSlots.length,
@@ -6764,6 +7078,54 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
     // 记录本次上传开始时的节点数量，用于计算新图片位置
     const baseNodeCount = nodes.length;
     let uploadIndex = 0;
+
+    if (hasAgentUploadSlots) {
+      const slotCount = agentUploadSlots.length;
+      const currentSlots = Array.from({ length: slotCount }, (_, index) => attachments[index] ?? null);
+      let handledCount = 0;
+
+      for (const file of Array.from(files)) {
+        const targetSlotIndex = currentSlots.findIndex((slot) => !slot);
+        if (targetSlotIndex < 0) break;
+
+        const uploaded = await uploadImageFileAndAddToCanvas(file, baseNodeCount + uploadIndex);
+        if (!uploaded) continue;
+
+        updateAttachmentAtAgentSlot(targetSlotIndex, {
+          url: uploaded.imageUrl,
+          name: uploaded.name,
+          contentType: uploaded.contentType,
+          assetId: uploaded.assetId,
+        });
+
+        currentSlots[targetSlotIndex] = {
+          url: uploaded.imageUrl,
+          name: uploaded.name,
+          contentType: uploaded.contentType,
+          assetId: uploaded.assetId,
+        };
+        handledCount += 1;
+        uploadIndex += 1;
+        if (!firstImageUrl) {
+          firstImageUrl = uploaded.imageUrl;
+        }
+      }
+
+      if (handledCount > 0) {
+        toast.success(
+          handledCount === 1
+            ? "图片已添加到智能体槽位"
+            : `已添加 ${handledCount} 张图片到智能体槽位`
+        );
+      }
+      if (handledCount < files.length) {
+        toast.info("智能体需要的图片槽位已满，剩余图片未添加");
+      }
+
+      void maybeAutoSetCanvasTitle(firstImageUrl);
+      e.target.value = '';
+      return;
+    }
     
     for (const file of Array.from(files)) {
       if (!file.type.startsWith('image/')) {
@@ -6922,8 +7284,11 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
   const hasInputOrAttachments = Boolean(
     input.trim() || attachments.length > 0 || (!isDetailGenerationMode && selectedSkillTag)
   );
+  const hasEnoughAgentUploads = !hasAgentUploadSlots || attachments.length >= agentRequiredUploadCount;
   const composerPlaceholder = isDetailGenerationMode
     ? "可选补充：风格、情绪氛围、禁用元素、期望生成张数（如：生成8张）"
+    : hasAgentUploadSlots
+      ? "请先上传产品图1、产品图2、模特图，再补充说明"
     : selectedSkillTag?.placeholder?.trim() ||
       (selectedSkillTag
         ? `请输入「${selectedSkillTag.name}」相关需求`
@@ -6937,7 +7302,7 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
       : 0;
   const canSend = isDetailGenerationMode
     ? detailValidationIssues.length === 0 && !isDetailGenerating
-    : hasInputOrAttachments;
+    : hasInputOrAttachments && hasEnoughAgentUploads;
   const currentVideoImageLimit = isVideoModel
     ? getVideoImageLimit(videoParams, currentModel?.providerType, currentModel?.videoToolCapability)
     : null;
@@ -8673,6 +9038,19 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
                 </button>
               </div>
             )}
+            {!isDetailGenerationMode && activeComposerMode === "agent" && selectedAgentPreset && (
+              <div className="mb-2 flex flex-wrap items-center gap-2">
+                <div className="inline-flex items-center gap-2 rounded-full bg-[#eef2ff] px-3 py-1.5 text-sm font-medium text-[#4338ca]">
+                  <Sparkles className="h-4 w-4" />
+                  <span>{selectedAgentPreset.name}</span>
+                </div>
+                {hasAgentUploadSlots ? (
+                  <div className="inline-flex items-center rounded-full bg-[#f4f5f7] px-3 py-1.5 text-xs text-[#5b6472]">
+                    请上传产品图1、产品图2、模特图
+                  </div>
+                ) : null}
+              </div>
+            )}
             <div
               ref={textareaRef as any}
               contentEditable={!isSubmitting}
@@ -8689,7 +9067,7 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
                   .replace(/\s+/g, " ")
                   .trim();
                 setInput(textContent);
-                if (!isVideoModel || videoFrameSlots.length === 0) {
+                if ((!isVideoModel || videoFrameSlots.length === 0) && !hasAgentUploadSlots) {
                   syncAttachmentsWithEditor(target);
                 }
               }}
@@ -8702,7 +9080,7 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
                   handleSend();
                 }
                 // 处理 Backspace 删除图片标签（只删除聊天框中的标签，不影响画布）
-                if (e.key === 'Backspace' && (!isVideoModel || videoFrameSlots.length === 0)) {
+                if (e.key === 'Backspace' && (!isVideoModel || videoFrameSlots.length === 0) && !hasAgentUploadSlots) {
                   const selection = window.getSelection();
                   if (selection && selection.rangeCount > 0) {
                     const range = selection.getRangeAt(0);
@@ -8741,6 +9119,74 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
             />
           </div>
 
+          {hasAgentUploadSlots && (
+            <div className="mb-11 px-4 pb-1">
+              <div className="flex flex-row gap-2 overflow-x-auto scrollbar-hide">
+                {agentUploadSlots.map((slot, index) => {
+                  const slotAttachment = attachments[index];
+                  const isUploading = agentSlotUploading === index;
+                  const isDisabled = isLoading || (index > 0 && !attachments[index - 1]);
+                  return (
+                    <button
+                      key={`canvas-agent-slot-${slot.slotIndex}-${index}`}
+                      type="button"
+                      onClick={() => {
+                        if (isDisabled) return;
+                        if (fillAgentSlotFromCanvasSelection(index)) {
+                          toast.success(`${slot.label}已从画布填充`);
+                          return;
+                        }
+                        agentSlotInputRefs.current[index]?.click();
+                      }}
+                      className={cn(
+                        "group relative flex h-24 min-w-24 shrink-0 items-center justify-center overflow-hidden rounded-xl border bg-muted/30 text-muted-foreground transition-colors",
+                        !isDisabled && "cursor-pointer hover:border-primary/60 hover:bg-muted/50",
+                        isDisabled && "cursor-not-allowed opacity-55"
+                      )}
+                    >
+                      {slotAttachment ? (
+                        <>
+                          <img
+                            src={buildPreviewImageUrl(slotAttachment.url, 256)}
+                            alt={slotAttachment.name || slot.label}
+                            className="h-full w-full object-cover"
+                            loading="lazy"
+                            decoding="async"
+                            onError={(event) => {
+                              if (event.currentTarget.src !== slotAttachment.url) {
+                                event.currentTarget.src = slotAttachment.url;
+                              }
+                            }}
+                          />
+                          <div className="absolute inset-x-0 bottom-0 bg-black/55 px-1 py-0.5 text-center text-[11px] text-white">
+                            {slot.label}
+                          </div>
+                          <div
+                            className="absolute right-1 top-1 rounded-full bg-black/65 px-1 text-[11px] text-white opacity-0 transition-opacity group-hover:opacity-100"
+                            onClick={(event) => {
+                              event.stopPropagation();
+                              updateAttachmentAtAgentSlot(index, null);
+                            }}
+                          >
+                            x
+                          </div>
+                        </>
+                      ) : (
+                        <div className="flex flex-col items-center gap-1 px-1">
+                          <span className="text-xl leading-none">+</span>
+                          <span className="text-center text-xs">{isUploading ? "上传中" : slot.label}</span>
+                        </div>
+                      )}
+                    </button>
+                  );
+                })}
+              </div>
+              <div className="mt-1.5 text-[11px] text-muted-foreground">
+                智能体需要上传 3 张图片：产品图1、产品图2、模特图；已选中画布图片时，点击槽位可直接填充，未选中则打开上传。
+              </div>
+            </div>
+          )}
+
           {isVideoModel && videoFrameSlots.length > 0 && (
             <div className="mb-11 px-4 pb-1">
               <div className="flex flex-row gap-2">
@@ -8834,6 +9280,20 @@ export function CanvasChatPanel({ chatId, initialMessages = [], onClose, onOpenF
                   }}
                 />
               ))}
+              {agentUploadSlots.map((_, index) => (
+                <input
+                  key={`canvas-agent-slot-input-${index}`}
+                  type="file"
+                  accept="image/*"
+                  className="hidden"
+                  ref={(element) => {
+                    agentSlotInputRefs.current[index] = element;
+                  }}
+                  onChange={(event) => {
+                    void handleAgentSlotFileChange(index, event);
+                  }}
+                />
+              ))}
               <DropdownMenu>
                 <DropdownMenuTrigger asChild>
                   <Button

components/canvas/nodes/ImageNode.tsx

@@ -34,6 +34,7 @@ import { fetchUploadWithRetry } from "@/lib/client/upload-with-retry";
 import { AngleEditDialog } from "@/components/angle-edit-dialog";
 import { SceneFusionDialog } from "@/components/scene-fusion-dialog";
 import { ImageAnnotator } from "@/components/image-annotator";
+import { AgentExecutionSnapshotDialog } from "@/components/agents/AgentExecutionSnapshotDialog";
 import type { PickedObject } from "@/store/canvas";
 
 // 已标注的物品标记（本地状态）
@@ -310,6 +311,7 @@ export function ImageNode({ node, isSelected }: ImageNodeProps) {
   const [fullscreenPreviewOpen, setFullscreenPreviewOpen] = useState(false);
   const [textEditPanelOpen, setTextEditPanelOpen] = useState(false);
   const [notePanelOpen, setNotePanelOpen] = useState(false);
+  const [snapshotDialogOpen, setSnapshotDialogOpen] = useState(false);
   const [textEditItems, setTextEditItems] = useState<ImageTextEditItem[]>(
     () => normalizeTextEditItems(data.textEditItems),
   );
@@ -361,6 +363,13 @@ export function ImageNode({ node, isSelected }: ImageNodeProps) {
   // 当前正在识别的点击位置
   const [currentClick, setCurrentClick] = useState<{ x: number; y: number } | null>(null);
   const displayTitle = getImageNodeTitle(data, node.id);
+  const agentExecutionSnapshotId =
+    typeof data.metadata?.agentExecutionSnapshotId === "string"
+      ? data.metadata.agentExecutionSnapshotId.trim()
+      : "";
+  const shortAgentExecutionSnapshotId = agentExecutionSnapshotId
+    ? agentExecutionSnapshotId.slice(0, 8)
+    : "";
   const isMultiSelected = isSelected && selectedNodeIds.length > 1;
   const compareSource = useMemo(() => {
     const targetIdentity = normalizeImageIdentityUrl(data.src || "");
@@ -2759,6 +2768,19 @@ export function ImageNode({ node, isSelected }: ImageNodeProps) {
             备注
           </div>
         )}
+        {shortAgentExecutionSnapshotId ? (
+          <button
+            type="button"
+            onClick={() => setSnapshotDialogOpen(true)}
+            className={cn(
+              "absolute left-2 inline-flex items-center gap-1 rounded-full bg-black/70 px-2 py-1 text-[11px] text-white shadow-sm transition-colors hover:bg-black/80",
+              data.note && !isSelected ? "top-10" : "top-2"
+            )}
+            title={`执行快照：${agentExecutionSnapshotId}`}
+          >
+            快照 {shortAgentExecutionSnapshotId}
+          </button>
+        ) : null}
       </div>
 
       {/* 角度编辑对话框 */}
@@ -2770,6 +2792,11 @@ export function ImageNode({ node, isSelected }: ImageNodeProps) {
           onComplete={handleAngleEditComplete}
         />
       )}
+      <AgentExecutionSnapshotDialog
+        snapshotId={agentExecutionSnapshotId || null}
+        open={snapshotDialogOpen}
+        onOpenChange={setSnapshotDialogOpen}
+      />
 
       {/* 批量处理对话框 */}
       <SceneFusionDialog

components/canvas/nodes/VideoNode.tsx

@@ -11,6 +11,7 @@ import {
   DropdownMenuTrigger,
 } from "@/components/ui/dropdown-menu";
 import { Slider } from "@/components/ui/slider";
+import { AgentExecutionSnapshotDialog } from "@/components/agents/AgentExecutionSnapshotDialog";
 import { toast } from "sonner";
 import { useCanvasStore } from "@/store/canvas";
 import type { CanvasNode, VideoNodeData } from "@/types/canvas";
@@ -23,11 +24,19 @@ interface VideoNodeProps {
 
 export function VideoNode({ node, isSelected }: VideoNodeProps) {
   const data = node.data as VideoNodeData;
+  const agentExecutionSnapshotId =
+    typeof data.metadata?.agentExecutionSnapshotId === "string"
+      ? data.metadata.agentExecutionSnapshotId.trim()
+      : "";
+  const shortAgentExecutionSnapshotId = agentExecutionSnapshotId
+    ? agentExecutionSnapshotId.slice(0, 8)
+    : "";
   const videoRef = useRef<HTMLVideoElement>(null);
   const [isHovered, setIsHovered] = useState(false);
   const [isPlaying, setIsPlaying] = useState(false);
   const [isMuted, setIsMuted] = useState(true);
   const [progress, setProgress] = useState(0);
+  const [snapshotDialogOpen, setSnapshotDialogOpen] = useState(false);
   
   const updateNode = useCanvasStore((state) => state.updateNode);
   const deleteNode = useCanvasStore((state) => state.deleteNode);
@@ -262,6 +271,22 @@ export function VideoNode({ node, isSelected }: VideoNodeProps) {
         </div>
       )}
 
+      {shortAgentExecutionSnapshotId ? (
+        <button
+          type="button"
+          onClick={() => setSnapshotDialogOpen(true)}
+          className="absolute top-2 left-2 inline-flex items-center gap-1 rounded-full bg-black/70 px-2 py-1 text-[11px] text-white shadow-sm transition-colors hover:bg-black/80"
+          title={`执行快照：${agentExecutionSnapshotId}`}
+        >
+          快照 {shortAgentExecutionSnapshotId}
+        </button>
+      ) : null}
+      <AgentExecutionSnapshotDialog
+        snapshotId={agentExecutionSnapshotId || null}
+        open={snapshotDialogOpen}
+        onOpenChange={setSnapshotDialogOpen}
+      />
+
       {/* 提示词显示 */}
       {data.prompt && isHovered && (
         <div className="absolute bottom-0 left-0 right-0 bg-gradient-to-t from-black/80 to-transparent p-3">

components/chat-shell.tsx

@@ -40,6 +40,7 @@ type ChatShellProps = {
   tutorialVideos: TutorialVideo[];
   tutorialEnabled: boolean;
   rankingEnabled: boolean;
+  agentEnabled: boolean;
 };
 
 export function ChatShell({
@@ -59,6 +60,7 @@ export function ChatShell({
   tutorialVideos,
   tutorialEnabled,
   rankingEnabled,
+  agentEnabled,
 }: ChatShellProps) {
   const pathname = usePathname();
   const isLovartHome = pathname === "/" && homeVariant === "lovart";
@@ -82,6 +84,7 @@ export function ChatShell({
         tutorialVideos={tutorialVideos}
         tutorialEnabled={tutorialEnabled}
         rankingEnabled={rankingEnabled}
+        agentEnabled={agentEnabled}
       />
       <div className="flex h-svh w-full overflow-hidden pl-16 [--sidebar-offset:4rem]">
         <AppSidebar user={user} />

components/top-navbar.tsx

@@ -2,7 +2,7 @@
 
 import { useRouter } from "next/navigation";
 import Link from "next/link";
-import { Settings, ExternalLink, Briefcase, Images, Sparkles, Trophy, Building2, Crown, Medal, Award, Edit, LayoutGrid, PlayCircle, X } from "lucide-react";
+import { Settings, ExternalLink, Briefcase, Images, Sparkles, Trophy, Building2, Crown, Medal, Award, Edit, LayoutGrid, PlayCircle, Wand2, X } from "lucide-react";
 import { Button } from "@/components/ui/button";
 import { useTheme } from "next-themes";
 import { useState, useEffect, useRef } from "react";
@@ -56,6 +56,7 @@ type TopNavbarProps = {
   tutorialVideos?: TutorialVideo[];
   tutorialEnabled?: boolean;
   rankingEnabled?: boolean;
+  agentEnabled?: boolean;
 };
 
 export function TopNavbar({
@@ -71,6 +72,7 @@ export function TopNavbar({
   tutorialVideos = [],
   tutorialEnabled = false,
   rankingEnabled = true,
+  agentEnabled = true,
 }: TopNavbarProps) {
   const router = useRouter();
   const { resolvedTheme } = useTheme();
@@ -134,6 +136,7 @@ export function TopNavbar({
     if (trimmed === "功能教程") return "教程";
     if (trimmed === "精选榜") return "榜单";
     if (trimmed === "图片编辑") return "编辑";
+    if (trimmed === "智能体") return "智能";
     if (trimmed === "无限画布") return "画布";
     if (trimmed === "我的图册") return "图册";
     if (trimmed === "系统设置") return "设置";
@@ -443,6 +446,21 @@ export function TopNavbar({
           <span className={navLabelClass}>{getShortLabel("图片编辑")}</span>
         </Button>
 
+        {/* 智能体模式 */}
+        {agentEnabled && (
+          <button
+            className={navItemBaseClass}
+            onClick={() => {
+              window.location.href = "/canvas?mode=agent";
+            }}
+            type="button"
+            title="智能体"
+          >
+            <Wand2 className="h-5 w-5" />
+            <span className={navLabelClass}>{getShortLabel("智能体")}</span>
+          </button>
+        )}
+
         {/* 无限画布 */}
         <button
           className={navItemBaseClass}

docs/智能体页面开发文档.md

@@ -1,17 +1,19 @@
 # 智能体页面开发文档（基于当前 `ai-chatbot-main`）
 
-> 目标：在现有“无限画布 + 画布对话 + 多模型生成”体系上，新增一个独立的“智能体”工作台页面。  
+> 目标：在现有“无限画布 + 画布对话 + 多模型生成”体系上，直接基于现有画布页改造出“智能体模式”。  
-> 页面形态：左侧半屏展示生成结果画布，右侧半屏展示智能体工具栏；支持公共/私有智能体、分类筛选、用户自建智能体、自建分类、预设词和参考图注入。
+> 页面形态：一侧为智能体界面，另一侧为现有对话框界面；底层仍使用画布项目、画布对话和媒体落盘能力。
 
 ---
 
 ## 1. 需求目标
 
-本期不是重做一套生成系统，而是在现有项目基础上新增“智能体工作台”：
+本期不是重做一套生成系统，而是在现有项目基础上，把现有画布页改造成“智能体模式”：
 
-1. 新增一个页面，名称为“智能体”
+1. 不新增独立业务形态，仍然基于现有画布界面
-2. 页面左半区复用现有无限画布能力，用于展示生成的图片/视频结果
+2. 画布页在“智能体模式”下采用左右双栏：
-3. 页面右半区为智能体面板，分为：
+   - 一侧为智能体面板
+   - 一侧为现有对话框界面
+3. 智能体面板分为：
    - 公共智能体区
    - 私有智能体区
 4. 公共智能体区支持按分类筛选查看
@@ -22,16 +24,20 @@
    - 类型：`生图模型` / `视频模型`
    - 自定义名称
    - 智能体分类
-   - 预设提示词
+   - 选择对应的模型
-   - 参考图
+   - 多个槽位配置
+   - 每个槽位一套提示词
+   - 每个槽位对应一组参考图
+   - 默认参数模板
    - 公开 / 私有
-7. 用户点击某个智能体后，可将其以“标签/卡片”的形式挂载到页面底部对话框
+7. 用户点击某个智能体后，可将其以“标签/卡片”的形式挂载到现有对话框界面
 8. 后端在发送时自动注入该智能体的预设提示词和参考图
-9. 用户只需：
+9. 用户只需在现有对话框中：
-   - 上传自己的图片
+   - 按槽位数量上传对应数量的图片
    - 可选补充提示词
    - 点击发送
-10. 后端完成生图或生视频后，结果直接落到左侧画布区显示
+10. 后端会按槽位顺序一张张生成
+11. 后端完成生图或生视频后，结果先展示在现有对话消息流里，并继续落盘到画布项目数据中
 
 ---
 
@@ -91,39 +97,48 @@
 5. 主要依赖前端拼 `hiddenPrompt`
 6. 无法形成正式智能体资产
 
-因此本功能建议升级为独立的数据模型，而不是继续放在 `SystemSettings` 或 `canvasSkillTags` 中。
+因此本功能建议升级为独立的数据模型，而不是继续放在 `SystemSettings` 或 `canvasSkillTags` 中；但前端页面层面不单独新开一页，而是直接复用当前画布页。
 
 ---
 
 ## 3. 产品形态建议
 
-### 3.1 页面路由
+### 3.1 页面入口
 
-建议新增独立工作台路由，而不是塞进现有 `CanvasChatPanel`：
+建议不新增 `/agents` 独立路由，直接基于现有画布页实现：
 
-- 入口页：`/agents`
+- 继续使用：`/canvas/[id]`
-- 工作页：`/agents/[id]`
+- 增加模式概念：
+  - 标准画布模式
+  - 智能体模式
 
 推荐行为：
 
-1. 访问 `/agents` 时自动打开最近一次智能体工作台
+1. 访问 `/canvas/[id]` 时可通过顶部入口或参数切到“智能体模式”
-2. 若没有工作台，则创建一个新的工作台
+2. 智能体模式底层仍复用 `CanvasProject`
-3. 工作台底层仍复用 `CanvasProject`
+3. 生成结果仍落到当前画布项目中
+4. 后续如需恢复可视化画布编辑，可在同一项目里切回标准画布模式
 
 ### 3.2 页面布局
 
 桌面端建议采用左右 1:1 分栏：
 
-- 左侧 50%：结果画布区
+- 左侧 50%：智能体工具区
-  - 复用 `CanvasWorkspace`
-  - 只展示当前智能体工作台结果
-  - 底部固定一个对话输入区
-- 右侧 50%：智能体工具区
   - 顶部：搜索 / 类型筛选 / 新建按钮
   - 公共区顶部：分类筛选栏（全部 + 分类 Tab/下拉）
   - 中部：公共智能体区
   - 下部：私有智能体区
   - 点击卡片后变为当前激活智能体
+- 右侧 50%：现有对话框界面
+  - 直接复用 `CanvasChatPanel`
+  - 保留原有消息流、上传、模型选择、发送能力
+  - 智能体只作为额外的预设注入层
+
+说明：
+
+- 首版不再把半屏留给 `CanvasWorkspace` 常驻展示
+- 结果展示以现有对话消息流为主
+- 生成结果仍继续落盘到画布项目，后续可以通过“切回标准画布模式”查看和编辑
 
 ### 3.3 交互约束
 
@@ -136,6 +151,8 @@ MVP 建议限制：
 3. 创建为公开后，立即出现在公共区
 4. 参考图统一绑定媒体资产
 5. 公共区默认只支持单分类筛选
+6. 每次执行都生成一份“执行快照”，用于复盘和复现
+7. 首版不改造原对话框核心交互，只在其上叠加智能体注入能力
 
 ---
 
@@ -145,48 +162,86 @@ MVP 建议限制：
 
 1. 用户点击“新建智能体”
 2. 弹出表单，填写：
+   - 智能体头像
    - 智能体名称
    - 智能体类型（生图 / 视频）
    - 可见性（公开 / 私有）
    - 智能体分类（选择已有分类 / 新建分类）
-   - 预设提示词
+   - 对应模型（生图智能体只能选生图模型，视频智能体只能选视频模型）
-   - 默认模型
+   - 多个槽位（支持新增/删除/排序）
-   - 参考图（可多张）
+   - 每个槽位填写：
+     - 提示词
+     - 参考图（可多张）
+     - 可选参数覆盖
 3. 前端先调用上传接口上传参考图
 4. 后端保存智能体记录
 5. 若勾选“公开”，则直接出现在公共区，并归入对应分类
 6. 若勾选“私有”，则出现在私有区；分类字段保留，便于后续扩展
 
+### 4.1.1 公共智能体另存为私有
+
+建议首版就支持：
+
+1. 用户在公共智能体卡片上点击“另存为我的智能体”
+2. 系统复制一份智能体到当前用户名下
+3. 默认写入：
+   - `visibility=private`
+   - `forkedFromPresetId=原公共智能体ID`
+4. 用户可继续修改：
+   - 名称
+   - 提示词
+   - 默认模型
+   - 默认参数模板
+   - 参考图
+
+这样更符合真实使用习惯：公共智能体用于发现，私有智能体用于个性化沉淀。
+
 ### 4.2 使用智能体
 
 1. 用户点击某个智能体卡片
 2. 底部对话框出现智能体标签
 3. 标签展示：
+   - 头像
    - 名称
    - 类型
    - 分类
    - 默认模型
-   - 参考图数量
+   - 槽位数量
-4. 用户上传图片
+   - 需要上传的图片数量
+4. 用户统一上传 3 张素材图：
+   - 产品图1
+   - 产品图2
+   - 模特图
 5. 用户可补充一句额外提示词
 6. 点击发送
 
+使用约束：
+
+- 即使智能体配置了 10 个槽位，前端也只提示上传 3 张素材图
+- 上传数量不足时，不允许发送
+- 这 3 张素材图会被后端在所有槽位中重复使用
+
 ### 4.3 后端执行
 
 1. 请求携带 `agentPresetId`
 2. 后端查询智能体定义
 3. 后端校验当前用户是否有权限使用
 4. 后端合并：
-   - 智能体预设提示词
+   - 智能体槽位列表
-   - 智能体参考图
+   - 每个槽位的提示词
-   - 用户上传图片
+   - 每个槽位的参考图
+   - 智能体默认参数模板
+   - 用户上传的 3 张素材图
    - 用户补充提示词
    - 当前工作台上下文
 5. 根据智能体类型进入：
    - 图片生成链路
    - 视频生成链路
-6. 结果通过现有流式消息返回
+6. 后端按槽位顺序串行执行
-7. 左侧画布自动新增图片或视频节点
+7. 每完成一个槽位，先回一条结果
+8. 结果通过现有流式消息返回
+9. 结果先回到现有对话框消息流
+10. 同时继续写入当前画布项目数据，便于后续切回标准画布查看
 
 ### 4.4 结果展示
 
@@ -202,32 +257,35 @@ MVP 建议限制：
 ### 5.1 总体原则
 
 1. 不新造一套生成后端，继续复用 `/api/chat`
-2. 不复制一套画布，继续复用 `CanvasWorkspace`
+2. 不新造独立页面，直接改造现有 `/canvas/[id]`
-3. 不继续使用系统设置存智能体，改为正式数据表
+3. 不复制现有对话框，继续复用 `CanvasChatPanel`
-4. 智能体 prompt 必须由服务端注入
+4. 不继续使用系统设置存智能体，改为正式数据表
-5. 结果继续走现有流式落画布逻辑
+5. 智能体 prompt 必须由服务端注入
+6. 结果继续走现有消息流展示与画布项目落盘逻辑
+7. 每次发送必须持久化一份不可变的执行快照
 
 ### 5.2 页面结构建议
 
 建议新增文件：
 
-- `app/agents/page.tsx`
-- `app/agents/[id]/page.tsx`
-- `app/agents/[id]/layout.tsx`
-- `components/agents/AgentWorkbench.tsx`
 - `components/agents/AgentPresetPanel.tsx`
 - `components/agents/AgentPresetCard.tsx`
 - `components/agents/AgentCategoryTabs.tsx`
 - `components/agents/AgentPresetDialog.tsx`
-- `components/agents/AgentComposer.tsx`
 - `services/agents/index.ts`
 - `types/agent.ts`
 - `store/agent-workbench.ts`
 
 建议修改文件：
 
+- `app/canvas/[id]/page.tsx`
+  - 增加“智能体模式”布局
+  - 将主界面切为“智能体面板 + 原对话框界面”
 - `components/top-navbar.tsx`
   - 新增“智能体”入口
+  - 入口行为建议直接跳转到最近画布并以 `mode=agent` 打开
+- `components/canvas/CanvasChatPanel.tsx`
+  - 接收 `agentPresetId` / 智能体标签注入
 - `app/(chat)/api/chat/schema.ts`
   - 扩展请求体，允许传入 `agentPresetId`
 - `app/(chat)/api/chat/route.ts`
@@ -239,32 +297,25 @@ MVP 建议限制：
 
 ### 5.3 组件复用策略
 
-#### A. 左侧结果区
+#### A. 智能体面板
 
-直接复用：
+新增左侧 `AgentPresetPanel`：
 
-- `CanvasWorkspace`
+- 负责公共/私有智能体列表
-- `store/canvas`
+- 分类筛选
-- `canvasService`
+- 新建智能体
+- 选择智能体
 
-#### B. 发送能力
+#### B. 对话能力
 
-建议从 `CanvasChatPanel.tsx` 中抽离一个可复用运行时层，例如：
+直接复用 `CanvasChatPanel`：
 
-- `hooks/useCanvasChatRuntime.ts`
+- 保留现有消息流
-  - 负责：
+- 保留现有上传/发送能力
-    - `sendMessage`
+- 在发送前额外注入 `agentPresetId`
-    - 模型选择
+- 在输入区显示当前选中的智能体标签
-    - 图片/视频参数
-    - 流式结果消费
-    - 结果写入画布
 
-然后：
+这样可以最大化复用现有画布页和现有对话框。
-
-- 旧页面继续使用 `CanvasChatPanel`
-- 新页面使用更轻量的 `AgentComposer`
-
-这样可以避免把 `CanvasChatPanel` 再复制一份。
 
 ---
 
@@ -282,12 +333,18 @@ MVP 建议限制：
 | `type` | varchar(16) | `image` / `video` |
 | `visibility` | varchar(16) | `public` / `private` |
 | `categoryId` | uuid nullable | 智能体分类 |
-| `promptTemplate` | text | 预设提示词 |
+| `promptTemplate` | text | 首槽位提示词摘要/检索字段 |
 | `description` | text | 可选描述 |
 | `defaultModelId` | uuid nullable | 默认模型，关联 `AiModel.id` |
-| `coverAssetId` | uuid nullable | 封面图，可选 |
+| `defaultParams` | jsonb nullable | 默认参数模板 |
+| `slotCount` | integer | 槽位数量 |
+| `requiredUploadCount` | integer | 需要上传的图片数量 |
+| `coverAssetId` | uuid nullable | 智能体头像 |
 | `status` | varchar(16) | `active` / `disabled` |
 | `sortOrder` | integer | 排序 |
+| `usageCount` | integer | 使用次数 |
+| `lastUsedAt` | timestamp nullable | 最近使用时间 |
+| `forkedFromPresetId` | uuid nullable | 从哪个公共智能体复制而来 |
 | `createdAt` | timestamp | 创建时间 |
 | `updatedAt` | timestamp | 更新时间 |
 
@@ -295,10 +352,35 @@ MVP 建议限制：
 
 - `visibility=public`：所有已登录用户可见
 - `visibility=private`：仅创建者可见
-- `promptTemplate` 为服务端最终合成提示词的主模板
+- `promptTemplate` 退化为摘要字段，真正执行以槽位配置为准
 - `categoryId` 首期主要用于公共区分类筛选，也可保留给私有区未来扩展
+- `defaultParams` 用于沉淀默认比例、张数、视频模式、时长等参数，不只存默认模型
+- `forkedFromPresetId` 用于支持“另存为我的智能体”和后续来源追踪
+- `usageCount`、`lastUsedAt` 便于做热门排序、最近使用和运营分析
 
-### 6.2 新增表：`AgentPresetCategory`
+### 6.2 新增表：`AgentPresetSlot`
+
+每个槽位代表“一套提示词 + 一组参考图 + 一张用户上传图”。
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `id` | uuid | 主键 |
+| `presetId` | uuid | 关联 `AgentPreset.id` |
+| `slotIndex` | integer | 槽位顺序，从 0 开始 |
+| `title` | varchar(128) nullable | 槽位名称 |
+| `promptTemplate` | text | 该槽位提示词 |
+| `defaultParams` | jsonb nullable | 槽位级参数覆盖 |
+| `requireUpload` | boolean | 是否要求用户上传对应图片 |
+| `uploadLabel` | varchar(128) nullable | 上传提示文案 |
+| `createdAt` | timestamp | 创建时间 |
+| `updatedAt` | timestamp | 更新时间 |
+
+说明：
+
+- 当前版本默认 `requiredUploadCount=3`
+- 执行时按 `slotIndex` 顺序串行处理
+- 后续若需要“部分槽位不上传图”，可通过 `requireUpload=false` 扩展
+### 6.3 新增表：`AgentPresetCategory`
 
 用于给公共智能体区提供分类筛选，也支持用户创建新分类。
 
@@ -320,8 +402,10 @@ MVP 建议限制：
 - 若智能体为公开，分类建议归入 `public` 或 `global`
 - 公共区默认展示 `public/global` 分类
 - 若后续要治理分类，可以再补后台合并/禁用能力
+- 分类创建时应先做 `slug` 标准化，再做重名复用，避免出现多个语义相同分类
+- 后续后台需支持“分类合并”，用于把重复分类归并到一个正式分类下
 
-### 6.3 新增表：`AgentPresetReferenceAsset`
+### 6.4 新增表：`AgentPresetReferenceAsset`
 
 用于绑定参考图，避免把 URL 直接写死在智能体表。
 
@@ -329,6 +413,7 @@ MVP 建议限制：
 |---|---|---|
 | `id` | uuid | 主键 |
 | `presetId` | uuid | 关联 `AgentPreset.id` |
+| `slotId` | uuid nullable | 关联 `AgentPresetSlot.id` |
 | `assetId` | uuid | 关联 `MediaAsset.id` |
 | `sortOrder` | integer | 参考图顺序 |
 | `createdAt` | timestamp | 创建时间 |
@@ -336,11 +421,11 @@ MVP 建议限制：
 说明：
 
 - 参考图建议全部走 `MediaAsset`
+- 每张参考图应归属于某个槽位
 - 后续若要做公共智能体复制、统计、缓存，会更稳定
+### 6.5 扩展表：`CanvasProject`
 
-### 6.4 扩展表：`CanvasProject`
+为了让同一个画布项目支持“标准画布模式 / 智能体模式”切换，建议给 `CanvasProject` 增加：
-
-为了让 `/agents` 能像 `/canvas` 一样恢复最近工作区，建议给 `CanvasProject` 增加：
 
 | 字段 | 类型 | 说明 |
 |---|---|---|
@@ -349,9 +434,44 @@ MVP 建议限制：
 默认值：
 
 - 旧画布默认 `canvas`
-- 智能体工作台创建时写 `agent`
+- 当画布被用作智能体模式项目时写 `agent`
 
-### 6.5 迁移建议
+### 6.6 新增表：`AgentPresetExecutionSnapshot`
+
+用于记录每一次发送时实际生效的智能体快照，保证结果可追溯、可复盘。
+
+| 字段 | 类型 | 说明 |
+|---|---|---|
+| `id` | uuid | 主键 |
+| `presetId` | uuid | 关联 `AgentPreset.id` |
+| `projectId` | uuid nullable | 工作台/画布项目 |
+| `chatId` | uuid nullable | 对话 ID |
+| `messageId` | uuid nullable | 触发消息 ID |
+| `userId` | uuid | 执行用户 |
+| `presetName` | varchar(128) | 执行时的智能体名称快照 |
+| `presetType` | varchar(16) | `image` / `video` |
+| `visibility` | varchar(16) | 执行时可见性 |
+| `categoryId` | uuid nullable | 执行时分类 |
+| `categoryName` | varchar(64) nullable | 执行时分类名快照 |
+| `promptTemplate` | text | 执行时预设词快照 |
+| `defaultModelId` | uuid nullable | 执行时默认模型 |
+| `resolvedModelId` | uuid nullable | 实际执行模型 |
+| `defaultParams` | jsonb nullable | 执行时默认参数模板 |
+| `slotCount` | integer | 执行时槽位数量 |
+| `requiredUploadCount` | integer | 执行时要求上传的图片数量 |
+| `slots` | jsonb | 执行时完整槽位快照 |
+| `referenceAssetIds` | jsonb | 执行时参考图资产 ID 列表 |
+| `referenceImageUrls` | jsonb | 执行时参考图 URL 列表 |
+| `userInputText` | text | 用户补充提示词 |
+| `requestPayload` | jsonb | 最终调用载荷摘要 |
+| `createdAt` | timestamp | 创建时间 |
+
+说明：
+
+- 该表是“不可变执行记录”，不随智能体后续编辑而改变
+- 用于排查“为什么同一个智能体今天出图和昨天不一样”
+- 也可用于后续做“复跑本次配置”
+### 6.7 迁移建议
 
 当前 `lib/db/migrations` 已存在到 `0052`，本期建议新增：
 
@@ -373,6 +493,7 @@ MVP 建议限制：
 - `type=image | video`
 - `categoryId=xxx`
 - `keyword=xxx`
+- `sort=latest | hottest | most_used`
 
 返回建议：
 
@@ -400,9 +521,23 @@ MVP 建议限制：
   "visibility": "private",
   "categoryId": "uuid",
   "newCategoryName": "电商主图",
-  "promptTemplate": "你是资深电商视觉设计师，请在保持主体一致的前提下优化主图视觉冲击力。",
   "defaultModelId": "uuid",
-  "referenceAssetIds": ["uuid1", "uuid2"]
+  "defaultParams": {
+    "aspectRatio": "1:1",
+    "imageCount": 1
+  },
+  "slots": [
+    {
+      "title": "第1张主图",
+      "promptTemplate": "请把用户上传的商品图做成干净高级的主图视觉。",
+      "referenceAssetIds": ["uuid1"]
+    },
+    {
+      "title": "第2张卖点图",
+      "promptTemplate": "请基于用户上传的商品图突出核心卖点与材质质感。",
+      "referenceAssetIds": ["uuid2"]
+    }
+  ]
 }
 ```
 
@@ -410,6 +545,10 @@ MVP 建议限制：
 
 - `categoryId` 和 `newCategoryName` 二选一
 - 若传了 `newCategoryName`，后端先创建或复用同名分类，再回填 `categoryId`
+- `slots` 至少 1 组，最多建议 20 组
+- 每个槽位对应 1 次顺序生成
+- 用户使用时统一只上传 3 张素材图：产品图1、产品图2、模特图
+- 分类创建前应做名称清洗、`slug` 标准化和重名复用
 
 #### `GET /api/agents/[id]`
 
@@ -423,6 +562,23 @@ MVP 建议限制：
 
 用途：删除智能体
 
+#### `POST /api/agents/[id]/fork`
+
+用途：把公共智能体复制为“我的智能体”
+
+请求体建议：
+
+```json
+{
+  "name": "我的商品主图增强"
+}
+```
+
+返回：
+
+- 一份新的 `visibility=private` 智能体
+- `forkedFromPresetId` 指向原公共智能体
+
 #### `GET /api/agent-categories`
 
 用途：获取智能体分类列表
@@ -490,21 +646,51 @@ agentPresetType?: "image" | "video";
 
 最终执行输入建议：
 
-1. 智能体预设提示词：来自 `AgentPreset.promptTemplate`
+1. 智能体槽位：来自 `AgentPresetSlot`
-2. 智能体参考图：来自 `AgentPresetReferenceAsset`
+2. 槽位参考图：来自 `AgentPresetReferenceAsset`
-3. 用户上传图：来自本轮 `message.parts`
+3. 智能体默认参数模板：来自 `AgentPreset.defaultParams`
-4. 用户补充提示词：来自本轮 `text part`
+4. 用户上传图：按槽位顺序映射
-5. 当前工作台上下文：可复用现有 `contextPayload`
+5. 用户补充提示词：来自本轮 `text part`
+6. 当前工作台上下文：可复用现有 `contextPayload`
 
 合成优先级建议：
 
 ```text
 系统路由/安全限制
-  -> 智能体 promptTemplate
+  -> 按槽位逐个取出 promptTemplate
+  -> 智能体 defaultParams
   -> 用户补充提示词
   -> 画布上下文 autoContext/contextPayload
 ```
 
+#### 顺序执行规则
+
+后端应按槽位串行执行，而不是一次并发全部发起：
+
+1. 校验用户上传图片数量是否满足 `requiredUploadCount`
+2. 用户统一上传 3 张素材图：产品图1、产品图2、模特图
+3. 每个槽位都会复用这 3 张素材图
+4. 依次类推，直到最后一个槽位
+5. 每个槽位完成后立即回传对应结果
+
+例如：
+
+- 智能体配置了 `10` 个槽位
+- 用户仍然只上传 `3` 张素材图
+- 则发送前前端明确提示“请上传产品图1、产品图2、模特图”
+- 后端按顺序一张张生成，共生成 10 次，并在每个槽位重复使用这 3 张素材图
+
+#### 执行快照写入策略
+
+每次使用智能体发送时，后端建议同步写入一条 `AgentPresetExecutionSnapshot`：
+
+1. 读取智能体定义
+2. 解析本轮用户上传图
+3. 生成本次执行的槽位展开结果
+4. 计算最终执行模型与参数
+5. 在真正调用模型前写入快照
+6. 将快照 ID 回写到消息或结果节点 metadata
+
 #### 类型路由规则
 
 - `image` 智能体：
@@ -543,18 +729,21 @@ agentPresetType?: "image" | "video";
 - 分类名称
 - 可见性标签（公共 / 私有）
 - 默认模型名称
+- 默认参数摘要
 - 参考图数量
+- 智能体头像
 - 创建人（公共区建议展示）
+- 使用次数 / 最近更新（公共区建议展示）
 
 ### 8.2 对话输入区
 
-底部输入区建议包含：
+现有对话框界面建议包含：
 
 1. 当前激活智能体标签
-2. 图片上传入口
+2. 原有图片上传入口
-3. 补充提示词输入框
+3. 原有输入框
-4. 模型选择器
+4. 原有模型选择器
-5. 发送按钮
+5. 原有发送按钮
 
 标签行为：
 
@@ -563,27 +752,61 @@ agentPresetType?: "image" | "video";
 - 切换到另一个智能体 -> 替换当前标签
 
 MVP 不建议一次选择多个智能体。
+首版尽量不改原有输入区主结构，只在现有输入区上增加“智能体标签”和必要的预设注入提示。
 
 ### 8.3 新建智能体弹窗建议
 
 表单字段建议：
 
-1. 智能体名称
+1. 智能体头像
-2. 智能体类型
+2. 智能体名称
-3. 可见性
+3. 智能体类型
-4. 分类选择
+4. 可见性
+5. 分类选择
    - 下拉展示已有分类
    - 支持“新建分类”
    - 选择“新建分类”后出现输入框
-5. 默认模型
+6. 对应模型选择
-6. 预设提示词
+   - 若类型为 `image`，只展示支持生图的模型
-7. 参考图上传
+   - 若类型为 `video`，只展示视频模型
+7. 默认参数模板
+8. 槽位配置区
+   - 支持新增/删除/排序槽位
+   - 每个槽位填写：
+     - 提示词
+     - 参考图
+     - 可选参数覆盖
+     - 上传提示语
 
 交互建议：
 
 - 当用户选择 `visibility=public` 时，分类必填
 - 当用户选择 `visibility=private` 时，分类可选
 - 若输入的新分类名已存在，后端直接复用
+- 当用户切换智能体类型时，模型下拉列表要立即联动刷新
+- 若用户已选模型与新类型不兼容，前端应清空当前模型并提示重新选择
+- 增加一个“预览执行内容”区域，展示将被注入的槽位数、默认模型、默认参数和总参考图数量
+- 即使槽位数为 10，也应明确展示“使用时需要上传 3 张素材图：产品图1、产品图2、模特图”
+
+### 8.3.1 使用时的上传校验
+
+当用户选中某个智能体后，现有对话框界面需要显示：
+
+- 该智能体共有多少个槽位
+- 本次必须上传多少张图片
+- 固定上传哪几类图片
+
+建议交互：
+
+1. 当前版本 `requiredUploadCount = 3`
+   - 输入区上方明确显示：`请上传产品图1、产品图2、模特图`
+2. 未达到数量时：
+   - 发送按钮禁用
+   - 或点击后给出明确报错
+3. 达到数量后：
+   - 在附件区按顺序展示 `产品图1 / 产品图2 / 模特图`
+4. 槽位生成阶段：
+   - 系统在每个槽位重复使用这 3 张素材图
 
 ### 8.4 模型选择策略
 
@@ -592,19 +815,22 @@ MVP 不建议一次选择多个智能体。
 - 生图智能体：只显示支持图片生成的模型
 - 视频智能体：只显示视频模型
 
+创建/编辑智能体时建议直接强制要求用户选择对应模型，不建议首版留空走系统默认，否则公共智能体可复现性会变差。
+
 默认行为：
 
 1. 若智能体配置了默认模型，则优先用该模型
-2. 若未配置，则走系统默认模型
+2. 若智能体配置了默认参数模板，则优先套用
-3. 用户可手动改模型，但必须满足类型限制
+3. 若未配置，则走系统默认模型和系统默认参数
+4. 用户可手动改模型，但必须满足类型限制
 
 ### 8.5 结果落画布
 
-继续复用 `CanvasChatPanel` 现有逻辑中的结果落图机制：
+继续复用 `CanvasChatPanel` 的现有结果消费机制：
 
-- 消费 `data-canvasMedia`
+- 前端主展示：现有对话消息流
-- 图片新增 `image` 节点
+- 项目持久化：继续写入当前画布项目
-- 视频新增 `video` 节点
+- 后续查看：可切回标准画布模式查看图片/视频节点
 
 建议给节点 `metadata` 增加：
 
@@ -612,6 +838,7 @@ MVP 不建议一次选择多个智能体。
 - `agentPresetName`
 - `agentCategoryId`
 - `agentCategoryName`
+- `agentExecutionSnapshotId`
 - `workspaceType: "agent"`
 
 ---
@@ -628,6 +855,7 @@ MVP 不建议一次选择多个智能体。
    - `promptTemplate`
    - `reference assets`
    - `default model`
+   - `default params`
 4. 服务端统一构造最终调用参数
 
 价值：
@@ -650,7 +878,42 @@ MVP 不建议一次选择多个智能体。
 - 再拼用户上传图
 - 去重后按模型能力截断
 
-### 9.3 公开/私有权限
+### 9.3 默认参数模板
+
+建议图片和视频智能体都支持默认参数模板：
+
+- 图片智能体：
+  - `aspectRatio`
+  - `imageCount`
+  - `imageSize`
+  - `negativePrompt`
+- 视频智能体：
+  - `mode`
+  - `duration`
+  - `ratio`
+  - `resolution`
+  - `cameraMove`
+
+这样智能体真正能沉淀“风格化工作方式”，而不是只沉淀一句提示词。
+
+### 9.3.1 对应模型选择规则
+
+创建或编辑智能体时，模型选择必须和智能体类型严格匹配：
+
+- `image` 智能体：
+  - 只能选择支持生图的模型
+  - 不能选择视频模型
+- `video` 智能体：
+  - 只能选择视频模型
+
+建议前后端都做校验：
+
+- 前端：只展示对应类型模型
+- 后端：再次校验提交的 `defaultModelId`
+
+后端若发现模型不匹配，直接拒绝保存，避免出现“视频智能体绑定生图模型”这类脏数据
+
+### 9.4 公开/私有权限
 
 私有智能体：
 
@@ -665,13 +928,24 @@ MVP 不建议一次选择多个智能体。
 
 - 默认所有登录用户可见
 - 若分类由普通用户创建，建议允许复用，但保留后续后台禁用/合并能力
+- 分类创建应遵循：名称清洗 -> `slug` 标准化 -> 重名复用 -> 后台可合并
 
-### 9.4 公共智能体是否立即发布
+### 9.5 公共智能体是否立即发布
 
 按当前需求，本期可直接采用：
 
 - 创建时选“公开” -> 立即出现在公共区对应分类下
 
+### 9.6 公共智能体复制为私有
+
+建议首版就支持：
+
+1. 公共智能体允许“另存为我的智能体”
+2. 复制后自动转为 `private`
+3. 保留原分类或允许用户重新选择分类
+4. 继承原智能体的参考图、提示词、默认模型、默认参数模板
+5. 记录 `forkedFromPresetId`
+
 ---
 
 ## 10. 与现有 `canvasSkillTags` 的兼容方案
@@ -680,15 +954,16 @@ MVP 不建议一次选择多个智能体。
 
 ### Phase A：先并行存在
 
-- 旧画布页面继续使用 `canvasSkillTags`
+- 旧画布页的常规模式继续可用
-- 新“智能体页面”使用 `AgentPreset`
+- 同一画布页新增“智能体模式”
+- `canvasSkillTags` 暂不移除
 
 ### Phase B：后续统一
 
 待智能体稳定后，可考虑：
 
-1. 将公共智能体映射到画布页面
+1. 用 `AgentPreset` 替换 `canvasSkillTags`
-2. 用 `AgentPreset` 替换 `canvasSkillTags`
+2. 标准画布模式和智能体模式共享同一套智能体预设
 3. 后台不再维护两套预设
 
 ---
@@ -700,26 +975,29 @@ MVP 不建议一次选择多个智能体。
 1. 新增 `AgentPreset`
 2. 新增 `AgentPresetCategory`
 3. 新增 `AgentPresetReferenceAsset`
-4. 扩展 `CanvasProject.workspaceType`
+4. 新增 `AgentPresetExecutionSnapshot`
-5. 增加 `lib/db/queries.ts` 查询函数
+5. 扩展 `CanvasProject.workspaceType`
-6. 增加 `/api/agents` 与 `/api/agent-categories`
+6. 增加 `lib/db/queries.ts` 查询函数
+7. 增加 `/api/agents` 与 `/api/agent-categories`
 
 ### 第二阶段：页面骨架
 
-1. 新增 `/agents`
+1. 改造 `app/canvas/[id]/page.tsx`
-2. 新增 `/agents/[id]`
+2. 增加“智能体模式”开关
-3. 左侧接入 `CanvasWorkspace`
+3. 左侧接入智能体列表
-4. 右侧接入智能体列表
+4. 右侧保留现有 `CanvasChatPanel`
 5. 接入公共区分类筛选
 6. 增加“新建智能体”弹窗
+7. 增加“另存为我的智能体”
 
 ### 第三阶段：发送链路
 
-1. 抽离 `CanvasChatPanel` 发送运行时
+1. `CanvasChatPanel` 支持接收 `agentPresetId`
-2. 新增 `AgentComposer`
+2. 发送时带 `agentPresetId`
-3. 发送时带 `agentPresetId`
+3. 后端 `/api/chat` 注入智能体定义
-4. 后端 `/api/chat` 注入智能体定义
+4. 写入执行快照
-5. 结果复用 `data-canvasMedia` 落到画布
+5. 结果继续走现有消息流展示
+6. 同时落到当前画布项目
 
 ### 第四阶段：体验增强
 
@@ -729,6 +1007,7 @@ MVP 不建议一次选择多个智能体。
 4. 参考图拖拽排序
 5. 最近使用智能体
 6. 公共智能体卡片展示创建者/热度
+7. 分类合并/禁用
 
 ---
 
@@ -736,21 +1015,31 @@ MVP 不建议一次选择多个智能体。
 
 满足以下条件即可视为第一版完成：
 
-1. 可以从页面入口进入“智能体”工作台
+1. 可以从现有画布页进入“智能体模式”
-2. 左侧能看到画布结果区域
+2. 一侧能看到智能体界面
-3. 右侧能看到公共区、私有区
+3. 一侧能看到现有对话框界面
 4. 公共区可以按分类筛选智能体
 5. 新建智能体时可以选择已有分类
 6. 新建智能体时可以创建新分类
-7. 可以新建一个生图智能体
+7. 创建分类时会做重名复用，不会重复创建近义同名分类
-8. 可以新建一个视频智能体
+8. 可以新建一个生图智能体
-9. 可以上传参考图并成功保存
+9. 可以新建一个视频智能体
-10. 可以选择公开或私有
+10. 创建智能体时必须能选择对应的生图模型或视频模型
-11. 点击智能体卡片后，底部输入区出现智能体标签
+11. 智能体支持多槽位配置
-12. 上传用户图片并发送后，后端会自动注入智能体预设
+12. 智能体支持上传头像并展示
-13. 生图结果会自动落到左侧画布
+13. 即使智能体有 10 个槽位，使用时也只提示上传 3 张素材图
-14. 生视频结果会自动落到左侧画布
+14. 上传数量不足时不允许发送
-15. 其他用户不能使用当前用户的私有智能体
+15. 智能体支持默认模型和默认参数模板
+16. 可以上传参考图并成功保存
+17. 可以选择公开或私有
+18. 公共智能体支持“另存为我的智能体”
+19. 点击智能体卡片后，底部输入区出现智能体标签
+20. 上传用户图片并发送后，后端会自动注入智能体预设
+21. 每次发送都会写入执行快照
+22. 生图结果会先出现在现有对话消息流中
+23. 生视频结果会先出现在现有对话消息流中
+24. 生成结果会同步落到当前画布项目
+25. 其他用户不能使用当前用户的私有智能体
 
 ---
 
@@ -768,7 +1057,6 @@ MVP 不建议一次选择多个智能体。
 
 ## 14. 一句话结论
 
-这次“智能体页面”最合适的实现方式是：
+这次“智能体功能”最合适的实现方式是：
-
-**复用现有无限画布和 `/api/chat` 生成主链路，在其上新增正式的 `AgentPreset + AgentPresetCategory` 数据模型与工作台页面，而不是重新做一套生成系统。**
 
+**不新开独立页面，而是直接改造现有 `/canvas/[id]`：一侧做智能体界面，一侧保留现有 `CanvasChatPanel`，底层继续复用 `CanvasProject + /api/chat + MediaAsset`。**

lib/db/migrations/0053_add_agent_presets.sql

@@ -0,0 +1,179 @@
+CREATE TABLE IF NOT EXISTS "AgentPresetCategory" (
+  "id" uuid PRIMARY KEY DEFAULT gen_random_uuid() NOT NULL,
+  "name" varchar(64) NOT NULL,
+  "slug" varchar(64) NOT NULL,
+  "scope" varchar(16) DEFAULT 'public' NOT NULL,
+  "createdBy" uuid,
+  "sortOrder" integer DEFAULT 0 NOT NULL,
+  "isEnabled" boolean DEFAULT true NOT NULL,
+  "createdAt" timestamp DEFAULT now() NOT NULL,
+  "updatedAt" timestamp DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetCategory" ADD CONSTRAINT "AgentPresetCategory_createdBy_User_id_fk" FOREIGN KEY ("createdBy") REFERENCES "public"."User"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+
+CREATE TABLE IF NOT EXISTS "AgentPreset" (
+  "id" uuid PRIMARY KEY DEFAULT gen_random_uuid() NOT NULL,
+  "ownerUserId" uuid NOT NULL,
+  "name" varchar(128) NOT NULL,
+  "type" varchar(16) DEFAULT 'image' NOT NULL,
+  "visibility" varchar(16) DEFAULT 'private' NOT NULL,
+  "categoryId" uuid,
+  "promptTemplate" text DEFAULT '' NOT NULL,
+  "description" text,
+  "defaultModelId" uuid,
+  "defaultParams" jsonb,
+  "coverAssetId" uuid,
+  "status" varchar(16) DEFAULT 'active' NOT NULL,
+  "sortOrder" integer DEFAULT 0 NOT NULL,
+  "usageCount" integer DEFAULT 0 NOT NULL,
+  "lastUsedAt" timestamp,
+  "forkedFromPresetId" uuid,
+  "createdAt" timestamp DEFAULT now() NOT NULL,
+  "updatedAt" timestamp DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPreset" ADD CONSTRAINT "AgentPreset_ownerUserId_User_id_fk" FOREIGN KEY ("ownerUserId") REFERENCES "public"."User"("id") ON DELETE cascade ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPreset" ADD CONSTRAINT "AgentPreset_categoryId_AgentPresetCategory_id_fk" FOREIGN KEY ("categoryId") REFERENCES "public"."AgentPresetCategory"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPreset" ADD CONSTRAINT "AgentPreset_defaultModelId_AiModel_id_fk" FOREIGN KEY ("defaultModelId") REFERENCES "public"."AiModel"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+
+ALTER TABLE "CanvasProject"
+  ADD COLUMN IF NOT EXISTS "workspaceType" varchar(16) DEFAULT 'canvas' NOT NULL;
+--> statement-breakpoint
+
+CREATE TABLE IF NOT EXISTS "AgentPresetExecutionSnapshot" (
+  "id" uuid PRIMARY KEY DEFAULT gen_random_uuid() NOT NULL,
+  "presetId" uuid NOT NULL,
+  "projectId" uuid,
+  "chatId" uuid,
+  "messageId" uuid,
+  "userId" uuid NOT NULL,
+  "presetName" varchar(128) NOT NULL,
+  "presetType" varchar(16) NOT NULL,
+  "visibility" varchar(16) NOT NULL,
+  "categoryId" uuid,
+  "categoryName" varchar(64),
+  "promptTemplate" text DEFAULT '' NOT NULL,
+  "defaultModelId" uuid,
+  "resolvedModelId" uuid,
+  "defaultParams" jsonb,
+  "referenceAssetIds" jsonb DEFAULT '[]'::jsonb NOT NULL,
+  "referenceImageUrls" jsonb DEFAULT '[]'::jsonb NOT NULL,
+  "userInputText" text,
+  "requestPayload" jsonb,
+  "createdAt" timestamp DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_presetId_AgentPreset_id_fk" FOREIGN KEY ("presetId") REFERENCES "public"."AgentPreset"("id") ON DELETE cascade ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_projectId_CanvasProject_id_fk" FOREIGN KEY ("projectId") REFERENCES "public"."CanvasProject"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_chatId_Chat_id_fk" FOREIGN KEY ("chatId") REFERENCES "public"."Chat"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_messageId_Message_v2_id_fk" FOREIGN KEY ("messageId") REFERENCES "public"."Message_v2"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_userId_User_id_fk" FOREIGN KEY ("userId") REFERENCES "public"."User"("id") ON DELETE cascade ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_categoryId_AgentPresetCategory_id_fk" FOREIGN KEY ("categoryId") REFERENCES "public"."AgentPresetCategory"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_defaultModelId_AiModel_id_fk" FOREIGN KEY ("defaultModelId") REFERENCES "public"."AiModel"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetExecutionSnapshot" ADD CONSTRAINT "AgentPresetExecutionSnapshot_resolvedModelId_AiModel_id_fk" FOREIGN KEY ("resolvedModelId") REFERENCES "public"."AiModel"("id") ON DELETE set null ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+
+CREATE TABLE IF NOT EXISTS "AgentPresetReferenceAsset" (
+  "id" uuid PRIMARY KEY DEFAULT gen_random_uuid() NOT NULL,
+  "presetId" uuid NOT NULL,
+  "assetId" uuid NOT NULL,
+  "sortOrder" integer DEFAULT 0 NOT NULL,
+  "createdAt" timestamp DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetReferenceAsset" ADD CONSTRAINT "AgentPresetReferenceAsset_presetId_AgentPreset_id_fk" FOREIGN KEY ("presetId") REFERENCES "public"."AgentPreset"("id") ON DELETE cascade ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetReferenceAsset" ADD CONSTRAINT "AgentPresetReferenceAsset_assetId_MediaAsset_id_fk" FOREIGN KEY ("assetId") REFERENCES "public"."MediaAsset"("id") ON DELETE cascade ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+
+CREATE INDEX IF NOT EXISTS "AgentPresetCategory_scope_sort_idx"
+  ON "AgentPresetCategory"("scope", "sortOrder", "name");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPresetCategory_slug_scope_idx"
+  ON "AgentPresetCategory"("slug", "scope");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPreset_owner_visibility_idx"
+  ON "AgentPreset"("ownerUserId", "visibility", "updatedAt");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPreset_category_visibility_idx"
+  ON "AgentPreset"("categoryId", "visibility", "updatedAt");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPreset_type_visibility_idx"
+  ON "AgentPreset"("type", "visibility", "updatedAt");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPresetExecutionSnapshot_preset_created_idx"
+  ON "AgentPresetExecutionSnapshot"("presetId", "createdAt");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPresetExecutionSnapshot_user_created_idx"
+  ON "AgentPresetExecutionSnapshot"("userId", "createdAt");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPresetReferenceAsset_preset_sort_idx"
+  ON "AgentPresetReferenceAsset"("presetId", "sortOrder");

lib/db/migrations/0054_add_agent_preset_slots.sql

@@ -0,0 +1,52 @@
+ALTER TABLE "AgentPreset"
+  ADD COLUMN IF NOT EXISTS "slotCount" integer DEFAULT 1 NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "AgentPreset"
+  ADD COLUMN IF NOT EXISTS "requiredUploadCount" integer DEFAULT 1 NOT NULL;
+--> statement-breakpoint
+
+CREATE TABLE IF NOT EXISTS "AgentPresetSlot" (
+  "id" uuid PRIMARY KEY DEFAULT gen_random_uuid() NOT NULL,
+  "presetId" uuid NOT NULL,
+  "slotIndex" integer NOT NULL,
+  "title" varchar(128),
+  "promptTemplate" text DEFAULT '' NOT NULL,
+  "defaultParams" jsonb,
+  "requireUpload" boolean DEFAULT true NOT NULL,
+  "uploadLabel" varchar(128),
+  "createdAt" timestamp DEFAULT now() NOT NULL,
+  "updatedAt" timestamp DEFAULT now() NOT NULL
+);
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetSlot" ADD CONSTRAINT "AgentPresetSlot_presetId_AgentPreset_id_fk" FOREIGN KEY ("presetId") REFERENCES "public"."AgentPreset"("id") ON DELETE cascade ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+
+ALTER TABLE "AgentPresetReferenceAsset"
+  ADD COLUMN IF NOT EXISTS "slotId" uuid;
+--> statement-breakpoint
+DO $$ BEGIN
+ ALTER TABLE "AgentPresetReferenceAsset" ADD CONSTRAINT "AgentPresetReferenceAsset_slotId_AgentPresetSlot_id_fk" FOREIGN KEY ("slotId") REFERENCES "public"."AgentPresetSlot"("id") ON DELETE cascade ON UPDATE no action;
+EXCEPTION
+ WHEN duplicate_object THEN null;
+END $$;
+--> statement-breakpoint
+
+ALTER TABLE "AgentPresetExecutionSnapshot"
+  ADD COLUMN IF NOT EXISTS "slotCount" integer DEFAULT 1 NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "AgentPresetExecutionSnapshot"
+  ADD COLUMN IF NOT EXISTS "requiredUploadCount" integer DEFAULT 1 NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "AgentPresetExecutionSnapshot"
+  ADD COLUMN IF NOT EXISTS "slots" jsonb DEFAULT '[]'::jsonb NOT NULL;
+--> statement-breakpoint
+
+CREATE INDEX IF NOT EXISTS "AgentPresetSlot_preset_slot_idx"
+  ON "AgentPresetSlot"("presetId", "slotIndex");
+--> statement-breakpoint
+CREATE INDEX IF NOT EXISTS "AgentPresetReferenceAsset_slot_sort_idx"
+  ON "AgentPresetReferenceAsset"("slotId", "sortOrder");

lib/db/migrations/0055_add_agent_enabled_site_setting.sql

@@ -0,0 +1,2 @@
+ALTER TABLE "SiteSettings"
+  ADD COLUMN IF NOT EXISTS "agentEnabled" boolean DEFAULT true NOT NULL;

lib/db/migrations/meta/_journal.json

@@ -274,6 +274,27 @@
       "when": 1772796000000,
       "tag": "0052_add_asset_memory_snapshots",
       "breakpoints": true
+    },
+    {
+      "idx": 39,
+      "version": "7",
+      "when": 1773146400000,
+      "tag": "0053_add_agent_presets",
+      "breakpoints": true
+    },
+    {
+      "idx": 40,
+      "version": "7",
+      "when": 1773148200000,
+      "tag": "0054_add_agent_preset_slots",
+      "breakpoints": true
+    },
+    {
+      "idx": 41,
+      "version": "7",
+      "when": 1773150000000,
+      "tag": "0055_add_agent_enabled_site_setting",
+      "breakpoints": true
     }
   ]
 }

lib/db/schema.ts

@@ -376,6 +376,7 @@ export const siteSettings = pgTable("SiteSettings", {
   rankingEnabled: boolean("rankingEnabled").notNull().default(true), // 是否显示精选榜按钮
   // 无限画布设置
   canvasEnabled: boolean("canvasEnabled").notNull().default(true), // 是否启用无限画布功能
+  agentEnabled: boolean("agentEnabled").notNull().default(true), // 是否启用智能体功能
   realtimeCanvasEnabled: boolean("realtimeCanvasEnabled").notNull().default(true), // 是否显示实时画布入口
   canvasDefaultModelId: varchar("canvasDefaultModelId", { length: 128 }), // 画布默认选择的模型ID（null表示智能自动）
   canvasAutoModeDefaultModelId: varchar("canvasAutoModeDefaultModelId", { length: 128 }), // 智能自动模式下实际使用的默认生图模型ID
@@ -496,6 +497,79 @@ export const generatedImage = pgTable("GeneratedImage", {
 
 export type GeneratedImage = InferSelectModel<typeof generatedImage>;
 
+// 智能体分类表
+export const agentPresetCategory = pgTable("AgentPresetCategory", {
+  id: uuid("id").primaryKey().notNull().defaultRandom(),
+  name: varchar("name", { length: 64 }).notNull(),
+  slug: varchar("slug", { length: 64 }).notNull(),
+  scope: varchar("scope", { enum: ["public", "private", "global"] })
+    .notNull()
+    .default("public"),
+  createdBy: uuid("createdBy").references(() => user.id, { onDelete: "set null" }),
+  sortOrder: integer("sortOrder").notNull().default(0),
+  isEnabled: boolean("isEnabled").notNull().default(true),
+  createdAt: timestamp("createdAt").notNull().defaultNow(),
+  updatedAt: timestamp("updatedAt").notNull().defaultNow(),
+});
+
+export type AgentPresetCategory = InferSelectModel<typeof agentPresetCategory>;
+
+// 智能体表
+export const agentPreset = pgTable("AgentPreset", {
+  id: uuid("id").primaryKey().notNull().defaultRandom(),
+  ownerUserId: uuid("ownerUserId")
+    .notNull()
+    .references(() => user.id, { onDelete: "cascade" }),
+  name: varchar("name", { length: 128 }).notNull(),
+  type: varchar("type", { enum: ["image", "video"] })
+    .notNull()
+    .default("image"),
+  visibility: varchar("visibility", { enum: ["public", "private"] })
+    .notNull()
+    .default("private"),
+  categoryId: uuid("categoryId").references(() => agentPresetCategory.id, {
+    onDelete: "set null",
+  }),
+  promptTemplate: text("promptTemplate").notNull().default(""),
+  description: text("description"),
+  defaultModelId: uuid("defaultModelId").references(() => aiModel.id, {
+    onDelete: "set null",
+  }),
+  defaultParams: jsonb("defaultParams").$type<Record<string, any>>(),
+  slotCount: integer("slotCount").notNull().default(1),
+  requiredUploadCount: integer("requiredUploadCount").notNull().default(1),
+  coverAssetId: uuid("coverAssetId"),
+  status: varchar("status", { enum: ["active", "disabled", "hidden", "blocked"] })
+    .notNull()
+    .default("active"),
+  sortOrder: integer("sortOrder").notNull().default(0),
+  usageCount: integer("usageCount").notNull().default(0),
+  lastUsedAt: timestamp("lastUsedAt"),
+  forkedFromPresetId: uuid("forkedFromPresetId"),
+  createdAt: timestamp("createdAt").notNull().defaultNow(),
+  updatedAt: timestamp("updatedAt").notNull().defaultNow(),
+});
+
+export type AgentPreset = InferSelectModel<typeof agentPreset>;
+
+// 智能体槽位表（每个槽位=一套提示词 + 一组参考图 + 一张用户上传图）
+export const agentPresetSlot = pgTable("AgentPresetSlot", {
+  id: uuid("id").primaryKey().notNull().defaultRandom(),
+  presetId: uuid("presetId")
+    .notNull()
+    .references(() => agentPreset.id, { onDelete: "cascade" }),
+  slotIndex: integer("slotIndex").notNull(),
+  title: varchar("title", { length: 128 }),
+  promptTemplate: text("promptTemplate").notNull().default(""),
+  defaultParams: jsonb("defaultParams").$type<Record<string, any>>(),
+  requireUpload: boolean("requireUpload").notNull().default(true),
+  uploadLabel: varchar("uploadLabel", { length: 128 }),
+  createdAt: timestamp("createdAt").notNull().defaultNow(),
+  updatedAt: timestamp("updatedAt").notNull().defaultNow(),
+});
+
+export type AgentPresetSlot = InferSelectModel<typeof agentPresetSlot>;
+
 // ============================================
 // 无限画布相关表
 // ============================================
@@ -509,6 +583,9 @@ export const canvasProject = pgTable("CanvasProject", {
   title: varchar("title", { length: 256 }).notNull().default("未命名画布"),
   description: text("description"),
   thumbnail: text("thumbnail"),
+  workspaceType: varchar("workspaceType", { enum: ["canvas", "agent"] })
+    .notNull()
+    .default("canvas"),
   // 存储完整画布状态（节点、边、视口等）
   canvasData: jsonb("canvasData").$type<{
     nodes: any[];
@@ -843,3 +920,78 @@ export const assetMemorySnapshot = pgTable("AssetMemorySnapshot", {
 });
 
 export type AssetMemorySnapshot = InferSelectModel<typeof assetMemorySnapshot>;
+
+// 智能体参考图绑定表
+export const agentPresetReferenceAsset = pgTable("AgentPresetReferenceAsset", {
+  id: uuid("id").primaryKey().notNull().defaultRandom(),
+  presetId: uuid("presetId")
+    .notNull()
+    .references(() => agentPreset.id, { onDelete: "cascade" }),
+  slotId: uuid("slotId").references(() => agentPresetSlot.id, {
+    onDelete: "cascade",
+  }),
+  assetId: uuid("assetId")
+    .notNull()
+    .references(() => mediaAsset.id, { onDelete: "cascade" }),
+  sortOrder: integer("sortOrder").notNull().default(0),
+  createdAt: timestamp("createdAt").notNull().defaultNow(),
+});
+
+export type AgentPresetReferenceAsset = InferSelectModel<typeof agentPresetReferenceAsset>;
+
+// 智能体执行快照表
+export const agentPresetExecutionSnapshot = pgTable("AgentPresetExecutionSnapshot", {
+  id: uuid("id").primaryKey().notNull().defaultRandom(),
+  presetId: uuid("presetId")
+    .notNull()
+    .references(() => agentPreset.id, { onDelete: "cascade" }),
+  projectId: uuid("projectId").references(() => canvasProject.id, {
+    onDelete: "set null",
+  }),
+  chatId: uuid("chatId").references(() => chat.id, { onDelete: "set null" }),
+  messageId: uuid("messageId").references(() => message.id, { onDelete: "set null" }),
+  userId: uuid("userId")
+    .notNull()
+    .references(() => user.id, { onDelete: "cascade" }),
+  presetName: varchar("presetName", { length: 128 }).notNull(),
+  presetType: varchar("presetType", { enum: ["image", "video"] }).notNull(),
+  visibility: varchar("visibility", { enum: ["public", "private"] }).notNull(),
+  categoryId: uuid("categoryId").references(() => agentPresetCategory.id, {
+    onDelete: "set null",
+  }),
+  categoryName: varchar("categoryName", { length: 64 }),
+  promptTemplate: text("promptTemplate").notNull().default(""),
+  defaultModelId: uuid("defaultModelId").references(() => aiModel.id, {
+    onDelete: "set null",
+  }),
+  resolvedModelId: uuid("resolvedModelId").references(() => aiModel.id, {
+    onDelete: "set null",
+  }),
+  defaultParams: jsonb("defaultParams").$type<Record<string, any>>(),
+  slotCount: integer("slotCount").notNull().default(1),
+  requiredUploadCount: integer("requiredUploadCount").notNull().default(1),
+  slots: jsonb("slots")
+    .$type<
+      Array<{
+        slotIndex: number;
+        title?: string | null;
+        promptTemplate: string;
+        defaultParams?: Record<string, any> | null;
+        requireUpload?: boolean;
+        uploadLabel?: string | null;
+        referenceAssetIds: string[];
+        referenceImageUrls: string[];
+      }>
+    >()
+    .notNull()
+    .default([]),
+  referenceAssetIds: jsonb("referenceAssetIds").$type<string[]>().notNull().default([]),
+  referenceImageUrls: jsonb("referenceImageUrls").$type<string[]>().notNull().default([]),
+  userInputText: text("userInputText"),
+  requestPayload: jsonb("requestPayload").$type<Record<string, any>>(),
+  createdAt: timestamp("createdAt").notNull().defaultNow(),
+});
+
+export type AgentPresetExecutionSnapshot = InferSelectModel<
+  typeof agentPresetExecutionSnapshot
+>;

services/agents/index.ts

@@ -0,0 +1,170 @@
+import type {
+  AgentCategory,
+  AgentCategoryCreatePayload,
+  AgentCreatePayload,
+  AgentPreset,
+  AgentPresetListResponse,
+  AgentUpdatePayload,
+} from "@/types/agent";
+
+interface ApiResponse<T = any> {
+  success: boolean;
+  data?: T;
+  error?: string;
+}
+
+async function parseJsonSafely<T>(response: Response): Promise<T | null> {
+  try {
+    return (await response.json()) as T;
+  } catch (_error) {
+    return null;
+  }
+}
+
+export const agentCategoryService = {
+  async getCategories(scope: "all" | "public" | "private" = "public"): Promise<ApiResponse<AgentCategory[]>> {
+    try {
+      const response = await fetch(`/api/agent-categories?scope=${scope}`, {
+        cache: "no-store",
+      });
+      const data = await parseJsonSafely<ApiResponse<AgentCategory[]>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "获取智能体分类失败" };
+      }
+      return data || { success: false, error: "获取智能体分类失败" };
+    } catch (_error) {
+      return { success: false, error: "获取智能体分类失败" };
+    }
+  },
+
+  async createCategory(payload: AgentCategoryCreatePayload): Promise<ApiResponse<AgentCategory>> {
+    try {
+      const response = await fetch("/api/agent-categories", {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify(payload),
+      });
+      const data = await parseJsonSafely<ApiResponse<AgentCategory>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "创建智能体分类失败" };
+      }
+      return data || { success: false, error: "创建智能体分类失败" };
+    } catch (_error) {
+      return { success: false, error: "创建智能体分类失败" };
+    }
+  },
+};
+
+export const agentService = {
+  async getPresets(params?: {
+    scope?: "all" | "public" | "private";
+    type?: "image" | "video";
+    categoryId?: string;
+    keyword?: string;
+    sort?: "latest" | "hottest" | "most_used";
+  }): Promise<ApiResponse<AgentPresetListResponse>> {
+    try {
+      const searchParams = new URLSearchParams();
+      if (params?.scope) searchParams.set("scope", params.scope);
+      if (params?.type) searchParams.set("type", params.type);
+      if (params?.categoryId) searchParams.set("categoryId", params.categoryId);
+      if (params?.keyword) searchParams.set("keyword", params.keyword);
+      if (params?.sort) searchParams.set("sort", params.sort);
+      const query = searchParams.toString();
+      const response = await fetch(`/api/agents${query ? `?${query}` : ""}`, {
+        cache: "no-store",
+      });
+      const data = await parseJsonSafely<ApiResponse<AgentPresetListResponse>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "获取智能体列表失败" };
+      }
+      return data || { success: false, error: "获取智能体列表失败" };
+    } catch (_error) {
+      return { success: false, error: "获取智能体列表失败" };
+    }
+  },
+
+  async getPreset(id: string): Promise<ApiResponse<AgentPreset>> {
+    try {
+      const response = await fetch(`/api/agents/${id}`, { cache: "no-store" });
+      const data = await parseJsonSafely<ApiResponse<AgentPreset>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "获取智能体失败" };
+      }
+      return data || { success: false, error: "获取智能体失败" };
+    } catch (_error) {
+      return { success: false, error: "获取智能体失败" };
+    }
+  },
+
+  async createPreset(payload: AgentCreatePayload): Promise<ApiResponse<AgentPreset>> {
+    try {
+      const response = await fetch("/api/agents", {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify(payload),
+      });
+      const data = await parseJsonSafely<ApiResponse<AgentPreset>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "创建智能体失败" };
+      }
+      return data || { success: false, error: "创建智能体失败" };
+    } catch (_error) {
+      return { success: false, error: "创建智能体失败" };
+    }
+  },
+
+  async updatePreset(id: string, payload: AgentUpdatePayload): Promise<ApiResponse<AgentPreset>> {
+    try {
+      const response = await fetch(`/api/agents/${id}`, {
+        method: "PUT",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify(payload),
+      });
+      const data = await parseJsonSafely<ApiResponse<AgentPreset>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "更新智能体失败" };
+      }
+      return data || { success: false, error: "更新智能体失败" };
+    } catch (_error) {
+      return { success: false, error: "更新智能体失败" };
+    }
+  },
+
+  async deletePreset(id: string): Promise<ApiResponse<AgentPreset>> {
+    try {
+      const response = await fetch(`/api/agents/${id}`, {
+        method: "DELETE",
+      });
+      const data = await parseJsonSafely<ApiResponse<AgentPreset>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "删除智能体失败" };
+      }
+      return data || { success: false, error: "删除智能体失败" };
+    } catch (_error) {
+      return { success: false, error: "删除智能体失败" };
+    }
+  },
+
+  async forkPreset(id: string, name?: string): Promise<ApiResponse<AgentPreset>> {
+    try {
+      const response = await fetch(`/api/agents/${id}/fork`, {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify(name ? { name } : {}),
+      });
+      const data = await parseJsonSafely<ApiResponse<AgentPreset>>(response);
+      if (!response.ok) {
+        return { success: false, error: data?.error || "复制智能体失败" };
+      }
+      return data || { success: false, error: "复制智能体失败" };
+    } catch (_error) {
+      return { success: false, error: "复制智能体失败" };
+    }
+  },
+};
+
+export default {
+  preset: agentService,
+  category: agentCategoryService,
+};

types/agent.ts

@@ -0,0 +1,159 @@
+export type AgentPresetType = "image" | "video";
+export type AgentPresetVisibility = "public" | "private";
+export type AgentCategoryScope = "public" | "private" | "global";
+export type AgentPresetStatus = "active" | "disabled" | "hidden" | "blocked";
+
+export interface AgentCategory {
+  id: string;
+  name: string;
+  slug: string;
+  scope: AgentCategoryScope;
+  createdBy?: string | null;
+  sortOrder: number;
+  isEnabled: boolean;
+  createdAt: string | Date;
+  updatedAt: string | Date;
+}
+
+export interface AgentReferenceAsset {
+  id: string;
+  assetId: string;
+  sortOrder: number;
+  originalUrl: string;
+  previewUrl?: string | null;
+  thumbnailUrl?: string | null;
+  mimeType?: string | null;
+  displayName?: string | null;
+}
+
+export interface AgentPresetSlot {
+  id?: string;
+  slotIndex: number;
+  title?: string | null;
+  promptTemplate: string;
+  defaultParams?: Record<string, any> | null;
+  requireUpload?: boolean;
+  uploadLabel?: string | null;
+  referenceAssets?: AgentReferenceAsset[];
+  referenceAssetIds?: string[];
+}
+
+export interface AgentModelSummary {
+  id: string;
+  modelId?: string | null;
+  displayName?: string | null;
+  avatarUrl?: string | null;
+}
+
+export interface AgentOwnerSummary {
+  id: string;
+  name?: string | null;
+  email?: string | null;
+}
+
+export interface AgentPreset {
+  id: string;
+  ownerUserId: string;
+  name: string;
+  type: AgentPresetType;
+  visibility: AgentPresetVisibility;
+  categoryId?: string | null;
+  promptTemplate: string;
+  description?: string | null;
+  defaultModelId?: string | null;
+  defaultParams?: Record<string, any> | null;
+  slotCount: number;
+  requiredUploadCount: number;
+  avatarAssetId?: string | null;
+  avatarUrl?: string | null;
+  coverAssetId?: string | null;
+  status: AgentPresetStatus;
+  sortOrder: number;
+  usageCount: number;
+  lastUsedAt?: string | Date | null;
+  forkedFromPresetId?: string | null;
+  createdAt: string | Date;
+  updatedAt: string | Date;
+  owner?: AgentOwnerSummary | null;
+  category?: AgentCategory | null;
+  defaultModel?: AgentModelSummary | null;
+  slots?: AgentPresetSlot[];
+  referenceAssets?: AgentReferenceAsset[];
+  referenceAssetIds?: string[];
+  referenceAssetCount?: number;
+}
+
+export interface AgentPresetListResponse {
+  categories: AgentCategory[];
+  public: AgentPreset[];
+  private: AgentPreset[];
+}
+
+export interface AgentExecutionSnapshot {
+  id: string;
+  presetId: string;
+  projectId?: string | null;
+  chatId?: string | null;
+  messageId?: string | null;
+  userId: string;
+  presetName: string;
+  presetType: AgentPresetType;
+  visibility: AgentPresetVisibility;
+  categoryId?: string | null;
+  categoryName?: string | null;
+  promptTemplate: string;
+  defaultModelId?: string | null;
+  resolvedModelId?: string | null;
+  defaultParams?: Record<string, any> | null;
+  slotCount: number;
+  requiredUploadCount: number;
+  slots: Array<{
+    slotIndex: number;
+    title?: string | null;
+    promptTemplate: string;
+    defaultParams?: Record<string, any> | null;
+    requireUpload?: boolean;
+    uploadLabel?: string | null;
+    referenceAssetIds: string[];
+    referenceImageUrls: string[];
+  }>;
+  referenceAssetIds: string[];
+  referenceImageUrls: string[];
+  userInputText?: string | null;
+  requestPayload?: Record<string, any> | null;
+  createdAt: string | Date;
+  defaultModel?: AgentModelSummary | null;
+  resolvedModel?: AgentModelSummary | null;
+}
+
+export interface AgentCreatePayload {
+  name: string;
+  type: AgentPresetType;
+  visibility: AgentPresetVisibility;
+  categoryId?: string;
+  newCategoryName?: string;
+  description?: string;
+  defaultModelId: string;
+  defaultParams?: Record<string, any>;
+  avatarAssetId?: string;
+  slots: AgentPresetSlot[];
+}
+
+export interface AgentUpdatePayload {
+  name?: string;
+  type?: AgentPresetType;
+  visibility?: AgentPresetVisibility;
+  categoryId?: string | null;
+  newCategoryName?: string;
+  description?: string | null;
+  defaultModelId?: string;
+  defaultParams?: Record<string, any> | null;
+  avatarAssetId?: string | null;
+  slots?: AgentPresetSlot[];
+  status?: AgentPresetStatus;
+}
+
+export interface AgentCategoryCreatePayload {
+  name: string;
+  scope?: "public" | "private";
+}

types/canvas/index.ts

@@ -68,6 +68,7 @@ export interface VideoNodeData {
   duration?: number;
   progress?: number; // 生成进度百分比
   coverImage?: string; // 用于生成视频的封面图
+  metadata?: Record<string, any>;
 }
 
 // 文本节点数据