mioku-plugin-chat 3.0.6 → 3.1.1

package/context.ts

@@ -38,7 +38,7 @@ export interface ChatPluginContext {
   ctx: MiokuContext;
   defaultConfig: ChatConfig;
   configProvider: ChatConfigProvider;
-  getConfig: (groupId?: number) => Promise<ChatConfig>;
+  getConfig: (groupId?: string) => Promise<ChatConfig>;
   db: ChatDatabase;
   aiInstance: AIInstance;
   workAIInstance: AIInstance;
@@ -75,8 +75,8 @@ export interface ChatPluginContext {
   runChat: RunChat;
   startCooldownTimer: (
     groupSessionId: string,
-    groupId: number,
-    selfId: number,
+    groupId: string,
+    selfId: string,
   ) => void;
   recordGroupMessageForLearning: (
     userMsg: ChatMessage,
@@ -91,12 +91,12 @@ export interface ChatRuntimeState {
 }
 
 export interface ChatHandlerState {
-  getConfig: (groupId?: number) => Promise<ChatConfig>;
+  getConfig: (groupId?: string) => Promise<ChatConfig>;
   matchMessageCommands?: (
     text: string,
   ) => Array<{ plugin: string; command: string }>;
   runtimeState: ChatRuntimeState;
-  pokeCooldowns: Map<number, number>;
+  pokeCooldowns: Map<string, number>;
 }
 
 export interface ChatRuntime {

package/core/base.ts

@@ -25,7 +25,7 @@ import {
 
 export interface SendAIResponseOptions {
   ctx: MiokuContext;
-  groupId: number;
+  groupId: string;
   messages: string[];
   config: ChatConfig;
   sentIndices?: Set<number>;
@@ -78,7 +78,7 @@ async function waitTypingDelay(
 
 export async function sendAIResponse(
   options: SendAIResponseOptions,
-  selfId: number,
+  selfId: string,
 ): Promise<Array<string | undefined>> {
   const {
     ctx,
@@ -252,11 +252,11 @@ export async function sendAIResponse(
 
 export async function sendMessage(
   ctx: MiokuContext,
-  groupId: number | undefined,
-  userId: number,
+  groupId: string | undefined,
+  userId: string,
   text: string,
   config: ChatConfig,
-  selfId: number,
+  selfId: string,
   audioService?: AudioServiceApi,
 ): Promise<void> {
   const typingDelayEnabled = config.enableTypingDelay ?? false;
@@ -555,8 +555,8 @@ async function buildMarkdownImage(
 
 async function dispatchSegments(
   bot: any,
-  groupId: number | undefined,
-  userId: number | undefined,
+  groupId: string | undefined,
+  userId: string | undefined,
   buildSegments: (imageSource?: string) => any[],
   fallbackImagePath?: string,
 ): Promise<string | undefined> {
@@ -583,8 +583,8 @@ async function dispatchSegments(
 
 async function sendByTarget(
   bot: Bot,
-  groupId: number | undefined,
-  userId: number | undefined,
+  groupId: string | undefined,
+  userId: string | undefined,
   segments: readonly any[],
 ): Promise<import("mioku").SentMessage | undefined> {
   if (groupId) {
@@ -668,12 +668,12 @@ export interface GroupHistoryResult {
 }
 
 export async function getGroupHistoryMessages(
-  groupId: number,
+  groupId: string,
   groupSessionId: string,
   ctx: MiokuContext,
   historyCount: number,
   db: ChatDatabase,
-  selfId: number,
+  selfId: string,
   mediaOptions?: {
     ai?: AIInstance;
     workingModel?: string;
@@ -710,8 +710,8 @@ export interface GroupInfoResult {
 
 export async function getGroupInfoData(
   ctx: MiokuContext,
-  groupId: number,
-  selfId: number,
+  groupId: string,
+  selfId: string,
   fallbackGroupName?: string,
 ): Promise<GroupInfoResult> {
   let groupName: string | undefined;
@@ -746,7 +746,7 @@ export async function getHumanizeContexts(
   groupSessionId: string,
   userName: string,
   history: ChatMessage[],
-  triggerUserId?: number,
+  triggerUserId?: string,
 ): Promise<HumanizeContextsResult> {
   const historyStartAt = history.length > 0 ? history[0].timestamp : undefined;
 
@@ -775,10 +775,10 @@ export async function getHumanizeContexts(
 export interface BuildToolContextOptions {
   ctx: MiokuContext;
   event: any;
-  selfId: number;
+  selfId: string;
   groupSessionId: string;
-  groupId?: number;
-  userId: number;
+  groupId?: string;
+  userId: string;
   config: ChatConfig;
   aiService: AIService;
   db: ChatDatabase;
@@ -867,7 +867,7 @@ export function buildToolContext(
 }
 
 export function saveBotMessages(
-  groupId: number,
+  groupId: string,
   groupSessionId: string,
   messages: string[],
   timestamp: number,
@@ -891,7 +891,7 @@ export function saveBotMessages(
       sessionId: groupSessionId,
       role: "assistant",
       content: msg ?? "",
-      userId: selfId ?? 0,
+      userId: String(selfId ?? ""),
       userName: botNickname ?? "Miku",
       userRole: "member",
       groupId,
@@ -904,7 +904,7 @@ export function saveBotMessages(
 
 export async function sendEmoji(
   ctx: MiokuContext,
-  groupId: number,
+  groupId: string,
   emojiPath: string | null | undefined,
   bot: Bot | undefined,
 ): Promise<void> {

package/core/chat-engine.ts

@@ -29,6 +29,7 @@ import {
   type StructuredUserInput,
 } from "../manage/group-structured-history";
 import { prepareImageUrlsForModel } from "./media/image-compress";
+import { mainModelSupportsVision } from "../utils/model";
 
 interface StructuredHistoryRunContext {
   manager: GroupStructuredHistoryManager;
@@ -116,7 +117,10 @@ export async function runChat(
   const currentUserMessages = hasStructuredHistory
     ? buildStructuredUserMessages(structuredHistory!.currentUserInputs)
     : [];
-  const directImageUrls = toolCtx.config.isMultimodal
+  const directImageUrls = mainModelSupportsVision(
+    toolCtx.aiService,
+    toolCtx.config.model,
+  )
     ? toolCtx.pendingImageUrls
     : undefined;
   const webSearchState = { count: 0 };
@@ -142,7 +146,7 @@ export async function runChat(
   const usageContext = {
     usageId,
     source: "chat",
-    botId: getEventNumber(toolCtx.event, "self_id"),
+    botId: getEventId(toolCtx.event, "self_id"),
     groupId: toolCtx.groupId,
     groupName: getEventString(toolCtx.event, "group_name"),
     userId: toolCtx.userId,
@@ -762,7 +766,7 @@ function sanitizeBrackets(text: string): string {
   let working = text.replace(FUNCTIONAL_BRACKET_PREFIX, (match) => {
     const idx = placeholders.length;
     placeholders.push(match);
-    return `${idx}`;
+    return `\u0000${idx}\u0000`;
   });
 
   let result = "";
@@ -772,8 +776,8 @@ function sanitizeBrackets(text: string): string {
   while (i < working.length) {
     const ch = working[i];
 
-    if (ch === "") {
-      const endIdx = working.indexOf("", i + 1);
+    if (ch === "\u0000") {
+      const endIdx = working.indexOf("\u0000", i + 1);
       if (endIdx > i) {
         const idx = parseInt(working.slice(i + 1, endIdx), 10);
         if (!Number.isNaN(idx) && placeholders[idx] !== undefined) {
@@ -960,7 +964,7 @@ ${failedSummary}
       ],
       usageContext: {
         source: "chat.tool-failure",
-        botId: getEventNumber(toolCtx.event, "self_id"),
+        botId: getEventId(toolCtx.event, "self_id"),
         groupId: toolCtx.groupId,
         groupName: getEventString(toolCtx.event, "group_name"),
         userId: toolCtx.userId,
@@ -984,12 +988,12 @@ ${failedSummary}
   return "我刚刚查这条信息时出了点问题，你可以换个关键词再试试，或者给我更具体一点的线索。";
 }
 
-function getEventNumber(event: unknown, key: string): number | undefined {
+function getEventId(event: unknown, key: string): string | undefined {
   if (!event || typeof event !== "object") return undefined;
   const value = (event as Record<string, unknown>)[key];
-  return typeof value === "number" && Number.isFinite(value)
-    ? value
-    : undefined;
+  if (value == null) return undefined;
+  const text = String(value).trim();
+  return text.length > 0 ? text : undefined;
 }
 
 function getEventString(event: unknown, key: string): string | undefined {

package/core/chat-turn.ts

@@ -2,7 +2,12 @@ import type { MiokuContext } from "mioku";
 import type { AITool, Bot, ChatRuntimePromptInjection } from "mioku";
 import type { ChatPluginContext, ChatRuntimeState } from "../context";
 import type { ChatConfig, ChatMessage, TargetMessage } from "../types";
-import { getGroupHistory, getBotRole, getQuotedContent } from "../utils";
+import {
+  getGroupHistory,
+  getBotRole,
+  getQuotedContent,
+  mainModelSupportsVision,
+} from "../utils";
 import { buildStructuredUserInputFromTarget } from "../manage/group-structured-history";
 
 export type RuntimeReplyContextType =
@@ -14,9 +19,9 @@ export type RuntimeReplyContextType =
 
 export interface ExecuteChatRuntimeRequestOptions {
   event?: any;
-  selfId?: number;
-  groupId?: number;
-  userId?: number;
+  selfId?: string;
+  groupId?: string;
+  userId?: string;
   config: ChatConfig;
   targetMessageContent?: string;
   promptInjections?: ChatRuntimePromptInjection[];
@@ -28,9 +33,9 @@ export interface ExecuteChatRuntimeRequestOptions {
 interface ResolvedRuntimeContext {
   event: any;
   isGroup: boolean;
-  groupId?: number;
-  userId: number;
-  selfId: number;
+  groupId?: string;
+  userId: string;
+  selfId: string;
   sessionId: string;
   personalSessionId?: string;
   senderName: string;
@@ -43,6 +48,53 @@ interface ResolvedRuntimeContext {
 const NO_NEW_MESSAGE =
   "[No new user message in this turn. Reply naturally based on the runtime instruction and recent context.]";
 
+/**
+ * 主模型不支持视觉时，引用消息里的图片交给视觉工作模型描述（结果按内容哈希入库复用）。
+ * 主模型支持视觉时返回 undefined：图片会直接附到请求里，不必重复描述。
+ */
+async function describeQuotedImage(
+  pluginCtx: ChatPluginContext,
+  cfg: ChatConfig,
+  imageUrl: string | undefined,
+  userId: string,
+  groupId: string | undefined,
+): Promise<string | undefined> {
+  if (
+    !imageUrl ||
+    mainModelSupportsVision(pluginCtx.aiService, cfg.model)
+  ) {
+    return undefined;
+  }
+  const visionAI =
+    pluginCtx.visionAIInstance ?? pluginCtx.aiService.getDefault();
+  if (!visionAI) return undefined;
+
+  try {
+    const { processImage } = await import("./media/image-analyzer");
+    const record = await processImage(
+      visionAI,
+      imageUrl,
+      cfg.multimodalWorkingModel,
+      pluginCtx.db,
+      {
+        runAIRequest: (request) =>
+          pluginCtx.runWithRateLimitGuard(request, {
+            userId,
+            groupId,
+            label: "quoted-image",
+            skipRetryOnRateLimit: true,
+          }),
+      },
+    );
+    return record?.description?.trim() || undefined;
+  } catch (err) {
+    pluginCtx.ctx.logger.error(
+      `[chat-turn] Failed to describe quoted image: ${err}`,
+    );
+    return undefined;
+  }
+}
+
 function buildRuntimeTargetMessageContent(
   ctx: MiokuContext,
   event: any,
@@ -61,8 +113,12 @@ function resolveRuntimeContext(
   if (options.event) {
     const event = options.event;
     const isGroup = event.message_type === "group";
-    const groupId: number | undefined = isGroup ? event.group_id : undefined;
-    const userId: number = event.user_id || event.sender?.user_id || 0;
+    const groupId: string | undefined = isGroup
+      ? String(event.group_id ?? "").trim() || undefined
+      : undefined;
+    const userId: string = String(
+      event.user_id ?? event.sender?.user_id ?? "",
+    ).trim();
     return {
       event,
       isGroup,
@@ -80,18 +136,18 @@ function resolveRuntimeContext(
     };
   }
 
-  if (typeof options.selfId !== "number") {
+  const selfId = String(options.selfId ?? "").trim();
+  if (!selfId) {
     throw new Error("Chat runtime requires either event or selfId");
   }
-  if (
-    typeof options.groupId !== "number" &&
-    typeof options.userId !== "number"
-  ) {
+  const optionGroupId = String(options.groupId ?? "").trim() || undefined;
+  const optionUserId = String(options.userId ?? "").trim();
+  if (!optionGroupId && !optionUserId) {
     throw new Error("Chat runtime requires groupId or userId");
   }
 
-  const isGroup = typeof options.groupId === "number";
-  const userId = options.userId ?? 0;
+  const isGroup = Boolean(optionGroupId);
+  const userId = optionUserId;
   const event = {
     self_id: options.selfId,
     message_type: isGroup ? "group" : "private",
@@ -110,15 +166,15 @@ function resolveRuntimeContext(
   return {
     event,
     isGroup,
-    groupId: options.groupId,
+    groupId: optionGroupId,
     userId,
-    selfId: options.selfId,
-    sessionId: options.groupId
-      ? `group:${options.groupId}`
+    selfId,
+    sessionId: optionGroupId
+      ? `group:${optionGroupId}`
       : `personal:${userId}`,
     personalSessionId:
-      options.groupId && userId ? `personal:${userId}` : undefined,
-    senderName: options.groupId ? "system" : String(userId),
+      optionGroupId && userId ? `personal:${userId}` : undefined,
+    senderName: optionGroupId ? "system" : String(userId),
     userRole: "member",
     userTitle: undefined,
     groupName: undefined,
@@ -133,10 +189,10 @@ export async function finalizeChatTurn(
     event: any;
     cfg: ChatConfig;
     result: { messages: string[]; emojiPath?: string | null };
-    groupId?: number;
+    groupId?: string;
     groupSessionId: string;
-    userId: number;
-    selfId: number;
+    userId: string;
+    selfId: string;
     toolCtx: { sentMessageIndices?: Set<number> };
     send: boolean;
     isLive: boolean;
@@ -247,8 +303,8 @@ export async function processChat(
 ): Promise<void> {
   const { ctx } = pluginCtx;
   const isGroup = e.message_type === "group";
-  const groupId: number | undefined = isGroup ? e.group_id : undefined;
-  const userId: number = e.user_id || e.sender?.user_id;
+  const groupId: string | undefined = isGroup ? e.group_id : undefined;
+  const userId: string = e.user_id || e.sender?.user_id;
   const selfId = options.replyBot ? Number(options.replyBot.bot_id) : e.self_id;
   const cfg = await pluginCtx.getConfig(groupId);
 
@@ -285,16 +341,14 @@ export async function processChat(
     }
     if (quotedInfo?.imageUrl) imageUrls.push(quotedInfo.imageUrl);
 
-    let messageContent = ctx.text(e) || "";
-    let extraContext = "";
-    if (quotedInfo) {
-      const parts = [
-        `[Quoted message #${quotedInfo.messageId} from ${quotedInfo.senderName}: ${quotedInfo.content}]`,
-      ];
-      if (quotedInfo.imageUrl) parts.push("[Quoted message contains an image]");
-      extraContext = parts.join(" ");
-    }
-    if (extraContext) messageContent = extraContext + " " + messageContent;
+    // 与历史拉取并行，省掉视觉描述带来的额外等待
+    const quotedImageNotePromise = describeQuotedImage(
+      pluginCtx,
+      cfg,
+      quotedInfo?.imageUrl,
+      userId,
+      groupId,
+    );
 
     const rawHistory = groupId
       ? await getGroupHistory(
@@ -346,6 +400,24 @@ export async function processChat(
       userId,
     );
 
+    const quotedImageNote = await quotedImageNotePromise;
+    let messageContent = ctx.text(e) || "";
+    if (quotedInfo) {
+      const quotedText =
+        quotedInfo.content || (quotedInfo.imageUrl ? "[image]" : "");
+      const parts = [
+        `[Quoted message #${quotedInfo.messageId} from ${quotedInfo.senderName}: ${quotedText}]`,
+      ];
+      if (quotedInfo.imageUrl) {
+        parts.push(
+          quotedImageNote
+            ? `[Quoted message image: ${quotedImageNote}]`
+            : "[Quoted message contains an image]",
+        );
+      }
+      messageContent = `${parts.join(" ")} ${messageContent}`;
+    }
+
     const targetMessage: TargetMessage = {
       userName: senderName,
       userId,

package/core/media/history-media.ts

@@ -70,12 +70,12 @@ export interface HistoryMediaProcessingOptions {
     error(message: string): void;
   };
   bot?: Bot;
-  groupId?: number;
+  groupId?: string;
   runAIRequest?<T>(request: () => Promise<T>): Promise<T | null>;
 }
 
 export interface GroupNoticeHistorySummary {
-  userId: number;
+  userId: string;
   userName: string;
   userRole: string;
   content: string;
@@ -556,7 +556,7 @@ async function extractVideoFrames(
         { timeout: 30_000 },
       );
       const buffer = await fs.readFile(outputPath);
-      frames.push(`data:image/jpeg);base64,${buffer.toString("base64")}`);
+      frames.push(`data:image/jpeg;base64,${buffer.toString("base64")}`);
     }
 
     return frames;
@@ -701,18 +701,18 @@ function extractNoticeText(notice: any): string {
 }
 
 function extractNoticeSender(notice: any): {
-  userId: number;
+  userId: string;
   userName: string;
 } {
-  const userId = Number(
-    notice?.user_id ||
-      notice?.sender_id ||
-      notice?.operator_id ||
-      notice?.poster_id ||
-      notice?.publisher_id ||
-      notice?.sender?.user_id ||
-      0,
-  );
+  const userId = String(
+    notice?.user_id ??
+      notice?.sender_id ??
+      notice?.operator_id ??
+      notice?.poster_id ??
+      notice?.publisher_id ??
+      notice?.sender?.user_id ??
+      "",
+  ).trim();
   const userName =
     notice?.sender?.card ||
     notice?.sender?.nickname ||
@@ -722,7 +722,7 @@ function extractNoticeSender(notice: any): {
     notice?.operator_name ||
     (userId ? String(userId) : "unknown");
   return {
-    userId: Number.isFinite(userId) ? userId : 0,
+    userId,
     userName: String(userName),
   };
 }

package/core/media/image-compress.ts

@@ -5,6 +5,13 @@ const IMAGE_MAX_BYTES = 1 * 1024 * 1024;
 const COMPRESS_MAX_WIDTH = 1280;
 const COMPRESS_JPEG_QUALITY = 80;
 
+const INLINE_MIME_TYPES = new Set([
+  "image/jpeg",
+  "image/png",
+  "image/webp",
+  "image/gif",
+]);
+
 export const QQ_IMAGE_FETCH_HEADERS = {
   "User-Agent":
     "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36",
@@ -13,26 +20,24 @@ export const QQ_IMAGE_FETCH_HEADERS = {
 
 const FETCH_HEADERS = QQ_IMAGE_FETCH_HEADERS;
 
-/**
- * 准备发给模型的图片 URL：体积超过 1MB 时压缩为 JPEG data URL，否则原样返回。
- * data URL（如 GIF 抽帧结果）与非 http(s) 链接直接放行。
- */
 export async function prepareImageUrlForModel(url: string): Promise<string> {
   if (url.startsWith("data:")) return url;
   if (!/^https?:\/\//i.test(url)) return url;
 
-  const size = await probeImageSize(url);
-  if (size !== null && size <= IMAGE_MAX_BYTES) return url;
-
   let buffer: Buffer;
+  let mimeType: string;
   try {
-    buffer = await downloadImageBuffer(url);
+    const downloaded = await downloadImage(url);
+    buffer = downloaded.buffer;
+    mimeType = downloaded.mimeType;
   } catch (err) {
     logger.warn(`[image-compress] download failed, using original: ${err}`);
     return url;
   }
 
-  if (buffer.length <= IMAGE_MAX_BYTES) return url;
+  if (buffer.length <= IMAGE_MAX_BYTES && INLINE_MIME_TYPES.has(mimeType)) {
+    return `data:${mimeType};base64,${buffer.toString("base64")}`;
+  }
 
   try {
     const compressed = await sharp(buffer)
@@ -44,8 +49,8 @@ export async function prepareImageUrlForModel(url: string): Promise<string> {
     );
     return `data:image/jpeg;base64,${compressed.toString("base64")}`;
   } catch (err) {
-    logger.warn(`[image-compress] compress failed, using original: ${err}`);
-    return url;
+    logger.warn(`[image-compress] compress failed, inlining original: ${err}`);
+    return `data:${mimeType};base64,${buffer.toString("base64")}`;
   }
 }
 
@@ -55,23 +60,29 @@ export async function prepareImageUrlsForModel(
   return Promise.all(urls.map(prepareImageUrlForModel));
 }
 
-async function probeImageSize(url: string): Promise<number | null> {
-  try {
-    const head = await fetch(url, { method: "HEAD", headers: FETCH_HEADERS });
-    if (!head.ok) return null;
-    const len = head.headers.get("content-length");
-    if (!len) return null;
-    const n = Number(len);
-    return Number.isFinite(n) ? n : null;
-  } catch {
-    return null;
-  }
-}
-
-async function downloadImageBuffer(url: string): Promise<Buffer> {
+async function downloadImage(
+  url: string,
+): Promise<{ buffer: Buffer; mimeType: string }> {
   const response = await fetch(url, { headers: FETCH_HEADERS });
   if (!response.ok) {
-    throw new Error(`download failed: ${response.status} ${response.statusText}`);
+    throw new Error(
+      `download failed: ${response.status} ${response.statusText}`,
+    );
+  }
+
+  const mimeType = normalizeImageMimeType(
+    response.headers.get("content-type") || "",
+  );
+  if (!mimeType) {
+    throw new Error(
+      `not an image response: ${response.headers.get("content-type") || "unknown"}`,
+    );
   }
-  return Buffer.from(await response.arrayBuffer());
+
+  return { buffer: Buffer.from(await response.arrayBuffer()), mimeType };
+}
+
+function normalizeImageMimeType(contentType: string): string | null {
+  const mimeType = contentType.split(";")[0]?.trim().toLowerCase() || "";
+  return mimeType.startsWith("image/") ? mimeType : null;
 }

package/core/media/segment.ts

@@ -28,7 +28,7 @@ export function buildHistoryMediaProcessingOptions(
     saveMediaSummary(summary: MediaSummaryRecord): void;
   },
   bot: Bot,
-  groupId: number,
+  groupId: string,
   log: HistoryMediaProcessingOptions["logger"],
   runAIRequest?: <T>(request: () => Promise<T>) => Promise<T | null>,
 ): HistoryMediaProcessingOptions {
@@ -81,7 +81,7 @@ export function getCardData(seg: unknown): string | null {
 
 export function isMediaAnalysisBlocked(
   config: ChatConfig,
-  userId: number,
+  userId: string,
 ): boolean {
   return Boolean(config.mediaAnalysisBlacklistUsers?.includes(userId));
 }

package/core/prompt.ts

@@ -274,7 +274,7 @@ function buildResponseFormatSection(
 - For action markers like [] , put them on their own line when they are meant to be a separate action.
 
 - **SPECIAL ACTIONS in your text (auto-parsed and removed from message):**
-  - Use [at:123456] in your text to @ someone (123456 is the QQ number)
+  - Use [at:<user_id>] in your text to @ someone (<user_id> is the platform user id shown in history)
   - Use [poke:123456] in your text to poke someone. IMPORTANT: when you plan to poke a user, DON't describe your poke actions.
   - Use [reply:message_id] at the START of a line to quote-reply that message (copy the message_id EXACTLY as shown after # in the chat history; it may contain letters, e.g. [reply:P2rW...pMB])
   - **You can use MULTIPLE [reply:xxx] markers in different lines to quote multiple messages!**`);