koishi-plugin-best-cave 2.6.10 → 2.7.1

package/lib/AIManager.d.ts

@@ -0,0 +1,113 @@
+import { Context, Logger } from 'koishi';
+import { Config, CaveObject, StoredElement } from './index';
+import { FileManager } from './FileManager';
+/**
+ * @description 数据库 `cave_meta` 表的完整对象模型。
+ */
+export interface CaveMetaObject {
+    cave: number;
+    keywords: string[];
+    description: string;
+    rating: number;
+}
+declare module 'koishi' {
+    interface Tables {
+        cave_meta: CaveMetaObject;
+    }
+}
+/**
+ * @class AIManager
+ * @description 负责 AI 分析（描述、评分、关键词）和 AI 查重。
+ * 通过与外部 AI 服务接口交互，实现对回声洞内容的深度分析和重复性检查。
+ */
+export declare class AIManager {
+    private ctx;
+    private config;
+    private logger;
+    private fileManager;
+    private http;
+    /**
+     * @constructor
+     * @param {Context} ctx - Koishi 的上下文对象。
+     * @param {Config} config - 插件的配置信息。
+     * @param {Logger} logger - 日志记录器实例。
+     * @param {FileManager} fileManager - 文件管理器实例。
+     */
+    constructor(ctx: Context, config: Config, logger: Logger, fileManager: FileManager);
+    /**
+     * @description 注册与 AI 功能相关的 `.ai` 子命令。
+     * @param {any} cave - 主 `cave` 命令实例。
+     */
+    registerCommands(cave: any): void;
+    /**
+     * @description 对新内容进行两阶段 AI 查重。
+     * @param {StoredElement[]} newElements - 新内容的元素数组。
+     * @param {{ sourceUrl: string, fileName: string }[]} newMediaToSave - 新内容中待上传的媒体文件信息。
+     * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+     * @returns {Promise<{ duplicate: boolean; id?: number }>} - 返回 AI 判断结果。
+     */
+    checkForDuplicates(newElements: StoredElement[], newMediaToSave: {
+        sourceUrl: string;
+        fileName: string;
+    }[], mediaBuffers?: {
+        fileName: string;
+        buffer: Buffer;
+    }[]): Promise<{
+        duplicate: boolean;
+        id?: number;
+    }>;
+    /**
+     * @description 分析单个回声洞，并将分析结果存入数据库。
+     * @param {CaveObject} cave - 需要分析的回声洞对象。
+     * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+     * @returns {Promise<void>}
+     */
+    analyzeAndStore(cave: CaveObject, mediaBuffers?: {
+        fileName: string;
+        buffer: Buffer;
+    }[]): Promise<void>;
+    /**
+     * @description 调用 AI 模型获取内容的分析结果。
+     * @param {StoredElement[]} elements - 内容的元素数组。
+     * @param {{ sourceUrl: string, fileName: string }[]} [mediaToSave] - (可选) 待保存的媒体文件信息。
+     * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+     * @returns {Promise<Omit<CaveMetaObject, 'cave'>>} - 返回分析结果对象。
+     */
+    private getAnalysis;
+    /**
+     * @description 使用 Jaccard 相似度系数计算两组关键词的相似度。
+     * @param {Set<string>} setA - 第一组关键词集合。
+     * @param {Set<string>} setB - 第二组关键词集合。
+     * @returns {number} - 返回 0 到 1 之间的相似度值。
+     */
+    private calculateKeywordSimilarity;
+    /**
+     * @description 准备发送给 AI 模型的请求体（Payload）。
+     * @param {string} prompt - 系统提示词。
+     * @param {string} schemaString - JSON Schema 字符串。
+     * @param {StoredElement[]} elements - 内容的元素数组。
+     * @param {{ sourceUrl: string, fileName: string }[]} [mediaToSave] - (可选) 待保存的媒体文件信息。
+     * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+     * @returns {Promise<{ payload: any }>} - 返回包含请求体的对象。
+     */
+    private preparePayload;
+    /**
+     * @description 准备用于 AI 精准查重的请求体（Payload）。
+     * @param {StoredElement[]} newElements - 新内容的元素。
+     * @param {CaveObject[]} existingCaves - 经过初筛的疑似重复的旧内容。
+     * @returns {Promise<{ payload: any }>} - 返回适用于查重场景的请求体。
+     */
+    private prepareDedupePayload;
+    /**
+     * @description 解析 AI 返回的分析响应。
+     * @param {any} response - AI 服务的原始响应对象。
+     * @returns {Omit<CaveMetaObject, 'cave'>} - 返回结构化的分析结果。
+     */
+    private parseAnalysisResponse;
+    /**
+     * @description 解析 AI 返回的查重响应。
+     * @param {any} response - AI 服务的原始响应对象。
+     * @returns {{ duplicate: boolean; id?: number }} - 返回查重结果。
+     */
+    private parseDedupeResponse;
+}

package/lib/Utils.d.ts

@@ -3,6 +3,7 @@ import { CaveObject, Config, StoredElement } from './index';
 import { FileManager } from './FileManager';
 import { HashManager, CaveHashObject } from './HashManager';
 import { PendManager } from './PendManager';
+import { AIManager } from './AIManager';
 /**
  * @description 构建一条用于发送的完整回声洞消息，处理不同存储后端的资源链接。
  * @param cave 回声洞对象。
@@ -47,7 +48,7 @@ export declare function getNextCaveId(ctx: Context, reusableIds: Set<number>): P
  * @param creationTime 统一的创建时间戳，用于生成文件名。
  * @returns 包含数据库元素和待保存媒体列表的对象。
  */
-export declare function processMessageElements(sourceElements: h[], newId: number, session: Session, config: Config, logger: Logger, creationTime: Date): Promise<{
+export declare function processMessageElements(sourceElements: h[], newId: number, session: Session, creationTime: Date): Promise<{
     finalElementsForDb: StoredElement[];
     mediaToSave: {
         sourceUrl: string;
@@ -55,20 +56,35 @@ export declare function processMessageElements(sourceElements: h[], newId: numbe
     }[];
 }>;
 /**
- * @description 异步处理文件上传、查重和状态更新的后台任务。
+ * @description 执行文本 (Simhash) 和图片 (pHash) 相似度查重。
+ * @returns 一个对象，指示是否发现重复项；如果未发现，则返回生成的哈希。
+ */
+export declare function performSimilarityChecks(ctx: Context, config: Config, hashManager: HashManager, finalElementsForDb: StoredElement[], downloadedMedia: {
+    fileName: string;
+    buffer: Buffer;
+}[]): Promise<{
+    duplicate: boolean;
+    message?: string;
+    textHashesToStore?: Omit<CaveHashObject, 'cave'>[];
+    imageHashesToStore?: Omit<CaveHashObject, 'cave'>[];
+}>;
+/**
+ * @description 异步处理文件上传和状态更新的后台任务。
  * @param ctx - Koishi 上下文。
  * @param config - 插件配置。
  * @param fileManager - FileManager 实例，用于保存文件。
  * @param logger - 日志记录器实例。
  * @param reviewManager - ReviewManager 实例，用于提交审核。
  * @param cave - 刚刚在数据库中创建的 `preload` 状态的回声洞对象。
- * @param mediaToSave - 需要下载和处理的媒体文件列表。
+ * @param downloadedMedia - 需要保存的媒体文件及其 Buffer。
  * @param reusableIds - 可复用 ID 的内存缓存。
  * @param session - 触发此操作的用户会话，用于发送反馈。
  * @param hashManager - HashManager 实例，如果启用则用于哈希计算和比较。
  * @param textHashesToStore - 已预先计算好的、待存入数据库的文本哈希对象数组。
+ * @param imageHashesToStore - 已预先计算好的、待存入数据库的图片哈希对象数组。
+ * @param aiManager - AIManager 实例，如果启用则用于 AI 分析。
  */
-export declare function handleFileUploads(ctx: Context, config: Config, fileManager: FileManager, logger: Logger, reviewManager: PendManager, cave: CaveObject, mediaToToSave: {
-    sourceUrl: string;
+export declare function handleFileUploads(ctx: Context, config: Config, fileManager: FileManager, logger: Logger, reviewManager: PendManager, cave: CaveObject, downloadedMedia: {
     fileName: string;
-}[], reusableIds: Set<number>, session: Session, hashManager: HashManager, textHashesToStore: Omit<CaveHashObject, 'cave'>[]): Promise<void>;
+    buffer: Buffer;
+}[], reusableIds: Set<number>, session: Session, hashManager: HashManager, textHashesToStore: Omit<CaveHashObject, 'cave'>[], imageHashesToStore: Omit<CaveHashObject, 'cave'>[], aiManager: AIManager | null): Promise<void>;

package/lib/index.d.ts

@@ -1,5 +1,6 @@
 import { Context, Schema } from 'koishi';
 import { CaveHashObject } from './HashManager';
+import { CaveMetaObject } from './AIManager';
 export declare const name = "best-cave";
 export declare const inject: string[];
 export declare const usage = "\n<div style=\"border-radius: 10px; border: 1px solid #ddd; padding: 16px; margin-bottom: 20px; box-shadow: 0 2px 5px rgba(0,0,0,0.1);\">\n  <h2 style=\"margin-top: 0; color: #4a6ee0;\">\uD83D\uDCCC \u63D2\u4EF6\u8BF4\u660E</h2>\n  <p>\uD83D\uDCD6 <strong>\u4F7F\u7528\u6587\u6863</strong>\uFF1A\u8BF7\u70B9\u51FB\u5DE6\u4E0A\u89D2\u7684 <strong>\u63D2\u4EF6\u4E3B\u9875</strong> \u67E5\u770B\u63D2\u4EF6\u4F7F\u7528\u6587\u6863</p>\n  <p>\uD83D\uDD0D <strong>\u66F4\u591A\u63D2\u4EF6</strong>\uFF1A\u53EF\u8BBF\u95EE <a href=\"https://github.com/YisRime\" style=\"color:#4a6ee0;text-decoration:none;\">\u82E1\u6DDE\u7684 GitHub</a> \u67E5\u770B\u672C\u4EBA\u7684\u6240\u6709\u63D2\u4EF6</p>\n</div>\n<div style=\"border-radius: 10px; border: 1px solid #ddd; padding: 16px; margin-bottom: 20px; box-shadow: 0 2px 5px rgba(0,0,0,0.1);\">\n  <h2 style=\"margin-top: 0; color: #e0574a;\">\u2764\uFE0F \u652F\u6301\u4E0E\u53CD\u9988</h2>\n  <p>\uD83C\uDF1F \u559C\u6B22\u8FD9\u4E2A\u63D2\u4EF6\uFF1F\u8BF7\u5728 <a href=\"https://github.com/YisRime\" style=\"color:#e0574a;text-decoration:none;\">GitHub</a> \u4E0A\u7ED9\u6211\u4E00\u4E2A Star\uFF01</p>\n  <p>\uD83D\uDC1B \u9047\u5230\u95EE\u9898\uFF1F\u8BF7\u901A\u8FC7 <strong>Issues</strong> \u63D0\u4EA4\u53CD\u9988\uFF0C\u6216\u52A0\u5165 QQ \u7FA4 <a href=\"https://qm.qq.com/q/PdLMx9Jowq\" style=\"color:#e0574a;text-decoration:none;\"><strong>855571375</strong></a> \u8FDB\u884C\u4EA4\u6D41</p>\n</div>\n";
@@ -35,6 +36,7 @@ declare module 'koishi' {
     interface Tables {
         cave: CaveObject;
         cave_hash: CaveHashObject;
+        cave_meta: CaveMetaObject;
     }
 }
 export interface Config {
@@ -55,6 +57,14 @@ export interface Config {
     secretAccessKey?: string;
     bucket?: string;
     publicUrl?: string;
+    enableAI: boolean;
+    aiEndpoint?: string;
+    aiApiKey?: string;
+    aiModel?: string;
+    AnalysePrompt?: string;
+    aiCheckPrompt?: string;
+    aiAnalyseSchema?: string;
+    aiCheckSchema?: string;
 }
 export declare const Config: Schema<Config>;
 export declare function apply(ctx: Context, config: Config): void;

package/lib/index.js

@@ -37,7 +37,7 @@ __export(src_exports, {
   usage: () => usage
 });
 module.exports = __toCommonJS(src_exports);
-var import_koishi3 = require("koishi");
+var import_koishi4 = require("koishi");
 
 // src/FileManager.ts
 var import_client_s3 = require("@aws-sdk/client-s3");
@@ -289,7 +289,7 @@ var DataManager = class {
     const newCavesFromConflicts = conflictingCaves.map((cave) => {
       maxId++;
       this.logger.info(`回声洞（${cave.id}）已转移至（${maxId}）`);
-      return { ...cave, maxId, status: "active" };
+      return { ...cave, id: maxId, status: "active" };
     });
     const finalCavesToUpsert = [...nonConflictingCaves, ...newCavesFromConflicts];
     if (finalCavesToUpsert.length > 0) await this.ctx.database.upsert("cave", finalCavesToUpsert);
@@ -425,6 +425,7 @@ async function cleanupPendingDeletions(ctx, fileManager, logger2, reusableIds) {
     idsToDelete.forEach((id) => reusableIds.add(id));
     await ctx.database.remove("cave", { id: { $in: idsToDelete } });
     await ctx.database.remove("cave_hash", { cave: { $in: idsToDelete } });
+    await ctx.database.remove("cave_meta", { cave: { $in: idsToDelete } });
   } catch (error) {
     logger2.error("清理回声洞时发生错误:", error);
   }
@@ -457,7 +458,7 @@ async function getNextCaveId(ctx, reusableIds) {
   return newId;
 }
 __name(getNextCaveId, "getNextCaveId");
-async function processMessageElements(sourceElements, newId, session, config, logger2, creationTime) {
+async function processMessageElements(sourceElements, newId, session, creationTime) {
   const mediaToSave = [];
   let mediaIndex = 0;
   const typeMap = { "img": "image", "image": "image", "video": "video", "audio": "audio", "file": "file", "text": "text", "at": "at", "forward": "forward", "reply": "reply", "face": "face" };
@@ -543,34 +544,47 @@ async function processMessageElements(sourceElements, newId, session, config, lo
   return { finalElementsForDb, mediaToSave };
 }
 __name(processMessageElements, "processMessageElements");
-async function handleFileUploads(ctx, config, fileManager, logger2, reviewManager, cave, mediaToToSave, reusableIds, session, hashManager, textHashesToStore) {
-  try {
-    const downloadedMedia = [];
-    const imageHashesToStore = [];
-    const allExistingImageHashes = hashManager ? await ctx.database.get("cave_hash", { type: "phash" }) : [];
-    for (const media of mediaToToSave) {
-      const buffer = Buffer.from(await ctx.http.get(media.sourceUrl, { responseType: "arraybuffer", timeout: 3e4 }));
-      downloadedMedia.push({ fileName: media.fileName, buffer });
-      if (hashManager && [".png", ".jpg", ".jpeg", ".webp"].includes(path2.extname(media.fileName).toLowerCase())) {
-        const imageHash = await hashManager.generatePHash(buffer, 256);
+async function performSimilarityChecks(ctx, config, hashManager, finalElementsForDb, downloadedMedia) {
+  const textHashesToStore = [];
+  const imageHashesToStore = [];
+  const combinedText = finalElementsForDb.filter((el) => el.type === "text" && typeof el.content === "string").map((el) => el.content).join(" ");
+  if (combinedText) {
+    const newSimhash = hashManager.generateTextSimhash(combinedText);
+    if (newSimhash) {
+      const existingTextHashes = await ctx.database.get("cave_hash", { type: "simhash" });
+      for (const existing of existingTextHashes) {
+        const similarity = hashManager.calculateSimilarity(newSimhash, existing.hash);
+        if (similarity >= config.textThreshold) return { duplicate: true, message: `文本与回声洞（${existing.cave}）的相似度（${similarity.toFixed(2)}%）超过阈值` };
+      }
+      textHashesToStore.push({ hash: newSimhash, type: "simhash" });
+    }
+  }
+  if (downloadedMedia.length > 0) {
+    const allExistingImageHashes = await ctx.database.get("cave_hash", { type: "phash" });
+    for (const media of downloadedMedia) {
+      if ([".png", ".jpg", ".jpeg", ".webp"].includes(path2.extname(media.fileName).toLowerCase())) {
+        const imageHash = await hashManager.generatePHash(media.buffer, 256);
         for (const existing of allExistingImageHashes) {
           const similarity = hashManager.calculateSimilarity(imageHash, existing.hash);
-          if (similarity >= config.imageThreshold) {
-            await session.send(`图片与回声洞（${existing.cave}）的相似度（${similarity.toFixed(2)}%）超过阈值`);
-            await ctx.database.upsert("cave", [{ id: cave.id, status: "delete" }]);
-            cleanupPendingDeletions(ctx, fileManager, logger2, reusableIds);
-            return;
-          }
+          if (similarity >= config.imageThreshold) return { duplicate: true, message: `图片与回声洞（${existing.cave}）的相似度（${similarity.toFixed(2)}%）超过阈值` };
         }
         imageHashesToStore.push({ hash: imageHash, type: "phash" });
+        allExistingImageHashes.push({ cave: 0, hash: imageHash, type: "phash" });
       }
     }
+  }
+  return { duplicate: false, textHashesToStore, imageHashesToStore };
+}
+__name(performSimilarityChecks, "performSimilarityChecks");
+async function handleFileUploads(ctx, config, fileManager, logger2, reviewManager, cave, downloadedMedia, reusableIds, session, hashManager, textHashesToStore, imageHashesToStore, aiManager) {
+  try {
+    if (aiManager) await aiManager.analyzeAndStore(cave, downloadedMedia);
     await Promise.all(downloadedMedia.map((item) => fileManager.saveFile(item.fileName, item.buffer)));
     const needsReview = config.enablePend && session.channelId !== config.adminChannel?.split(":")[1];
     const finalStatus = needsReview ? "pending" : "active";
     await ctx.database.upsert("cave", [{ id: cave.id, status: finalStatus }]);
     if (hashManager) {
-      const allHashesToInsert = [...textHashesToStore, ...imageHashesToStore].map((h4) => ({ ...h4, cave: cave.id }));
+      const allHashesToInsert = [...textHashesToStore, ...imageHashesToStore].map((h5) => ({ ...h5, cave: cave.id }));
       if (allHashesToInsert.length > 0) await ctx.database.upsert("cave_hash", allHashesToInsert);
     }
     if (finalStatus === "pending" && reviewManager) {
@@ -745,7 +759,7 @@ var HashManager = class {
   async generateHashesForHistoricalCaves() {
     const allCaves = await this.ctx.database.get("cave", { status: "active" });
     const existingHashes = await this.ctx.database.get("cave_hash", {});
-    const existingHashSet = new Set(existingHashes.map((h4) => `${h4.cave}-${h4.hash}-${h4.type}`));
+    const existingHashSet = new Set(existingHashes.map((h5) => `${h5.cave}-${h5.hash}-${h5.type}`));
     if (allCaves.length === 0) return "无需补全回声洞哈希";
     this.logger.info(`开始补全 ${allCaves.length} 个回声洞的哈希...`);
     let hashesToInsert = [];
@@ -819,7 +833,7 @@ var HashManager = class {
     const textThreshold = options.textThreshold ?? this.config.textThreshold;
     const imageThreshold = options.imageThreshold ?? this.config.imageThreshold;
     const allHashes = await this.ctx.database.get("cave_hash", {});
-    const allCaveIds = [...new Set(allHashes.map((h4) => h4.cave))];
+    const allCaveIds = [...new Set(allHashes.map((h5) => h5.cave))];
     const textHashes = /* @__PURE__ */ new Map();
     const imageHashes = /* @__PURE__ */ new Map();
     for (const hash of allHashes) {
@@ -965,6 +979,255 @@ function hexToBinary(hex) {
 }
 __name(hexToBinary, "hexToBinary");
 
+// src/AIManager.ts
+var import_koishi3 = require("koishi");
+var path3 = __toESM(require("path"));
+var AIManager = class {
+  /**
+   * @constructor
+   * @param {Context} ctx - Koishi 的上下文对象。
+   * @param {Config} config - 插件的配置信息。
+   * @param {Logger} logger - 日志记录器实例。
+   * @param {FileManager} fileManager - 文件管理器实例。
+   */
+  constructor(ctx, config, logger2, fileManager) {
+    this.ctx = ctx;
+    this.config = config;
+    this.logger = logger2;
+    this.fileManager = fileManager;
+    this.http = ctx.http;
+    this.ctx.model.extend("cave_meta", {
+      cave: "unsigned",
+      keywords: "json",
+      description: "text",
+      rating: "unsigned"
+    }, {
+      primary: "cave"
+    });
+  }
+  static {
+    __name(this, "AIManager");
+  }
+  http;
+  /**
+   * @description 注册与 AI 功能相关的 `.ai` 子命令。
+   * @param {any} cave - 主 `cave` 命令实例。
+   */
+  registerCommands(cave) {
+    cave.subcommand(".ai", "分析回声洞", { hidden: true, authority: 4 }).usage("分析尚未分析的回声洞，补全回声洞记录。").action(async ({ session }) => {
+      if (session.channelId !== this.config.adminChannel?.split(":")) return "此指令仅限在管理群组中使用";
+      try {
+        const allCaves = await this.ctx.database.get("cave", { status: "active" });
+        const analyzedCaveIds = new Set((await this.ctx.database.get("cave_meta", {})).map((meta) => meta.cave));
+        const cavesToAnalyze = allCaves.filter((cave2) => !analyzedCaveIds.has(cave2.id));
+        if (cavesToAnalyze.length === 0) return "无需分析回声洞";
+        await session.send(`开始分析 ${cavesToAnalyze.length} 个回声洞...`);
+        let totalSuccessCount = 0;
+        for (let i = 0; i < cavesToAnalyze.length; i += 5) {
+          const batch = cavesToAnalyze.slice(i, i + 5);
+          this.logger.info(`[${totalSuccessCount}/${cavesToAnalyze.length}] 正在分析 ${batch.length} 条回声洞...`);
+          await Promise.all(batch.map((cave2) => this.analyzeAndStore(cave2)));
+          totalSuccessCount += batch.length;
+        }
+        return `已分析 ${totalSuccessCount} 个回声洞`;
+      } catch (error) {
+        this.logger.error("已中断分析回声洞:", error);
+        return `分析回声洞失败：${error.message}`;
+      }
+    });
+    cave.subcommand(".desc <id:posint>", "查询回声洞").action(async ({}, id) => {
+      if (!id) return "请输入要查看的回声洞序号";
+      try {
+        const [meta] = await this.ctx.database.get("cave_meta", { cave: id });
+        if (!meta) return `回声洞（${id}）尚未分析`;
+        const keywordsText = meta.keywords.join(", ");
+        const report = [
+          `回声洞（${id}）分析结果：`,
+          `描述：${meta.description}`,
+          `关键词：${keywordsText}`,
+          `评分：${meta.rating}/100`
+        ];
+        return import_koishi3.h.text(report.join("\n"));
+      } catch (error) {
+        this.logger.error(`查询回声洞（${id}）失败:`, error);
+        return "查询失败，请稍后再试";
+      }
+    });
+  }
+  /**
+   * @description 对新内容进行两阶段 AI 查重。
+   * @param {StoredElement[]} newElements - 新内容的元素数组。
+   * @param {{ sourceUrl: string, fileName: string }[]} newMediaToSave - 新内容中待上传的媒体文件信息。
+   * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+   * @returns {Promise<{ duplicate: boolean; id?: number }>} - 返回 AI 判断结果。
+   */
+  async checkForDuplicates(newElements, newMediaToSave, mediaBuffers) {
+    try {
+      const newAnalysis = await this.getAnalysis(newElements, newMediaToSave, mediaBuffers);
+      if (!newAnalysis || newAnalysis.keywords.length === 0) return { duplicate: false };
+      const newKeywords = new Set(newAnalysis.keywords);
+      const allMeta = await this.ctx.database.get("cave_meta", {});
+      const potentialDuplicates = [];
+      for (const meta of allMeta) {
+        const existingKeywords = new Set(meta.keywords);
+        const similarity = this.calculateKeywordSimilarity(newKeywords, existingKeywords);
+        if (similarity * 100 >= 80) {
+          const [cave] = await this.ctx.database.get("cave", { id: meta.cave });
+          if (cave) potentialDuplicates.push(cave);
+        }
+      }
+      if (potentialDuplicates.length === 0) return { duplicate: false };
+      const { payload } = await this.prepareDedupePayload(newElements, potentialDuplicates);
+      const response = await this.http.post(`${this.config.aiEndpoint}:generateContent?key=${this.config.aiApiKey}`, payload, { headers: { "Content-Type": "application/json" }, timeout: 9e4 });
+      return this.parseDedupeResponse(response);
+    } catch (error) {
+      this.logger.error("查重回声洞出错:", error);
+      return { duplicate: false };
+    }
+  }
+  /**
+   * @description 分析单个回声洞，并将分析结果存入数据库。
+   * @param {CaveObject} cave - 需要分析的回声洞对象。
+   * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+   * @returns {Promise<void>}
+   */
+  async analyzeAndStore(cave, mediaBuffers) {
+    try {
+      const analysisResult = await this.getAnalysis(cave.elements, void 0, mediaBuffers);
+      if (analysisResult) await this.ctx.database.upsert("cave_meta", [{ cave: cave.id, ...analysisResult }]);
+    } catch (error) {
+      this.logger.error(`分析回声洞（${cave.id}）失败:`, error);
+    }
+  }
+  /**
+   * @description 调用 AI 模型获取内容的分析结果。
+   * @param {StoredElement[]} elements - 内容的元素数组。
+   * @param {{ sourceUrl: string, fileName: string }[]} [mediaToSave] - (可选) 待保存的媒体文件信息。
+   * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+   * @returns {Promise<Omit<CaveMetaObject, 'cave'>>} - 返回分析结果对象。
+   */
+  async getAnalysis(elements, mediaToSave, mediaBuffers) {
+    const { payload } = await this.preparePayload(this.config.AnalysePrompt, this.config.aiAnalyseSchema, elements, mediaToSave, mediaBuffers);
+    if (!payload.contents) return null;
+    const response = await this.http.post(`${this.config.aiEndpoint}:generateContent?key=${this.config.aiApiKey}`, payload, { headers: { "Content-Type": "application/json" }, timeout: 6e4 });
+    return this.parseAnalysisResponse(response);
+  }
+  /**
+   * @description 使用 Jaccard 相似度系数计算两组关键词的相似度。
+   * @param {Set<string>} setA - 第一组关键词集合。
+   * @param {Set<string>} setB - 第二组关键词集合。
+   * @returns {number} - 返回 0 到 1 之间的相似度值。
+   */
+  calculateKeywordSimilarity(setA, setB) {
+    const intersection = new Set([...setA].filter((x) => setB.has(x)));
+    const union = /* @__PURE__ */ new Set([...setA, ...setB]);
+    return union.size === 0 ? 0 : intersection.size / union.size;
+  }
+  /**
+   * @description 准备发送给 AI 模型的请求体（Payload）。
+   * @param {string} prompt - 系统提示词。
+   * @param {string} schemaString - JSON Schema 字符串。
+   * @param {StoredElement[]} elements - 内容的元素数组。
+   * @param {{ sourceUrl: string, fileName: string }[]} [mediaToSave] - (可选) 待保存的媒体文件信息。
+   * @param {{ fileName: string; buffer: Buffer }[]} [mediaBuffers] - (可选) 已下载的媒体文件 Buffer。
+   * @returns {Promise<{ payload: any }>} - 返回包含请求体的对象。
+   */
+  async preparePayload(prompt, schemaString, elements, mediaToSave, mediaBuffers) {
+    const parts = [{ text: prompt }];
+    const combinedText = elements.filter((el) => el.type === "text" && el.content).map((el) => el.content).join("\n");
+    if (combinedText) parts.push({ text: combinedText });
+    const mediaMap = new Map(mediaBuffers?.map((m) => [m.fileName, m.buffer]));
+    const imageElements = elements.filter((el) => el.type === "image" && el.file);
+    for (const el of imageElements) {
+      try {
+        let buffer;
+        if (mediaMap.has(el.file)) {
+          buffer = mediaMap.get(el.file);
+        } else if (mediaToSave) {
+          const item = mediaToSave.find((m) => m.fileName === el.file);
+          if (item) buffer = Buffer.from(await this.ctx.http.get(item.sourceUrl, { responseType: "arraybuffer" }));
+        } else {
+          buffer = await this.fileManager.readFile(el.file);
+        }
+        if (buffer) {
+          const mimeType = path3.extname(el.file).toLowerCase() === ".png" ? "image/png" : "image/jpeg";
+          parts.push({ inline_data: { mime_type: mimeType, data: buffer.toString("base64") } });
+        }
+      } catch (error) {
+        this.logger.warn(`分析内容（${el.file}）失败:`, error);
+      }
+    }
+    if (parts.length <= 1) return { payload: {} };
+    try {
+      const schema = JSON.parse(schemaString);
+      return { payload: { contents: [{ parts }], generationConfig: { response_schema: schema } } };
+    } catch (error) {
+      this.logger.error("解析JSON Schema失败:", error);
+      return { payload: {} };
+    }
+  }
+  /**
+   * @description 准备用于 AI 精准查重的请求体（Payload）。
+   * @param {StoredElement[]} newElements - 新内容的元素。
+   * @param {CaveObject[]} existingCaves - 经过初筛的疑似重复的旧内容。
+   * @returns {Promise<{ payload: any }>} - 返回适用于查重场景的请求体。
+   */
+  async prepareDedupePayload(newElements, existingCaves) {
+    const formatContent = /* @__PURE__ */ __name((elements) => elements.filter((el) => el.type === "text").map((el) => el.content).join(" "), "formatContent");
+    const payloadContent = JSON.stringify({
+      new_content: { text: formatContent(newElements) },
+      existing_contents: existingCaves.map((cave) => ({ id: cave.id, text: formatContent(cave.elements) }))
+    });
+    const fullPrompt = `${this.config.aiCheckPrompt}
+
+以下是需要处理的数据:
+${payloadContent}`;
+    try {
+      const schema = JSON.parse(this.config.aiCheckSchema);
+      return { payload: { contents: [{ parts: [{ text: fullPrompt }] }], generationConfig: { response_schema: schema } } };
+    } catch (error) {
+      this.logger.error("解析查重JSON Schema失败:", error);
+      return { payload: {} };
+    }
+  }
+  /**
+   * @description 解析 AI 返回的分析响应。
+   * @param {any} response - AI 服务的原始响应对象。
+   * @returns {Omit<CaveMetaObject, 'cave'>} - 返回结构化的分析结果。
+   */
+  parseAnalysisResponse(response) {
+    try {
+      const content = response.candidates.content.parts.text;
+      const parsed = JSON.parse(content);
+      const keywords = Array.isArray(parsed.keywords) ? parsed.keywords : [];
+      return {
+        keywords,
+        description: parsed.description || "无",
+        rating: Math.max(0, Math.min(100, parsed.rating || 0))
+      };
+    } catch (error) {
+      this.logger.error("分析响应解析失败:", error, "原始响应:", JSON.stringify(response));
+      return { keywords: [], description: "解析失败", rating: 0 };
+    }
+  }
+  /**
+   * @description 解析 AI 返回的查重响应。
+   * @param {any} response - AI 服务的原始响应对象。
+   * @returns {{ duplicate: boolean; id?: number }} - 返回查重结果。
+   */
+  parseDedupeResponse(response) {
+    try {
+      const content = response.candidates.content.parts.text;
+      const parsed = JSON.parse(content);
+      if (parsed.duplicate === true && parsed.id) return { duplicate: true, id: Number(parsed.id) };
+      return { duplicate: false };
+    } catch (error) {
+      this.logger.error("查重响应解析失败:", error, "原始响应:", JSON.stringify(response));
+      return { duplicate: false };
+    }
+  }
+};
+
 // src/index.ts
 var name = "best-cave";
 var inject = ["database"];
@@ -980,30 +1243,77 @@ var usage = `
   <p>🐛 遇到问题？请通过 <strong>Issues</strong> 提交反馈，或加入 QQ 群 <a href="https://qm.qq.com/q/PdLMx9Jowq" style="color:#e0574a;text-decoration:none;"><strong>855571375</strong></a> 进行交流</p>
 </div>
 `;
-var logger = new import_koishi3.Logger("best-cave");
-var Config = import_koishi3.Schema.intersect([
-  import_koishi3.Schema.object({
-    perChannel: import_koishi3.Schema.boolean().default(false).description("启用分群模式"),
-    enableName: import_koishi3.Schema.boolean().default(false).description("启用自定义昵称"),
-    enableIO: import_koishi3.Schema.boolean().default(false).description("启用导入导出"),
-    adminChannel: import_koishi3.Schema.string().default("onebot:").description("管理群组 ID"),
-    caveFormat: import_koishi3.Schema.string().default("回声洞 ——（{id}）|—— {name}").description("自定义文本（参见 README）")
+var logger = new import_koishi4.Logger("best-cave");
+var Config = import_koishi4.Schema.intersect([
+  import_koishi4.Schema.object({
+    perChannel: import_koishi4.Schema.boolean().default(false).description("启用分群模式"),
+    enableName: import_koishi4.Schema.boolean().default(false).description("启用自定义昵称"),
+    enableIO: import_koishi4.Schema.boolean().default(false).description("启用导入导出"),
+    adminChannel: import_koishi4.Schema.string().default("onebot:").description("管理群组 ID"),
+    caveFormat: import_koishi4.Schema.string().default("回声洞 ——（{id}）|—— {name}").description("自定义文本（参见 README）")
   }).description("基础配置"),
-  import_koishi3.Schema.object({
-    enablePend: import_koishi3.Schema.boolean().default(false).description("启用审核"),
-    enableSimilarity: import_koishi3.Schema.boolean().default(false).description("启用查重"),
-    textThreshold: import_koishi3.Schema.number().min(0).max(100).step(0.01).default(90).description("文本相似度阈值 (%)"),
-    imageThreshold: import_koishi3.Schema.number().min(0).max(100).step(0.01).default(90).description("图片相似度阈值 (%)")
+  import_koishi4.Schema.object({
+    enablePend: import_koishi4.Schema.boolean().default(false).description("启用审核"),
+    enableSimilarity: import_koishi4.Schema.boolean().default(false).description("启用查重"),
+    textThreshold: import_koishi4.Schema.number().min(0).max(100).step(0.01).default(90).description("文本相似度阈值 (%)"),
+    imageThreshold: import_koishi4.Schema.number().min(0).max(100).step(0.01).default(90).description("图片相似度阈值 (%)")
   }).description("复核配置"),
-  import_koishi3.Schema.object({
-    localPath: import_koishi3.Schema.string().description("文件映射路径"),
-    enableS3: import_koishi3.Schema.boolean().default(false).description("启用 S3 存储"),
-    publicUrl: import_koishi3.Schema.string().description("公共访问 URL").role("link"),
-    endpoint: import_koishi3.Schema.string().description("端点 (Endpoint)").role("link"),
-    bucket: import_koishi3.Schema.string().description("存储桶 (Bucket)"),
-    region: import_koishi3.Schema.string().default("auto").description("区域 (Region)"),
-    accessKeyId: import_koishi3.Schema.string().description("Access Key ID").role("secret"),
-    secretAccessKey: import_koishi3.Schema.string().description("Secret Access Key").role("secret")
+  import_koishi4.Schema.object({
+    enableAI: import_koishi4.Schema.boolean().default(false).description("启用 AI"),
+    aiEndpoint: import_koishi4.Schema.string().description("端点 (Endpoint)").role("link").default("https://generativelanguage.googleapis.com/v1beta"),
+    aiApiKey: import_koishi4.Schema.string().description("密钥 (Key)").role("secret"),
+    aiModel: import_koishi4.Schema.string().description("模型").default("gemini-2.5-flash"),
+    AnalysePrompt: import_koishi4.Schema.string().role("textarea").default(`你是一位内容分析专家。请分析我提供的内容，总结关键词，概括内容并进行评分。`).description("分析提示词 (Prompt)"),
+    aiAnalyseSchema: import_koishi4.Schema.string().role("textarea").default(
+      `{
+        "type": "object",
+        "properties": {
+          "keywords": {
+            "type": "array",
+            "items": { "type": "string" },
+            "description": "使用尽可能多的关键词准确形容内容"
+          },
+          "description": {
+            "type": "string",
+            "description": "概括或描述这部分内容"
+          },
+          "rating": {
+            "type": "integer",
+            "description": "对内容的综合质量进行评分",
+            "minimum": 0,
+            "maximum": 100
+          }
+        },
+        "required": ["keywords", "description", "rating"]
+      }`
+    ).description("分析输出模式 (JSON Schema)"),
+    aiCheckPrompt: import_koishi4.Schema.string().role("textarea").default(`你是一位内容查重专家。请判断我提供的"新内容"是否与"已有内容"重复或高度相似。`).description("查重提示词 (Prompt)"),
+    aiCheckSchema: import_koishi4.Schema.string().role("textarea").default(
+      `{
+        "type": "object",
+        "properties": {
+          "duplicate": {
+            "type": "boolean",
+            "description": "新内容是否与已有内容重复"
+          },
+          "id": {
+            "type": "integer",
+            "description": "如果重复，此为第一个重复的已有内容的ID"
+          }
+        },
+        "required": ["duplicate"]
+      }`
+    ).description("查重输出模式 (JSON Schema)")
+  }).description("模型配置"),
+  import_koishi4.Schema.object({
+    localPath: import_koishi4.Schema.string().description("文件映射路径"),
+    enableS3: import_koishi4.Schema.boolean().default(false).description("启用 S3 存储"),
+    publicUrl: import_koishi4.Schema.string().description("公共访问 URL").role("link"),
+    endpoint: import_koishi4.Schema.string().description("端点 (Endpoint)").role("link"),
+    bucket: import_koishi4.Schema.string().description("存储桶 (Bucket)"),
+    region: import_koishi4.Schema.string().default("auto").description("区域 (Region)"),
+    accessKeyId: import_koishi4.Schema.string().description("Access Key ID").role("secret"),
+    secretAccessKey: import_koishi4.Schema.string().description("Secret Access Key").role("secret")
   }).description("存储配置")
 ]);
 function apply(ctx, config) {
@@ -1025,6 +1335,7 @@ function apply(ctx, config) {
   const reviewManager = config.enablePend ? new PendManager(ctx, config, fileManager, logger, reusableIds) : null;
   const hashManager = config.enableSimilarity ? new HashManager(ctx, config, logger, fileManager) : null;
   const dataManager = config.enableIO ? new DataManager(ctx, config, fileManager, logger) : null;
+  const aiManager = config.enableAI ? new AIManager(ctx, config, logger, fileManager) : null;
   ctx.on("ready", async () => {
     try {
       const staleCaves = await ctx.database.get("cave", { status: "preload" });
@@ -1049,7 +1360,7 @@ function apply(ctx, config) {
       const randomId = candidates[Math.floor(Math.random() * candidates.length)].id;
       const [randomCave] = await ctx.database.get("cave", { ...query, id: randomId });
       const messages = await buildCaveMessage(randomCave, config, fileManager, logger, session.platform);
-      for (const message of messages) if (message.length > 0) await session.send(import_koishi3.h.normalize(message));
+      for (const message of messages) if (message.length > 0) await session.send(import_koishi4.h.normalize(message));
     } catch (error) {
       logger.error("随机获取回声洞失败:", error);
       return "随机获取回声洞失败";
@@ -1061,34 +1372,38 @@ function apply(ctx, config) {
       if (session.quote?.elements) {
         sourceElements = session.quote.elements;
       } else if (content?.trim()) {
-        sourceElements = import_koishi3.h.parse(content);
+        sourceElements = import_koishi4.h.parse(content);
       } else {
         await session.send("请在一分钟内发送你要添加的内容");
         const reply = await session.prompt(6e4);
         if (!reply) return "等待操作超时";
-        sourceElements = import_koishi3.h.parse(reply);
+        sourceElements = import_koishi4.h.parse(reply);
       }
       const newId = await getNextCaveId(ctx, reusableIds);
       const creationTime = /* @__PURE__ */ new Date();
-      const { finalElementsForDb, mediaToSave } = await processMessageElements(sourceElements, newId, session, config, logger, creationTime);
+      const { finalElementsForDb, mediaToSave } = await processMessageElements(sourceElements, newId, session, creationTime);
       if (finalElementsForDb.length === 0) return "无可添加内容";
-      const textHashesToStore = [];
-      if (hashManager) {
-        const combinedText = finalElementsForDb.filter((el) => el.type === "text" && typeof el.content === "string").map((el) => el.content).join(" ");
-        if (combinedText) {
-          const newSimhash = hashManager.generateTextSimhash(combinedText);
-          if (newSimhash) {
-            const existingTextHashes = await ctx.database.get("cave_hash", { type: "simhash" });
-            for (const existing of existingTextHashes) {
-              const similarity = hashManager.calculateSimilarity(newSimhash, existing.hash);
-              if (similarity >= config.textThreshold) return `文本与回声洞（${existing.cave}）的相似度（${similarity.toFixed(2)}%）超过阈值`;
-            }
-            textHashesToStore.push({ hash: newSimhash, type: "simhash" });
-          }
+      const hasMedia = mediaToSave.length > 0;
+      const downloadedMedia = [];
+      if (hasMedia) {
+        for (const media of mediaToSave) {
+          const buffer = Buffer.from(await ctx.http.get(media.sourceUrl, { responseType: "arraybuffer", timeout: 3e4 }));
+          downloadedMedia.push({ fileName: media.fileName, buffer });
         }
       }
+      let textHashesToStore = [];
+      let imageHashesToStore = [];
+      if (hashManager) {
+        const checkResult = await performSimilarityChecks(ctx, config, hashManager, finalElementsForDb, downloadedMedia);
+        if (checkResult.duplicate) return checkResult.message;
+        textHashesToStore = checkResult.textHashesToStore;
+        imageHashesToStore = checkResult.imageHashesToStore;
+      }
+      if (aiManager) {
+        const duplicateResult = await aiManager.checkForDuplicates(finalElementsForDb, mediaToSave, downloadedMedia);
+        if (duplicateResult && duplicateResult.duplicate) return `内容与回声洞（${duplicateResult.id}）重复`;
+      }
       const userName = (config.enableName ? await profileManager.getNickname(session.userId) : null) || session.username;
-      const hasMedia = mediaToSave.length > 0;
       const needsReview = config.enablePend && session.channelId !== config.adminChannel?.split(":")[1];
       const initialStatus = hasMedia ? "preload" : needsReview ? "pending" : "active";
       const newCave = await ctx.database.create("cave", {
@@ -1101,9 +1416,10 @@ function apply(ctx, config) {
         time: creationTime
       });
       if (hasMedia) {
-        handleFileUploads(ctx, config, fileManager, logger, reviewManager, newCave, mediaToSave, reusableIds, session, hashManager, textHashesToStore);
+        handleFileUploads(ctx, config, fileManager, logger, reviewManager, newCave, downloadedMedia, reusableIds, session, hashManager, textHashesToStore, imageHashesToStore, aiManager);
       } else {
-        if (hashManager && textHashesToStore.length > 0) await ctx.database.upsert("cave_hash", textHashesToStore.map((h4) => ({ ...h4, cave: newCave.id })));
+        if (aiManager) await aiManager.analyzeAndStore(newCave);
+        if (hashManager && textHashesToStore.length > 0) await ctx.database.upsert("cave_hash", textHashesToStore.map((h5) => ({ ...h5, cave: newCave.id })));
         if (initialStatus === "pending") reviewManager.sendForPend(newCave);
       }
       return needsReview ? `提交成功，序号为（${newCave.id}）` : `添加成功，序号为（${newCave.id}）`;
@@ -1118,7 +1434,7 @@ function apply(ctx, config) {
       const [targetCave] = await ctx.database.get("cave", { ...getScopeQuery(session, config), id });
       if (!targetCave) return `回声洞（${id}）不存在`;
       const messages = await buildCaveMessage(targetCave, config, fileManager, logger, session.platform);
-      for (const message of messages) if (message.length > 0) await session.send(import_koishi3.h.normalize(message));
+      for (const message of messages) if (message.length > 0) await session.send(import_koishi4.h.normalize(message));
     } catch (error) {
       logger.error(`查看回声洞（${id}）失败:`, error);
       return "查看失败，请稍后再试";
@@ -1135,7 +1451,7 @@ function apply(ctx, config) {
       await ctx.database.upsert("cave", [{ id, status: "delete" }]);
       const caveMessages = await buildCaveMessage(targetCave, config, fileManager, logger, session.platform, "已删除");
       cleanupPendingDeletions(ctx, fileManager, logger, reusableIds);
-      for (const message of caveMessages) if (message.length > 0) await session.send(import_koishi3.h.normalize(message));
+      for (const message of caveMessages) if (message.length > 0) await session.send(import_koishi4.h.normalize(message));
     } catch (error) {
       logger.error(`标记回声洞（${id}）失败:`, error);
       return "删除失败，请稍后再试";
@@ -1146,7 +1462,7 @@ function apply(ctx, config) {
       const adminChannelId = config.adminChannel?.split(":")[1];
       if (session.channelId !== adminChannelId) return "此指令仅限在管理群组中使用";
       try {
-        const aggregatedStats = await ctx.database.select("cave", { status: "active" }).groupBy(["userId", "userName"], { count: /* @__PURE__ */ __name((row) => import_koishi3.$.count(row.id), "count") }).execute();
+        const aggregatedStats = await ctx.database.select("cave", { status: "active" }).groupBy(["userId", "userName"], { count: /* @__PURE__ */ __name((row) => import_koishi4.$.count(row.id), "count") }).execute();
         if (!aggregatedStats.length) return "目前没有回声洞投稿";
         const userStats = /* @__PURE__ */ new Map();
         for (const stat of aggregatedStats) {
@@ -1185,6 +1501,7 @@ ${caveIds}`;
   if (dataManager) dataManager.registerCommands(cave);
   if (reviewManager) reviewManager.registerCommands(cave);
   if (hashManager) hashManager.registerCommands(cave);
+  if (aiManager) aiManager.registerCommands(cave);
 }
 __name(apply, "apply");
 // Annotate the CommonJS export names for ESM import in node:

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "koishi-plugin-best-cave",
   "description": "功能强大、高度可定制的回声洞。支持丰富的媒体类型、内容查重、人工审核、用户昵称、数据迁移以及本地/S3 双重文件存储后端。",
-  "version": "2.6.10",
+  "version": "2.7.1",
   "contributors": [
     "Yis_Rime <yis_rime@outlook.com>"
   ],