koishi-plugin-hds-interlude 0.1.4-beta2 → 0.1.4-beta3

package/BEGINNER_GUIDE.md

@@ -1,6 +1,6 @@
 # HDS Interlude 新手引导
 
-适用版本：`0.1.4-beta2`
+适用版本：`0.1.4-beta3`
 
 HDS Interlude 是 Koishi 的持续叙事聊天插件。插件使用共享主剧本保存角色状态、关系分支、已发生事件、待处理计划和长期记忆。用户消息会进入当前活动场景；主模型在同一次请求中续写已经发生的生活，并决定是否发送、延迟发送或暂不发送消息。实时写作读取一条按时间排序的活动场景记录：最近剧本文字、真实用户消息和已经成功投递的角色消息在同一条线上。剧本引子、场景外近期事实和长期记忆负责更早的历史。
 

package/CONFIGURATION_GUIDE.md

@@ -1,6 +1,6 @@
 # HDS Interlude 配置指南
 
-适用版本：`0.1.4-beta2`
+适用版本：`0.1.4-beta3`
 
 第一次安装先看 `BEGINNER_GUIDE.md`。本文件严格按照 Koishi Console 的显示顺序说明当前字段；旧版本已经移除或隐藏的字段集中列在末尾，不再混入正常配置流程。
 
@@ -167,6 +167,12 @@ Embedding 地址留空时，插件会尝试从标准 `/chat/completions` 地址
 
 转写成功后，主模型收到的用户事件会带有 `[用户语音转写]` 标记，并和当前文本、图片一起构成同一个回合。该功能需要 SnowLuma 支持原始 OneBot `fetch_ptt_text` 动作；NapCat 或其它实现不支持时会安全降级。它不把音频文件或 base64 写入 HDSI 数据库。
 
+### 4.4 chatActions：QQ 原生表情
+
+`nativeFaces` 控制是否允许结构化 QQ 原生小表情；关闭后主模型不会收到对应字段，插件也不会发送 face 段。
+
+`expressionThreshold` 是表达严格度，不是固定发送频率。HDSI 会同时校验模型意愿与当前回复文字是否具有相同的非语言含义：低分或语义不相符时不投递。`0.70` 是平衡值；`0.90` 以上非常克制；`0.95` 及以上接近关闭，适合只保留纯文字聊天的场景。
+
 ## 5. sharedStory：共享剧本
 
 当前运行时固定为“同一机器人账号一个活动主剧本”，因此不再显示旧版 `enabled` 开关。

package/DEPLOYMENT_GUIDE.md

@@ -45,7 +45,7 @@ Koishi 第一次启动会准备运行环境，耐心等待即可。
 
 ```powershell
 cd C:\Users\你的用户名\AppData\Roaming\Koishi\Desktop\data\instances\default
-npm install C:\路径\koishi-plugin-hds-interlude-0.1.3.tgz
+npm install --save-exact C:\路径\koishi-plugin-hds-interlude-0.1.4-beta3.tgz
 ```
 
 安装完成后回到 Koishi Console，添加或启用 `hds-interlude`。

package/README.md

@@ -4,7 +4,7 @@
 
 HDS Interlude 是一个面向 Koishi 一对一与多参与者场景的持续叙事聊天框架。它让用户消息、角色的沉默、延迟回复、主动联系和自动推进，都成为同一段生活剧本中自然可见的部分，并由一次主叙事写作连贯地决定。
 
-当前版本：`0.1.4-beta2`（测试版）。提供持续生活剧本、结构化投递、群聊意愿、多提供商模型连接与可选聊天动作。
+当前版本：`0.1.4-beta3`（测试版）。提供持续生活剧本、结构化投递、群聊意愿、多提供商模型连接与可选聊天动作。
 
 ## 文档导航
 
@@ -259,13 +259,13 @@ npm install koishi-plugin-hds-interlude@beta
 使用本地预发布包时，可在 Koishi 实例目录执行：
 
 ```bash
-npm install /absolute/path/to/koishi-plugin-hds-interlude-0.1.4-beta2.tgz
+npm install /absolute/path/to/koishi-plugin-hds-interlude-0.1.4-beta3.tgz
 ```
 
 Windows 示例：
 
 ```powershell
-npm install C:\dev\HDS-Interlude\plugins\hds-interlude\release\koishi-plugin-hds-interlude-0.1.4-beta2.tgz
+npm install C:\dev\HDS-Interlude\plugins\hds-interlude\release\koishi-plugin-hds-interlude-0.1.4-beta3.tgz
 ```
 
 安装后重新加载 Koishi，再在 Console 启用插件。

package/docs/CHANGELOG.md

@@ -1,5 +1,12 @@
 # 版本记录
 
+## 0.1.4-beta3
+
+- 入站 QQ 原生表情现在会在私聊、群聊与引用内容中翻译为稳定语义文字；经典系统表情复用 QFace 表，并补齐当前具名新增表情。未知未来 ID 保留“名称未收录”标记，不再交由模型猜测。
+- QQ 原生小表情的 `expressionThreshold` 改为语义校准阈值：模型声明的意愿必须与当前回复文字相符才能投递，不能再用 `willingness: 1.0` 穿透高阈值；`0.95` 以上为接近关闭档。
+- 收束原生表情提示词：日常回复不应附带表情，`nativeFace` 是可省略的表达动作，不是发送许可字段。
+- 部署与配置文档同步 beta3 本地包路径，并说明原生表情开关和阈值含义。
+
 ## 0.1.4-beta2
 
 - 优化聊天动作提示：私聊仅启用原生表情时不再注入群消息引用说明，减少无关上下文。

package/lib/index.d.ts

@@ -6,7 +6,7 @@ declare module 'koishi' {
     }
 }
 export declare const name = "hds-interlude";
-export declare const version = "0.1.4-beta2";
+export declare const version = "0.1.4-beta3";
 export declare const inject: {
     required: string[];
     optional: string[];

package/lib/index.js

@@ -1,6 +1,8 @@
+var __create = Object.create;
 var __defProp = Object.defineProperty;
 var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
 var __getOwnPropNames = Object.getOwnPropertyNames;
+var __getProtoOf = Object.getPrototypeOf;
 var __hasOwnProp = Object.prototype.hasOwnProperty;
 var __name = (target, value) => __defProp(target, "name", { value, configurable: true });
 var __export = (target, all) => {
@@ -15,6 +17,14 @@ var __copyProps = (to, from, except, desc) => {
   }
   return to;
 };
+var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__getProtoOf(mod)) : {}, __copyProps(
+  // If the importer is in node compatibility mode or this is not an ESM
+  // file that has been converted to a CommonJS file using a Babel-
+  // compatible transform (i.e. "__esModule" has not been set), then set
+  // "default" to the CommonJS "module.exports" for node compatibility.
+  isNodeMode || !mod || !mod.__esModule ? __defProp(target, "default", { value: mod, enumerable: true }) : target,
+  mod
+));
 var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
 
 // src/index.ts
@@ -30,6 +40,7 @@ __export(src_exports, {
   ZHIPU_FIRST_VISIBLE_TOKEN_TIMEOUT: () => ZHIPU_FIRST_VISIBLE_TOKEN_TIMEOUT,
   ZHIPU_OFFICIAL_CHAT_ENDPOINT: () => ZHIPU_OFFICIAL_CHAT_ENDPOINT,
   apply: () => apply,
+  calibratedNativeFaceWillingness: () => calibratedNativeFaceWillingness,
   configuredProviders: () => configuredProviders,
   createCompactor: () => createCompactor,
   createEmbedder: () => createEmbedder,
@@ -1045,7 +1056,7 @@ function chatActionInstruction(capabilities) {
     instructions.push(`The protagonist may add at most one lightweight message reaction without sending text: "messageReactions":[{"messageRef":"msg-...","reaction":"${capabilities.reactions.join("|")}"}]. Keep groupReply explicit, using mode=none when reacting without text.`);
   }
   if (capabilities.nativeFaces?.length) {
-    instructions.push(`For a subtle native QQ face, return nativeFace: {"semantic":"${capabilities.nativeFaces.join("|")}","willingness":0.0-1.0}. Use it only when a small nonverbal expression genuinely adds meaning; it is sent only when willingness reaches ${capabilities.expressionThreshold ?? 0.7}. Do not write bracketed face labels in reply text.`);
+    instructions.push(`For a subtle native QQ face, return nativeFace: {"semantic":"${capabilities.nativeFaces.join("|")}","willingness":0.0-1.0}. Omit nativeFace for routine wording: it is not a permission field and never needs to accompany a reply. Use it only when the reply text itself clearly carries the same nonverbal meaning; do not raise willingness to 1.0 to force a send. It is calibrated against reply text and is sent only when it reaches ${capabilities.expressionThreshold ?? 0.7}; at thresholds above 0.90, omit the field unless an expression is truly indispensable. Do not write bracketed face labels in reply text.`);
   }
   return instructions.join("\n");
 }
@@ -1766,7 +1777,7 @@ function isRecord2(value) {
 __name(isRecord2, "isRecord");
 
 // src/meta.ts
-var HDS_INTERLUDE_VERSION = "0.1.4-beta2";
+var HDS_INTERLUDE_VERSION = "0.1.4-beta3";
 
 // src/logging.ts
 var import_node_util = require("node:util");
@@ -2052,6 +2063,118 @@ function clamp3(value, min, max) {
 }
 __name(clamp3, "clamp");
 
+// src/qq-face.ts
+var qface = __toESM(require("qface"));
+var QQ_NATIVE_FACE_NAME_EXTENSIONS = {
+  349: "坚强",
+  350: "贴贴",
+  351: "敲敲",
+  352: "咦",
+  353: "拜托",
+  354: "尊嘟假嘟",
+  355: "耶",
+  356: "666",
+  357: "裂开",
+  358: "骰子",
+  359: "包剪锤",
+  360: "亲亲",
+  361: "狗狗笑哭",
+  362: "好兄弟",
+  363: "狗狗可怜",
+  364: "超级赞",
+  365: "狗狗生气",
+  366: "芒狗",
+  367: "狗狗疑问",
+  368: "奥特笑哭",
+  369: "彩虹",
+  370: "祝贺",
+  371: "冒泡",
+  372: "气呼呼",
+  373: "忙",
+  374: "波波流泪",
+  375: "超级鼓掌",
+  376: "跺脚",
+  377: "嗨",
+  378: "企鹅笑哭",
+  379: "企鹅流泪",
+  380: "真棒",
+  381: "路过",
+  382: "emo",
+  383: "企鹅爱心",
+  384: "晚安",
+  385: "太气了",
+  386: "呜呜呜",
+  387: "太好笑",
+  388: "太头疼",
+  389: "太赞了",
+  390: "太头秃",
+  391: "太沧桑",
+  392: "龙年快乐",
+  393: "新年中龙",
+  394: "新年大龙",
+  395: "略略略",
+  396: "狼狗",
+  397: "抛媚眼",
+  398: "超级ok",
+  399: "tui",
+  400: "快乐",
+  401: "超级转圈",
+  402: "别说话",
+  403: "出去玩",
+  404: "闪亮登场",
+  405: "好运来",
+  406: "姐是女王",
+  407: "我听听",
+  408: "臭美",
+  409: "送你花花",
+  410: "么么哒",
+  411: "一起嗨",
+  412: "开心",
+  413: "摇起来",
+  415: "划龙舟",
+  416: "中龙舟",
+  417: "大龙舟",
+  419: "火车",
+  420: "中火车",
+  421: "大火车",
+  424: "续标识",
+  425: "求放过",
+  426: "玩火",
+  427: "偷感",
+  428: "收到",
+  429: "蛇年快乐",
+  430: "蛇身",
+  431: "蛇尾"
+};
+function attributeValue(attributes, key) {
+  const match = new RegExp(`(?:^|[\\s,])${key}\\s*=\\s*(?:"([^"]*)"|'([^']*)'|([^\\s>]+))`, "i").exec(attributes);
+  return (match?.[1] ?? match?.[2] ?? match?.[3] ?? "").trim();
+}
+__name(attributeValue, "attributeValue");
+function qqNativeFaceName(id) {
+  const key = String(id ?? "").trim();
+  if (!key) return void 0;
+  const extension = QQ_NATIVE_FACE_NAME_EXTENSIONS[key];
+  if (extension) return extension;
+  const face = qface.get(key);
+  return face?.QDes?.replace(/^\//, "").trim() || void 0;
+}
+__name(qqNativeFaceName, "qqNativeFaceName");
+function describeQQNativeFace(id) {
+  const key = String(id ?? "").trim();
+  if (!key) return "[QQ 原生表情（未提供 ID）]";
+  const name2 = qqNativeFaceName(key);
+  return name2 ? `[QQ 原生表情：${name2}（ID: ${key}）]` : `[QQ 原生表情（ID: ${key}；名称未收录）]`;
+}
+__name(describeQQNativeFace, "describeQQNativeFace");
+function normalizeQQNativeFaceSegments(content) {
+  return String(content ?? "").replace(/<face\b([^>]*)>(?:<\/face>)?/gi, (_match, attributes) => describeQQNativeFace(attributeValue(attributes, "id"))).replace(/\[CQ:face,([^\]]*)\]/gi, (_match, attributes) => describeQQNativeFace(attributeValue(attributes, "id"))).replace(/<mface\b([^>]*)>(?:<\/mface>)?/gi, (_match, attributes) => {
+    const name2 = attributeValue(attributes, "summary") || attributeValue(attributes, "name");
+    return name2 ? `[QQ 商城表情：${name2}]` : "[QQ 商城表情]";
+  });
+}
+__name(normalizeQQNativeFaceSegments, "normalizeQQNativeFaceSegments");
+
 // src/types.ts
 var emptyStorySetting = /* @__PURE__ */ __name(() => ({
   character: { name: "Unnamed character", profile: "" },
@@ -2747,12 +2870,13 @@ var InterludeService = class extends import_koishi.Service {
     const senderId = normalizeAccountId(session.userId);
     const senderName = await this.groupSenderName(groupId, senderId, session);
     const quote = describeQuotedMessage(session, story.setting.character.name);
+    const messageContent = normalizeQQNativeFaceSegments(session.content);
     const accepted = await this.serial(story.id, async () => {
       const current = await this.getStory(story.id);
       const entry = await this.appendEntry(current.id, {
         kind: "group-message",
         actor: "user",
-        content: session.content,
+        content: messageContent,
         occurredAt: now.toISOString(),
         metadata: {
           groupId,
@@ -2773,7 +2897,7 @@ var InterludeService = class extends import_koishi.Service {
       speaker: formatGroupSpeaker(senderName, senderId),
       ...messageId ? { messageId, messageRef: groupMessageRef(accepted.id) } : {},
       ...quote ? { quote } : {},
-      content: session.content,
+      content: messageContent,
       occurredAt: now,
       direction: "user"
     }, mentionedBot, quotedBot);
@@ -3098,7 +3222,8 @@ ${message.content}`).join("\n\n");
     const allowed = new Set(capabilities?.nativeFaces ?? []);
     if (!allowed.size) return void 0;
     const draft = decision.nativeFace;
-    if (draft && allowed.has(draft.semantic) && normalizeExpressionThreshold(draft.willingness) >= (capabilities?.expressionThreshold ?? this.expressionThreshold)) {
+    const replyContent = decision.groupReply?.content ?? decision.interaction?.reply?.content ?? "";
+    if (draft && allowed.has(draft.semantic) && calibratedNativeFaceWillingness(draft.semantic, draft.willingness, replyContent) >= (capabilities?.expressionThreshold ?? this.expressionThreshold)) {
       return draft.semantic;
     }
     return void 0;
@@ -3336,7 +3461,7 @@ ${message.content}`).join("\n\n");
   describeVisionEvent(session) {
     const raw = String(session.content ?? "");
     const sources = extractSessionImageSources(session);
-    const text2 = raw.replace(/<\/?(?:img|image|audio|record)\b[^>]*>/gi, "").replace(/\[CQ:(?:image|record),[^\]]*\]/gi, "").trim();
+    const text2 = normalizeQQNativeFaceSegments(raw).replace(/<\/?(?:img|image|audio|record)\b[^>]*>/gi, "").replace(/\[CQ:(?:image|record),[^\]]*\]/gi, "").trim();
     const content = text2;
     return { content, sources };
   }
@@ -6268,6 +6393,24 @@ function normalizeExpressionThreshold(value) {
   return Number.isFinite(number) ? Math.max(0, Math.min(1, number)) : 0.7;
 }
 __name(normalizeExpressionThreshold, "normalizeExpressionThreshold");
+function calibratedNativeFaceWillingness(semantic, willingness, replyContent) {
+  const text2 = String(replyContent ?? "").replace(/<sep\/>/g, " ").trim();
+  if (!text2) return 0;
+  const patterns = {
+    smile: /(?:微笑|开心|高兴|谢谢|好耶|好呀|可以|行吧|嘿|哈哈)/i,
+    laugh: /(?:哈{2,}|笑死|好笑|乐|绷不住|蚌埠|草|救命)/i,
+    sweat: /(?:流汗|尴尬|无语|服了|麻了|救命|离谱|完了|累|忙|不知道怎么说)/i,
+    awkward: /(?:尴尬|那个|呃|emm|……|\.{3,}|我真的|怎么说呢)/i,
+    heart: /(?:喜欢|爱你|抱抱|可爱|谢谢|好耶|开心|高兴)/i,
+    surprised: /(?:不会吧|真的假的|居然|什么|怎么会|\?{1,}|？{1,}|!{1,}|！{1,})/i,
+    sad: /(?:难过|哭|委屈|可怜|遗憾|心疼|唉)/i,
+    angry: /(?:生气|气死|烦|闭嘴|别[再乱闹说]|离谱|过分|你.*(?:啊|吧|？|!|！))/i
+  };
+  const semanticMatch = patterns[semantic].test(text2);
+  const evidence = semanticMatch ? 0.9 : 0.2;
+  return Math.min(0.9, normalizeExpressionThreshold(willingness) * (0.25 + evidence * 0.75));
+}
+__name(calibratedNativeFaceWillingness, "calibratedNativeFaceWillingness");
 function targetableMessageId(value) {
   const id = String(value ?? "").trim();
   return /^-?\d+$/.test(id) && id !== "0" ? id : void 0;
@@ -6318,7 +6461,7 @@ function describeQuotedMessage(session, characterName = "主角") {
 }
 __name(describeQuotedMessage, "describeQuotedMessage");
 function normalizeQuotedMessageContent(value) {
-  const raw = String(value ?? "");
+  const raw = normalizeQQNativeFaceSegments(value);
   const content = raw.replace(/<(?:img|image)\b[^>]*\/?>(?:<\/(?:img|image)>)?/gi, "[图片]").replace(/<(?:audio|record)\b[^>]*\/?>(?:<\/(?:audio|record)>)?/gi, "[语音]").replace(/<video\b[^>]*\/?>(?:<\/video>)?/gi, "[视频]").replace(/<(?:face|mface)\b[^>]*\/?>(?:<\/(?:face|mface)>)?/gi, "[表情]").replace(/<at\b[^>]*(?:name|id)=["']?([^\s"'>]+)[^>]*\/?>(?:<\/at>)?/gi, "[@$1]").replace(/\[CQ:image,[^\]]*\]/gi, "[图片]").replace(/\[CQ:record,[^\]]*\]/gi, "[语音]").replace(/\[CQ:video,[^\]]*\]/gi, "[视频]").replace(/\[CQ:face,[^\]]*\]/gi, "[表情]").replace(/<[^>]+>/g, "").replace(/[\r\n]+/g, " ").replace(/\s{2,}/g, " ").trim();
   return clip(content, 1500);
 }
@@ -7855,6 +7998,7 @@ __name(looksLikeInterludeCommand, "looksLikeInterludeCommand");
   ZHIPU_FIRST_VISIBLE_TOKEN_TIMEOUT,
   ZHIPU_OFFICIAL_CHAT_ENDPOINT,
   apply,
+  calibratedNativeFaceWillingness,
   configuredProviders,
   createCompactor,
   createEmbedder,

package/lib/meta.d.ts

@@ -1 +1 @@
-export declare const HDS_INTERLUDE_VERSION = "0.1.4-beta2";
+export declare const HDS_INTERLUDE_VERSION = "0.1.4-beta3";

package/lib/narrator.d.ts

@@ -230,7 +230,7 @@ export declare function toPromptPayload(request: NarrativeRequest): {
     }[];
     durableFacts: {
         participantId: string;
-        scope: "promise" | "character" | "world" | "relationship" | "event";
+        scope: "character" | "world" | "relationship" | "event" | "promise";
         content: string;
         importance: number;
         confidence: number;
@@ -378,7 +378,7 @@ export declare function toPromptPayload(request: NarrativeRequest): {
     groupContext: {
         messages: {
             occurredAt: string;
-            direction: "character" | "user";
+            direction: "user" | "character";
             quote?: import("./types").QuotedMessageContext;
             senderId: string;
             senderName: string;

package/lib/qq-face.d.ts

@@ -0,0 +1,4 @@
+export declare function qqNativeFaceName(id: unknown): string;
+export declare function describeQQNativeFace(id: unknown): string;
+/** Convert adapter markup into stable narrator-visible meaning without asking a model to infer an icon. */
+export declare function normalizeQQNativeFaceSegments(content: unknown): string;

package/lib/service.d.ts

@@ -687,6 +687,13 @@ export declare function extractSessionVoiceCount(session: Pick<Session, 'content
 /** Voice and typed text share one user event. The explicit marker lets the
  * narrator distinguish recognized speech from ordinary typed text. */
 export declare function mergeUserMessageWithVoiceTranscripts(text: string, transcripts: string[], detected?: number): string;
+/**
+ * A model's willingness is an intent estimate, not a transport permission.
+ * Native faces need a visible-text counterpart so a model cannot turn every
+ * routine reply into a face merely by returning willingness=1. The 0.90 cap
+ * deliberately makes thresholds above 0.90 an effective near-disable mode.
+ */
+export declare function calibratedNativeFaceWillingness(semantic: NativeFaceSemantic, willingness: unknown, replyContent: unknown): number;
 export declare function describeQuotedMessage(session: Session, characterName?: string): QuotedMessageContext | undefined;
 export declare function normalizeQuotedMessageContent(value: unknown): string;
 export declare function normalizeAllowedReactions(value: unknown): ChatReactionName[];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "koishi-plugin-hds-interlude",
-  "version": "0.1.4-beta2",
+  "version": "0.1.4-beta3",
   "description": "A time-aware, persistent narrative chat framework for Koishi.",
   "main": "lib/index.js",
   "types": "lib/index.d.ts",
@@ -14,7 +14,8 @@
   "koishi": {"service": {"optional": ["puppeteer"]}},
   "peerDependencies": {"koishi": "^4.18.7", "koishi-plugin-puppeteer": "^3.9.0"},
   "peerDependenciesMeta": {"koishi-plugin-puppeteer": {"optional": true}},
-  "devDependencies": {"koishi-plugin-puppeteer": "^3.9.0"}
+  "devDependencies": {"koishi-plugin-puppeteer": "^3.9.0"},
+  "dependencies": {"qface": "^1.4.1"}
 }