dsh-novel-writer 2.5.1 → 2.5.5

package/README.en.md

@@ -8,7 +8,7 @@ English | [**中文**](./README.md)
 [![GitHub release](https://img.shields.io/github/v/release/siweina/dsh-novel-writer.svg?style=flat-square)](https://github.com/siweina/dsh-novel-writer/releases)
 [![DSH plugin](https://img.shields.io/badge/DSH-plugin-4b8bbe.svg?style=flat-square)](https://github.com/deepseek-ai/deepseek-harness)
 
-A novel-writing assistant plugin for **DeepSeek Harness (DSH)** (v2.5.0): chapter library management, sentence-pattern analysis, emotion purification & quantification, **12-axis vibe spectrum**, **style portrait report**, plot & settings management, local semantic search (0 token), webnovel signal detection, batch import, and AI-assisted continuation writing. Zero third-party dependencies on host.
+A novel-writing assistant plugin for **DeepSeek Harness (DSH)** (v2.5.5): chapter library management, sentence-pattern analysis, emotion purification & quantification, **12-axis vibe spectrum**, **style portrait report**, plot & settings management, local semantic search (0 token), webnovel signal detection, batch import, and AI-assisted continuation writing. Zero third-party dependencies on host.
 
 ---
 

package/README.md

@@ -8,7 +8,7 @@
 [![GitHub release](https://img.shields.io/github/v/release/siweina/dsh-novel-writer.svg?style=flat-square)](https://github.com/siweina/dsh-novel-writer/releases)
 [![DSH plugin](https://img.shields.io/badge/DSH-plugin-4b8bbe.svg?style=flat-square)](https://github.com/deepseek-ai/deepseek-harness)
 
-为 **DeepSeek Harness (DSH)** 打造的小说写作助手插件（v2.5.0）：章节库管理、句式分析、情感净化与量化、氛围光谱、**风格画像报告**、伏笔设定管理、本地语义检索（0 token）、网文信号识别、批量导入与 AI 续写辅助。宿主端零第三方依赖，浏览器端仅依赖 Web GUI 自带的 react。
+为 **DeepSeek Harness (DSH)** 打造的小说写作助手插件（v2.5.5）：章节库管理、句式分析、情感净化与量化、氛围光谱、**风格画像报告**、伏笔设定管理、本地语义检索（0 token）、网文信号识别、批量导入与 AI 续写辅助。宿主端零第三方依赖，浏览器端仅依赖 Web GUI 自带的 react。
 
 ---
 

package/lib/embedding.js

@@ -6,6 +6,7 @@
  */
 import path from "node:path";
 import fs from "node:fs";
+import { createHash } from "node:crypto";
 import { createRequire } from "node:module";
 import { fileURLToPath, pathToFileURL } from "node:url";
 
@@ -74,9 +75,14 @@ async function isAvailable() {
 async function embed(text) {
   const s = await initEngine();
   if (!s.available) return null;
-  const enc = s.tokenizer.encode(String(text).slice(0, 4000));
-  const ids = Array.from(enc.ids);
-  const mask = Array.from(enc.attention_mask);
+  // v2.5.0 修复轮 6：空/纯空白输入防护（空输入会产生无意义向量）
+  const raw = String(text ?? "");
+  if (raw.trim() === "") return null;
+  const enc = s.tokenizer.encode(raw.slice(0, 4000));
+  // v2.5.0 修复轮 6：token 级截断（bge 上限 512 token）——字符截断下 4000 个中文字符≈2000+ token
+  // 远超模型上限会导致 OrtRun() 报错（超长 query 首次必现"语义引擎不可用"）
+  const ids = Array.from(enc.ids).slice(0, 512);
+  const mask = Array.from(enc.attention_mask).slice(0, 512);
   const ort = s.ort;
   const iids = new ort.Tensor("int64", BigInt64Array.from(ids, BigInt), [1, ids.length]);
   const attn = new ort.Tensor("int64", BigInt64Array.from(mask, BigInt), [1, mask.length]);
@@ -114,11 +120,14 @@ function cosine(a, b) {
   return dot / (Math.sqrt(na) * Math.sqrt(nb));
 }
 
-/** 语义检索（缓存索引 128 维，query 同步降采样）。 */
+/** 语义检索（query 维度跟随索引：首次直出 512 维 / 缓存命中 128 维）。 */
 async function search(query, index, k = 5) {
   const qv = await embed(query);
   if (!qv || !index || index.length === 0) return [];
-  const qvCache = qv.filter((_, i) => i % 4 === 0);
+  // v2.5.0 修复轮 5：首次 built 的索引为 512 维（embedMany 直出），缓存命中为 128 维（saveIndex 降采样）——
+  // query 必须与 index 同维度，否则 cosine(128,512) 长度不等恒返回 0（首次检索全 0 分、排序无意义）
+  const dim = index[0]?.vec?.length ?? 0;
+  const qvCache = dim === qv.length ? qv : qv.filter((_, i) => i % 4 === 0);
   return index
     .map((item) => ({ ...item, score: cosine(qvCache, item.vec) }))
     .sort((x, y) => y.score - x.score)
@@ -205,13 +214,25 @@ function cachePath(root, book) {
   return path.join(root, EMBEDDING_CACHE_ROOT, "embedding", safe + ".json");
 }
 
-/** 保存索引（128 维降采样压缩）。 */
-function saveIndex(root, book, index) {
+/** 内容指纹（sha1，id+text）：章节内容/章节集变化 → 指纹变化 → 索引自动重建（不依赖版本号）。 */
+function fingerprint(chunks) {
+  const h = createHash("sha1");
+  for (const c of chunks || []) {
+    h.update(String(c.id));
+    h.update("\u0000");
+    h.update(String(c.text));
+    h.update("\u0001");
+  }
+  return h.digest("hex");
+}
+
+/** 保存索引（128 维降采样压缩；fp=构建输入内容指纹，用于章节更新后失效重建）。 */
+function saveIndex(root, book, index, fp = "") {
   try {
     const dir = path.dirname(cachePath(root, book));
     fs.mkdirSync(dir, { recursive: true });
     const compact = index.map((item) => ({ id: item.id, t: item.text.slice(0, 120), v: item.vec.filter((_, i) => i % 4 === 0) }));
-    fs.writeFileSync(cachePath(root, book), JSON.stringify({ model: "bge-small-zh-v1.5", dim: 128, items: compact }), "utf8");
+    fs.writeFileSync(cachePath(root, book), JSON.stringify({ model: "bge-small-zh-v1.5", dim: 128, fp, items: compact }), "utf8");
     return true;
   } catch { return false; }
 }
@@ -226,6 +247,16 @@ function loadIndex(root, book) {
   } catch { return null; }
 }
 
+/** 读取索引缓存 + 内容指纹（旧格式无 fp 时 fp=null，视为失效需重建；不靠版本号失效）。 */
+function loadIndexMeta(root, book) {
+  try {
+    const p = cachePath(root, book);
+    if (!fs.existsSync(p)) return { items: null, fp: null };
+    const data = JSON.parse(fs.readFileSync(p, "utf8"));
+    return { items: data.items.map((item) => ({ id: item.id, text: item.t, vec: item.v })), fp: data.fp ?? null };
+  } catch { return { items: null, fp: null }; }
+}
+
 /** 全文 → 语义段落（按空行/150 字切块，带章节标记）。 */
 function chunkText(text) {
   const raw = String(text);
@@ -249,4 +280,4 @@ function chunkText(text) {
 /** 显式释放（仅重置单例，WASM 内存由 GC 回收）。 */
 function reset() { enginePromise = null; engine = { available: false, error: null, tokenizer: null, session: null, ort: null, wasmPaths: null }; }
 
-export { status, isAvailable, embed, embedMany, cosine, search, saveIndex, loadIndex, chunkText, reset, detectImplicitEmotions, IMPLICIT_EMOTION_PROTOTYPES, MODEL_DIR, cachePath };
+export { status, isAvailable, embed, embedMany, cosine, search, saveIndex, loadIndex, loadIndexMeta, fingerprint, chunkText, reset, detectImplicitEmotions, IMPLICIT_EMOTION_PROTOTYPES, MODEL_DIR, cachePath };

package/lib/index.js

@@ -855,10 +855,13 @@ async function enrichSemanticImplicit(state, root, book, result, exec) {
       const text = await readTextFile(join(dir, chapter.file), exec);
       for (const p of embedding.chunkText(text)) chunks.push({ ...p, id: chapter.file + "|" + p.id });
     }
-    let index = embedding.loadIndex(root, book);
-    if (!index || index.length === 0) {
+    // v2.5.0 修复轮 4：内容指纹失效重建——章节更新后旧缓存不再命中（不靠版本号）
+    const { items: cachedIndex, fp: cachedFp } = embedding.loadIndexMeta(root, book);
+    const indexFp = embedding.fingerprint(chunks);
+    let index = cachedIndex;
+    if (!index || index.length === 0 || cachedFp !== indexFp) {
       index = await embedding.embedMany(chunks);
-      embedding.saveIndex(root, book, index);
+      embedding.saveIndex(root, book, index, indexFp);
     }
     const implicit = await embedding.detectImplicitEmotions(index);
     if (result.emotion?.quantification) {
@@ -1707,10 +1710,13 @@ function registerNovelStyleCheck(ctx, config) {
             for (const p of embedding.chunkText(allTexts[i])) allChunks.push({ ...p, id: String(i) + "|" + p.id });
           }
           const styleCacheKey = book + "__style_" + String(chapterArg).replace(/[\\/:*?"<>|]/g, "_");
-          let index = embedding.loadIndex(root, styleCacheKey);
-          if (!index || index.length === 0) {
+          // v2.5.0 修复轮 4：内容指纹失效重建——章节更新后旧缓存不再命中，避免语义对比静默失效（不靠版本号）
+          const { items: cachedIndex, fp: cachedFp } = embedding.loadIndexMeta(root, styleCacheKey);
+          const indexFp = embedding.fingerprint(allChunks);
+          let index = cachedIndex;
+          if (!index || index.length === 0 || cachedFp !== indexFp) {
             index = await embedding.embedMany(allChunks);
-            embedding.saveIndex(root, styleCacheKey, index);
+            embedding.saveIndex(root, styleCacheKey, index, indexFp);
           }
           // 目标章段落向量 vs 其他章段落向量，平均余弦作为语义风格相似度
           const targetChunks = allChunks.filter((ch) => ch.id.startsWith("0|"));
@@ -1818,7 +1824,7 @@ function normalizePlotEntry(entry) {
 }
 
 
-/** v0.8.0 设定管理（四张表：人物/地点/道具/时间线）。 */
+/** v0.8.0 设定管理（五张表：人物/地点/道具/时间线/世界观用语规范，worldview 于 v0.9.0 加入）。 */
 function settingsFile(root, book) {
   return join(novelDataDir(root), "settings", sanitizeSegment(book, "book") + ".json");
 }
@@ -1960,7 +1966,7 @@ function registerNovelStyleReport(ctx, config) {
       try {
         if ((await embedding.isAvailable())) {
           const emb = await import("./embedding.js");
-          semantic = await semanticStyleDistances(text, emb);
+          semantic = await semanticStyleDistances(text, emb, { root, book });
         }
       } catch { /* 语义距离失败不影响报告 */ }
       // 组装报告文本
@@ -1991,8 +1997,8 @@ function registerNovelStyleReport(ctx, config) {
       lines.push("  意象：负 " + Math.round((implicit.negative ?? 0) * 100) + "% / 正 " + Math.round((implicit.positive ?? 0) * 100) + "% / 歧义 " + Math.round((implicit.ambiguousRatio ?? 0) * 100) + "%");
       lines.push("  隐性情绪：" + (Object.keys(q.semanticImplicit?.distribution || {}).map((k) => k + "×" + q.semanticImplicit.distribution[k]).join(" ") || "（无）"));
       lines.push("");
-      lines.push("五、氛围光谱（什么味道，0~1）");
-      for (const ax of vibe.axes.slice(0, 8)) {
+      lines.push("五、氛围光谱（什么味道，0~1，共 12 轴）");
+      for (const ax of vibe.axes) {
         const bar = "█".repeat(Math.round(ax.score * 12)).padEnd(12, "░");
         lines.push("  " + ax.name.padEnd(5) + " " + bar + " " + ax.score.toFixed(2));
       }
@@ -2068,7 +2074,7 @@ function extractTopKeywords(text, topN) {
 function registerNovelSettings(ctx, config) {
   ctx.tools.register({
     name: "novel_settings",
-    description: "设定管理（四张表）：人物卡/地点卡/道具清单/时间线。list/add/update/delete 按 category 维护；scan 用规则扫描章节提取人物/道具候选供登记。用于续写时保持人物性格、环境、道具与时间线一致。",
+    description: "设定管理（五张表）：人物卡/地点卡/道具清单/时间线/世界观用语规范。list/add/update/delete 按 category 维护；scan 用规则扫描章节提取人物/道具候选供登记；detect 自动判断世界观文化基准。用于续写时保持人物性格、环境、道具、时间线与用语风格一致。",
     parameters: {
       type: "object",
       properties: {
@@ -2883,12 +2889,15 @@ function registerNovelSemanticSearch(ctx, config) {
         const text = await readTextFile(join(dir, chapter.file), exec);
         for (const p of embedding.chunkText(text)) chunks.push({ ...p, id: chapter.file + "|" + p.id });
       }
-      let index = embedding.loadIndex(root, book);
+      // v2.5.0 修复轮 4：内容指纹失效重建——章节更新后旧缓存不再命中（不靠版本号）
+      const { items: cachedIndex, fp: cachedFp } = embedding.loadIndexMeta(root, book);
+      const indexFp = embedding.fingerprint(chunks);
+      let index = cachedIndex;
       let cache = "hit";
-      if (!index || index.length === 0) {
+      if (!index || index.length === 0 || cachedFp !== indexFp) {
         index = await embedding.embedMany(chunks);
         cache = "built";
-        embedding.saveIndex(root, book, index);
+        embedding.saveIndex(root, book, index, indexFp);
       }
       const results = await embedding.search(query, index, top);
       return {
@@ -3378,7 +3387,7 @@ const WORKFLOW_TEXT = [
   "8. 工具链提示：续写/分析前可用 novel_plot 查看未回收伏笔（open 条目，含类型/优先级/提及章节）；写完新章节可用 novel_style_check 做风格自检（相似度+偏差+细节密度），novel_plot scan 自动更新伏笔提及记录；",
   "   世界观一致性：续写前先确认文化基准——novel_settings category=worldview action=detect 自动判断（或人工 add/update），动笔时对照其 bannedWords/recommended 用词，避免中西意象混搭（如欧式背景不写老夫/上香/时辰）；novel_continuity_check 会扫描用语冲突候选；",
   "   语用一致性：不只管词，还要管'怎么说话'——对照 worldview 的 speechStyle（title 称谓规范/honorBad 客套禁词/ritualBadPatterns 仪式禁式/tone 语气），人物开口前检查称谓是否欧式（Miss+名）、客套是否避免'提点/承蒙/在下'、宗教仪式是否点烛而非烧香/上X柱香、对话是否口语化不文言；",
-  "   novel_settings 维护四张设定表（人物/地点/道具/时间线），novel_summary 保存每章摘要（长书续写先读摘要再按需细读），novel_continuity_check 输出设定矛盾候选；",
+  "   novel_settings 维护五张设定表（人物/地点/道具/时间线/世界观用语规范），novel_summary 保存每章摘要（长书续写先读摘要再按需细读），novel_continuity_check 输出设定矛盾候选；",
   "   novel_sentence_analysis 结果自动缓存到书库数据目录 <root>/.novel-writer/analysis/（reportFile 字段），novel_keywords 结果同样落盘，需要重算时传 fresh=true。"
 ].join("\n");
 

package/lib/vibe.js

@@ -4,6 +4,9 @@
  * 纯规则加权 0 token；不做"贴标签"，只输出数值坐标 + 组合结论 + 证据链。
  */
 
+import fs from "node:fs";
+import path from "node:path";
+
 const clamp01 = (v) => Math.max(0, Math.min(1, v));
 
 /** 解析 evidence 词频数组 → { 词: 次数 } */
@@ -47,15 +50,16 @@ function semSignals(dist) {
 // v2.2.0 网文动作/对话/套路词群（参考 webnovel-writer genre-tropes 精选，封顶权重防污染）
 const WEB_NOVEL_SIGNALS = {
   // 词源标注：★=genre-tropes.md 现成套路词（lib/lexicons/webnovel-tropes.md），无标注=补充词
-  fluff: ["宝贝", "老婆", "亲爱的", "宠溺", "宠", "哄", "撒娇", "搂", "吻", "抱紧", "摸头", "腻", "甜", "心头一软", "小傻瓜", "乖乖", "甜宠★", "拉扯★", "白月光★"],
-  tearjerker: ["虐", "心碎", "对不起", "眼泪", "分手", "绝望", "崩溃★", "哽咽", "追妻★", "火葬场★", "求你", "别走", "疼", "悔婚★", "虐心★", "决绝★", "卑微★", "挽回★", "拒绝★", "替身★"],
-  blaze: ["冲", "杀", "斩", "轰", "燃", "战意", "怒吼", "碾压", "横扫", "秒杀★", "一招", "破", "震", "吞噬★", "越级★", "击杀★", "暴涨★", "大比★", "秘境★", "夺宝★", "宗门★", "练气★", "筑基★", "突破", "绝技★", "凶兽★", "宝物★", "金手指★", "天才★", "觉醒★", "传承★", "武道★", "念力★", "雷电★", "火焰★", "超能力★", "医术★", "体质★", "戒指★", "老爷爷★", "重生★", "签到★"],
+  // v2.5.0 修复轮 7：单字词全部改双字（"战/杀/血/死/慌/怕/疼/腻/查"等命中普通词会误判）
+  fluff: ["宝贝", "老婆", "亲爱的", "宠溺", "宠", "哄", "撒娇", "搂", "吻", "抱紧", "摸头", "甜", "心头一软", "小傻瓜", "乖乖", "甜宠★", "拉扯★", "白月光★"],
+  tearjerker: ["心碎", "对不起", "眼泪", "分手", "绝望", "崩溃★", "哽咽", "追妻★", "火葬场★", "求你", "别走", "心疼", "悔婚★", "虐心★", "决绝★", "卑微★", "挽回★", "拒绝★", "替身★"],
+  blaze: ["冲锋", "斩杀", "轰鸣", "燃烧", "震碎", "战意", "怒吼", "碾压", "横扫", "秒杀★", "一招", "吞噬★", "越级★", "击杀★", "暴涨★", "大比★", "秘境★", "夺宝★", "宗门★", "练气★", "筑基★", "突破", "绝技★", "凶兽★", "宝物★", "金手指★", "天才★", "觉醒★", "传承★", "武道★", "念力★", "雷电★", "火焰★", "超能力★", "医术★", "体质★", "戒指★", "老爷爷★", "重生★", "签到★"],
   absurd: ["离谱", "无语", "笑死", "尴尬", "吐槽", "什么鬼", "疯了", "惊了", "搞什么", "还有这种", "打脸★", "装逼★", "扮猪吃虎★", "嘲讽★", "震惊★", "下跪★", "跪地★", "退婚★", "废物★", "赌石★", "透视★", "隐藏身份★", "高攀不起", "悔婚★", "看不起★"],
   nightmare: ["诡异", "毛骨悚然", "不祥", "低语", "畸形", "腐烂", "扭曲", "窒息", "心悸", "不对劲", "渗人", "爬行", "凶兽★"],
-  angst: ["完了", "完了完了", "怎么办", "发抖", "紧张", "不安", "慌", "怕", "救命"],
+  angst: ["完了", "完了完了", "怎么办", "发抖", "紧张", "不安", "慌张", "慌乱", "害怕", "惧怕", "救命"],
   heartwarming: ["温柔", "安心", "踏实", "轻声", "轻轻", "抚", "哄睡", "暖意", "治愈"],
-  mystery: ["线索", "真相", "谜", "疑点", "秘密", "调查", "查", "发现", "蛛丝马迹★", "身份反转★", "震撼★"],
-  dark: ["血", "尸", "死", "折磨", "地狱", "残忍", "冷血", "杀"],
+  mystery: ["线索", "真相", "谜", "疑点", "秘密", "调查", "发现", "蛛丝马迹★", "身份反转★", "震撼★"],
+  dark: ["鲜血", "尸体", "死亡", "屠杀", "血泊", "杀戮", "折磨", "地狱", "残忍", "冷血"],
   lonesome: ["一个人", "独自", "没人", "寂寞", "想家", "陌生", "空荡荡", "孤零零"],
   aesthetic: ["月色", "清辉", "烟雨", "荷塘", "落英", "余韵", "浮光", "静默", "素净", "微凉", "风过", "细碎"],
   sensual: ["呼吸", "发烫", "贴近", "肌肤", "颤栗", "酥麻", "灼热", "喘息", "缠绕", "柔软", "耳畔", "温存"]
@@ -140,7 +144,8 @@ export function computeVibe(detect, emotion, text = "") {
   const linkThemes = (axis, base) => {
     for (const l of THEME_AXIS_LINK) {
       if (l.axis === axis && l.themes.some((t) => themes.includes(t) || theme.includes(t))) {
-        if ((base ?? 0) >= l.needBase || l.bonus >= 0.25) push(axis, l.bonus, 1.2, "题材联动:" + l.themes[0]);
+        // v2.5.0 修复轮 7：题材联动必须已有基础信号（去掉 bonus>=0.25 绕过，"提了一句恐怖"不再强拉噩梦感）
+        if ((base ?? 0) >= l.needBase) push(axis, l.bonus, 1.2, "题材联动:" + l.themes[0]);
         return;
       }
     }
@@ -213,7 +218,8 @@ export function computeVibe(detect, emotion, text = "") {
   linkThemes("mystery", axes.mystery?.score || 0);
 
   // ⑧ 热血激昂
-  const fightWords = freqIn([...(ev.modern || []), ...(ev.western || []), ...(ev.eastern || [])], ["战", "刀", "剑", "冲", "杀", "拳"]);
+  // v2.5.0 修复轮 7：战斗词改双字（单字"战/杀"命中"战战兢兢/抹杀"误判热血）
+  const fightWords = freqIn([...(ev.modern || []), ...(ev.western || []), ...(ev.eastern || [])], ["战斗", "冲锋", "斩杀", "刀光", "剑影", "铁拳", "战意", "厮杀", "刀剑", "拳风", "刀锋", "剑锋", "热血", "烈焰", "战鼓", "号角", "拔剑", "挥剑", "挥刀", "杀伐"]);
   push("blaze", clean === "anger" ? 0.8 : 0.15, 2.5, "clean=anger");
   push("blaze", Math.min(1, fightWords * 0.3), 1.5, "战斗词群");
   push("blaze", pos * 0.4, 1, "意象正向");
@@ -223,7 +229,8 @@ export function computeVibe(detect, emotion, text = "") {
 
   // ⑨ 荒诞无厘头（absurdWords 已在顶部定义）
   push("absurd", Math.min(1, absurdWords * 0.12), 1.2, "网络生活词");
-  push("absurd", emotion.dominant !== emotion.cleanDominant ? 0.55 : 0.08, 1.3, "情绪反差(表象≠内核)");
+  // v2.5.0 修复轮 7：反差信号需净化确有动作（caveat）才计，且降权（原来有污染时几乎必触发、权重偏大）
+  push("absurd", (emotion.dominant !== emotion.cleanDominant && emotion.caveat) ? 0.3 : 0.08, 1, "情绪反差(表象≠内核)");
   if (jocular) push("absurd", 0.4, 1.2, "戏谑语气");
   wnv("absurd", 1.1);
   linkThemes("absurd", axes.absurd?.score || 0);
@@ -267,7 +274,6 @@ export function computeVibe(detect, emotion, text = "") {
   const evidenceCount = (emotion.caveat ? 1 : 0) + (implicit.totalHits > 0 ? 1 : 0) + (Object.keys(dist).length > 0 ? 1 : 0) + (westernCount > 0 ? 1 : 0) + (modernCount > 0 ? 1 : 0) + (horror || mystery ? 1 : 0);
   const confidence = clamp01(topScore * 0.55 + evidenceCount * 0.07);
 
-  const combo = top.map((t) => t.name).join("+");
   // v2.5.0：不再由规则贴结论——判断交给大模型读报告
   const conclusion = "（测量数据已输出，请由大模型结合全部维度判断风格气质）";
   const evidence = [
@@ -297,24 +303,68 @@ export const STYLE_PROTOTYPES = {
   "怀旧乡愁": ["巷口的桂花还是那个味道", "老屋的木门吱呀作响，像在说别走", "她想起很多年前的黄昏，也是这样下雨", "照片泛黄了，可那时候的笑还是真的", "故乡的月亮，总是比别处圆"],
   "孤独疏离": ["她一个人坐在角落里，看着热气升起来", "没有人叫她的名字", "这座城很大，可没有一盏灯是为她亮的", "隔着一层雾，什么都够不着", "她习惯了把话咽回去"]
 };
+/** 等距抽样：超过 max 段时全书均匀取 max 段（首尾保留），避免只取开头序章失真。 */
+function sampleEvenly(arr, max) {
+  if (arr.length <= max) return arr;
+  const out = [];
+  const step = (arr.length - 1) / (max - 1);
+  for (let i = 0; i < max; i++) out.push(arr[Math.round(i * step)]);
+  return out;
+}
+
+/** 风格原型句向量内存缓存（STYLE_PROTOTYPES 固定 12 类×5 句，跨书复用，进程内一次）。 */
+const PROTO_VEC_CACHE = new Map();
+
+/** 语义距离按书缓存路径：<root>/.novel-writer/embedding/<书>__styleproto.json */
+function styleProtoCachePath(root, book) {
+  const safe = String(book).replace(/[\\/:*?"<>|]/g, "_");
+  return path.join(root, ".novel-writer", "embedding", safe + "__styleproto.json");
+}
+
 /**
  * 计算全书与各风格原型的语义距离表。
  * @param {string} text 全书文本
- * @param {object} emb embedding 模块（embed/cosine/chunkText）
+ * @param {object} emb embedding 模块（embed/cosine/chunkText/fingerprint）
+ * @param {object} [opts] 可选 { root, book }——提供后段落向量按书缓存（内容指纹失效重建，不靠版本号）
  * @returns {Array<{name:string,score:number}>} 按相似度降序
  */
-export async function semanticStyleDistances(text, emb) {
+export async function semanticStyleDistances(text, emb, opts = {}) {
   const results = [];
   try {
     if (!emb || typeof emb.embed !== "function") return results;
-    const chunks = emb.chunkText ? emb.chunkText(text).slice(0, 60) : [{ text: text.slice(0, 2000) }];
-    if (chunks.length === 0) return results;
-    const bookVecs = [];
-    for (const ch of chunks) {
+    // v2.5.0 修复轮 7：全书等距抽样 ≤60 段（原 slice(0,60) 只取开头，序章/引子会失真）
+    const allChunks = emb.chunkText ? emb.chunkText(text) : [{ text: text.slice(0, 2000) }];
+    if (allChunks.length === 0) return results;
+    const chunks = sampleEvenly(allChunks, 60);
+    const fp = typeof emb.fingerprint === "function" ? emb.fingerprint(chunks) : null;
+    // 段落向量：优先复用按书缓存（fp 匹配才命中；不匹配/缺失/损坏则重建）
+    let bookVecs = null;
+    let cacheFile = null;
+    if (opts?.root && fp) {
       try {
-        const vec = await emb.embed(ch.text);
-        if (vec && vec.length > 0) bookVecs.push(vec);
-      } catch { /* 单段失败跳过 */ }
+        cacheFile = styleProtoCachePath(opts.root, opts.book);
+        if (fs.existsSync(cacheFile)) {
+          const data = JSON.parse(fs.readFileSync(cacheFile, "utf8"));
+          if (data.fp === fp && Array.isArray(data.bookVecs) && data.bookVecs.length === chunks.length) {
+            bookVecs = data.bookVecs;
+          }
+        }
+      } catch { /* 缓存损坏 → 重建 */ }
+    }
+    if (!bookVecs) {
+      bookVecs = [];
+      for (const ch of chunks) {
+        try {
+          const vec = await emb.embed(ch.text);
+          if (vec && vec.length > 0) bookVecs.push(vec);
+        } catch { /* 单段失败跳过 */ }
+      }
+      if (cacheFile && fp && bookVecs.length > 0) {
+        try {
+          fs.mkdirSync(path.dirname(cacheFile), { recursive: true });
+          fs.writeFileSync(cacheFile, JSON.stringify({ fp, bookVecs }), "utf8");
+        } catch { /* 写缓存失败不影响结果 */ }
+      }
     }
     if (bookVecs.length === 0) return results;
     const dim = bookVecs[0].length;
@@ -322,14 +372,18 @@ export async function semanticStyleDistances(text, emb) {
     for (const vec of bookVecs) for (let i = 0; i < dim; i++) bookMean[i] += vec[i] / bookVecs.length;
     for (const [name, sentences] of Object.entries(STYLE_PROTOTYPES)) {
       try {
-        const pVecs = [];
-        for (const s of sentences) {
-          const vec = await emb.embed(s);
-          if (vec && vec.length === dim) pVecs.push(vec);
+        let pMean = PROTO_VEC_CACHE.get(name);
+        if (!pMean || pMean.length !== dim) {
+          const pVecs = [];
+          for (const s of sentences) {
+            const vec = await emb.embed(s);
+            if (vec && vec.length === dim) pVecs.push(vec);
+          }
+          if (pVecs.length === 0) continue;
+          pMean = new Array(dim).fill(0);
+          for (const vec of pVecs) for (let i = 0; i < dim; i++) pMean[i] += vec[i] / pVecs.length;
+          PROTO_VEC_CACHE.set(name, pMean);
         }
-        if (pVecs.length === 0) continue;
-        const pMean = new Array(dim).fill(0);
-        for (const vec of pVecs) for (let i = 0; i < dim; i++) pMean[i] += vec[i] / pVecs.length;
         const score = emb.cosine(bookMean, pMean);
         if (Number.isFinite(score)) results.push({ name, score: Math.round(score * 1000) / 1000 });
       } catch { /* 单原型失败跳过 */ }

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-novel-writer",
-  "version": "2.5.1",
-  "description": "小说写作助手插件（v2.5.0）：句式/情感/意象分析、伏笔设定管理、本地语义检索、氛围光谱、风格画像报告（测量与判断分离）。",
+  "version": "2.5.5",
+  "description": "小说写作助手插件（v2.5.5）：句式/情感/意象分析、伏笔设定管理、本地语义检索、氛围光谱、风格画像报告（测量与判断分离）。",
   "type": "module",
   "main": "lib/index.js",
   "exports": {

package/skills/novel-writing/SKILL.md

@@ -31,7 +31,7 @@
 
 ## 设定管理（v0.8.0 新增）
 
-- §BT§novel_settings§BT§：四张表（人物卡/地点卡/道具清单/时间线），list/add/update/delete/scan；
+- §BT§novel_settings§BT§：五张表（人物卡/地点卡/道具清单/时间线/世界观用语规范），list/add/update/delete/scan/detect；
 - 登记新人物/新地点/道具去向，时间线记录"第几天/倒计时"；
 - §BT§novel_continuity_check§BT§：续写前跑一次，输出设定矛盾候选（数字口径/人物缺场/别名/重复）。