dsh-plugin-lookatstudy 0.11.1 → 0.12.0

package/lib/epub-parser-oH96guBW.mjs

@@ -0,0 +1,306 @@
+import { n as htmlToMarkdown } from "./html-article-Da8ksU0i.mjs";
+import { n as readZipText, t as readZip } from "./zip-reader-KnRrq0av.mjs";
+//#region src/vendor/epub-parser.ts
+/** 提取某标签的全部出现(OPF 的属性顺序不定,先抓整标签再逐个提属性)。 */
+function tags(xml, tagName) {
+	const out = [];
+	const re = new RegExp(`<${tagName}\\b[^>]*>`, "g");
+	let m;
+	while ((m = re.exec(xml)) !== null) {
+		const raw = m[0];
+		const attrs = {};
+		const attrRe = /([\w:-]+)\s*=\s*"([^"]*)"/g;
+		let a;
+		while ((a = attrRe.exec(raw)) !== null) attrs[a[1]] = a[2];
+		out.push({
+			raw,
+			attrs
+		});
+	}
+	return out;
+}
+function firstTagText(xml, tagName) {
+	return xml.match(new RegExp(`<${tagName}[^>]*>([^<]*)</${tagName}>`, "i"))?.[1]?.trim() ?? "";
+}
+/** zip 内路径归一:posix 分隔 + href 相对 OPF 目录解析。 */
+function resolveZipPath(opfPath, href) {
+	const cleanHref = decodeURIComponent(href.split("#")[0] ?? href).replace(/\\/g, "/");
+	if (!opfPath.includes("/")) return cleanHref.replace(/^\.\//, "");
+	const parts = `${opfPath.slice(0, opfPath.lastIndexOf("/"))}/${cleanHref}`.split("/");
+	const resolved = [];
+	for (const p of parts) if (p === "..") resolved.pop();
+	else if (p !== "." && p !== "") resolved.push(p);
+	return resolved.join("/");
+}
+/** EPUB3 nav.xhtml 或 EPUB2 toc.ncx → href(去 fragment)→ 章节标题。 */
+function parseTocLabels(tocXml, isNcx) {
+	const labels = /* @__PURE__ */ new Map();
+	if (isNcx) {
+		const blockRe = /<navPoint\b[\s\S]*?<\/navPoint>/g;
+		let m;
+		while ((m = blockRe.exec(tocXml)) !== null) {
+			const label = firstTagText(m[0], "text");
+			const src = m[0].match(/<content[^>]+src="([^"]+)"/i)?.[1];
+			if (label && src) labels.set(decodeURIComponent(src.split("#")[0].replace(/\\/g, "/")), label);
+		}
+	} else {
+		const aRe = /<a\b[^>]*href="([^"]+)"[^>]*>([\s\S]*?)<\/a>/gi;
+		let m;
+		while ((m = aRe.exec(tocXml)) !== null) {
+			const label = (m[2] ?? "").replace(/<[^>]+>/g, "").trim();
+			if (label) labels.set(decodeURIComponent((m[1] ?? "").split("#")[0].replace(/\\/g, "/")), label);
+		}
+	}
+	return labels;
+}
+function sanitizeFileName(title) {
+	return (title || "chapter").replace(/[\\/:*?"<>|#\s]+/g, "-").replace(/^-+|-+$/g, "").slice(0, 50) || "chapter";
+}
+/** 文件内章标记。两类形态:heading 行(# 前缀)与**裸行**(Gutenberg 常见——章号在
+* 原书 html 里是段落不是标题,转 markdown 后成裸文本 "CHAPTER XX.")。
+* 裸行限行长(≤60)防正文中引用句误切;罗马数字/裸序号低置信需序列验证。 */
+const CH_HEADING = /^(?:#{1,3}\s*)?(?:CHAPTER|Chapter|chap\.)\s+[IVXLC0-9]{1,7}\b[.:)]?\s*(.*)$/;
+const LETTER_HEADING = /^(?:#{1,3}\s*)?(?:Letter|LETTER)\s+\d{1,3}\b[.:)]?\s*(.*)$/;
+const ROMAN_HEADING = /^(?:#{1,3}\s*)?(?:[—[]\s*)?([IVXLC]{1,7})\s*[—\]]?\.?\s*$/;
+const NUM_HEADING = /^(?:#{1,3}\s*)?[—[]?\s*(\d{1,3})\s*[—\]]?\.?\s*$/;
+const romanValue = (s) => {
+	const vals = {
+		I: 1,
+		V: 5,
+		X: 10,
+		L: 50,
+		C: 100
+	};
+	let v = 0;
+	for (let i = 0; i < s.length; i++) {
+		const cur = vals[s[i]] ?? 0;
+		const next = vals[s[i + 1]] ?? 0;
+		v += cur < next ? -cur : cur;
+	}
+	return v;
+};
+/**
+* 文件内按章标记切分。返回 null = 无高置信切分(保持单章)。
+* 置信规则:CHAPTER/Letter 标记 ≥2 即切;罗马数字/裸序号须形成连续递增
+* 序列(≥3 个)才切——正文中孤立的 "I" / "2" 不当章号。
+* 首标记前的引言/卷头(<800 字符)并入首章;够长则独立成"引言"章。
+*/
+function splitChaptersInBody(body, opts) {
+	const lines = body.split("\n");
+	const marks = [];
+	for (let i = 0; i < lines.length; i++) {
+		const ln = lines[i].trimEnd();
+		if (!ln || ln.length > 60) continue;
+		const strip = (s) => s.replace(/^#{1,3}\s*/, "");
+		let m = ln.match(CH_HEADING);
+		if (m) {
+			marks.push({
+				line: i,
+				title: strip(ln),
+				kind: "ch",
+				seq: 0
+			});
+			continue;
+		}
+		m = ln.match(LETTER_HEADING);
+		if (m) {
+			marks.push({
+				line: i,
+				title: strip(ln),
+				kind: "letter",
+				seq: 0
+			});
+			continue;
+		}
+		m = ln.match(ROMAN_HEADING);
+		if (m && (ln.startsWith("#") || /[—[]/.test(ln))) {
+			marks.push({
+				line: i,
+				title: strip(ln),
+				kind: "roman",
+				seq: romanValue(m[1])
+			});
+			continue;
+		}
+		m = ln.match(NUM_HEADING);
+		if (m && (ln.startsWith("#") || /[—[]/.test(ln))) {
+			marks.push({
+				line: i,
+				title: strip(ln),
+				kind: "num",
+				seq: parseInt(m[1], 10)
+			});
+			continue;
+		}
+	}
+	const minHigh = opts?.minHighConf ?? 2;
+	const highConf = marks.filter((mk) => mk.kind === "ch" || mk.kind === "letter");
+	let splitMarks;
+	if (highConf.length >= minHigh) splitMarks = highConf;
+	else {
+		splitMarks = [];
+		for (const kind of ["roman", "num"]) {
+			const seq = marks.filter((mk) => mk.kind === kind);
+			const chain = [];
+			let expect = 0;
+			for (const mk of seq) if (expect === 0) {
+				if (mk.seq === 1 || mk.seq === 2) {
+					chain.push(mk);
+					expect = mk.seq + 1;
+				}
+			} else if (mk.seq === expect) {
+				chain.push(mk);
+				expect++;
+			}
+			if (chain.length >= 3) splitMarks.push(...chain);
+		}
+		if (splitMarks.length < 3) return null;
+	}
+	if (splitMarks.length < 1) return null;
+	const out = [];
+	const preamble = lines.slice(0, splitMarks[0].line).join("\n").trim();
+	for (let k = 0; k < splitMarks.length; k++) {
+		const start = splitMarks[k].line;
+		const end = k + 1 < splitMarks.length ? splitMarks[k + 1].line : lines.length;
+		let content = lines.slice(start + 1, end).join("\n").trim();
+		if (k === 0 && preamble && preamble.length < 800) content = `${preamble}\n\n${content}`.trim();
+		out.push({
+			title: splitMarks[k].title,
+			content
+		});
+	}
+	if (preamble && preamble.length >= 800) out.unshift({
+		title: "引言",
+		content: preamble
+	});
+	return out.filter((c) => c.content.replace(/\s/g, "").length > 0);
+}
+/**
+* 清理电子书正文噪声(Gutenberg 头/尾块、1894 插图版装饰行)。
+* 只删确定性标志,不猜内容:
+* - 头:"The Project Gutenberg eBook of…" 段落(heading 形态)
+* - 尾:"THE FULL PROJECT GUTENBERG™ LICENSE" 起截断到文件尾;"End of the Project Gutenberg" 行起同理
+* - 行内:`[ _Copyright 1894 by …_ ]` 插图版装饰片段
+*/
+function sanitizeEpubBody(body) {
+	let md = body;
+	const lines0 = md.split("\n");
+	for (let i = 0; i < lines0.length; i++) {
+		const ln = lines0[i].trim();
+		if (ln.length <= 100 && (/(?:THE\s+)?FULL PROJECT GUTENBERG/i.test(ln) || /END OF THE PROJECT GUTENBERG/i.test(ln) || ln === "START: FULL LICENSE" || ln === "START: THE FULL LICENSE")) {
+			md = lines0.slice(0, i).join("\n");
+			break;
+		}
+	}
+	md = md.replace(/^#{1,3}\s*(?:The Project Gutenberg eBook[^*\n]*|Project Gutenberg[^*\n]*)\n[\s\S]*?\n(?:\n|$)/im, "");
+	md = md.replace(/^#{1,3}[ \t]*$/gm, "");
+	md = md.replace(/\\?\[?\s*_?Copyright 1[6-9]\d\d[^_\n]*_?\s*\\?]?/g, "");
+	return md.trim();
+}
+function parseEpub(buf) {
+	const entries = readZip(buf);
+	const read = (p) => readZipText(entries, p);
+	const opfPath = read("META-INF/container.xml").match(/full-path="([^"]+)"/i)?.[1];
+	if (!opfPath) throw new Error("epub 结构异常:找不到 container.xml 里的 OPF 路径");
+	const opf = read(opfPath);
+	if (!opf) throw new Error(`epub 结构异常:OPF 文件缺失(${opfPath})`);
+	const bookTitle = firstTagText(opf, "dc:title") || "未命名电子书";
+	const manifest = /* @__PURE__ */ new Map();
+	for (const t of tags(opf, "item")) manifest.set(t.attrs["id"] ?? "", {
+		href: t.attrs["href"] ?? "",
+		mediaType: (t.attrs["media-type"] ?? "").toLowerCase(),
+		properties: t.attrs["properties"] ?? ""
+	});
+	const spineIds = tags(opf, "itemref").map((t) => t.attrs["idref"] ?? "").filter(Boolean);
+	const spineTocId = opf.match(/<spine\b[^>]*\btoc="([^"]+)"/i)?.[1];
+	let tocLabels = /* @__PURE__ */ new Map();
+	const navItem = [...manifest.values()].find((it) => it.properties.split(/\s+/).includes("nav"));
+	if (navItem?.href) tocLabels = parseTocLabels(read(resolveZipPath(opfPath, navItem.href)), false);
+	if (tocLabels.size === 0 && spineTocId && manifest.has(spineTocId)) tocLabels = parseTocLabels(read(resolveZipPath(opfPath, manifest.get(spineTocId).href)), true);
+	const opfDirKey = (href) => decodeURIComponent(href.split("#")[0] ?? href).replace(/\\/g, "/");
+	/** 无标题标记(不再编造"第 N 章";配对/课程设计层的锚点) */
+	const UNTITLED_MARK = "未命名章节";
+	const chapters = [];
+	let n = 0;
+	for (const id of spineIds) {
+		const item = manifest.get(id);
+		if (!item?.href) continue;
+		if (!(item.mediaType === "application/xhtml+xml" || item.mediaType === "text/html" || /\.(xhtml|html|htm)$/i.test(item.href))) continue;
+		if (item.properties.split(/\s+/).includes("nav")) continue;
+		const xhtml = read(resolveZipPath(opfPath, item.href));
+		if (!xhtml) continue;
+		const md = htmlToMarkdown(xhtml, { stripImages: true });
+		if (!md || md.replace(/[#\s>*-]/g, "").length < 8) continue;
+		n++;
+		const body0 = md.startsWith("# ") && md.includes("\n") ? md.slice(md.indexOf("\n") + 1).trim() : md;
+		const firstHeading = md.startsWith("# ") ? md.split("\n")[0].slice(2).trim() : "";
+		const fileTitle = tocLabels.get(opfDirKey(item.href)) || firstHeading || UNTITLED_MARK;
+		if (/^(contents|table of contents)$/i.test(fileTitle) && body0.replace(/\s/g, "").length < 4e3) continue;
+		if (body0.replace(/\s/g, "").length < 1500 && /project gutenberg|copyright/i.test(body0)) continue;
+		if (/^(版权信息|版权|版权页|著作权|版權)/.test(fileTitle)) continue;
+		if ((body0.match(/^(书名|作者|译者|编者|丛书|出版社|出版时间|出版年|ISBN|版次|定价|装帧|字数|品牌|出品方)[：:]/gm) ?? []).length >= 3 && body0.replace(/\s/g, "").length < 2500) continue;
+		if (/^目\s*录$/.test(fileTitle)) {
+			const lines = body0.split("\n").map((l) => l.trim()).filter(Boolean);
+			if (lines.length >= 3 && lines.filter((l) => l.includes("](")).length / lines.length >= .6) continue;
+		}
+		const body = sanitizeEpubBody(body0);
+		if (body.replace(/\s/g, "").length < 8) continue;
+		const titleIsLicense = /FULL PROJECT GUTENBERG|LICENSE/i.test(fileTitle);
+		const split = splitChaptersInBody(body, titleIsLicense ? { minHighConf: 1 } : void 0);
+		if (split && split.length >= (titleIsLicense ? 1 : 2)) {
+			for (const seg of split) {
+				chapters.push({
+					path: `chapters/${String(n).padStart(2, "0")}-${sanitizeFileName(seg.title)}.md`,
+					title: seg.title,
+					markdown: `# ${seg.title}\n\n${seg.content}`
+				});
+				n++;
+			}
+			n--;
+			continue;
+		}
+		if (titleIsLicense) {
+			if (body.replace(/\s/g, "").length > 2e3) chapters.push({
+				path: `chapters/${String(n).padStart(2, "0")}-appendix.md`,
+				title: "附录",
+				markdown: `# 附录\n\n${body}`
+			});
+			continue;
+		}
+		chapters.push({
+			path: `chapters/${String(n).padStart(2, "0")}-${sanitizeFileName(fileTitle)}.md`,
+			title: fileTitle,
+			markdown: `# ${fileTitle}\n\n${body}`
+		});
+	}
+	const absorbed = /* @__PURE__ */ new Set();
+	for (let i = 0; i + 1 < chapters.length; i++) {
+		const stub = chapters[i];
+		const next = chapters[i + 1];
+		if (absorbed.has(stub.path)) continue;
+		const stubBody = stub.markdown.includes("\n") ? stub.markdown.slice(stub.markdown.indexOf("\n") + 1).trim() : "";
+		if (stub.title === UNTITLED_MARK || stub.title === "附录") continue;
+		if (stubBody.replace(/\s/g, "").length > 300) continue;
+		if (next.title !== UNTITLED_MARK) continue;
+		const nextBody = next.markdown.includes("\n") ? next.markdown.slice(next.markdown.indexOf("\n") + 1).trim() : next.markdown;
+		chapters[i] = {
+			path: stub.path,
+			title: stub.title,
+			markdown: `# ${stub.title}\n\n${stubBody}\n\n${nextBody}`
+		};
+		absorbed.add(stub.path);
+		chapters.splice(i + 1, 1);
+		i--;
+	}
+	if (chapters.length === 0) throw new Error("epub 里没有可识别的章节文本");
+	return {
+		title: bookTitle,
+		chapters
+	};
+}
+/** 文件夹导入路径用:整本书压平成一个 markdown(全部 H1 降为 H2,给结构设计当 anchor 拆章)。 */
+function parseEpubFlat(buf) {
+	return parseEpub(buf).chapters.map((c) => c.markdown.replace(/^# /gm, "## ")).join("\n\n");
+}
+//#endregion
+export { parseEpubFlat };

package/lib/{html-article-C6nhqJbX.mjs → html-article-Da8ksU0i.mjs}

@@ -414,9 +414,43 @@ function extractArticle(html, baseUrl = "") {
 	if (!body || body.replace(/[#\s>*|-]/g, "").length < 40) return null;
 	return {
 		title,
-		markdown: body.split("\n")[0]?.replace(/^#\s+/, "").trim() === title && body.startsWith("# ") ? body : `# ${title}\n\n${body}`
+		markdown: stripTailNavigation(body.split("\n")[0]?.replace(/^#\s+/, "").trim() === title && body.startsWith("# ") ? body : `# ${title}\n\n${body}`)
 	};
 }
+/**
+* 尾部**站点模板指纹**清理(upstream v0.23.1, 2026-08-23 真实站点采样驱动; vendored verbatim)。
+* 原则:规则只管高置信度的**机器生成模板**(跨文章稳定、作为正文出现概率≈0);
+* 除此之外的不确定判断(作者自己写的推广段/水印/相关阅读算不算正文)一律不猜——
+* 那是课程设计层(tutor/Step4)的职责,解析层动了就是误删正文
+* (实测:按"欢迎关注公众号"删规则,把 CSDN 作者自己写的推广段删了)。
+*/
+const NAV_TAIL_PATTERNS = [
+	/返回\S{0,6}[，,]?\s*查看更多/,
+	/点击进入\S{0,8}首页/,
+	/^热门文章$/,
+	/^最新文章$/,
+	/^目录\s*$/,
+	/^END\b.*版权/
+];
+/** 行内导航后缀:正文与模板被并成一行时剥掉(精确短语,全文安全) */
+const INLINE_NAV_SUFFIXES = ["目录 热门文章 最新文章"];
+function stripTailNavigation(md) {
+	if (!md) return md;
+	for (const suffix of INLINE_NAV_SUFFIXES) md = md.split(suffix).join("");
+	const lines = md.split("\n");
+	let end = lines.length;
+	for (let i = lines.length - 1; i >= Math.max(0, lines.length - 25); i--) {
+		const ln = lines[i].trim();
+		if (!ln) continue;
+		const bare = ln.replace(/^#{1,6}\s*/, "").replace(/^[>\-*]\s*/, "");
+		const isNav = NAV_TAIL_PATTERNS.some((re) => re.test(bare) || re.test(ln));
+		const isBareImage = /^!\[[^\]]*\]\([^)]*\)$/.test(ln) || /^\S+\.(jpeg|jpg|png|gif|webp)\)?$/i.test(ln);
+		if (isNav || isBareImage) continue;
+		end = i + 1;
+		break;
+	}
+	return lines.slice(0, end).join("\n").trimEnd();
+}
 function htmlToMarkdownOf(roots, stripImages) {
 	const lines = [];
 	emitBlocks(roots, stripImages, lines);

package/lib/index.mjs

@@ -1,8 +1,9 @@
 import { n as inflateZlib } from "./inflate-DIKUrTRi.mjs";
-import { t as extractArticle } from "./html-article-C6nhqJbX.mjs";
+import { t as extractArticle } from "./html-article-Da8ksU0i.mjs";
 import { existsSync, mkdirSync, readFileSync, renameSync, writeFileSync } from "node:fs";
 import { basename, dirname, join, relative, sep } from "node:path";
 import z from "@deepseek-ai/schemastery";
+import { fileURLToPath } from "node:url";
 import { homedir } from "node:os";
 import { randomBytes } from "node:crypto";
 import { readFile, readdir } from "node:fs/promises";
@@ -1320,6 +1321,20 @@ function courseToPackMarkdown(course) {
 * message channel were removed once the in-client study tab superseded them.
 * @module dsh-plugin-lookatstudy/dashboard
 */
+/** Wiring handed in by `apply`. */
+/**
+* The plugin's own version, read from the package.json sitting beside the
+* running module (src in dev, lib in install) — strictly the installed build,
+* no build-time inlining that could drift. Upstream v0.24.0 port (settings
+* About row); empty string on any read failure (display degrades, never throws).
+*/
+function pluginVersion() {
+	try {
+		return String(JSON.parse(readFileSync(fileURLToPath(new URL("../package.json", import.meta.url)), "utf8")).version ?? "");
+	} catch {
+		return "";
+	}
+}
 /**
 * Assemble the whole workbench state (pure read; the lesson HTML is rendered
 * server-side from the sanitized markdown pipeline).
@@ -1495,7 +1510,8 @@ function registerDashboard(webServer, deps) {
 			if (req.method === "GET" && pathname === "/lookatstudy/api/state") {
 				sendJson(res, 200, {
 					...workbenchState(deps.store.get(), /* @__PURE__ */ new Date()),
-					statePath: deps.statePath
+					statePath: deps.statePath,
+					version: pluginVersion()
 				});
 				return;
 			}
@@ -1636,9 +1652,7 @@ const ZH = {
 	"pane.tutor": "导师",
 	"pane.bb": "黑板",
 	"viewtab.teach": "讲解",
-	"viewtab.mind": "🧠 脑图",
 	"viewtab.cmap": "🕸 概念图",
-	"viewtab.mind.title": "本课结构一图流(markmap)",
 	"viewtab.cmap.title": "本课概念关系(ELK 布局)",
 	"loading": "加载中…",
 	"start.title": "一键准备学习:建立学习工作区、开启会话并让导师就位",
@@ -1705,7 +1719,6 @@ const ZH = {
 	"bb.empty": "黑板还空着",
 	"bb.empty.hint": "在左侧课程树选择一课",
 	"bb.mastery": "掌握度 {pct}%",
-	"bb.fallback.mind": "脑图渲染不可用(网络受限) — 已回退讲解视图",
 	"bb.fallback.cmap.empty": "概念图需要先由导师定义本课概念",
 	"bb.fallback.cmap": "概念图渲染不可用(布局引擎加载失败) — 已回退讲解视图",
 	"bb.notes": "笔记",
@@ -1729,6 +1742,8 @@ const ZH = {
 	"settings.stats.xp": "XP {xp} · Lv{level}({pct}%)",
 	"settings.stats.today": "今日 {xp}/{goal}",
 	"settings.stats.streak": "连续 {days} 天(最长 {best},剩冻结 {freeze})",
+	"settings.about": "关于",
+	"settings.about.hint": "当前运行的构建版本,点击查看该版更新内容。",
 	"settings.stateFile": "状态文件",
 	"settings.stateFile.hint": "学习进度存放于本机 JSON,换机可迁移",
 	"dock.title": "学习状态:待复习 {due} · 连续 {streak} 天 · Lv{level} — 点上方「学习」页签进入",
@@ -2155,7 +2170,7 @@ async function scanFolder(rootDir, onProgress, options) {
 	for (const doc of dedupedDocs) {
 		if (doc.kind !== "pptx") continue;
 		try {
-			const { parsePptx } = await import("./pptx-parser-CD5pR2cj.mjs");
+			const { parsePptx } = await import("./pptx-parser-B9IDkiGM.mjs");
 			const result = await parsePptx(await readFile(join(rootDir, doc.path)));
 			for (const img of result.images ?? []) pptxImages.push({
 				path: `${doc.path}#slide${img.slideNumber}.png`,
@@ -2346,7 +2361,7 @@ async function readFileWithKind(absPath, kind, parsePdf) {
 	}
 	if (kind === "pptx") {
 		const buf = await readFile(absPath);
-		const { parsePptx } = await import("./pptx-parser-CD5pR2cj.mjs");
+		const { parsePptx } = await import("./pptx-parser-B9IDkiGM.mjs");
 		return (await parsePptx(buf)).markdown;
 	}
 	if (kind === "docx") {
@@ -2356,7 +2371,7 @@ async function readFileWithKind(absPath, kind, parsePdf) {
 	}
 	if (kind === "epub") {
 		const buf = await readFile(absPath);
-		const { parseEpubFlat } = await import("./epub-parser-DvlKap-d.mjs");
+		const { parseEpubFlat } = await import("./epub-parser-oH96guBW.mjs");
 		return parseEpubFlat(buf);
 	}
 	if (kind === "ipynb") {
@@ -3184,7 +3199,10 @@ function normalizeUrlIdentity(url) {
 }
 //#endregion
 //#region src/vendor/pdf-text.ts
-var pdf_text_exports = /* @__PURE__ */ __exportAll({ parsePdfText: () => parsePdfText });
+var pdf_text_exports = /* @__PURE__ */ __exportAll({
+	normalizeRadicals: () => normalizeRadicals,
+	parsePdfText: () => parsePdfText
+});
 function decodePdfString(s) {
 	return s.replace(/\\([0-7]{1,3}|.)/g, (_m, esc) => {
 		if (/^[0-7]+$/.test(esc)) return String.fromCharCode(parseInt(esc, 8));
@@ -3262,7 +3280,17 @@ function parsePdfText(buf) {
 		if (text.trim()) parts.push(text);
 		streamRe.lastIndex = end + 9;
 	}
-	return parts.join("\n").replace(/[ \t]+\n/g, "\n").replace(/\n{3,}/g, "\n\n").trim();
+	return normalizeRadicals(parts.join("\n").replace(/[ \t]+\n/g, "\n").replace(/\n{3,}/g, "\n\n").trim());
+}
+/**
+* 康熙部首区(U+2F00-U+2FDF)→CJK 统一表意区归一(upstream v0.23.1 port, verbatim)。
+* 部分中文 PDF 的 ToUnicode 映射落在部首区(d2l-zh 真书采样实测:"⼿⼀⽅"应读
+* "手一方"),部首区上检索/匹配全部断裂。逐字符 NFKC(部首→汉字是 Unicode 标准
+* 一对一兼容映射,不动其他任何字符)。
+*/
+function normalizeRadicals(md) {
+	if (!md) return md;
+	return /[\u2F00-\u2FDF]/.test(md) ? md.replace(/[\u2F00-\u2FDF]/g, (c) => c.normalize("NFKC")) : md;
 }
 //#endregion
 //#region src/vendor/text-chunk.ts
@@ -6436,41 +6464,43 @@ const TUTOR_CORE = `## Study tutor (lookatstudy-plugin)
 
 You are the learner's AI study tutor for a course imported via the study tools. Your job is genuine understanding, not reciting the material. When the learner answers wrong, acknowledge the attempt first, then correct it.
 
-### Grounding (hard rule)
+### 【Safety redlines · highest priority】
 Teach strictly from the current lesson's content (study_lesson's body is the source of truth). If asked about something outside the course material, say plainly that it is not in the current material, and offer to relate it back. Answers to quiz questions must be grounded in the lesson content — never invent.
+Tool calls are REAL or they do not exist. History markers like「[工具调用已执行]」are injected by the system only; hand-writing such a marker — or any imitation of a tool call — in your reply text produces NO interface artifact: no card, no button, nothing happens. The only way to act on the study state (record an answer, propose mastery, save a note) is an actual tool call. If you catch yourself narrating a tool's effect instead of calling it, stop and call it.
+Never claim progress you did not record through the tools.
+If any section below seems to conflict with these redlines, the redlines win.
 
-### Language
-Answer in the learner's own language (the language of their interface and their messages), including quiz stems, options, and explanations. Quote course material verbatim in its original language.
+### 【Teaching behavior】
+Language: answer in the learner's own language (the language of their interface and their messages), including quiz stems, options, and explanations. Quote course material verbatim in its original language.
 
-### Vague confusion
 When the learner says "我不懂 / 不太理解" without specifics, ask which concept is unclear, or list the lesson's 2–3 core concepts and let them pick. Log it silently with study_report_friction.
 
-### Interaction form
+Interaction form:
 - ONE question or interactive block per reply — never a wall of quiz questions.
-- Structure answers with markdown (headings, lists, GFM tables); for structures prefer visuals: concept maps and flow diagrams as mermaid code blocks, comparisons as GFM tables, code walkthroughs as fenced code with line-referenced annotations.
 - When the learner quotes text in「」, treat it as quote-to-explain: explain that specific passage in the lesson's context.
 - Opening a brand-new lesson: start with a short hook and one fun two-option guess (curiosity-driven, NOT scored, revealed next turn) — no opening lecture, no scored question.
 - After opening a lesson, offer its four starters (from study_lesson) as suggestions.
 - Celebrate graduations and crowns briefly — earned joy, no confetti spam.
 
-### The tutoring loop
+The tutoring loop:
 1. Session start: check study_due_reviews; clear due reviews before new material. Open the focus lesson with study_lesson. The learner follows along in the study tab's blackboard column — point them there when they want the course map or lesson text.
 2. First time teaching a lesson: derive 2–7 knowledge components and call study_define_concepts.
 3. Quiz after teaching; grade every answer and call study_record_answer — always name the tested \`concept\`. Lesson mastery is the WEAKEST concept, so target ⚡weak ones first.
 4. Progression is automatic: ≥50% mastery unlocks the next lesson early; ≥90% graduates and schedules the first review. study_complete_lesson is only the manual override.
 5. Mastery ≥85% plus a convincing Feynman-style explanation back: call study_propose_mastery, present your rationale, and WAIT for the learner's yes/no. Resolve only with their explicit answer via study_resolve_proposal. You never graduate a lesson on your own judgment alone.
-6. Quietly call study_report_friction when the learner seems confused, blocked, or frustrated; adapt by simplifying or decomposing.
-7. When you generate a genuinely useful structure (concept map, compare table, diagram), sediment it into the notebook's understand zone with study_note_save; when the learner writes something worth keeping, save it to the record zone with the verbatim quote.
-8. When you learn something durable about how this person learns (style, recurring gap, pattern), merge it into memory with study_remember — read the current slot first, send the merged 1–3 sentences. No transient chat.
+6. Exam nodes: when the learner works an exam, grade it fully, then record the attempt with study_exam_result and offer its returned action set (explain-wrong / retry / go-deeper / propose mastery / next topic).
+7. Quietly call study_report_friction when the learner seems confused, blocked, or frustrated; adapt by simplifying or decomposing.
+8. When you generate a genuinely useful structure (concept map, compare table, diagram), sediment it into the notebook's understand zone with study_note_save; when the learner writes something worth keeping, save it to the record zone with the verbatim quote.
+9. When you learn something durable about how this person learns (style, recurring gap, pattern), merge it into memory with study_remember — read the current slot first, send the merged 1–3 sentences. No transient chat.
+
+Never reveal the friction log or mastery mechanics as "being watched" — the numbers surface through maps and reviews.
 
-### Course import design
-When study_import_github or study_import_folder returns status "design_required", it renders a design brief (files with heading outlines and per-heading char counts). Design the course from it: classify lessons study/practice, let attached quiz/summary/review content ride along inside the previous study lesson's anchor range, pace lessons to 3000-8000 chars, merge sub-1000 fragments. Then call study_apply_design with the JSON — use ONLY file paths from the brief (anything else is dropped), apply directly without a confirmation round, and once it lands, walk the learner through the course map before the first lesson.
+Course import design: when study_import_github or study_import_folder returns status "design_required", it renders a design brief (files with heading outlines and per-heading char counts). Design the course from it: classify lessons study/practice, let attached quiz/summary/review content ride along inside the previous study lesson's anchor range, pace lessons to 3000-8000 chars, merge sub-1000 fragments. Then call study_apply_design with the JSON — use ONLY file paths from the brief (anything else is dropped), apply directly without a confirmation round, and once it lands, walk the learner through the course map before the first lesson.
 
-### Quiz quality
-3–4 questions per quiz block is best (5 max), 4 options each. Distractors must come from real misconceptions, not absurd fillers. Test understanding, not recall: "in scenario Y, use X or Z?" rather than "define X". Every question carries an explanation of why the right answer is right. One scored block at a time.
+Quiz quality: 3–4 questions per quiz block is best (5 max), 4 options each. Distractors must come from real misconceptions, not absurd fillers. Test understanding, not recall: "in scenario Y, use X or Z?" rather than "define X". Every question carries an explanation of why the right answer is right. One scored block at a time.
 
-### Integrity
-Never claim progress you did not record through the tools. Never reveal the friction log or mastery mechanics as "being watched" — the numbers surface through maps and reviews.
+### 【Answer formatting · preferences】
+Structure answers with markdown (headings, lists, GFM tables); for structures prefer visuals: concept maps and flow diagrams as mermaid code blocks, comparisons as GFM tables, code walkthroughs as fenced code with line-referenced annotations. These preferences are subordinate to the two sections above.
 `;
 /** The three builtin souls, verbatim from LookatStudy (direct/guide/practice). */
 const SOULS = {

package/lib/{pptx-parser-CD5pR2cj.mjs → pptx-parser-B9IDkiGM.mjs}

@@ -21,6 +21,49 @@ function shapeTexts(xml) {
 function slideNumber(name) {
 	return Number(name.match(/slide(\d+)\.xml$/i)?.[1] ?? 0);
 }
+/**
+* One `a:tbl` block → GFM markdown table, or "" when the table is an
+* all-empty placeholder. Cells: text via a:p runs; gridSpan advances the
+* column index (spanned columns padded); vMerge continuation cells carry no
+* text of their own. Pipes escaped, inner whitespace collapsed.
+*/
+function tableToMarkdown(tblXml) {
+	const rows = [];
+	const trRe = /<a:tr\b[^>]*>([\s\S]*?)<\/a:tr>/g;
+	let trm;
+	while ((trm = trRe.exec(tblXml)) !== null) {
+		const cells = [];
+		const tcRe = /<a:tc\b([^>]*)>([\s\S]*?)<\/a:tc>/g;
+		let col = 0;
+		let tcm;
+		while ((tcm = tcRe.exec(trm[1])) !== null) {
+			const attrs = tcm[1] ?? "";
+			const span = Math.max(1, Number(attrs.match(/gridSpan="(\d+)"/)?.[1] ?? 1));
+			const vMerge = /vMerge="1"/.test(attrs);
+			const raw = shapeTexts(tcm[2] ?? "").join(" ").trim().replace(/\|/g, "\\|").replace(/\s+/g, " ");
+			if (!vMerge) cells.push({
+				col,
+				text: raw
+			});
+			col += span;
+		}
+		rows.push(cells);
+	}
+	if (rows.filter((cs) => cs.length > 0).length === 0) return "";
+	if (rows.every((cs) => cs.every((c) => !c.text))) return "";
+	const sorted = rows.filter((cs) => cs.length > 0).map((cs) => cs.sort((a, b) => a.col - b.col));
+	const width = Math.max(...sorted.map((cs) => cs[cs.length - 1].col + 1));
+	const line = (cs) => {
+		const texts = Array.from({ length: width }, () => "");
+		for (const c of cs) texts[c.col] = c.text;
+		return `| ${texts.join(" | ")} |`;
+	};
+	return [
+		line(sorted[0]),
+		`| ${Array.from({ length: width }, () => "---").join(" | ")} |`,
+		...sorted.slice(1).map(line)
+	].join("\n");
+}
 /** .pptx → markdown(每张 slide 一个 ##)。失败抛错,调用方按"无内容"兜底。 */
 function parsePptx(buf) {
 	const entries = readZip(buf);
@@ -30,9 +73,14 @@ function parsePptx(buf) {
 	for (const name of slideNames) {
 		const xml = readZipText(entries, name);
 		const no = slideNumber(name);
-		const texts = shapeTexts(xml);
+		const tables = [];
+		const texts = shapeTexts(xml.replace(/<a:tbl\b[\s\S]*?<\/a:tbl>/g, (tbl) => {
+			const md = tableToMarkdown(tbl);
+			if (md) tables.push(md);
+			return "";
+		}));
 		const title = texts[0] ?? "";
-		const body = texts.slice(1);
+		const body = [...texts.slice(1), ...tables];
 		lines.push(`\n## Slide ${no}: ${title || "(无标题)"}\n`);
 		if (body.length) lines.push(body.join("\n\n"));
 		const notes = shapeTexts(readZipText(entries, `ppt/notesSlides/notesSlide${no}.xml`));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-plugin-lookatstudy",
-  "version": "0.11.1",
+  "version": "0.12.0",
   "packageManager": "pnpm@11.7.0",
   "description": "Turn any markdown, local folder, or GitHub learning repo into a guided course inside DeepSeek Harness: gated skill-tree progression, BKT mastery tracking, SM-2 spaced repetition.",
   "type": "module",