Strip active content from extracted HTML before PDF print

Defuddle-extracted article content is injected raw into the PDF HTML
and rendered by a real browser (browserless). As defense-in-depth
against a compromised or malicious source page, remove script/style/
iframe/object/embed elements and on* event-handler / javascript: href
attributes from the parsed DOM before serializing it into the template.

src/lib/clean-pdf/__tests__/template.test.ts

@@ -42,6 +42,14 @@ describe("buildCleanHtml", () => {
     expect(new Set(ids).size).toBe(3);
   });
 
+  it("вырезает script и inline-обработчики из контента", () => {
+    const html = buildCleanHtml({ ...base, contentHtml: `<p onclick="alert(1)">t</p><script>alert(2)</script><a href="javascript:alert(3)">x</a>` });
+    expect(html).not.toContain("<script>alert(2)");
+    expect(html).not.toContain("onclick");
+    expect(html).not.toContain("javascript:alert(3)");
+    expect(html).toContain(">t<");
+  });
+
   it("переключает тёмную тему", () => {
     const light = buildCleanHtml({ ...base, contentHtml: "<p>x</p>" });
     const dark = buildCleanHtml({ ...base, theme: "dark", contentHtml: "<p>x</p>" });

src/lib/clean-pdf/template.ts

@@ -42,6 +42,17 @@ function prepareContent(contentHtml: string): { html: string; toc: TocEntry[] }
     toc.push({ id, text, level: h.tagName === "H2" ? 2 : 3 });
   });
 
+  // Defuddle отдаёт очищенный, но не полностью доверенный HTML исходной
+  // страницы — вырезаем активное содержимое перед печатью в реальном браузере.
+  doc.querySelectorAll("script, style, iframe, object, embed").forEach((el) => el.remove());
+  doc.querySelectorAll("*").forEach((el) => {
+    for (const attr of [...el.attributes]) {
+      if (attr.name.startsWith("on") || (attr.name === "href" && attr.value.trim().toLowerCase().startsWith("javascript:"))) {
+        el.removeAttribute(attr.name);
+      }
+    }
+  });
+
   return { html: doc.body.innerHTML, toc };
 }