A browser extension that lets you summarize any webpage and ask questions using AI.
fix(extraction): skip embeds and strip literal iframe/embed markup from text
Exclude object/embed/frame subtrees like iframe; sanitize CMS embed snippets in
Readability and legacy pipelines.
+44
-4
+44
-4
scripts/content.js
+44
-4
scripts/content.js
···
78
78
"style",
79
79
"noscript",
80
80
"iframe",
81
+
"embed",
82
+
"object",
83
+
"frame",
81
84
"nav",
82
85
"aside",
83
86
"form",
···
85
88
"input",
86
89
];
87
90
91
+
/**
92
+
* Subtrees to omit from plain-text collection. `textContent` includes script/style/template
93
+
* bodies, which pulls in Astro/React/Vite inline bundles when walking `main div` etc.
94
+
* Also skip embed-like tags (iframe/object/embed) so news players do not dump URLs or attrs.
95
+
*/
96
+
const TEXT_SUBTREE_EXCLUDE_TAGS = new Set([
97
+
...EXCLUDE_TAGS,
98
+
"template",
99
+
]);
100
+
101
+
function shouldExcludeTextSubtree(tagName) {
102
+
return TEXT_SUBTREE_EXCLUDE_TAGS.has(tagName.toLowerCase());
103
+
}
104
+
105
+
/**
106
+
* Strip embed markup that appears as literal text in article bodies (CMS/oEmbed fallbacks).
107
+
* Complements subtree skipping — some sites still surface tags as visible copy.
108
+
*/
109
+
function sanitizeLiteralEmbedMarkup(text) {
110
+
if (!text || typeof text !== "string") {
111
+
return text;
112
+
}
113
+
let t = text;
114
+
t = t.replace(/<iframe\b[\s\S]{0,20000}?<\/iframe>/gi, "\n");
115
+
t = t.replace(/<iframe\b[^>]{0,8000}\/?>/gi, "\n");
116
+
t = t.replace(/<\/iframe>/gi, "");
117
+
t = t.replace(/<embed\b[^>]{0,8000}\/?>/gi, "\n");
118
+
t = t.replace(/<object\b[\s\S]{0,20000}?<\/object>/gi, "\n");
119
+
return t;
120
+
}
121
+
88
122
function extractWithReadability() {
89
123
const documentClone = document.cloneNode(true);
90
124
const reader = new Readability(documentClone);
···
126
160
}
127
161
128
162
let content = article.textContent || "";
163
+
content = sanitizeLiteralEmbedMarkup(content);
129
164
130
165
content = content
131
166
.replace(/[^\S\n]+/g, " ")
···
134
169
135
170
extractedText += content;
136
171
137
-
const bodyTextLen = (article.textContent || "").trim().length;
172
+
const bodyTextLen = content.trim().length;
138
173
139
174
let wasTruncated = false;
140
175
if (extractedText.length > MAX_LENGTH) {
···
177
212
try {
178
213
const elements = document.querySelectorAll(selector);
179
214
for (const el of elements) {
180
-
const content = el.textContent.trim();
215
+
const content = getTextContent(el).trim();
181
216
if (content.length < 20 || seen.has(content.substring(0, 100))) continue;
182
217
183
218
const style = window.getComputedStyle(el);
···
200
235
if (text.length < 500 && !wasTruncated) {
201
236
const allParagraphs = document.querySelectorAll("p");
202
237
for (const p of allParagraphs) {
203
-
const content = p.textContent.trim();
238
+
const content = getTextContent(p).trim();
204
239
if (content.length > 30 && !seen.has(content.substring(0, 100))) {
205
240
const style = window.getComputedStyle(p);
206
241
if (style.display === "none" || style.visibility === "hidden") continue;
···
404
439
} else if (node.nodeType === Node.ELEMENT_NODE) {
405
440
const tagName = node.tagName.toLowerCase();
406
441
442
+
if (shouldExcludeTextSubtree(tagName)) {
443
+
continue;
444
+
}
445
+
407
446
if (["br", "p", "div", "li"].includes(tagName)) {
408
447
text += " " + getTextContent(node) + " ";
409
448
} else {
···
416
455
}
417
456
418
457
function cleanExtractedText(text, shouldTruncate = true) {
419
-
let cleaned = text
458
+
let cleaned = sanitizeLiteralEmbedMarkup(text);
459
+
cleaned = cleaned
420
460
.replace(/[^\S\n]+/g, " ")
421
461
.replace(/\n{3,}/g, "\n\n")
422
462
.replace(/^\s+|\s+$/g, "");