@aglyn/plugins-ai 1.0.0-beta.288 → 1.0.0-beta.290

package/src/lib/site-import/site-import-html.js

@@ -32,15 +32,25 @@
  *
  * The output is DATA. Text from a page reaches a model only fenced as
  * untrusted (`fenceUntrustedText`), never as instructions.
+ *
+ * What a page is made of is kept as the page shows it (AGL-4083): words that
+ * sit in tags side by side stay apart ("Sign in" and "Start free", never
+ * "Sign inStart free"); a link is a block of its own, `mailto:` and `tel:`
+ * ones too, so an address or a phone number survives; a menu entry is named
+ * by its first line, not by its mega-menu description; a `<picture>`'s
+ * source stands in for an `<img>` with no address; and the head's share card,
+ * icons, site name and Organization structured data come along for the
+ * header, the footer and the site's own facts.
  *=========================================*/ /** One run of a page's content, in the order the page draws it. */ import { _ as _extends } from "@swc/helpers/_/_extends";
 /** Ceilings that keep one odd page from becoming the whole inventory. */ export const SITE_IMPORT_PAGE_LIMITS = {
-    blocks: 400,
-    blockChars: 2000,
+    /** A pricing page's plan table runs past 400 (AGL-4083); a wave's read keeps fewer, under its own budget. */ blocks: 2000,
+    /** A legal page's paragraph runs past 2,000 characters. */ blockChars: 10000,
     images: 100,
     links: 300,
-    navLinks: 60,
+    /** A mega-menu's entries and the footer's columns together. */ navLinks: 150,
     forms: 10,
-    fieldsPerForm: 30
+    fieldsPerForm: 30,
+    icons: 12
 };
 /** Elements removed with everything inside them. */ const REMOVED_WITH_CONTENT = [
     'script',
@@ -105,6 +115,52 @@ const BLOCK_TAGS = {
     'br',
     'body'
 ]);
+/**
+ * Tags a browser lays out apart from the words beside them (AGL-4083): two
+ * links, or a link and a picture, side by side are two runs of words, never
+ * one ("Sign in" and "Start free", not "Sign inStart free"). Every other tag
+ * inside a block styles words where they are (a `<span>` or a `<strong>`
+ * mid-sentence) and adds no space; block-level tags already end their block.
+ */ const RUN_BOUNDARY = new Set([
+    'a',
+    'img',
+    'button',
+    'input',
+    'select',
+    'textarea',
+    'picture',
+    'video',
+    'audio',
+    'iframe',
+    'object',
+    'embed',
+    'canvas',
+    'svg',
+    'meter',
+    'progress',
+    'output'
+]);
+/** Where one run of words ends and the next begins, resolved when a block is written. */ const RUN_BREAK = '\u0001';
+/**
+ * The runs of a block joined as a reader sees them: a space between two
+ * runs, except before closing punctuation or after an opening one, where
+ * the page wrote none ("<span>Austin</span>, Texas" is "Austin, Texas").
+ */ const RUN_BREAKS = new RegExp(`[\\s${RUN_BREAK}]*${RUN_BREAK}[\\s${RUN_BREAK}]*`, 'g');
+function joinRuns(text) {
+    if (!text.includes(RUN_BREAK)) return text;
+    return text.replace(RUN_BREAKS, (match, offset, whole)=>{
+        var _whole_, _whole_1, _whole_2, _whole_3;
+        if (/\s/.test(match)) return ' ';
+        const before = (_whole_ = whole[offset - 1]) != null ? _whole_ : '';
+        const after = (_whole_1 = whole[offset + match.length]) != null ? _whole_1 : '';
+        if (!before || !after) return '';
+        if (/[.,;:!?%)\]}\u2019\u201d\u00bb]/.test(after) || /[([{\u2018\u201c\u00ab]/.test(before)) return '';
+        // A straight quote opens after a space or a bracket, and closes before one.
+        if (/["']/.test(before) && /^$|[\s([{]/.test((_whole_2 = whole[offset - 2]) != null ? _whole_2 : '')) return '';
+        if (/["']/.test(after) && /^$|[\s.,;:!?)\]}]/.test((_whole_3 = whole[offset + match.length + 1]) != null ? _whole_3 : '')) return '';
+        return ' ';
+    });
+}
 const NAMED_ENTITIES = {
     amp: '&',
     lt: '<',
@@ -176,7 +232,7 @@ const TAG = /<(\/?)([a-zA-Z][\w:-]*)((?:[^>"']|"[^"]*"|'[^']*')*)>/g;
     }).filter(Boolean);
 }
 function collapse(text) {
-    return text.replace(/\s+/g, ' ').trim();
+    return joinRuns(text).replace(/\s+/g, ' ').trim();
 }
 function clip(text, max) {
     return text.length > max ? `${text.slice(0, max - 1).trimEnd()}…` : text;
@@ -203,9 +259,12 @@ function removeElements(html, tags) {
         if (kept.length < SITE_IMPORT_PAGE_LIMITS.forms) {
             const fields = [];
             for (const match of form.matchAll(/<(input|textarea|select)\b((?:[^>"']|"[^"]*"|'[^']*')*)>/gi)){
+                var _attributes_ariahidden;
                 const attributes = parseAttributes(match[2]);
                 const type = match[1].toLowerCase() === 'input' ? (attributes['type'] || 'text').toLowerCase() : match[1].toLowerCase();
                 if (type === 'hidden' || type === 'submit' || type === 'button' || type === 'image') continue;
+                // A field no visitor sees is a trap for bots or a widget's shadow, not a question (AGL-4083).
+                if (((_attributes_ariahidden = attributes['aria-hidden']) != null ? _attributes_ariahidden : '').toLowerCase() === 'true' || attributes['tabindex'] === '-1') continue;
                 fields.push({
                     type,
                     name: clip(attributes['name'] || attributes['id'] || '', 80)
@@ -241,19 +300,154 @@ function removeElements(html, tags) {
     const first = (_String_split_ = String(srcset != null ? srcset : '').split(',')[0]) == null ? void 0 : _String_split_.trim().split(/\s+/)[0];
     return first || undefined;
 }
+/**
+ * Where a link a page shows goes (AGL-4083): an absolute http(s) address
+ * without its fragment, a `mailto:` with an address in it, or a `tel:` with
+ * a number in it; `null` for anything else (a script, a jump on the page).
+ */ export function resolveSiteLink(raw, base) {
+    var _ref;
+    var _resolveSiteUrl;
+    const value = String(raw != null ? raw : '').trim();
+    const mail = /^mailto:([^?\s<>"']+@[^?\s<>"']+)/i.exec(value);
+    if (mail) return `mailto:${decodeURIComponent(mail[1]).toLowerCase()}`;
+    const phone = /^tel:([+\d][\d\s().-]{2,30})$/i.exec(value);
+    if (phone) return `tel:${phone[1].replace(/[\s().-]/g, '')}`;
+    if (value.startsWith('#')) return null;
+    return (_ref = (_resolveSiteUrl = resolveSiteUrl(value, base)) == null ? void 0 : _resolveSiteUrl.toString()) != null ? _ref : null;
+}
+/**
+ * A `<picture>` whose `<img>` has no address of its own (a placeholder, or
+ * none at all) is read from its first `<source>` (AGL-4083); a `<picture>`
+ * whose `<img>` has one keeps it. The sources go with the void elements.
+ */ function readPictures(html) {
+    return html.replace(/<picture\b(?:[^>"']|"[^"]*"|'[^']*')*>([\s\S]*?)<\/picture\s*>/gi, (whole, inner)=>{
+        var _attributes_alt;
+        const img = /<img\b((?:[^>"']|"[^"]*"|'[^']*')*)>/i.exec(inner);
+        const attributes = img ? parseAttributes(img[1]) : {};
+        const own = attributes['src'] || attributes['data-src'] || firstSrcset(attributes['srcset'] || attributes['data-srcset']);
+        if (own && !/^data:/i.test(own.trim())) return whole;
+        let source;
+        for (const match of inner.matchAll(/<source\b((?:[^>"']|"[^"]*"|'[^']*')*)>/gi)){
+            const sourceAttributes = parseAttributes(match[1]);
+            source = firstSrcset(sourceAttributes['srcset'] || sourceAttributes['data-srcset']);
+            if (source) break;
+        }
+        if (!source) return whole;
+        const quote = (value)=>value.replace(/&/g, '&amp;').replace(/"/g, '&quot;').replace(/</g, '&lt;');
+        return ` <img src="${quote(source)}" alt="${quote((_attributes_alt = attributes['alt']) != null ? _attributes_alt : '')}"> `;
+    });
+}
+/** The head's icons, the largest first within each kind. */ function readIcons(html, pageUrl) {
+    const icons = [];
+    for (const match of html.matchAll(/<link\b((?:[^>"']|"[^"]*"|'[^']*')*)>/gi)){
+        var _attributes_rel, _match, _attributes_sizes;
+        var _attributes_type;
+        const attributes = parseAttributes(match[1]);
+        const rels = ((_attributes_rel = attributes['rel']) != null ? _attributes_rel : '').toLowerCase().split(/\s+/);
+        const rel = rels.includes('apple-touch-icon') || rels.includes('apple-touch-icon-precomposed') ? 'apple-touch-icon' : rels.includes('icon') ? 'icon' : null;
+        if (!rel) continue;
+        const url = resolveSiteUrl(attributes['href'], pageUrl);
+        if (!url) continue;
+        const sizes = [
+            ...(_match = ((_attributes_sizes = attributes['sizes']) != null ? _attributes_sizes : '').match(/(\d+)\s*x\s*(\d+)/gi)) != null ? _match : []
+        ];
+        const size = sizes.reduce((most, entry)=>Math.max(most, ...entry.split(/x/i).map((side)=>Number(side) || 0)), 0);
+        icons.push({
+            url: url.toString(),
+            rel,
+            size: size || null,
+            type: ((_attributes_type = attributes['type']) == null ? void 0 : _attributes_type.trim()) || null
+        });
+        if (icons.length >= SITE_IMPORT_PAGE_LIMITS.icons) break;
+    }
+    return icons.sort((a, b)=>{
+        var _b_size, _a_size;
+        return a.rel === b.rel ? ((_b_size = b.size) != null ? _b_size : 0) - ((_a_size = a.size) != null ? _a_size : 0) : a.rel === 'apple-touch-icon' ? -1 : 1;
+    });
+}
+const ORGANIZATION_TYPE = /(?:^|\b)(?:Organization|Corporation|LocalBusiness|NGO|EducationalOrganization|MedicalOrganization|SportsOrganization|Store|Restaurant|ProfessionalService|HomeAndConstructionBusiness|HealthAndBeautyBusiness|FoodEstablishment|AutomotiveBusiness|LegalService|FinancialService|Dentist|Physician|LodgingBusiness|EntertainmentBusiness)$/;
+/**
+ * The first Organization (or kind of local business) the page's structured
+ * data describes, read before scripts are thrown away. Malformed data, or
+ * none, is `null`; nothing in it is ever run.
+ */ function readOrganization(html) {
+    var _node_contactPoint, _node_sameAs;
+    const nodes = [];
+    const visit = (value, depth)=>{
+        if (depth > 4 || nodes.length > 50 || !value || typeof value !== 'object') return;
+        if (Array.isArray(value)) {
+            for (const entry of value)visit(entry, depth + 1);
+            return;
+        }
+        const node = value;
+        nodes.push(node);
+        if (node['@graph']) visit(node['@graph'], depth + 1);
+    };
+    for (const match of html.matchAll(/<script\b[^>]*\btype\s*=\s*["']?application\/ld\+json["']?[^>]*>([\s\S]*?)<\/script\s*>/gi)){
+        try {
+            visit(JSON.parse(match[1]), 0);
+        } catch (unused) {
+        // Structured data that does not parse says nothing.
+        }
+    }
+    const typeOf = (node)=>{
+        var _node_type;
+        return [].concat((_node_type = node['@type']) != null ? _node_type : []).map(String);
+    };
+    const node = nodes.find((entry)=>typeOf(entry).some((type)=>ORGANIZATION_TYPE.test(type)));
+    if (!node) return null;
+    const text = (value, max = 300)=>typeof value === 'string' ? clip(collapse(value), max) : '';
+    const urlOf = (value)=>{
+        const raw = typeof value === 'string' ? value : value && typeof value === 'object' ? value['url'] : '';
+        return typeof raw === 'string' && /^https?:\/\//i.test(raw.trim()) ? raw.trim() : '';
+    };
+    const contact = [].concat((_node_contactPoint = node['contactPoint']) != null ? _node_contactPoint : []).find((entry)=>entry && typeof entry === 'object');
+    const address = (()=>{
+        const value = node['address'];
+        if (typeof value === 'string') return text(value);
+        if (!value || typeof value !== 'object') return '';
+        const parts = value;
+        const region = [
+            text(parts['addressRegion']),
+            text(parts['postalCode'])
+        ].filter(Boolean).join(' ');
+        const country = parts['addressCountry'];
+        return [
+            text(parts['streetAddress']),
+            text(parts['addressLocality']),
+            region,
+            typeof country === 'string' ? text(country) : text(country == null ? void 0 : country['name'])
+        ].filter(Boolean).join(', ');
+    })();
+    return {
+        name: text(node['name']),
+        description: text(node['description'], 1000),
+        url: urlOf(node['url']),
+        logo: urlOf(node['logo']),
+        telephone: text(node['telephone']) || text(contact == null ? void 0 : contact['telephone']),
+        email: (text(node['email']) || text(contact == null ? void 0 : contact['email'])).replace(/^mailto:/i, ''),
+        address,
+        sameAs: [].concat((_node_sameAs = node['sameAs']) != null ? _node_sameAs : []).map((entry)=>urlOf(entry)).filter(Boolean).slice(0, 20)
+    };
+}
 /**
  * The words, structure, pictures and links of one page.
  *
  * `isSameSite(url)` says whether an address is on the site being imported;
  * only those become `links`, and images are marked by it.
  */ export function extractSiteImportPage(html, pageUrl, isSameSite) {
-    var _ref;
-    var _parseAttributes_lang;
+    var _ref, _ref1;
+    var _resolveSiteUrl, _parseAttributes_lang;
     const empty = {
         title: '',
         description: '',
         lang: null,
         canonical: null,
+        ogImage: null,
+        siteName: '',
+        icons: [],
+        brand: null,
+        organization: null,
         blocks: [],
         images: [],
         navLinks: [],
@@ -268,7 +462,10 @@ function removeElements(html, tags) {
     if (PASSWORD_INPUT.test(source)) return _extends({}, empty, {
         loginForm: true
     });
-    source = removeElements(source, REMOVED_WITH_CONTENT).replace(REMOVED_VOID, ' ');
+    // Read before scripts go: the structured data is a script's body.
+    const organization = readOrganization(source);
+    const icons = readIcons(source, pageUrl);
+    source = removeElements(readPictures(source), REMOVED_WITH_CONTENT).replace(REMOVED_VOID, ' ');
     // Forms next, so a field list is read before the controls go.
     const forms = takeForms(source);
     source = removeElements(forms.html, FORM_CONTROLS);
@@ -284,6 +481,8 @@ function removeElements(html, tags) {
     const titleMatch = /<title\b[^>]*>([\s\S]*?)<\/title\s*>/i.exec(source);
     const title = clip(collapse(decodeHtmlEntities((_ref = titleMatch == null ? void 0 : titleMatch[1]) != null ? _ref : '')) || metaContent('og:title', 'property'), 300);
     const description = clip(metaContent('description', 'name') || metaContent('og:description', 'property'), 1000);
+    const ogImage = (_ref1 = (_resolveSiteUrl = resolveSiteUrl(metaContent('og:image', 'property') || metaContent('og:image:url', 'property') || metaContent('twitter:image', 'name'), pageUrl)) == null ? void 0 : _resolveSiteUrl.toString()) != null ? _ref1 : null;
+    const siteName = clip(metaContent('og:site_name', 'property') || metaContent('application-name', 'name'), 120);
     const htmlTag = /<html\b((?:[^>"']|"[^"]*"|'[^']*')*)>/i.exec(source);
     const lang = htmlTag ? ((_parseAttributes_lang = parseAttributes(htmlTag[1])['lang']) == null ? void 0 : _parseAttributes_lang.trim()) || null : null;
     let canonical = null;
@@ -291,9 +490,9 @@ function removeElements(html, tags) {
         var _attributes_rel;
         const attributes = parseAttributes(match[1]);
         if (((_attributes_rel = attributes['rel']) != null ? _attributes_rel : '').toLowerCase().split(/\s+/).includes('canonical')) {
-            var _ref1;
-            var _resolveSiteUrl;
-            canonical = (_ref1 = (_resolveSiteUrl = resolveSiteUrl(attributes['href'], pageUrl)) == null ? void 0 : _resolveSiteUrl.toString()) != null ? _ref1 : null;
+            var _ref2;
+            var _resolveSiteUrl1;
+            canonical = (_ref2 = (_resolveSiteUrl1 = resolveSiteUrl(attributes['href'], pageUrl)) == null ? void 0 : _resolveSiteUrl1.toString()) != null ? _ref2 : null;
             break;
         }
     }
@@ -306,47 +505,90 @@ function removeElements(html, tags) {
     const navLinks = [];
     const links = [];
     const linkSeen = new Set();
+    let brand = null;
     let navDepth = 0;
     let headerDepth = 0;
     let footerDepth = 0;
     let current = null;
     let buffer = '';
-    let anchor = null;
+    /** The link being read: its menu address (http(s) only), where it goes, its words and its first picture. */ let anchor = null;
+    /** The links read into the block being built, written after it. */ let pending = [];
     const region = ()=>footerDepth > 0 ? 'footer' : headerDepth > 0 ? 'header' : 'main';
     // A site without a <nav> keeps its menu in the header; its links are the nav.
     const inNav = ()=>navDepth > 0 || !hasNav && headerDepth > 0;
+    const room = ()=>blocks.length < SITE_IMPORT_PAGE_LIMITS.blocks;
     const flush = ()=>{
         var _ref;
-        const text = clip(collapse(decodeHtmlEntities(buffer)), SITE_IMPORT_PAGE_LIMITS.blockChars);
+        const raw = buffer;
+        const written = pending;
         buffer = '';
+        pending = [];
+        if (anchor && collapse(decodeHtmlEntities(raw.slice(anchor.start)))) anchor.crossed = true;
+        if (anchor) anchor.start = 0;
         const kind = (_ref = current == null ? void 0 : current.kind) != null ? _ref : 'paragraph';
         const level = current == null ? void 0 : current.level;
+        const loose = current === null;
         current = null;
-        if (!text || navDepth > 0 || blocks.length >= SITE_IMPORT_PAGE_LIMITS.blocks) return;
-        const last = blocks[blocks.length - 1];
-        if (last && last.kind !== 'image' && last.text === text) return;
-        if (kind === 'heading') {
-            blocks.push({
-                kind,
-                level: level != null ? level : 2,
-                text,
-                region: region()
-            });
-        } else {
-            blocks.push({
-                kind,
-                text,
-                region: region()
-            });
+        if (navDepth > 0) return;
+        const text = clip(collapse(decodeHtmlEntities(raw)), SITE_IMPORT_PAGE_LIMITS.blockChars);
+        // Loose words that are only links (a row of buttons, an address in a
+        // card) are those links, each a block of its own; words around a link
+        // keep their block, and the link follows it.
+        let rest = raw;
+        for (const link of [
+            ...written
+        ].sort((a, b)=>b.start - a.start))rest = `${rest.slice(0, link.start)}${rest.slice(link.end)}`;
+        const onlyLinks = loose && written.length > 0 && !collapse(decodeHtmlEntities(rest));
+        if (text && !onlyLinks && room()) {
+            const last = blocks[blocks.length - 1];
+            if (!(last && last.kind !== 'image' && last.kind !== 'link' && last.text === text)) {
+                if (kind === 'heading') {
+                    blocks.push({
+                        kind,
+                        level: level != null ? level : 2,
+                        text,
+                        region: region()
+                    });
+                } else {
+                    blocks.push({
+                        kind,
+                        text,
+                        region: region()
+                    });
+                }
+            }
+        }
+        for (const link of written){
+            if (!room()) break;
+            const last = blocks[blocks.length - 1];
+            if ((last == null ? void 0 : last.kind) === 'link' && last.href === link.href && last.text === link.text) continue;
+            const inline = !onlyLinks || link.crossed;
+            blocks.push(_extends({
+                kind: 'link',
+                text: link.text,
+                href: link.href,
+                region: link.region
+            }, inline ? {
+                inline: true
+            } : {}));
         }
     };
+    /** A boundary between two runs of words: in the block, and in the link being read. */ const boundary = (block)=>{
+        buffer += RUN_BREAK;
+        if (!anchor) return;
+        anchor.text += RUN_BREAK;
+        if (block && anchor.lines[anchor.lines.length - 1] !== '') anchor.lines.push('');
+    };
     for (const match of body.matchAll(new RegExp(`${TAG.source}|([^<]+)|<`, 'g'))){
         var _match_;
         const text = match[4];
         if (text !== undefined || match[2] === undefined) {
             const run = text != null ? text : '<';
             buffer += run;
-            if (anchor) anchor.text += run;
+            if (anchor) {
+                anchor.text += run;
+                anchor.lines[anchor.lines.length - 1] += run;
+            }
             continue;
         }
         const closing = match[1] === '/';
@@ -355,22 +597,51 @@ function removeElements(html, tags) {
         if (tag === 'a') {
             if (closing) {
                 if (anchor) {
-                    const label = clip(collapse(decodeHtmlEntities(anchor.text)), 120);
-                    if (inNav() && label && navLinks.length < SITE_IMPORT_PAGE_LIMITS.navLinks) {
+                    var _anchor_lines_find;
+                    const label = clip(collapse(decodeHtmlEntities(anchor.text)), 300);
+                    const firstLine = clip(collapse(decodeHtmlEntities((_anchor_lines_find = anchor.lines.find((line)=>collapse(decodeHtmlEntities(line)))) != null ? _anchor_lines_find : '')), 120);
+                    if (inNav() && anchor.url && firstLine && navLinks.length < SITE_IMPORT_PAGE_LIMITS.navLinks) {
                         navLinks.push({
-                            text: label,
+                            text: firstLine,
                             url: anchor.url.toString()
                         });
                     }
+                    if (!brand && anchor.region !== 'footer' && (headerDepth > 0 || navDepth > 0) && anchor.url && isSameSite(anchor.url) && anchor.url.pathname === '/') {
+                        brand = {
+                            logo: anchor.image,
+                            text: label,
+                            region: anchor.region
+                        };
+                    }
+                    if (!inNav() && anchor.href && label) {
+                        pending.push({
+                            text: label,
+                            href: anchor.href,
+                            start: anchor.start,
+                            end: buffer.length,
+                            crossed: anchor.crossed,
+                            region: anchor.region
+                        });
+                    }
                     anchor = null;
                 }
+                boundary(false);
                 continue;
             }
+            boundary(false);
             const url = resolveSiteUrl(attributes['href'], pageUrl);
-            anchor = url ? {
+            anchor = {
                 url,
-                text: ''
-            } : null;
+                href: resolveSiteLink(attributes['href'], pageUrl),
+                text: '',
+                lines: [
+                    ''
+                ],
+                image: null,
+                start: buffer.length,
+                crossed: false,
+                region: region()
+            };
             if (url && isSameSite(url) && links.length < SITE_IMPORT_PAGE_LIMITS.links) {
                 const key = url.toString();
                 if (!linkSeen.has(key)) {
@@ -382,27 +653,31 @@ function removeElements(html, tags) {
         }
         if (tag === 'img' && !closing) {
             const url = resolveSiteUrl(attributes['src'] || attributes['data-src'] || firstSrcset(attributes['srcset'] || attributes['data-srcset']), pageUrl);
+            boundary(false);
             if (url && images.length < SITE_IMPORT_PAGE_LIMITS.images) {
                 var _attributes_alt;
                 const key = url.toString();
                 const alt = clip(collapse((_attributes_alt = attributes['alt']) != null ? _attributes_alt : ''), 300);
+                if (anchor && !anchor.image) anchor.image = {
+                    url: key,
+                    alt
+                };
                 if (!imageSeen.has(key)) {
                     imageSeen.add(key);
                     images.push({
                         url: key,
                         alt,
-                        sameSite: isSameSite(url)
+                        sameSite: isSameSite(url),
+                        region: region()
                     });
                     if (navDepth === 0) {
                         flush();
-                        if (blocks.length < SITE_IMPORT_PAGE_LIMITS.blocks) {
-                            blocks.push({
-                                kind: 'image',
-                                url: key,
-                                alt,
-                                region: region()
-                            });
-                        }
+                        if (room()) blocks.push({
+                            kind: 'image',
+                            url: key,
+                            alt,
+                            region: region()
+                        });
                     }
                 }
             }
@@ -410,6 +685,7 @@ function removeElements(html, tags) {
         }
         if (tag === 'nav' || tag === 'header' || tag === 'footer') {
             flush();
+            boundary(true);
             const step = closing ? -1 : 1;
             if (tag === 'nav') navDepth = Math.max(0, navDepth + step);
             if (tag === 'header') headerDepth = Math.max(0, headerDepth + step);
@@ -420,9 +696,11 @@ function removeElements(html, tags) {
         if (heading || BLOCK_TAGS[tag]) {
             if (closing) {
                 if ((current == null ? void 0 : current.tag) === tag) flush();
+                boundary(true);
                 continue;
             }
             flush();
+            boundary(true);
             current = heading ? {
                 kind: 'heading',
                 level: Number(heading[1]),
@@ -436,10 +714,14 @@ function removeElements(html, tags) {
         if (BREAK_TAGS.has(tag)) {
             if (tag === 'br' && current) {
                 buffer += ' ';
+                if (anchor) anchor.text += ' ';
                 continue;
             }
             flush();
+            boundary(true);
+            continue;
         }
+        if (RUN_BOUNDARY.has(tag)) boundary(false);
     }
     flush();
     return {
@@ -447,6 +729,11 @@ function removeElements(html, tags) {
         description,
         lang,
         canonical,
+        ogImage,
+        siteName,
+        icons,
+        brand,
+        organization,
         blocks,
         images,
         navLinks,