feat(katalog): set aus einem link uebernehmen

Amazon, idealo, geizhals, Hersteller-Shops und Set-Datenbanken: Produkt- daten aus der Seite (schema.org, Open Graph, Amazon-Produktseite), sonst aus dem Link selbst, ergaenzt ueber Rebrickable (Lego) und Merlins Bricks (BlueBrixx, Cobi). Fuellt das Formular vor, das Bild wird beim Speichern lokal abgelegt. Abrufe nur an oeffentliche Adressen (safeFetch). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-10-25 16:03:00 +01:00
parent 498a5c6226
commit 95a28640e1
9 changed files with 1271 additions and 39 deletions
@@ -0,0 +1,380 @@
+import { isValidEan, normalizeEan } from './ean';
+
+/**
+ * Reads set data out of a pasted link and, where the page can be fetched, out
+ * of its structured data (schema.org JSON-LD, Open Graph).
+ *
+ * Pure: no network, no database. The server action in katalog/link-actions.ts
+ * does the fetching and decides what to trust.
+ */
+
+export type LinkHints = {
+  manufacturer?: string;
+  setNumber?: string;
+  name?: string;
+  ean?: string;
+  imageUrl?: string;
+  pieceCount?: number;
+  year?: number;
+};
+
+/** Brands we recognise in links and page data, with how we spell them. */
+const BRANDS: Array<{ name: string; aliases: string[] }> = [
+  { name: 'Lego', aliases: ['lego'] },
+  { name: 'Playmobil', aliases: ['playmobil'] },
+  { name: 'Cobi', aliases: ['cobi'] },
+  { name: 'BlueBrixx', aliases: ['bluebrixx', 'bluebrixxpro', 'bb'] },
+  { name: 'Mega Bloks', aliases: ['megabloks', 'megaconstrux', 'mega'] },
+  { name: 'Mould King', aliases: ['mouldking'] },
+  { name: 'Cada', aliases: ['cada'] },
+  { name: 'Sluban', aliases: ['sluban'] },
+  { name: 'Qman', aliases: ['qman', 'keeppley'] },
+  { name: 'Xingbao', aliases: ['xingbao'] },
+];
+
+function key(value: string): string {
+  return value.toLowerCase().replace(/[^a-z0-9]/g, '');
+}
+
+/** "LEGO®", "lego", "BlueBrixx Pro" -> "Lego", "BlueBrixx". */
+export function canonicalBrand(value: string | undefined | null): string | undefined {
+  if (!value) return undefined;
+  const wanted = key(value);
+  if (!wanted) return undefined;
+  for (const brand of BRANDS) {
+    if (brand.aliases.some((alias) => wanted === alias || (alias.length > 3 && wanted.startsWith(alias)))) {
+      return brand.name;
+    }
+  }
+  return undefined;
+}
+
+/** Where a link points, in words for the UI. */
+export function sourceLabel(url: URL): string {
+  const host = url.hostname.replace(/^www\./, '');
+  if (/(^|\.)amazon\./.test(host) || /^amzn\./.test(host)) return 'Amazon';
+  if (/(^|\.)idealo\./.test(host)) return 'idealo';
+  if (/(^|\.)geizhals\./.test(host)) return 'geizhals';
+  if (/(^|\.)brickmerge\./.test(host)) return 'brickmerge';
+  if (/(^|\.)brickfact\./.test(host)) return 'Brickfact';
+  if (/(^|\.)merlinsbricks\./.test(host)) return 'Merlins Bricks';
+  if (/(^|\.)ebay\./.test(host)) return 'eBay';
+  return host;
+}
+
+// Words that sit in shop slugs but never in a set name.
+const STOPWORDS = new Set([
+  'set', 'sets', 'bausatz', 'bausteine', 'klemmbausteine', 'baustein', 'spielzeug', 'spielset',
+  'kinder', 'jungen', 'maedchen', 'mädchen', 'ab', 'jahre', 'jahren', 'jahr', 'teile', 'teilig',
+  'geschenk', 'geschenkidee', 'the', 'and', 'for', 'kids', 'boys',
+  'girls', 'toy', 'toys', 'building', 'kit', 'bricks', 'brick', 'preisvergleich', 'dp', 'html',
+  'neu', 'ovp', 'new', 'sealed',
+]);
+
+const KEEP_LOWER = new Set(['und', 'mit', 'der', 'die', 'das', 'von', 'im', 'am', 'zum', 'zur']);
+
+function looksLikeYear(value: string): boolean {
+  const year = Number(value);
+  return year >= 1950 && year <= 2040;
+}
+
+/**
+ * A set number in a list of slug tokens: 4 to 7 digits, longer ones first,
+ * and four-digit years only as a last resort. Cobi's "2-30500" style is kept.
+ */
+function findSetNumber(tokens: string[]): string | undefined {
+  const joined = tokens.join('-');
+  const cobi = /(?:^|-)(\d-\d{5})(?:-|$)/.exec(joined);
+  if (cobi) return cobi[1];
+
+  const numbers = tokens.filter((token) => /^\d{4,7}$/.test(token));
+  const best =
+    numbers.find((token) => token.length >= 5) ?? numbers.find((token) => !looksLikeYear(token));
+  return best;
+}
+
+function titleCase(word: string): string {
+  if (KEEP_LOWER.has(word)) return word;
+  return word.charAt(0).toUpperCase() + word.slice(1);
+}
+
+/** The descriptive part of a shop URL, split into words. */
+function slugTokens(url: URL): string[] {
+  const segments = decodeURIComponent(url.pathname)
+    .split('/')
+    .filter(Boolean)
+    .map((segment) => segment.replace(/\.html?$/i, ''));
+
+  let slug = '';
+  const host = url.hostname;
+
+  if (/amazon\./.test(host)) {
+    // amazon.de/LEGO-60398-Familienhaus-Elektroauto/dp/B0...: the part before /dp/.
+    const dp = segments.findIndex((segment) => segment === 'dp' || segment === 'gp');
+    slug = dp > 0 ? segments[dp - 1]! : '';
+  } else if (/idealo\./.test(host)) {
+    // .../OffersOfProduct/202595377_-familienhaus-mit-elektroauto-60398-lego
+    slug = (segments.at(-1) ?? '').replace(/^\d+_-?/, '');
+  } else if (/geizhals\./.test(host)) {
+    // geizhals.de/lego-city-familienhaus-...-60398-a2875000
+    slug = (segments.at(-1) ?? '').replace(/-a\d+$/, '');
+  } else {
+    // Generic: the longest segment with dashes is usually the product slug.
+    slug = [...segments].sort((a, b) => b.split(/[-_]/).length - a.split(/[-_]/).length)[0] ?? '';
+  }
+
+  return slug
+    .toLowerCase()
+    .split(/[-_+\s]+/)
+    .map((token) => token.replace(/[^\p{L}\p{N}]/gu, ''))
+    .filter(Boolean);
+}
+
+/** Everything the address alone tells us - useful when the page blocks us. */
+export function hintsFromUrl(url: URL): LinkHints {
+  const host = url.hostname.replace(/^www\./, '');
+
+  // merlinsbricks.com/sets/bb-108782, /sets/cobi-2400: brand and number in one.
+  const merlin = /merlinsbricks\./.test(host)
+    ? /\/sets\/([a-z]+)-([\w-]+?)\/?$/i.exec(url.pathname)
+    : null;
+  if (merlin) {
+    return { manufacturer: canonicalBrand(merlin[1]), setNumber: merlin[2] };
+  }
+
+  const tokens = slugTokens(url);
+  const setNumber = findSetNumber(tokens);
+
+  let manufacturer: string | undefined;
+  for (const token of tokens) {
+    manufacturer = canonicalBrand(token);
+    if (manufacturer) break;
+  }
+  manufacturer ??= canonicalBrand(host.split('.').at(-2));
+
+  const words = tokens.filter(
+    (token) =>
+      token !== setNumber &&
+      !/^\d+$/.test(token) &&
+      !STOPWORDS.has(token) &&
+      !canonicalBrand(token) &&
+      token.length > 1,
+  );
+  const name = words.length > 0 ? words.slice(0, 8).map(titleCase).join(' ') : undefined;
+
+  return { manufacturer, setNumber, name };
+}
+
+// --- page data ------------------------------------------------------------
+
+const ENTITIES: Record<string, string> = {
+  amp: '&', lt: '<', gt: '>', quot: '"', apos: "'", nbsp: ' ', '#39': "'",
+  auml: 'ä', ouml: 'ö', uuml: 'ü', Auml: 'Ä', Ouml: 'Ö', Uuml: 'Ü', szlig: 'ß',
+};
+
+function decodeEntities(value: string): string {
+  return value
+    .replace(/&#(\d+);/g, (_, code: string) => String.fromCodePoint(Number(code)))
+    .replace(/&#x([\da-f]+);/gi, (_, code: string) => String.fromCodePoint(parseInt(code, 16)))
+    .replace(/&([a-z]+|#39);/gi, (match, name: string) => ENTITIES[name] ?? match);
+}
+
+function metaContent(html: string, property: string): string | undefined {
+  const escaped = property.replace(/[.:]/g, '\\$&');
+  const pattern = new RegExp(
+    `<meta[^>]+(?:property|name)=["']${escaped}["'][^>]*content=["']([^"']*)["']|<meta[^>]+content=["']([^"']*)["'][^>]*(?:property|name)=["']${escaped}["']`,
+    'i',
+  );
+  const match = pattern.exec(html);
+  const value = match?.[1] ?? match?.[2];
+  return value ? decodeEntities(value).trim() : undefined;
+}
+
+type JsonObject = Record<string, unknown>;
+
+function isObject(value: unknown): value is JsonObject {
+  return typeof value === 'object' && value !== null && !Array.isArray(value);
+}
+
+/** All schema.org Product objects on the page, however they are nested. */
+function jsonLdProducts(html: string): JsonObject[] {
+  const products: JsonObject[] = [];
+  const pattern = /<script[^>]+application\/ld\+json[^>]*>([\s\S]*?)<\/script>/gi;
+
+  const visit = (node: unknown) => {
+    if (Array.isArray(node)) return node.forEach(visit);
+    if (!isObject(node)) return;
+    const type = node['@type'];
+    const types = Array.isArray(type) ? type : [type];
+    if (types.includes('Product') || types.includes('ProductGroup')) products.push(node);
+    if (node['@graph']) visit(node['@graph']);
+  };
+
+  for (const match of html.matchAll(pattern)) {
+    try {
+      visit(JSON.parse(match[1]!.trim()));
+    } catch {
+      // A broken block on someone else's page is not our problem.
+    }
+  }
+  return products;
+}
+
+function text(value: unknown): string | undefined {
+  if (typeof value === 'string') return decodeEntities(value).trim() || undefined;
+  if (typeof value === 'number') return String(value);
+  if (isObject(value)) return text(value.name) ?? text(value.url);
+  if (Array.isArray(value)) return text(value[0]);
+  return undefined;
+}
+
+function cleanEan(value: unknown): string | undefined {
+  const raw = text(value);
+  if (!raw) return undefined;
+  const ean = normalizeEan(raw);
+  return isValidEan(ean) ? ean : undefined;
+}
+
+function absoluteUrl(value: string | undefined, base: URL): string | undefined {
+  if (!value) return undefined;
+  try {
+    const url = new URL(value, base);
+    return url.protocol === 'https:' || url.protocol === 'http:' ? url.toString() : undefined;
+  } catch {
+    return undefined;
+  }
+}
+
+/** "462 Teile", "450 pieces" in a description. */
+function piecesFrom(value: string | undefined): number | undefined {
+  const match = value && /(\d{1,3}(?:[.,]\d{3})*|\d+)\s*(?:teilen?|pieces|pcs|steinen?|bausteinen?|parts)\b/i.exec(value);
+  if (!match) return undefined;
+  const pieces = Number(match[1]!.replace(/[.,]/g, ''));
+  return pieces > 0 && pieces < 100_000 ? pieces : undefined;
+}
+
+/** "from 2025", "aus 2023", "Jahr: 2021" in a description. */
+function yearFrom(value: string | undefined): number | undefined {
+  const match = value && /\b(?:from|aus|jahr:?|released|erschienen)\s+((?:19|20)\d{2})\b/i.exec(value);
+  return match ? Number(match[1]) : undefined;
+}
+
+function stripTags(value: string): string {
+  return decodeEntities(value.replace(/<[^>]+>/g, ' '))
+    .replace(/[‎‏]/g, '')
+    .replace(/\s+/g, ' ')
+    .trim();
+}
+
+/**
+ * Amazon has no schema.org data, but a stable page skeleton: the title in
+ * #productTitle, the brand in the byline, pictures as "hiRes" in a script and
+ * the facts in the "Produktinformation" table or bullet list.
+ */
+function hintsFromAmazon(html: string): LinkHints {
+  const facts = new Map<string, string>();
+  for (const match of html.matchAll(/<th[^>]*>([\s\S]*?)<\/th>\s*<td[^>]*>([\s\S]*?)<\/td>/gi)) {
+    facts.set(stripTags(match[1]!).toLowerCase(), stripTags(match[2]!));
+  }
+  for (const match of html.matchAll(
+    /<span class="a-text-bold">([\s\S]*?)<\/span>\s*<span[^>]*>([\s\S]*?)<\/span>/gi,
+  )) {
+    facts.set(stripTags(match[1]!).replace(/[:\s]+$/, '').toLowerCase(), stripTags(match[2]!));
+  }
+  const fact = (...labels: string[]) => {
+    for (const [label, value] of facts) {
+      if (labels.some((wanted) => label.includes(wanted))) return value;
+    }
+    return undefined;
+  };
+
+  const title = /id="productTitle"[^>]*>([\s\S]*?)<\/span>/i.exec(html);
+  const byline = /id="bylineInfo"[^>]*>([\s\S]*?)<\/a>/i.exec(html);
+  const brandText = byline
+    ? stripTags(byline[1]!).replace(/^(marke|brand):\s*/i, '').replace(/^besuche den\s+|-store$|\s+store$/gi, '')
+    : fact('marke', 'hersteller');
+  const image = /"hiRes":"(https:[^"]+)"/.exec(html) ?? /data-old-hires="(https:[^"]+)"/.exec(html);
+  const pieces = Number((fact('anzahl der teile', 'teileanzahl', 'anzahl teile') ?? '').replace(/\D/g, ''));
+
+  return {
+    manufacturer: canonicalBrand(brandText),
+    setNumber: fact('modellnummer', 'herstellerreferenz')?.replace(/\s/g, ''),
+    name: title ? stripTags(title[1]!) : undefined,
+    ean: cleanEan(fact('ean', 'gtin')),
+    imageUrl: image?.[1],
+    pieceCount: pieces > 0 && pieces < 100_000 ? pieces : undefined,
+  };
+}
+
+export function hintsFromHtml(html: string, pageUrl: URL): LinkHints {
+  const hints = /(^|\.)amazon\./.test(pageUrl.hostname) ? hintsFromAmazon(html) : hintsFromStructuredData(html, pageUrl);
+
+  // A shop's own article number (an ASIN, a long SKU) is not a set number.
+  if (hints.setNumber && !/^[\d-]{3,12}$/.test(hints.setNumber)) delete hints.setNumber;
+
+  // Titles usually carry the number: "LEGO City 60398 Familienhaus".
+  if (!hints.setNumber && hints.name) {
+    const fromName = findSetNumber(hints.name.toLowerCase().split(/[\s()[\]:,/|]+/).filter(Boolean));
+    if (fromName) hints.setNumber = fromName;
+  }
+  return hints;
+}
+
+function hintsFromStructuredData(html: string, pageUrl: URL): LinkHints {
+  const product = jsonLdProducts(html)[0];
+  const description = text(product?.description) ?? metaContent(html, 'og:description');
+
+  const hints: LinkHints = {
+    manufacturer: canonicalBrand(text(product?.brand)) ?? canonicalBrand(metaContent(html, 'product:brand')),
+    setNumber: text(product?.mpn) ?? text(product?.sku),
+    name: text(product?.name) ?? metaContent(html, 'og:title'),
+    ean: cleanEan(product?.gtin13) ?? cleanEan(product?.gtin) ?? cleanEan(product?.gtin12) ?? cleanEan(product?.gtin8),
+    imageUrl: absoluteUrl(text(product?.image) ?? metaContent(html, 'og:image'), pageUrl),
+    pieceCount: piecesFrom(description),
+    year: yearFrom(description),
+  };
+  return hints;
+}
+
+/** First value wins per field: pass the most trusted source first. */
+export function mergeHints(...sources: Array<LinkHints | null | undefined>): LinkHints {
+  const merged: LinkHints = {};
+  for (const source of sources) {
+    if (!source) continue;
+    for (const [field, value] of Object.entries(source) as Array<[keyof LinkHints, never]>) {
+      if (merged[field] === undefined && value !== undefined && value !== '') merged[field] = value;
+    }
+  }
+  return merged;
+}
+
+/**
+ * Shop titles carry everything: "LEGO® City Familienhaus mit Elektroauto
+ * 60398", "BlueBrixx - Technisches Hilfswerk™ Vehicle | Set 108782". Keeps
+ * the part that is the set's own name.
+ */
+export function cleanSetName(name: string, manufacturer?: string, setNumber?: string): string {
+  let result = name.replace(/[®™©]/g, '');
+
+  // Site names and set references after a separator.
+  result = result.split(/\s+[|•·]\s+/)[0]!;
+
+  // Price-comparison titles: "... ab € 4,20 (2026)", "... ab 52,99 €".
+  result = result.replace(/\s+ab\s+(?:€\s*)?\d[\d.,]*\s*(?:€|EUR)?.*$/i, '');
+
+  if (setNumber) {
+    const escaped = setNumber.replace(/[-]/g, '\\-');
+    result = result.replace(new RegExp(`\\(?\\b${escaped}\\b\\)?`, 'g'), '');
+  }
+
+  if (manufacturer) {
+    for (const brand of BRANDS.filter((entry) => entry.name === manufacturer)) {
+      for (const alias of [brand.name, ...brand.aliases]) {
+        if (alias.length < 3) continue;
+        result = result.replace(new RegExp(`^\\s*${alias}\\b\\s*[-–:]?\\s*`, 'i'), '');
+      }
+    }
+  }
+
+  return result.replace(/\s{2,}/g, ' ').replace(/^[\s\-–:]+|[\s\-–:,]+$/g, '').trim();
+}
@@ -0,0 +1,66 @@
+import { BlockList, isIP } from 'node:net';
+
+/**
+ * Which addresses the server may fetch on a user's behalf.
+ *
+ * The link import downloads pages and pictures from URLs people paste. Without
+ * a guard, a link to http://192.168.1.1/ would make the server talk to the
+ * router - or to any other device in the home network. Only public unicast
+ * addresses pass. Pure, so it is unit tested.
+ */
+
+const blocked = new BlockList();
+
+for (const [network, prefix] of [
+  ['0.0.0.0', 8], // "this" network
+  ['10.0.0.0', 8], // private
+  ['100.64.0.0', 10], // carrier-grade NAT
+  ['127.0.0.0', 8], // loopback
+  ['169.254.0.0', 16], // link-local, cloud metadata
+  ['172.16.0.0', 12], // private
+  ['192.0.0.0', 24], // IETF protocol assignments
+  ['192.0.2.0', 24], // documentation
+  ['192.168.0.0', 16], // private
+  ['198.18.0.0', 15], // benchmarking
+  ['198.51.100.0', 24], // documentation
+  ['203.0.113.0', 24], // documentation
+  ['224.0.0.0', 4], // multicast
+  ['240.0.0.0', 4], // reserved, broadcast
+] as const) {
+  blocked.addSubnet(network, prefix, 'ipv4');
+}
+
+for (const [network, prefix] of [
+  ['::', 128], // unspecified
+  ['::1', 128], // loopback
+  ['64:ff9b::', 96], // NAT64 - would reach IPv4 space behind it
+  ['100::', 64], // discard
+  ['2001:db8::', 32], // documentation
+  ['fc00::', 7], // unique local
+  ['fe80::', 10], // link-local
+  ['ff00::', 8], // multicast
+] as const) {
+  blocked.addSubnet(network, prefix, 'ipv6');
+}
+
+export function isPublicAddress(address: string): boolean {
+  const version = isIP(address);
+  if (version === 0) return false;
+
+  if (version === 6) {
+    // An IPv4 address in IPv6 clothing is judged as what it really is.
+    const mapped = /^::ffff:(\d+\.\d+\.\d+\.\d+)$/i.exec(address);
+    if (mapped) return isPublicAddress(mapped[1]!);
+    return !blocked.check(address, 'ipv6');
+  }
+
+  return !blocked.check(address, 'ipv4');
+}
+
+/** Only plain web URLs on the standard ports; no credentials in the URL. */
+export function isAllowedWebUrl(url: URL): boolean {
+  if (url.protocol !== 'https:' && url.protocol !== 'http:') return false;
+  if (url.username || url.password) return false;
+  if (url.port && url.port !== '80' && url.port !== '443') return false;
+  return true;
+}
@@ -0,0 +1,167 @@
+import 'server-only';
+import { lookup as dnsLookup, type LookupAddress } from 'node:dns';
+import http from 'node:http';
+import https from 'node:https';
+import { isIP, type LookupFunction } from 'node:net';
+import { Readable } from 'node:stream';
+import zlib from 'node:zlib';
+import { isAllowedWebUrl, isPublicAddress } from './net-guard';
+
+/**
+ * Fetches a URL a user pasted - and nothing else.
+ *
+ * - Every hostname is resolved first and refused if any of its addresses is
+ *   not public (see net-guard.ts). The check happens inside the socket's DNS
+ *   lookup, so a second resolution cannot swap in a private address.
+ * - Redirects are followed by hand, and every hop passes the same checks.
+ * - Size and time are capped, compressed bodies included.
+ */
+
+export class SafeFetchError extends Error {}
+
+export type SafeResponse = {
+  status: number;
+  url: string;
+  contentType: string;
+  body: Buffer;
+};
+
+type Options = {
+  maxBytes?: number;
+  timeoutMs?: number;
+  accept?: string;
+};
+
+const MAX_REDIRECTS = 4;
+
+// A regular browser identity: some shops answer anything else with an error
+// page. It is one request per paste, the same a browser preview would make.
+const USER_AGENT =
+  'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/141.0 Safari/537.36';
+
+const guardedLookup = ((
+  hostname: string,
+  options: { all?: boolean; family?: number },
+  callback: (error: Error | null, address: string | LookupAddress[], family?: number) => void,
+) => {
+  dnsLookup(hostname, { family: options.family ?? 0, all: true }, (error, addresses) => {
+    if (error) return callback(error, '', 4);
+    if (addresses.length === 0 || addresses.some((entry) => !isPublicAddress(entry.address))) {
+      return callback(new SafeFetchError(`${hostname} zeigt auf eine interne Adresse.`), '', 4);
+    }
+    if (options.all) return callback(null, addresses);
+    callback(null, addresses[0]!.address, addresses[0]!.family);
+  });
+}) as unknown as LookupFunction;
+
+function checkUrl(url: URL) {
+  if (!isAllowedWebUrl(url)) throw new SafeFetchError('Nur normale http- und https-Links.');
+
+  // Literal IPs never reach the DNS lookup, so they are checked here.
+  const host = url.hostname.replace(/^\[|\]$/g, '');
+  if (isIP(host) && !isPublicAddress(host)) {
+    throw new SafeFetchError('Interne Adressen werden nicht abgerufen.');
+  }
+}
+
+function decode(response: http.IncomingMessage): Readable {
+  switch ((response.headers['content-encoding'] ?? '').toLowerCase()) {
+    case 'gzip':
+    case 'x-gzip':
+      return response.pipe(zlib.createGunzip());
+    case 'deflate':
+      return response.pipe(zlib.createInflate());
+    case 'br':
+      return response.pipe(zlib.createBrotliDecompress());
+    default:
+      return response;
+  }
+}
+
+function requestOnce(url: URL, options: Required<Options>, signal: AbortSignal) {
+  return new Promise<{ status: number; location: string | null; contentType: string; body: Buffer }>(
+    (resolve, reject) => {
+      const client = url.protocol === 'https:' ? https : http;
+      const request = client.get(
+        url,
+        {
+          lookup: guardedLookup,
+          signal,
+          headers: {
+            'user-agent': USER_AGENT,
+            accept: options.accept,
+            'accept-language': 'de-DE,de;q=0.9,en;q=0.6',
+            'accept-encoding': 'gzip, deflate, br',
+          },
+        },
+        (response) => {
+          const status = response.statusCode ?? 0;
+          const location = response.headers.location ?? null;
+          const contentType = response.headers['content-type'] ?? '';
+
+          if (status >= 300 && status < 400 && location) {
+            response.resume();
+            resolve({ status, location, contentType, body: Buffer.alloc(0) });
+            return;
+          }
+
+          const chunks: Buffer[] = [];
+          let size = 0;
+          const stream = decode(response);
+          stream.on('data', (chunk: Buffer) => {
+            size += chunk.byteLength;
+            if (size > options.maxBytes) {
+              request.destroy();
+              stream.destroy();
+              reject(new SafeFetchError('Die Seite ist zu groß.'));
+              return;
+            }
+            chunks.push(chunk);
+          });
+          stream.on('end', () =>
+            resolve({ status, location: null, contentType, body: Buffer.concat(chunks) }),
+          );
+          stream.on('error', reject);
+        },
+      );
+      request.on('error', reject);
+    },
+  );
+}
+
+export async function safeFetch(rawUrl: string | URL, options: Options = {}): Promise<SafeResponse> {
+  const settings: Required<Options> = {
+    maxBytes: options.maxBytes ?? 3 * 1024 * 1024,
+    timeoutMs: options.timeoutMs ?? 10_000,
+    accept: options.accept ?? 'text/html,application/xhtml+xml;q=0.9,*/*;q=0.5',
+  };
+  const signal = AbortSignal.timeout(settings.timeoutMs);
+
+  let url = new URL(rawUrl);
+  for (let hop = 0; hop <= MAX_REDIRECTS; hop += 1) {
+    checkUrl(url);
+
+    let result;
+    try {
+      result = await requestOnce(url, settings, signal);
+    } catch (error) {
+      if (error instanceof SafeFetchError) throw error;
+      if (signal.aborted) throw new SafeFetchError('Die Seite hat zu lange gebraucht.');
+      throw new SafeFetchError('Die Seite war nicht erreichbar.');
+    }
+
+    if (result.location) {
+      url = new URL(result.location, url);
+      continue;
+    }
+
+    return {
+      status: result.status,
+      url: url.toString(),
+      contentType: result.contentType,
+      body: result.body,
+    };
+  }
+
+  throw new SafeFetchError('Zu viele Weiterleitungen.');
+}