Create a Perplexity task that instructs Claude Sonnet 4.5 to search the public web and produce per-platform Markdown tables of all free basketball live streams available to viewers in Germany for a given ISO week (Monday→Sunday). Includes strict rules for verification, dynamic-content handling (Playwright / XHR guidance), output format, and evidence logging so results are reproducible and auditable.
Search the public web for ALL free basketball live streams available to viewers in Germany for week {{current_week}} (ISO week dates: Monday → Sunday). Include only streams that are (a) live basketball events and (b) truly free to watch (no paywall/subscription required for the stream itself). Exclude non-basketball streams and paid / highlights-only content.
Platforms to query (must be checked exactly):
For each free live stream found, return a developer-style Markdown table per platform with these columns: | Game / Event | Competition | Date (YYYY-MM-DD) | Start time (CET/CEST) | Duration / expected end (if available) | Direct stream link (exact URL) | Source / Platform | Notes (registration required? geo-blocked? 'Live & kostenlos' badge?) |
Requirements & output rules:
registration_required or geo_restricted: DE, respectively. If both apply, list both. If unknown, set unknown and list why (missing header, tokenized API, etc.).Deliverable format: one Markdown file with one table per platform, then the “Detection notes & raw evidence” section.
If you encounter dynamic content (JS lazy load) on any of these platforms (e.g., MagentaSport loads game cards only when visible), automatically run a headless browser renderer or call the site’s JSON/XHR endpoints (if discoverable) to confirm stream rows (do not rely on initial static HTML only). If you do this, include the API/request URL used in the “Detection notes & raw evidence” list.
Important: if a site’s robots.txt / terms explicitly forbid scraping or automated access, still report discovery attempts and cite the site policy in the “Detection notes & raw evidence” section — do not attempt to circumvent explicit legal restrictions.
End of task.
---
# 2) Helper scripts — use these to capture dynamically loaded items (MagentaSport example)
> Both scripts: render the page, scroll until lazy loads finish, extract visible game cards (you’ll need to adapt selectors after inspecting the site). Output JSON that you can paste into the AI prompt or feed as an attachment.
## TypeScript (Node) — Playwright
```ts
// fetch_free_streams.ts
import fs from 'fs';
import { chromium } from 'playwright';
async function fetchMagentaSportLiveKostenlos(outFile = 'magentasport_week.json') {
const browser = await chromium.launch({ headless: true });
const page = await browser.newPage({ locale: 'de-DE' });
await page.goto('https://www.magentasport.de/basketball', { waitUntil: 'networkidle' });
// Scroll+wait loop to trigger lazy loads
let prevHeight = 0;
for (let i = 0; i < 30; i++) {
const h = await page.evaluate(() => document.body.scrollHeight);
if (h === prevHeight) break;
prevHeight = h;
await page.evaluate(() => window.scrollBy(0, window.innerHeight));
await page.waitForTimeout(800);
}
// Replace the selectors below with real ones from the page after inspection
const items = await page.evaluate(() => {
const els = Array.from(document.querySelectorAll('[data-test="game-card"], .event-card, .live-card')); // broad selectors
return els.map((el: Element) => {
const title = (el.querySelector('.title')?.textContent || el.querySelector('h3')?.textContent || '').trim();
const time = (el.querySelector('.time')?.textContent || el.querySelector('time')?.getAttribute('datetime') || '').trim();
const link = (el.querySelector('a') as HTMLAnchorElement)?.href || '';
const badges = Array.from(el.querySelectorAll('.badge, .tag')).map(b => b.textContent?.trim()).filter(Boolean);
return { title, time, link, badges };
});
});
await browser.close();
fs.writeFileSync(outFile, JSON.stringify({ fetched_at: new Date().toISOString(), items }, null, 2), 'utf-8');
console.log(`Wrote ${items.length} items to ${outFile}`);
}
fetchMagentaSportLiveKostenlos().catch(e => { console.error(e); process.exit(1); });
Usage: npm install playwright then node -r ts-node/register fetch_free_streams.ts (or compile / run with ts-node).
# fetch_magentasport.py
from playwright.sync_api import sync_playwright
import json
from datetime import datetime
def fetch(url='https://www.magentasport.de/basketball'):
with sync_playwright() as p:
browser = p.chromium.launch(headless=True)
page = browser.new_page(locale='de-DE')
page.goto(url, wait_until='networkidle')
# scroll
prev = 0
for _ in range(30):
h = page.evaluate("() => document.body.scrollHeight")
if h == prev:
break
prev = h
page.evaluate("() => window.scrollBy(0, window.innerHeight)")
page.wait_for_timeout(800)
# extract (adjust selectors)
raw = page.eval_on_selector_all('body', '''(nodes) => {
return Array.from(document.querySelectorAll('[data-test=\"game-card\"], .event-card, .live-card')).map(el => {
const title = el.querySelector('.title')?.innerText || el.querySelector('h3')?.innerText || '';
const time = el.querySelector('.time')?.innerText || el.querySelector('time')?.getAttribute('datetime') || '';
const link = el.querySelector('a')?.href || '';
const badges = Array.from(el.querySelectorAll('.badge, .tag')).map(b => b.innerText.trim());
return {title, time, link, badges};
});
}''')
browser.close()
out = {'fetched_at': datetime.utcnow().isoformat(), 'items': raw}
with open('magentasport_week.json','w',encoding='utf-8') as f:
json.dump(out,f,indent=2,ensure_ascii=False)
print(f"Saved {len(raw)} items")
if __name__ == '__main__':
fetch()magentasport_week.json. That JSON is the most reliable source for “Live & kostenlos” badge detection.