wikipedia.org

find-the-trending-world-news-for-today

Installation

Adds this website's skill for your agents

 

Summary

Return today's trending world news from Wikipedia — curated ITN headlines from the Main Page plus today's Current Events portal subpage bucketed by category (Business, Disasters, Politics, etc.). Read-only; uses the public MediaWiki API with no auth or anti-bot stealth.

FIG. 01
FIG. 02
FIG. 03
SKILL.md
266 lines

Wikipedia Trending World News — Today

Purpose

Return today's trending world news as Wikipedia frames it: a small set of editorially curated headlines (from Template:In_the_news, the same items shown on the Main Page "In the news" box) plus a fuller list of today's events organized by category (from the dated Portal:Current_events subpage). Output is a single JSON object keyed by date, top_headlines, categories, and source URLs. Read-only — never edits or posts anything.

When to Use

  • "What's trending in world news on Wikipedia today?"
  • A morning-briefing agent that wants a calmer, more curated alternative to algorithmic news feeds (Wikipedia ITN updates a few times per day, vetted by ITN editors).
  • A research agent that wants today's events bucketed by topic (Business and economy, Disasters and accidents, Law and crime, Politics and elections, Armed conflicts and attacks, Sports, etc.) — the portal subpage groups items that way.
  • Cross-checking what a mainstream-media outlet is leading with vs. what Wikipedia's volunteer editors deem significant.

Workflow

The recommended path is the MediaWiki public API — no auth, no cookies, no anti-bot Verified, no Browserbase session needed. Two HTTP requests to https://en.wikipedia.org/w/api.php are sufficient. Browser navigation is a fallback that costs a Browserbase session quota and gives the same data.

1. Compute today's date in Wikipedia's portal-subpage format

The portal subpage title is Portal:Current_events/{YYYY}_{Month}_{D}, where:

  • {YYYY} is the 4-digit year.
  • {Month} is the full English month name (January, February, …, December) — not zero-padded number.
  • {D} is the day of month, not zero-padded (so 5, 18, 31 — 05 will 404).

Use UTC, not local time, to match Wikipedia's clock. Example for today (2026-05-18): Portal:Current_events/2026_May_18.

TODAY_PORTAL=$(date -u +'%Y_%B_%-d')   # → 2026_May_18

2. Fetch today's portal subpage as wikitext

Wikitext is much easier to parse than HTML (categories are wrapped in '''bold''' and bullets start with *). Use the MediaWiki parse action:

GET https://en.wikipedia.org/w/api.php
    ?action=parse
    &page=Portal:Current_events/{TODAY_PORTAL}
    &format=json
    &prop=wikitext

Through Browserbase Fetch:

export BROWSERBASE_API_KEY="$BB_API_KEY"
browse cloud fetch \
  "https://en.wikipedia.org/w/api.php?action=parse&page=Portal:Current_events/${TODAY_PORTAL}&format=json&prop=wikitext"

The response is a Browserbase envelope; the page wikitext is in JSON.parse(envelope.content).parse.wikitext['*']. The body shape is:

{{Current events|year=2026|month=05|day=18|content=
<!-- All news items below this line -->
'''Business and economy'''
*[[Australia–China relations]]
**[[Australia]]n treasurer [[Jim Chalmers]] orders … [https://hongkongfp.com/… (AFP via HKFP)]

'''Disasters and accidents'''
*At least 13 people are killed … [https://www.news18.com/… (CNN-News18)]
…
<!-- All news items above this line -->}}

3. Parse wikitext into categories

  • Each '''Category Name''' line opens a new category bucket.
  • Lines starting with * (single asterisk) are first-level bullets — either standalone events or topic-header bullets (a wiki-link with no sentence).
  • Lines starting with ** (two asterisks) are sub-bullets that describe the parent topic. Fold sub-bullets into their parent when the parent is just a wiki-link header (no terminal punctuation, < ~80 chars): emit one combined string "{Parent topic}: {Sub-bullet text}". If the parent bullet is already a full sentence, emit it as-is and treat sub-bullets as siblings.
  • Strip wiki markup before emitting:
    • [[Page|Display]] → Display
    • [[Page]] → Page
    • '''bold''' → bold (strip the triple-quotes; ITN sometimes bolds key phrases)
    • ''italic'' → italic
    • [https://url (Source name)] → (Source name) (keep the parenthesized source citation; drop the URL)
    • [https://url text] → text
  • Ignore lines starting with <!--, {{, }}, or blank — those are template wrappers.

4. Fetch the curated Template:In_the_news headlines

ITN sits on the Main Page and is what most readers think of as "Wikipedia's trending news." It's a separately-edited template with 3–5 curated headlines from the last several days (not necessarily today). Same API:

GET https://en.wikipedia.org/w/api.php
    ?action=parse
    &page=Template:In_the_news
    &format=json
    &prop=text          # ← use rendered HTML here, not wikitext

Use rendered HTML for ITN (not wikitext): the template includes {{In the news/footer | currentevents = … | recentdeaths = … }} blocks that are messy to parse from raw markup but cleanly separated in the rendered output. The lead headlines are the top-level <ul><li>…</li></ul> items appearing before the <h2 id="…ongoing"> / <h2 id="…recent_deaths"> sub-headers; the "ongoing" and "recent deaths" lists come after.

A simple extraction: grab the first 5 <li>…</li> matches from the response, strip HTML tags, html-unescape — these are the curated headlines. Skip the inner <ul> sub-lists inside recentdeaths/currentevents blocks (they're shorter strings without a terminal period, easy to filter out by length+content).

5. Optional: pageviews "what's trending" signal

If "trending" should include what readers are clicking on (not just what editors curated), add a third call to the Wikimedia pageviews API:

GET https://wikimedia.org/api/rest_v1/metrics/pageviews/top/en.wikipedia/all-access/{YYYY}/{MM}/{DD}

This returns the top 1000 most-viewed pages on English Wikipedia for the given day. Filter out housekeeping pages (Main_Page, Special:Search, Wikipedia:*) and you get a real-time view of which articles are surging — often a leading indicator that an event is breaking before ITN editors curate it. Returns within a few hours' lag from real-time. Same browse cloud fetch path; no auth.

6. Assemble output

Merge into a single JSON object — see the Expected Output section. Cite source URLs for each surface so downstream agents can deep-link a user to the Wikipedia page.

Browser fallback

If the MediaWiki API is unreachable (very rare — Wikipedia's API has multi-9s uptime and serves billions of requests/day), fall back to navigating the rendered pages directly:

sid=$(browse cloud sessions create --keep-alive | jq -r .id)
export BROWSE_SESSION="$sid"

# Main Page (ITN box is in the right column under "In the news")
browse open "https://en.wikipedia.org/wiki/Main_Page" --remote
browse wait load --remote
browse get markdown body --remote | sed -n '/^## In the news/,/^## On this day/p'

# Today's portal subpage
TODAY_PORTAL=$(date -u +'%Y_%B_%-d')
browse open "https://en.wikipedia.org/wiki/Portal:Current_events/${TODAY_PORTAL}" --remote
browse wait load --remote
browse get markdown body --remote

browse cloud sessions update "$sid" --status REQUEST_RELEASE

browse get markdown body returns a clean markdown rendering with the category headings as ### and the bullets as - — easier to parse than HTML but slower and costlier than the API path (one session + 2 page loads vs. 2 raw HTTP calls).

No Verified or proxies are required for either path. Wikipedia explicitly welcomes bots that follow the API etiquette guidelines — set a descriptive User-Agent if making many requests, and keep concurrency low.

Site-Specific Gotchas

  • The portal subpage day-of-month is not zero-padded. Portal:Current_events/2026_May_5 exists; Portal:Current_events/2026_May_05 returns a "page does not exist" stub. Use date -u +'%-d' (GNU) or date -u +'%e' | tr -d ' ' (BSD), not %d. The month name is the full English name (May, not 5 or 05 or MAY).
  • Use UTC, not local time. The portal page rolls over at 00:00 UTC. A request made at 23:30 UTC for "today" in a positive-offset timezone will hit the right page; the same request from an agent that uses local time may try a future-dated page that doesn't exist yet (or worse, hit yesterday's stale page after UTC midnight).
  • Future-dated portal subpages return a 200 OK with empty content (no error). The MediaWiki parse API returns a valid JSON envelope with parse.wikitext['*'] containing only the {{Current events|year=…|month=…|day=…|content=}} wrapper and no bullets. Detect this by counting bullets — if zero events parsed AND the date is > now_utc, the page hasn't been written yet (try yesterday's). If zero events AND date is <= now_utc early in the UTC morning, the page may not have its first edit yet — try the previous day.
  • ITN bullets are typically 1–3 days old, not "today." Template:In_the_news is editorially curated, not auto-rolling. The headlines you read at any given moment are usually from items posted across the previous 24–72 hours; some stay up longer. Don't claim ITN bullets are "today's news" — frame them as "Wikipedia's currently-featured headlines." Use the portal subpage when you specifically need today's events.
  • ITN includes "Ongoing" and "Recent deaths" sub-lists which are not headlines. The rendered template has 3 sections — lead headlines, then an Ongoing block (Iran war, Russo-Ukrainian war, Sudanese civil war — long-running stories), then a Recent deaths block (6 names, no description). The lead headlines are the top-level <li>s before any <h2>. The "ongoing" entries are 1–3 words each (no terminal period) and the recent-deaths entries are just names — they're easy to filter by length + the absence of a sentence-ending period, or by detecting the <h2 id="…ongoing"> separator and only keeping <li>s before it.
  • Sub-bullet structure is inconsistent across categories. Some categories use the *topic-link\n**sub-bullet-detail pattern (Politics, Business commonly); other categories put the full sentence on the * line with no ** sub-bullets (Disasters, Law). A parser that only emits * lines drops the actual content for Politics-style entries; a parser that only emits ** lines drops the content for Disasters-style entries. Walk the wikitext top-down, hold the most recent * line in a "pending topic" buffer, and flush as "{topic}: {sub}" whenever a ** arrives; flush the * standalone when the next *, ''', or end-of-content hits without a ** in between.
  • Source citations are in square-bracket external-link syntax, not native wiki templates. Each bullet ends with [https://example.com/article (Source Name)] or [https://example.com/article Source Name]. To preserve attribution, regex-extract \[https?://\S+\s+(?:\(([^)]+)\)|([^\]]+))\] and keep the parenthesized name or trailing text as the source field. To preserve the original URL, capture the full bracketed group before stripping.
  • Wikitext is much easier than rendered HTML for the portal page. The rendered HTML embeds <style> blocks, a navbar with edit/history/watch links, ARIA region wrappers, and class="current-events-content-heading" divs that don't always appear as expected (today's portal omitted them entirely — the categories were only in '''bold''' wikitext). Always pull prop=wikitext for the portal subpage. For ITN, the opposite is true — pull prop=text (rendered HTML) because wikitext has nested templates ({{In the news/footer}}) that are simpler to read after expansion.
  • No anti-bot at all on api.wikipedia.org. No CAPTCHAs, no rate-limit blocks at reasonable volumes, no User-Agent sniffing that rejects requests (though the API etiquette page asks you to set a descriptive UA). Browserbase --proxies or --verified flags are not needed — bare HTTP from any IP works. The skill was developed and verified with no Verified, no proxies, and a single bare browse cloud fetch call per surface.
  • The pageviews API has a ~24-hour lag, sometimes more. wikimedia.org/api/rest_v1/metrics/pageviews/top/en.wikipedia/all-access/{Y}/{M}/{D} for today often returns a 404 ("data not available") until the daily roll-up is processed (typically late the next day UTC). For a real-time-ish trending signal, request yesterday's date — or use the per-hour endpoint …/per-article/en.wikipedia/all-access/all-agents/{article}/hourly/… for a specific article. Don't depend on pageviews for "what broke in the last hour."
  • Portal:Current_events (unsuffixed) shows the last ~7 days, not today. That redirect lands on a transclusion of the last week's subpages all stacked together. Don't parse it for "today's events" — you'll mix in yesterday and the day before. Always use the dated subpage Portal:Current_events/{Year}_{Month}_{Day}.
  • User-Agent should be descriptive. Wikimedia explicitly requests it (policy) and will throttle or eventually block anonymous python-requests/2.x or empty UAs. browse cloud fetch sets a Browserbase UA by default which is acceptable; if you raw-curl, include something like MyAgent/1.0 (https://example.com; bot@example.com).

Expected Output

A single JSON object. The top_headlines array is Template:In_the_news lead bullets (curated, 0–5 items, frequently 1–3 days old). The categories object is today's portal subpage events, bucketed by editor-assigned category. Optional trending_pageviews is the day's most-viewed articles (lagged 24h). as_of_utc reflects the date used to address the portal subpage, not request time.

{
  "as_of_utc": "2026-05-18",
  "sources": {
    "portal_subpage": "https://en.wikipedia.org/wiki/Portal:Current_events/2026_May_18",
    "in_the_news_template": "https://en.wikipedia.org/wiki/Template:In_the_news",
    "pageviews_api": "https://wikimedia.org/api/rest_v1/metrics/pageviews/top/en.wikipedia/all-access/2026/05/18"
  },
  "top_headlines": [
    {
      "text": "Bulgaria, represented by Dara with the song \"Bangaranga\", wins the Eurovision Song Contest.",
      "source": "Template:In_the_news",
      "note": "curated by ITN editors; may be 1–3 days old"
    },
    {
      "text": "The World Health Organization declares the Ebola epidemic in the Democratic Republic of the Congo and Uganda a public health emergency of international concern.",
      "source": "Template:In_the_news"
    },
    {
      "text": "Niuean prime minister Dalton Tagelagi is re-elected for a third term.",
      "source": "Template:In_the_news"
    },
    {
      "text": "The Philippine Senate goes into lockdown after the attempted arrest of senator Ronald dela Rosa.",
      "source": "Template:In_the_news"
    }
  ],
  "categories": {
    "Business and economy": [
      {
        "topic": "Australia–China relations",
        "text": "Australian treasurer Jim Chalmers orders several China-linked shareholders to divest their stakes in Northern Minerals under foreign investment laws aimed at protecting the country's rare earths sector.",
        "source_name": "AFP via HKFP",
        "source_url": "https://hongkongfp.com/2026/05/18/australia-orders-china-linked-investors-to-sell-stakes-in-rare-earths-firm/"
      }
    ],
    "Disasters and accidents": [
      {
        "text": "At least 13 people are killed and more than 20 others are injured, most seriously, when a tempo carrying wedding guests collides with a container truck on the Mumbai-Ahmedabad Highway in Palghar district, Maharashtra, India.",
        "source_name": "CNN-News18",
        "source_url": "https://www.news18.com/india/13-killed-dozens-injured-in-tempo-truck-crash-on-mumbai-ahmedabad-highway-ws-l-10098391.html"
      }
    ],
    "Law and crime": [
      {
        "text": "Four people are killed and eight others are injured in two spree shootings, including one inside a restaurant, in Mersin, Mediterranean region, Turkey. A widespread manhunt is underway to catch the suspect.",
        "source_name": "Insider Paper",
        "source_url": "https://insiderpaper.com/4-dead-8-hurt-as-gunman-opens-fire-in-southern-turkey-reports/"
      }
    ],
    "Politics and elections": [
      {
        "topic": "Second impeachment of Sara Duterte",
        "text": "The Philippine senate resumes the impeachment process against Vice President Sara Duterte following last week's lockdown.",
        "source_name": "Reuters",
        "source_url": "https://www.reuters.com/world/asia-pacific/philippines-convene-court-vp-impeachment-amid-political-turmoil-2026-05-18/"
      }
    ]
  },
  "trending_pageviews": [
    { "article": "Eurovision_Song_Contest_2026", "views": 1842301, "rank": 1 },
    { "article": "Dara_(Bulgarian_singer)",      "views":  612408, "rank": 2 }
  ]
}

Categories that appear in the portal but have no items today should be omitted (or emitted as empty arrays — both are acceptable, agents using this skill should treat absence as "no events in that category today"). The canonical category labels used by ITN editors are: Armed conflicts and attacks, Arts and culture, Business and economy, Disasters and accidents, Health and environment, International relations, Law and crime, Politics and elections, Science and technology, and Sports — though not all appear every day. Always read category names from the wikitext rather than hardcoding the list.

Edge cases the consumer should be ready for:

  • Empty top_headlines (rare — ITN almost always has at least one bullet, but it can lag during slow news cycles).
  • Empty categories (the portal subpage exists with zero events — happens early UTC morning before the first edit lands; retry against today - 1 day if you need non-empty data).
  • Missing trending_pageviews (the pageviews endpoint 404s on today's date until the daily roll-up is processed — degrade gracefully, return only top_headlines + categories).
  • Wikitext that includes unexpected templates (e.g., {{Wikinews|…}} or {{cite news|…}}) inline within a bullet — strip with the same \{\{[^}]*\}\} regex you use for the outer {{Current events|…|content=…}} wrapper, or accept that the rendered text may contain template-syntax fragments and emit them anyway.
Wikipedia Trending World News · browse.sh