xiaohongshu.com

get-trending-content

Installation

Adds this website's skill for your agents

 

Summary

Extract the currently-trending short-video posts from Xiaohongshu's explore feed (homefeed_recommend), returning each as a 1-line title plus author, like count, duration, and canonical URL. Read-only — no login, like, or follow.

FIG. 01
FIG. 02
FIG. 03
FIG. 04
SKILL.md
205 lines

Xiaohongshu Trending Short Videos

Purpose

Return the currently-trending short-video posts on Xiaohongshu (小红书 / RedNote) as a list of 1-line descriptions plus structured metadata (author, like-count, duration, canonical URL). The data source is the homefeed_recommend channel — Xiaohongshu's algorithmic "for-you" feed served to logged-out visitors of https://www.xiaohongshu.com/explore, which is the platform's de-facto trending surface for unauthenticated traffic. Read-only — never click into a note, like, follow, comment, or attempt the login QR/SMS flow.

When to Use

  • "What's trending on Xiaohongshu right now?" / "Show me hot short-video posts."
  • Daily / hourly snapshots of the explore feed for content-trend dashboards.
  • Pre-screening trending videos by category (穿搭 / 美食 / 彩妆 / 影视 / 职场 / 情感 / 家居 / 游戏 / 旅行 / 健身) by switching the category slug.
  • Anywhere you'd otherwise scrape rendered cards from /explore. The cards are noise; the underlying window.__INITIAL_STATE__.feed.feeds[] blob has everything in clean JSON.

Workflow

The Xiaohongshu explore page is server-side-rendered with the first batch of feed items embedded inline in window.__INITIAL_STATE__.feed.feeds[]. This is the cheap path — one page-load gives you 20–30 ranked items (mix of type: "video" and type: "normal" (image)) with author, like count, video duration, and the xsec_token needed to construct canonical note URLs. No homefeed API call is required and none is even fired during initial load — the homefeed XHR only fires on scroll-to-load-more, and that request is signed (X-S, X-S-Common, X-t, x-rap-param) by Xiaohongshu's obfuscated xhs-secsdk bundle, so it's not callable from outside a real browser without re-implementing their signature scheme.

Do NOT call browse cloud fetch against /explore and try to parse the result — the same URL served to a non-browser client returns the SSR shell with feeds: [] empty (verified 2026-05-21: shell is 596 KB but feed.feeds is empty arrays). The SSR server gates the populated feed on a real-browser fingerprint. Use browse open … --remote with a stealth + residential-proxy session.

Recommended path — SSR initial-state extraction

  1. Create a stealth + residential-proxy Browserbase session. Both flags are mandatory; bare sessions render the shell but the feed comes back empty (the EdgeOne edge gates by TLS/UA fingerprint).

    sid=$(browse cloud sessions create --keep-alive --verified --proxies \
  | node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")

  2. Open the explore page. The login QR-code modal opens automatically — it's an overlay only; the feed is rendered behind it and the SSR JSON is already in the DOM. You can ignore the modal entirely (or press Escape for a clean screenshot — it doesn't affect data extraction).

    browse open "https://www.xiaohongshu.com/explore" --remote --session "$sid"
browse wait load --remote --session "$sid"
browse wait timeout 3000 --remote --session "$sid"   # let hydration settle

  3. Grab the rendered HTML body.

    browse get html body --remote --session "$sid" > /tmp/explore.html

  4. Extract window.__INITIAL_STATE__ and parse feed.feeds[]. The blob is JSON-shaped (Vue SSR serialization), the feeds:[ … ] array sits inside state.feed. Walk brackets from the "feeds":[ literal until depth returns to 0 — substring extraction is more reliable than regex on this 45 KB blob.

    const html = fs.readFileSync('/tmp/explore.html', 'utf8');
const m = html.match(/window\.__INITIAL_STATE__=(\{[\s\S]*?\})<\/script>/);
const blob = m[1];
const fi = blob.indexOf('"feeds":[');
let depth = 0, start = fi + 8, end = start;
for (let i = start; i < blob.length; i++) {
  if (blob[i] === '[') depth++;
  else if (blob[i] === ']') { if (--depth === 0) { end = i + 1; break; } }
}
const feeds = JSON.parse(blob.slice(start, end));

  5. Filter to videos and project the fields. noteCard.type === "video" is the video filter; "normal" is an image-only post. displayTitle is the 1-line description we want.

    const videos = feeds
  .filter(f => f?.noteCard?.type === 'video')
  .map(f => ({
    title: f.noteCard.displayTitle,                       // the 1-liner
    author: f.noteCard.user.nickname || f.noteCard.user.nickName,
    like_count: f.noteCard.interactInfo.likedCount,       // pre-formatted CN string e.g. "1.4万"
    duration_seconds: f.noteCard.video?.capa?.duration,
    note_id: f.id,
    xsec_token: f.xsecToken,
    url: `https://www.xiaohongshu.com/explore/${f.id}?xsec_token=${f.xsecToken}&xsec_source=pc_feed`,
    note_type: 'video'
  }));

  6. Release the session.

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

Optional — category filter

To restrict the feed to a single category instead of the default homefeed_recommend mix, append ?channel_id={category} to the URL. Observed category ids (from state.channel.categories[] on the same page load):

channel_idCategory
homefeed_recommend推荐 (default — mixed)
homefeed.fashion_v3穿搭
homefeed.food_v3美食
homefeed.cosmetics_v3彩妆
homefeed.movie_and_tv_v3影视
homefeed.career_v3职场
homefeed.love_v3情感
homefeed.household_product_v3家居
homefeed.gaming_v3游戏
homefeed.travel_v3旅行
homefeed.fitness_v3健身

There is no homefeed.video_feed_v3 channel — all categories are mixed video+image, filter client-side on noteCard.type === "video".

Want more than 25 items? (Not recommended)

Each SSR load returns 20–30 ranked items. Subsequent items come from POST https://edith.xiaohongshu.com/api/sns/web/v1/homefeed, but every request must carry valid X-S, X-S-Common, X-t, and x-rap-param headers computed by Xiaohongshu's obfuscated client signature SDK. Reproducing those headers outside the browser is not viable — they rotate the algorithm regularly. If you genuinely need more than one SSR-page worth of items, drive the browser further: scroll the live page (browse mouse scroll 400 400 0 3000) and the JS will fire the signed homefeed POST itself, then re-extract feed.feeds[] from the updated DOM (it's mutated in place).

Site-Specific Gotchas

  • browse cloud fetch returns the empty SSR shell. Same URL, same --proxies flag, but the EdgeOne / Tencent CDN gates the populated feed on a real-browser fingerprint (TLS, UA, sec-ch-ua headers). Verified 2026-05-21: cloud fetch returned 596 KB of HTML with "feeds":[] in the initial state, while a browse open browser session got the populated 25-item feed inline. Always use a full browser session.
  • Stealth + residential proxy is mandatory. A bare Browserbase session loads /explore but the SSR-embedded feed is empty. Both --verified and --proxies flags are required.
  • The login QR modal is cosmetic, not a wall. It overlays the feed but the data is already in the DOM behind it. You do not need to dismiss it for extraction. Press Escape only if you want a clean screenshot.
  • likedCount is a pre-formatted Chinese-style string, not an integer. Values like "1.4万" (14,000), "9.2万" (92,000), "1185", "2万" (20,000). If you need a numeric value, parse: trim "万" and multiply by 10,000; "亿" × 100,000,000.
  • displayTitle may be truncated. Cards show ~22 Chinese chars before truncation (the card is fixed-width). The SSR JSON does NOT contain a separate full-title field — displayTitle is what you get. For the full body text, you'd need to GET /explore/{noteId}?xsec_token=… and parse the note-detail page (out of scope for trending).
  • xsec_token is mandatory in the canonical URL. Direct /explore/{noteId} without the token redirects to the explore root or returns a 404-like state. Always carry the xsecToken from the same SSR snapshot — tokens are scoped to the request session and may stop working after a few hours.
  • noteCard.type values observed: "video" (short video, has video.capa.duration seconds) and "normal" (image post, no video block). No other types seen in the recommend feed. Default mix is roughly 50/50 video/image (verified iter-1: 12 video / 13 normal out of 25).
  • user.nickname vs user.nickName — both fields exist on the same object and usually have the same value, but for some users only nickName is populated (camelCase) and for others only nickname (lowercase). Always coalesce: user.nickname || user.nickName.
  • The feeds array starts at SSR index 0 but Xiaohongshu marks individually-loaded items with ssrRendered: true. Items appended later by the signed homefeed XHR do not carry this flag — useful for telling "first paint" items apart from scroll-loaded items if you ever extend the skill.
  • No homefeed.video_feed_v3 channel exists despite plausible-looking task hints. Video isolation is client-side only. (The task prompt may suggest this URL — it returns the same mixed recommend feed.)
  • Direct calls to POST /api/sns/web/v1/homefeed are non-viable. They require X-S / X-S-Common / X-t / x-rap-param signatures generated by xhs-secsdk / mnscore obfuscated JS. Don't waste time trying to reverse-engineer the signature scheme — it rotates. Use the browser path.
  • The /website/hot-list endpoint (热搜 / hot search) is a different thing — it lists trending search keywords, not trending videos. Don't confuse it with this skill's surface.
  • Scrolling-triggered API calls return 204 first, then 200 — the 204 is the CORS preflight (OPTIONS), the 200 is the actual POST. Both must succeed for the in-browser scroll-to-load-more to populate the next batch.
  • Geographic accessibility: Xiaohongshu is open globally without a CN-IP requirement (verified from US-region residential-proxy session, 2026-05-21). The login wall only blocks personalized features (search, follow, save) — public trending feed is anonymous-readable.

Expected Output

{
  "success": true,
  "source": "explore-ssr-initial-state",
  "channel": "homefeed_recommend",
  "fetched_at": "2026-05-21T23:02:30Z",
  "video_count": 12,
  "videos": [
    {
      "title": "李若彤｜好吃不胖更抗炎～再不为三餐发愁",
      "author": "李若彤",
      "like_count": "8664",
      "duration_seconds": 244,
      "note_id": "6453db04000000001300c1bb",
      "xsec_token": "ABmEUv4MbDhlwDICDI6NNR8RoC0SFuGxPHWnM2uq0iBuU=",
      "url": "https://www.xiaohongshu.com/explore/6453db04000000001300c1bb?xsec_token=ABmEUv4MbDhlwDICDI6NNR8RoC0SFuGxPHWnM2uq0iBuU=&xsec_source=pc_feed",
      "note_type": "video"
    },
    {
      "title": "新疆旅行vlog🍃是我梦里才会出现的场景啊",
      "author": "林森Live",
      "like_count": "1.4万",
      "duration_seconds": 286,
      "note_id": "6479822c0000000013002886",
      "xsec_token": "ABtkcQnFP-73URrNUeGSyeTRerpGI9UYTnt_w79xFVNqw=",
      "url": "https://www.xiaohongshu.com/explore/6479822c0000000013002886?xsec_token=ABtkcQnFP-73URrNUeGSyeTRerpGI9UYTnt_w79xFVNqw=&xsec_source=pc_feed",
      "note_type": "video"
    },
    {
      "title": "加油啊，宝………",
      "author": "鹿十元",
      "like_count": "9.2万",
      "duration_seconds": 287,
      "note_id": "644fb4db0000000007038e63",
      "xsec_token": "ABbpHsLUVDfzFGQhVGvLVuGtVX4bVBjsmPJILobMCubdg=",
      "url": "https://www.xiaohongshu.com/explore/644fb4db0000000007038e63?xsec_token=ABbpHsLUVDfzFGQhVGvLVuGtVX4bVBjsmPJILobMCubdg=&xsec_source=pc_feed",
      "note_type": "video"
    }
  ],
  "error_reasoning": null
}

Failure outcomes (less common — the SSR path is stable):

// Stealth-less session: shell renders but feed is empty
{ "success": false, "videos": [], "error_reasoning": "SSR returned empty feeds[] — session is missing --verified/--proxies stealth or was fingerprinted as bot." }

// Non-browser fetch path: same symptom as above
{ "success": false, "videos": [], "error_reasoning": "browse cloud fetch returns empty feed; must use browse open with a real browser session." }