threads-feed-scan

name

threads-feed-scan

description

Scan Threads feed posts from the homepage and collect candidate AI posts quickly. Use when browsing Threads via browser relay, especially when the page is dynamic and direct step-by-step DOM reads are too slow or unstable. Supports a page-injected JavaScript collector that auto-scrolls, accumulates post data into an in-page list, pauses/resumes, and lets the agent read the collected list instead of depending on every live DOM update.

Threads Feed Scan

Use homepage-first scanning.

• Prefer staying on https://www.threads.com/.
• Avoid opening profile pages unless the user explicitly wants profile inspection.
• If richer data is needed, allow opening a single post in another tab for view-count lookup only.
• For follower count, prefer hovering the author's avatar or handle card on the homepage.

Default workflow

1. Confirm the connected tab is the Threads homepage.
2. If the page is too dynamic for reliable live scraping, inject a page-side collector.
3. Let the collector auto-scroll and store normalized items in window.__threadCollector.items.
4. Periodically read window.__threadCollector.stats() or items.
5. Filter the collected items according to the user's criteria.
6. Pause or stop the collector when asked.

When to use the page-injected collector

Use the collector when:

• the relay DOM updates lag behind visible scrolling
• repeated snapshot calls are too slow
• the user wants high-frequency feed scanning
• you need accumulation over time rather than one-screen inspection

Collector interface

Create a global object named:

• window.__threadCollector

Expected methods/fields:

• items: collected normalized post objects
• scan(): inspect current viewport/cards and append unseen items
• scrollStep(): scroll downward
• stats(): return status summary
• stop(): stop timer and mark running=false
• running: boolean
• startedAt: timestamp
• seen: dedupe set

Suggested normalized item shape

{
  key,
  postLink,
  userHref,
  username,
  timeText,
  timeDesc,
  bodyText,
  likes,
  replies,
  reposts,
  shares,
  capturedAt
}

Recommended extraction rules

• Prefer cards containing both:
  • interaction labels like 讚, 回覆, 轉發, 分享
  • time labels like 分鐘前, 小時前, 剛剛, 大約
• Extract post link from a[href*="/post/"]
• Extract username from a non-post profile link
• Parse counts including plain numbers and 萬
• Remove obvious UI noise:
  • 首頁, 搜尋, 通知, 建立, 更多, 發佈
  • standalone metric labels
  • duplicated username lines
  • tab labels like 為你推薦
• Expect false positives on the first version; iterate on card selection and body cleanup

High-frequency mode

Use a short interval such as 1.5s–2.5s.

Recommended loop:

1. scan()
2. scrollStep()
3. repeat on interval

Keep scrolling increments moderate rather than huge to reduce missed cards.

Pause / resume behavior

When the user says pause or stop:

• call window.__threadCollector.stop()
• confirm no more auto-scroll is happening

If resuming:

• either restart the interval on the existing collector
• or create a fresh collector if the page state was lost

Practical constraints

• Threads relay pages may desync from visible browser state
• live DOM reads can lag behind the actual screen
• homepage cards may not show view counts directly
• use single-post tabs only when necessary to fetch views
• do not assume the first parse is clean; validate sample items before trusting the list

Minimal browser-evaluate pattern

Inject with browser act evaluate and return whether the collector started successfully.

After waiting a few seconds, read:

window.__threadCollector?.stats()

To inspect raw items:

window.__threadCollector?.items.slice(-20)

To stop:

window.__threadCollector?.stop()

Success condition

Success means:

• the collector is running or paused intentionally
• items are accumulating without manual step-by-step reading
• the agent can read the accumulated list directly
• the user can interrupt at any time