X Content Extraction

X Content Extraction & Review

Recovering X article/post text is non-trivial because X gates long-form Articles behind a login wall and the standard web-scraping path is unreliable. This skill is the fallback ladder that actually works, plus a review pattern for the promotional articles that dominate X AI-content.

Trigger

• User shares an x.com/.../status/... or x.com/i/article/... link.
• User says "review this article on X", "read this post", "what does this X thread say".
• web_extract / Firecrawl returned empty, timed out (504 UPSTREAM_ERROR), or said "Website Not Supported" for an X URL.

Recovery ladder (try in order)

1. Browser snapshot (works for most posts). browser_navigate to the status URL. The accessibility-tree snapshot often reveals the post text, author, timestamp, and view count without a login. Long-form Articles show only a cover card + blurb here (the click-to-open article view is login-gated) — see step 3.

2. t.co redirect resolution. X post cover cards link through t.co/<id>. Following that redirect lands on x.com/i/article/<ARTICLE_ID> (extracted from the redirect_after_login query param). Note the ARTICLE_ID even if login is required.

3. xurl raw API (best for X Articles, requires xurl auth). If xurl is authenticated on the machine, the bundled social-media/xurl skill documents the exact call:

   xurl --app APP_NAME '/2/tweets/<ARTICLE_ID>?expansions=author_id&tweet.fields=created_at,article'
   

   then read data.article.plain_text from the JSON. This bypasses the login wall because it hits the API, not the rendered page. (APP_NAME = the user's authenticated X app; don't inline secrets — see the xurl skill's secret-safety rules.)

4. Mirror / search fallback. web_search the article title or a distinctive quoted phrase — promo pieces are often reposted to Instagram, YouTube, Skool, or blogs where the full text is readable. Last resort only.

Review pattern for promotional X articles

AI-content X articles are frequently affiliate lead-gen. Don't take the framing at face value. Apply:

• Separate marketing claim from technical fact. Verify each load-bearing claim against an official source (e.g. for "Hermes ships xurl" → the Hermes docs; for model claims → the vendor's own announcement). Mark verified (✅), overstated (⚠️), or unverified.
• Spot the funnel. Search the author handle + the promo's named product/person. Paid communities ("boardroom", "$X custom automations", coaching), affiliate-portal domains (goaffpro, *-goaffpro.com), and "affiliate partner / we earn commission" terms in their Terms signal the piece is a funnel, not a neutral tutorial.
• Flag hidden costs the pitch omits: you still need your own developer app + OAuth credentials, API rate limits apply, and "set-and-forget" quality is on you.
• State the access caveat first. If you could not read the full article (login wall / tool failure), say so explicitly before reviewing — review the public framing + verified claims, not an imagined full text.

Pitfalls

• web_extract on x.com reliably fails (504 timeout or "not supported") — skip it first, go straight to the browser or xurl.
• Clicking the article's cover card in the browser often throws "Something went wrong" / a login interstitial. Don't loop on it — capture the cover blurb, then use xurl (step 3).
• xurl read <ID> is for posts, NOT Articles — Articles require the raw /2/tweets endpoint with tweet.fields=article. Using read on an article ID returns nothing.
• X search tool (x_search) may be credit-blocked ("personal-team-blocked: spending-limit") — don't rely on it; the browser + web_search ladder is more robust.

Support files

• references/gated-x-article-recovery.md — worked example (cover-card → t.co → article ID → xurl call) and the funnel-spotting checklist distilled.