| name | x-post-fetch-and-media |
| description | Use when the user supplies an X/Twitter status URL and wants its post or thread fetched, read, inspected, or summarized without interaction. Do not use for posting, liking, replying, feed browsing, or generic web research. |
X Post Fetch
Fetch an exact X status URL with web_fetch and report only content actually extracted. For several exact status URLs, use web_fetch_batch and evaluate every result independently.
Workflow
- Confirm the URL is an
x.com or twitter.com status URL.
- Call
web_fetch on that exact URL. Leave include_media and include_thread false for a simple exact-post read; this uses X's official oEmbed endpoint without opening a browser.
- Require at least one extracted tweet/post. A success flag with
tweets: [], a zero count, placeholder text, or an extraction message is failure—not a captured post.
- Report author/handle, timestamp, text, thread order/count, metrics, and media hints only when present.
- Clearly label deleted, unavailable, login-gated, rate-limited, or extraction-blocked results. Do not guess the post text.
- Stop when the fetch satisfies the request.
For an explicitly requested thread, set include_thread: true and require more than the target status only when the URL actually belongs to a thread. Browser extraction can be login-gated or rate-limited, so report an incomplete thread instead of treating the exact post as the whole thread.
Media
Do not turn a read request into a download workflow. If the user explicitly asks to download, transcribe, or analyze attached media, capture the post first and then call web_fetch with include_media: true or hand the confirmed media to the appropriate media workflow.
Before claiming a media operation succeeded, verify the output file exists, has non-zero bytes, and has the expected media type. A URL, attempted download, or tool success flag alone is not proof.
Browser escalation
Use the X browser automation skill only when the user asks for interaction or when they explicitly want a browser fallback. Browser visibility is not proof of programmatic extraction; report any login wall or inaccessible content honestly.
Failure contract
Return a failure/caveat when:
- no posts were extracted;
- the status is deleted or unavailable;
- X requires authentication the current path does not have;
- extraction times out or is blocked;
- requested media cannot be validated.
The fail-closed payload validation is covered by scripts/test-phase3-fail-closed.mjs. The live smoke test in scripts/test-x-live-smoke.mjs has also captured a known public NASA status through Prometheus's real web_fetch path, including its target ID, author, handle, timestamp, text, and non-zero count without starting a media workflow. A later login wall, deletion, rate limit, or empty result is still a request-level failure; never reuse the fixture text as a fallback.
