Twitter / X (script-mode)

Read-only access to twitterapi.io endpoints. 13 functions covering tweets, users, followers, replies, threads, quotes, articles, and trends.

All requests go through sc-proxy via core.httpclient.proxiedget. The TWITTERAPIKEY env var is auto-injected server-side, no local key needed on the agent machine.

Script Usage

Standard invocation pattern:

python3 - <<'EOF'
import sys, json
sys.path.insert(0, "/data/workspace/skills/twitter")
from exports import twitter_user_info, twitter_user_tweets

profile = twitter_user_info(username="vitalikbuterin")
print(json.dumps(profile, indent=2))

recent = twitter_user_tweets(username="vitalikbuterin")
print(f"got {len(recent.get('tweets', []))} tweets")
EOF

Tweet ID extraction from URL: the last path segment of any x.com/{user}/status/{id} or twitter.com/{user}/status/{id} URL is the tweet ID. Pass it as a string (Python int will lose precision on long IDs).

Function Reference (signatures)

All 13 functions live in exports.py. Returns are dicts straight from twitterapi.io — keys vary per endpoint, inspect once before scripting.

Tweet endpoints

User endpoints

username is the handle WITHOUT @ (e.g. "elonmusk", not "@elonmusk"). Pagination: when a response includes next_cursor, pass it back as cursor on the next call.

When to use this skill

• ANY x.com/... or twitter.com/... URL → start here, NOT web_fetch

(Twitter blocks scrapers).

• Single tweet detail → twittergettweets([tweet_id]).
• "What's @user been posting?" → twitterusertweets.
• KOL discovery / cashtag mentions → twittersearchtweets("$SOL min_faves:50").
• Trending topics → twittergettrends.

Billing & cost control (READ before bulk/scheduled use)

twitterapi.io bills per item actually returned, not per request and NOT by any "max_results" you ask for. sc-proxy charges = returned-item-count × unit (tweets 45 / profiles 54 / followers 45 credits; 100k credits = $1; 3× upstream). Min 1 item per request.

The lasttweets / usertweets trap: the upstream /twitter/user/lasttweets endpoint has no page-size parameter — it always returns up to 20 tweets per page. There is no maxresults / pageSize lever, and twitterusertweets() does not accept one. So "I only need 5" still fetches and bills for ~20. Slicing the result client-side does NOT save money — the charge is already counted at the proxy from the upstream response.

⭐ Polling for "new tweets from account X" → use search, NOT last_tweets

This is the biggest, most common waste. twitterusertweets() (upstream lasttweets) has no page-size param and always bills a full ~20-tweet page every call, even when nothing new was posted. The official twitterapi.io guide recommends the advancedsearch endpoint instead, which our skill already exposes as twittersearchtweets():

# Cheap polling pattern — bills only the tweets actually in the window.
# When NO new tweet exists, the call is billed as 1 item (not 20).
import time
since = int(last_check_unix)
until = int(time.time())
q = f"from:{handle} include:nativeretweets since_time:{since} until_time:{until}"
res = twitter_search_tweets(q)   # queryType defaults to Latest

Official pricing (upstream; our proxy bills 3×):

• tweets found → $0.00015 per returned tweet
• no tweets found → $0.00015 for the whole call (vs last_tweets' ~20× that)

Per-call cost in our billing makes the difference obvious:

• last_tweets → ~$0.009/call (20 tweets every time)
• advanced_search empty window → ~$0.00045/call (1 item) — ~20× cheaper

Frequency vs monthly cost (single account, upstream): hourly $0.11 · 30min $0.22 · 15min $0.43 · 5min $1.30 · 1min $6.48.

Other cost levers

• Use get_tweets([ids]) when IDs are known — pay only for those exact

tweets, not a 20-item page.

• Followers/followings bill per returned profile (default page 200 → 200

billed). Only paginate as far as needed. For ID-only graph work use the bulk followers-IDs endpoint (lightweight).

• Tighten search queries (minfaves, sincetime/until_time, lang) so fewer

pages are needed.

Note: twitterapi.io also sells a managed stream/webhook product. **We do NOT
subscribe to it** — do not use the /oapi/xuserstream/* or
/oapi/tweet_filter/* endpoints. For any account-monitoring need, the
advanced_search polling pattern above is the correct and only approach here.

Error handling

• 402 Credits is not enough → upstream proxy credits exhausted; tell user

to top up. Don't retry.

• 429 → rate limited; surface to user, don't auto-retry.
• 404 user not found → suggest verifying the handle spelling.

Version Policy (hard rule)

This skill is script-mode (delivery: script). It does NOT register runtime tools — agent must read_file SKILL.md and call functions via bash + python3. The legacy tools.py / init.py files are kept for backward compatibility but are no longer the preferred entry point.

Bump rules:

• Any signature change, env-var change, or sc-proxy contract change → MAJOR
• New function added, response schema clarified → MINOR
• Bug fix or doc-only change → PATCH