---

name: twitter-x-hub description: > Skill for reading and writing Twitter/X data with Python + UV, with zero third-party dependencies (pure standard library), by directly passing the auth_token + ct0 cookies for authentication. In the Minis environment, cookies can be retrieved automatically with the browser_use tool by navigating to x.com and then using the get_cookies action, with no manual copying required. Supports scraping the home timeline, following list, bookmarks (including bookmark folders), search, user profiles, user tweets, likes, tweet details (single tweet or with replies), List timelines, followers/following lists, and write operations such as posting, deleting, liking, retweeting, and bookmarking. This skill must be triggered when the user mentions "scraping Twitter data," "get X tweets," "Twitter timeline," "X bookmarks," "search tweets," "twitter-x-hub," "request Twitter with cookies," "Twitter GraphQL," or any scenario that requires programmatic reading or writing of Twitter/X data.

twitter-x-hub

Modified from: public-clis/twitter-cli (formerly jackwener/twitter-cli) This skill simplifies the original repository as follows: it removes the browser-cookie3/rich/click/PyYAML/curl_cffi/ xclienttransaction/beautifulsoup4 dependencies and replaces them with a pure standard library implementation; authentication now uses directly supplied cookies instead of automatic browser extraction; Twitter Article rendering and image upload functionality have been removed.

---

File Structure

/var/minis/skills/twitter-x-hub/
├── SKILL.md
├── pyproject.toml              # UV project configuration (zero third-party dependencies)
└── scripts/
    ├── __init__.py
    ├── models.py               # Data models (Tweet, Author, Metrics, UserProfile, BookmarkFolder)
    ├── parser.py               # GraphQL response parsing (split from client.py, synced from upstream v0.8.6)
    ├── client.py               # GraphQL client (core logic)
    └── cli.py                  # Command-line entry point (argparse)

---

Authentication

The Twitter/X internal GraphQL API uses two cookies for authentication:

Cookie	Description
auth_token	User login credential (OAuth Session Token)
ct0	CSRF Token, also used as the X-Csrf-Token request header

Method 1: Retrieve automatically with the browser_use tool (recommended, preferred in the Minis environment)

In Minis, you can use the browser_use tool to navigate directly to x.com, then use the get_cookies action to read cookies, with no manual copying required. After retrieval, store them in environment variables immediately to avoid exposing plaintext values in the conversation context.

Steps:

1. Use browser_use navigate to open https://x.com and confirm you are logged in.
2. Use browser_use get_cookies to retrieve all cookies.
   • The tool returns an offload env file path, such as /var/minis/offloads/env_cookies_xxx.sh.
   • Raw cookie values will not appear in the conversation.
3. Load the file, then use the cookies:

. /var/minis/offloads/env_cookies_xxx.sh
export TWITTER_AUTH_TOKEN="$COOKIE_AUTH_TOKEN"
export TWITTER_CT0="$COOKIE_CT0"

Method 2: Set environment variables manually

Copy auth_token and ct0 from browser DevTools -> Application -> Cookies -> https://x.com, then store them in Minis environment variables (Settings -> Environments): TWITTER_AUTH_TOKEN + TWITTER_CT0

Passing credentials (three methods, in descending priority)

1. Environment variables: TWITTER_AUTH_TOKEN + TWITTER_CT0 (recommended)
2. CLI parameters: --auth-token <value> --ct0 <value>
3. Pass directly in code: TwitterClient(auth_token=..., ct0=...)

---

Quick Start

Environment Setup

# Confirm UV is available
which uv || pip install uv

# Enter the skill directory
cd /var/minis/skills/twitter-x-hub

CLI Usage

# Scrape the home For-You timeline (20 items by default)
uv run python -m scripts.cli feed

# Scrape the Following timeline, 30 items, JSON output
uv run python -m scripts.cli feed --type following --max 30 --json

# Search tweets (Top/Latest/Photos/Videos)
uv run python -m scripts.cli search "Claude Code" --tab Latest --max 20

# Bookmarks
uv run python -m scripts.cli bookmarks --max 50

# Bookmark folder list (new)
uv run python -m scripts.cli bookmark-folders

# User profile
uv run python -m scripts.cli user elonmusk

# User tweets
uv run python -m scripts.cli user-posts elonmusk --max 20

# User likes
uv run python -m scripts.cli user-likes elonmusk --max 20

# Tweet details (including reply thread)
uv run python -m scripts.cli tweet 1234567890

# Quickly retrieve a single tweet (new, faster than the tweet command)
uv run python -m scripts.cli tweet-by-id 1234567890

# List timeline
uv run python -m scripts.cli list 1539453138322673664

# Followers / following lists (first use the user command to get user_id)
uv run python -m scripts.cli followers <user_id> --max 50
uv run python -m scripts.cli following <user_id> --max 50

# Post a tweet / reply
uv run python -m scripts.cli post "Hello from twitter-x-hub!"
uv run python -m scripts.cli post "reply text" --reply-to 1234567890

# Like / retweet / bookmark
uv run python -m scripts.cli like 1234567890
uv run python -m scripts.cli retweet 1234567890
uv run python -m scripts.cli bookmark 1234567890

Use environment variables to avoid passing credentials each time

export TWITTER_AUTH_TOKEN="xxxx"
export TWITTER_CT0="yyyy"

uv run python -m scripts.cli feed --max 30 --json

Call as a Python library

import os, json, dataclasses
from scripts.client import TwitterClient

client = TwitterClient(
    auth_token=os.environ["TWITTER_AUTH_TOKEN"],
    ct0=os.environ["TWITTER_CT0"],
)

# Scrape the home timeline
tweets = client.fetch_home_timeline(count=20)
for t in tweets:
    print(f"@{t.author.screen_name}: {t.text[:80]}")
    print(f"  ❤️ {t.metrics.likes}  🔁 {t.metrics.retweets}  👁 {t.metrics.views}  🔖 {t.metrics.bookmarks}")

# Search (Latest tab)
results = client.fetch_search("AI agent", count=10, product="Latest")

# Single tweet (quick, no replies)
tweet = client.fetch_tweet_by_id("1234567890")

# Bookmark folders
folders = client.fetch_bookmark_folders()

# User profile
user = client.fetch_user("elonmusk")
print(user.id, user.followers_count)

# JSON serialization
data = [dataclasses.asdict(t) for t in tweets]
print(json.dumps(data, ensure_ascii=False, indent=2))

---

Core Implementation Principles

Authentication Mechanism

Uses browser cookies (auth_token + ct0) plus a hard-coded public Bearer Token to impersonate a Chrome browser request to Twitter's internal GraphQL API.

Three-level QueryId resolution (automatically handles API changes)

1. In-memory cache (fastest)
2. Hard-coded FALLBACK_QUERY_IDS (constant fallback)
   → If 404, the queryId has expired; proceed to the next level
3. Fetch the latest queryId from github.com/fa0311/twitter-openapi
   → If it is still unavailable, scan the x.com JS Bundle and extract it with a regular expression

URL Optimization (synced from upstream v0.8)

• Keys with a value of False in the features dictionary are not sent, avoiding overly long URLs (414 errors).

Pagination & Rate Limiting

• Each response includes a cursor; pages are fetched automatically until the count limit is reached.
• The default request interval is 1.5 seconds + ±30% random jitter. HTTP 429 triggers exponential backoff and retry.
• Write operations use a random delay of 1.5 to 4 seconds.

Parser Split (synced from upstream v0.7+)

• parser.py was split out from client.py and includes independent functions such as parse_tweet_result, parse_timeline_response, and parse_user_result, making unit testing and reuse easier.

---

CLI Subcommand Quick Reference

Subcommand	Description	Key Parameters
feed	Home timeline	--type for-you|following, --max, --json
bookmarks	Bookmarks	--max, --json
bookmark-folders	Bookmark folder list ⭐ New	--json
search	Search	query, --tab Top|Latest|Photos|Videos, --max, --json
user	User profile	screen_name, --json
user-posts	User tweets	screen_name, --max, --json
user-likes	User likes	screen_name, --max, --json
tweet	Tweet details + replies	tweet_id, --max, --json
tweet-by-id	Single tweet (quick) ⭐ New	tweet_id, --json
list	List timeline	list_id, --max, --json
followers	Followers list	user_id, --max, --json
following	Following list	user_id, --max, --json
post	Post a tweet	text, --reply-to
delete	Delete a tweet	tweet_id
like / unlike	Like/unlike	tweet_id
retweet / unretweet	Retweet/unretweet	tweet_id
bookmark / unbookmark	Bookmark/unbookmark	tweet_id

All subcommands support the --auth-token / --ct0 parameters, which can also be replaced with environment variables.

---

Change Log (synced from upstream)

v0.8.6 Sync (2026-04-08)

• Full QueryId update: Real-time scanning from the x.com JS bundle (main.0e98bc8a.js) updated all IDs, including HomeTimeline, HomeLatestTimeline, UserTweets, SearchTimeline, Likes, TweetDetail, TweetResultByRestId, ListLatestTweetsTimeline, Followers, Following, CreateTweet, and others.
• New QueryIds: TweetResultByRestId, BookmarkFoldersSlice, BookmarkFolderTimeline
• New commands: tweet-by-id (quick single-tweet retrieval), bookmark-folders (bookmark folders)
• models.py: Added the bookmarks field to Metrics; added the article_title, article_text, and is_subscriber_only fields to Tweet; added the BookmarkFolder dataclass.
• parser.py: Split from client.py into an independent module; fixed the new API structure (core.name/core.screen_name); parse_tweet_result supports full text from note_tweet (long tweets with "Show more"); added _unwrap_visibility to handle TweetWithVisibilityResults; fixed parse_user_result so the joined date is read from core.created_at.
• URL optimization: False values in features are not sent, avoiding 414 errors.
• SearchTimeline restriction: Starting in late 2025, X requires the x-client-transaction-id header. This header is generated by xclienttransaction (a C extension), which cannot be installed in the iSH/Alpine environment. Therefore, the search command is temporarily unavailable in this environment. Alternative: use browser_use to navigate to the search page and extract the DOM.

---

Notes

• Cookies are typically valid for several weeks to several months. After they expire, they must be retrieved from the browser again.
• Use a dedicated secondary account to reduce the risk of your main account being flagged by risk controls.
• Write operations, such as posting tweets and liking tweets, carry a higher risk of triggering risk controls than read operations. Use them at your discretion.
• max_count has a hard limit of 500 to prevent accidentally sending a large number of requests.
• Upstream uses curl_cffi for TLS fingerprint spoofing. This skill uses stdlib urllib instead. If you encounter risk controls, you can try passing the complete cookie string through the cookie_string parameter to strengthen the fingerprint.