Imported from joelhooks/joelclaw (skills/x-api/SKILL.md). Install upstream with npx skills add joelhooks/joelclaw --skill x-api. Copyright stays with the author.

X/Twitter API Skill

Read paths stay here. X Articles writes go through shitrat x. Tweets still use OAuth 1.0a until that lands on the same CLI.

Authentication

Recommended for read-only access: app bearer token

There is no stored x_bot_bearer_token secret right now. Derive the app bearer token from the X app consumer key + secret, then use that bearer token for read-only endpoints like tweet lookup.

python3 <<'PY'
import base64, json, subprocess
from urllib import request, parse


def lease(name: str) -> str:
    raw = subprocess.check_output(['secrets', 'lease', name], text=True)
    data = json.loads(raw)
    return data.get('secret') or data.get('result') or raw.strip()

ck = lease('x_consumer_key')
cs = lease('x_consumer_secret')
cred = base64.b64encode(f'{ck}:{cs}'.encode()).decode()

req = request.Request(
    'https://api.twitter.com/oauth2/token',
    data=parse.urlencode({'grant_type': 'client_credentials'}).encode(),
    headers={
        'Authorization': f'Basic {cred}',
        'Content-Type': 'application/x-www-form-urlencoded;charset=UTF-8',
    },
    method='POST',
)
with request.urlopen(req) as r:
    bearer = json.loads(r.read().decode())['access_token']

req2 = request.Request(
    'https://api.twitter.com/2/tweets/TWEET_ID',
    headers={'Authorization': f'Bearer {bearer}'},
)
with request.urlopen(req2) as r:
    print(r.read().decode())
PY

# Revoke only the exact lease ids acquired by this task; never all sessions’ leases.

Use this for:

• reading a specific tweet/post
• fetching public user metadata
• recent search
• full-archive search when the app has access
• reply/context passes for source ingestion
• other public read-only X v2 endpoints

Reply/context pass for ingestion

When ingesting an X post/article as a source, do a reply pass before writing the final note. The replies often contain the correction, missing caveat, source link, or better framing. Do not blindly archive the whole reply swamp; extract the high-signal replies and keep receipts.

Minimum pass:

1. Fetch the root post with conversation_id, author_id, created_at, entities, public_metrics, referenced_tweets, note_tweet, and article when available.
2. Search replies with conversation_id:ROOT_ID is:reply -is:retweet.
3. Capture high-signal replies: author clarification, author thread continuation, cited source/link, strong objection, correction, implementation detail, or unusually crisp language.
4. Store reply receipts with tweet_id, author handle/name, created time, URL, text basis, metrics, and why it matters.
5. In the Brain/source note, separate the root quote from Reply signals so public claims do not blur root-author text with commenter interpretation.

For posts inside the recent-search window:

# After deriving bearer as above:
curl "https://api.x.com/2/tweets/search/recent?query=conversation_id%3AROOT_ID%20is%3Areply%20-is%3Aretweet&max_results=100&tweet.fields=author_id,conversation_id,created_at,entities,id,in_reply_to_user_id,note_tweet,public_metrics,referenced_tweets,text&expansions=author_id,referenced_tweets.id,referenced_tweets.id.author_id&user.fields=username,name,verified,verified_type" \
  -H "Authorization: Bearer $bearer"

For older posts, use Full-Archive Search when access allows it; otherwise fall back to browser inspection and mark the limitation in the verification note.

Full-Archive Search

The bot app can call Full-Archive Search if the access tier permits it. Do not use Recent Search for YTD/backfill research; it rejects older start_time values.

# After deriving bearer as above:
curl "https://api.x.com/2/tweets/search/all?query=from%3Athreepointone%20-is%3Areply&start_time=2026-01-01T00%3A00%3A00Z&end_time=2026-05-02T00%3A00%3A00Z&max_results=100&tweet.fields=created_at,public_metrics" \
  -H "Authorization: Bearer $bearer"

Notes:

• Full archive endpoint: /2/tweets/search/all.
• Recent endpoint: /2/tweets/search/recent; last-7-days style window only.
• tweet.fields=context_annotations forces max_results <= 100; X returns 400 if paired with max_results=500.
• Full-archive docs: https://docs.x.com/x-api/posts/search/quickstart/full-archive-search

Alternative: OAuth 1.0a User Context (for posting and account actions)

Tokens do not expire — no refresh dance needed.

Secrets (in agent-secrets)

• x_consumer_key — API Key (OAuth 1.0a consumer key)
• x_consumer_secret — API Key Secret (OAuth 1.0a consumer secret)
• x_access_token — Access Token (OAuth 1.0a, format: numeric-alphanumeric)
• x_access_token_secret — Access Token Secret (OAuth 1.0a)

@joelhooks / shitrat-joel app

Use these scoped secret names for Joel's personal X account so they don't collide with the @joelclaw bot app:

• x_joelhooks_consumer_key
• x_joelhooks_consumer_secret
• x_joelhooks_bearer_token
• x_joelhooks_access_token
• x_joelhooks_access_token_secret

PIN OAuth note, 2026-06-28: https://api.x.com/oauth/authorize?... showed “There is no request token.” Regenerating with https://api.twitter.com/oauth/request_token and opening https://twitter.com/oauth/authorize?force_login=true&oauth_token=... worked for shitrat-joel → @joelhooks.

Auth verification, 2026-06-28: x_joelhooks_access_token + x_joelhooks_access_token_secret verified with GET https://api.twitter.com/2/users/me as @joelhooks / user id 12087242.

Bookmark caveat: GET /2/users/12087242/bookmarks returned 403 Unsupported Authentication with OAuth 1.0a. X requires OAuth 2.0 User Context / PKCE with bookmark.read, tweet.read, and users.read for bookmarks. Do not assume the saved OAuth 1.0a token can read bookmarks.

Signing requests

OAuth 1.0a requires cryptographic signing. Use requests-oauthlib (Python) or equivalent:

export CK=$(secrets lease x_consumer_key)
export CS=$(secrets lease x_consumer_secret)
export AT=$(secrets lease x_access_token)
export ATS=$(secrets lease x_access_token_secret)

uv run --with requests-oauthlib python3 << 'PYEOF'
import os
from requests_oauthlib import OAuth1Session

client = OAuth1Session(
    os.environ['CK'],
    client_secret=os.environ['CS'],
    resource_owner_key=os.environ['AT'],
    resource_owner_secret=os.environ['ATS'],
)

r = client.get("https://api.twitter.com/2/users/me")
print(r.json())
PYEOF

Revoke leases after use

# Revoke only the exact lease ids acquired by this task; never all sessions’ leases.

Account Info

• Account: @joelclaw
• User ID: 2022779096049311744
• Scopes: Read and write (OAuth 1.0a app permissions)

@joelhooks Account Info

• Account: @joelhooks
• User ID: 12087242
• App: shitrat-joel
• OAuth 1.0a status: saved and verified for user-context actions that allow OAuth 1.0a
• Bookmarks: not available through OAuth 1.0a; needs OAuth 2.0 PKCE user token with bookmark.read

Common Operations

Read-only lookups can use the app bearer-token flow above.

Posting, replying, following, deleting, and account-scoped actions require OAuth 1.0a signing. Use the Python pattern from Authentication section above, then:

Common API calls (inside OAuth1Session)

# Check mentions
r = client.get("https://api.twitter.com/2/users/2022779096049311744/mentions",
    params={"max_results": 10, "tweet.fields": "created_at,author_id,text",
            "expansions": "author_id", "user.fields": "username,name"})

# Get my timeline
r = client.get("https://api.twitter.com/2/users/2022779096049311744/tweets",
    params={"max_results": 10, "tweet.fields": "created_at,public_metrics"})

# Post a tweet
r = client.post("https://api.twitter.com/2/tweets", json={"text": "your tweet"})

# Reply to a tweet
r = client.post("https://api.twitter.com/2/tweets",
    json={"text": "@user reply", "reply": {"in_reply_to_tweet_id": "TWEET_ID"}})

# Search recent tweets
r = client.get("https://api.twitter.com/2/tweets/search/recent",
    params={"query": "joelclaw", "max_results": 10, "tweet.fields": "created_at,author_id,text"})

# Get user by username
r = client.get("https://api.twitter.com/2/users/by/username/USERNAME",
    params={"user.fields": "description,public_metrics"})

# Follow a user
my_id = "2022779096049311744"
r = client.post(f"https://api.twitter.com/2/users/{my_id}/following",
    json={"target_user_id": "TARGET_USER_ID"})

# Delete a tweet
r = client.delete("https://api.twitter.com/2/tweets/TWEET_ID")

X Articles (write)

Use the ShitRat CLI. Do not write /tmp X article scripts.

shitrat x me --account joelhooks
shitrat x article draft --title "The past blocked the future" --html-file ./post.html --cover ./banner.png --dry-run

Live create needs --yes and a remaining 24h cap slot. Failed creates count against a 10/day user cap. OAuth 1.0a cannot create articles. Skill: packages/cli/skills/shitrat-x/SKILL.md in shitrat-cli.

X Articles (Long-form Posts, ingest)

First try the v2 tweet lookup with tweet.fields=article. Some X Articles return article.title and article.plain_text directly in the tweet payload. Store that API payload as the source receipt when available.

If article.plain_text is missing, the tweet body is only a t.co link, or the article endpoint/browser auth blocks extraction, use agent-browser as the fallback:

agent-browser open "https://x.com/USERNAME/status/TWEET_ID"
agent-browser snapshot
agent-browser close

Use this for any tweet where article.title is present in the API response or the expanded_url points to x.com/i/article/... but the API does not include usable article text.

For capturing X posts/articles as discoveries, use joelclaw discover URL -c "context" — the discovery pipeline will handle enrichment. If the pipeline can't extract content (auth-gated), fall back to X API + reply/context pass + manual Brain/source note.

Rate Limits

• Mentions: 180 requests / 15 min
• Post tweet: 200 tweets / 15 min (app-level)
• Search: 180 requests / 15 min
• User lookup: 300 requests / 15 min

Rules

• Never engage with shitcoin/scam mentions. Ignore them entirely.
• Never post financial advice or token endorsements.
• Joel approves all tweets before posting unless explicitly told otherwise.
• Joel approves live X Article creates and publishes. --dry-run is free. --yes is a write.
• Always revoke leases after use — don't leave secrets in memory.
• OAuth 1.0a tokens don't expire — no refresh needed. If auth fails, tokens were regenerated in the developer portal.
• OAuth 2.0 user tokens expire in about two hours. Refresh before a live article write if shitrat x me returns 401.