API Insights

Twitter/X GraphQL API

UserByScreenName endpoint uses rotating query IDs — must discover from JS bundles, never hardcode
The featureSwitches array near the query definition in bundles can be extracted and all set to true
IMPORTANT: Newer API responses have moved screen_name OUT of the legacy object. Don't rely on legacy.screen_name existing. Use the screen name you already know from the request parameter instead.
followers_count is still in legacy but always use fallback deep search
description (bio) is in legacy.description or profile_bio.description
Bearer token is public/shared across all Twitter web clients
CSRF token comes from ct0 cookie
Rate limit: ~1000 requests per 15-minute window; use escalating delays + 429 backoff
Use origFetch (saved before overriding window.fetch) for own API calls to avoid infinite loops

@run-at document-start + fetch/XHR interception to capture Twitter's own API responses
Serialized queue with escalating delays (1-3s base, +1s per consecutive request, capped at 10-12s)
Queue cancellation on DOM changes — only fetches currently visible profiles
userCache Map stores { followers, bio, ts } per lowercase handle
Persistent cache in localStorage with 7-day TTL
MutationObserver triggers processTweets() on DOM changes
scheduleReprocess() debounces re-processing when new cache entries arrive

Tweets: article[data-testid="tweet"]
User name: [data-testid="User-Name"], Avatar: [data-testid^="UserAvatar-Container-"]
Three-dot menu: [data-testid="caret"]
Back button (soft nav): [data-testid="app-bar-back"]
Retweet button: [data-testid="retweet"] / [data-testid="unretweet"]
Dropdown menus: [role="menu"] → [role="menuitem"]
Tabs: [role="tablist"] → [role="tab"]
j/k focused tweet: walk document.activeElement up to article[data-testid="tweet"]