X API Skill

This skill is a set of guidelines and tools for interacting with X (Twitter) via the X API (https://console.x.com/) and docs (https://docs.x.com/overview). Use the suite of .md files in /references for more detailed information and examples.

Compatibility: requires curl and python3; expects network access to docs.x.com and api.x.com.

When to Apply

Reference these guidelines when:

• Performing any action on X (Twitter) that can be executed via API requests (post/reply/search, DMs, bookmarks, follows, auth, DMs, spaces, etc.)
• Understanding how to use the X (Twitter) API
• Setting up X (Twitter) API credentials (https://console.x.com/)
• Understanding the X (Twitter) API documentation (https://docs.x.com/overview)

Example prompts of when to apply this skill

• “Post this text: ...”
• “Reply to this post id ... with: ...”
• “Quote-post ... and say: ...”
• “Delete my post id ...”
• “Search recent posts for query and return the top 20 results”
• “Set up a filtered stream rule for query and show me how to read results”
• “Send a DM to @username saying ...”
• “Bookmark post id ...”
• “Follow @username”

What this skill needs from you (tokens)

This skill never asks you to paste secrets. It expects tokens via environment variables:

• Read public data: X_BEARER_TOKEN (app-only bearer token)
• Do anything as a user (post/reply/delete, bookmarks, DMs, follow/unfollow, etc.): X_USER_ACCESS_TOKEN (OAuth user context)

If an endpoint requires OAuth 1.0a (some legacy surfaces), the agent should call that out explicitly and avoid recommending it unless required.

Setup: create an app (onboarding)

If you haven’t created an app/credentials yet, start in the Developer Console:

• Create an app
• Generate credentials (Bearer Token for read-only; OAuth 2.0 Client ID/Secret for user actions)
• Allowlist callback URLs for OAuth (exact match; local dev should use http://127.0.0.1)

Reference: references/apps-and-credentials.md

Documentation index (for this skill’s sources)

To discover official X docs pages quickly, the agent can start from:

• https://docs.x.com/llms.txt

What the agent should output

When answering, prefer a small, concrete plan and then the exact request(s):

• Endpoint(s) (method + path)
• Auth mode (bearer vs user context) and required scopes when known
• curl command(s) using env vars (never inline tokens)
• Request shaping: fields / expansions only when needed
• Pagination: a stop condition (page cap, result cap, time bounds)
• Rate limits: how to behave on 429 using response headers
• Errors: how to interpret 401/403/429/5xx and what to fix vs retry

Important behavior rules

• Default to least-privileged: use bearer only for read-only public data; use user context for writes.
• Be explicit about side effects: for write actions, state what will happen and confirm before executing.
• Avoid bulk actions by default: if a user asks for bulk deletes/DMs/follows, require a cap and mention rate limits + compliance considerations.
• Use stable identifiers: treat IDs as strings (do not parse as numbers).
• Minimize follow-up calls: use expansions + *.fields to avoid N+1 patterns when it meaningfully reduces requests.

Automation & spam compliance

All usage must follow X’s Automation Rules (https://help.x.com/en/rules-and-policies/x-automation) and Developer Agreement (https://developer.x.com/en/developer-terms/agreement-and-policy). Key rules:

• Consent before outreach: never send automated replies or DMs without explicit opt-in from the recipient.
• Honor opt-outs immediately: if someone asks to stop being contacted, stop.
• No bulk/aggressive actions: no mass following, unfollowing, liking, or DMing. Always cap and pace automated actions.
• No duplicate content: never post identical or substantially similar content across multiple accounts.
• Bot disclosure: if operating a bot account, the profile bio must clearly state it is a bot and who is responsible for it.
• No trend manipulation: do not automatically post about trending topics or attempt to manipulate trends.
• No keyword-only replies: sending automated replies based on keyword searches alone (without prior opt-in) is prohibited.

Deeper references (optional)

If you need the detailed docs map or topic guides:

• references/getting-started.md
• references/docs-map.md
• references/workflows.md
• references/request-shapes.md
• references/fundamentals-auth.md
• references/apps-and-credentials.md