Skip to main content

x-intel

Read X (Twitter) posts, search, timelines and user profiles via the official xurl CLI.

ソース情報

リポジトリ
mateaix/mateclaw
ソースの最終更新活動
2026年5月11日 09:22
検出された SKILL.md の言語
英語
スター
1,162
フォーク
346

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

SKILL.md を表示中

SKILL.md
ソースの指示 · 読み取り専用プレビュー
name
x_intel
description
Read X (Twitter) posts, search, timelines and user profiles via the official xurl CLI.
nameZh
X 情报采集
nameEn
X Intel
version
1.0.0
icon
🐦
author
MateClaw
optional
true
tags
["x","twitter","social-media","research","xurl"]
platforms
["linux","macos"]
dependencies
{"commands":["xurl"],"tools":["execute_shell_command"]}
# x_intel — X (Twitter) information gathering `x_intel` lets an agent pull posts, search results, timelines and user profiles from X (Twitter) through `xurl`, the X developer platform's official CLI. **This skill is read-only by design** — it intentionally omits posting, replying, deleting, DM-sending and any other write surface. For a separate publishing skill, see follow-up work. Use this skill for: - looking up a single post by ID or URL - searching posts with the X search query syntax (`from:user`, `lang:en`, `#hashtag`, ...) - reading the agent operator's home timeline, mentions, bookmarks, likes - inspecting a user profile by handle - walking the social graph (who someone follows / is followed by) - raw read access to any X API v2 GET endpoint when the shortcuts don't fit --- ## Credential safety (mandatory) Critical rules when invoked inside an agent session: - **Never** read, print, parse, summarize, upload or quote `~/.xurl` into chat context. It is a YAML token store. - **Never** ask the user to paste credentials/tokens into the conversation. - **Never** suggest or run the auth commands with inline secrets in an agent session. - **Never** pass `--verbose` / `-v` — it prints auth headers to stdout. - The only credential-touching command this skill ever runs is `xurl auth status` (status only, no secrets). Forbidden flags in any agent-issued command (each accepts inline secrets): `--bearer-token`, `--consumer-key`, `--consumer-secret`, `--access-token`, `--token-secret`, `--client-id`, `--client-secret`. App registration and the OAuth 2.0 PKCE flow must be performed by the user **outside** the agent session (see "User setup" below). Tokens persist in `~/.xurl` (YAML); OAuth 2.0 refreshes automatically. --- ## Install The agent should verify, not install. Direct the user to install if missing. ```bash # Shell script (Linux + macOS, installs to ~/.local/bin, no sudo) curl -fsSL https://raw.githubusercontent.com/xdevplatform/xurl/main/install.sh | bash # Homebrew (macOS) brew install --cask xdevplatform/tap/xurl # Go (cross-platform) go install github.com/xdevplatform/xurl@latest ``` Verify: ```bash xurl --help xurl auth status ``` --- ## User setup (user runs these, NOT the agent) The agent must not perform these steps — they involve pasting secrets. Direct the user to this section verbatim. 1. Open the X developer dashboard: <https://developer.x.com/en/portal/dashboard> 2. In the app's User Authentication Settings, set the redirect URI to `http://localhost:8080/callback` and the app type to **Web app, automated app or bot**. 3. Copy the app's Client ID and Client Secret. 4. Register the app locally: ```bash xurl auth apps add my-app --client-id YOUR_CLIENT_ID --client-secret YOUR_CLIENT_SECRET ``` 5. Authenticate (this opens a browser for OAuth 2.0 PKCE): ```bash xurl auth oauth2 --app my-app ``` If X returns `UsernameNotFound` or a 403 on the post-OAuth `/2/users/me` lookup, pass the handle explicitly (xurl v1.1.0+): ```bash xurl auth oauth2 --app my-app YOUR_HANDLE ``` 6. Mark this app as the default so all commands use it: ```bash xurl auth default my-app ``` 7. Verify: ```bash xurl auth status xurl whoami ``` > **Most common mistake:** omitting `--app my-app` from `xurl auth oauth2`. The OAuth token then lands in the built-in `default` profile, which has no client-id/client-secret, and every later read fails. Re-run `xurl auth oauth2 --app my-app` and `xurl auth default my-app` to fix. --- ## Read-only command reference All commands return JSON to stdout. The agent parses JSON directly; no extra tooling needed. | Action | Command | | --- | --- | | Who is the bound account | `xurl whoami` | | Look up a user | `xurl user @handle` | | Read one post (ID or URL) | `xurl read POST_ID` | | Search posts | `xurl search "QUERY" -n 10` | | Home timeline | `xurl timeline -n 20` | | Mentions of bound account | `xurl mentions -n 20` | | Bookmarks list | `xurl bookmarks -n 20` | | Likes list | `xurl likes -n 20` | | Following list | `xurl following -n 50` | | Followers list | `xurl followers -n 50` | | Another user's graph | `xurl following --of HANDLE -n 20` | | Auth status | `xurl auth status` | Notes: - `POST_ID` accepts a full `https://x.com/user/status/...` URL — xurl extracts the ID. - Handles work with or without the leading `@`. ### Search query language X's search supports operators inside the quoted query string: ```bash xurl search "from:elonmusk -is:retweet" -n 20 xurl search "#buildinpublic lang:en since:2026-01-01" -n 25 xurl search "OR" -n 10 # literal OR — must be quoted xurl search "(rust OR go) lang:en" -n 10 xurl search "to:NASA -is:reply" -n 10 ``` Common operators: `from:`, `to:`, `@`, `#`, `is:retweet`, `is:reply`, `is:quote`, `lang:`, `since:`, `until:`, `has:media`, `has:links`. See the X search syntax docs for the full list. --- ## Raw v2 read access For anything beyond the shortcuts, hit any v2 GET endpoint directly: ```bash # Public user fields xurl /2/users/by/username/elonmusk?user.fields=public_metrics,description,verified # Single tweet with metrics + author expansion xurl /2/tweets/1234567890?tweet.fields=public_metrics,created_at&expansions=author_id # Recent search with extra fields (paid tier) xurl /2/tweets/search/recent?query=langchain&tweet.fields=created_at,public_metrics&max_results=25 # Full URLs also work xurl https://api.x.com/2/users/me ``` Streaming endpoints are auto-detected; force with `-s` if needed. **Streaming endpoints can be expensive — do not start one without confirming intent with the user.** --- ## Common workflows ### Profile a user ```bash xurl user @handle xurl /2/users/by/username/handle?user.fields=public_metrics,description,verified,created_at xurl following --of handle -n 20 # who they pay attention to ``` ### Triage a trending term ```bash xurl search "topic lang:en -is:retweet" -n 25 # Pick interesting IDs from the JSON, then drill in: xurl read 1234567890 xurl user @ORIGINAL_POSTER ``` ### Catch up on activity ```bash xurl whoami xurl mentions -n 20 xurl timeline -n 20 xurl bookmarks -n 10 ``` ### Conversation context ```bash xurl read https://x.com/user/status/1234567890 # Conversation expansion via raw v2 xurl /2/tweets/search/recent?query=conversation_id:1234567890&max_results=25 ``` --- ## Output format Every command emits X API v2 shape JSON to stdout: ```json { "data": { "id": "1234567890", "text": "Hello world!" }, "includes": { "users": [{ "id": "...", "username": "..." }] } } ``` Errors are also JSON: ```json { "errors": [ { "message": "Not authorized", "code": 403 } ] } ``` The non-zero exit code distinguishes errors from empty results. --- ## Agent workflow 1. Verify prerequisites: `xurl --help` (the command exists) and `xurl auth status` (the user has at least one app with `oauth2` tokens, marked `▸` as default). 2. **Parse `auth status` output before any other command.** If the default app shows `oauth2: (none)` but a non-default app has valid tokens, instruct the user to run `xurl auth default <that-app>` — this is the most common config glitch and does not require a re-login. 3. If `auth status` shows no apps or no tokens, **stop**. Tell the user to follow the "User setup" section. Do not attempt to register apps or run any auth flow yourself. 4. Start with the cheapest read first (`xurl whoami` / `xurl user @handle` / `xurl search ... -n 3`) to confirm reachability and the request shape. 5. Treat 401 / 403 / 429 distinctly: 401 → re-auth needed, 403 → scope or plan, 429 → wait and retry (X rate-limits per-endpoint). 6. Never paste `~/.xurl` content back into the conversation, even when troubleshooting. 7. When in doubt about cost: X's API has paid tiers and per-endpoint rate limits. Do not run unbounded loops or streams without the user's explicit confirmation. --- ## Troubleshooting | Symptom | Cause | Fix | | --- | --- | --- | | `auth status` shows `oauth2: (none)` on default | Token saved to built-in `default` profile (no client-id/secret) | Re-run `xurl auth oauth2 --app my-app` then `xurl auth default my-app` | | `unauthorized_client` during OAuth | App type set to "Native App" in X dashboard | Change to "Web app, automated app or bot" | | `UsernameNotFound` / 403 right after OAuth | X not returning username from `/2/users/me` | `xurl auth oauth2 --app my-app YOUR_HANDLE` (xurl v1.1.0+) | | 401 on every read | Token expired or wrong default app | Check `xurl auth status` — verify `▸` points to the app with oauth2 tokens | | `client-forbidden` / `client-not-enrolled` | X platform enrollment | Developer dashboard → Apps → Manage → Production environment | | `CreditsDepleted` | $0 balance on X API | Buy credits in Developer Console → Billing | | 429 on search/timeline | Hit per-endpoint rate limit | Pause, retry with smaller `-n`, or wait for the reset window | --- ## Notes - **Cost:** X API access is paid for meaningful usage. Many failures are plan or rate-limit problems, not skill problems. - **Scopes:** OAuth 2.0 tokens use broad scopes; a 403 on a specific read usually means the token is missing a scope — have the user re-run `xurl auth oauth2`. - **Token refresh:** OAuth 2.0 tokens auto-refresh; nothing to do. - **Multiple apps:** `xurl --app NAME ...` runs one read against a specific app without changing the default. - **Token storage:** `~/.xurl` is YAML. Treat it like a private key. Never read or send it to LLM context. --- ## Attribution - Underlying CLI: <https://github.com/xdevplatform/xurl> (X developer platform). - This skill wraps the CLI's read commands and documents agent-side safety rules. No code is shipped beyond this SKILL.md.
GitHubで見る