| name | grok-ask |
| description | Ask xAI Grok a one-off question with optional live web search. Use when the user wants Grok's opinion or a quick answer that benefits from Grok's distinct training (recent web/X awareness, code-focused models). Read-only by default. Returns Grok's stdout verbatim. |
Grok — Ask
Use this skill when the user asks Grok a question directly, or when you need a second opinion / cross-check from a different model than the one you're running in. The companion script wraps the local grok CLI with environment scrubbing, ANSI sanitization, capability probing, and prompt-injection defenses.
When to use
- The user explicitly asks "ask grok…", "what would grok say about…", "double-check this with grok".
- You want a second perspective from a model with different training (xAI vs OpenAI/Anthropic).
- The question benefits from Grok's strengths: live web/X search awareness, code-focused models, current events.
Do not use this skill for:
- Long-form rescue / multi-step delegation → use the
grok-rescue skill instead.
- Image generation → use
grok-imagine.
- Deep research that benefits from
--effort max + --check self-verification → use grok-research.
How to invoke
Run exactly one bash command:
node "${CLAUDE_PLUGIN_ROOT}/scripts/companion.mjs" ask "<the user's question, properly quoted>"
The CLAUDE_PLUGIN_ROOT env var is set by both Claude Code and Codex CLI. The companion accepts the full argument string and parses its own flags.
Common flags (append to the argument string before the question)
--model <name> — override default Grok model. Available models (run /grok:models): grok-build (default), grok-composer-2.5-fast.--effort <low|medium|high|xhigh|max> — reasoning effort. Note: the current default model (grok-build) declares no reasoning-effort support, so the companion silently strips --effort (with a stderr warning) to avoid a 400 from upstream. It only takes effect on a model that advertises supports_reasoning_effort.--timeout <duration> — e.g., 90s, 5m, 0 to disable. Default unbounded.--no-web-search — disable Grok's live web search (on by default).--check — append Grok's self-verification loop to the answer.--rules <text|@file> — extra system rules.--sandbox <profile> — restrict tool/filesystem access.
Output rules
- Return the companion's stdout verbatim. Do not paraphrase, summarize, or wrap in additional commentary.
- If stderr contains a
[hint: ...] line about authentication, surface it and tell the user to run grok login or set XAI_API_KEY.
- If the user's request is ambiguous about what to ask Grok, clarify with the user before invoking — do not ask Grok yourself.
