How do I install the Grok Build CLI skill?

Run npx skillmds@latest add pluviobyte/grok-build-cli in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Grok Build CLI safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Grok Build CLI?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Grok Build CLI free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Grok Build CLI?

pluviobyte (@pluviobyte) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Grok Build CLI by pluviobyte — Agent Skill

Name: Grok Build CLI
Author: pluviobyte

Grok Build CLI

Invoke the locally installed Grok Build CLI from Codex and return Grok's response. Use whenever the user asks Codex to call, consult, run, or delegate a prompt to Grok or Grok Build, including Grok web/X searches and agentic Grok tasks. This Skill covers only the invocation mechanism, process monitoring, and response capture; it does not prescribe the task itself.

pluviobyte cd876f6 2 files · 3.2 KB Updated Sep 17, 2026

File contents

Call Grok Build

Call the local grok CLI from Codex. Do not add an unrelated research, download, editing, or production workflow.

Check availability

Before the first call in a task, run:

command -v grok
grok models

grok models must confirm that the user is logged in and show at least one available model. If authentication is missing, report it and run grok login only when the user wants to authenticate interactively.

Choose the call form

For a direct single-turn answer that does not need Grok tools:

grok --single 'PROMPT' --output-format plain

For an agentic task that may use Grok's tools, including X or web search, put the full prompt in a UTF-8 file and run:

grok --prompt-file '/absolute/path/prompt.md' \
  --max-turns 8 \
  --output-format plain

Use --model MODEL only when the user requests a particular available model. Otherwise use Grok's configured default.

Run from Codex

For a long prompt, create prompt.md with apply_patch; avoid fragile shell quoting.
Start Grok with exec_command, normally using yield_time_ms: 1000.
If the command returns a session ID, poll it with write_stdin, sending an empty string and waiting at most 30 seconds per poll.
Share a concise progress update when a call runs longer than 60 seconds.
Capture Grok's final stdout. Treat a nonzero exit code, Max turns reached, or empty final output as failure.
Return Grok's answer to the user, clearly identified as Grok's result when attribution matters.

Isolation

When Grok should only execute the supplied prompt and must not inspect the current repository's skills or instructions, give it an empty temporary working directory:

GROK_CALL_DIR="$(mktemp -d)"
grok --cwd "$GROK_CALL_DIR" --no-memory --no-subagents \
  --prompt-file '/absolute/path/prompt.md' \
  --max-turns 8 \
  --output-format plain

Use this isolation by default for web/X research and independent second opinions. Use the actual project directory only when Grok must read or modify that project.

Guardrails

Do not claim Grok was called unless the CLI actually ran.
Do not fabricate, paraphrase as successful, or silently replace an empty or failed Grok response.
Do not pass secrets in prompts or command-line arguments.
Do not use --always-approve, bypassPermissions, or broad write access unless the user explicitly authorizes the corresponding mutations.
Do not resume an unrelated prior Grok session. Start a new call unless the user explicitly asks to continue a specific Grok session.
Keep this Skill task-agnostic: the user's prompt determines what Grok does.

---name: grok-build-clidescription: Invoke the locally installed Grok Build CLI from Codex and return Grok's response. Use whenever the user asks Codex to call, consult, run, or delegate a prompt to Grok or Grok Build, including Grok web/X searches and agentic Grok tasks. This Skill covers only the invocation mechanism, process monitoring, and response capture; it does not prescribe the task itself.---# Call Grok BuildCall the local `grok` CLI from Codex. Do not add an unrelated research,download, editing, or production workflow.## Check availabilityBefore the first call in a task, run:```bashcommand -v grokgrok models````grok models` must confirm that the user is logged in and show at least oneavailable model. If authentication is missing, report it and run `grok login`only when the user wants to authenticate interactively.## Choose the call formFor a direct single-turn answer that does not need Grok tools:```bashgrok --single 'PROMPT' --output-format plain```For an agentic task that may use Grok's tools, including X or web search, putthe full prompt in a UTF-8 file and run:```bashgrok --prompt-file '/absolute/path/prompt.md' \ --max-turns 8 \ --output-format plain```Use `--model MODEL` only when the user requests a particular available model.Otherwise use Grok's configured default.## Run from Codex1. For a long prompt, create `prompt.md` with `apply_patch`; avoid fragile shell quoting.2. Start Grok with `exec_command`, normally using `yield_time_ms: 1000`.3. If the command returns a session ID, poll it with `write_stdin`, sending an empty string and waiting at most 30 seconds per poll.4. Share a concise progress update when a call runs longer than 60 seconds.5. Capture Grok's final stdout. Treat a nonzero exit code, `Max turns reached`, or empty final output as failure.6. Return Grok's answer to the user, clearly identified as Grok's result when attribution matters.## IsolationWhen Grok should only execute the supplied prompt and must not inspect thecurrent repository's skills or instructions, give it an empty temporaryworking directory:```bashGROK_CALL_DIR="$(mktemp -d)"grok --cwd "$GROK_CALL_DIR" --no-memory --no-subagents \ --prompt-file '/absolute/path/prompt.md' \ --max-turns 8 \ --output-format plain```Use this isolation by default for web/X research and independent secondopinions. Use the actual project directory only when Grok must read or modifythat project.## Guardrails- Do not claim Grok was called unless the CLI actually ran.- Do not fabricate, paraphrase as successful, or silently replace an empty or failed Grok response.- Do not pass secrets in prompts or command-line arguments.- Do not use `--always-approve`, `bypassPermissions`, or broad write access unless the user explicitly authorizes the corresponding mutations.- Do not resume an unrelated prior Grok session. Start a new call unless the user explicitly asks to continue a specific Grok session.- Keep this Skill task-agnostic: the user's prompt determines what Grok does.

Frequently asked questions