Call Grok Build Directly from Codex

Where this fits

You're in a Codex session, and the user says "Ask Grok Build to research this thread, then summarize." You need a reliable bridge between Codex and the locally installed Grok CLI without inventing a fragile shell workflow. grok-build-cli is that bridge: it checks auth, chooses the right invocation, and returns Grok's output so Codex can continue the task. It delegates the prompt; it does not prescribe what the prompt should be.

Why agents benefit

• Explicit availability check: command -v grok and grok models catch missing login before wasting a turn.
• Two call forms tuned for the request: --single for direct answers, --prompt-file with --max-turns for agentic tasks that may need X/web search.
• Process monitoring built into the skill, so Codex can wait, observe, and capture output instead of fire-and-forget.
• Local execution keeps the flow inside the user's existing toolchain; no cloud proxy or API key to manage.
• The skill scopes itself only to invocation mechanics, making it composable with research and analysis skills.

Practical scenarios

Cross-model sanity check

User has a Codex-generated plan and wants a second opinion from Grok. Codex calls grok --single "Review this plan for edge cases" --output-format plain, captures the reply, and merges both perspectives into the final recommendation.

Delegating a web/X research task

A user asks Codex to "have Grok Build find recent announcements from @OpenAI on X and summarize them." The skill writes the full prompt to a UTF-8 file and calls grok --prompt-file ... --max-turns 8, letting Grok use its own tools while Codex monitors the process.

Authenticated environment setup

During a fresh setup, Codex checks availability, finds no auth, and asks the user if they want to run grok login interactively. No blind execution, no hidden assumptions.

Add it to your agent workflow

Install from the registry:

npx skills add Pluviobyte/rnskill --skill grok-build-cli

Then delegate a prompt naturally:

grok --single 'Explain the tradeoffs of this data model in 3 bullets' --output-format plain

Compare before adopting

Before adding a Grok bridge, evaluate how each option behaves when Grok is not installed or logged in. Check whether the skill was recently updated and whether the author responds to issues. Compare against a plain grok subprocess inside your own Codex instructions, and weigh whether a maintained skill is worth the dependency. Also assess fit: does the skill's narrow invocation-only scope align with your workflow, or do you need a tool that also plans the prompt?

Why it is worth tracking

Pluviobyte/rnskill has 1,421 stars and a focused README, which suggests real usage rather than an abandoned experiment. The skill matters because it models a small, reusable pattern: isolate external CLI invocation in a skill, keep the task prompt separate, and let the agent orchestrate. It is worth evaluating now if you already use Grok Build and Codex side by side. More context is on the skill page: https://www.openagentskill.com/skills/pluviobyte-grok-build-cli.