grok-first

Delegates scoped implementation and review tasks to Grok CLI from Claude Code sessions.

4|Updated Jun 2, 2026
One-click install
npx skills add https://github.com/sukbearai/king-ai --skill grok-first-sukbearai
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: grok-first
Source: https://github.com/sukbearai/king-ai/tree/main/packages/skills/skills/grok-first
Command: npx skills add https://github.com/sukbearai/king-ai --skill grok-first-sukbearai

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Claude Code sessions often spend expensive context on mechanical implementation, bulk exploration, or second-opinion analysis that another model could handle. This Skill routes that scoped work to Grok CLI in headless mode while Claude retains design, specification, and verification ownership. ## Core Features & Use Cases - Scoped Delegation: Sends frozen specs for implementation, refactors, bug fixes, test writing, and CI fixes to Grok CLI with explicit working directory, model, and output controls. - Alternate-Model Review: Obtains second-opinion debugging, code review, or risk analysis from Grok while Claude keeps final judgment. - Session Continuation: Uses --continue or --resume flags for cheap follow-up corrections against stored Grok sessions instead of fresh runs. - Use Case: You have a well-defined refactor across a repo. Freeze the spec in Claude, delegate the mechanical edits to Grok headless mode, then review the diff and run tests yourself before shipping. ## Quick Start Ask Claude to use the grok-first skill to delegate a scoped implementation task to Grok CLI with a frozen spec and then verify the resulting diff.

Frequently Asked Questions about grok-first

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I delegate a coding task to Grok CLI from Claude Code?▼

Write the prompt to a temp file containing the goal, repo paths, constraints, and expected proof, then run grok with -p, --cwd, --output-format plain, and --always-approve. Read the saved output file and verify the diff yourself.

What tasks should be delegated to Grok versus kept in Claude?▼

Delegate implementation from frozen specs, refactors, test writing, CI fixes, and bulk exploration. Keep design, architecture, naming, tiny edits under 20 lines, destructive operations, and all review of Grok output in Claude.

Can I use this skill inside a Grok or Codex session?▼

No. The skill includes a session gate that restricts it to Claude Code only. It must never be applied in Grok, Codex, or other harnesses, and Grok must never be invoked from a Grok session.

How do I request follow-up fixes on a Grok delegation?▼

Use the --continue flag to resume the most recent Grok session in the current directory, or --resume with a specific session ID. Send a focused correction prompt with exact files and failing proof rather than starting a fresh run.

Why does Grok delegation fail or produce wrong results?▼

The external Grok process starts with zero Claude session context, so vague prompts cause failures. Include the goal, exact repo paths, constraints, non-goals, expected proof, and output shape in every prompt, and take over directly after two failed rounds.