convez.ai
Skill
gitguffaw

grok-claude-skill

gitguffaw/grok-claude-skill

Ver no GitHub
InicialPluginScore · de 0 a 100

A Claude Code skill that drives the local Grok CLI (xAI) as an inner coding/analysis agent via headless grok -p

Instalação

$npx convez-ai gitguffaw/grok-claude-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
1
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 4 semanas
Versão
v1.0.8
Issues abertas
1
Contribuidores
1

README

A tradução para o português deste README ainda está na fila.

Grok skill for Claude Code

A Claude Code skill that drives the local Grok CLI (xAI) as Claude's inner coding and analysis agent. When Claude should delegate work to Grok instead of relying only on its own tools, this skill teaches it to route everything through headless grok -p with explicit control over model, effort, web search, tool scope, permissions, sessions, subagents/personas, best-of-N, and self-check.

What it does

Outer Claude treats Grok as a second agent. The skill encodes:

  • Spine: every delegation goes through headless grok -p (the one-shot agent run). grok agent (ACP / stdio / serve) is for IDE integration, not delegation, and the interactive TUI is left for the user — Claude does not drive it.
  • Modes: Analyze (read-only investigation), Exec (bounded change), Review (findings-only), Parallel (--best-of-n + subagents/personas), Session (multi-turn / machine-readable).
  • Modifiers: model (-m), effort (--effort low|medium|high|xhigh|max), output format (--output-format json|streaming-json), permissions (--always-approve / --permission-mode / --allow/--deny / --sandbox), tool scope (--tools / --disallowed-tools), self-check (--check), plan mode, worktree isolation, and rules.
  • Guardrails: the installed binary is treated as the source of truth over the bundled docs (verified drift is documented), models are resolved live with grok models (never pinned), and grok inspect is used to discover the real capability surface.

Requirements

  • The grok CLI installed and authenticated (grok login, or XAI_API_KEY for headless/CI). See https://x.ai/cli.
  • Claude Code — the skill is loaded and invoked by Claude, not run directly.

Verified against grok 0.2.39.

Install

Clone into your Claude skills directory as Grok:

git clone https://github.com/gitguffaw/grok-claude-skill ~/.claude/skills/Grok

Claude Code discovers it on the next launch. (Grok itself also reads ~/.claude/skills, so the same folder doubles as a Grok skill.)

Layout

SKILL.md                      # routing: modes, modifiers, launch rule, hard rules
references/QuickRef.md         # flag-by-flag CLI surface
references/LaunchPatterns.md   # copy-paste launch lines per mode/modifier
Workflows/                     # Exec, Analyze, Review, Parallel, Session

License

Apache License 2.0.

Ver o original no GitHub

Bom para

Relacionados