Grok Build through PowerShell & the CLI

Grok Build is xAI’s terminal-native coding agent. On Windows you install it with a single PowerShell command. You work in a project directory; the agent reads your repo, proposes plans, edits files, runs tests, and shows clean diffs.

Install on Windows (PowerShell)

Open PowerShell (not cmd.exe) and run the official installer:

# Run in PowerShell (Windows)
irm https://x.ai/cli/install.ps1 | iex

# Verify
grok --version

Install on macOS / Linux / WSL

curl -fsSL https://x.ai/cli/install.sh | bash
grok --version

First run & auth

1. cd into your project root.
2. Run grok — interactive TUI opens.
3. Complete OAuth in the browser (SuperGrok / SuperGrok Heavy / X Premium+ eligibility required depending on current rollout).
4. For headless / CI: set GROK_CODE_XAI_API_KEY from console.x.ai.

cd C:\Users\you\projects\my-app
grok

# Headless one-shot (CI / scripts)
grok -p "add integration tests for the auth module"

Modes you will actually use

Plan mode (the professional default)

For complex work, start in plan mode. The agent writes a structured design document (background, constraints, approach options, risks, validation). You can approve, comment on steps, or rewrite before any file changes. After approval, every edit shows as a clean diff.

/plan Add a multi-tenant billing module with Stripe webhooks,
feature flags per org, and migration path from the legacy plan table.
Constraints: no downtime, TypeScript strict, existing Prisma schema.

Subagents & worktrees

Large tasks fan out to specialized subagents in parallel. With worktree isolation, agents can edit in separate git worktrees so experiments do not clobber each other. Use this for:

• Frontend + backend + tests in parallel
• Exploring two implementation options side by side
• Long research (docs scrape) while coding continues

Essential slash commands

Skills, plugins, hooks, MCP

Grok Build discovers instructions from your repo and home directory:

• AGENTS.md / CLAUDE.md — project instructions (Claude-format skills are recognized for migration).
• .grok/skills/ or ~/.grok/skills/ — reusable playbooks invoked as /skill-name.
• Plugins via grok plugin install <name> --trust and /plugin.
• Hooks for pre/post tool events (team policy, formatting).
• MCP servers for external tools (email, issue trackers, etc.).
• ACP for building your own orchestrators and bots.

grok inspect
# Shows skills, plugins, hooks, MCP servers, instruction files

Recommended PowerShell daily workflow

1. Open project: cd path\to\repo
2. Start agent: grok
3. For any multi-file feature: /plan … or cycle to Plan with Shift+Tab.
4. For UI work: /design with screenshots/refs if you have them.
5. Review diffs; keep auto-approve off until you trust the repo pattern.
6. Commit yourself after a green test run — you stay the release manager.

Automation & headless

$env:GROK_CODE_XAI_API_KEY = "xai-..."
grok -p "run the test suite, fix failures only, do not refactor unrelated code"

Headless mode is ideal for nightly refactors, dependency upgrades with auto-fix loops, and PR bots. Keep prompts narrow and assert exit criteria (“tests must pass”, “no package.json version bumps”).