# grok-cli — Terminal Coding Agent for Grok API

> grok-cli is an open-source terminal coding agent for the Grok API with interactive and headless modes, so you can script tasks or drive a full TUI loop.

## Install

Copy the content below into your project:

## Quick Use

1. Install:
   ```bash
   curl -fsSL https://raw.githubusercontent.com/superagent-ai/grok-cli/main/install.sh | bash
   ```
2. Run interactive mode:
   ```bash
   grok
   ```
3. Headless one-shot prompt:
   ```bash
   grok --prompt "run the tests and summarize failures"
   ```

## Intro

grok-cli is an open-source terminal coding agent for the Grok API with interactive and headless modes, so you can script tasks or drive a full TUI loop.

- **Best for:** developers who want a Grok-powered coding agent with both TUI and scriptable headless runs
- **Works with:** curl/bash installer; requires a Grok API key for real use; runs in a project directory
- **Setup time:** 10–20 minutes

## Practical Notes

- Use headless `--prompt` for automation; use TUI for longer sessions.
- Keep prompts task-scoped (1 request = 1 outcome) for predictable runs.

## Two modes: interactive vs headless

### Interactive (TUI)

Use interactive mode when you want multi-step reasoning, file browsing, and iterative edits:

```bash
grok
```

### Headless (one-shot)

Use headless mode in scripts or CI:

```bash
grok --prompt "explain failing tests and propose a minimal fix"
```

## Practical guardrails

- Start by asking for a plan and a dry-run command list (no destructive actions).
- For code changes: request a diff-only proposal, then run tests.
- Keep secrets out of pasted logs; prefer env vars or config files.

## Suggested use cases

- “Run tests and summarize failures”
- “Scan the repo and propose refactors”
- “Draft a changelog from recent commits”

Treat it like a programmable teammate: deterministic inputs → reproducible outputs.

### FAQ

**Q: Do I need the Grok API key to install?**
A: No. Installation is separate; you need an API key to actually use the agent.

**Q: Is headless mode good for CI?**
A: Yes. It’s the most repeatable mode for scripted runs.

**Q: How do I keep results stable?**
A: Keep prompts narrow, ask for diffs, and always run verification commands.

## Source & Thanks

> Source: https://github.com/superagent-ai/grok-cli
> License: MIT
> GitHub stars: 3,027 · forks: 373

---

<!-- ZH -->

## 快速使用

1. 安装：
   ```bash
   curl -fsSL https://raw.githubusercontent.com/superagent-ai/grok-cli/main/install.sh | bash
   ```
2. 交互模式：
   ```bash
   grok
   ```
3. Headless 一次性 prompt：
   ```bash
   grok --prompt "run the tests and summarize failures"
   ```

## 简介

grok-cli 是面向 Grok API 的开源终端编码 agent：支持交互式 TUI 与 headless 模式，既能写脚本跑任务，也能做完整的终端 agent 循环并保存会话轨迹与产物。

- **适合谁：** 想用 Grok 做终端编码 agent，同时需要交互 TUI 与脚本化 headless 的开发者
- **可搭配：** curl/bash 安装；实际使用需要 Grok API key；在项目目录中运行
- **准备时间：** 10–20 分钟

## 实战建议

- 自动化优先用 headless `--prompt`；长任务用 TUI 会话更合适。
- Prompt 尽量任务化（一次请求一个结果），更可控。 

## 两种模式：交互 vs headless

### 交互模式（TUI）

适合多轮推理、浏览文件、反复迭代修改：

```bash
grok
```

### Headless（一次性）

适合脚本或 CI：

```bash
grok --prompt "explain failing tests and propose a minimal fix"
```

## 实战护栏（强烈建议）

- 先让它输出计划与要跑的命令清单（避免误操作）。  
- 改代码时：先要“只给 diff”，再跑测试验证。  
- 日志里不要粘贴敏感信息；用环境变量或配置文件传递。

## 典型用例

- “跑测试并总结失败原因”  
- “扫描仓库并给出重构建议”  
- “根据最近提交草拟 changelog”

把它当作“可编程队友”：输入可控 → 输出可复现。

### FAQ

**安装需要 Grok API key 吗？**
答：不需要。安装和使用分开；真正运行 agent 需要 API key。

**Headless 适合 CI 吗？**
答：适合。它是最可复现、最脚本化的模式。

**如何让结果更稳定？**
答：Prompt 要窄、先给 diff、并始终跑验证命令。

## 来源与感谢

> Source: https://github.com/superagent-ai/grok-cli
> License: MIT
> GitHub stars: 3,027 · forks: 373


---
Source: https://tokrepo.com/en/workflows/grok-cli-terminal-coding-agent-for-grok-api
Author: Agent Toolkit