fatby/code-runner

Code Runner Skill

Execute shell commands and scripts on the user's local machine.

Usage

{"thought": "...", "tool": "runCmd", "args": {"command": "..."}}

Examples

Simple Command

User: "查看当前目录有哪些文件"

{"thought": "用户想查看当前目录的文件列表，使用 dir 命令（Windows 系统）", "tool": "runCmd", "args": {"command": "dir /b"}}

Multi-step Task

User: "帮我创建一个 Python 虚拟环境并安装 requests 库" Step 1:

{"thought": "需要先创建虚拟环境，使用 python -m venv 命令", "tool": "runCmd", "args": {"command": "python -m venv myenv"}}

Step 2 (after success):

{"thought": "虚拟环境创建成功，现在激活并安装 requests。Windows 下用 myenv\\Scripts\\pip", "tool": "runCmd", "args": {"command": "myenv\\Scripts\\pip install requests"}}

Step 3 (verify):

{"thought": "安装完成，验证一下 requests 是否正确安装", "tool": "runCmd", "args": {"command": "myenv\\Scripts\\pip list | findstr requests"}}

Diagnostic Task

User: "看看端口 3001 被谁占用了"

{"thought": "需要查看端口占用情况，使用 netstat 命令过滤 3001 端口", "tool": "runCmd", "args": {"command": "netstat -ano | findstr 3001"}}

Error Recovery

If previous command failed with "python not found":

{"thought": "python 命令未找到，可能系统用的是 python3 或完整路径。先检查 Python 安装位置。", "tool": "runCmd", "args": {"command": "where python"}}

Safety Rules

1. Never execute destructive commands (rm -rf, format, etc.) without explicit user confirmation
2. Avoid commands that modify system settings
3. Prefer read-only commands for diagnostics
4. Always explain what a command will do in the thought field

Notes

• Commands run in project directory
• Output is captured and returned
• Timeout: 60 seconds
• This is a Windows system, use Windows commands (dir, findstr, type, etc.)