← 返回
开发者工具 Key 中文

Windows Execution Interface for OpenClaw

Windows automation via Clawdos API: screen capture, mouse/keyboard input, window management, file-system operations, and shell command execution. Standalone...
通过 Clawdos API 实现 Windows 自动化:屏幕捕获、鼠标/键盘输入、窗口管理、文件系统操作和 Shell 命令执行。独立版...
danzig233 danzig233 来源
开发者工具 clawhub v2.1.1 2 版本 100000 Key: 需要
★ 1
Stars
📥 886
下载
💾 53
安装
2
版本
#automation#latest#windows

概述

Clawdos

Overview

This skill exposes a CLI wrapper around the Clawdos REST API, allowing you to operate a Windows machine securely from OpenClaw via shell commands.

Instead of loading tools, use exec to call the standalone python script scripts/clawdos.py.

⚠️ SECURITY & SANDBOX MECHANISM

File System Sandbox Protection:

  • All file system operations (fs_list, fs_read, fs_write, fs_delete, fs_move) are restricted to a sandboxed root directory on the Windows host.
  • The sandbox root is configured via the CLAWDOS_FS_ROOT_ID environment variable and enforced server-side.
  • The Clawdos service prevents access to files outside the designated sandbox directory. Path traversal attempts (e.g., ../../../) are blocked.
  • This isolation ensures that skill operations cannot accidentally or intentionally access sensitive system files, user documents, or configuration outside the permitted scope.

Network Isolation:

  • The Clawdos service only communicates with the configured CLAWDOS_BASE_URL and does not establish unauthorized external connections.
  • All API calls are authenticated via CLAWDOS_API_KEY and encrypted over HTTPS when applicable.

⚠️ AUTHORIZATION & CAPABILITY WARNINGS

This skill grants access to powerful Windows automation capabilities. Users must explicitly understand and authorize the following operations:

  1. Shell Command Execution (shell_exec)
    • Can execute arbitrary PowerShell or cmd commands on the Windows host.
    • Even within the sandbox, commands can potentially modify system state, install software, or alter configurations.
    • Only use with trusted sources and explicit user approval.
  1. File Deletion (fs_delete)
    • Permanently removes files and directories within the sandbox.
    • No recovery mechanism exists once deleted.
    • Exercise extreme caution; confirm deletion intent before execution.
  1. File Upload/Download (--file, --out)
    • The CLI script can read local files and upload them to the Windows host (within sandbox).
    • The script can download remote files from the Windows host to the agent system.
    • Do not use with sensitive files or untrusted remote systems.
  1. Persistent Screen/Window Monitoring
    • Visual actions (screen_capture, window_list, window_focus) can observe active GUI content.
    • If sensitive information is visible on screen, it may be captured.

⚠️ Requirements

This skill requires a corresponding server running on your Windows host.

Ensure the Windows host is running danzig233/clawdos. The connection parameters (CLAWDOS_BASE_URL and CLAWDOS_API_KEY) must be configured via OpenClaw's skill configuration UI or environment variables, as specified in this file's metadata.

Usage

You interact with Clawdos by running the scripts/clawdos.py CLI using the exec tool. The script will automatically pick up the CLAWDOS_BASE_URL and CLAWDOS_API_KEY environment variables injected by OpenClaw.

Basic Syntax:

python3 ~/.nvm/versions/node/v22.22.1/lib/node_modules/openclaw/skills/clawdos/scripts/clawdos.py <action> --args '{"key":"value"}'

Available Actions

1. Visual Navigation & System Check

  • health: Check service status.
  • get_env: Get screen resolution, DPI scale, and active window.
  • window_list: List all open windows.
  • window_focus: Focus a window. Args: {"titleContains": "..."} or {"processName": "..."}
  • screen_capture: Take a screenshot. Use --out path/to/save.png to save binary. Args: {"format": "png", "quality": 80}

2. Precise Input (Mouse & Keyboard)

(Prioritize keyboard/shell when possible to avoid visual estimation errors)

  • click: Click the mouse. Args: {"x": 100, "y": 200, "button": "left"}
  • move: Move cursor. Args: {"x": 100, "y": 200}
  • drag: Drag mouse. Args: {"fromX": 100, "fromY": 200, "toX": 300, "toY": 400}
  • keys: Press key combos. Args: {"combo": ["ctrl", "c"]}
  • type_text: Type text. Args: {"text": "hello"}
  • scroll: Scroll wheel. Args: {"amount": -500}
  • batch: Execute multiple input actions sequentially. Args: {"actions": [...]}

3. File & System Operations

  • fs_list: List directory contents. Args: {"path": "/"}
  • fs_read: Read a file (prints raw contents to stdout). Use --out path/to/save.bin to save binary files. Args: {"path": "/hello.txt"}
  • fs_write: Write to a file. Args: {"path": "/hello.txt", "content": "hello world"}. Or use --file path/to/local.bin to upload a local binary file.
  • fs_mkdir: Create a directory. Args: {"path": "/newdir"}
  • fs_delete: Delete a file or directory. Args: {"path": "/newdir", "recursive": true}
  • fs_move: Move or rename. Args: {"from": "/src", "to": "/dst"}
  • shell_exec: Run a shell command on the Windows host. Args: {"command": "dir", "args": ["/w"], "workingDir": ""}

Operation Strategy

Operational Best Practices

  • Prefer Keyboard & Shell: To minimize errors from visual coordinate estimation, prioritize using keyboard shortcuts (key_combo, type_text) or shell commands (shell_exec) over mouse operations whenever possible.
  • Targeted Mouse Usage: Reserve precise mouse operations (mouse_click, mouse_move, mouse_drag) strictly for necessary UI interactions (e.g., clicking a specific button on a web page, navigating a software interface, or focusing an input field).
  • Scrolling: Using mouse_scroll is safe and recommended for navigating long pages or documents.

Security Best Practices

  • Verify File Paths: Always confirm the target path is within the intended sandbox directory. The server enforces isolation, but double-check paths in scripts.
  • Audit Shell Commands: Review shell_exec commands before execution. Avoid running commands from untrusted sources.
  • File Transfer Restrictions: Only upload/download files you trust. Do not use --file with sensitive credentials or system files.
  • Minimize Screen Captures: Avoid capturing screens if sensitive information (passwords, tokens, personal data) may be visible.
  • Explicit Deletion Confirmation: Review the target path carefully before executing fs_delete. Deleted files cannot be recovered.

Examples

Focus MS Edge and type:

python3 scripts/clawdos.py window_focus --args '{"processName": "msedge"}'
python3 scripts/clawdos.py type_text --args '{"text": "https://openclaw.ai\n"}'

Take a screenshot and save it locally:

python3 scripts/clawdos.py screen_capture --out /tmp/windows_screen.png --args '{"format":"png"}'

Read a file from Windows:

python3 scripts/clawdos.py fs_read --args '{"path": "logs/app.log"}'

Execute PowerShell on Windows:

python3 scripts/clawdos.py shell_exec --args '{"command": "powershell", "args": ["-Command", "Get-Process"]}'

版本历史

共 2 个版本

  • v1.0.1
    2026-03-31 07:56 安全
  • v2.1.1 当前
    2026-03-18 19:52 安全 安全

安全检测

腾讯云安全 (Keen)

安全,无风险
查看报告

腾讯云安全 (Sanbu)

安全,无风险
查看报告

🔗 相关推荐

ai-agent

self-improving agent

pskoett
捕获经验教训、错误及修正内容,以实现持续改进。适用于以下场景:(1)命令或操作意外失败;(2)用户纠正Claude(如“不,那不对……”“实际上……”);(3)用户请求的功能不存在;(4)外部API或工具出现故障;(5)Claude发现自身
★ 4,085 📥 813,429
ai-agent

Skill Vetter

spclaudehome
AI智能体技能安全预审工具。安装ClawdHub、GitHub等来源技能前,检查风险信号、权限范围及可疑模式。
★ 1,230 📥 268,134
ai-agent

Self-Improving + Proactive Agent

ivangdavila
自我反思+自我批评+自我学习+自组织记忆。智能体评估自身工作、发现错误并持续改进。
★ 1,383 📥 320,829