A comprehensive interface for AI agents to control sandboxed Linux desktops using the E2B SDK.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install e2b-desktop
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install e2b-desktop using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is E2B Desktop Sandbox Manager?
The E2B Desktop skill enables AI coding agents to interact with a headless Ubuntu environment featuring the XFCE desktop. By wrapping the E2B Desktop Python SDK into efficient bash scripts, it provides a seamless bridge for agents to perform complex GUI tasks within a secure, isolated sandbox. This utility is a vital addition to your collection of Openclaw Skills for building robust computer-use agent loops.
Agents can perform a full range of Human Interface Device (HID) actions, from precision mouse movements to complex keyboard shortcuts. The skill handles state management automatically, allowing sandboxes to persist across script executions, which is essential for multi-step workflows and complex debugging sessions.
E2B Desktop Sandbox Manager Use Cases
- Building computer-use agent loops that require vision and GUI interaction.
- Automated software testing and QA within a sandboxed Linux environment.
- Securely running and debugging desktop applications like Firefox or VS Code without impacting the host.
- Capturing periodic screenshots to monitor system states for AI analysis and decision-making.
- Streaming VNC output to observe agent actions in real-time during development.
How E2B Desktop Sandbox Manager Works
- Initialize a new sandbox using start_sandbox.sh, which provisions a Linux instance and stores the session ID in a local state file.
- Execute shell commands or launch specific applications inside the environment to set up the agent workspace.
- Capture the current visual state using the screenshot utility to provide visual context to the AI model.
- Translate agent decisions into mouse movements, clicks, or keystrokes via dedicated script wrappers that interface with the E2B SDK.
- Optionally initiate a VNC stream to provide a live URL for visual monitoring of the sandbox desktop.
- Clean up resources by terminating the sandbox instance and removing local state when tasks are complete.
E2B Desktop Sandbox Manager Setup
To get started with this skill, install the necessary Python package and configure your API credentials:
pip install e2b-desktop
export E2B_API_KEY=your_api_key_here
Ensure the scripts in the scripts/ directory have execution permissions. You can then start your first sandbox session:
./scripts/start_sandbox.sh --resolution 1280x800 --stream
E2B Desktop Sandbox Manager Data Schema & Taxonomy
The skill manages its environment through local state files and predictable file paths to ensure compatibility across different Openclaw Skills implementations:
| Component | Description |
|---|---|
~/.e2b_state |
Stores the active sandbox ID to enable persistent sessions across multiple script calls. |
/tmp/e2b_screenshot.png |
The default destination for captured PNG frames used by vision-capable agents. |
E2B_SANDBOX_ID |
Environment variable that can override the local state for specific operations. |
| Script Exit Codes | Commands executed via run_command.sh return the actual exit code of the sandboxed process for error handling. |
E2B Desktop Sandbox Manager Advanced Features
- Persistent sandboxes that survive script exits, enabling long-running or multi-session agent tasks.
- Integrated VNC streaming support with optional authentication for secure, real-time visualization.
- Multi-key combination support (e.g., ctrl+alt+delete) for navigating complex OS shortcuts.
- Automated coordinate scaling to match any custom sandbox resolution defined at startup.
- Detailed window management capabilities, including the ability to list all open windows or focus on specific applications.
SKILL.md
Loading
Related Openclaw Skills
CLAW Betting AI Advisory
An AI-powered advisory engine providing intelligent bet recommendations, risk analysis, and bankroll management for gaming platforms.
Gemini Painter
A high-performance AI art generation tool that integrates Gemini Imagen 3 into Openclaw Skills for seamless text-to-image creation.
Android Remote Control
A technical skill for programmatically controlling Android devices through ADB and uiautomator2 integration.
Persistent User Memory
A long-term memory layer for OpenClaw that enables personalized interactions by persisting user preferences and patterns across sessions.
Steel Browser
A cloud-based browser control skill using Steel.dev and Playwright for reliable AI web automation.
Pamela Calls
An advanced voice AI skill for making instant, low-latency outbound phone calls with automated IVR navigation.
Featured*
