How fast is the sandbox execution?

When using the optional daemon mode, AgentKernel maintains a pool of pre-warmed sandboxes, enabling command execution in under 125ms.

Can I restrict network access for specific commands?

Yes, you can use the 'restrictive' security profile to completely disable network access and set the filesystem to read-only for maximum security.

AgentKernel is a specialized tool that allows AI agents to execute code and terminal commands in secure, isolated containers or microVMs to protect the host system from unintended side effects.

Which programming languages are supported?

It automatically detects and supports a wide range of environments including Python, Node.js, Rust, Go, Ruby, Java, PHP, C/C++, and Terraform.

AgentKernel Sandbox

Name: AgentKernel Sandbox
Author: thrashr888

bythrashr888

•

安全与测试

Executes commands and code in secure, isolated microVMs and containers to protect your host system during AI-driven development.

AgentKernel provides a high-performance execution environment for AI agents like Claude Code, allowing them to run untrusted code, shell commands, and complex build processes within secure, hardware-isolated sandboxes. With sub-125ms boot times and support for dozens of runtimes including Python, Node.js, and Rust, it enables safe code execution and testing without risking the integrity of the host machine or exposing sensitive credentials. The skill integrates seamlessly via the Model Context Protocol (MCP), offering granular control over security profiles and network access to balance developer productivity with robust system protection.

主要功能

01Full MCP server integration for direct tool usage within Claude Code

02Sub-125ms boot times with pre-warmed sandbox pools in daemon mode

03Configurable security profiles ranging from restrictive (no network) to permissive

0426 GitHub stars

05Hardware-level isolation using Firecracker microVMs and secure containers

06Automatic runtime detection for Python, Node.js, Rust, Go, and 10+ other languages

使用场景

01Running and testing untrusted or AI-generated code snippets safely

02Executing build and test commands in a clean, reproducible environment without side effects

03Isolating shell operations that could potentially modify the host system state or files

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add thrashr888/agentkernel agentkernel

For use in Claude.ai and ChatGPT

Download Skill