Agentkernel
- 56 repo stars
- Updated July 26, 2026
- thrashr888/agentkernel
agentkernel is a Claude Code plugin and CLI that runs commands in secure isolated sandboxes with auto-detected runtimes so developers can execute agent-requested code safely.
About
agentkernel is a Claude Code marketplace plugin wrapping a CLI that runs commands in secure, isolated microVM sandboxes with sub-125ms boot times. Developers reach for it when AI agents need to execute tests, builds, or scripts without trusting host-level side effects. The run command auto-detects language runtimes from commands and project files, while persistent sandboxes support longer CI-style work. Optional signed receipts help verify what actually executed inside the sandbox.
- agentkernel run auto-detects runtime from command or project files and cleans up after execution
- Persistent named sandboxes with templates, TTL, git-branch naming, and exec into running environments
- Supports many languages via Docker images detected from package manifests and Procfiles
- Optional signed execution receipts with verify and replay commands
- Installable via Homebrew, install script, or cargo install agentkernel
Agentkernel by the numbers
- Data as of Jul 27, 2026 (Skillselion catalog sync)
/plugin marketplace add thrashr888/agentkernelAdd your badge
Show developers this marketplace is listed on Skillselion. Paste this into your README.
| repo stars | ★ 56 |
|---|---|
| Last updated | July 26, 2026 |
| Repository | thrashr888/agentkernel ↗ |
How do I let AI coding agents run project commands without exposing my host filesystem and environment to unrestricted execution?
Run agent-requested commands in secure, isolated microVM sandboxes with auto-detected language runtimes and optional signed receipts.
Who is it for?
Developers running agent-driven test, build, or script execution who need fast hardware-isolated sandboxes with minimal setup.
Skip if: Workflows that only need lightweight permission prompts inside Claude Code and never execute untrusted or agent-generated shell commands.
What you get
Command output from an ephemeral or named sandbox, optionally a signed execution receipt, with automatic cleanup after run completes.
Plugins in this marketplace
1 plugin - install individually after you add the marketplace.
Recommended Marketplaces
FAQ
How does runtime detection work?
Agentkernel picks a Docker image from the command, project files like package.json or Cargo.toml, Procfile entries, or an Agentkernel.toml config.
What is the fastest way to run a one-off command?
Use Agentkernel run <command>, which creates a temporary sandbox, executes, and cleans up automatically.
Can I keep a sandbox after a failed run?
Yes. Pass --keep to Agentkernel run or create a named sandbox with Agentkernel sandbox create for persistent debugging.