antomanc

Agbox — DevOps skill for Claude Code

DevOps community

Turn a fresh Debian/Ubuntu machine into a home for AI coding agents (Claude Code, Codex, opencode) in one command: toolchains, Docker, Playwright, Tailscale, snapshots.

How to install Agbox

This entry records only its repository, not the path inside it, so there is no exact command to give. Open antomanc/agbox and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agbox does

Turn a fresh Debian/Ubuntu machine into a home for AI coding agents (Claude Code, Codex, opencode) in one command: toolchains, Docker, Playwright, Tailscale, snapshots. Plain Bash, idempotent.

Alternatives in DevOps

Claude Code Router — Use Claude Code as the foundation for coding infrastructure, allowing you to decide how to interact with the m 30.1k ★
Mirrord — Run any process, on your machine or in an AI agent's environment, as if it were a pod in your Kubernetes clust 5.3k ★
Skill Claw — OpenClaw instance administration — manage hosts across macOS, Ubuntu/Debian, Docker, OCI, and Proxmox 2.8k ★

README

agbox

**Turn a fresh Debian/Ubuntu machine into a smooth home for AI coding agents, in one command.**

Mini PC in a closet, old laptop, or $5 VPS: agbox sets it up so Claude Code, Codex and opencode can install things, run builds, start databases and test web apps **without tripping over sandboxes, AppArmor, sudo prompts or missing toolchains**.

curl -fsSL https://raw.githubusercontent.com/antomanc/agbox/main/install.sh | bash

[![CI](https://github.com/antomanc/agbox/actions/workflows/ci.yml/badge.svg)](https://github.com/antomanc/agbox/actions) [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)

Why

Running agents on a headless Linux box usually means a pile of small failures:

bubblewrap sandboxes blocked by Ubuntu's AppArmor
node found in your shell but not in the agent's
password prompts mid-task
services that die on reboot
a database installed three different ways

agbox fixes these once, with defaults that come from actually running agents on a box like this.

What you get


Agents	Claude Code, Codex, opencode, with one shared `AGENTS.md` that tells them what this machine is and how to work on it
No friction	Non-root user with passwordless sudo. Agents run without prompts, because the machine is the sandbox (why)
Toolchains	mise: Node, Python, uv, pnpm... working in interactive shells, agent shells and systemd services
Services	Docker Engine + Compose for Postgres/Mongo/Redis. Optional always-on t3code and Hermes Agent
Web testing	Playwright headless Chromium with all system libraries
Access	Tailscale + Tailscale SSH from anywhere, no open ports
Safety net	btrfs snapshots every hour (roll back anything an agent breaks), auto security updates, key-only SSH on VPSs
Small machines