Containerizer Sandbox

Run AI coding agents (Claude Code, etc.) in isolated Apptainer containers on HPC clusters with SLURM job scheduling.

Uses a two-layer container strategy: base-system.sif (Ubuntu 22.04 + Python, Node.js, core tools) and base-agent.sif (adds Claude Code + sandbox environment).

Get Started

Paste this into your coding agent (Claude Code, Cursor, Gemini CLI, etc.):

Read INSTALL.md from https://github.com/Kamaleswaran-Lab/containerizer and follow the instructions to install HPC Agent Sandbox and start running containerized agents.

That's it. The agent handles the rest.

Quick Start

# Launch an interactive sandbox session
sandbox shell --task task.yaml

# SSH mode for IDE access
sandbox shell --task task.yaml --ssh

# Build container images from definitions
sandbox build

# View task audit trail
sandbox audit --task-id <id>

# Clean up old tasks
sandbox cleanup --older-than 7d

# Stop a running task
sandbox stop --task-id <id>

CLI Subcommands

Command	Description
shell	Launch an interactive sandbox session
build	Build .sif container images from definition files
audit	Show audit information for a completed task
cleanup	Remove old task directories and logs
stop	Cancel a running sandbox SLURM job

AI Agent Integration

Three skills are provided in skills/ for AI agent integration:

• sandbox-install — automated setup of the sandbox CLI and dependencies
• sandbox-configure — interactive wizard to generate a task.yaml
• sandbox-reference — CLI command reference for day-to-day use

Requirements

• Python 3.11+
• Apptainer container runtime
• SLURM workload manager
• uv package manager