Sandbox Skill

Create and manage isolated container environments on DigitalOcean App Platform for AI agent code execution and testing workflows.

What This Skill Does

• Creates ephemeral, single-use sandbox containers for running untrusted code
• Manages hot pools of pre-warmed sandboxes for instant acquisition (~50ms vs ~30s cold start)
• Provides SDK patterns for AI agent workflows (code interpreters, iterative development)

When to Use This Skill

Use sandbox when you need to:

• Execute untrusted code in isolation (AI code interpreters)
• Run long-running or stateful agent workflows
• Test in isolated environments before production

NOT for debugging existing apps — use the troubleshooting skill for that.

Quick Start

from do_app_sandbox import Sandbox

# Create a sandbox
sandbox = Sandbox.create(image="python")
result = sandbox.exec("python3 -c 'print(2+2)'")
print(result.stdout)  # 4
sandbox.delete()

Key Decisions

Decision	Choice	Rationale
SDK over CLI	Python SDK only	AI agents use programmatic access
Hot pool default	SandboxManager	Eliminates 30s cold start for agents
Image choice	python, node	Pre-built at ghcr.io/bikramkgupta

Files

• SKILL.md — Full documentation with decision trees and patterns
• reference/cold-sandbox.md — Single sandbox creation patterns
• reference/hot-pool.md — SandboxManager for pre-warmed pools
• reference/use-cases.md — AI agent and testing patterns
• reference/positioning.md — When to use sandbox vs Lambda

Integration

• ← troubleshooting: Different use case (debug existing vs create new)
• → designer: Can include sandbox-compatible images in app specs