docker-docker-sandbox-agent（Docker Docker sandbox Agent 部署）

Docker Sandbox

Execute and verify code securely inside isolated temporary containers.

Problem

Agents frequently generate code (Python, JS, Shell scripts) that they need to test before presenting a final solution. Running this code directly on the host machine poses significant security and stability risks.

Solution

The docker-sandbox skill establishes a pattern for agents to provision a temporary, isolated container environment using Docker. Code executes inside the container, captures standard output and error, and cleans up immediately after completion.

Features

• ✅ Host Isolation - Runs code securely away from the host system's file tree and environment variables.
• ✅ Ephemeral Execution - Containers are automatically destroyed (--rm) after the process exits.
• ✅ Multi-environment - Supports Python, Node.js, Go, or any Docker-supported runtime.
• ✅ Resource Controls - Hard constraints on memory and CPU avoid runaway processes.

Usage Guide (For Agents)

When evaluating or testing code, use the following execution abstractions via the host terminal. Prepare your code in a local temporary directory (e.g., ./.sandbox/), then mount it in the container.

1. Python Verification

Run a python script securely with constrained memory and CPU.

# Create a test script
mkdir -p .sandbox
echo 'print("Hello from Docker Sandbox!")' > .sandbox/main.py

# Execute in python sandbox
docker run --rm \
    --memory="512m" \
    --cpus="1.0" \
    --network none \
    -v "$(pwd)/.sandbox:/app" \
    -w /app \
    python:3.10-slim python main.py

2. Node.js Verification

Evaluate JavaScript / Node.js safely.

docker run --rm \
    --memory="512m" \
    --cpus="1.0" \
    --network none \
    -v "$(pwd)/.sandbox:/app" \
    -w /app \
    node:18-alpine node main.js

3. Bash/Shell Verification

Test shell scripts in a generic Alpine environment.

docker run --rm -v "$(pwd)/.sandbox:/app" -w /app alpine sh script.sh

Security Guidelines

1. Mount Minimization: Never mount sensitive host directories (e.g., /etc, ~/.ssh, or /) into the sandbox. Mount only the specifically designated .sandbox or task-related directory.
2. Network Isolation: By default, include --network none in the command to prevent the code from exfiltrating data or initiating unwanted network requests, unless network access is functionally necessary for the test.
3. Privileges: Never use --privileged mode or run containers mapped directly to the root user of the host if preventable.