Container sandbox

Skill scoobydont-666/shared-claude-skills/skills/container-sandbox

Run risky or destructive operations inside ephemeral Docker containers for isolation. Use when executing untrusted scripts, destructive Ansible playbooks, or operations that could damage the host.From its SKILL.md

Install

npx -y skills add scoobydont-666/shared-claude-skills --skill container-sandbox

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

reads credentialsReads from 1 credential source: `~/.ssh`.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
runs commandsInstructs the agent to run 7 commands, including `docker run --rm -it -v "$(pwd):/project:ro" -v /tmp/sandbox-work:/work -w /work ubuntu:24.04 bash -c "cp -r /project/* /work/ && bash"` and 6 more.

SKILL.md

2.8 KB, 689 tokens by cl100k_base, as published. Nobody here has run it

Container Sandbox — Isolated Execution Environment

Run potentially dangerous operations inside ephemeral Docker containers. The host filesystem is protected — containers get read-only project mounts with a writable overlay.

When to Use

Destructive Ansible playbooks (--force, --limit testing)
Untrusted scripts from external sources
Operations that modify system state (rm -rf, package installs)
Testing migrations or schema changes before applying to production
Running code from unknown repos

Quick Start

Basic sandbox (Ubuntu + project files):

docker run --rm -it \
  -v "$(pwd):/project:ro" \
  -v /tmp/sandbox-work:/work \
  -w /work \
  ubuntu:24.04 \
  bash -c "cp -r /project/* /work/ && bash"

Python sandbox (with project deps):

docker run --rm -it \
  -v "$(pwd):/project:ro" \
  -v /tmp/sandbox-work:/work \
  -w /work \
  python:3.12-slim \
  bash -c "cp -r /project/* /work/ && pip install -r requirements.txt 2>/dev/null; bash"

GPU sandbox (for CUDA operations):

docker run --rm -it --gpus all \
  -v "$(pwd):/project:ro" \
  -v /tmp/sandbox-work:/work \
  -w /work \
  nvidia/cuda:13.1-devel-ubuntu24.04 \
  bash

Ansible dry-run sandbox:

docker run --rm -it \
  -v /opt/ai-project:/project:ro \
  -v ~/.ssh:/root/.ssh:ro \
  -w /work \
  --network host \
  willhallonline/ansible:latest \
  bash -c "cp -r /project/* /work/ && ansible-playbook -i inventory playbook.yml --check --diff"

Extracting Results

After running operations in the sandbox:

# Copy results out of the sandbox work directory
docker cp <container_id>:/work/results/ ./sandbox-results/

# Or use the /tmp/sandbox-work mount (persists after --rm if pre-created)
ls /tmp/sandbox-work/

Safety Rules

Always mount project directories as :ro (read-only)
Use /tmp/sandbox-work or /work for writable operations
Use --rm to auto-cleanup containers after exit
Use --network none for maximum isolation (no internet access)
Never mount / or /home as writable into a sandbox
GPU sandboxes (--gpus all) still share the host GPU — use for testing, not isolation

Integration with Claude Code

When Claude needs to run something risky, instruct it to:

Build the Docker command with appropriate mounts
Execute the command via Bash tool
Capture stdout/stderr from the container
Extract any produced files from the work directory
Clean up: rm -rf /tmp/sandbox-work

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.