sandbox-claude-inside

Run Claude Code inside an isolated ca-sandbox container with token authentication.

139|7|Updated May 8, 2026
One-click install
npx skills add https://github.com/arbiterForge/codeArbiter --skill sandbox-claude-inside
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: sandbox-claude-inside
Source: https://github.com/arbiterForge/codeArbiter/tree/main/plugins/ca-sandbox/skills/sandbox-claude-inside
Command: npx skills add https://github.com/arbiterForge/codeArbiter --skill sandbox-claude-inside

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill lets you run Claude Code inside a hardened ca-sandbox container so agent loops stay isolated from the host and follow explicit security boundaries.

Core Features & Use Cases

  • Isolated agent execution: Runs Claude in a host-FS-isolated container instead of the machine you are working on.
  • Credential-safe auth: Uses an env-injected CLAUDE_CODE_OAUTH_TOKEN with a named HOME volume and no host bind of ~/.claude.
  • Guarded networking and teardown: Enforces offline or Anthropic-only egress, blocks token/source co-mounts, and cleans up containers and volumes by default.
  • Use Case: Start a sandboxed Claude loop for a repo when you need strong isolation, reproducible CLI pinning, and explicit lifecycle control.

Quick Start

Use the sandbox-claude-inside skill to run Claude Code in a hardened ca-sandbox container with the desired egress posture and token-handling rules enforced.

Frequently Asked Questions about sandbox-claude-inside

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run Claude Code in an isolated container safely?

Run Claude Code in an isolated container by using a hardened ca-sandbox that enforces host-FS separation, env-injected CLAUDE_CODE_OAUTH_TOKEN authentication, and controlled egress to keep agent loops secure.

Why does running Claude agents on untrusted source code require a sandbox?

Running Claude agents on untrusted source code requires a sandbox to enforce explicit security boundaries, preventing unauthorized host access and separating the agent loop from your local filesystem.

Can I prevent credential leaks when authenticating Claude Code in a sandbox?

Yes, you prevent credential leaks by using an env-injected CLAUDE_CODE_OAUTH_TOKEN with a named HOME volume, ensuring the sandbox blocks token and source co-mounts while rejecting host binds of ~/.claude.

How do I enforce egress control and lifecycle teardown for Claude Code containers?

Enforce egress control and lifecycle teardown by applying offline or Anthropic-only networking rules to the ca-sandbox, which automatically cleans up containers and volumes after the agent loop finishes.

Does Claude Code require a pinned installation for sandboxed agent loops?

Yes, a pinned Claude Code installation is required to ensure reproducible CLI access and stable execution within the isolated ca-sandbox container during hardened agent loops.

What are the limitations of using a ca-sandbox for Claude Code isolation?

Limitations include enforced offline or Anthropic-only egress, strict rejection of co-mounting source code with tokens, and mandatory container teardown rules that prevent persistent host-level volume bindings.