Code Container

Code Container: Isolated Docker environment for your autonomous coding harness.

Overview

• Project Isolation: One container per project with complete isolation
• State Persistence: All changes and packages persist between sessions
• Shared Resources: npm cache, pip cache, and Claude history shared across projects
• Security: Changes within a container don't affect your host or other projects

Prerequisites

• Docker — Docker Desktop or Docker Engine
• A POSIX-Compatible System — Linux, macOS, WSL

Initial Setup

1. Install as Global Command

To use the container command from anywhere, create a symlink in a PATH-tracked folder:

ln -s "$(pwd)/container.sh" /usr/local/bin/container

2. Configure Harnesses

Copy configurations into this repo (shared across all containers):

# Script to copy harness configs
./copy-configs.sh

Or, if copying manually:

# OpenCode
cp -R ~/.config/opencode/ ./.opencode/
# Codex
cp -R ~/.codex/ ./.codex/
# Claude Code
cp -R ~/.claude/ ./.claude/ && cp ~/.claude.json container.claude.json

3. Build Docker Image

container --build    # Run once, or when rebuilding

Includes: Ubuntu 24.04 LTS, Node.js 22 LTS, Python 3, Claude Code, OpenCode, OpenAI Codex CLI, git. Add other tools by modifying the Dockerfile.

Primary Usage

Navigate to any project and run container to mount the project and open the container.

cd /path/to/your/project
container                    # Enter container

Inside the container: Start your harness and develop like normal.

opencode                     # Start OpenCode
codex                        # Start OpenAI Codex
npm install <package>        # Persists per container
pip install <package>        # Persists per container
exit                         # Auto-stops container on exit

Container state is saved. Next invocation resumes where you left off. AI conversations and settings persist across all projects.

Container Isolation

Destructive actions are localized inside containers. You can let your harness run with full permissions.

.opencode/opencode.json

{
  "permission": "allow"
}

.codex/config.toml

approval_policy = "never"
sandbox_mode = "danger-full-access"

.claude/settings.json

{
  "permissions": {
    "allow": [
      "*",
      "Bash"
    ]
  }
}

Common Commands

container                  # Enter the container
container --list           # List all containers
container --stop           # Stop current project's container
container --remove         # Remove current project's container
container --build          # Rebuild Docker image

# With an explicit path:
container /path/to/project
container --stop /path/to/project
container --remove /path/to/project

What Persists

Per-Container:

• All installed system packages, npm packages, Python packages
• All file modifications, databases, shell history
• Container filesystem state

Shared Across All Projects:

• Harness configuration and conversation history
• npm and pip download caches
• Python user packages

Read-only from Host:

• Git configuration, SSH keys

Simultaneous Work

You and your harness can work on the same project simultaneously.

Safe: Reading files, editing files, most development operations

Avoid: Simultaneous Git operations from both sides, installing conflicting node_modules

Recommended Workflow: Let your harness run autonomously in the container while you work; review changes and commit.

Customization

Add tools/packages — Edit Dockerfile and rebuild:

RUN apt-get update && apt-get install -y postgresql-client redis-tools && rm -rf /var/lib/apt/lists/*

Add shared volumes (caches, config, etc.) — Edit the docker run -it command in container.sh:

-v "$SCRIPT_DIR/new-shared-dir:/root/target-path"

Security

• SSH keys and Git config mounted read-only
• Project isolation prevents cross-contamination across containers
• Host filesystem protected (access limited to mounted directories)

Limitations:

• Network access still available; information may still be exfiltrated over network
• Project files can still be deleted by harness; always use upstream version control