rubyist

Cubicle

This is my personal, sandboxed Docker setup for running AI coding agents with full autonomy.

Why I Built This

• I want one place to store and version my configs, shared context, and examples.
• I want to be able to iterate quickly on commands, skills, prompts across multiple models.
• I want to learn what these agents can and cannot do for me.
• I want to run agents in “allow everything” mode without risking my host system.

What It Runs

Agents:

• Claude Code
• Codex
• GitHub Copilot CLI

Each agent gets:

• 8GB memory
• Access to my repos and notes (mounted from the host)
• Full permissions inside its container
• A shared workspace so they can work on the same code

Quick Start

cp .env.example .env
# add your API keys
docker-compose up -d
docker exec -it claude-cubicle claude

See DOCKER_SETUP.md for the detailed setup.

Next Steps (for me)

• Build a workflow where agents pull tasks from Todoist, complete them, and open a pull request. Document that in CLAUDE.md and AGENTS.md.
• Build a CLI to run a specific agent in a specific repo.
• Test whether agents can collaborate on an options trading dashboard in Streamlit, using my matplotlib prototypes.
• Delegating to cloud agents