Treat your coding agents like developers

6 hours ago

yolobox was created to safely sandbox AI coding agents, preventing them from accidentally damaging the home directory by running destructive commands.
The single-agent workflow doesn't scale because multiple agents editing the same Git checkout, filesystem, and Docker Compose setup leads to conflicts and resource collisions.
Instead of using Git worktrees, which require manual setup and lack local state like dependencies and env files, yolobox forks full copies of the project folder for each agent.
Each agent gets a named fork (e.g., 'alice') with its own complete workspace, including ignored files, dependencies, and runtime isolation via unique Docker Compose project names.
To manage multiple web apps, yolobox uses a local reverse proxy with friendly URLs (e.g., alice.myapp.localhost) instead of random ports, improving usability.
This approach treats agents like developers, providing parallelism and isolation similar to human teams, with agents pushing branches for review and integration.
The workflow enables multiple agents to work simultaneously on tasks like bug fixes, feature prototyping, and refactoring, reducing coordination friction.

Hasty Briefsbeta