I’ve been experimenting with running Cursor CLI agents inside project-specific Docker containers. The setup is dead simple — a <name>.cursor-agent file and a Bash function that creates or resumes the container when you run cursor-agent.
It’s been great for isolation, reproducibility, and security. Curious to hear if others in the Cursor community have tried something similar, or if this could be worth formalizing into a shared convention.
Repo: https://github.com/cleaton/cli-agent-container