🛠 agent-run: sandboxing coding agents
A lightweight Rust-based tool, agent-run, has been released to securely run coding agents (Claude, OpenCode, or Pi) in an isolated environment (sandbox). The project is less than 1 MB and uses bwrap to protect Linux systems.
🌍 It solves the security problem when working with autonomous AI agents by providing a simple isolation mechanism without the need to deploy heavy virtual machines.
👤 It allows for safe experimentation with AI coders on a local machine by restricting their access solely to the project's working files.
Source 1: https://github.com/sin-ack/agent-run