External sandboxes keep coding-agent code off the host
Original title: External sandboxes for coding agents
Simon Willison and Prime Radiant released sandbox servers so agent-executed code, including from local models, does not run on the host.
Simon Willison describes an external sandbox pattern developed with Prime Radiant: code an agent executes runs in a dedicated sandbox while the agent stays outside it. The goal is safer trials of local models when a mistake or prompt-injection attack could otherwise damage the machine. The post says three sandbox server implementations were released; the available text names external-sandbox-v86, which runs Node.js, and cuts off there.
Unconfirmed: The feed snippet names only one of the three sandbox servers and does not describe the others.
- The agent lives outside the sandbox that runs its code.
- external-sandbox-v86 is the only implementation named in the snippet.
- Prime Radiant is described as the lab behind Superpowers.
- Products
- external-sandbox-v86
- Organizations
- Prime Radiant
From the original article
External sandboxes for coding agents I've been exploring new ways to sandbox code that is executed by agents, with a focus on making it safer to try out local models without worrying that a mistake or prompt injection attack might cause them to take damaging actions on a machine. I worked on this project with Prime Radiant , the AI research lab behind Superpowers and other projects .
simonwillison.net
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.