AI
External sandboxes for coding agents
Publication: Simon Willison's Weblog
Author: Not available in collected metadata
Source published: 17/11/2026, 22:52:29 UTC
Briefing created: 17/11/2026, 23:47:57 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 17/11/2026, 22:57:24 UTC.
Briefings
The user reports collaboration with Prime Radiant to develop external sandboxes for coding agents, specifically to mitigate risks from local model mistakes or prompt injection. Three sandbox server implementations were released, each running a dedicated sandbox environment (Alpine Linux or Quicksand) while the agent operates externally. These servers provide a common localhost API for executing code and include optional allow-list proxies and credential replacement mechanisms. The project focuses on safely testing local models without exposing them to machine damage.
Key points
- Three sandbox server implementations were released: external-sandbox-v86 (Node.js + Alpine Linux), external-sandbox-container (Python + Apple Containers), and external-sandbox-quicksand (Python + Microsoft Quicksand).
- All implementations offer a common localhost API for agents to execute code and include optional HTTP/HTTPS allow-list proxies for host control and credential replacement.
Context
The user worked with Prime Radiant, the AI research lab behind Superpowers and other projects, to explore external sandboxing patterns for safer code execution by coding agents.
Limitations
Evidence is limited to the user's blog post and project descriptions; specific security vulnerabilities, incident reports, or quantitative impact metrics are not provided.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 17/11/2026, 23:47:57 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeidjghztr4idl3ymsjnhjciv2m44cedseslnmr7fls5pulve5ay5z4