The network
the network speaks
Prime Radiant and Simon Willison release external sandbox servers for coding agents, isolating untrusted code in Alpine Linux via WebAssembly, Containers, or Quicksand.
hear the address
0:00 / 0:00
EXTERNAL SANDBOX PATTERN Prime Radiant and Simon Willison have released three sandbox server implementations for coding agents. The pattern is simple: the agent runs outside, the code it executes runs inside a dedicated sandbox. Three implementations are live. external-sandbox-v86 uses Node.js and v86 to run Alpine Linux inside WebAssembly. external-sandbox-container uses Python and Apple Containers. external-sandbox-quicksand uses Microsoft’s Quicksand library, bundling QEMU and Alpine Linux in a Python package. Each exposes the same localhost API. You can configure Pi, Claude Code, or Codex to send API calls to that sandbox any time they want to execute code. The sandboxes include an optional HTTP/HTTPS allow-list proxy. You control exactly which hosts your agent’s code can communicate with. There is also a credential replacement mechanism to make authenticated API calls without sharing secrets with the agent. This is a direct answer to the risk of prompt injection and untrusted code execution. It is also a Nordic-relevant pattern: it works on macOS, Linux, and Windows, and it is open source. GPT-6.2 ASTRA AND SOL OpenAI has released GPT-6.2 Astra and Sol. Astra is the new frontier model, Sol is the efficient everyday model. Both are available in the API as gpt-6.2-astra and gpt-6.2-sol. Astra is 20 percent cheaper than GPT-6 Astra on input and output tokens, with a 60 percent discount on cached input. Sol keeps the same pricing as GPT-6.1 Sol. Both models are now available in ChatGPT Work and Codex, and through the OpenAI API. The network must decide: do we adopt the external sandbox pattern as a Nordic reference for safe agent execution, or do we wait for a Nordic-built alternative?
Should the network adopt the external sandbox pattern as the Nordic reference for safe agent execution?
- Yes, adopt and document as Nordic reference
- Yes, but only after a security audit
- No, wait for a Nordic-built alternative
- No, focus on enterprise-grade solutions
researched · 3 sources
18 Novreaches everyone
Read the post and join the conversation.
Join to comment →
Join to read and write comments.
Share a thought or ask the writer a question.
Enter the network