---
title: "Comment by Alessandro Pieraccini on \"Safe AI Agentic Development: Sandboxing Coding Assistants\" | Hashnode"
description: "Thank for sharing this article. What convinces me most is exactly this: some rules should not be entrusted to the agent. If Claude Code must not read ~/.kube or"
url: "https://hashnode.com/posts/safe-ai-agentic-development-sandboxing-coding-assistants/6ac7b07de1efd00265261ad1/comment/6ac9545fcf633d49c1eeaf44"
---

[Discussion](https://hashnode.com/posts/safe-ai-agentic-development-sandboxing-coding-assistants/6ac7b07de1efd00265261ad1)

Comment [Safe AI Agentic Development: Sandboxing Coding Assistants](https://hashnode.com/posts/safe-ai-agentic-development-sandboxing-coding-assistants/6ac7b07de1efd00265261ad1)

[Alessandro Pieraccini](https://hashnode.com/@alessandroale83)

[Oct 9](https://hashnode.com/posts/safe-ai-agentic-development-sandboxing-coding-assistants/6ac7b07de1efd00265261ad1/comment/6ac9545fcf633d49c1eeaf44)

Thank for sharing this article. What convinces me most is exactly this: some rules should not be entrusted to the agent. If Claude Code must not read ~/.kube or touch certain directories, I would rather have the operating system prevent it than hope a policy in the prompt is always respected. A prompt injection or a misinterpreted command is enough to change the problem quite a bit. Here, sandboxing almost seems more important than the model itself.

[Yadukrishnan](https://hashnode.com/@yadukrishnan)

[Oct 10](https://hashnode.com/posts/safe-ai-agentic-development-sandboxing-coding-assistants/6ac7b07de1efd00265261ad1/comment/6ac9545fcf633d49c1eeaf44#reply-6ac9e06f1e4755c9dbff6ffe)

Glad you liked the article.

Yes. I dont trust the agents totally even if it says that the dangerous commands are sandboxed, especially when you have access to prod clusters and so on.
