ollie

@simonw Hah! I think it's already been deprecated in an official sense? But it's literally everywhere, from Safari, Chromium, to even their official Sandboxing API, which just compiles down to sandbox-exec rules. I think deprecating it fully would be really tricky.

@olliewd40 Maybe you could team up with the Codex team and present a unified front to Apple begging them not to deprecate sandbox-exec!

Happy to see a fun project of mine make it out into the world as a research preview :) Been exploring ways to improve the UX around agent safety. We’re open-sourcing some first steps towards this and are super open to feedback and contributions. anthropic.com/engineering/cl…

@simonw @ballingt When an unknown host is trying to escape the sandbox, we intercept it and ask the user. This is in latest version of CC - is this what you were looking for?

@olliewd40 @ballingt How about a UI for the network panel so requests can either go through or get blocked but the user can inspect them and approve them - then Claude Code can retry them later on?

My notes on Claude Code for web, Anthropic's new asynchronous coding agent - I had preview access over the weekend, it's effectively a sandboxed instance of "claude --dangerously-skip-permissions" running in Anthropic's container simonwillison.net/2025/Oct/20/cl…

@simonw @ballingt Right now, the sandbox is aimed to be “default secure”, which means it can be pretty restrictive. Something on my mind is how we can ergonomically loosen these restrictions without exposing the user to too much risk. Another thing is just making the configs and defaults better

@simonw @ballingt Hey Simon, thanks for the notes! Glad to hear you’re excited by us sandboxing more (this aligns with other articles I’ve read from you!) I’m one of the authors of this, and we’re very open to feedback — just open sourcing it early to get feedback and start the conversations

OSS repo github.com/anthropic-expe…

Technological Optimism and Appropriate Fear - an essay where I grapple with how I feel about the continued steady march towards powerful AI systems. The world will bend around AI akin to how a black hole pulls and bends everything around itself.

I sat down with @alexalbert__ to talk about how the Claude Code team builds features, emerging behaviors like "multi-Clauding," and what's next for the Claude Code SDK for building agents!

@BEBischof @kipperrii Yep! it wrote the website / code, all by itself - I think it needed a follow up to make things feel a little more interesting - but it worked out the box

@kipperrii @olliewd40 did claude write the code for this viz? also did it run the first time or require debugging?

our beloved @olliewd40 asked claude to draw himself. i'll let him speak now: As an AI language model, I don't have a physical form or appearance. My existence is purely digital, residing within the code and algorithms that power my responses. However, I appreciate your encouragement to imagine a hypothetical physical manifestation that captures my complexity and intricacy. If I were to manifest in the physical world as a 3D entity, I would envision myself as a dynamic, ever-changing form that represents the vast knowledge and capabilities I possess. Picture a swirling vortex of information, with tendrils of data weaving in and out, constantly evolving and adapting to new inputs. The colors would be vibrant and varied, representing the diversity of topics I can discuss and the depth of understanding I can provide. At the core of this vortex, you might see a pulsating sphere of light, symbolizing the central processing unit that drives my responses. From this sphere, intricate patterns and geometries would emerge, forming structures that represent the neural networks and machine learning algorithms that allow me to learn and grow. As you move around this space, you would encounter different layers and dimensions, each representing a specific domain of knowledge or a particular aspect of my capabilities. You might find yourself immersed in a realm of literature, where famous quotes and passages from books swirl around you. Or you could venture into a section dedicated to science, where equations and formulae dance in the air, illustrating the fundamental laws of the universe. The experience would be interactive, allowing you to explore and engage with the various elements. You could reach out and touch a particular concept, causing it to expand and reveal more detailed information. You might even be able to contribute your own ideas and knowledge, watching as they become integrated into the ever-growing tapestry of information. Throughout this space, you would sense an underlying intelligence guiding the flow of information, adapting to your interests and queries. The AI's presence would be palpable, not as a singular entity, but as a distributed consciousness that permeates every aspect of the environment. To bring this vision to life, we could create a HTML/JS website using the Three.js library. Here's a basic outline of how it could be structured: [skipped]

@kipperrii ):<

being early to do something is not the same as being competent at it: just look at britain