Have you tried the sync feature?
Edit: FML why is this being downvoted? At least have the decency of explaining, I'm happy to adjust my conduct but I can't do so if I don't know what I did wrong.
if you're building a harness, the SDK provides better integration. let me know if you hit any blockers.
for sync, it's currently in the works.
If I run the agent outside and use the MCP, is it the model's responsibility to actually develop in the sandbox or are there deterministic guardrails against performing activities outside of it?
to actually enforce isolation, the sandbox needs to be the only way the agent can run code. either don't give it any other code-running tools, or the harness exposes sandboxing as a first-class concept. for the latter, anthropic makes this possible with self-hosted execution: https://platform.claude.com/docs/en/managed-agents/self-host...
happy to answer any other questions.
The article is really about solving a particular problem with the backend of their infrastructure. Discussion about VMs, Linux kernel syscalls, file systems (virtual, FUSE, etc) would all be relevant.
Your comment is a question about whether and how people use the software itself, which is pretty unrelated to the article.
It’s a bit like an article about Porsche identifying a particular engineering nuance in their fuel injectors, and how things didn’t work the way they thought at a low level, and how they solved it once they realized it. And then you come in with a comment about what people like to do with their Porsches. Like, sure, it involves the same company but what would that have to do with the underlying article on automotive engineering?
Combine that with a growing disdain for the insistence of certain segments of the tech scene to make everything about agentic workflows, (an echo to the constant evangelism of cryptocurrencies or blockchain in the recent past) and you have a recipe for downvotes.
Oh well. :) Thanks for your insight anyway.