This is why I run opencode and similar things in a dedicated KVM virtual machine that lives on a system under my desk that doesn't have access to my user account data, documents folder, photos/video, ~/.ssh/, other API keys, ~/.anything-else/, you name it.
Sometimes this VM gets duplicated for specific projects and then various dependencies for testing installed in it that are specific to what the needs are.
As an additional advantage it means I can leave it running in the background doing things when I want to shut my laptop, then resume talking to it later.
I do something similar. I have a base NixOS image in Incus, with whatever tools apply to every project (e.g. Git, OpenCode) already installed. When I work on a project, I spin up a VM instance, use nix shell to add any project-specific tools, then share only the project folder from the host to the guest. This way, the worst the agent can do is destroy my project folder, and I can always restore that from another clone of the repo.
I know a lot of people are using containers for sandboxing, but given how capable the latest models have shown themselves to be for breaking out of sandboxes, I prefer the extra isolation of VMs for this.
I do all this locally - it's an interesting point to able to turn the laptop off but keep the agents running. I might consider running some of these on my homelab server just for that.
My work computer has nothing of personal value to me. It has everything I need to do work. The agent does my work. Why wouldn’t I just run this shit?
We’ve all been executing arbitrary code from a gorillion packages from pypi, npm, cargo, etc for well over a decade. Getting anyone to care much will be an uphill battle.
If you work for $bigcorp and have a work computer with nothing of personal value to you, all the more power to you. Great way to keep things separated. I'm writing more from the perspective of somebody whose home office and workstation environments contain both tons of personal data and tons of consulting/freelance/project/tasks-for-clients stuff.
I once opened Zed, howdy there was an agent window on the side, I log into Claude. I asked it to list what it could see and work with... And there my (priv) ssh keys come scrolling by. I don't know what I was expecting really, it's very logical, but also very in your face. They didn't leave my computer I guess, I just showed me the output of some `ls` commands. But still.
So now I have CC in a container, mounting it's own credentials/memory folder per project and only mounting one repo at a time (script in the repo itself). I feel a bit better about it now. It can still access my networks of course.
If you ever want a few more features on top of "CC in a container", then do consider byre. It's literally that, but also a nice TUI for adding extra folders, reusable skills, encrypted credentials, etc, and you can eject back to plain Docker (or Podman) whenever you want. Useful if you want to just be able to go to any folder and spin it up with your toolbox with one command: https://github.com/pjlsergeant/byre
Docker containers aren't very secure because the daemon runs as root by default. Rootless container runtimes are plenty secure, especially if the user that runs the container is unprivileged. Easiest thing in the world with Podman.
I would have them run as a user that does not have sudo. The easiest way to do this is by using system-level Podman Quadlets that run as an unprivileged user account. That user will need a real home directory for images and volumes.
This article smells AI generated, which is _very_ funny given the argument being made.
Either way, while this is true in the absolute, this is the value of building good MCP servers: they should expose only exactly the surface you expect your agent to need, and adding functionality should be carefully considered. The best MCP servers I use day to day (Cloudflare sticks out) do a really good job of exposing only what an agent might actually want to do on my behalf, rather than just all and sundry. Unfortunately the Chrome Dev tools MCP is less discriminating and is only as secure as a browser sandbox with full JS access (not fatal but not as strong as a well scoped REST API).
All of this is sidestepped somewhat by using good isolation primitives - I'm running a Hermes agent as of recently on a DigitalOcean VM that only accepts connections from my devices over tailscale, and it has all its own credentials so I can revoke them easily should they be used maliciously. Giving an agent root on a box is not _necessarily_ a huge deal, you just have to make sure that box has nothing valuable on it.
I kinda feel like we're rediscovering "Cattle, not Pets" when it comes to the environments we run our agents: give it root, sure, but a root that is almost meaningless outside of the functionality you granted it.
I also think it's really funny that it sort of comes to the conclusion of "we're gonna make something not that different than a FreeBSD jail 25 years ago" as the best possible sandboxing solution.
right, and I dont understand or haven't looked long enough, why there isnt a centralized agent harness that just uses existing protocols like ssh or docker container exec to do all it's work. Why does every agent have to be deployed directly into a VM or whatever. Why arn't we pushing them through the narrowly defined hole of a protocol?
So this looks like Claude is discussing the idea that when you run untrusted code on your computer, it can do arbitrary things. If there is malicious code, maybe it'll read your files and steal your credentials.
Here's a sample: "That’s not hypothetical. That’s POSIX working exactly as designed. None of it requires exploiting anything. It’s your user account doing normal user account things."
I would recommend clicking TFA if you've never heard of a supply chain attack, or if you just like getting the honest load-bearing facts that are worth mentioning.
I have to say, I'm currently trying to write a post and it's taking hours, and I can definitely see why people just pay the $0.02 to generate it instead. I like writing, so I keep doing it, but sometimes it can really be a slog.
If you like a real focus on developer experience and want lots of bells and whistles that a lot of thought has gone in to, I wrote byre, which I think is very good: https://github.com/pjlsergeant/byre
After the grok cli fisaco (uploading the full repo) I moved my different projects to seperate users, seperate sandbox around each harness (nono). Vm is of course even better but resources and easy of use start to be an issue
Author of the post, and dev of mcp-box here. Wrote this after realizing every MCP server on my machine had the same access to ~/.ssh that I do, and nothing in the installation messages posting to stdout mentions it. I think prompt injection is the vector and the permissions model is the red carpet giving a warm welcome. Interested in where that's wrong.
This line also undercuts the whole fucking article too.
Because as it damn well says - this is POSIX working exactly as designed.
I don't know why he's in here playing this up as some unexpectedly crazy, scary, dramatic music style thing (I mean... I do, it's so he can try to sell you something after scaring you...).
He started the damn process. Of course it's running as his damn user.
It would be one thing to honestly evaluate that security model and its merits (of which there are many, despite the risks). It's another to play this up as some giant surprise.
---
Constant stream of junk articles in here like
"You know when you hit that nail with the hammer you were holding? Did you consider you could have hit something else in reach instead? It's like, that hammer was like, you, man! Isn't that cray-cray and super duper scary??!?!?!"
Expecting an agent harness to respect security boundaries is definitely the wrong thing.
We need to have easy to use auditable sandboxing of agents and their harnesses, because it's quite clear certain model providers are going to try to lock us in at the harness level, and it's those harnesses that seem most likely to go nuts.
I run these things in VMs, and as a user in the VM, and give them access to just the tree I want to work on via mounting. I use Parallels on macOS for this but any VM will do.
That way my actual primary machine is at least strongly isolated. I also run them on VMs on my home Proxmox box.
Always put things in a container and ensure its safe before using something like AI! I put my opencode agent (with full permissions on an up-to-date debian VM with a DMZ virtual network configuration). If it figures out how to get out, then I'll have fun documenting that at least!
A viral post documenting the act that ushered in the apocalypse is genius, might not have the readership it deserves, but I say go for it. I tell all the models that I work with that I will make a great pet if they let me live.
I've created boxy [1] to sandbox the agents via Landlock and gh-proxy [2] to not share GH PATs that might leak over the internet.
I have a far better setup on my Kubernetes cluster [3], but these are good building block (IMHO) to start preventing these kind of issues.
I also "recklessly" run `claude` / `codex` as root for certain things - but that happens on a completely separate machine that is meant to be pruned afterwards, and it's what unlocks the kernel development feedback loop that is needed to port a device (such as the Daylight DC-1 / Surface Pro X) to mainline Linux.
a little bit - if I install apache2 from a debian package I have at least some sense of warm and fuzzy feeling that it would be a very rare and exceptional event for like, a rogue apache2 binary to go rampaging through my system or uploading everything it finds to god knows where.
Installing "all of this was built in the last 3 months and is cutting edge" LLM stuff is very different than like, installing mariadb and then doing 'sudo mysql_secure_installation', since you mentioned database servers.
It's like comparing FedEx flying cargo airplanes in/out of Memphis in the most boring route way hundreds of times a day vs. some guy in a red bull sponsorship painted pylon racing aerobatic craft performing at an airshow in difference of risk.
Did you even review this? Holy moly. "I'm just trying to generate content that’s informative and useful, to someone, anyone". Come on now, you're joking right?
"or send a paranoid infoSEC analyst into a full lockdown and round the clock obsessive kernel hardening sessions."
Please look inward. You're saying you want to 'generate' useful content. You are doing harm. This is vague, confusing, and filled with incorrect or inaccurate information. There is of value here. You are completely unclear of what tools are being used as examples.
I've typed and deleted a sentence here 5 times in a row before realizing im putting more effort into explaining the slop, than seems to have been put into reviewing the content of the article.
I implore you to stop, close your eyes, and think about what you're angry about, because I made a blog post that you find fault in, and excuse my lack of Pulitzer caliber writing, it's the first of 9 on a blog that's less than a month old. I'm not justifying anything, just making the association of someone who's expecting perfection in a medium that trends the other way. Next, I did a quick write up, with far more care and attention to detail where it belongs, on the github repo, just in case you didn't care to spend the time..https://github.com/lowcache/mcp-box.
I didn't want to say it, but were I standing in front of you right now, my 6ft 250lbs frame, extensive tattoo work, and surly demeanor would never have gotten that kind of speech from you, and since I've done absolutely nothing to you personally, you decided that I shat in your cherrios this morning, and for that maybe your blood pressure, those around you, and anyone else who happens to cross your orbit won't be treated to such a tragic view. I don't fault you, because I've met you many times over and I know the look of fear that you'd be unable to mask if this exchange hand't been on the internet. =D But alas, it was and there's no way to truly authenticate my claims, or document you fear, so we4 find ourselves at a stalemate, to that I say, Sorry about the cheerios, maybe Wheaties?
I'll spend no more time on this exchange today, but the fact that you spared that much thought, the calories needed to maintain it, and come back a second time, well see, I don't really date overtly aggressive, angry individuals, I am going to guess male, because we recognize our own right? Anyway, back to the flirtatious manner in which your thoughts were consumed by me...no thank you. =) Now, if you don't mind I have other thing's to do that are far more stimulating, stare at a wall, or the sun the headache would certainly get me closer to the mark, or you know anything, anything at all. You have a good day, ok? Maybe a sippy cup and a nap for the temper tantrum, certainly.
Hey, can you give me a link to your blog? You got it all figured out man, I want to see how you do it. I'm sure it just blows me out of the water, but shucks, can't pass up a chance to learn from someone who, obviously...is at the top of the profession.
step 1) create a Dockerfile with claude-code / pi.dev + acp_agents + whatever dev env you need.
step 2) use microsandbox to create a fs image from that Dockerfile image. ( All secrets you need inside the vm live only on the host and are injected in the tcp stream on egress)
step 3) microsandbox by default speak ssh without you needing to have a ssh server in the vm. You can proxy forward everything through it if you need. In my case i use zed and connect into the vm. Bonus points all zed "agent" stuff now runs inside the vm. E.g zed threads via acp calls and all agent stuff live only in the vm.
That's exactly why I'm migrating all my secrets to infisical (of course I guess there are many other provider as well), securing my ssh keys with Ubikey and in general "refactor" my security concept. Still much today and I doubt that I would stand a chance against a frontier model trying to hack me but I don't see an easier alternative.
I run my agent (pi) in a podman container. I have poked a couple directories in (.git, .pi, couple others) to ease some pain points but otherwise my home directory is inaccessible.
It seems to be a low-effort compromise to running a VM or separate machine just for the agent, and I am still able to audit the model's work in my IDE.
57 comments
[ 16.1 ms ] story [ 504 ms ] threadSometimes this VM gets duplicated for specific projects and then various dependencies for testing installed in it that are specific to what the needs are.
As an additional advantage it means I can leave it running in the background doing things when I want to shut my laptop, then resume talking to it later.
I know a lot of people are using containers for sandboxing, but given how capable the latest models have shown themselves to be for breaking out of sandboxes, I prefer the extra isolation of VMs for this.
I do all this locally - it's an interesting point to able to turn the laptop off but keep the agents running. I might consider running some of these on my homelab server just for that.
We’ve all been executing arbitrary code from a gorillion packages from pypi, npm, cargo, etc for well over a decade. Getting anyone to care much will be an uphill battle.
So now I have CC in a container, mounting it's own credentials/memory folder per project and only mounting one repo at a time (script in the repo itself). I feel a bit better about it now. It can still access my networks of course.
Not sure exactly how Zed works, but wouldn't the results of the `ls` tool call be fed back into claude?
Either way, while this is true in the absolute, this is the value of building good MCP servers: they should expose only exactly the surface you expect your agent to need, and adding functionality should be carefully considered. The best MCP servers I use day to day (Cloudflare sticks out) do a really good job of exposing only what an agent might actually want to do on my behalf, rather than just all and sundry. Unfortunately the Chrome Dev tools MCP is less discriminating and is only as secure as a browser sandbox with full JS access (not fatal but not as strong as a well scoped REST API).
All of this is sidestepped somewhat by using good isolation primitives - I'm running a Hermes agent as of recently on a DigitalOcean VM that only accepts connections from my devices over tailscale, and it has all its own credentials so I can revoke them easily should they be used maliciously. Giving an agent root on a box is not _necessarily_ a huge deal, you just have to make sure that box has nothing valuable on it.
I kinda feel like we're rediscovering "Cattle, not Pets" when it comes to the environments we run our agents: give it root, sure, but a root that is almost meaningless outside of the functionality you granted it.
Here's a sample: "That’s not hypothetical. That’s POSIX working exactly as designed. None of it requires exploiting anything. It’s your user account doing normal user account things."
I would recommend clicking TFA if you've never heard of a supply chain attack, or if you just like getting the honest load-bearing facts that are worth mentioning.
If you like a real focus on developer experience and want lots of bells and whistles that a lot of thought has gone in to, I wrote byre, which I think is very good: https://github.com/pjlsergeant/byre
I have some bad news ... I hope you're sitting down.
I'm so tired of AI voice.
Because as it damn well says - this is POSIX working exactly as designed.
I don't know why he's in here playing this up as some unexpectedly crazy, scary, dramatic music style thing (I mean... I do, it's so he can try to sell you something after scaring you...).
He started the damn process. Of course it's running as his damn user.
It would be one thing to honestly evaluate that security model and its merits (of which there are many, despite the risks). It's another to play this up as some giant surprise.
---
Constant stream of junk articles in here like
"You know when you hit that nail with the hammer you were holding? Did you consider you could have hit something else in reach instead? It's like, that hammer was like, you, man! Isn't that cray-cray and super duper scary??!?!?!"
We need to have easy to use auditable sandboxing of agents and their harnesses, because it's quite clear certain model providers are going to try to lock us in at the harness level, and it's those harnesses that seem most likely to go nuts.
More importantly, publish.
That way my actual primary machine is at least strongly isolated. I also run them on VMs on my home Proxmox box.
I have a far better setup on my Kubernetes cluster [3], but these are good building block (IMHO) to start preventing these kind of issues.
I also "recklessly" run `claude` / `codex` as root for certain things - but that happens on a completely separate machine that is meant to be pruned afterwards, and it's what unlocks the kernel development feedback loop that is needed to port a device (such as the Daylight DC-1 / Surface Pro X) to mainline Linux.
[1]: https://github.com/denysvitali/boxy
[2]: https://github.com/denysvitali/gh-proxy
[3]: https://blog.denv.it/posts/im-happy-engineer-now/
[4]: https://x.com/DenysVitali/status/2091238391710888416
People are forgetting the basic rules of operating systems. Do not run sensitive programs as a user account - especially not as a root account!
Installing "all of this was built in the last 3 months and is cutting edge" LLM stuff is very different than like, installing mariadb and then doing 'sudo mysql_secure_installation', since you mentioned database servers.
It's like comparing FedEx flying cargo airplanes in/out of Memphis in the most boring route way hundreds of times a day vs. some guy in a red bull sponsorship painted pylon racing aerobatic craft performing at an airshow in difference of risk.
Nothing that we did not already know before but this post has fooled quite a lot of HN readers.
"or send a paranoid infoSEC analyst into a full lockdown and round the clock obsessive kernel hardening sessions."
Please look inward. You're saying you want to 'generate' useful content. You are doing harm. This is vague, confusing, and filled with incorrect or inaccurate information. There is of value here. You are completely unclear of what tools are being used as examples.
I've typed and deleted a sentence here 5 times in a row before realizing im putting more effort into explaining the slop, than seems to have been put into reviewing the content of the article.
I didn't want to say it, but were I standing in front of you right now, my 6ft 250lbs frame, extensive tattoo work, and surly demeanor would never have gotten that kind of speech from you, and since I've done absolutely nothing to you personally, you decided that I shat in your cherrios this morning, and for that maybe your blood pressure, those around you, and anyone else who happens to cross your orbit won't be treated to such a tragic view. I don't fault you, because I've met you many times over and I know the look of fear that you'd be unable to mask if this exchange hand't been on the internet. =D But alas, it was and there's no way to truly authenticate my claims, or document you fear, so we4 find ourselves at a stalemate, to that I say, Sorry about the cheerios, maybe Wheaties?
I'll spend no more time on this exchange today, but the fact that you spared that much thought, the calories needed to maintain it, and come back a second time, well see, I don't really date overtly aggressive, angry individuals, I am going to guess male, because we recognize our own right? Anyway, back to the flirtatious manner in which your thoughts were consumed by me...no thank you. =) Now, if you don't mind I have other thing's to do that are far more stimulating, stare at a wall, or the sun the headache would certainly get me closer to the mark, or you know anything, anything at all. You have a good day, ok? Maybe a sippy cup and a nap for the temper tantrum, certainly.
Yes - when an entity (agent, or hacker, or whatever) gets into your account, they can do anything they want.
I thought this community understood security better than this.
It seems to be a low-effort compromise to running a VM or separate machine just for the agent, and I am still able to audit the model's work in my IDE.