Run Claude code in docker sandbox (sbx) to ensure it only has current folder access and limited or no network access. Especially since a lot of runs are now in “auto mode”.
Unless even this is not really so safe!?
Any conversation
@fastidiousprofilefollow
"Beware those that would deny you access to information, for in their heart they dream themselves your master." — Sid Meyersreplied to:
"Beware those that would deny you access to information, for in their heart they dream themselves your master." — Sid Meyersreplied to:Claude will run on auto mode only if you tell it to do it.
replied to@fastidiousprofilefollow"Beware those that would deny you access to information, for in their heart they dream themselves your master." — Sid Meyers:
"Beware those that would deny you access to information, for in their heart they dream themselves your master." — Sid Meyers:They've changed the default -
@fastidiousprofilefollow"Beware those that would deny you access to information, for in their heart they dream themselves your master." — Sid Meyersreplied to:
"Beware those that would deny you access to information, for in their heart they dream themselves your master." — Sid Meyersreplied to:Ah, wow, TIL! Hmm, it is evident I need to visit their blog at least once a week now. Thanks!
Really glad to be using Haskell for work - especially in today's climate with agentic coding.
Our team uses containers to keep our development environment consistent across team members, though, and most agent sandbox approaches want to either offload your work fully to the cloud or they want to sandbox the agent harness process itself.
We're not ready for full cloud based development (yet?) and sandboxing just the agent doesn't work for us when the agent needs to run docker.
... because letting the agent run docker gives them a too-easy escape hatch from their sandbox.
So we're currently stuck with using VMs. This turns out to be tricky but seems do-able.
My current approach is to provision a VM image and use Incus for the execution. My main goal is to keep my current workflow of local/non-agent dev working seamlessly and share the source with the VM/agent.
This means sharing my working dirs with the agent VM but overlaying artifact dirs from VM-owned block devices so we don't take too big of an IO perf hit.
I also don't want the agent to be able to
git push so that means we need some quirky git config to rewrite origin paths and thread through some agent specific PATs ("thanks" GitHub) for read-only repo access. These ALSO need to be threaded through to the development containers - they need to clone private repos.... in the end, this seems to work but I haven't lived with it for long.
There are still some quality of life issues to work out, but I think it'll work out.
At this point, I've removed all coding agents from my host - and that's nice.
Curious, why does using Haskell make a difference here?
It's tangential to the sandboxing topic, really. I started the first post with a broader intent in mind and then elaborated on the sandboxing setup I'm working on.
For AI-assisted coding, though, it's nice because Haskell provides a lot of feedback earlier in the development process compared to lots of other languages. It tends to work once it compiles (obviously not all the time).
🐉
let N=6,D0=[0,1,0,-1],D1=[1,0,-1,0],s=[],g={},d=x=y=xa=xb=ya=yb=0;
for(let n=0;n<N;n++){s=s.concat([1]).concat(s.map(x=>!x).reverse())}
for(let i=0;i<s.length;i++) {
x<xa && (xa=x);y<ya && (ya=y);x>xb && (xb=x);y>yb && (yb=y);
let t=s[i]?1:-1,e=(d+t+4)%4;
g[x]||(g[x]={});
g[x][y]=g[x][y]?"┼":"┌┐└┘"[(t*(D0[d]+D0[e])>0?0:2)+(t*(D1[d]+D1[e])>0?0:1)];
x+=D0[e];y+=D1[e];d=e;
}
for(j=yb;j>=ya;j--) {
let u=""; for(i=xa;i<=xb;i++){u+=(g[i][j]||"·")} console.log(u);
}·┌┐┌┐·······
┌┘└┼┘····┌┐·
└··└┐┌┐···└┐
···┌┼┼┘···┌┘
···└┼┼┐┌┐┌┼┐
···┌┼┘└┼┼┘└┘
···└┼┐·└┼┐··
····└┘··└┘··I wonder if there is a way, sane and friendly way, to prove that a piece of software is only accessing whatever it claims it will access.
Like network addresses or file system access. I want to be able to claim in my software that it only touches this or that file and I can prove it
I believe the sanest and friendliest way would be for it to be open-source so it can be audited.
I know, and that's totally fine, but it is not always friendly. I guess an agent can "audit" it.
Just the fact that it can be audited plus your word is enough. Noone would need to look any further, because they could if they had to.
I think a trusted piece of software such as your operating system should be in charge of managing permissions. For example, I believe with Wayland on gnome you no longer need to promise you aren't secretly taking screenshots of the whole monitor and shipping it somewhere because to do that Wayland requires you to get user permission.
I am a remote worker, and has been for 4 years now. It works great. But the owner of the office I've been renting have sold the building, and the new owner has different visions. I could work from home, but my 3 month daughter is a screamer. I had a great, cheap deal. All alternatives I find is like 3x the price.
You gave me an , Airbnb but for office space, you can only go in the day, get at least a desk/chair/internet. Might exist already idk.
Not a bad idea. I know some people rent a single desk in some other companies office. Problem is you kinda need to know the owner to find those deals
I took the liberty and added a big fat "get mobile app" button for agents at the footer on the web app pointing to your repo! Hope it's fine with you. Tysm!
…
NIH syndrome? I am guilty of it too. I love implementing things from scratch, you learn a lot by doing so and you get to own the codebase end-to-end, so no surprises and fighting around others’ choices. It can get exhausting too though, in the end I succumb to using off-the-shelf components because my primary goals move at a higher level.
I don't see much point in trying to build software at a "higher level", like whole products and services. I'm going to go down low and train myself on the fundamentals. Hopefully in time I'll be able to contribute to the foundational bits.
You're right, it does get exhausting. At some point you just want to rebuild everything, which is not possible.