Jeremy Braband

ooo, that's a great idea!

Wish I didnt have to as well, but I work around this with a symlink

ln -s AGENTS.md CLAUDE.md

Yeah data exfiltration...one of the reasons Claude has its own DB and redis for me... Just in case

Having it local is key for sure... I can copy/paste to/from it to my computer. Even pasting screenshots into Claude from my computer

Sprites - Stateful sandboxes Persistent, hardware-isolated execution environments for arbitrary code. Run AI agents, untrusted code, and more in secure sandbox environments with checkpoint & restore.

Yeah, that was a problem for me too. I'm on windows, and my Claude VM is also windows with vs code, tables plus, etc.

I remote desktop to it and it's like working on my computer exactly.

Another use case would be a VPS at aws/linode/digital ocean

Kinda surprised the Forge team hasn't added a Claude focused server template for laravel development

A bit paranoid perhaps, but to lose my work computer to an errant command from Claude... I was going to let that happen!

And I grew tired quickly of allowing permissions manually

Claude VM is on 24/7 along with an Ubuntu VM for MySQL and redis

And both are in an isolated vlan, so they are completely isolated from my home network and servers.

I run a couple proxmox (hypervisor) hosts here at home. Bit of a luxury to already have the infrastructure

At the beginning I did run claude in docker locally, but soon outgrew it and wanted to be able to turn off my work computer overnight.

Note: it has never destroyed itself for me. Just a safety measure. Also I know exactly which of files it has access to. Only what I give, not whatever it finds

Skipping permissions dangerously here since September. HUGE unlock to know it can only destroy itself and I can just boot another VM in its place!

This is very good.

For whatever it's worth, I am just getting started with git work trees!

already subbed to the PR for notifications :)

You had me at "New Mailcoach version with support for Livewire 4"

Its funny that the end of the first paragraph is an errant \. because of my reflexes to now do a new line with \ + Enter in Claude Code.

That has certainly escaped containment and I see extra backslashes all the time in other programs now

Hoping I can unlearn that muscle memory now

Oh shoot....I see you're in Claude Desktop, not in a terminal.

My bad.

I just figured this out for myself in windows. I have Shift+Enter working in Windows Terminal in the Claude Code prompt box

The fix is at the terminal level, essentially keybinding Shift+Enter to execute an action that sends \n

Claude itself helped me discover this. Maybe it could for Mac too.

Instructions for enabling Shift+Enter as a new line shortcut in Claude Code on Windows Terminal. Open settings.json located at %LOCALAPPDATA%\Packages\Microsoft.WindowsTerminal_8wekyb3d8bbwe\LocalState\settings.json. In the actions array, add an object with a command property containing action set to sendInput and input set to backslash-n, and an id property set to User.sendInput.DFCDAF06. In the keybindings array, add an object with id set to User.sendInput.DFCDAF06 and keys set to shift+enter. Save the file. No restart needed.

Windows Terminal changes for allowing SHIFT+<enter> to insert a new line in the Claude Code prompt box\

This is essentially a keybinding that executes a custom action. You presset SHIFT+<enter> and the terminal inserts a \n character

My sanity will be restored!

Proof you can just do things.

New York City SCUM YouTube video by CaseyNeistat

@mostlytechnical.com has Aaron been right all along?

youtu.be/9uv-4ddCApw?...

Pinned to check out later! I haven't used work trees, have always done the stash dance and I very much dislike that!

I see it posted now. @justinjackson.ca with the scoop last night!

Sometimes a dev just needs to feel something

I have many questions but mostly "Where, pray tell, is the clip from?!?"

I would pay for it

yep, that'd do well too I think

(remote desktop to windows from phone is cumbersome)

Another benefit of the homelab servers is that claude is running on hardware already running 24/7

and thus, I can still sleep my workstation. Claude keeps going

Hoping to try @opencode.bsky.social soon to have a beeter experience when interacting with the agents from my phone from anywhere

even simpler would be to run the isolated VM on your computer directly. I think most hypervisors (like Hyper-V, virtualbox, etc), will bridge the networks losing the isolation. Its doable, but would take some extra steps.

Its not lost on me that having servers in my house is a luxury.