Seems to work so far. I've manually checked various things, rw access, access to sensitive folders/files etc.
So far, I've only used it on two small projects. My major projects I've been clicking through prompts and recently switched to "auto" mode.
I'm not quite sure why anyone would trust "--dangerously-skip-permissions". I've seen these agents go off-piste far too many times, installing unnecessary packages, environments, calling sudo & creating files in all sorts of places.
There's a page on various sandboxing strategies on their site:
I've seen various comments in several topics with folk baking their own sandboxes. That's great. Although I'd be inclined to try Anthropic's solutions initially.
+1 for gpakosz/.tmux which I also wanted to mention here. I have it synced across all machines and a personal .tmux.conf.local that modifies some settings to my liking.
I'm not sure I trust this. A quick search finds a Psychology Today article about it along with a single reference. I lazily suspect the result is based on some type of questionnaire.
The way "chain of thought" is used in LLMs to improve reasoning demonstrates, to me at least, the value of capturing intermediate steps in some rich compressed structure. Nothing beats that than words and sentences (see them or hear them). A lot of ideas can't be captured with just photos alone imho.
I'm struggling to believe you have a dozen new vacuum cleaners in your shed. It's quite an extraordinary claim. Are you willing to share some evidence?
Sure. It's cold, rainy, and midnight but here's what came up in my email when searching "vacuum". It's not all, and you can see some I didn't reply to but -
I was looking forward to seeing 12 neatly stacked & boxed vacuum cleaners in a dimly lit corner of your shed ;-)
That said, thanks for sharing the emails/headers.
It's curious that Amazon hasn't flagged you for purchasing & reviewing multiple similar items in such a short span of time. I would imagine it would be quite easy to spot someone who's bought and reviewed 12 vacuum cleaners in a 2 or 3 year window.
I feel I'm in the same boat. For several months I've been thinking my GPU was on its way out (it's a pretty old 2080 now). My desktop freezes randomly. I can log into it remotely but all the usb devices stop working and the screen goes blank. l took a good look at the logs and noticed a bunch of pageflip timeouts followed by usb disconnections. I later discovered the Nvidia forums seem to have many recent complaints (with similar logs) especially around their latest drivers and Plasma + Wayland compatibility.
In a, thankfully past, role working remotely for an antipodal organisation with badly configured networking; often, the lag between typing several characters on the keyboard and those characters appearing on my screen could be measured in seconds! Vi key bindings were a godsend as I could send commands (eg global search & edit etc) and be confident they were being applied before recieving a (delayed) visual update. I feel my experience seems to echo (albeit slightly) that of Bill Joy's vi development on a 300 baud modem!
https://github.com/anthropic-experimental/sandbox-runtime
This is combined with "auto" mode.
Seems to work so far. I've manually checked various things, rw access, access to sensitive folders/files etc.
So far, I've only used it on two small projects. My major projects I've been clicking through prompts and recently switched to "auto" mode.
I'm not quite sure why anyone would trust "--dangerously-skip-permissions". I've seen these agents go off-piste far too many times, installing unnecessary packages, environments, calling sudo & creating files in all sorts of places.
There's a page on various sandboxing strategies on their site:
https://code.claude.com/docs/en/sandbox-environments
I've seen various comments in several topics with folk baking their own sandboxes. That's great. Although I'd be inclined to try Anthropic's solutions initially.