Post

Hillel
@hillelwayne.com

This job is the first time I've had unfettered access to Claude code. So, uh, how do people like to customize their Claude rigs? Special skills, hooks, other stuff? I'm not great at using programs in their vanilla state

0 reposts12 likes

djeikyb.bsky.social
@djeikyb.bsky.social

similar boat, and i'd love to read more experiences of folks letting claude execute things locally but isolated (vs renting a vm eg ec2). my co-workers seem to be giving heaps of permissions, which doesn't suit me

Chraz
@edu.runlinea.com

Highest leverage customization is a deny list before any skill pack. Migrations, lockfiles, anything under infra/. Most vanilla pain is the agent editing the wrong surface.

Amyville Hoyrror
@amyhoy.bsky.social

matt pocock's skills are surprisingly useful (given how simple they are) and i also use the ed3d plugins for more structured/critical work. they really do up the code quality. i added a ccstatusline plugin to track session context

Sviatoslav Abakumov
@perlence.bsky.social

If you have a thousand line init.vim, I think you'll love the Pi harness. The mind expanded moment for me was when I learned that the clanker can write extensions for itself. It also works with enterprise (paid per token) Claude subscriptions AFAIK.

Barry Kelly
@barrkel.bsky.social

Don't forget to ask Claude these things also. Don't spend time tweaking things. Tell Claude to to tweak things. E.g. tell Claude to write a skill. Get Claude to evaluate the skill in subagents, reflect on what well and what didn't, and iterate on the skill.

FlowRMM
@flowrmmbyflowdevs.bsky.social

Unfettered access is exactly when I want a hook that shows the exact shell or file write and waits. Skills help after that; the first customization is making the agent stop so you can see what it is about to do on the box.