Running Claude Code on my own machine was easy. Getting it onto shared compute my whole team could trigger was the hard part. There's plenty written about the local side. A lot less about the team side.
I made that move because of how a broken deploy plays out for us. I'm the only DevOps engineer on my team. A CloudFormation deploy fails. A Slack notification fires. And more often than not, someone pings me to ask what went wrong.
I get why. AWS isn't everyone's day to day, and a CREATE_FAILED event with a rollback behind it isn't the friendliest thing to read. The pings weren't the real problem, though. A broken deploy that hinges on one person doesn't scale.
So I decided to build my way out of it. I'd give the team a starting point on a broken deploy without pinging me. It wouldn't fix the problem, but it'd tell them what broke and where to start.
What I built
