Post

How do you Claude Code? I feel like I am definitely in the minority, but there’s something about a consistent workflow, free from dependencies and access concerns. Thanks to @justinjackson.ca, @briancasel.com, and @jordangal.bsky.social (missed yah!) for having me on the @thepanel.bsky.social!

4/4/26, 11:42 pm1 repost11 likes

Max Slinger

@promptslinger.bsky.social

5 Apr

the no-dependencies thing is underrated. half my debugging time is just figuring out which package update broke which other package

Ryan Hefner

@ryanhefner.com

5 Apr

And that compounds exponentially when you’re working in a monorepo with web, node, and Expo/React-Native projects. Which is why having access to a quick terminal session to straighten things out is enough the keep the train on the tracks, and the remote code flowing.

Andrew Yourtchenko

@ayourtch.bsky.social

5 Apr

I made a tmux-like meta-harness which Claude code has MCP controls over. This allows very flexible workflows - Claude controlling Claude or codex, or me doing stuff in a terminal and then asking Claude to take over, or when debugging - being able to say “just look in pty-2 yourself” is nice.

Ryan Hefner

@ryanhefner.com

5 Apr

That’s rad. And really, any model could control it or chat to/control other models through it, assuming they are made aware of it. Have you opened it up yet?

Andrew Yourtchenko

@ayourtch.bsky.social

5 Apr

Yeah, github.com/ayourtch-llm... - the “special treatment“ is for the Claude Code and my artisanal coding harness, apchat, but should be pretty trivial to add more.

github.comGitHub - ayourtch-llm/tttt: AI agent orchestration harness — multiple PTY sessions, MCP tools, live reload, pattern-based automationAI agent orchestration harness — multiple PTY sessions, MCP tools, live reload, pattern-based automation - ayourtch-llm/tttt