I’m writing this from my phone, connected to the dev server from my previous post.

I’m using Wispr Flow to dictate into Terminus, an iOS terminal app. Terminus is SSH’d into a Vultr VM over Tailscale, running Claude Code in a tmux session.

The pieces: I already had the VM. SSH works over Tailscale from iOS. tmux keeps sessions alive across disconnects. Claude Code has a --dangerously-skip-permissions flag for unattended operation. Together, I can do development while I sleep.

I keep multiple tmux tabs open for different projects. Each one maintains its own context.

Integrations

Claude connects to GitHub through the API. I file issues from Slack and Claude picks them up. It runs web searches, executes code locally, and reports back. When something finishes or needs attention, I get a Slack notification. I can respond from there.

I’ll be out somewhere, think of something I want to change, and say it. By the time I’m back at my desk, it’s usually done.

Ideas I’m exploring

Visual feedback through GitHub issues. Claude could fetch a webpage, screenshot it, and post the image as an issue comment. I’d see the actual page instead of relying on Claude’s description.

Application development snapshots. Claude could screenshot the app at various stages, populate it with test data, and post the images to an issue thread. A visual history of the build that I can scroll through and discuss.

The primitives for this already exist. I’m still figuring out which workflows are worth building.