What's the first thing you'd send your coding agent from your phone?

Caraka goes live tomorrow, and I'd rather ask this before the launch than after.

The problem I built it for is small and stupid. I start a task with Claude Code, walk away from my desk, and come back hours later to find it stopped at the first thing that needed my approval. Once it waited seven hours for a one-word answer.

Every workaround I tried technically worked and was miserable: SSH from my phone, remote desktop, tmux. Typing terminal commands on a 6-inch screen is a punishment, not a solution.

So Caraka is deliberately small. It is not another assistant. No model, no agent loop, no tools of its own, no extra API bill. You send a task from Telegram, Discord or WhatsApp, and the coding agent already installed on your machine does the work. Every write or command needs a single-use approval that expires, and plain chat text can never approve anything.

I asked this same question on LinkedIn two days ago. Three people who don't know each other converged on the same answer, and it wasn't what I expected. They didn't want more control. They wanted to see. "Task complete" tells you nothing if you can't look at the result. That reframed my roadmap in a day.

So I'll ask here too, and I'll read every reply:

What would you actually send your agent while you're away from your desk? And what would you need to see back before you'd trust the answer?

MIT, self-hosted, npx caraka init.