Remote-control Claude Code from your phone with Cowork

The setup

Cowork puts a chat box on my phone that dispatches tasks to the Claude Code already running on my machine. That machine is set up the way I work: binaries authenticated, git aliases in place, opusplan as the default model, and deploy webhooks wired up. I send an instruction from anywhere — a coffee queue, a train, bed — and it runs against a real checkout at home.

What I sent, and what the agent did

I sent one line from my phone:

for my imsj front page project, change the appbusleh.imsj.dev to
busleh.imsj.dev and use gg to commit and push

It didn’t recognise “imsj”, so instead of editing the wrong repo it asked:

Don’t see an “imsj” project among your known Claude Code workspaces… What’s the path to the imsj front page repo?

I replied imsj-frontpage. It found the repo, swapped the domain, and committed and pushed with my gg alias:

Starting a code task in /Users/jason/workspace/code/imsj-frontpage — swapping the domain and pushing via gg now.

So the whole thing was: I typed one line and answered one question. The agent did the edit, the commit, and the push on my machine, with my tools and my git config. I front-load the boundaries the same way as in making Claude Code smarter — do the change, don’t touch anything unrelated, ask if ambiguous, report back — because I’m not at the keyboard to course-correct. And because I run opusplan, Opus planned the change and Sonnet did the typing, which is why a terse one-liner still landed as a clean, focused commit.

I couldn’t watch it — the webhook told me it worked

Dispatching from my phone, I had no real visibility into what the agent did. I could read its summary, but I couldn’t watch the work. What confirmed it wasn’t the agent saying “done” — it was the deploy pipeline the push triggered, which pinged Discord when it finished:

A green Success on main, the commit it produced, and a link to the run. When you hand work to an agent you can’t observe, the pipeline around it — CI, deploy notifications, health checks, rollback — is what tells you the truth. Proper DevOps matters more than ever when you only ever see the effects.