Quarterdeck

Why I built it

I run a lot of Claude Code agents at once, and for months the bottleneck was me. Not the models. Me. I couldn’t tell which of the five needed me right now without alt-tabbing through every terminal, and by the time I finished the loop one had stalled and another had guessed wrong on a decision it should have just asked me about. This is not a product the factory ships. It is a piece of the system itself: the thing I built so I could run a fleet of agents without becoming the bottleneck.

The moment

Every monitor for Claude Code stops at the same place: it shows you a status and goes quiet. That is the cheap half. The expensive half is the agent that needs a decision only I can make, and a dashboard can’t carry my answer back. So Quarterdeck adds the return trip. Any session can ask me a question through an always-on-top popup while it works, and my answer routes straight back into that session. A twenty-minute run stops and asks “which of these two migrations?” instead of guessing.

The hard part

Getting the plumbing honest. Status comes from Claude Code’s own hook events, not from polling or scraping transcripts. The ask channel is a local MCP server bound to your own machine, nothing leaves the box. The permission takeover is fail-open by design, so if the app isn’t running or I don’t answer, Claude Code just falls through to its normal terminal prompt and no agent is ever left stuck.

I wrote up the full argument, why a reach-back channel matters more than another dashboard, in Autonomous agents don’t need a dashboard. The code is open on GitHub.