If you use Claude Code seriously, you've hit this wall:
You give Claude a complex task, switch to another window, come back 20 minutes later — and it's been completely frozen the entire time, silently waiting for your approval. No notification. No sound. No focus switch. Nothing.
It kills your productivity. It kills Claude's momentum. And it happens constantly.
So I built claude-alert.
It hooks directly into Claude Code and fires a native macOS notification the instant an approval prompt appears. Your terminal snaps into focus automatically. You approve in two seconds and Claude gets straight back to work.
Why this matters:
The whole point of Claude Code is autonomous work — you shouldn't have to sit there watching it. But right now, stepping away means gambling on how long Claude will silently idle before you notice. For long tasks this can mean losing 30–60 minutes of compute time because you didn't catch a single prompt.
claude-alert closes that gap. Walk away. Do real work. Trust that you'll be called back the moment you're needed.
What you get today:
- Native macOS banner notifications (actual system notifications, not a third-party overlay)
- Sound alerts so you hear it even if you're across the room
- Terminal auto-focus so you're ready to act immediately
- Full local audit log of every approval event at
~/.claude-notifier/audit.json - Quiet hours support so it doesn't ping you at 2am
- Zero telemetry — everything stays on your machine
Install in one line (no Xcode needed):
npx claude-alert install
There's also a Tier 2 install (requires Xcode) with an animated robot in your menu bar and richer notifications.
What's coming next:
The feedback so far has been clear — people want more granular control. So here's what's in progress:
📋 Per-command allowlist
Whitelist the specific commands you personally trust. Maybe npm install is fine to run silently, but you always want to see anything touching git. You define the list — nothing else slips through. This is deliberately surgical, not a blanket bypass. A "trust everything" switch would defeat the entire purpose of the tool.
📁 Project-level config
Your personal side project and a client's production codebase shouldn't share the same rules. With project-level config, drop a .claude-alert file into any repo to override your global settings. Strict on prod. Relaxed on experiments. Claude adapts to the context automatically.
What I'd love to know:
- Does the silent-pause problem resonate with you? How much time have you lost to it?
- What commands would you put on your personal allowlist?
- What would you configure differently between projects?
- Anything else that would make this a must-have for your workflow?
This is still early and your feedback will directly shape what ships next.
GitHub: github.com/wilbert-t/claude-buddy
Appreciate any feedback, even harsh 🙏
Top comments (0)