A few days ago I kept having the same thought every time I used Claude Code: the agent is doing a ton of work, and there’s nothing good to look at while it does it.
Token usage, context filling up, which skills fired, what tools ran, when it paused, when something broke: all of it real, all of it buried in scrollback. So I’d tail files, squint at the terminal, and hope.
That bugged me more than it probably should have. That feeling became Conan.
I wanted something I’d actually want to look at
Here’s the honest version: I love a good dashboard. Charts, gauges, a little telemetry ticking across the screen. I find it genuinely fun to watch. And I live in Claude Code all day. So what I really wanted was something pretty sitting next to me while the agent works: not a tool I’d check when something broke, but a view I’d want to keep an eye on.
Before I wrote any code, I wrote down what it should be: a terminal-primary app that wraps Claude Code and surfaces what it’s doing, right where you’re already working. Beautiful enough to stare at, useful enough to earn the screen space.
A “simple” app is never simple
Nobody warns you that a “simple” paid app isn’t actually one thing. It’s the desktop app, a marketing site, and a little service that hands out licenses: three separate projects, each with its own personality and its own way of surprising you.
The marketing site was the one I underestimated. I figured it’d be the easy afternoon; it became the rabbit hole. And the part I’m still grinning about is the launch video.
I didn’t want to learn a video editor, so I reached for Remotion, a framework all its own that turns React into real, rendered video. That meant the trailer became code: components, props, timing I could tweak and re-render like anything else in the repo. Change a line, run it again, new cut. No timeline to fight, no keyframes to nudge by hand.
The clutch part was getting Claude Code to drive it well. I wrote a skill (basically a playbook that teaches the agent how to build Remotion compositions the right way), and after that the trailer was mostly prompting. What I’d have lost days to in After Effects took an afternoon. And there was something perfect about it: the tool I built to watch Claude Code work, with its own trailer made by Claude Code.
Did watching it actually change anything?
More than I expected. There’s a 60% rule a lot of people who live in these agents swear by: once context creeps past ~60%, you hand off or compact. Not because the agent hits a wall: it doesn’t crash, it just quietly starts to rot. It stops following instructions, contradicts itself, drops details you gave it earlier. I’d nodded along to the rule for months; now I actually follow it, because I can finally see the context climbing toward the line. And I know whether my skills fire on their own, instead of assuming they did and hoping.
That’s the funny part. I built it because I wanted something nice to look at, and the looking is what changed how I work. Turns out pretty and useful were the same feature all along.
Today it’s on Product Hunt
From a vague itch to a signed, shipping product. It took longer than I wanted, broke in more places than I expected, and I’d do it again tomorrow.
If you live in Claude Code, come see what your agent’s actually up to.