A creative OS that runs beside Claude Code. While you build, it keeps the reasoning: the experiments, the visual design, the decisions. Then it composes them into pages you can open, read, and share.
solo build · research · design · code · sound
song • illusion by aether
Quest + experiments
A quest is the project. Experiments are the tasks you try inside it. As you build with Claude Code, the panel logs each one the moment it starts and marks it complete when it ships. You never stop to fill anything in, and the quest stays up-to-date on its own.
Keep
Keep is the part that stays yours. Screenshots, videos, and notes worth holding onto. Drop them in and they land in real files on your machine, filed to that build. Later they become the visuals that carry the brief and the case story.
The skill + the hook
The skill teaches the agent Unicorn’s language: quests, experiments, what to log and when. The hook runs at session start: open a terminal, type unicorn, and the agent has your quest and your files in front of it.
Everything Unicorn writes lands in real files on your machine, organized where you can actually reach them: the keeps, the build files, the data. Then the Mirror composes them into HTML you open like a page: a weekly brief and a living case story. There’s a reason for that: agents read markdown; designers don’t. So the agent gets its .md memory, and you get something you’d genuinely show another human.
Claude Code already saves every session, so the record exists. But a transcript is everything at once, not the decision, the fix, or the sentence that finally landed. What you actually need later: the handful of moments that mattered, ready to hand back to the agent as context.
With AI, a week of building can ship in an afternoon, so there is no “writing it up afterward.” By the time you could, you’re three builds past it. The documentation has to happen during. Unicorn splits that job: the agent logs the experiments automatically, and you keep the screenshots, clips, and lines worth holding. When it’s time to tell the story, the pieces are already filed.
Build for where the model is going, not where it is today.
Boris Cherny, creator of Claude Code. Watching his talk live at Claude Code’s 1st birthday, San Francisco.
When I heard this, Unicorn was an idea I had just started building. The line rearranged my thinking: stop designing inside today’s limits, because that is not the reality we build in anymore. So Unicorn was shaped for the agent it would grow up beside. The skill now fills the Quest panel while I build, and the Mirror composes the story as it happens. What the next model unlocks, I find out by building. That is the job now.
Design + engineer in one head.
The ultimate pragmatist.
I reached out to four design engineers on LinkedIn and interviewed each of them. All four had been guests on Dive Club, and every conversation gave me a real look at where product design is heading.
A few of the takeaways:
“Nobody is going to be as good at engineering as the AI. But if you do not know the principles, you cannot direct it. It goes off the rails.”
“I can ship in ten minutes. So now I have two weeks to make it the most delightful experience, not just bold the text.”
“How do you show growth? The work I made when nobody was making me do it.”
“Someone told me nobody knows how to do this yet. That became the leveler. I have as much of a shot at learning it as anyone.”
“I keep a vault. I treat it as my long term memory. If you have worked with AI at all, you know context is the huge thing.”
One of them named the gap exactly: you end up building a separate memory brain for all the context. It will be a solved problem someday. It is not yet. That gap is what Unicorn fills while you build, not after, so you are not digging for context later.

Product thinking. Documenting your own work mid-flow is friction, so Unicorn hands that job to the agent. It logs each experiment as it happens, and the panels update while you build. You never stop to fill them in. The panels live in the corners; the terminal stays in the center.
Keep anchors it. Screenshots, videos, and notes land in real files that live next to the work: one place per project instead of a desktop full of strays, readable by you today and by the agent next session.
The Mirror composes it back. Your experiments and your keeps become a weekly brief and a living case story: HTML pages you can open, send, and build a portfolio from. You chose every moment in them. Unicorn keeps the material together.
Unicorn’s language was cute, not sticky.
If it only entertained, it failed.
It accumulates while you build.
Catch what matters. File it to the quest.
Everything within reach. I use it daily.
I designed the pixel unicorn in Figma first, and the design didn’t translate. A terminal draws each character in a single color, and its cells are twice as tall as they are wide.
So I had Claude Code build a tool for designing inside those limits: paint a grid, preview the literal block characters a shell will print, export the result. The unicorn it produced became the whole brand: the terminal hook, the app icon, the installer. The tool grew into Critter Maker: pick two colors, name your critter, give it a voice, and print it in any terminal. The unicorn was just the first.
Everyone who tested it early landed on the same note: it should fill itself in. Stopping the work to document the work defeats the point.
The skill and the SessionStart hook are that alignment: the agent writes the case story as the work happens, a casestory.md for its own memory, composed HTML for yours. Every new model raises what the agent can do, and Unicorn grows with it, so the story writes more of itself each time.
Build the tool you need, even if it barely works today. The next model will meet you there.
Unicorn went further than I imagined. It’s a real damn desktop app; keeping the story so I
can stay in the work.
The clips on this page are from the real desktop app. The Unicorn preview is a clickable version to play around in.
open the unicorn preview