TokenBreak — Move while Claude resets
Claude has token limits. When you hit them, you wait: 5 minutes for a session pause, 15 if you hit the 5-hour cap, 30 if you hit the daily ceiling. I turned that dead time into something useful.
Dead time that happens
multiple times a day
Claude has token limits. When you hit them, you wait: 5 minutes for a session pause, 15 if you hit the 5-hour cap, 30 if you hit the daily ceiling. This isn't a bug — it's the system working as designed.
The problem is what happens during that wait. Most people stare at the screen, open another tab, check their phone. They stay in the same physical spot, in the same mental state, doing nothing productive. And that pattern repeats several times a day if you work intensively with AI.
That time already existed. It just needed a better destination.
The idea was simple enough
that it took one day to build
If you already know how long you have to wait, you can receive an exercise routine calibrated exactly to that time. No equipment. Without leaving the room. Without installing an app.
The user hits the limit, opens TokenBreak, picks how much time they have, and gets a sequence of exercises across 4 categories: mobility, strength, breathing, and eyes. Each category is designed to counteract exactly what happens to the body after hours in front of a screen.
The routine is randomized every time and never repeats two exercises from the same category back to back. 43 exercises in the pool. Zero friction.
Three screens.
No account, no app, no friction.
The entire experience is three states:
Available in English and Spanish with a language toggle in the header. Both versions have their own URL and independent SEO within andresbuonaiuto.com.
How it was built:
no new dependencies, in one day
TokenBreak lives inside andresbuonaiuto.com as a React component embedded in Astro pages. It added zero new dependencies to the project — React and TypeScript were already there. All styles are inline. No CSS framework, no router, no global state.
The public GitHub repo also includes a skill.md: a prompt ready to paste into any AI (Claude, ChatGPT, Codex) that regenerates the full app from scratch, with color and font customization built in.
Why it's on GitHub
and not just on my site
The Claude user community is technical. Many have their own sites or want a customized version. Publishing the repo as a standalone with clear instructions has more reach than keeping it closed.
The skill.md is the most interesting part: anyone can copy it, paste it into their favorite AI, answer three questions about colors and fonts, and have the app running locally in under 5 minutes. That's native distribution for the agent era.
TokenBreak isn't a startup. It's a small tool that solves a specific moment in the AI work day. That also has value.