How to use Claude to build a gym tracker from your phone
Agent Arnold grew from a personal workout split into a free gym tracker built with Claude Code. The author reviewed most code and used phone SSH sessions alongside daily app use.
news.ycombinator.com
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build [gym tracker] around [my workout split] using [chosen stack]. Add [requested routine features] and [AI coach requirements]. Keep data export available, explain generated changes for review, and address [bugs observed during daily use].
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source poster / publisher
- bojanstef4
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author shared a free gym tracker with an AI coach, offline-capable React PWA, data export, and an anonymous leaderboard, while continuing to collect feature requests.
- Human involvement
- The author reviewed most code and used the app daily, reporting that this exposed bugs, sometimes during gym sessions.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Built the initial project around the author's own workout split using Claude Code.
- Added routines after friends expressed interest in using the tracker.
- Forwarded incoming feature requests to CLI agents.
The author built Agent Arnold, a gym tracker with an AI coach, using Claude Code. They said they deliberately wrote no code themselves, although they reviewed most of it. The project began as their own workout split before friends wanted to try it, leading to routines and further feature requests.
The author described forwarding those requests to CLI agents. As a daily user, they also encountered bugs directly, sometimes while at the gym and connected through SSH from a phone between sets. This made personal use part of the development feedback loop.
The application used FastAPI and SQLite on a Linux machine, with a React progressive web app that worked offline. Claude also powered the in-app coach through the Anthropic API. Other reported components included Telegram OAuth, Resend magic-link authentication, Sentry error tracking, and Foursquare Places for gym locations.
The shared app was free, supported full data export, and included an anonymous global leaderboard. The author presented vibe coding as imperfect while continuing to invite requests from people dissatisfied with their existing trackers.
Who it's for: Developers building a personally used application and willing to review generated code and respond to everyday bugs.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · bojanstef4
Frequently asked questions
Was Claude used for development or for the app's coach?
Both. The author said the entire project was built with Claude Code, and the AI coach used Claude through the Anthropic API.
How did the author supervise development?
They reviewed most of the code, forwarded feature requests to CLI agents, and caught bugs through daily use, sometimes accessing the project by phone SSH at the gym.
More details
Remaining steps
- Reviewed most generated code and caught bugs through daily app use and phone SSH access.
About this account
Claude Code generated the application code while the author passed along requirements and reviewed most of the implementation.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47599944. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: bojanstef4 on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · SSH · FastAPI · SQLite · React · Anthropic API · Sentry
Fitness apps · Phone development · PWA · AI coach
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.