This is a recording
These are real runs of the app with a made-up skater, played back here. Claude can't run on this public page: the live app runs on a Mac with Claude Code (npm run app), on a Claude subscription and never a paid API key. The two runs marked "stand-in" show what happens when a plan fails a check; they use the recorded replies the tests use in Claude's place.
Plan a session
Pick a recorded run to play it back.
What happened
What Claude was given
Claude's reply, as it came back
How it works
- Code reads the skater's MySkate backup with MySkate's own code and turns it into numbered facts: ratings, landings, homework, goals, the next test and competition.
- Claude picks 2 to 6 skills, the minutes for each, a reason for each that cites the facts, and a focus copied from the skater's own notes. It has no tools: it only reads the facts and answers.
- Code checks every part: real skills, minutes that fill the ice time, every number traced to the facts, the focus in the skater's own words. A plan that fails goes back once with the problems; if it fails again, MySkate's own suggestion is used.
Built for TCSS 490, Artifact 3. The code, and exactly what Claude does and what the code checks.