tschau-sepp/AGENTS.md

1.2 KiB

Tschausepp - Jass Card Tracker

Mobile-first React app for tracking Schaffhauser (Jass) card game rounds via camera.

Tech Stack

  • React 18, TypeScript, Vite
  • TensorFlow.js (lite mode, CDN-hosted)
  • HTML5 Camera API (environment-facing preference)
  • localStorage (JSON, max 100 games)

Developer Commands

  • npm run typecheck: Run TypeScript check
  • npm run build: Build production version
  • npm run preview: Preview production build
  • Dev Server: Always running on localhost:5173. Do not start manually.

Architecture & Logic

  • Screens: Setup → Camera (with Detection/Assignment) → Results → History.
  • Detection: Custom image processing + color analysis for suits (Schellen, Schilten, Eicheln, Rosen).
  • Assignment: Uses radial sectors to assign detected cards to the nearest player.
  • Storage: Max 100 games in localStorage; follows games → rounds → cards structure.

Reference Docs

  • ML_SETUP_GUIDE.md: ML pipeline architecture, training, and TF.js deployment.
  • DETECTION_IMPROVEMENT_PLAN.md: Roadmap for improving detection stability and accuracy.
  • swiss_jass_suits.md: Domain reference for Jass suit colors and iconography.