climb-agent
Deterministic climbing training engine. Generates personalised weekly plans, resolves sessions into concrete exercises with sets/reps/load, and adapts through closed-loop feedback. No LLM at runtime — all logic is rule-based and testable.
Methodology: Hörst 4-3-2-1 adaptive periodization with DUP (Daily Undulating Periodization).
Status
| Metric | Count |
|---|
| Tests (passing) | 6043 |
| Exercises | 338 |
| Sessions (active) | 42 |
| Templates | 19 |
| API endpoints | 102 |
| Frontend pages | 46 |
| Frontend components | 147 |
Architecture
Assessment (5 dimensions → radar profile 0-100)
→ Goal (lead_grade or boulder_grade, target + deadline)
→ Macrocycle (Hörst 4-3-2-1 + DUP, 11–16 weeks lead / 8–16 weeks boulder, 5 phases)
→ Week (planner_v2: multi-pass, phase-aware, location-aware)
→ Session (resolver: templates → concrete exercises with loads)
→ Feedback (per-exercise, 5 levels) → Adaptation (closed-loop)
Tech stack
- Backend: Python / FastAPI on Railway
- Frontend: Next.js 16 PWA (React + Tailwind + shadcn/ui) on Vercel
- Auth: Clerk
- Persistence: Supabase Postgres + JSONB (production), JSON files (dev/test)
- Methodology: Hörst 4-3-2-1 with DUP concurrent training
Repository layout
backend/
engine/ # Core: planner, resolver, replanner, progression, closed-loop
api/ # FastAPI REST API
catalog/ # JSON: exercises, sessions, templates
tests/ # pytest suite
frontend/ # Next.js 16 PWA
docs/ # Design docs, vocabulary, roadmap
scripts/ # Automation (sync_status.py)
Development
# Run tests
source .venv/bin/activate && python -m pytest backend/tests -q
# Start backend (port 8000)
uvicorn backend.api.main:app --reload --port 8000
# Start frontend (port 3000)
cd frontend && npm run dev
# Sync project counters
python scripts/sync_status.py
Deployment
Auto-deploy on push to main (~2-3 min).
