feat: implement AI-curated playlist service and dashboard integration

- Added hierarchical AGENTS.md knowledge base
- Implemented PlaylistService with 6h themed and 24h devotion mix logic
- Integrated AI theme generation for 6h playlists via Gemini/OpenAI
- Added /playlists/refresh and metadata endpoints to API
- Updated background worker with scheduled playlist curation
- Created frontend PlaylistsSection, Tooltip components and integrated into Dashboard
- Added Alembic migration for playlist tracking columns
- Fixed Docker healthcheck with curl installation

AGENTS.md

@@ -0,0 +1,85 @@
+# PROJECT KNOWLEDGE BASE
+
+**Generated:** 2025-12-30
+**Branch:** main
+
+## OVERVIEW
+
+Personal music analytics dashboard polling Spotify 24/7. Core stack: Python (FastAPI, SQLAlchemy, SQLite) + React (Vite, Tailwind, AntD). Integrates AI (Gemini) for listening narratives.
+
+## STRUCTURE
+
+```
+.
+├── backend/            # FastAPI API & Spotify polling worker
+│   ├── app/            # Core logic (services, models, schemas)
+│   ├── alembic/        # DB migrations
+│   └── tests/          # Pytest suite
+├── frontend/           # React application
+│   └── src/            # Components & application logic
+├── docs/               # Technical & architecture documentation
+└── docker-compose.yml  # Production orchestration
+```
+
+## WHERE TO LOOK
+
+| Task | Location | Notes |
+|------|----------|-------|
+| Modify API endpoints | `backend/app/main.py` | FastAPI routes |
+| Update DB models | `backend/app/models.py` | SQLAlchemy ORM |
+| Change polling logic | `backend/app/ingest.py` | Worker & ingestion logic |
+| Add analysis features | `backend/app/services/stats_service.py` | Core metric computation |
+| Update UI components | `frontend/src/components/` | React/AntD components |
+| Adjust AI prompts | `backend/app/services/narrative_service.py` | LLM integration |
+
+## CODE MAP (KEY SYMBOLS)
+
+| Symbol | Type | Location | Role |
+|--------|------|----------|------|
+| `SpotifyClient` | Class | `backend/app/services/spotify_client.py` | API wrapper & token management |
+| `StatsService` | Class | `backend/app/services/stats_service.py` | Metric computation & report generation |
+| `NarrativeService` | Class | `backend/app/services/narrative_service.py` | LLM (Gemini/OpenAI) integration |
+| `ingest_recently_played` | Function | `backend/app/ingest.py` | Primary data ingestion entry |
+| `Track` | Model | `backend/app/models.py` | Central track entity with metadata |
+| `PlayHistory` | Model | `backend/app/models.py` | Immutable log of listening events |
+
+### Module Dependencies
+
+```
+[run_worker.py] ───> [ingest.py] ───> [spotify_client.py]
+                                  └───> [reccobeats_client.py]
+[main.py] ─────────> [services/] ───> [models.py]
+```
+
+## CONVENTIONS
+
+- **Single Container Multi-Process**: `backend/entrypoint.sh` starts worker + API (Docker anti-pattern, project-specific).
+- **SQLite Persistence**: Production uses SQLite (`music.db`) via Docker volumes.
+- **Deduplication**: Ingestion checks `(track_id, played_at)` unique constraint before insert.
+- **Frontend State**: Minimal global state; primarily local component state and API fetching.
+
+## ANTI-PATTERNS (THIS PROJECT)
+
+- **Manual DB Edits**: Always use Alembic migrations for schema changes.
+- **Sync in Async**: Avoid blocking I/O in FastAPI routes (GeniusClient is currently synchronous).
+- **Hardcoded IDs**: Avoid hardcoding Spotify/Playlist IDs; use `.env` configuration.
+
+## COMMANDS
+
+```bash
+# Backend
+cd backend && uvicorn app.main:app --reload
+python backend/run_worker.py
+
+# Frontend
+cd frontend && npm run dev
+
+# Tests
+cd backend && pytest tests/
+```
+
+## NOTES
+
+- Multi-arch Docker builds (`amd64`, `arm64`) automated via GHA.
+- `ReccoBeats` service used for supplemental audio features (energy, valence).
+- Genius API used as fallback for lyrics and artist images.