Update README and CLAUDE.md for Home Assistant integration

CLAUDE.md

@@ -40,8 +40,8 @@ journalctl --user -u battery-tracker -f
 `config.py` reads `DATABASE_URL` from the environment, falling back to `sqlite:///batteries.db`. Swapping to MariaDB is entirely a matter of setting that env var — no code changes needed. All column types are restricted to `Integer`, `String`, `Text`, `Boolean`, `DateTime` for MariaDB compatibility. Status is stored as `String(20)` (not `Enum`) to avoid DDL differences between SQLite and MariaDB.
 
 ### Models
-- `Battery`: label (unique), brand, status (`available`/`installed`/`retired`), device_id (FK nullable, `ondelete=SET NULL`), notes, size, chemistry, capacity_mah, tested_capacity_mah, tested_date, charge_cycles, purchase_date, storage_location
-- `Device`: name (unique), battery_slots, device_type, notes
+- `Battery`: label (unique), brand, status (`available`/`installed`/`retired`), device_id (FK nullable, `ondelete=SET NULL`), notes, size, chemistry, capacity_mah, tested_capacity_mah, tested_date, charge_cycles, purchase_date, storage_location, battery_percentage
+- `Device`: name (unique), battery_slots, device_type, notes, ha_entity_id
 - Helper methods: `Battery.is_available/installed/retired()`, `Device.installed_count()`, `Device.installed_brands()`, `Device.has_mixed_brands()`
 
 ### Business rules (enforced in routes, not DB constraints)
@@ -52,6 +52,10 @@ journalctl --user -u battery-tracker -f
 - Bulk install into device: capacity check before write; batteries already in target device are skipped; retired batteries are skipped
 - Unretire: sets status back to `available`
 - Unassign with `next` form field → redirects to that URL (must start with `/`); falls back to dashboard
+- Logging a charge entry → sets `battery.battery_percentage = 100`
+
+### Home Assistant integration (optional)
+`ha_client.py` wraps the HA REST API (`GET /api/states/<entity_id>`). `ha_poller.py` runs a daemon thread started in `create_app` only when `HOMEASSISTANT_URL` and `HOMEASSISTANT_API_KEY` are set. The poller queries all `Device` rows with `ha_entity_id IS NOT NULL`, fetches the current percentage from HA, and writes it to `battery_percentage` on each installed battery in that device. The poller uses its own `sessionmaker` session (not the request-scoped `scoped_session`). When HA is not configured the app behaves exactly as before — all HA UI is gated on `ha_enabled` passed to templates.
 
 ### Adding new columns to existing DB
 `create_all()` won't add columns to existing tables. Run via Python:
@@ -66,5 +70,7 @@ Always snapshot the DB first: `cp batteries.db batteries.db.$(date +%Y-%m-%d).sn
 ### Testing
 Tests use a temporary file-based SQLite DB (via `tempfile.mkstemp`) created fresh per test — avoids SQLite in-memory per-connection isolation issues. The `seeded_client` fixture in `conftest.py` pre-populates via HTTP POST calls (not direct DB access), so tests exercise the full stack. Battery IDs in tests are positional (id=1 is always the first battery POSTed by the fixture).
 
+`TestConfig` sets `HOMEASSISTANT_URL = None` to prevent the HA poller thread from starting during tests. HA integration tests live in `tests/test_ha_integration.py` and use a separate `ha_app` fixture with a fake HA URL; they call `poller._poll_once()` directly rather than waiting for the background thread to fire. HA API calls are mocked with `unittest.mock.patch("ha_client.requests.get")`.
+
 ### MariaDB migration
 `migrate_to_mariadb.py` opens two SQLAlchemy sessions simultaneously (SQLite source, MariaDB destination), migrates Devices first (FK dependency), then Batteries with explicit `id=` values to preserve FK links, then resets `AUTO_INCREMENT` via `text("ALTER TABLE ...")`. Takes `MARIADB_URL` from env or CLI arg. See `MIGRATION.md` for the full procedure.