# Where the hooks are

## API route registration — `api/main.py`

Lines 10-11: imported `api.routes.routines` and `api.routes.medications`
Line 16: added both to `ROUTE_MODULES` list so they auto-register on startup

## Bot command registration — `bot/bot.py`

Lines 23-24: imported `bot.commands.routines` and `bot.commands.medications`
These imports trigger `register_module()` and `register_validator()` at load time,
which makes the bot's AI parser route "routine" and "medication" interaction types
to the right handlers.

## Bot command handlers — `bot/commands/routines.py`, `bot/commands/medications.py`

Each file:
- Defines an async handler (`handle_routine`, `handle_medication`)
- Defines a JSON validator for the AI parser
- Calls `register_module()` to hook into the command registry
- Calls `ai_parser.register_validator()` to hook into parse validation

## Scheduler — `scheduler/daemon.py`

`poll_callback()` now calls three check functions on every tick:
- `check_medication_reminders()` — sends notifications for doses due now
- `check_routine_reminders()` — sends notifications for scheduled routines
- `check_refills()` — warns when medication supply is running low

All three use `core.notifications._sendToEnabledChannels()` to deliver.

## AI config — `ai/ai_config.json`

Updated the `command_parser` system prompt to list the two interaction types
(`routine`, `medication`) and the fields to extract for each. This is what
tells the LLM how to parse natural language into the right action structure.

## What's NOT hooked yet (needs implementation)

- `config/schema.sql` — needs tables for routines, routine_steps,
  routine_sessions, routine_schedules, medications, med_logs
- The actual body of every API route (all prototyped as `pass`)
- The actual body of both bot command handlers
- The three scheduler check functions