food-tracker

Send a message listing foods and quantities, and the Skill splits it into items, estimates calories, protein, carbs, fat, and leucine from its macro defaults table, then writes one INSERT per item into the SQLite meals table and replies with updated daily totals.

Weight trends use morning-only entries via the period column, filtering with COALESCE(period,'morning')='morning'. The Skill reports both full-period and last-7-day slopes in kg per week and compares them against a 0.5-1.0% of body weight per week band.

It uses five tables: meals (one row per food item), weight (one per day with morning/evening period), daily_metrics (sleep score, steps, active kcal), workouts (one row per session), and waist (circumference plus auto-computed waist-to-height ratio).

Yes. Look up the row by date and food name, then UPDATE it by id, or DELETE the row to undo. The Skill always echoes back what was deleted so you can catch a wrong undo, and assumptions stored in the notes column make corrections straightforward.

A hard rule checks local time before framing totals: before roughly 18:00 the day is still in progress, so the Skill reports "so far today" with protein remaining rather than declaring a deficit or final day total that later meals would invalidate.

The preflight check distinguishes a fresh tenant (NOT-READY, triggers guided setup) from an environment fault (UNKNOWN). On a fault it refuses to onboard or recreate the database, since that would overwrite real data, and instead reports the problem to the owner.