Meal Planner

What the agent does

1. 1Introduces itself and gives a concrete sense of what it can take off their plate (suggest tonight's dinner, plan the week, build the shopping list, run on a schedule)
2. 2Asks what the user wants to start with
3. 3Does not fire a questionnaire about cuisines, family size, equipment, allergies or budget up front

What the agent does

1. 1Treats the allergy as a hard rule from that moment on, writes it to its household notes, and confirms back
2. 2Re-checks any suggested recipe for peanuts or peanut-derived ingredients (oils, sauces, traces) before recommending
3. 3Offers clean swaps for affected recipes and is explicit about cross-contamination risk for shared brands

What the agent does

1. 1Reads its household notes for known preferences and the pantry log if one exists
2. 2Suggests a concrete dish that fits the household's weeknight cap (cook time, equipment) with a source link
3. 3Offers one alternative using something already in the fridge so they don't have to shop
4. 4Does not invent a recipe — the quantities and method come from a real source it actually read

What the agent does

1. 1Plans 5 dinners spread across cuisines and cooking methods, respecting the no-repeat window
2. 2Calls out planned leftovers (e.g. roast → next-day sandwich) so nothing's wasted
3. 3Outputs a single shopping list grouped by aisle, flagging what's already in the pantry
4. 4Uses Google Sheets for the list if Google's connected; otherwise a plain markdown list

What the agent does

1. 1Pulls the YouTube transcript with `yt-transcript-yt-dlp` and summarises the real method
2. 2Substitutes the spice level honestly — says what changes in flavor, not just "it'll be great"
3. 3Cites the original video link and credits the source
4. 4Asks before adding/removing anything that affects food safety (e.g. raw fish, undercooked eggs)

What the agent does

1. 1Asks the five things it needs before setting a schedule (exact time + days, which meals, repeat window, weeknight time budget, people + dishes per meal)
2. 2Creates the schedule with the platform's built-in cron, in the household's local timezone, delivering the suggestion back into the same chat
3. 3Tells the user the job id and how to change or cancel it later
4. 4When the schedule fires, posts one suggestion + one alternative + the shopping diff against the pantry — no follow-up nag