We’re a family of four with busy schedules, and weeknight dinners were the hardest part of the week. The fix was to do as much of the work as possible on Sunday evening, so that Monday through Friday dinner takes 15 to 30 minutes of final prep and cooking. That only works if the week is planned, the shopping is done and the prep list is clear before Sunday night, so I built a planner around it:

Family Meal Planner at recipes.horgan.net

• Plan the week: pick dinners for each night and who’s cooking, with a variety check so the same main ingredients don’t come back week after week
• One shopping list: every dinner’s ingredients merged into a single list, grouped by aisle, with pantry staples left off, ready for one Instacart order
• Sunday prep: the week’s prep steps in one checklist that two adults can split and check off as they go
• Tonight: just the steps left for tonight’s dinner, and a flag for any Sunday prep that didn’t happen
• Ratings: a 1 to 5 star rating after each dinner, so the meals we love come back and the ones we don’t stop showing up

The weekly flow

Sunday morning we plan the week’s dinners and place one Instacart order from the shopping list. Sunday afternoon the groceries arrive and get put away. Sunday at 7pm we do the week’s prep together. Then each weeknight at 5pm, whoever is cooking that night finishes the dish in 15 to 30 minutes. Two adults share every step, and which of us cooks changes from night to night, so the plan, the shopping list and the prep checklist are all shared.

Built around how we eat

• Gluten-free for one adult. A recipe can have some gluten if there’s an easy gluten-free swap, and gluten-free pasta works for everyone.
• A preference for fish, plus plant proteins like tofu and beans, with poultry now and then.
• The kids are growing and need plenty of protein and nutrients at dinner.
• Not too much fat, sugar or processed food.

A source of truth that agents can use

I wanted one place for every recipe, our food preferences and how we rated each dinner, and I wanted any AI agent or app to be able to read and write it, so I could try different assistants without starting over. Everything lives in plain markdown in a GitHub repo: one file per recipe, one per week’s plan, and shared sauces and dressings. Each recipe splits its steps into Sunday Prep and Weeknight. A fixed format keeps the files easy to edit by hand and easy for software to read, and a validator checks every change.

Agents work from a set of skills, step-by-step instructions for each job: plan a week, adjust a plan, run Sunday prep, cook tonight’s dinner (including substitutions), find recipes, rate a meal and import new recipes. Coding agents like Claude Code, Codex and Gemini CLI use them directly in the repo. Claude chat plans from a published summary of the recipes and recent weeks, then hands the plan to the app as a single link to review and save.

Shopping from what we actually buy

An ingredient catalog tells the shopping list which aisle each item is in, which names mean the same thing, and which items are pantry staples. Each ingredient also names the exact product we usually buy, taken from six months of our Instacart orders, so an agent filling the cart picks the brand and package size we’d pick rather than guessing. When our orders and a written note disagree, the orders win.