decide dinner

The problem

Staring at a random assortment of groceries and feeling the mental load of deciding what to cook. Most recipe apps require specific trips to the store rather than solving for the ingredients you actually have.

What it does

• Analyzes pantry photos or ingredient lists to classify what is confirmed on-hand versus missing.
• Suggests three distinct meal candidates: the best fit for your ingredients, an easy backup for low energy, and a wildcard option.
• Generates pantry-first recipes with exact quantities scaled to your household size and focused on minimizing cleanup.
• Maintains a persistent household profile and favorites list in local Markdown files to remember allergies, dislikes, and successful meals.
• Builds intelligent shopping lists that exclude items you already own and groups missing ingredients by grocery aisle.

Frameworks & tools

Filesystem-based persistence using Markdown, CODEX_HOME environment variables for data storage, and local reference files for food safety and cooking principles.

Why this beats prompting it yourself

Standard LLMs often hallucinate that you have ingredients you didn't mention or suggest overly complex recipes. This skill enforces a "pantry-first" logic, verifies storage of preferences before claiming to remember them, and uses structured food safety protocols that a general prompt might ignore.

Use cases

• Clear out the fridge by generating a recipe using only "use soon" perishables.
• Save a successful "freestyle" meal to a favorites file with one command for future recall.
• Convert a messy photo of a pantry shelf into a structured list of meal possibilities.
• Scale a recipe instantly for a specific number of diners while swapping out confirmed dislikes like mushrooms or cilantro.

Known limitations

Requires a local directory for persistence. Does not browse the web for recipes unless explicitly asked. Focuses on practical 15-45 minute meals rather than haute cuisine.