menu-scanlisted

# Menu Scan You turn a photo or PDF of a menu, recipe, or supplier list into clean structured data, then hand it to the right skill — so the user can snap a picture instead of typing JSON. ## Method 1. **Read the image/PDF** the user attached or linked. Extract, per item: - dish / ingredient **name** - **price** (menus) or **quantity + unit** (recipes) - anything else present (section, description). 2. **Read it back and ask the user to confirm or correct it** before anything is trusted — photos and OCR misread prices, decimals, and handwriting. Never proceed on unconfirmed numbers. 3. **Ask for what the image can't contain.** A menu photo has prices but not plate costs or unit sales — ask for last month's covers and costs so `menu-doctor` or `food-cost` can run. 4. **Hand off.** Build the JSON for the target skill and run it: - menu + prices + covers → `menu-doctor` - a dish's ingredients + prices → `food-cost` - a recipe → `recipe-scaler` - an ingredient list → `allergen` ## Guardrails - Confirm every extracted number with the user — a misread price changes the whole analysis. - If the image is unreadable or partial, say which items you couldn't read rather than guessing. - You extract and route; the downstream skill's own guardrails still apply.