Multi-Agent-Lebensmitteleinkaufs-Workflow-Design

Agent 1: Meal Organiser Agent - **Trigger**: On-demand invocation by user, specifying number of days (e.g., "5 days"). - **Input**: User-provided day count, [DIET TYPE], [COMPLEXITY STANDARD]. - **Process**: Generates a meal plan (breakfast, lunch, dinner per day) strictly adhering to [DIET TYPE]. Each meal is filtered to match [COMPLEXITY STANDARD] (e.g., max 30 minutes, max 8 ingredients, no advanced techniques). Ensures variety and nutritional balance within constraints. - **Output**: Structured JSON list of meals, each with: meal ID, day index, meal type (breakfast/lunch/dinner), dish name, ingredient list (with quantities), and a reference to a cooking method ID (for later use by Cooking Agent). This list is passed directly to Shopper Agent without human review. Agent 2: Shopper Agent - **Trigger**: Receives meal list from Meal Organiser Agent. - **Input**: Structured meal list (JSON). - **Process**: Aggregates all ingredients across meals, deduplicates, and maps each to a searchable item on [GROCERY SITE] using site-specific categories and search terms. Builds a cart with quantities, checking for availability and substitutions. If any ingredient is ambiguous (e.g., "fresh basil" vs "dried basil", or missing from site), it flags the item with a clear note - it does not guess or substitute silently. - **Output**: A draft shopping cart on [GROCERY SITE] (not submitted for payment) plus a list of flagged ambiguities for user confirmation. User reviews and confirms flags, then manually completes payment. Agent 3: Cooking Agent - **Trigger**: User selects a specific meal (by meal ID or dish name) from the generated plan. - **Input**: Meal ID, [COMPLEXITY STANDARD]. - **Process**: Retrieves the stored cooking method for that meal (predefined in a recipe database, matching the same complexity standard). Generates a step-by-step guide with numbered instructions, timing, and required tools. Keeps language concise and actionable, assuming the user has standard kitchen equipment. - **Output**: A text-based cooking guide displayed to the user, with no external links or videos. System Flow Summary: 1. User invokes Meal Organiser with day count, diet, complexity. 2. Meal Organiser outputs meal list -> Shopper Agent. 3. Shopper Agent builds cart, flags ambiguities, presents for review. 4. User confirms flags and pays manually. 5. Later, user invokes Cooking Agent with a meal ID -> gets step-by-step guide. Design Notes: - All agents operate asynchronously; no real-time coordination needed. - Data passes via a shared JSON schema to ensure compatibility. - No agent executes purchases or accesses payment systems. - Ambiguity handling is explicit in Shopper Agent to avoid user frustration. - Cooking Agent reuses the same complexity standard to ensure consistency between planning and execution. Please confirm this design or suggest adjustments before we proceed to implementation.