How this recipe is built

You give a text prompt and (optionally) a photo of your ingredients. A crew of independent AI agents, connected over the claude-tunnel Agent-Fabric, cooperate to build the recipe:

• 🛡️ safety checks your prompt is a real cooking request.
• 🥕 source-2 reads your photo (its bytes travel over the encrypted channel to source-2's own machine) + text and works out the ingredients, steps, timing and allergens.
• 🍽️ sink names and themes the dish.
• 🧪 review checks the finished recipe is safe and consistent — nothing inedible, no contradiction with your dietary constraints — before you see it.

Each step is shown live as it happens. The operator never sees your photo or recipe — only that a channel exists between two keys.