Developer Exposes How a Validation Error Leaked Into a Chatbot's UI and the Fix

A developer building a conversational onboarding bot encountered a bug where an internal schema validation error — instead of a friendly follow-up question — was displayed directly to users in the chat window. The root cause was conflating two separate concerns: running a conversation and producing a validated data object, both within the same loop. Each conversational turn became a potential point of failure where validation errors could leak into the UI. After patching individual leaks with try/catch blocks proved insufficient, the developer restructured the flow so that schema validation happens only once, after the conversation ends, keeping internal errors entirely separate from chat output. The fix relied on a 'turnaround mode' in the Shapecraft library, which lets the AI model drive the conversation freely while deferring structured data extraction to a single, isolated step at the end.