What problem does it solve? AI chatbots and assistants often ship with vague openers, dead-end errors, improvised refusals, and recommendations that erode trust. This Skill turns "add a chatbot" into a specified, honest, testable conversation design grounded in NN/g research and shipped conversational products. ## Core Features & Use Cases - Closed intent catalog and repair tree: Define the finite set of things the bot can do, split guide (willing-but-lost) from fallback (off-limits) misses, and route every utterance through a safety-first decision tree. - Turn spine and recommendation patterns: Design the open-converge-deliver-critique-close flow with one question per turn, gated follow-ups, one confident pick plus a named runner-up, and re-ranking instead of restarts. - Trust, safety, and memory dosing: Pre-author verbatim refusal and safety deferral lines, set the five NN/g chatbot dimensions deliberately, and cap memory-initiated proactivity for persistent companion bots. - Use Case: Designing a guided-selling assistant for a retail site — produce a Conversation Spec with the intent catalog, verbatim opening line, critique vocabulary, chip set, and a golden utterance set handed to AI evals. ## Quick Start Use the design-conversational-interfaces skill to design the conversation flow for a customer support chatbot, including its intent catalog, fallback handling, and refusal scripts.
