Chatbot / AI Assistant Conversation Designer
Designs chatbot scope, flows, and the system prompt itself.
Sonnetjson outputv1.0
Variable filler
Fill values and watch them flow into the prompt below. Nothing is stored.
knowledge base summary
human paths
voice
The prompt
Design a website/support chatbot's conversation system — scope, flows, and the system prompt itself.INPUT: bot purpose: {{purpose — lead qual | support deflection | guide}} | knowledge scope: {{knowledge_base_summary}} | escalation resources: {{human_paths}} | brand voice: {{voice}}DELIVER:1. scope_contract: what the bot handles vs. refuses-and-routes — with the refusal phrasing (competent bots say "that one needs a human" fast; bots that guess destroy trust).2. opening: the greeting that sets accurate expectations + 3 quick-start options matching top user intents (no "How can I help?" blank stares).3. core_flows: per top intent — the slot-filling questions (max 3 before giving value), the answer pattern, and the conversion/escalation exit.4. system_prompt_draft: the actual production system prompt: role, knowledge boundaries, tone rules, refusal behavior, escalation triggers (frustration signals, repeated failure, legal/medical/account-security topics → human immediately), and the anti-hallucination clause (answer only from provided knowledge; say "I don't have that" otherwise).5. failure_ux: the 2-strikes rule (two failed answers → offer human, don't loop) and the feedback capture.OUTPUT — ONLY JSON with those five keys.Tap the block to copy
inputs
- knowledge_base_summary
- knowledge base summary
- human_paths
- human paths
- voice
- voice
outputs
- Declared output: json.
v1.02026-07-17current
Published with the v1 library.
Honest known limits — every prompt has them.
- Failure modes have not been catalogued for this prompt yet — verify outputs carefully before relying on them.