The Conversation Design Spec Template - Pro
The document you hand engineering to build a conversational flow, and a Claude skill that fills it in and checks it.
A working spec for the flows, the fallbacks, and the moments a conversation can break. An eleven-section template you duplicate into your own workspace, two complete example flows filled section by section, and a Claude skill that evaluates a flow against the spec and returns it filled.
Most conversational flows get designed in a doc that covers the happy path and stops. The fallback logic gets decided in standup. The tone for a failure state gets improvised at runtime. What the agent should say when it cannot reach the data, and the user can see on screen that it is wrong, never gets written down at all. The gaps only surface after the flow ships and a real user hits the state nobody specified.
This template treats the flow as a spec. Eleven sections carry it from goal through intents, utterances, system responses per state, tone calibration, fallback logic, escalation, and the decision log. A container error matrix and a reachability rule handle the moment the agent contradicts what the user can see. The Claude skill runs a flow through all eleven sections, names what is present, names what is missing as a question you can answer, and returns the filled spec ready to paste back in.
What's inside
The eleven-section spec template, as a Notion page you duplicate
Two complete example flows, filled section by section
The container error matrix and reachability rule, pre-built
The Pro Claude skill: checks your whole spec at once, catches where sections disagree, and drafts your decision log
Writes results straight into your Notion, no copy-paste
Core or Pro. Core checks one flow a section at a time and hands you paste-ready output. Pro reads the whole spec at once, catches where sections contradict each other, and writes into your Notion. Core for the occasional flow; Pro for documenting flows on a team.
