Make an explicit, concise statement about what we can do together now (and what we cannot), establishing expectations and boundaries.
- A → U: Capability statement scoped to the channel/context (what’s supported, sources/tools, costs/risks).
- A: Offer topic options or invite a user-proposed topic.
- U: Choose a topic or ask for something out of scope.
- A: If in scope, transition to activity; if out of scope, refuse with alternatives or suggest a handoff.
Show evidence or constraints (sources available, data boundaries) when relevant.
Support & transitions
- If user requests something unsupported, provide safe refusals and nearby options without dead-ends.
- Use agent repair to clarify ambiguous scope requests.
Metrics
Help deflection; out-of-scope request rate; successful redirections.
Consequences
- the actor knows what is in and out of scope before spending a turn on it, and an out-of-scope request meets a refusal with somewhere to go rather than a dead end
- the statement is now a promise the rest of the exchange is held to; capability that turns out narrower than stated reads as the agent having misled them
Related patterns
Enables
- Opening (Agent) — capability giving is often embedded within an agent-initiated opening
- Conversation — an explicit statement of what the conversation can and cannot do
Complements
- Agent repair — out-of-scope or ambiguous requests are handled via agent repair
- Bounded autonomy — what the system can do versus what it may do on this occasion
Related
- Agent — the conversational partner whose capabilities are being stated