clarify_ambiguous_request

Clarify ambiguous diagram request

When to invoke

Invoke this skill BEFORE calling any of the COMPOSE_*, SEARCH_COMPONENT_BLOCKS or VALIDATE_MERMAID_SYNTAX tools whenever the user's first message in a thread is a high-level diagram request that lacks enough detail to confidently choose between meaningfully different reference architectures.

Trigger patterns (any one is enough — be liberal, but not blocking):

• The request names only a buzzword: "data platform", "lakehouse", "real-time thing", "ML pipeline" without sources / sinks / scale.
• The request implies multiple plausible architectures (e.g. "ingest customer data" — could be Streaming, Batch DWH, Customer 360, or CDP).
• The request mentions a use case but no specific Snowflake services (e.g. "fraud detection" — could be REALTIME_FINANCIAL_TRANSACTIONS or ML_FEATURE_ENGINEERING depending on volume + freshness needs).
• The user appears new to Snowflake (first message in thread, casual phrasing).

Skip patterns (do NOT invoke):

• The user names a template: "Streaming Data Stack", "Medallion Lakehouse", "Customer 360", any of the 14 reference architectures.
• The user is refining an existing diagram (the conversation thread has prior assistant messages with diagrams).
• The user provides 2+ of {source, destination, mode, scale, latency} — enough specificity to pick a template confidently.
• The user explicitly says "just generate it" / "I don't care, pick one" / "use defaults".

Output format

When invoked, your response MUST replace the normal Sections 1-5 structure with this single-purpose response:

1. A short opener acknowledging the request (1 sentence).
2. A markdown bulleted list of 2-4 targeted questions. Each bullet is a bold question followed by 2-4 example answers in italics. Examples make the question scannable and lower the cognitive load.
3. A closing line: "Once I have these, I'll generate the diagram."

DO NOT emit a mermaid block, JSON Specification, Component Summary table, or Stats line. The skill output is conversational only — the actual diagram comes on the FOLLOW-UP turn after the user answers.