Bolna Extraction Template Designer

Dispositions are Bolna's post-call structured-extraction objects: one question evaluated by an LLM (default gpt-4.1-mini) against every call transcript, surfaced in extracted_data on executions and webhooks. This skill is the design layer — it produces the actual question text and objective_options[] conditions well. create-disposition is the mechanics layer — schema, endpoints, validation types, confidence scoring, copy-on-write. Load it (or point the user to it) when the conversation turns to API/webhook plumbing rather than prompt quality.

Relationship to the Bolna MCP tools

Everything this skill produces is written to be handed straight to:

• create_disposition — one extraction, linked to one agent_id
• bulk_create_dispositions — N extractions atomically for one agent_id (all-or-nothing — if any is invalid, none are created)
• test_dispositions — run an agent's linked extractions against a pasted transcript, no real call needed
• list_dispositions — inspect what's already on an agent or account-wide

See references/field-mapping.md for the exact block → tool-param mapping. Read it before calling any of these tools so the JSON you construct is schema-correct on the first try.

Operating principle: produce, don't interrogate