How does the booking workflow map onto record links and agent tool actions?

A booking environment can be built around a chain of linked records. Identifiers follow a consistent style so that foreign keys are easy to check.

- `APP-2187` approval, with conditions and approver role - `FAC-2290` facility, holding limit, term and product code - `LOAN-5530` loan account, linked to the facility - `SCH-5530-A` repayment schedule, linked to the account - `COL-9031` collateral, linked to the facility - `DD-7714` payment instruction, linked to the account - `HO-0418` servicing handoff, with status and notes

Typical tool actions an agent could call:

- `get_approval(app_id)` and `list_conditions(app_id)` - `create_account(facility_id, idempotency_key)` - `generate_schedule(account_id, options)` - `link_collateral(facility_id, collateral_id)` - `lodge_instruction(account_id, payload)` - `raise_exception(app_id, type, owner_role)` - `release_to_servicing(account_id)`

Each action returns a structured result and an error code where relevant, so the environment can model rejected calls, timeouts and validation failures.

Task records can be supplied as JSONL, with one task per line holding the starting state reference, the instruction text and the expected end state. Scope the exact tool names and payloads to match your agent framework in a pilot.