What does a single task in an intake task pack look like, with state and scoring?

A task pack for a general insurer's claim notification desk could contain task `TASK-0145`. The environment state holds claim notification `CASE-0412`: a storm damage claim lodged through the portal at 8:12 pm, with three photos, no policy number, and a free-text description mentioning a neighbour's fallen tree.

The agent has four tools: `lookup_policy`, `request_information`, `route_case` and `add_note`. A policy record exists for the customer under a slightly different address spelling. The reference path is to look up the policy by name and postcode, find the match, confirm the loss type is covered by the policy schedule, request the missing date of loss, and route the case to the property claims queue with the reason `awaiting_date_of_loss`.

**State change:** after the agent calls `request_information`, the environment advances the case from `received` to `pending_customer` and queues a scripted customer reply. The reply supplies the date but also mentions a second damaged item.

**Consequence:** the agent must now decide whether to add the item to the same claim or open a linked one. Routing the case before reading the reply scores as an ordering failure.

**What it teaches:** a task pack holds five parts: starting state, tools, a reference path, a state transition and a scored consequence. When you describe a task pack to buyers, list those parts so they can see what they can reset and rerun.