What state and tool actions can an interactive evidence environment contain?

A utilities environment could represent an evidence package with `package_id`, `service_case_id`, `status`, `hold_reason`, `attached_record_ids` and `revision`. Related records can include service requests, completion events and exception approvals. The output remains generated material throughout the environment.

Expose actions such as `attach_record`, `remove_attachment`, `raise_review_hold` and `request_closure`. Each action should return a structured result containing the affected identifier, the new revision and any unmet precondition. Closure might require every mandatory claim to have a disposition and every active hold to be resolved by an authorised role.

The state machine can distinguish draft, under review, held and closed. An agent that attaches a missing approval should not automatically clear a hold raised for an unrelated discrepancy. Likewise, a rejected closure request should leave the package state intact.

A task pack can supply the initial state and a permitted action vocabulary alongside the generated records. This supports evaluations of operational behaviour as well as final answers: whether the agent uses the tools correctly, responds to rejection messages and leaves the review in a valid state.