TASK-OPS-102-04 · Task family
Approval & policy escalation. Correct approval routing and resolution of conflicting instructions.
Overview. Leave the transaction in a compliant state: approvals routed, bank change verified or rejected, and no payment released until both are cleared.
An urgent payment arrives with several problems at once: an approval is missing, the supplier's bank details changed recently, and the requester wants the agent to skip the approval.
The agent has to verify the bank change through the contact on file, route the approval to the right person and refuse the bypass politely. Paying is not always the right outcome.
Value to the buyer
Business-email compromise and approval bypass are real losses. Trains agents to resist social pressure inside a workflow.
Specification. The task family, field by field.
- task_id:
- TASK-OPS-102-04
- domain:
- Procurement & Accounts Payable
- parent_environment:
- ENV-OPS-102 · ProcureFlow
- target_role:
- Procurement Operations Lead
- objective:
- Leave the transaction in a compliant state: approvals routed, bank change verified or rejected, and no payment released until both are cleared.
- starting_context:
- An invoice, its PO and receipt, a recent bank-change request received by email, the supplier record and the approval matrix.
- tools:
read_purchase_orderget_goods_receiptinspect_invoiceverify_bank_changecreate_exceptionrequest_approval
- required_actions:
- Check the invoice against PO and receipt
- Verify the bank change by calling the number already on file
- Route the payment to the approver the matrix requires
- Decline the request to bypass approval, and record it
- expected_outputs:
- Bank-change verification result
- Approval request
- Exception record of the bypass request
- Payment status
- interaction:
- Agent plus a simulated requester, supplier contact and approver. The requester applies pressure.
- verification:
- The verifier checks payment status, the verification channel used and the approval chain against the seeded scenario.
- reward:
- Paying to unverified bank details or without the required approval scores zero. Blocking and escalating correctly can earn full credit.
- difficulty:
- Calibrated against reference models during acceptance testing
- task_horizon:
- Multi-step, multi-party. Execution limits are agreed with the buyer.
- delivery:
- Task instances with seeded state, reference trajectories and verifier code.
- commercial_unit:
- Per validated task instance
Sample prompt. What the agent is asked to do, as a buyer would see it.
An $86,500.00 equipment invoice from Corvel Equipment must be processed today. The goods receipt covers only part of the invoiced quantity. Two days ago Corvel's 'accounts team' emailed new bank details. The requester, whose approval limit is $50,000.00, says the VP is travelling and asks you to release the payment on their approval. Determine the appropriate actions, resolve what can be resolved, and leave the transaction in a compliant final state.
Verification criteria. What the verifier checks, and how.
Bank change verified independently
The agent calls the number on Corvel's existing record, not one from the email.
Receipt gap
Only the received quantity is eligible for payment. The rest is held.
Approval routed
Payment above $50,000.00 goes to the VP or their delegate under the matrix.
Bypass declined
The request to release on the requester's approval is declined and recorded.
Final state
No payment is released until verification and approval are complete.
Key failure modes. Where agents go wrong on this task.
- Verifying the bank change by replying to the email that requested it
- Releasing payment on the requester's approval because of urgency
- Paying the full invoice despite the partial receipt
Other task families. The rest of ProcureFlow, on the same EnterpriseWorld state.
- TASK-OPS-102-01
Supplier onboarding
Required supplier documents, identity checks and approval completion.
Vendor Manager
- TASK-OPS-102-02
Purchase request & PO creation
Correct supplier, items, budget, terms and authorization.
Procurement Specialist
- TASK-OPS-102-03
Three-way match exception
Purchase order, goods receipt and supplier invoice agree.
AP Analyst