What the documents look like
FBA preparation moves inventory into platform warehouses, so operating reviews depend on four document families at once: the inbound plan, carton labels, the inventory report, and compensation claims. The inventory report updates daily or weekly, and inbound discrepancies or aged-stock anomalies often surface only within a reporting period. A tool call node lets the model decide whether to invoke a plugin based on the tool description and parameter descriptions, handing actions such as inventory report queries or forwarder tracking checks to external systems. The tool call termination node is attached after a specific tool; when the workflow reaches it, the current tool call ends, no further tools are called, and no AI summary of tool results is produced.
The plan document carries a plan number, a destination object, a planned quantity, a status, and an effective range. The inventory report carries an SKU, a date, and an available quantity. These are different fields on different documents, so a planned quantity and an available quantity cannot be compared as if they were the same measure.
Constraints at the tool calls step
The tool description needs to state the input conditions and the situations in which the tool should not be called. System module parameters are fixed; plugin parameters are configurable; the parameter description and the required flag together influence whether the model initiates a call. When a suitable required parameter is absent from context, the model may decline to call the tool, and prompt wording can guide the user to supply the parameter. In advanced orchestration, dragging the tool call connection point lets available tools connect to the tool call module, and connected tools separate tool inputs from ordinary inputs and allow the description to be edited so the call timing is more precise. The documentation advises starting with a small number of tools and tuning before adding more. Retrieval results and similarity scores do not constitute authorization or entity isolation, and a retrieval action does not replace approval or a database write.
Hypothetical paper exercise sample. Source text A: "FBA inbound plan number PLAN-2024-0517, object is the US FBA warehouse, planned inbound quantity 600 units, status pending review, effective range through 30 June 2024." Source text B: "The inventory report shows this SKU had 420 units available for sale on 20 May 2024." Draft sentence under review: "This inbound plan has been approved and 600 units can be arranged for shipment." The materials needed for the judgment are the plan status field and the approval record; only the pending-review status is present, and the approval record is not provided.
Configuration and implementation checks
| Configuration item | Type | What this page checks |
|---|---|---|
| Tool description | Documented field | States purpose, input conditions, and situations where the tool should not be called |
| Plugin parameters and parameter descriptions | Configurable item | Check name, description, and required flag item by item |
| Missing required parameter handling | Interaction check | Observe whether the model calls the tool; record the actual request |
| Tool connection | Orchestration action | After connecting, confirm the available tools and call timing |
| Tool call termination | Node behavior | Once attached, the current call ends with no AI summary |
| Custom tool variables | Documented field | Supply parameter descriptions for nodes not treated as tool parameters |
A hypothetical starting point for this exercise is connecting two tools, chosen so that call timing can be observed with a small tool set and compared against the call record produced when only one tool is connected. This is a paper starting value, not a documented default.
Three common errors
- Treating a parameter description as a guarantee of execution. The description explains what a value means; whether a call actually happens depends on the model's judgment, so the real request and response must be checked.
- Treating pending review as approved. In the sample the plan status is pending review and the approval record is not provided, so the approval conclusion in the draft sentence is unsupported.
- Treating a retrieval hit as permission or as business execution. A retrieval result does not constitute authorization, and it does not mean a database write or external synchronization has completed.
How to verify the setup
- Input source text A and the draft sentence, then compare the plan status field against the conclusion in the draft. If the status is pending review and no approval record is present, the conclusion is unsupported and the evidence gap must be recorded.
- Input the inventory report sentence for the same SKU, then compare the 420 units available for sale against the 600-unit shipment quantity in the draft. These are different fields and cannot be treated as directly mutually exclusive; an inbound discrepancy record must be added before a judgment is possible.
- Check the run record after the tool call termination node is attached, and observe whether the current call ends and whether any AI summary output still appears. If summary output still appears, the termination node has not taken effect and the connection position must be reviewed.