Review an AI coding handoff with a three-row evidence table

A reviewer should be able to check an agent's work while the original conversation is closed. Give the reviewer the change, the evidence that supports it, and the decision still needed. The following example uses a fictional export screen.

Start with one observable change

The request was to keep selected filters when a user exports an order list. The agent reported that the export uses the same filters as the visible table. Read the proposed change and find where it copies filters into the export request. Capture the file and symbol, then check whether a test actually covers the user change.

Claim to checkEvidence to hand overReview result
CSV export uses the selected status.exportOrders reads the selected status. A fixture has one paid order and one cancelled order.Run with the paid filter. The CSV should contain one data row, for the paid order.
Export respects the visible date filter.The changed function forwards dates, but the current fixture contains only orders from the same day.Open: add an order outside the range and verify that it is absent.
Users receive an empty file when there are no results.The brief did not specify whether to generate a header-only CSV.Ask the product owner to choose the empty-state behavior before declaring completion.

Make the handoff repeatable

Record the revision you reviewed and the command or manual steps you used. Keep the fixture with the change. Mark the date test open until it is run against a mixed fixture. Use the agent's summary to find the right files, then check those files directly.

Orkas is one workspace for this kind of handoff. Its desktop Projects can keep instructions, tasks and selected Library sources together while Commander coordinates work. See the Orkas guide to Projects and agents for setup. If you are using a remote model, use a fictitious or cleared fixture because relevant inputs may be sent to your selected provider.

Prepared on behalf of Orkas. The export scenario is illustrative.