Use the What to check checklist
Use the generated checklist to perform the manual checks that still matter.
Last updated 16 September 2026
Use the generated checklist to perform the manual checks that still matter.
Before you start
What to check is generated after relevant file changes. It is a review aid, not evidence that the actions have already been performed.
The generated checklist remains at 0 of 1 checked. The successful file-presence command above it does not establish that a screen reader announces the navigation correctly.
Step by step
Open the completed implementation conversation and find What to check below the result and changed files.
Expand the checklist if it is collapsed. Read each action and its expected result, including any additional detail shown for the item.
Perform the described journey in the running app with the required setup. For an accessibility check, use a screen reader or accessibility inspector instead of guessing from the visual layout.
Check an item only after observing its expected result. Leave a failed or unperformed check unchecked.
When a check fails, tell the agent the action, observed result, expected result, and relevant context. Review its correction and repeat that check.
Review the count at the top of the card before delivery. Also consider requirements outside the suggested checklist if the task has additional risks.
What to expect
The checklist records which suggested manual checks you actually completed. In the screenshot, 0 of 1 checked means the accessibility journey still needs human verification.
Troubleshooting and useful details
Preparing checks means generation is still running; Couldn’t prepare checks offers Retry. No manual checks suggested is not a guarantee that the software is correct. A checkmark is your review record, not a replacement for tests, code review, or checking the original requirements.
