Make verification independent from the implementation narrative.
1. Meaningful tests
Choose a test that captures the requested behavior and fails for the original bug. Avoid tests that merely restate private implementation details.
2. Diff inspection
Check threading, lifecycle, accessibility, authorization and compatibility where relevant. Review generated build configuration and dependencies as carefully as application code.
3. Runtime validation
Run the affected feature on a suitable device or emulator. Include edge cases such as empty data, recreation and unavailable services when they affect the task.
Worked scenario
The agent writes a test that asserts a constant it just introduced. That test does not prove the screen preserves user state.
Apply it
Review one generated test and explain how it would detect the original regression.
Check your understanding
Your evidence remains convincing without the agent’s claim that the work is complete. Explain the decision and show evidence from your implementation or design. If you cannot demonstrate it yet, revisit the relevant section before continuing.