Review software quality
Define the outcome and checks before starting an agent. Review the resulting source changes, recorded results and unresolved findings against that scope.
Agent Studio keeps the task, test and build evidence, review findings and follow-up work together. Select checks that fit the change; their presence alone does not establish correctness.
Define acceptance criteria
Record the project’s relevant lint, test, architecture and review requirements. A frontend change may need responsive captures and component tests; a backend migration may need API checks and a rollback plan.
Screenshots grouped by capture time and project. Source labels remain visible, including mocked. A screenshot alone does not establish that a test passed.
Inspect what was checked
Read command output, diffs, screenshots and review notes. Check their source revision and coverage. Request missing evidence before accepting the task. A screenshot documents one captured state.
Reproduce a visual result
Use the screenshot’s source, route, viewport and capture recipe to understand the recorded state. When the interface changes, rerun the recipe and review the updated image.
Automatic checks and human review
Configured automatic review checks the delivered work and can request rework or escalate a finding. Successful delivery and automatic gates permit integration before human review. A person still decides whether to accept the task.
Exception example: DEMO-9 has exhausted its reissue budget. Open findings and pending integration remain visible beside the decision controls.
Make findings actionable
Name the affected source, explain the consequence and state the required correction. Focus attention on architecture, dependencies, accessibility and behavior that automated checks do not adequately cover.
Resolve mismatches
Compare implementation, documentation and public claims. When they disagree, record the source and observed behavior, then repair the mismatch through a scoped task.