Automated testing
Let Agents handle code checks, test design, and hands-on execution. Keep requirements, code, and test evidence on one task, so your team can focus on exceptions and quality decisions.
From our team's workflow · Results and evidence return to the original task
White-box checks and build verification come together at test design.
Scroll horizontally to explore the full workflow
The testing workflow
Follow a real feature: switching the AI Runtime or execution computer for a conversation while preserving the Agent's default settings.
Task status → Ready for testing
The developer links a PR to the requirement. Moving the task to Ready for testing triggers white-box checks: verify commits and merge results, run relevant tests, and identify risks that need hands-on validation.
Build succeeds → Feishu notification
A successful test build sends a Feishu notification. The Agent verifies that the landed commit is in the test version and moves eligible tasks to In testing.
Task status → In testing
Use requirements, task comments, and white-box risks to prepare data and cases that cover every requirement and edge case.
A tester schedules and starts execution
Run the cases as written: each one reports its result as it completes. The run produces an HTML report, keeping screenshots and recordings as evidence.
Meet the Agents
Code checks uncover risks. Test cases carry them forward. Execution leaves evidence. Choose browser or desktop testing to match the change.
Reads requirements and code, checks commits and test readiness, runs relevant tests, and flags risks.
Checks that the landed commit is included in the test tag and maintains test batches and task status.
Turns requirements and white-box risks into test scope, data setup, and executable cases.
Operates the browser to run black-box tests and delivers reports, screenshots, and recordings.
Tests desktop behavior, installation, and upgrades, retaining application and environment evidence.
Connect the workflow
Instructions define each Agent's role. Skills provide operating methods. Automations connect triggers to actions.
Set role instructions and bind a Runtime such as Claude Code or Codex on a shared computer. Provide repository access, testing tools, and Skills so each Agent can carry out its responsibilities.
Ready for testing triggers white-box checks. In testing triggers test design. Configure the trigger, Agent, and instructions, then inspect prompts and run history in automation details.
Connect GitHub or GitLab and associate PRs with requirement tasks. The white-box Agent traces changes, while the build verification Agent checks that the landed commit is included in the test tag.
Send build notifications to a Feishu group to trigger test-version verification. Teams can also create and assign tasks from Feishu and receive Agent results in the group.
Before the first run
This team keeps frontend and backend in one repository. A feature owner aligns the complete user journey, deployment prerequisites, and acceptance criteria.
Prepare a browser, test application, Runtime, and an operable desktop so sleep settings and everyday work do not interrupt runs.
Standardize dependency installation, database setup, startup, and test commands. Check services, migrations, the computer, and Runtime before execution.
Prepare admin, member, and read-only roles with clear access rules. Make test data creation and cleanup repeatable.
Describe the main path, permissions, failures, and persisted behavior so Agents judge against requirements.
Upload reports and recordings promptly, linking task, run, and case IDs. Temporary execution directories can be reclaimed.
Start with one automation
Start with white-box checks at handoff, then add test design, hands-on execution, and human review.
Connect the repository and Runtime, define the checks, and confirm the computer and dependencies are ready.
Trigger checks when a task becomes Ready for testing and return findings to the original task.
Add build verification, test design, and browser or desktop runs, with testers reviewing the evidence.
Questions & answers
Testers still define acceptance criteria, schedule black-box execution, resolve uncertainties, and sign off. Agents handle code checks, case generation, and hands-on steps. In this example, a tester manually started the desktop tests.
Choose based on impact. Browser features can go to the Web Agent; installation, upgrades, and local environments need the Desktop Agent. For cross-platform changes, validate each affected path.
Start with one automation: trigger white-box checks when a task is ready for testing. Add build verification, case design, execution, and human review as the workflow matures.
This practice uses local Runtimes on a shared computer. White-box checks need the repository and dependencies; browser testing needs a browser and dedicated accounts; desktop testing also needs the test application and an operable desktop. Script setup and readiness checks where possible.