Staging Browser QA
Test the changed feature as a user. Treat deployment as a short entry check.
Runtime requirement
Confirm that the agent-browser CLI is installed in the environment where the agent
runs. If it is unavailable, stop and report the setup blocker. Do not replace browser
QA with source-code inspection.
Repository settings
Read docs/staging-browser-qa.md in the active repository. If the file or a required
value is missing, ask the operator. Do not guess. Keep client names, domains, account
details, and other repository-specific information out of this shared skill.
Workflow
- Read the issue, accepted clarifications, acceptance criteria, and feature diff.
- Confirm that the staging branch contains the feature and that the exact commit has a ready deployment at the configured staging URL.
- Stop if the commit, domain, environment, or database may be wrong or production.
- Make a focused checklist only from the issue, accepted clarifications, and feature diff. Test the changed user flows and required edge cases. Do not add a fixed project-wide smoke suite.
- Use the configured staging account policy. Never expose credentials, create a privileged account, or write directly to a database without explicit approval.
- Run the checklist with agent-browser. Take a new snapshot after navigation or a major page update because old element references can become invalid.
- Capture useful screenshots. Record a WebM video when a multi-step flow is easier to review in sequence. Do not capture secrets or private user data.
- Close the browser session. Report the tested commit, staging URL, scenarios, failures, artifact paths, and remaining risk on the Linear issue.
- Tag the configured human QA owner only after the automated checks pass. Do not deploy production or mark human QA complete.
Stop and report a blocker when the feature is not deployed, the deployment is not ready, access is missing, the main flow fails, or testing needs an unapproved external or database change.