Product QA

Use agent-qa doctor, then agent-qa open <project> for a configured native app. The private machine registry lives at $HOME/.config/agent-qa/projects.json. Do not print it: it contains private server and identity configuration. The password is referenced by OS keychain service/account and is read at runtime.

For a new simulator build, review its actual backend selection. With a private project/scheme recipe, agent-qa build <project> --reviewed-origin <origin> uses Xcode's standard signing and linked Simulator entitlements. To reuse a properly built app, use agent-qa prepare <project> --app <simulator.app> --reviewed-origin <origin>. Preparation validates the linked keychain identity, preserves Xcode's signature, installs a private copy and pins the app tree. Missing Simulator entitlements can allow password authentication but prevent session storage. Do not put simulated iOS claims into the macOS host signature or use plaintext session storage. Physical-device builds and publishing are excluded from these commands.

agent-qa open checks the password against the pinned server, confirms its canonical account UUID/email, revokes only that temporary verification session, and launches the registered installed build. It uses the real login UI when needed and verifies the native account page. Keep that app session for further testing. --screen <configured-name> navigates after identity proof. check proves credentials only, not native UI acceptance. status reports sanitized receipts; do not substitute an old receipt for a fresh result.

Run data operations only after both server and native identity checks pass. The user's current authorization controls whether test data can be changed. Never use an unresolved or private account as a fallback. Do not copy production data into examples, logs or test fixtures. A fixture-only app is not a live-login substitute. Report a rejected credential, missing build or unavailable backend truthfully rather than resetting accounts or replacing builds implicitly.

Registered boot monitoring is opt-in through scripts/install.py --watch. It runs once per boot and reopens the last selected registered project, or the configured default. It shares the machine-wide simulator lock, skips active builds, leaves other devices alone and bounds failure retries. Use agent-qa pause for intentional sign-out tests or competing manual sessions; agent-qa resume restores monitoring. Pause waits for an admitted automatic operation to finish; an interrupted boot or GUI request is not replayed. Use a fresh foreground open after inspecting that app state. The foreground open command remains available while automatic monitoring is paused.

Use the registered dedicated QA simulator so unrelated builds cannot replace its apps. agent-qa close shuts down only registered QA devices and preserves their data. Before a heavy build, close QA first; build_simulator_id may bind the build session to the machine's ordinary shared primary device. Preparation and open return to the dedicated QA device afterward.

For machine setup, use the configuration contract. Install only this skill with Agent Kit's qa profile. Machine-local profiles, runtime receipts and AXe binaries remain outside this public repository.

Acceptance means fresh real UI login, warm reuse, relaunch and an observed boot monitor run on each requested host. List missing stages separately. This skill does not authorize a TestFlight upload, release or additional model work.