On this page

On this page

Get started

Mac OpenClaw ID sign-in testing

Mac OpenClaw ID sign-in testing

The acceptance condition is the original Mac Control UI shows the signed-in account after authentication. Opening an issuer page, signing in in Safari or Chrome, or passing browser-only tests is not sufficient.

The Mac host presents a separate, unprivileged login window and returns the matching callback to the requesting dashboard document. The Gateway still starts PKCE and redeems the code; the Control UI still stores the resulting account. This is separate from native onboarding's sign-in to discover and join claws. Ordinary browser popup behavior and Gateway authorization rules are unchanged.

The app and an external browser remain separate clients. This does not import browser cookies or tokens, or automatically sign every client into an account. An existing Safari/Chrome login may therefore still require authentication in the app's login window. Once completed there, the app must show the account.

1. Prepare an isolated Mac

Use a disposable Tart guest or credentialless macOS Actions worker, with macOS 26.2+ and Xcode 26.4+ / Swift 6.3. Follow Mac development setup for the current Node/pnpm and build requirements. A temporary HOME or a test filter on an operator desktop is not isolation. Do not mount the host's Keychain, OpenClaw state, or browser profiles into the guest. Keep its ordinary system temporary-directory configuration.

Record the checkout SHA, macOS version, Xcode version, and app build before each run. Do the synthetic tests before signing into the live issuer. Use a disposable test account and claw for the later live flow.

Inside the guest, from the repository root:

bash
git status -sbgit rev-parse HEADsw_versxcodebuild -versionpnpm install --frozen-lockfileswift build --package-path apps/macos --build-system native --build-testsswift test --package-path apps/macos --build-system native --skip-build \  --filter DashboardOpenClawIDSignInTests

Do not interpret compilation alone, zero selected tests, skipped tests, or a process exit without the Swift Testing summary as a pass.

2. What the native regression proves

The suite uses the real DashboardWindowController, WK script-message transport, an ephemeral HTTP issuer fixture, a second WKWebView, and real callback navigation. It does not replace the native receiver with a JavaScript mock.

It checks:

  • successful code delivery to the original document, without navigating that document away from the dashboard;
  • issuer denial and mismatched callback state;
  • cancellation when the person closes the login window or dashboard;
  • rejection of subframe requests, documents outside the dashboard mount, and callbacks outside that mount;
  • no dashboard/device-auth bridges in the issuer page, and closure of the login window after settlement.

The fixture supplies the external issuer, not the asserted native response. It does not prove real issuer compatibility, Gateway token redemption, or the full account-menu click. Section 4 covers those remaining boundaries.

Red/green comparison

Use separate worktrees inside the disposable guest; do not switch a checkout that is running the app or tests. For the original report, the inspected base was 428d2047b766fcf0e8d426c1d967b0c8697b8625.

From the candidate checkout, create a base worktree and copy only the new regression test into it:

bash
git worktree add --detach ../openclaw-id-before 428d2047b766fcf0e8d426c1d967b0c8697b8625cp apps/macos/Tests/OpenClawIPCTests/DashboardOpenClawIDSignInTests.swift \  ../openclaw-id-before/apps/macos/Tests/OpenClawIPCTests/

Install dependencies and build/run the same focused suite from the base worktree. Its success cases should fail at the absent native sign-in return channel, not at compilation, fixture startup, or a missing test dependency. Run the suite from the candidate checkout again and require it to pass. Record both results; a base setup failure is not red proof. Preserve the base checkout until the before-app capture is complete.

3. GitHub Actions and remaining focused checks

The repository's existing macos-swift lane discovers the Swift test file. Require the lane's build and test results for the candidate head SHA, not an older revision. Inside a normally prepared disposable Actions Mac worker, after its test build, the existing isolated launcher also supports:

bash
node scripts/test-macos-native.mts named \  --package-path apps/macos --build-system native --enable-code-coverage \  --skip-build --filter DashboardOpenClawIDSignInTests

Do not fake Actions environment markers on an ordinary desktop to get through this launcher's admission checks. Use the Tart command above in a clean guest. Draft PRs do not automatically run the ordinary CI workflow; follow the repository's authorized draft/CI dispatch process, or supply the Tart results before marking the PR ready. Do not treat a skipped draft job as proof. For a maintainer-authorized canonical draft targeting main, the supported dispatch is scripts/pr ci-dispatch <PR_NUMBER>; verify the resulting run selected the candidate head. This is not a request to mark an untested PR ready.

The web-side contract checks that a native reply is redeemed and stored in the app's account owner, while cancellation never falls back to an external browser:

bash
node scripts/run-vitest.mjs \  ui/src/lib/openclaw-id/browser-sign-out.test.ts \  ui/src/components/openclaw-id/account-menu.test.tsx \  ui/src/lib/open-external-url.test.ts --maxWorkers=1pnpm tsgo:uipnpm native:i18n:verifyscripts/lint-swift.sh macosscripts/format-swift.sh macosgit diff --check

Those web tests cover a separate client contract; they do not substitute for the native suite or the full-app flow below. Existing URL safety, sign-out and owner/member tests remain useful independent contracts. Redundant popup-helper call-shape checks are not sign-in proof.

4. Test the actual app against OpenClaw ID

Build both the app and the Gateway-served Control UI from the candidate. Updating only the web assets leaves an older native host without the new bridge. The normal package command builds both:

bash
scripts/package-mac-app.shopen dist/OpenClaw.app

Use the normal signing identity. For an otherwise credentialless disposable VM, the documented development-only alternative is ALLOW_ADHOC_SIGNING=1 scripts/package-mac-app.sh; this is not evidence of production signing or TCC persistence. See signing.

Use This Mac, its test Gateway, and the OpenClaw ID Labs configuration already known to work in the browser. If using a separate source Gateway, also follow the native transport build instructions. Verify the app is connected to that Gateway, rather than a host-machine Gateway or an older installed app. Never paste tokens into the test report.

Exercise these scenarios through the actual account menu:

  1. Fresh app sign-in: click Sign in with OpenClaw ID. A usable login window appears; complete the real issuer's authentication. It closes, and reopening the original account menu shows the account and Sign out. The original chat stays open. Check account-specific actions appropriate to this test claw.
  2. Browser already signed in: sign into the same loopback Gateway in the guest's external browser first. Then start sign-in from the app. Complete any app-local authentication; the app must now show its signed-in account too. Do not count the external browser's success screen as an app success.
  3. Persistence: reload the app dashboard and close/reopen it. The same app profile and Gateway origin should retain the unexpired account. Quit/relaunch the app and verify persistence for its persistent local dashboard store.
  4. Cancel and retry: close the login window before completing authentication. The app stays signed out and reports failure rather than waiting ten minutes. A new click opens a fresh usable attempt. Also test issuer denial.
  5. Window lifetime: start sign-in, then close or reload the dashboard, or switch Gateway. The login window must close and must not sign a replacement dashboard into the old attempt. Repeated clicks should not leave orphan windows.
  6. Issuer failure: temporarily disconnect the guest network after opening login. Verify a visible failure or issuer error and a working cancel/retry. Restore networking before the next scenario.
  7. Browser parity: repeat sign-in in a normal browser. Its existing popup, callback, and account menu still work. Ordinary links in the Mac dashboard retain their existing behavior.
  8. Issuer methods: exercise the real account's password/passkey/social method as applicable. A provider refusing embedded WebKit is a failed live check, not something to work around with a spoofed user agent or weaker policy. Report it before treating the patch as ready.

5. Evidence to return

Record base/candidate SHAs, toolchain, focused test summaries, and the result of each full-app scenario. Include a clean before/after pair of the actual Mac window and account menu, with matching theme/window size and synthetic or redacted identity. An optional short recording should show the click and return to the original app. Never capture passwords, one-time codes, callback queries, tokens, recovery material, or private conversations.

Keep native compile/test results, fixture transport proof, and live account sign-in results separate. A candidate remains unverified on macOS until those results are collected; browser-only green tests cannot close that gap.

Was this useful?