[Jason Morse (JasonVMo)]

Platforms Impacted

• iOS
• macOS
• win32 (Office)
• windows
• android

Description of changes

Replaces the Storybook-specific desktop lifecycle with an application-neutral desktop-test harness on top of the
existing desktop-driver, and keeps Storybook as an optional integration. The PR also:

• qualifies the refactor on Win32 and macOS;
• moves desktop-testing knowledge into package-owned agent guidance;
• lets yarn test run a subset of a package's tests.

Legacy desktop Appium and all mobile lanes are unchanged. They stay until native semantic parity and multi-day
reliability qualification are complete (#4345).

Summary

• @fluentui-react-native/desktop-test (new):
  • one configuration schema for app targets and suites;
  • native lifecycle and process ownership for installed apps, RN Test App, and REX hosts;
  • standalone WDIO/Jasmine execution;
  • one authenticated reporting collector with live diagnostics, durable logs, and JSON, JUnit, console, and GitHub
    Actions output.
• Storybook (storybook-desktop, apps/storybook): uses the shared lifecycle and reporting, while keeping
  isolated CSF callback workers, authenticated readiness, exact app leases, and platform exclusions.
• desktop-driver:
  • binds native visibility and focus to existing WDIO browsers;
  • strengthens Windows attach identity;
  • forwards trusted launch timeouts and unpackaged working directories;
  • exposes UI Automation required-form state, reported as unsupported on other providers;
  • reads native checkbox and switch state as checked state, without changing selected semantics or legacy Appium
    behavior.
• FluentTester and E2E: apps/fluent-tester, apps/win32, and apps/E2E declare canonical desktop targets and
  suites in desktop.config.mts without importing test tooling into the app. E2E page objects gain native desktop
  adapters.
• macOS fixes (31098f4):
  • the reporting root is canonicalized (/var to /private/var);
  • the harness no longer sends stop frames to producers whose Unix sockets have half-closed;
  • Storybook requests the renderer-scoped target ID;
  • RN Test App launches the exact built product, and fails closed on a bundle-identifier, renderer, or embedded-bundle
    mismatch;
  • retried attempts are labeled by ordinal;
  • FluentTester buttons scroll until their click point is unobscured;
  • the RNmacOS Yarn patch now exposes accessible Fabric <Text> paragraphs to accessibility.
• Package-owned guidance (a27723f):
  • desktop-driver, desktop-test, storybook-desktop, and storybook-desktop-runtime each get an agent
    AGENTS.md, a user README.md, and an indexed references/ directory.
  • New rules R-PKGDOC-001 to R-PKGDOC-008 are enforced by yarn check-agent-instructions.
  • New task yarn check-package-contents, part of buildci, checks actual yarn pack inventories. files
    allowlists keep build caches out of the four packages.
  • New skills desktop-testing and maintain-desktop-testing. agentic-storybook-development now covers only the
    app toolchain.
  • driver-next-plan.md and desktop-driver/PLAN.md are removed; Infrastructure: Complete Desktop Driver production release qualification #4345 tracks the remaining qualification.
• Focused test runs (e33dae0): fluentui-scripts jest, behind yarn test and yarn update-snapshots, used to
  pass every discovered test file to Jest, so yarn test popover ran the whole suite.
  • It now accepts path selectors, -t, --findRelatedTests, -o, and --changedSince.
  • Platform-suffixed tests still run only on their own platform.
  • New rule R-VALID-004 has agents iterate on focused runs, then run yarn clean-lage-cache and yarn lage buildci
    once as final acceptance.
• Code scanning and CI (3c40367):
  • CodeQL js/polynomial-redos: the duplicated secret-redaction regex was quadratic on a keyword followed by many
    spaces (7.3 s on 200 KB). One shared linear-time pattern now gives identical output (0.2 ms), with a regression
    test.
  • CodeQL js/bad-code-sanitization: desktop-test contract fixtures spliced JSON.stringify'd paths into executed
    Node source. The source is now constant, and values arrive through process.argv or import.meta.url.
  • check-agent-instructions checked agent-map paths on the file system. The generated, gitignored
    storybook.requires.ts therefore passed locally but failed the clean CI checkout. It now checks against Git and
    exempts sources declared policy: generated-….
  • New reference references/agents/repository/code-scanning.md (R-SCAN-001 to R-SCAN-003) is cited from the root
    safeguards and the review and desktop maintenance skills.

Findings worth reviewing

• Runner behavior, 77 packages: with fluentui-scripts jest, a bare argument that used to do nothing now narrows
  the run, and one that matches no test file exits 1. Unselected runs are unchanged, except for a one-line tip printed
  locally (not under CI or TF_BUILD) in packages with 10 or more test files. No pipeline or package script passes
  positional arguments to the runner.
• RNmacOS patch: the Fabric <Text> accessibility fix belongs upstream in react-native-macos. Until then, the Yarn
  patch is required for accessible Fabric paragraphs.
• Infrastructure: Complete Desktop Driver production release qualification #4345: its description links packages/agentic/desktop-driver/PLAN.md as the remaining-work inventory, which this
  PR removes. Its acceptance criteria still list the work; update the description when this merges.
• knip.config.ts: a narrow Windows portability fix for backslash-separated workspace keys, found during Windows
  validation.
• Windows baseline: two unmatched Windows baseline patterns fail closed rather than silently reducing coverage.

Deferred and follow-ups

• Infrastructure: Complete Desktop Driver production release qualification #4345:

  • equivalent-build old and new comparisons;
  • complete assertion mapping;
  • FluentTester runs on Windows Fabric;
  • the remaining Windows Storybook renderers;
  • live GitHub Actions presentation;
  • the 100-run, 14-day reliability gate.

  Retiring desktop Appium waits on these.

• Extend package-owned agent guidance to agentic and framework-base packages #4388: extend package-owned guidance to the remaining packages/agentic/* packages and framework-base.

• Harness cleanup: a failed readiness hook can leave "owned reporting producers did not finish their shutdown
  handshake" noise. A crashed session leaves the per-user interactive lock behind, and the harness correctly refuses to
  run until it is removed.

• Dead scripts: 23 packages declare a start-test script that calls a nonexistent fluentui-scripts jest-watch.

Verification

PR head 3c40367, macOS, October 22, 2026

Check	Result
yarn clean-lage-cache, then yarn lage buildci	Passed: 175 tasks, 0 failed
yarn lage test-links	Passed: 88 tasks
yarn change:check	Passed
yarn check-package-contents	Passed: four inventories
yarn check-agent-instructions with the generated storybook.requires.ts removed, as in CI	Passed
packages/agentic/desktop-test: yarn test	Passed: 25 tests, including the new redaction tests
apps/storybook: yarn storybook smoke --macos --mode stories-and-tests (Fabric), at d683ca8	Passed: 199 stories rendered and 81 story tests passed (280/280), 0 skipped, 0 retries

Focused test runner, macOS

• yarn workspace @fluentui-react-native/scripts test: passed, 49/49, including selection unit tests that load the
  installed Jest's option table.
• In packages/agentic/components:
  • yarn test popover selected 2 of 66 files, 19 tests, in 2.5 s; the unfiltered yarn test still ran 66 suites and
    561 tests in 5.3 s.
  • yarn update-snapshots button selected 4 files with no snapshot changes.
  • A typo exited 1.
  • --findRelatedTests on a shared primitive selected 16 files; a missing source path exited 1.
  • With a temporary uncommitted edit, -o and --changedSince=HEAD selected the 5 related tests.
  • --changedSince=<commit>^ selected 25 tests from committed history.
• Multi-platform and react-platform packages:
  • In packages/components/Button, yarn test ColorTokens.test.win32 ran only the win32 platform group.
  • packages/theming/default-theme (five platforms) passed.
  • packages/agentic/desktop-test (react platform) passed.

macOS qualification at 31098f4, October 22, 2026 (after merging main at 60121de)

No skips or retries:

Lane	Fabric	Paper
Storybook lifecycle	280/280	280/280
FluentTester pilot	2/2	2/2
FluentTester legacy (62 tests)	62/62	62/62

• Package suites:
  • desktop-test contracts: 16/16;
  • storybook-desktop: 175 passed, 2 skipped;
  • opt-in macOS desktop-driver tests: 68 passed, 2 skipped.
• buildci: the first run hit a Jest worker SIGSEGV in the unchanged design package, whose tests had all passed. The
  rerun passed.

Win32 at 238db96, October 21, 2026 (before merging main)

• yarn workspace @fluentui-react-native/e2e-testing desktop-test --target win32 test --suite behavioral-pilot: passed
  37/37. These are the unchanged Button, Checkbox, Switch, directional FocusZone, and Callout assertions.
• With STORYBOOK_SMOKE_STORY=components-button--default,
  yarn workspace @fluentui-react-native/agentic-components-storybook desktop-test --target win32 test --suite stories:
  passed 154/154, which is 152 navigations plus both native Button callbacks.
• With FURN_NATIVE_DRIVER_TEST=1, yarn workspace @fluentui-react-native/desktop-driver test --runInBand: 68 passed;
  2 macOS-only tests skipped.
• Package suites:
  • harness: 15 passed;
  • Storybook: 175 passed;
  • E2E adapters: 4 passed;
  • test:appium: 3 discovery and configuration tests passed; this is not legacy native execution.
• After yarn clean-lage-cache, yarn lage buildci --concurrency 4: passed.

Not run

• Windows Fabric and Win32 were not rerun after merging main (Add Popover, Menu, MenuButton, and Dropdown for macOS, Windows, and Win32 #4386) or after the macOS fixes. The Win32 evidence
  above is from 238db96.
• FluentTester on Windows Fabric, and the remaining Windows Storybook renderers.
• The Paper Storybook smoke at the PR head. Since Paper passed 280/280 at 31098f4, only documentation, the Jest
  runner, and the four packages' files allowlists have changed, and none of these depend on the renderer. The macOS
  Paper Storybook CI job runs on this PR.
• The focused runner on Windows: backslash selectors are covered by unit tests only, and the --runInBand path under
  TF_BUILD is unexercised.
• Live GitHub Actions presentation and the reliability gate.
• VoiceOver and Narrator, including the newly exposed Fabric <Text> paragraphs.
• Live model routing for the new evaluation fixtures, which check-agent-instructions validates structurally only.

Pull request checklist

This PR has considered (when applicable):

• Automated Tests
• Documentation and examples
• Keyboard Accessibility
• Voiceover
• Internationalization and Right-to-left Layouts

[github-actions]

Bundle size report

Tree-shaken, minified production esbuild bundles with React and React Native runtimes externalized.

Scenario	Modules-Mac (Δ)	Modules-Win (Δ)	Size-Mac (Δ)	Size-Win (Δ)
components-button	64 (+2)	65 (+2)	50.63k (+687b)	50.55k (+687b)
components-catalog	214 (+55)	215 (+55)	205.49k (+50.22k)	205.41k (+50.22k)
design-theme-state	34 (+1)	35 (+1)	30.15k (+42b)	30.08k (+42b)
design-themed-root	38 (+1)	39 (+1)	32.12k (+42b)	32.04k (+42b)
design-color-lib	6 (+0)	6 (+0)	6.33k (+0b)	6.33k (+0b)
design-theme-state-legacy-source	35 (+1)	36 (+1)	31.81k (+42b)	31.73k (+42b)
design-legacy-theme-from-flex	58 (+1)	59 (+1)	97.77k (+42b)	97.63k (+42b)

The job is advisory: size changes are reported but do not fail the pull request. Bundle or analysis errors still fail.

---

View bundle-size workflow run