Add the desktop-test harness, package-owned desktop testing guidance, and focused test runs

.changeset/desktop-package-guidance.md

@@ -0,0 +1,9 @@
+---
+"@fluentui-react-native/desktop-driver": patch
+"@fluentui-react-native/desktop-test": patch
+"@fluentui-react-native/storybook-desktop": patch
+"@fluentui-react-native/storybook-desktop-runtime": patch
+---
+
+Add package-owned agent guidance and user entry points, and restrict published
+contents to the package files required at runtime and for documentation.

.changeset/desktop-test-macos-fixes.md

@@ -0,0 +1,14 @@
+---
+'@fluentui-react-native/desktop-test': patch
+'@fluentui-react-native/storybook-desktop': patch
+---
+
+Fix macOS desktop runs: canonicalize the reporting root so artifacts under
+symlinked temporary directories stay confined, skip stop requests to producers
+whose Unix-domain sockets have already half-closed, and request the canonical
+desktop target that Storybook's embedded driver registers for Fabric and Paper
+smoke tests. macOS RN Test App sessions without explicit identity now launch
+the project's exact built product instead of any Launch Services copy sharing
+its bundle identifier, and fail closed when its bundle identifier, renderer, or
+embedded JavaScript bundle does not match. Console test labels number retried
+attempts by ordinal rather than by WDIO's sparse attempt keys.

.changeset/neutral-desktop-harness.md

@@ -0,0 +1,26 @@
+---
+'@fluentui-react-native/desktop-test': minor
+'@fluentui-react-native/desktop-driver': minor
+'@fluentui-react-native/storybook-desktop': minor
+'@fluentui-react-native/tester': patch
+---
+
+Add an application-neutral desktop-test harness with canonical app/suite
+configuration, native lifecycle ownership, standalone WDIO/Jasmine execution,
+and shared event, JSON, JUnit, console, and GitHub Actions reporting.
+
+Make Storybook an optional integration while preserving isolated callbacks,
+authenticated readiness, platform exclusions, and owned PowerShell services.
+Bind native visibility/focus to existing WDIO browsers, retain delayed-render
+waits, strengthen Windows attach identity, and forward trusted launch timeouts
+and unpackaged working directories.
+
+Expose authoritative UI Automation required-form state, preserving unsupported
+results on other providers. Adapt native checkbox/switch reads to checked state
+without changing selected semantics or legacy Appium behavior.
+
+Expose the FluentTester application's canonical desktop targets for referenced
+standalone suites without importing test or Storybook tooling into the app.
+
+Legacy desktop Appium and mobile lanes remain until native semantic parity and
+multi-day reliability qualification are complete.

.github/skills/agentic-component-authoring/SKILL.md

@@ -66,9 +66,11 @@ references used, the implementation order, and the validation plan.
 7. Add focused tests and stories that exercise the public API and the resolved native output.
 8. Reconcile and ratify the local contract against the realized public types, tests, stories, and platform evidence.
    Upstream changes require explicit re-review and never overwrite the local contract.
-9. Run the smallest declared validation command while iterating. Finish with package format, lint, build, and tests;
-   run the Storybook bundle for story changes and the root build when public types, manifests, or project references
-   change. Finish with `yarn clean-lage-cache` and then `yarn lage buildci` (R-VALID-003) and report results truthfully (R-VALID-002).
+9. Iterate with focused validation (R-VALID-004): `yarn test <component>` from the package, `yarn build` after type
+   changes, and `yarn test --findRelatedTests <files>` after editing shared code. Before handover, run the package
+   format, lint, build, and whole test suite, the Storybook bundle for story changes, and the root build when public
+   types, manifests, or project references change. Then run `yarn clean-lage-cache` and `yarn lage buildci` once as
+   final acceptance (R-VALID-003) and report results truthfully (R-VALID-002).
 
 Do not divide one component implementation into separate sub-agent or sub-skill phases. Its types, state, styling, and
 rendering form one contract and should remain in one implementation context.

.github/skills/agentic-storybook-development/SKILL.md

@@ -1,31 +1,31 @@
 ---
 name: agentic-storybook-development
-description: Develop, run, bundle, or troubleshoot the agentic-components Storybook app on macOS or Windows, including CocoaPods, Xcode, Metro, Fabric, and offline builds.
+description: Develop, bundle, build, or troubleshoot the agentic-components Storybook app in apps/storybook on macOS, Windows, or Win32, including app catalog configuration, Metro, CocoaPods, native projects, and renderer builds. Not for configuring or running desktop tests or changing desktop package source.
 license: MIT
 ---
 
-# Agentic Storybook development
+# Agentic Storybook app development
 
-Work on the native Storybook application in `apps/storybook`.
+This workflow covers the `apps/storybook` consumer and its native toolchain.
+For the shared Storybook integration API and CLI use
+[`storybook-desktop` references](../../../packages/agentic/storybook-desktop/references/README.md);
+for app identity, discovery, and diagnostics use the app's
+[`agent-map.yaml`](../../../apps/storybook/agent-map.yaml) and
+[`AGENTS.md`](../../../apps/storybook/AGENTS.md).
 
-## Workflow
+1. Inspect `desktop.config.mts` and the app's declared scripts; do not invent
+   aliases or direct native runner commands.
+2. Use the package's platform recipes for prep, bundle, build, smoke, and
+   diagnostics. Treat Pods, workspaces, generated solutions, build directories,
+   and DerivedData as disposable. Fix the owning hand-authored source.
+3. Reinstall/regenerate affected native dependency sources and use a clean
+   build after dependency or workaround changes.
+4. Validate the JavaScript bundle and affected native build. Run the app's
+   format/lint and root validation when shared manifests or references change.
+5. Finish code/configuration changes with `yarn clean-lage-cache` then
+   `yarn lage buildci` (R-VALID-003).
 
-1. Read the repository `AGENTS.md`, then the compact
-   [agent map](../../../apps/storybook/agent-map.yaml). Read the
-   [Storybook instructions](../../../apps/storybook/AGENTS.md), including its native diagnostics commands, and
-   [README](../../../apps/storybook/README.md) only as needed for the task.
-2. Inspect the Storybook `package.json` and run its declared workspace scripts; do not invent direct runner commands.
-3. Use `storybook-agent` to query runtime story IDs, select stories, and sweep renderability. Reproduce the first
-   failure from the Storybook workspace so pnpm-linked React Native tooling resolves from the app.
-4. Treat native workspaces, Pods, lockfiles, generated solutions, build directories, and DerivedData as disposable
-   outputs. Fix the owning manifest, Podfile, configuration, or script instead.
-5. After dependency or workaround changes, regenerate the affected native dependency source and use a clean build so
-   stale output cannot produce a false success.
-   When adding a codegen-enabled workspace dependency to existing macOS Pods, the first pod install can generate its
-   spec without refreshing ReactCodegen's source list. If the clean build then reports a missing generated spec header,
-   run `pods:macos` once more before changing native configuration.
-6. Validate the JavaScript bundle and the affected native build. Run Storybook format/lint and the root build when
-   manifests or shared package references changed. Finish with `yarn clean-lage-cache` and then `yarn lage buildci` (R-VALID-003).
-
-Keep component API, slot, token, and rendering work in the `agentic-component-authoring` skill; use this skill for the
-Storybook application and its native toolchains.
+For test configuration, helper provisioning, test runs, smoke reports, or CI
+lanes, use `desktop-testing`. For package source changes, use
+`maintain-desktop-testing`. Component API and assertion work stays with
+`agentic-component-authoring`.

.github/skills/desktop-testing/SKILL.md

@@ -0,0 +1,37 @@
+---
+name: desktop-testing
+description: Configure, run, or troubleshoot native desktop tests using desktop-test, desktop-driver, and storybook-desktop, including app targets and suites, helpers, permissions, WDIO suites, Storybook story tests and smoke runs, reports, and CI. Not for package source changes or the Storybook app build toolchain.
+license: MIT
+---
+
+# Desktop testing
+
+Package references own commands, schemas, and troubleshooting. Read only the
+matching topic; do not copy its content here.
+
+| Task                                     | Read                                                                                                                                                                                                                                                                         |
+| ---------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Add an app target or suite               | [`desktop-test` configuration](../../../packages/agentic/desktop-test/references/configuration.md)                                                                                                                                                                           |
+| Run stages, suites, or inspect results   | [`desktop-test` commands](../../../packages/agentic/desktop-test/references/commands.md) and [reporting](../../../packages/agentic/desktop-test/references/reporting.md)                                                                                                     |
+| Prepare a platform or provision a helper | [`desktop-test` platforms](../../../packages/agentic/desktop-test/references/platforms.md) and [`desktop-driver` helpers](../../../packages/agentic/desktop-driver/references/native-helpers.md)                                                                             |
+| Integrate Storybook or run callbacks     | [`storybook-desktop` integration](../../../packages/agentic/storybook-desktop/references/integration.md), [commands](../../../packages/agentic/storybook-desktop/references/commands.md), and [tests](../../../packages/agentic/storybook-desktop/references/story-tests.md) |
+| Fix E2E adapters or diagnose failures    | [`desktop-test` WDIO suites](../../../packages/agentic/desktop-test/references/wdio-suites.md) and [lifecycle](../../../packages/agentic/desktop-test/references/lifecycle.md)                                                                                               |
+
+## Workflow
+
+1. Identify the consumer workspace; read its config and declared scripts. For
+   `apps/storybook`, also read `AGENTS.md` and `agent-map.yaml`.
+2. Reproduce with the narrowest declared stage, suite, story, or filter.
+   Provision the verified helper explicitly; never guess flags or call runner
+   internals.
+3. Do not change privacy settings without authorization or overlap interactive
+   desktop workloads.
+4. Report failures, skips, unsupported capabilities, and incomplete work
+   distinctly. Skips are not passes.
+5. If a published package changed, add its changeset. Finish code/config
+   changes with `yarn clean-lage-cache` then `yarn lage buildci`.
+
+For package source changes or reviews, use `maintain-desktop-testing`. For the
+`apps/storybook` build/toolchain, use `agentic-storybook-development`.
+Component story assertions stay with `agentic-component-authoring`; v0/v1
+FluentTester assertions stay with `maintain-v0-v1-component`.

.github/skills/implement-native-component/SKILL.md

@@ -37,7 +37,8 @@ ownership and qualification rules (R-NATIVE-001 through R-NATIVE-003).
 4. Implement with the TypeScript specification, codegen configuration, native props, events, commands, and registration
    kept consistent. Preserve public APIs and registration names (R-NATIVE-002) and keep fork imports in platform files
    (R-REPO-002).
-5. Validate the affected boundary with declared scripts (R-VALID-001): package build, test, and lint; native-core
+5. Validate the affected boundary with declared scripts (R-VALID-001), running focused package tests while iterating
+   (R-VALID-004): package build, test, and lint; native-core
    `codegen:windows` when Windows specs change; native builds through the owning app, such as
    `yarn storybook build --macos` or `--windows` from `apps/storybook`; and runtime evidence through Storybook smoke or
    `wdio` story tests. macOS Storybook commands default to Fabric: add `--paper` to `prep`, `build`, `run`, or `smoke`

.github/skills/maintain-desktop-testing/SKILL.md

@@ -0,0 +1,45 @@
+---
+name: maintain-desktop-testing
+description: Change, review, or plan source changes to desktop-driver, desktop-test, storybook-desktop, or storybook-desktop-runtime, including native helpers. Applies package invariants, documentation, native and consumer validation, and changesets.
+license: MIT
+---
+
+# Maintain desktop testing packages
+
+| Owner                                         | Entry                                                                        |
+| --------------------------------------------- | ---------------------------------------------------------------------------- |
+| WebDriver service, clients, helpers           | [`desktop-driver`](../../../packages/agentic/desktop-driver/AGENTS.md)       |
+| App/suite config, lifecycle, WDIO, reporting  | [`desktop-test`](../../../packages/agentic/desktop-test/AGENTS.md)           |
+| Storybook integration, CLI, extraction, tests | [`storybook-desktop`](../../../packages/agentic/storybook-desktop/AGENTS.md) |
+| On-device bridge, Metro, theme, chrome        | [`runtime`](../../../packages/agentic/storybook-desktop-runtime/AGENTS.md)   |
+
+Read the matching contributing reference from the package index.
+
+## Mode first
+
+- **Review/plan:** read-only. Do not implement, run write-mode formatting,
+  build, create changesets, launch apps, or run native contracts. Report
+  validation not run.
+- **Implementation:** continue below.
+
+## Implementation
+
+1. Change the lowest package that owns the behavior; update dependent
+   references when needed. Preserve exact identity, fail-closed behavior,
+   cancellation, ownership-safe cleanup, and explicit unsupported/skip status.
+   Contract fixtures that spawn Node or generate specs pass values through
+   arguments, not spliced source (R-SCAN-002).
+2. For standalone Swift/C++ helpers, read the provider README and
+   `native/PROTOCOL.md`; keep protocol schema and docs in sync. Apply
+   `implement-native-component` as a native overlay. React-module and Office
+   bridging-header steps do not apply to these standalone helpers.
+3. Update owning references and affected README/AGENTS files in the same
+   change.
+4. Iterate with `yarn test <name>` for the files you change (R-VALID-004), then run package format, lint, build, and
+   test. Native changes also need the matching-OS helper contract and affected
+   consumer lane; JS tests do not qualify native behavior.
+5. Add a patch changeset for each changed published package. Finish with
+   `yarn clean-lage-cache` then `yarn lage buildci`.
+
+When delegating, pass the package AGENTS.md and selected reference paths;
+subagents do not inherit them automatically.

.github/skills/maintain-v0-v1-component/SKILL.md

@@ -30,9 +30,10 @@ v0 and v1 are maintenance-only (R-V0-001 and R-V1-001). A request for a new comp
    editing.
 5. Make the smallest change that fixes the problem. Preserve public props, types, exports and aliases, display names,
    `componentRef`, token contracts, and native registration names (R-V0-002, R-V0-003, R-V1-002, R-V1-003).
-6. Validate with the owning package's declared scripts (R-VALID-001): build, test, and lint, plus the framework
-   packages named in the recipe when framework semantics change. Review snapshot diffs before accepting them. Report
-   each check's outcome (R-VALID-002), and finish with `yarn clean-lage-cache` and then `yarn lage buildci` (R-VALID-003).
+6. Validate with the owning package's declared scripts (R-VALID-001), narrowest first (R-VALID-004): `yarn test <component>`
+   while iterating, then the package's build, test, and lint, plus the framework packages named in the recipe when
+   framework semantics change. Review snapshot diffs before accepting them. Report each check's outcome (R-VALID-002),
+   and finish with one `yarn clean-lage-cache` and then `yarn lage buildci` (R-VALID-003).
 7. Add a changeset for published packages (R-REPO-003).
 
 Stop and confirm scope before migrating between frameworks, creating a component, or changing a public API. Those

.github/skills/review-component/SKILL.md

@@ -32,9 +32,11 @@ license: MIT
    | Framework-base pattern | [Framework-base instructions](../../../packages/framework-base/AGENTS.md)                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                                            |
 
 4. Check repository-wide rules: explicit exports (R-REPO-001), platform fork imports (R-REPO-002), changesets
-   (R-REPO-003), and no copied private source (R-REPO-004).
-5. Inspect validation claims. Run non-mutating checks only when they add evidence; never run snapshot-update or fix
-   modes. Record each check as passed, failed, skipped, not run, or unsupported (R-VALID-002).
+   (R-REPO-003), no copied private source (R-REPO-004), and the code-scanning patterns: linear-time regular expressions
+   and no values spliced into executed code (R-SCAN-001, R-SCAN-002).
+5. Inspect validation claims. Run non-mutating checks only when they add evidence, preferring focused tests over the
+   whole suite (R-VALID-004); never run snapshot-update or fix modes. Record each check as passed, failed, skipped, not
+   run, or unsupported (R-VALID-002).
 
 ## Output