ci: drop the two electron-builder suites from the Windows lane

.github/workflows/desktop-install-windows.yml

@@ -20,8 +20,16 @@ name: Desktop install (Windows)
 #     `desktop-electron-pin.test.ts` asserts the dependency, the lockfile and
 #     `build.electronVersion` all agree, which is what keeps electron-builder
 #     from packaging a different Electron than `npm ci` installed.
-#   * The `scripts/*.test.mjs` build guards, which gate `prebuilder` and
-#     `postbuild` and therefore run on Windows during a real packaging run.
+#   * The `scripts/*.test.mjs` build guards that gate `prebuilder`/`postbuild`
+#     — the ones whose module graph does not reach `electron-builder`. The two
+#     that do (`before-pack`, `stage-native-deps`) fail to load under vitest on
+#     Windows with `SyntaxError: Invalid or unexpected token` while loading
+#     cleanly on ubuntu. That is a vitest/vite transform problem with that
+#     dependency, not this repo's code, and it is a packaging-stage concern
+#     rather than an install-path one — so it is recorded as a finding rather
+#     than chased here. Observed, not assumed: the four listed in
+#     `test:desktop:win-install` were seen passing on a Windows runner
+#     (21 tests, 0 failures) in the run that surfaced the other two.
 #
 # It deliberately does NOT run the whole `electron` vitest project: that also
 # holds POSIX-only suites (ssh-connection asserts ControlMaster multiplexing,

apps/desktop/package.json

@@ -52,7 +52,7 @@
     "fix": "npm run lint:fix && npm run fmt",
     "test:ui": "vitest run --project ui",
     "test:desktop:platforms": "vitest run --project electron",
-    "test:desktop:win-install": "vitest run --project electron desktop-electron-pin assert-dist-built assert-win-vcruntime before-pack stage-native-deps write-build-stamp",
+    "test:desktop:win-install": "vitest run --project electron desktop-electron-pin assert-dist-built assert-win-vcruntime write-build-stamp",
     "test": "vitest run",
     "preview": "node scripts/assert-root-install.mjs && vite preview --host 127.0.0.1 --port 4174",
     "check:test:desktop:platforms": "npm run test:desktop:platforms",

docs/system-log/2026-08-09.md

@@ -228,3 +228,40 @@ I should have checked the diff before naming a culprit.
   - The two path-literal tests remain Linux-only by accident rather than
     design. Both are one-line fixes (`path.resolve` on the fixture root;
     `path.join` for the expectation) whenever someone wants them cross-platform.
+
+## 2026-08-09T17:40:00Z — Windows job: drop the electron-builder suites (Claude Code)
+
+- **Agent/tool:** Claude Code
+- **Repos:** hermes-agent
+- **Done:**
+  - The narrowed job from the 17:05 entry ran and got **21 tests passed, 0 test
+    failures** — but two of the six *suites failed to load*, so the job was
+    still red. Both merged before their job finished (#173 and #174 are not
+    required checks), so `main` carried a red job through both.
+  - **Observed green on Windows** (quoted from the run, not inferred):
+    `desktop-electron-pin.test.ts` 4 tests, `assert-win-vcruntime.test.mjs`
+    6, `assert-dist-built.test.mjs` 5, `write-build-stamp.test.mjs` 6.
+    The pin contract and the VC++ guard both hold on the platform they are
+    about, which is the whole point.
+  - **The two failures share one cause:** `stage-native-deps.mjs` does
+    `import { Arch } from 'electron-builder'`, and `before-pack.mjs` imports
+    from `stage-native-deps.mjs`. Both suites therefore pull `electron-builder`
+    into the vitest module graph, and both fail with the *same*
+    `SyntaxError: Invalid or unexpected token` on Windows while loading cleanly
+    on ubuntu. The four that pass import no such thing.
+  - **Not chased further, deliberately.** It is a vitest/vite transform problem
+    with a third-party package on Windows, not this repo's code, and those two
+    are packaging-stage guards rather than install-path checks — so including
+    them was scope creep on my part, twice over. Dropped from
+    `test:desktop:win-install`; the cause is recorded in the workflow header so
+    the next person starts from the diagnosis rather than the symptom.
+  - **Ruled out first:** BOM, CRLF, and stray control characters were compared
+    across all five `.mjs` suites — no difference between the failing and
+    passing files, which is what pointed at the import graph instead.
+- **Follow-up:**
+  - **`electron-builder` cannot be imported under vitest on Windows.** Worth
+    knowing before anyone tries to add packaging tests to a Windows lane. Not
+    filed as a repo bug because the defect is very likely upstream.
+  - Still open from 17:05: Windows' no-mux SSH path has no coverage, and the
+    two POSIX-literal tests (`update-relaunch:54`, `windows-hermes-path:177`)
+    are Linux-only by accident.