[jayleaton]

The first nightly built as Ada (run 43445916151) failed at Desktop Windows x64 / Verify and arm64 / Verify, on test-agents-windows-installer.ps1:52, where Start-Process could not find Uninstall Ada.exe.

The installer was right; the check was wrong. electron-builder's NSIS template defines UNINSTALL_FILENAME as Uninstall ${PRODUCT_FILENAME}.exe. PRODUCT_FILENAME is appInfo.productFilename, which is win.executableName when one is set. Since #238 pins the exe to "T3 Agents", an Ada build ships:

	Ada build
Exe	%LOCALAPPDATA%\Programs\t3agents\T3 Agents.exe
Uninstaller	…\t3agents\Uninstall T3 Agents.exe
Registry UninstallString	"…\t3agents\Uninstall T3 Agents.exe" /currentuser (same define, so Settings → Apps works)
Apps entry DisplayName, shortcuts	Ada <version>, Ada.lnk (from the product name)

The uninstaller file name is the same before and after the rename, so an existing "T3 Agents" install's uninstaller is replaced in place on update.

Changes

• The exe-name rule moves from build-desktop-artifact.ts to a dependency-free scripts/lib/windows-package-names.ts, which also derives the uninstaller name. Packaging, the payload check and the release Verify step all use it. Verify runs it with node and passes both names to the PowerShell check, so no names are hard-coded in the test.
• The check now uninstalls through the registered UninstallString, the path Settings → Apps uses. It asserts that the string names the installed uninstaller, and that uninstalling removes the Apps entry and the shortcuts. The update path from "T3 Agents" (one Ada shortcut afterwards, an Ada Apps entry) is still covered, and now ends in that same uninstall.
• New fork-windows-installer-check.yml, path-filtered to the installer files, the check, the packaging script and release-desktop.yml. It builds an unsigned Ada installer through the nightly's own release-desktop.yml (bundle, Linux x64 for the WSL CLI archive, then Windows x64 with a windows-2025 Verify) and runs the same Verify step. It's about 4 jobs and roughly 15 minutes, only on PRs that touch those paths, plus workflow_dispatch. This PR triggers it.

Merge risk

Low. Packaging output is unchanged; only the check, its inputs and a new CI workflow change. The new workflow passes no secrets.

vp run check:changed passes. The PowerShell runs only on Windows; this PR's Fork Windows installer check is its proof.

Opus 5.5 via Claude Code in Ada.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	21.5 KiB	21.5 KiB	0 B (0.0%)	24.4 KiB	✅
Codex	Live turn messages	1	1	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.9 KiB	21.9 KiB	0 B (0.0%)	24.4 KiB	✅
Claude	Live turn messages	1	1	0 (0.0%)	8	✅

Baseline: 89b5cb1 · PR result: 0dbe6ce · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.