Greptile Summary

This PR closes a Windows UX regression where every ACP spawn would flash a visible cmd.exe console window by adding windowsHide: true to the two resolution paths (direct and shell-fallback) that were previously missing the flag in resolveWindowsSpawnProgramCandidate / applyWindowsSpawnProgramPolicy.

• Core fix (src/plugin-sdk/windows-spawn.ts): windowsHide: true added to the direct return (line 252) and to the shell-fallback return inside applyWindowsSpawnProgramPolicy (line 275). The remaining paths (node-entrypoint, exe-entrypoint) already carried the flag, so all four resolution branches are now consistent.
• New test (process.test.ts): "sets windowsHide on direct exe resolution on windows" exercises the previously uncovered direct path with an absolute .exe path on a simulated win32 runtime.
• Updated test (process.test.ts): The "falls back to shell mode" expectation is updated to assert windowsHide: true, aligning with the production change.
• Housekeeping (process.test.ts, mcp-proxy.test.ts): Raw spawn() calls in test files now include windowsHide: true to suppress console flash during CI runs on Windows — these do not affect test logic.

Confidence Score: 5/5

• This PR is safe to merge — it is a minimal, targeted bug fix with full test coverage and no behavioral change on non-Windows platforms.
• The change is confined to two one-line additions in well-isolated helper functions. All resolution paths are now consistently marked with windowsHide: true on Windows. The fix propagates automatically to all downstream spawn call sites through the existing forwarding logic in applyWindowsSpawnProgramPolicy. New and updated tests cover both previously-untested paths. The change is a no-op on macOS/Linux.
• No files require special attention.

_{Last reviewed commit: 8d36db3}

I reviewed this change and the implementation looks sound. The patch fills the two missing Windows resolution outcomes in src/plugin-sdk/windows-spawn.ts—direct and shell-fallback—by setting windowsHide: true, which matches the gap here.

I also checked the surrounding call sites, and the resolved windowsHide value is already propagated through the relevant ACPX / ACP client / lobster spawn paths, so this should flow through as intended without additional plumbing in those layers.

The tests line up with the change: there’s a new direct-.exe resolution assertion, the shell-fallback expectation now checks windowsHide: true, and the small test-only raw spawn() updates are consistent with avoiding Windows console flashes during test runs.

From a code and scope perspective, this looks reasonable to land.