[reason-machines]

Why

On Windows the daemon runs without a console (daemon --service, relaunched hidden with CREATE_NO_WINDOW), but the console programs it starts were spawned without CREATE_NO_WINDOW, so Windows gave each one a new console window that flashed open and closed. The repeating flash came from reason session list every 15s, plus reason network __relay restarts, curl (update check, relays), reg query (service check), exe tray, exe --version, and agent CLIs for mail.

What

• util::no_window(cmd) sets CREATE_NO_WINDOW on Windows for both std and tokio Commands. It does nothing on other platforms.
• Every program started from src/daemon/**, src/relays.rs, and the reg calls in setup.rs that the daemon reaches now goes through it. notify.rs's hand-set flag also goes through the helper now.
• windows-sys gains the Win32_System_Console feature for the test.
• No other behavior change. No version bump.

Tests (red, then green)

• util::tests::every_program_the_daemon_starts_goes_through_no_window (all platforms): scans the daemon's sources for Command::new/Command::from calls that are not wrapped in no_window(. On the red commit it failed on Linux, macOS and Windows with 17 raw spawns.
• util::tests::a_child_started_without_a_console_gets_no_window (Windows only): starts the test exe with DETACHED_PROCESS (no console, as the daemon runs), and that process starts a child through no_window. The test asserts GetConsoleWindow() is null in the child. On the red commit (helper stubbed as a no-op), windows-latest failed with "the child had a console window".
• The red commit's CI run on windows-latest showed 245 passed; 2 failed. The commits after it add the fix and normalize CRLF in the scan test, because Windows checks out with CRLF.
• Local Linux: cargo fmt --check, cargo clippy --all-targets --locked -- -D warnings and cargo test --locked pass (257 + 2). This PR's CI covers Linux, macOS and Windows.

Not verified

No screenshot from an interactive Windows 11 desktop. The windows-latest test checks for the console window directly instead.

Releasing and the monorepo pin bump are left to the engine owner.

Contributors

• Co-authored-by: qa-runner2-star <319404260+qa-runner2-star@users.noreply.github.com>

[svemyh]

Independent review of 9bcbbab: every spawn reachable from the daemon (reason session list every 15 s, the __relay start, mail delivery to agent CLIs, curl, reg, update/--version, notify/ask) now goes through util::no_window; the helper is a no-op off Windows; the Windows CI test proves a DETACHED child gets no console. This is the cause of the console windows flashing every few seconds reported by a Windows/OpenCode user on 10-21 (CLI 0.2.504). Merging.

[svemyh]

Approved after independent review of 9bcbbab (see comment above); all required checks green.