Fast follow-up from the #1764 review (non-blocking): (1) attachWindowCmd.onExit (home.go:~8321) has no test constructing it; (2) the plain-terminal onExit site (~8800) sits outside the source guard; (3) the new workDirIfFollowing gate (home.go:~13024-13056) is only guarded by a substring check that an inverted gate would pass; (4) isReloading is never set by any test. Also welcome: one on-real-fleet before/after measurement so 'instant' becomes measured rather than extrapolated from the idle-sandbox numbers.