diff --git a/cmd/cbzoptimizer/commands/watch_command.go b/cmd/cbzoptimizer/commands/watch_command.go
index 181d81c..b3bcb3e 100644
--- a/cmd/cbzoptimizer/commands/watch_command.go
+++ b/cmd/cbzoptimizer/commands/watch_command.go
@@ -1,6 +1,7 @@
 package commands
 
 import (
+	"errors"
 	"fmt"
 	"io/fs"
 	"os"
@@ -158,6 +159,10 @@ func addRecursiveWatch(watcher *fsnotify.Watcher, rootPath string) error {
 			return nil
 		}
 		if err := watcher.Add(path); err != nil {
+			if errors.Is(err, fs.ErrPermission) {
+				log.Warn().Err(err).Str("path", path).Msg("Skipping unreadable directory while setting up watch")
+				return nil
+			}
 			return fmt.Errorf("failed to watch directory %s: %w", path, err)
 		}
 		return nil
@@ -208,15 +213,21 @@ type eventDebouncer struct {
 	delay   time.Duration
 	onQuiet func(path string)
 
-	mu     sync.Mutex
-	timers map[string]*time.Timer
+	mu       sync.Mutex
+	timers   map[string]*debounceTimer
+	inFlight sync.WaitGroup
+	stopping bool
+}
+
+type debounceTimer struct {
+	timer *time.Timer
 }
 
 func newEventDebouncer(delay time.Duration, onQuiet func(path string)) *eventDebouncer {
 	return &eventDebouncer{
 		delay:   delay,
 		onQuiet: onQuiet,
-		timers:  make(map[string]*time.Timer),
+		timers:  make(map[string]*debounceTimer),
 	}
 }
 
@@ -226,18 +237,27 @@ func (d *eventDebouncer) Trigger(path string) {
 	d.mu.Lock()
 	defer d.mu.Unlock()
 
-	if existing, exists := d.timers[path]; exists {
-		existing.Stop()
+	if d.stopping {
+		return
 	}
 
-	var timer *time.Timer
-	timer = time.AfterFunc(d.delay, func() {
+	if existing, exists := d.timers[path]; exists {
+		if existing.timer.Stop() {
+			d.inFlight.Done()
+		}
+	}
+
+	entry := &debounceTimer{}
+	d.inFlight.Add(1)
+	entry.timer = time.AfterFunc(d.delay, func() {
+		defer d.inFlight.Done()
+
 		d.mu.Lock()
 		// Only the still-current timer for this path is allowed to clear the
 		// entry and fire onQuiet. If Trigger raced with this callback and
 		// already installed a newer timer, this stale invocation must not
 		// delete that newer entry or fire early.
-		owns := d.timers[path] == timer
+		owns := d.timers[path] == entry && !d.stopping
 		if owns {
 			delete(d.timers, path)
 		}
@@ -246,25 +266,32 @@ func (d *eventDebouncer) Trigger(path string) {
 			d.onQuiet(path)
 		}
 	})
-	d.timers[path] = timer
+	d.timers[path] = entry
 }
 
 // Stop cancels all pending timers.
 func (d *eventDebouncer) Stop() {
 	d.mu.Lock()
-	defer d.mu.Unlock()
+	d.stopping = true
 	for path, timer := range d.timers {
-		timer.Stop()
+		if timer.timer.Stop() {
+			d.inFlight.Done()
+		}
 		delete(d.timers, path)
 	}
+	d.mu.Unlock()
+	d.inFlight.Wait()
 }
 
 // optimizeQueue is a small worker pool that runs Optimize jobs off of the
 // fsnotify event loop, so a slow conversion never blocks event draining.
 type optimizeQueue struct {
-	jobs    chan string
-	wg      sync.WaitGroup
-	options *utils2.OptimizeOptions
+	jobs     chan string
+	wg       sync.WaitGroup
+	mu       sync.RWMutex
+	stopped  bool
+	options  *utils2.OptimizeOptions
+	optimize func(options *utils2.OptimizeOptions) error
 }
 
 func newOptimizeQueue(workerCount int, options *utils2.OptimizeOptions) *optimizeQueue {
@@ -272,8 +299,9 @@ func newOptimizeQueue(workerCount int, options *utils2.OptimizeOptions) *optimiz
 		workerCount = 1
 	}
 	q := &optimizeQueue{
-		jobs:    make(chan string, 64),
-		options: options,
+		jobs:     make(chan string, 64),
+		options:  options,
+		optimize: utils2.Optimize,
 	}
 	q.wg.Add(workerCount)
 	for i := 0; i < workerCount; i++ {
@@ -307,18 +335,26 @@ func (q *optimizeQueue) process(path string) {
 
 	options := *q.options
 	options.Path = path
-	if err := utils2.Optimize(&options); err != nil {
+	if err := q.optimize(&options); err != nil {
 		log.Error().Err(err).Str("file", path).Msg("Error processing file")
 	}
 }
 
 // Enqueue submits path for processing by the worker pool.
 func (q *optimizeQueue) Enqueue(path string) {
+	q.mu.RLock()
+	defer q.mu.RUnlock()
+	if q.stopped {
+		return
+	}
 	q.jobs <- path
 }
 
 // Stop closes the job queue and waits for in-flight jobs to finish.
 func (q *optimizeQueue) Stop() {
+	q.mu.Lock()
+	q.stopped = true
 	close(q.jobs)
+	q.mu.Unlock()
 	q.wg.Wait()
 }
diff --git a/cmd/cbzoptimizer/commands/watch_command_test.go b/cmd/cbzoptimizer/commands/watch_command_test.go
index 6d171b7..485c3fd 100644
--- a/cmd/cbzoptimizer/commands/watch_command_test.go
+++ b/cmd/cbzoptimizer/commands/watch_command_test.go
@@ -3,6 +3,7 @@ package commands
 import (
 	"os"
 	"path/filepath"
+	"runtime"
 	"sync"
 	"sync/atomic"
 	"testing"
@@ -11,6 +12,7 @@ import (
 	utils2 "github.com/belphemur/CBZOptimizer/v2/internal/utils"
 	"github.com/fsnotify/fsnotify"
 	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
 )
 
 func TestIsComicArchive(t *testing.T) {
@@ -55,9 +57,20 @@ func TestShouldProcessWatchEvent(t *testing.T) {
 }
 
 func TestAddRecursiveWatchSkipsUnreadableSubdirectory(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("permission bits are not reliably enforced on Windows")
+	}
+
 	root := t.TempDir()
 	nested := filepath.Join(root, "nested")
 	assert.NoError(t, os.MkdirAll(nested, 0o755))
+	assert.NoError(t, os.Chmod(nested, 0o000))
+	defer func() {
+		assert.NoError(t, os.Chmod(nested, 0o755))
+	}()
+	if _, err := os.ReadDir(nested); err == nil {
+		t.Skip("cannot make nested directory unreadable in this environment")
+	}
 
 	watcher, err := fsnotify.NewWatcher()
 	assert.NoError(t, err)
@@ -99,11 +112,11 @@ func TestEventDebouncerCoalescesBurstsIntoSingleCall(t *testing.T) {
 	// Simulate a burst of events for the same path.
 	for i := 0; i < 5; i++ {
 		debouncer.Trigger("/tmp/chapter.cbz")
-		time.Sleep(5 * time.Millisecond)
 	}
 
-	time.Sleep(50 * time.Millisecond)
-	assert.EqualValues(t, 1, atomic.LoadInt32(&calls))
+	require.Eventually(t, func() bool {
+		return atomic.LoadInt32(&calls) == 1
+	}, 2*time.Second, 10*time.Millisecond)
 }
 
 func TestEventDebouncerHandlesMultiplePaths(t *testing.T) {
@@ -119,22 +132,41 @@ func TestEventDebouncerHandlesMultiplePaths(t *testing.T) {
 	debouncer.Trigger("/tmp/a.cbz")
 	debouncer.Trigger("/tmp/b.cbz")
 
-	time.Sleep(50 * time.Millisecond)
-
-	mu.Lock()
-	defer mu.Unlock()
-	assert.Equal(t, 1, seen["/tmp/a.cbz"])
-	assert.Equal(t, 1, seen["/tmp/b.cbz"])
+	require.Eventually(t, func() bool {
+		mu.Lock()
+		defer mu.Unlock()
+		return seen["/tmp/a.cbz"] == 1 && seen["/tmp/b.cbz"] == 1
+	}, 2*time.Second, 10*time.Millisecond)
 }
 
 func TestOptimizeQueueSkipsMissingPath(t *testing.T) {
 	q := newOptimizeQueue(1, &utils2.OptimizeOptions{})
 	defer q.Stop()
+	var calls int32
+	done := make(chan struct{}, 1)
+	existing := filepath.Join(t.TempDir(), "existing.cbz")
+	require.NoError(t, os.WriteFile(existing, []byte("data"), 0o644))
+	q.optimize = func(options *utils2.OptimizeOptions) error {
+		if options.Path == existing {
+			atomic.AddInt32(&calls, 1)
+			select {
+			case done <- struct{}{}:
+			default:
+			}
+		}
+		return nil
+	}
 
 	// Enqueue a path that doesn't exist; process should skip it without
 	// panicking or blocking since it never reaches utils2.Optimize.
 	q.Enqueue(filepath.Join(t.TempDir(), "missing.cbz"))
+	q.Enqueue(existing)
 
-	// Give the worker a moment to drain the job.
-	time.Sleep(50 * time.Millisecond)
+	select {
+	case <-done:
+	case <-time.After(2 * time.Second):
+		t.Fatal("timed out waiting for existing path to be processed")
+	}
+
+	assert.EqualValues(t, 1, atomic.LoadInt32(&calls))
 }