fix(channels): fix memory hazards in channel abstraction layer

Address 7 memory/architecture issues affecting long-running gateway
processes on embedded devices (<10MB RAM):

- Fix dispatcher busy-wait: remove select+default pattern that caused
  CPU spin after context cancellation; SubscribeOutbound handles ctx
  internally
- Add TTL janitor for typingStops/placeholders sync.Map entries to
  prevent unbounded accumulation when outbound paths fail
- Reduce queue buffers from 100 to 16 slots (~84% memory reduction)
- Optimize SplitMessage with index-based rune operations to reduce
  intermediate string/rune allocations
- Replace uuid.New() with atomic counter + random prefix for media
  scope IDs (eliminates per-call crypto/rand syscall)
- Lazy channel worker creation: defer goroutine+buffer allocation
  until channel.Start() succeeds

pkg/bus/bus.go

@@ -11,6 +11,8 @@ import (
 // ErrBusClosed is returned when publishing to a closed MessageBus.
 var ErrBusClosed = errors.New("message bus closed")
 
+const defaultBusBufferSize = 16
+
 type MessageBus struct {
 	inbound       chan InboundMessage
 	outbound      chan OutboundMessage
@@ -21,9 +23,9 @@ type MessageBus struct {
 
 func NewMessageBus() *MessageBus {
 	return &MessageBus{
-		inbound:       make(chan InboundMessage, 100),
-		outbound:      make(chan OutboundMessage, 100),
-		outboundMedia: make(chan OutboundMediaMessage, 100),
+		inbound:       make(chan InboundMessage, defaultBusBufferSize),
+		outbound:      make(chan OutboundMessage, defaultBusBufferSize),
+		outboundMedia: make(chan OutboundMediaMessage, defaultBusBufferSize),
 		done:          make(chan struct{}),
 	}
 }

pkg/bus/bus_test.go

@@ -67,7 +67,7 @@ func TestPublishInbound_ContextCancel(t *testing.T) {
 
 	// Fill the buffer
 	ctx := context.Background()
-	for i := 0; i < 100; i++ {
+	for i := 0; i < defaultBusBufferSize; i++ {
 		if err := mb.PublishInbound(ctx, InboundMessage{Content: "fill"}); err != nil {
 			t.Fatalf("fill failed at %d: %v", i, err)
 		}
@@ -194,7 +194,7 @@ func TestPublishInbound_FullBuffer(t *testing.T) {
 	ctx := context.Background()
 
 	// Fill the buffer
-	for i := 0; i < 100; i++ {
+	for i := 0; i < defaultBusBufferSize; i++ {
 		if err := mb.PublishInbound(ctx, InboundMessage{Content: "fill"}); err != nil {
 			t.Fatalf("fill failed at %d: %v", i, err)
 		}

pkg/channels/base.go

@@ -2,10 +2,13 @@ package channels
 
 import (
 	"context"
+	"crypto/rand"
+	"encoding/binary"
+	"encoding/hex"
+	"strconv"
 	"strings"
 	"sync/atomic"
-
-	"github.com/google/uuid"
+	"time"
 
 	"github.com/sipeed/picoclaw/pkg/bus"
 	"github.com/sipeed/picoclaw/pkg/config"
@@ -14,6 +17,28 @@ import (
 	"github.com/sipeed/picoclaw/pkg/media"
 )
 
+var (
+	fastIDCounter uint64
+	fastIDPrefix  string
+)
+
+func init() {
+	// One-time read from crypto/rand for a unique prefix (single syscall).
+	var b [8]byte
+	if _, err := rand.Read(b[:]); err != nil {
+		// fallback to time-based prefix
+		binary.BigEndian.PutUint64(b[:], uint64(time.Now().UnixNano()))
+	}
+	fastIDPrefix = hex.EncodeToString(b[:])
+}
+
+// fastID generates a unique ID using a random prefix and an atomic counter.
+// Much cheaper than uuid.New() which calls crypto/rand.Read on every invocation.
+func fastID() string {
+	n := atomic.AddUint64(&fastIDCounter, 1)
+	return fastIDPrefix + strconv.FormatUint(n, 16)
+}
+
 type Channel interface {
 	Name() string
 	Start(ctx context.Context) error
@@ -264,7 +289,7 @@ func (c *BaseChannel) GetPlaceholderRecorder() PlaceholderRecorder {
 func BuildMediaScope(channel, chatID, messageID string) string {
 	id := messageID
 	if id == "" {
-		id = uuid.New().String()
+		id = fastID()
 	}
 	return channel + ":" + chatID + ":" + id
 }

pkg/channels/manager.go

@@ -26,14 +26,30 @@ import (
 )
 
 const (
-	defaultChannelQueueSize = 100
+	defaultChannelQueueSize = 16
 	defaultRateLimit        = 10 // default 10 msg/s
 	maxRetries              = 3
 	rateLimitDelay          = 1 * time.Second
 	baseBackoff             = 500 * time.Millisecond
 	maxBackoff              = 8 * time.Second
+
+	janitorInterval = 10 * time.Second
+	typingStopTTL   = 5 * time.Minute
+	placeholderTTL  = 10 * time.Minute
 )
 
+// typingEntry wraps a typing stop function with a creation timestamp for TTL eviction.
+type typingEntry struct {
+	stop      func()
+	createdAt time.Time
+}
+
+// placeholderEntry wraps a placeholder ID with a creation timestamp for TTL eviction.
+type placeholderEntry struct {
+	id        string
+	createdAt time.Time
+}
+
 // channelRateConfig maps channel name to per-second rate limit.
 var channelRateConfig = map[string]float64{
 	"telegram": 20,
@@ -73,14 +89,14 @@ type asyncTask struct {
 // Implements PlaceholderRecorder.
 func (m *Manager) RecordPlaceholder(channel, chatID, placeholderID string) {
 	key := channel + ":" + chatID
-	m.placeholders.Store(key, placeholderID)
+	m.placeholders.Store(key, placeholderEntry{id: placeholderID, createdAt: time.Now()})
 }
 
 // RecordTypingStop registers a typing stop function for later invocation.
 // Implements PlaceholderRecorder.
 func (m *Manager) RecordTypingStop(channel, chatID string, stop func()) {
 	key := channel + ":" + chatID
-	m.typingStops.Store(key, stop)
+	m.typingStops.Store(key, typingEntry{stop: stop, createdAt: time.Now()})
 }
 
 // preSend handles typing stop and placeholder editing before sending a message.
@@ -90,16 +106,16 @@ func (m *Manager) preSend(ctx context.Context, name string, msg bus.OutboundMess
 
 	// 1. Stop typing
 	if v, loaded := m.typingStops.LoadAndDelete(key); loaded {
-		if stop, ok := v.(func()); ok {
-			stop() // idempotent, safe
+		if entry, ok := v.(typingEntry); ok {
+			entry.stop() // idempotent, safe
 		}
 	}
 
 	// 2. Try editing placeholder
 	if v, loaded := m.placeholders.LoadAndDelete(key); loaded {
-		if placeholderID, ok := v.(string); ok && placeholderID != "" {
+		if entry, ok := v.(placeholderEntry); ok && entry.id != "" {
 			if editor, ok := ch.(MessageEditor); ok {
-				if err := editor.EditMessage(ctx, msg.ChatID, placeholderID, msg.Content); err == nil {
+				if err := editor.EditMessage(ctx, msg.ChatID, entry.id, msg.Content); err == nil {
 					return true // edited successfully, skip Send
 				}
 				// edit failed → fall through to normal Send
@@ -156,7 +172,6 @@ func (m *Manager) initChannel(name, displayName string) {
 			setter.SetPlaceholderRecorder(m)
 		}
 		m.channels[name] = ch
-		m.workers[name] = newChannelWorker(name, ch)
 		logger.InfoCF("channels", "Channel enabled successfully", map[string]any{
 			"channel": displayName,
 		})
@@ -285,11 +300,11 @@ func (m *Manager) StartAll(ctx context.Context) error {
 				"channel": name,
 				"error":   err.Error(),
 			})
+			continue
 		}
-	}
-
-	// Start per-channel workers
-	for name, w := range m.workers {
+		// Lazily create worker only after channel starts successfully
+		w := newChannelWorker(name, channel)
+		m.workers[name] = w
 		go m.runWorker(dispatchCtx, name, w)
 		go m.runMediaWorker(dispatchCtx, name, w)
 	}
@@ -298,6 +313,9 @@ func (m *Manager) StartAll(ctx context.Context) error {
 	go m.dispatchOutbound(dispatchCtx)
 	go m.dispatchOutboundMedia(dispatchCtx)
 
+	// Start the TTL janitor that cleans up stale typing/placeholder entries
+	go m.runTTLJanitor(dispatchCtx)
+
 	// Start shared HTTP server if configured
 	if m.httpServer != nil {
 		go func() {
@@ -342,17 +360,25 @@ func (m *Manager) StopAll(ctx context.Context) error {
 
 	// Close all worker queues and wait for them to drain
 	for _, w := range m.workers {
-		close(w.queue)
+		if w != nil {
+			close(w.queue)
+		}
 	}
 	for _, w := range m.workers {
-		<-w.done
+		if w != nil {
+			<-w.done
+		}
 	}
 	// Close all media worker queues and wait for them to drain
 	for _, w := range m.workers {
-		close(w.mediaQueue)
+		if w != nil {
+			close(w.mediaQueue)
+		}
 	}
 	for _, w := range m.workers {
-		<-w.mediaDone
+		if w != nil {
+			<-w.mediaDone
+		}
 	}
 
 	// Stop all channels
@@ -487,40 +513,39 @@ func (m *Manager) dispatchOutbound(ctx context.Context) {
 	logger.InfoC("channels", "Outbound dispatcher started")
 
 	for {
-		select {
-		case <-ctx.Done():
+		msg, ok := m.bus.SubscribeOutbound(ctx)
+		if !ok {
 			logger.InfoC("channels", "Outbound dispatcher stopped")
 			return
-		default:
-			msg, ok := m.bus.SubscribeOutbound(ctx)
-			if !ok {
-				continue
-			}
+		}
 
-			// Silently skip internal channels
-			if constants.IsInternalChannel(msg.Channel) {
-				continue
-			}
+		// Silently skip internal channels
+		if constants.IsInternalChannel(msg.Channel) {
+			continue
+		}
 
-			m.mu.RLock()
-			_, exists := m.channels[msg.Channel]
-			w, wExists := m.workers[msg.Channel]
-			m.mu.RUnlock()
+		m.mu.RLock()
+		_, exists := m.channels[msg.Channel]
+		w, wExists := m.workers[msg.Channel]
+		m.mu.RUnlock()
 
-			if !exists {
-				logger.WarnCF("channels", "Unknown channel for outbound message", map[string]any{
-					"channel": msg.Channel,
-				})
-				continue
-			}
+		if !exists {
+			logger.WarnCF("channels", "Unknown channel for outbound message", map[string]any{
+				"channel": msg.Channel,
+			})
+			continue
+		}
 
-			if wExists {
-				select {
-				case w.queue <- msg:
-				case <-ctx.Done():
-					return
-				}
+		if wExists && w != nil {
+			select {
+			case w.queue <- msg:
+			case <-ctx.Done():
+				return
 			}
+		} else if exists {
+			logger.WarnCF("channels", "Channel has no active worker, skipping message", map[string]any{
+				"channel": msg.Channel,
+			})
 		}
 	}
 }
@@ -529,40 +554,39 @@ func (m *Manager) dispatchOutboundMedia(ctx context.Context) {
 	logger.InfoC("channels", "Outbound media dispatcher started")
 
 	for {
-		select {
-		case <-ctx.Done():
+		msg, ok := m.bus.SubscribeOutboundMedia(ctx)
+		if !ok {
 			logger.InfoC("channels", "Outbound media dispatcher stopped")
 			return
-		default:
-			msg, ok := m.bus.SubscribeOutboundMedia(ctx)
-			if !ok {
-				continue
-			}
+		}
 
-			// Silently skip internal channels
-			if constants.IsInternalChannel(msg.Channel) {
-				continue
-			}
+		// Silently skip internal channels
+		if constants.IsInternalChannel(msg.Channel) {
+			continue
+		}
 
-			m.mu.RLock()
-			_, exists := m.channels[msg.Channel]
-			w, wExists := m.workers[msg.Channel]
-			m.mu.RUnlock()
+		m.mu.RLock()
+		_, exists := m.channels[msg.Channel]
+		w, wExists := m.workers[msg.Channel]
+		m.mu.RUnlock()
 
-			if !exists {
-				logger.WarnCF("channels", "Unknown channel for outbound media message", map[string]any{
-					"channel": msg.Channel,
-				})
-				continue
-			}
+		if !exists {
+			logger.WarnCF("channels", "Unknown channel for outbound media message", map[string]any{
+				"channel": msg.Channel,
+			})
+			continue
+		}
 
-			if wExists {
-				select {
-				case w.mediaQueue <- msg:
-				case <-ctx.Done():
-					return
-				}
+		if wExists && w != nil {
+			select {
+			case w.mediaQueue <- msg:
+			case <-ctx.Done():
+				return
 			}
+		} else if exists {
+			logger.WarnCF("channels", "Channel has no active worker, skipping media message", map[string]any{
+				"channel": msg.Channel,
+			})
 		}
 	}
 }
@@ -644,6 +668,40 @@ func (m *Manager) sendMediaWithRetry(ctx context.Context, name string, w *channe
 	})
 }
 
+// runTTLJanitor periodically scans the typingStops and placeholders maps
+// and evicts entries that have exceeded their TTL. This prevents memory
+// accumulation when outbound paths fail to trigger preSend (e.g. LLM errors).
+func (m *Manager) runTTLJanitor(ctx context.Context) {
+	ticker := time.NewTicker(janitorInterval)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case now := <-ticker.C:
+			m.typingStops.Range(func(key, value any) bool {
+				if entry, ok := value.(typingEntry); ok {
+					if now.Sub(entry.createdAt) > typingStopTTL {
+						if _, loaded := m.typingStops.LoadAndDelete(key); loaded {
+							entry.stop() // idempotent, safe
+						}
+					}
+				}
+				return true
+			})
+			m.placeholders.Range(func(key, value any) bool {
+				if entry, ok := value.(placeholderEntry); ok {
+					if now.Sub(entry.createdAt) > placeholderTTL {
+						m.placeholders.Delete(key)
+					}
+				}
+				return true
+			})
+		}
+	}
+}
+
 func (m *Manager) GetChannel(name string) (Channel, bool) {
 	m.mu.RLock()
 	defer m.mu.RUnlock()
@@ -680,13 +738,12 @@ func (m *Manager) RegisterChannel(name string, channel Channel) {
 	m.mu.Lock()
 	defer m.mu.Unlock()
 	m.channels[name] = channel
-	m.workers[name] = newChannelWorker(name, channel)
 }
 
 func (m *Manager) UnregisterChannel(name string) {
 	m.mu.Lock()
 	defer m.mu.Unlock()
-	if w, ok := m.workers[name]; ok {
+	if w, ok := m.workers[name]; ok && w != nil {
 		close(w.queue)
 		<-w.done
 		close(w.mediaQueue)
@@ -712,7 +769,7 @@ func (m *Manager) SendToChannel(ctx context.Context, channelName, chatID, conten
 		Content: content,
 	}
 
-	if wExists {
+	if wExists && w != nil {
 		select {
 		case w.queue <- msg:
 			return nil

pkg/channels/manager_test.go

@@ -632,3 +632,233 @@ func TestSendWithRetry_PreSendEditsPlaceholder(t *testing.T) {
 		t.Fatal("expected Send to NOT be called when placeholder was edited")
 	}
 }
+
+// --- Dispatcher exit tests (Step 1) ---
+
+func TestDispatcherExitsOnCancel(t *testing.T) {
+	mb := bus.NewMessageBus()
+	defer mb.Close()
+
+	m := &Manager{
+		channels: make(map[string]Channel),
+		workers:  make(map[string]*channelWorker),
+		bus:      mb,
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan struct{})
+
+	go func() {
+		m.dispatchOutbound(ctx)
+		close(done)
+	}()
+
+	// Cancel context and verify the dispatcher exits quickly
+	cancel()
+
+	select {
+	case <-done:
+		// success
+	case <-time.After(2 * time.Second):
+		t.Fatal("dispatchOutbound did not exit within 2s after context cancel")
+	}
+}
+
+func TestDispatcherMediaExitsOnCancel(t *testing.T) {
+	mb := bus.NewMessageBus()
+	defer mb.Close()
+
+	m := &Manager{
+		channels: make(map[string]Channel),
+		workers:  make(map[string]*channelWorker),
+		bus:      mb,
+	}
+
+	ctx, cancel := context.WithCancel(context.Background())
+	done := make(chan struct{})
+
+	go func() {
+		m.dispatchOutboundMedia(ctx)
+		close(done)
+	}()
+
+	cancel()
+
+	select {
+	case <-done:
+		// success
+	case <-time.After(2 * time.Second):
+		t.Fatal("dispatchOutboundMedia did not exit within 2s after context cancel")
+	}
+}
+
+// --- TTL Janitor tests (Step 2) ---
+
+func TestTypingStopJanitorEviction(t *testing.T) {
+	m := newTestManager()
+
+	var stopCalled atomic.Bool
+	// Store a typing entry with a creation time far in the past
+	m.typingStops.Store("test:123", typingEntry{
+		stop:      func() { stopCalled.Store(true) },
+		createdAt: time.Now().Add(-10 * time.Minute), // well past typingStopTTL
+	})
+
+	// Run janitor with a short-lived context
+	ctx, cancel := context.WithCancel(context.Background())
+
+	// Manually trigger the janitor logic once by simulating a tick
+	go func() {
+		// Override janitor to run immediately
+		now := time.Now()
+		m.typingStops.Range(func(key, value any) bool {
+			if entry, ok := value.(typingEntry); ok {
+				if now.Sub(entry.createdAt) > typingStopTTL {
+					if _, loaded := m.typingStops.LoadAndDelete(key); loaded {
+						entry.stop()
+					}
+				}
+			}
+			return true
+		})
+		cancel()
+	}()
+
+	<-ctx.Done()
+
+	if !stopCalled.Load() {
+		t.Fatal("expected typing stop function to be called by janitor eviction")
+	}
+
+	// Verify entry was deleted
+	if _, loaded := m.typingStops.Load("test:123"); loaded {
+		t.Fatal("expected typing entry to be deleted after eviction")
+	}
+}
+
+func TestPlaceholderJanitorEviction(t *testing.T) {
+	m := newTestManager()
+
+	// Store a placeholder entry with a creation time far in the past
+	m.placeholders.Store("test:456", placeholderEntry{
+		id:        "msg_old",
+		createdAt: time.Now().Add(-20 * time.Minute), // well past placeholderTTL
+	})
+
+	// Simulate janitor logic
+	now := time.Now()
+	m.placeholders.Range(func(key, value any) bool {
+		if entry, ok := value.(placeholderEntry); ok {
+			if now.Sub(entry.createdAt) > placeholderTTL {
+				m.placeholders.Delete(key)
+			}
+		}
+		return true
+	})
+
+	// Verify entry was deleted
+	if _, loaded := m.placeholders.Load("test:456"); loaded {
+		t.Fatal("expected placeholder entry to be deleted after eviction")
+	}
+}
+
+func TestPreSendStillWorksWithWrappedTypes(t *testing.T) {
+	m := newTestManager()
+	var stopCalled bool
+	var editCalled bool
+
+	ch := &mockMessageEditor{
+		mockChannel: mockChannel{
+			sendFn: func(_ context.Context, _ bus.OutboundMessage) error {
+				return nil
+			},
+		},
+		editFn: func(_ context.Context, chatID, messageID, content string) error {
+			editCalled = true
+			if messageID != "ph_id" {
+				t.Fatalf("expected messageID ph_id, got %s", messageID)
+			}
+			return nil
+		},
+	}
+
+	// Use the new wrapped types via the public API
+	m.RecordTypingStop("test", "chat1", func() {
+		stopCalled = true
+	})
+	m.RecordPlaceholder("test", "chat1", "ph_id")
+
+	msg := bus.OutboundMessage{Channel: "test", ChatID: "chat1", Content: "response"}
+	edited := m.preSend(context.Background(), "test", msg, ch)
+
+	if !stopCalled {
+		t.Fatal("expected typing stop to be called via wrapped type")
+	}
+	if !editCalled {
+		t.Fatal("expected EditMessage to be called via wrapped type")
+	}
+	if !edited {
+		t.Fatal("expected preSend to return true")
+	}
+}
+
+// --- Lazy worker creation tests (Step 6) ---
+
+func TestLazyWorkerCreation(t *testing.T) {
+	m := newTestManager()
+
+	ch := &mockChannel{
+		sendFn: func(_ context.Context, _ bus.OutboundMessage) error {
+			return nil
+		},
+	}
+
+	// RegisterChannel should NOT create a worker
+	m.RegisterChannel("lazy", ch)
+
+	m.mu.RLock()
+	_, chExists := m.channels["lazy"]
+	_, wExists := m.workers["lazy"]
+	m.mu.RUnlock()
+
+	if !chExists {
+		t.Fatal("expected channel to be registered")
+	}
+	if wExists {
+		t.Fatal("expected worker to NOT be created by RegisterChannel (lazy creation)")
+	}
+}
+
+// --- FastID uniqueness test (Step 5) ---
+
+func TestBuildMediaScope_FastIDUniqueness(t *testing.T) {
+	seen := make(map[string]bool)
+
+	for i := 0; i < 1000; i++ {
+		scope := BuildMediaScope("test", "chat1", "")
+		if seen[scope] {
+			t.Fatalf("duplicate scope generated: %s", scope)
+		}
+		seen[scope] = true
+	}
+
+	// Verify format: "channel:chatID:id"
+	scope := BuildMediaScope("telegram", "42", "")
+	parts := 0
+	for _, c := range scope {
+		if c == ':' {
+			parts++
+		}
+	}
+	if parts != 2 {
+		t.Fatalf("expected scope to have 2 colons (channel:chatID:id), got: %s", scope)
+	}
+}
+
+func TestBuildMediaScope_WithMessageID(t *testing.T) {
+	scope := BuildMediaScope("discord", "chat99", "msg123")
+	expected := "discord:chat99:msg123"
+	if scope != expected {
+		t.Fatalf("expected %s, got %s", expected, scope)
+	}
+}

pkg/channels/split.go

@@ -19,6 +19,7 @@ func SplitMessage(content string, maxLen int) []string {
 	}
 
 	runes := []rune(content)
+	totalLen := len(runes)
 	var messages []string
 
 	// Dynamic buffer: 10% of maxLen, but at least 50 chars if possible
@@ -30,9 +31,11 @@ func SplitMessage(content string, maxLen int) []string {
 		codeBlockBuffer = maxLen / 2
 	}
 
-	for len(runes) > 0 {
-		if len(runes) <= maxLen {
-			messages = append(messages, string(runes))
+	start := 0
+	for start < totalLen {
+		remaining := totalLen - start
+		if remaining <= maxLen {
+			messages = append(messages, string(runes[start:totalLen]))
 			break
 		}
 
@@ -42,77 +45,88 @@ func SplitMessage(content string, maxLen int) []string {
 			effectiveLimit = maxLen / 2
 		}
 
+		end := start + effectiveLimit
+
 		// Find natural split point within the effective limit
-		msgEnd := findLastNewlineRunes(runes[:effectiveLimit], 200)
-		if msgEnd <= 0 {
-			msgEnd = findLastSpaceRunes(runes[:effectiveLimit], 100)
+		msgEnd := findLastNewlineInRange(runes, start, end, 200)
+		if msgEnd <= start {
+			msgEnd = findLastSpaceInRange(runes, start, end, 100)
 		}
-		if msgEnd <= 0 {
-			msgEnd = effectiveLimit
+		if msgEnd <= start {
+			msgEnd = end
 		}
 
 		// Check if this would end with an incomplete code block
-		candidate := runes[:msgEnd]
-		unclosedIdx := findLastUnclosedCodeBlockRunes(candidate)
+		unclosedIdx := findLastUnclosedCodeBlockInRange(runes, start, msgEnd)
 
 		if unclosedIdx >= 0 {
 			// Message would end with incomplete code block
 			// Try to extend up to maxLen to include the closing ```
-			if len(runes) > msgEnd {
-				closingIdx := findNextClosingCodeBlockRunes(runes, msgEnd)
-				if closingIdx > 0 && closingIdx <= maxLen {
+			if totalLen > msgEnd {
+				closingIdx := findNextClosingCodeBlockInRange(runes, msgEnd, totalLen)
+				if closingIdx > 0 && closingIdx-start <= maxLen {
 					// Extend to include the closing ```
 					msgEnd = closingIdx
 				} else {
 					// Code block is too long to fit in one chunk or missing closing fence.
 					// Try to split inside by injecting closing and reopening fences.
-					fenceRunes := runes[unclosedIdx:]
-					headerEnd := findNewlineInRunes(fenceRunes)
+					headerEnd := findNewlineFrom(runes, unclosedIdx)
 					var header string
 					if headerEnd == -1 {
 						header = strings.TrimSpace(string(runes[unclosedIdx : unclosedIdx+3]))
 					} else {
-						header = strings.TrimSpace(string(runes[unclosedIdx : unclosedIdx+headerEnd]))
+						header = strings.TrimSpace(string(runes[unclosedIdx:headerEnd]))
 					}
 					headerEndIdx := unclosedIdx + len([]rune(header))
 					if headerEnd != -1 {
-						headerEndIdx = unclosedIdx + headerEnd
+						headerEndIdx = headerEnd
 					}
 
 					// If we have a reasonable amount of content after the header, split inside
 					if msgEnd > headerEndIdx+20 {
 						// Find a better split point closer to maxLen
-						innerLimit := maxLen - 5 // Leave room for "\n```"
-						betterEnd := findLastNewlineRunes(runes[:innerLimit], 200)
+						innerLimit := start + maxLen - 5 // Leave room for "\n```"
+						if innerLimit > totalLen {
+							innerLimit = totalLen
+						}
+						betterEnd := findLastNewlineInRange(runes, start, innerLimit, 200)
 						if betterEnd > headerEndIdx {
 							msgEnd = betterEnd
 						} else {
 							msgEnd = innerLimit
 						}
-						chunk := strings.TrimRight(string(runes[:msgEnd]), " \t\n\r") + "\n```"
+						chunk := strings.TrimRight(string(runes[start:msgEnd]), " \t\n\r") + "\n```"
 						messages = append(messages, chunk)
-						remaining := strings.TrimSpace(header + "\n" + string(runes[msgEnd:]))
+						remaining := strings.TrimSpace(header + "\n" + string(runes[msgEnd:totalLen]))
+						// Replace the tail of runes with the reconstructed remaining
 						runes = []rune(remaining)
+						totalLen = len(runes)
+						start = 0
 						continue
 					}
 
 					// Otherwise, try to split before the code block starts
-					newEnd := findLastNewlineRunes(runes[:unclosedIdx], 200)
-					if newEnd <= 0 {
-						newEnd = findLastSpaceRunes(runes[:unclosedIdx], 100)
+					newEnd := findLastNewlineInRange(runes, start, unclosedIdx, 200)
+					if newEnd <= start {
+						newEnd = findLastSpaceInRange(runes, start, unclosedIdx, 100)
 					}
-					if newEnd > 0 {
+					if newEnd > start {
 						msgEnd = newEnd
 					} else {
 						// If we can't split before, we MUST split inside (last resort)
-						if unclosedIdx > 20 {
+						if unclosedIdx-start > 20 {
 							msgEnd = unclosedIdx
 						} else {
-							msgEnd = maxLen - 5
-							chunk := strings.TrimRight(string(runes[:msgEnd]), " \t\n\r") + "\n```"
+							splitAt := start + maxLen - 5
+							if splitAt > totalLen {
+								splitAt = totalLen
+							}
+							chunk := strings.TrimRight(string(runes[start:splitAt]), " \t\n\r") + "\n```"
 							messages = append(messages, chunk)
-							remaining := strings.TrimSpace(header + "\n" + string(runes[msgEnd:]))
+							remaining := strings.TrimSpace(header + "\n" + string(runes[splitAt:totalLen]))
 							runes = []rune(remaining)
+							totalLen = len(runes)
+							start = 0
 							continue
 						}
 					}
@@ -120,29 +134,30 @@ func SplitMessage(content string, maxLen int) []string {
 			}
 		}
 
-		if msgEnd <= 0 {
-			msgEnd = effectiveLimit
+		if msgEnd <= start {
+			msgEnd = start + effectiveLimit
 		}
 
-		messages = append(messages, string(runes[:msgEnd]))
-		remaining := strings.TrimSpace(string(runes[msgEnd:]))
-		runes = []rune(remaining)
+		messages = append(messages, string(runes[start:msgEnd]))
+		// Advance start, skipping leading whitespace of next chunk
+		start = msgEnd
+		for start < totalLen && (runes[start] == ' ' || runes[start] == '\t' || runes[start] == '\n' || runes[start] == '\r') {
+			start++
+		}
 	}
 
 	return messages
 }
 
-// findLastUnclosedCodeBlockRunes finds the last opening ``` that doesn't have a closing ```
-// Returns the rune position of the opening ``` or -1 if all code blocks are complete
-func findLastUnclosedCodeBlockRunes(runes []rune) int {
+// findLastUnclosedCodeBlockInRange finds the last opening ``` that doesn't have a closing ```
+// within runes[start:end]. Returns the absolute rune index or -1.
+func findLastUnclosedCodeBlockInRange(runes []rune, start, end int) int {
 	inCodeBlock := false
 	lastOpenIdx := -1
 
-	for i := 0; i < len(runes); i++ {
-		if i+2 < len(runes) && runes[i] == '`' && runes[i+1] == '`' && runes[i+2] == '`' {
-			// Toggle code block state on each fence
+	for i := start; i < end; i++ {
+		if i+2 < end && runes[i] == '`' && runes[i+1] == '`' && runes[i+2] == '`' {
 			if !inCodeBlock {
-				// Entering a code block: record this opening fence
 				lastOpenIdx = i
 			}
 			inCodeBlock = !inCodeBlock
@@ -156,36 +171,21 @@ func findLastUnclosedCodeBlockRunes(runes []rune) int {
 	return -1
 }
 
-// findNextClosingCodeBlockRunes finds the next closing ``` starting from a rune position
-// Returns the rune position after the closing ``` or -1 if not found
-func findNextClosingCodeBlockRunes(runes []rune, startIdx int) int {
-	for i := startIdx; i < len(runes); i++ {
-		if i+2 < len(runes) && runes[i] == '`' && runes[i+1] == '`' && runes[i+2] == '`' {
+// findNextClosingCodeBlockInRange finds the next closing ``` starting from startIdx
+// within runes[startIdx:end]. Returns the absolute index after the closing ``` or -1.
+func findNextClosingCodeBlockInRange(runes []rune, startIdx, end int) int {
+	for i := startIdx; i < end; i++ {
+		if i+2 < end && runes[i] == '`' && runes[i+1] == '`' && runes[i+2] == '`' {
 			return i + 3
 		}
 	}
 	return -1
 }
 
-// findNewlineInRunes finds the first newline character in a rune slice.
-// Returns the rune index of the newline or -1 if not found.
-func findNewlineInRunes(runes []rune) int {
-	for i, r := range runes {
-		if r == '\n' {
-			return i
-		}
-	}
-	return -1
-}
-
-// findLastNewlineRunes finds the last newline character within the last N runes
-// Returns the rune position of the newline or -1 if not found
-func findLastNewlineRunes(runes []rune, searchWindow int) int {
-	searchStart := len(runes) - searchWindow
-	if searchStart < 0 {
-		searchStart = 0
-	}
-	for i := len(runes) - 1; i >= searchStart; i-- {
+// findNewlineFrom finds the first newline character starting from the given index.
+// Returns the absolute index or -1 if not found.
+func findNewlineFrom(runes []rune, from int) int {
+	for i := from; i < len(runes); i++ {
 		if runes[i] == '\n' {
 			return i
 		}
@@ -193,17 +193,32 @@ func findLastNewlineRunes(runes []rune, searchWindow int) int {
 	return -1
 }
 
-// findLastSpaceRunes finds the last space character within the last N runes
-// Returns the rune position of the space or -1 if not found
-func findLastSpaceRunes(runes []rune, searchWindow int) int {
-	searchStart := len(runes) - searchWindow
-	if searchStart < 0 {
-		searchStart = 0
+// findLastNewlineInRange finds the last newline within the last searchWindow runes
+// of the range runes[start:end]. Returns the absolute index or start-1 (indicating not found).
+func findLastNewlineInRange(runes []rune, start, end, searchWindow int) int {
+	searchStart := end - searchWindow
+	if searchStart < start {
+		searchStart = start
 	}
-	for i := len(runes) - 1; i >= searchStart; i-- {
+	for i := end - 1; i >= searchStart; i-- {
+		if runes[i] == '\n' {
+			return i
+		}
+	}
+	return start - 1
+}
+
+// findLastSpaceInRange finds the last space/tab within the last searchWindow runes
+// of the range runes[start:end]. Returns the absolute index or start-1 (indicating not found).
+func findLastSpaceInRange(runes []rune, start, end, searchWindow int) int {
+	searchStart := end - searchWindow
+	if searchStart < start {
+		searchStart = start
+	}
+	for i := end - 1; i >= searchStart; i-- {
 		if runes[i] == ' ' || runes[i] == '\t' {
 			return i
 		}
 	}
-	return -1
+	return start - 1
 }