feat: implement channel healthcheck and sentinel auto-healing

pkg/channels/base.go

@@ -17,6 +17,7 @@ type Channel interface {
 	Send(ctx context.Context, msg bus.OutboundMessage) error
 	IsRunning() bool
 	IsAllowed(senderID string) bool
+	HealthCheck(ctx context.Context) error
 }
 
 type BaseChannel struct {

pkg/channels/manager.go

@@ -217,6 +217,31 @@ func (m *Manager) StopAll(ctx context.Context) error {
 	return nil
 }
 
+func (m *Manager) CheckHealth(ctx context.Context) map[string]error {
+	m.mu.RLock()
+	defer m.mu.RUnlock()
+
+	results := make(map[string]error)
+	for name, channel := range m.channels {
+		results[name] = channel.HealthCheck(ctx)
+	}
+	return results
+}
+
+func (m *Manager) RestartChannel(ctx context.Context, name string) error {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+
+	channel, ok := m.channels[name]
+	if !ok {
+		return fmt.Errorf("channel %s not found", name)
+	}
+
+	logger.InfoCF("channels", "Restarting channel", map[string]interface{}{"channel": name})
+	_ = channel.Stop(ctx)
+	return channel.Start(ctx)
+}
+
 func (m *Manager) dispatchOutbound(ctx context.Context) {
 	logger.InfoC("channels", "Outbound dispatcher started")
 
@@ -271,10 +296,7 @@ func (m *Manager) GetStatus() map[string]interface{} {
 
 	status := make(map[string]interface{})
 	for name, channel := range m.channels {
-		status[name] = map[string]interface{}{
-			"enabled": true,
-			"running": channel.IsRunning(),
-		}
+		status[name] = map[string]interface{}{}
 	}
 	return status
 }

pkg/channels/telegram.go

@@ -74,6 +74,16 @@ func (c *TelegramChannel) SetTranscriber(transcriber *voice.GroqTranscriber) {
 	c.transcriber = transcriber
 }
 
+func (c *TelegramChannel) HealthCheck(ctx context.Context) error {
+	if !c.IsRunning() {
+		return fmt.Errorf("telegram bot not running")
+	}
+	hCtx, cancel := withTelegramAPITimeout(ctx)
+	defer cancel()
+	_, err := c.bot.GetMe(hCtx)
+	return err
+}
+
 func (c *TelegramChannel) Start(ctx context.Context) error {
 	if c.IsRunning() {
 		return nil

pkg/sentinel/service.go

@@ -1,12 +1,14 @@
 package sentinel
 
 import (
+	"context"
 	"fmt"
 	"os"
 	"path/filepath"
 	"sync"
 	"time"
 
+	"clawgo/pkg/channels"
 	"clawgo/pkg/config"
 	"clawgo/pkg/lifecycle"
 	"clawgo/pkg/logger"
@@ -23,6 +25,7 @@ type Service struct {
 	runner     *lifecycle.LoopRunner
 	mu         sync.RWMutex
 	lastAlerts map[string]time.Time
+	mgr        *channels.Manager
 }
 
 func NewService(cfgPath, workspace string, intervalSec int, autoHeal bool, onAlert AlertFunc) *Service {
@@ -40,6 +43,10 @@ func NewService(cfgPath, workspace string, intervalSec int, autoHeal bool, onAle
 	}
 }
 
+func (s *Service) SetManager(mgr *channels.Manager) {
+	s.mgr = mgr
+}
+
 func (s *Service) Start() {
 	if !s.runner.Start(s.loop) {
 		return
@@ -76,6 +83,7 @@ func (s *Service) runChecks() {
 	issues := s.checkConfig()
 	issues = append(issues, s.checkMemory()...)
 	issues = append(issues, s.checkLogs()...)
+	issues = append(issues, s.checkChannels()...)
 
 	if len(issues) == 0 {
 		return
@@ -86,6 +94,34 @@ func (s *Service) runChecks() {
 	}
 }
 
+func (s *Service) checkChannels() []string {
+	if s.mgr == nil {
+		return nil
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+
+	health := s.mgr.CheckHealth(ctx)
+	var issues []string
+	for name, err := range health {
+		if err != nil {
+			msg := fmt.Sprintf("sentinel: channel %s health check failed: %v", name, err)
+			issues = append(issues, msg)
+			if s.autoHeal {
+				go func(n string) {
+					logger.InfoCF("sentinel", "Attempting auto-heal for channel", map[string]interface{}{"channel": n})
+					if rErr := s.mgr.RestartChannel(context.Background(), n); rErr != nil {
+						logger.ErrorCF("sentinel", "Auto-heal restart failed", map[string]interface{}{"channel": n, "error": rErr.Error()})
+					} else {
+						logger.InfoCF("sentinel", "Auto-heal successful", map[string]interface{}{"channel": n})
+					}
+				}(name)
+			}
+		}
+	}
+	return issues
+}
+
 func (s *Service) checkConfig() []string {
 	_, err := os.Stat(s.cfgPath)
 	if err != nil {