feat: stabilize swarm engine and enhance observability

Summary of changes:
1. Hard-coded Limits (Fixed): Added SummarizeThreshold and connected limits to config.
2. Silent Database Failures (Fixed): Enhanced error handling and logging for SQLite and NodeActor transitions.
3. "Zombie" Agents (Fixed): Implemented StopAll/Stop methods for graceful shutdown and linked to AgentLoop.
4. Memory ID Corruption (Verified): Ensured valid float formatting and added timestamps for uniqueness.
5. Code Quality: Corrected typos in system prompts and improved /swarm command error reporting.

pkg/agent/loop.go

@@ -133,6 +133,9 @@ func (al *AgentLoop) Run(ctx context.Context) error {
 
 func (al *AgentLoop) Stop() {
 	al.running = false
+	if al.swarm != nil {
+		al.swarm.Stop()
+	}
 }
 
 func (al *AgentLoop) ProcessDirect(ctx context.Context, content, sessionKey string) (string, error) {

pkg/swarm/config/config.go

@@ -21,6 +21,7 @@ type LimitsConfig struct {
 	MaxRetries         int           `json:"max_retries" env:"PICOCLAW_SWARM_MAX_RETRIES"`
 	MaxIterations      int           `json:"max_iterations" env:"PICOCLAW_SWARM_MAX_ITERATIONS"`
 	PruningMsgKeep     int           `json:"pruning_msg_keep" env:"PICOCLAW_SWARM_PRUNING_MSG_KEEP"`
+	SummarizeThreshold int           `json:"summarize_threshold" env:"PICOCLAW_SWARM_SUMMARIZE_THRESHOLD"`
 }
 
 type ResilienceConfig struct {
@@ -50,6 +51,7 @@ func DefaultSwarmConfig() SwarmConfig {
 			MaxRetries:         3,
 			MaxIterations:      10,
 			PruningMsgKeep:     6,  // Keep last 6 messages when pruning
+			SummarizeThreshold: 20, // Summarize after 20 messages
 		},
 		Resilience: ResilienceConfig{
 			RetryBackoff: 2 * time.Second,

pkg/swarm/memory/chromem_store.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"runtime"
+	"time"
 
 	"github.com/philippgille/chromem-go"
 	"github.com/sipeed/picoclaw/pkg/swarm/core"
@@ -42,7 +43,7 @@ func (s *ChromemStore) SaveFact(ctx context.Context, fact core.Fact) error {
 	meta["source"] = fact.Source
 
 	doc := chromem.Document{
-		ID:       fmt.Sprintf("%s_%d", fact.SwarmID, fact.Confidence),
+		ID:       fmt.Sprintf("%s_%f_%d", fact.SwarmID, fact.Confidence, time.Now().UnixNano()),
 		Content:  fact.Content,
 		Metadata: meta,
 	}

pkg/swarm/memory/sqlite_store.go

@@ -17,14 +17,23 @@ func NewSQLiteStore(dbPath string) (*SQLiteStore, error) {
 	db, err := sql.Open("sqlite", dbPath)
 	if err != nil { return nil, err }
 	s := &SQLiteStore{db: db}
-	s.init()
+	if err := s.init(); err != nil {
+		return nil, err
+	}
 	return s, nil
 }
 
-func (s *SQLiteStore) init() {
+func (s *SQLiteStore) init() error {
-	s.db.Exec(`CREATE TABLE IF NOT EXISTS swarms (id TEXT PRIMARY KEY, goal TEXT, status TEXT, created_at DATETIME);`)
+	if _, err := s.db.Exec(`CREATE TABLE IF NOT EXISTS swarms (id TEXT PRIMARY KEY, goal TEXT, status TEXT, created_at DATETIME);`); err != nil {
-	s.db.Exec(`CREATE TABLE IF NOT EXISTS nodes (id TEXT PRIMARY KEY, swarm_id TEXT, parent_id TEXT, role JSON, task TEXT, status TEXT, output TEXT, stats JSON);`)
+		return err
-	s.db.Exec(`CREATE TABLE IF NOT EXISTS facts (swarm_id TEXT, content TEXT, confidence REAL, source TEXT, metadata JSON);`)
+	}
+	if _, err := s.db.Exec(`CREATE TABLE IF NOT EXISTS nodes (id TEXT PRIMARY KEY, swarm_id TEXT, parent_id TEXT, role JSON, task TEXT, status TEXT, output TEXT, stats JSON);`); err != nil {
+		return err
+	}
+	if _, err := s.db.Exec(`CREATE TABLE IF NOT EXISTS facts (swarm_id TEXT, content TEXT, confidence REAL, source TEXT, metadata JSON);`); err != nil {
+		return err
+	}
+	return nil
 }
 
 func (s *SQLiteStore) CreateSwarm(ctx context.Context, sw *core.Swarm) error {
@@ -35,8 +44,10 @@ func (s *SQLiteStore) CreateSwarm(ctx context.Context, sw *core.Swarm) error {
 func (s *SQLiteStore) GetSwarm(ctx context.Context, id core.SwarmID) (*core.Swarm, error) {
 	row := s.db.QueryRowContext(ctx, "SELECT id, goal, status, created_at FROM swarms WHERE id=?", id)
 	var sw core.Swarm
-	err := row.Scan(&sw.ID, &sw.Goal, &sw.Status, &sw.CreatedAt)
+	if err := row.Scan(&sw.ID, &sw.Goal, &sw.Status, &sw.CreatedAt); err != nil {
-	return &sw, err
+		return nil, err
+	}
+	return &sw, nil
 }
 
 func (s *SQLiteStore) UpdateSwarm(ctx context.Context, sw *core.Swarm) error {
@@ -45,21 +56,32 @@ func (s *SQLiteStore) UpdateSwarm(ctx context.Context, sw *core.Swarm) error {
 }
 
 func (s *SQLiteStore) ListSwarms(ctx context.Context, status core.SwarmStatus) ([]*core.Swarm, error) {
-	rows, _ := s.db.QueryContext(ctx, "SELECT id, goal, status, created_at FROM swarms WHERE status=?", status)
+	rows, err := s.db.QueryContext(ctx, "SELECT id, goal, status, created_at FROM swarms WHERE status=?", status)
+	if err != nil {
+		return nil, err
+	}
 	defer rows.Close()
 	var list []*core.Swarm
 	for rows.Next() {
 		var sw core.Swarm
-		rows.Scan(&sw.ID, &sw.Goal, &sw.Status, &sw.CreatedAt)
+		if err := rows.Scan(&sw.ID, &sw.Goal, &sw.Status, &sw.CreatedAt); err != nil {
+			return nil, err
+		}
 		list = append(list, &sw)
 	}
 	return list, nil
 }
 
 func (s *SQLiteStore) CreateNode(ctx context.Context, n *core.Node) error {
-	role, _ := json.Marshal(n.Role)
+	role, err := json.Marshal(n.Role)
-	stats, _ := json.Marshal(n.Stats)
+	if err != nil {
-	_, err := s.db.ExecContext(ctx, "INSERT INTO nodes VALUES (?, ?, ?, ?, ?, ?, ?, ?)", n.ID, n.SwarmID, n.ParentID, role, n.Task, n.Status, n.Output, stats)
+		return err
+	}
+	stats, err := json.Marshal(n.Stats)
+	if err != nil {
+		return err
+	}
+	_, err = s.db.ExecContext(ctx, "INSERT INTO nodes VALUES (?, ?, ?, ?, ?, ?, ?, ?)", n.ID, n.SwarmID, n.ParentID, role, n.Task, n.Status, n.Output, stats)
 	return err
 }
 
@@ -67,36 +89,57 @@ func (s *SQLiteStore) GetNode(ctx context.Context, id core.NodeID) (*core.Node,
 	row := s.db.QueryRowContext(ctx, "SELECT * FROM nodes WHERE id=?", id)
 	var n core.Node
 	var role, stats []byte
-	row.Scan(&n.ID, &n.SwarmID, &n.ParentID, &role, &n.Task, &n.Status, &n.Output, &stats)
+	if err := row.Scan(&n.ID, &n.SwarmID, &n.ParentID, &role, &n.Task, &n.Status, &n.Output, &stats); err != nil {
-	json.Unmarshal(role, &n.Role)
+		return nil, err
-	json.Unmarshal(stats, &n.Stats)
+	}
+	if err := json.Unmarshal(role, &n.Role); err != nil {
+		return nil, err
+	}
+	if err := json.Unmarshal(stats, &n.Stats); err != nil {
+		return nil, err
+	}
 	return &n, nil
 }
 
 func (s *SQLiteStore) UpdateNode(ctx context.Context, n *core.Node) error {
-	stats, _ := json.Marshal(n.Stats)
+	stats, err := json.Marshal(n.Stats)
-	_, err := s.db.ExecContext(ctx, "UPDATE nodes SET status=?, output=?, stats=? WHERE id=?", n.Status, n.Output, stats, n.ID)
+	if err != nil {
+		return err
+	}
+	_, err = s.db.ExecContext(ctx, "UPDATE nodes SET status=?, output=?, stats=? WHERE id=?", n.Status, n.Output, stats, n.ID)
 	return err
 }
 
 func (s *SQLiteStore) GetSwarmNodes(ctx context.Context, id core.SwarmID) ([]*core.Node, error) {
-	rows, _ := s.db.QueryContext(ctx, "SELECT * FROM nodes WHERE swarm_id=?", id)
+	rows, err := s.db.QueryContext(ctx, "SELECT * FROM nodes WHERE swarm_id=?", id)
+	if err != nil {
+		return nil, err
+	}
 	defer rows.Close()
 	var list []*core.Node
 	for rows.Next() {
 		var n core.Node
 		var role, stats []byte
-		rows.Scan(&n.ID, &n.SwarmID, &n.ParentID, &role, &n.Task, &n.Status, &n.Output, &stats)
+		if err := rows.Scan(&n.ID, &n.SwarmID, &n.ParentID, &role, &n.Task, &n.Status, &n.Output, &stats); err != nil {
-		json.Unmarshal(role, &n.Role)
+			return nil, err
-		json.Unmarshal(stats, &n.Stats)
+		}
+		if err := json.Unmarshal(role, &n.Role); err != nil {
+			return nil, err
+		}
+		if err := json.Unmarshal(stats, &n.Stats); err != nil {
+			return nil, err
+		}
 		list = append(list, &n)
 	}
 	return list, nil
 }
 
 func (s *SQLiteStore) SaveFact(ctx context.Context, f core.Fact) error {
-	meta, _ := json.Marshal(f.Metadata)
+	meta, err := json.Marshal(f.Metadata)
-	_, err := s.db.ExecContext(ctx, "INSERT INTO facts VALUES (?, ?, ?, ?, ?)", f.SwarmID, f.Content, f.Confidence, f.Source, meta)
+	if err != nil {
+		return err
+	}
+	_, err = s.db.ExecContext(ctx, "INSERT INTO facts VALUES (?, ?, ?, ?, ?)", f.SwarmID, f.Content, f.Confidence, f.Source, meta)
 	return err
 }
 
@@ -109,12 +152,17 @@ func (s *SQLiteStore) SearchFacts(ctx context.Context, id core.SwarmID, q string
 		args = []any{"%" + q + "%", limit}
 	}
 
-	rows, _ := s.db.QueryContext(ctx, query, args...)
+	rows, err := s.db.QueryContext(ctx, query, args...)
+	if err != nil {
+		return nil, err
+	}
 	defer rows.Close()
 	var res []core.FactResult
 	for rows.Next() {
 		var content string
-		rows.Scan(&content)
+		if err := rows.Scan(&content); err != nil {
+			return nil, err
+		}
 		res = append(res, core.FactResult{Content: content, Score: 1.0})
 	}
 	return res, nil

pkg/swarm/prompt/prompts.go

@@ -13,7 +13,7 @@ SWARM CONTEXT:
 - Your Role: %s
 
 INSTRUCTIONS:
-1. FOCUS: Stick strictly to your assigned role. Do not halllucinate capabilities you don't have.
+1. FOCUS: Stick strictly to your assigned role. Do not hallucinate capabilities you don't have.
 2. COLLABORATION: If you need information you can't get, ask for it clearly.
 3. OUTPUT: Provide clear, structured reasoning.
 4. TOOLS: Use available tools to gather facts. Do not guess.

pkg/swarm/runtime/node.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"log/slog"
 
+	"github.com/sipeed/picoclaw/pkg/swarm/config"
 	"github.com/sipeed/picoclaw/pkg/swarm/core"
 	"github.com/sipeed/picoclaw/pkg/tools"
 )
@@ -17,11 +18,12 @@ type NodeActor struct {
 	LLM    core.LLMClient
 	Tools  *tools.ToolRegistry
 	Policy *core.PolicyChecker
+	Config config.SwarmConfig
 
 	peerInsights []string // Buffer for thoughts heard from peers
 }
 
-func NewNodeActor(data *core.Node, bus core.EventBus, store core.SwarmStore, llm core.LLMClient, toolRegistry *tools.ToolRegistry, policy *core.PolicyChecker) *NodeActor {
+func NewNodeActor(data *core.Node, bus core.EventBus, store core.SwarmStore, llm core.LLMClient, toolRegistry *tools.ToolRegistry, policy *core.PolicyChecker, cfg config.SwarmConfig) *NodeActor {
 	return &NodeActor{
 		Data:         data,
 		Bus:          bus,
@@ -29,13 +31,16 @@ func NewNodeActor(data *core.Node, bus core.EventBus, store core.SwarmStore, llm
 		LLM:          llm,
 		Tools:        toolRegistry,
 		Policy:       policy,
+		Config:       cfg,
 		peerInsights: make([]string, 0),
 	}
 }
 
 func (n *NodeActor) Start(ctx context.Context) {
 	n.Data.Status = core.NodeStatusRunning
-	n.Store.UpdateNode(ctx, n.Data)
+	if err := n.Store.UpdateNode(ctx, n.Data); err != nil {
+		slog.Error("Failed to update node status to running", "node", n.Data.ID, "error", err)
+	}
 
 	// Subscribe to peer events
 	sub, _ := n.Bus.Subscribe("node.events", func(e core.Event) {
@@ -69,15 +74,31 @@ func (n *NodeActor) run(ctx context.Context) {
 	allowedTools := n.getToolsForRole()
 	model := n.Data.Role.Model
 
-	for i := 0; i < 10; i++ { // Max 10 iterations
+	maxIterations := n.Config.Limits.MaxIterations
-		// Progressive Summarization: If history > 20 messages, compress the middle part
+	if maxIterations <= 0 {
-		if len(messages) > 20 {
+		maxIterations = 10
+	}
+
+	summarizeThreshold := n.Config.Limits.SummarizeThreshold
+	if summarizeThreshold <= 0 {
+		summarizeThreshold = 20
+	}
+
+	for i := 0; i < maxIterations; i++ {
+		// Progressive Summarization: If history > threshold messages, compress the middle part
+		if len(messages) > summarizeThreshold {
 			slog.Info("Context threshold reached, performing progressive summarization", "node", n.Data.ID)
 
 			// 1. Identify context to summarize
 			// Header: System (0) + User Goal (1)
-			// Tail: Last 6 messages
+			// Tail: Last N messages
-			toSummarize := messages[2 : len(messages)-6]
+			keepCount := n.Config.Limits.PruningMsgKeep
+			if keepCount <= 0 {
+				keepCount = 6
+			}
+
+			if len(messages) > keepCount+2 {
+				toSummarize := messages[2 : len(messages)-keepCount]
 
 				summaryPrompt := "Briefly summarize the key findings, data points, and progress from the conversation above. Focus on facts discovered so far. Be very concise."
 				summaryMessages := append(toSummarize, core.Message{Role: "user", Content: summaryPrompt})
@@ -85,8 +106,8 @@ func (n *NodeActor) run(ctx context.Context) {
 				summaryResp, err := n.LLM.Chat(ctx, summaryMessages, nil, model)
 				if err == nil {
 					// 2. Reconstruct history: [Header] + [Summary] + [Tail]
-				tail := messages[len(messages)-6:]
+					tail := messages[len(messages)-keepCount:]
-				newHistory := make([]core.Message, 0, 10)
+					newHistory := make([]core.Message, 0, keepCount+3)
 					newHistory = append(newHistory, messages[0:2]...) // Keep System + Goal
 					newHistory = append(newHistory, core.Message{
 						Role:    "system",
@@ -100,6 +121,7 @@ func (n *NodeActor) run(ctx context.Context) {
 					})
 				}
 			}
+		}
 
 		// Inject Peer Insights
 		if len(n.peerInsights) > 0 {
@@ -177,7 +199,9 @@ func (n *NodeActor) getToolsForRole() []core.ToolDef {
 func (n *NodeActor) complete(ctx context.Context, out string) {
 	n.Data.Output = out
 	n.Data.Status = core.NodeStatusCompleted
-	n.Store.UpdateNode(ctx, n.Data)
+	if err := n.Store.UpdateNode(ctx, n.Data); err != nil {
+		slog.Error("Failed to update node status to completed", "node", n.Data.ID, "error", err)
+	}
 	n.Bus.Publish("node.events", core.Event{
 		Type: core.EventNodeCompleted, SwarmID: n.Data.SwarmID, NodeID: n.Data.ID,
 		Payload: map[string]any{"output": out},
@@ -186,7 +210,9 @@ func (n *NodeActor) complete(ctx context.Context, out string) {
 
 func (n *NodeActor) fail(ctx context.Context, err error) {
 	n.Data.Status = core.NodeStatusFailed
-	n.Store.UpdateNode(ctx, n.Data)
+	if errUpdate := n.Store.UpdateNode(ctx, n.Data); errUpdate != nil {
+		slog.Error("Failed to update node status to failed", "node", n.Data.ID, "error", errUpdate)
+	}
 	n.Bus.Publish("node.events", core.Event{
 		Type: core.EventNodeFailed, SwarmID: n.Data.SwarmID, NodeID: n.Data.ID,
 		Payload: map[string]any{"error": err.Error()},

pkg/swarm/runtime/orchestrator.go

@@ -41,14 +41,19 @@ func (o *Orchestrator) SetSharedMemory(m core.SharedMemory) { o.memory = m }
 
 func (o *Orchestrator) SpawnSwarm(ctx context.Context, goal string) (core.SwarmID, error) {
 	id := core.SwarmID(uuid.New().String())
-	o.store.CreateSwarm(ctx, &core.Swarm{ID: id, Goal: goal, Status: core.SwarmStatusActive, CreatedAt: time.Now()})
+	if err := o.store.CreateSwarm(ctx, &core.Swarm{ID: id, Goal: goal, Status: core.SwarmStatusActive, CreatedAt: time.Now()}); err != nil {
+		return "", fmt.Errorf("failed to create swarm in store: %w", err)
+	}
 
 	o.mu.Lock()
-	sCtx, cancel := context.WithCancel(context.Background())
+	sCtx, cancel := context.WithCancel(ctx)
 	o.activeSwarms[id] = cancel
 	o.mu.Unlock()
 
-	go o.RunSubTask(sCtx, id, "", "Manager", goal)
+	go func() {
+		defer o.StopSwarm(id)
+		o.RunSubTask(sCtx, id, "", "Manager", goal)
+	}()
 	return id, nil
 }
 
@@ -64,7 +69,9 @@ func (o *Orchestrator) RunSubTask(ctx context.Context, sid core.SwarmID, pid cor
 		Role: core.Role{Name: roleName, SystemPrompt: rc.SystemPrompt, Model: model},
 		Task: task, Status: core.NodeStatusPending,
 	}
-	o.store.CreateNode(ctx, node)
+	if err := o.store.CreateNode(ctx, node); err != nil {
+		return "", fmt.Errorf("failed to create node in store: %w", err)
+	}
 
 	nt := o.tools.Clone()
 	if roleName == "Manager" { nt.Register(NewDelegateTool(o, sid, node.ID)) }
@@ -77,19 +84,27 @@ func (o *Orchestrator) RunSubTask(ctx context.Context, sid core.SwarmID, pid cor
 	fail := make(chan error, 1)
 	sub, _ := o.bus.Subscribe("node.events", func(e core.Event) {
 		if e.NodeID == node.ID {
-			if e.Type == core.EventNodeCompleted { done <- e.Payload["output"].(string) }
+			if e.Type == core.EventNodeCompleted { 
+				output, _ := e.Payload["output"].(string)
+				done <- output 
+			}
 			if e.Type == core.EventNodeFailed { fail <- fmt.Errorf("%v", e.Payload["error"]) }
 		}
 	})
 	defer sub.Unsubscribe()
 
-	NewNodeActor(node, o.bus, o.store, o.llm, nt, o.policyChecker).Start(ctx)
+	NewNodeActor(node, o.bus, o.store, o.llm, nt, o.policyChecker, o.config).Start(ctx)
+
+	timeout := o.config.Limits.GlobalTimeout
+	if timeout <= 0 {
+		timeout = 15 * time.Minute
+	}
 
 	select {
 	case out := <-done: return out, nil
 	case err := <-fail: return "", err
 	case <-ctx.Done(): return "", ctx.Err()
-	case <-time.After(15 * time.Minute): return "", core.ErrTaskTimeout
+	case <-time.After(timeout): return "", core.ErrTaskTimeout
 	}
 }
 
@@ -101,3 +116,12 @@ func (o *Orchestrator) StopSwarm(id core.SwarmID) {
 		delete(o.activeSwarms, id)
 	}
 }
+
+func (o *Orchestrator) StopAll() {
+	o.mu.Lock()
+	defer o.mu.Unlock()
+	for id, cancel := range o.activeSwarms {
+		cancel()
+		delete(o.activeSwarms, id)
+	}
+}

pkg/swarm/service.go

@@ -41,7 +41,7 @@ func NewService(dbPath string, provider providers.LLMProvider, registry *tools.T
 }
 
 func (s *Service) listen() {
-	s.Bus.Subscribe("node.events", func(e core.Event) {
+	if _, err := s.Bus.Subscribe("node.events", func(e core.Event) {
 		msg := ""
 		switch e.Type {
 		case core.EventNodeThinking: msg = fmt.Sprintf("🤖 [%s]: %s", e.NodeID[:4], e.Payload["content"])
@@ -51,7 +51,9 @@ func (s *Service) listen() {
 		if msg != "" {
 			select { case s.Outbound <- msg: default: }
 		}
-	})
+	}); err != nil {
+		fmt.Printf("Warning: Swarm event subscription failed: %v\n", err)
+	}
 }
 
 func (s *Service) HandleCommand(ctx context.Context, input string) string {
@@ -61,10 +63,16 @@ func (s *Service) HandleCommand(ctx context.Context, input string) string {
 	switch args[1] {
 	case "spawn":
 		goal := strings.Join(args[2:], " ")
-		id, _ := s.Orchestrator.SpawnSwarm(ctx, goal)
+		id, err := s.Orchestrator.SpawnSwarm(ctx, goal)
+		if err != nil {
+			return fmt.Sprintf("❌ Failed to spawn swarm: %v", err)
+		}
 		return fmt.Sprintf("🚀 Swarm ID: %s", id)
 	case "list":
-		swarms, _ := s.Store.ListSwarms(ctx, core.SwarmStatusActive)
+		swarms, err := s.Store.ListSwarms(ctx, core.SwarmStatusActive)
+		if err != nil {
+			return fmt.Sprintf("❌ Failed to list swarms: %v", err)
+		}
 		out := "Active Swarms:\n"
 		for _, sw := range swarms { out += fmt.Sprintf("- %s: %s\n", sw.ID, sw.Goal) }
 		return out
@@ -75,3 +83,9 @@ func (s *Service) HandleCommand(ctx context.Context, input string) string {
 	}
 	return "Unknown command"
 }
+
+func (s *Service) Stop() {
+	if s.Orchestrator != nil {
+		s.Orchestrator.StopAll()
+	}
+}