Merge branch 'upstream-main' into feat/subturn-poc

pkg/agent/loop.go

@@ -76,7 +76,8 @@ type processOptions struct {
 }
 
 const (
-	defaultResponse           = "I've completed processing but have no response to give. Increase `max_tool_iterations` in config.json."
+	defaultResponse           = "The model returned an empty response. This may indicate a provider error or token limit."
+	toolLimitResponse         = "I've reached `max_tool_iterations` without a final response. Increase `max_tool_iterations` in config.json if this task needs more tool steps."
 	sessionKeyAgentPrefix     = "agent:"
 	metadataKeyAccountID      = "account_id"
 	metadataKeyGuildID        = "guild_id"
@@ -1130,7 +1131,11 @@ func (al *AgentLoop) runAgentLoop(
 
 	// 4. Handle empty response
 	if finalContent == "" {
-		finalContent = opts.DefaultResponse
+		if iteration >= agent.MaxIterations && agent.MaxIterations > 0 {
+			finalContent = toolLimitResponse
+		} else {
+			finalContent = opts.DefaultResponse
+		}
 	}
 
 	// 5. Save final assistant message to session
@@ -1221,6 +1226,7 @@ func (al *AgentLoop) handleReasoning(
 }
 
 // runLLMIteration executes the LLM call loop with tool handling.
+// Returns (finalContent, iteration, error).
 func (al *AgentLoop) runLLMIteration(
 	ctx context.Context,
 	agent *AgentInstance,
@@ -1248,6 +1254,13 @@ func (al *AgentLoop) runLLMIteration(
 		}
 	}
 
+	// Check if both the provider and channel support streaming
+	streamProvider, providerCanStream := agent.Provider.(providers.StreamingProvider)
+	var streamer bus.Streamer
+	if providerCanStream && !opts.NoHistory && !constants.IsInternalChannel(opts.Channel) {
+		streamer, _ = al.bus.GetStreamer(ctx, opts.Channel, opts.ChatID)
+	}
+
 	// Determine effective model tier for this conversation turn.
 	// selectCandidates evaluates routing once and the decision is sticky for
 	// all tool-follow-up iterations within the same turn so that a multi-step
@@ -1364,6 +1377,16 @@ func (al *AgentLoop) runLLMIteration(
 			al.activeRequests.Add(1)
 			defer al.activeRequests.Done()
 
+			// Use streaming when available (streamer obtained, provider supports it)
+			if streamer != nil && streamProvider != nil {
+				return streamProvider.ChatStream(
+					ctx, messages, providerToolDefs, activeModel, llmOpts,
+					func(accumulated string) {
+						streamer.Update(ctx, accumulated)
+					},
+				)
+			}
+
 			if len(activeCandidates) > 1 && al.fallback != nil {
 				fbResult, fbErr := al.fallback.Execute(
 					ctx,
@@ -1500,15 +1523,31 @@ func (al *AgentLoop) runLLMIteration(
 			if finalContent == "" && response.ReasoningContent != "" {
 				finalContent = response.ReasoningContent
 			}
+
+			// If we were streaming, finalize the message (sends the permanent message)
+			if streamer != nil {
+				if err := streamer.Finalize(ctx, finalContent); err != nil {
+					logger.WarnCF("agent", "Stream finalize failed", map[string]any{
+						"error": err.Error(),
+					})
+				}
+			}
+
 			logger.InfoCF("agent", "LLM response without tool calls (direct answer)",
 				map[string]any{
 					"agent_id":      agent.ID,
 					"iteration":     iteration,
 					"content_chars": len(finalContent),
+					"streamed":      streamer != nil,
 				})
 			break
 		}
 
+		// Tool calls detected — cancel any active stream (draft auto-expires)
+		if streamer != nil {
+			streamer.Cancel(ctx)
+		}
+
 		normalizedToolCalls := make([]providers.ToolCall, 0, len(response.ToolCalls))
 		for _, tc := range response.ToolCalls {
 			normalizedToolCalls = append(normalizedToolCalls, providers.NormalizeToolCall(tc))

pkg/agent/loop_test.go

@@ -420,6 +420,29 @@ func (m *countingMockProvider) GetDefaultModel() string {
 	return "counting-mock-model"
 }
 
+type toolLimitOnlyProvider struct{}
+
+func (m *toolLimitOnlyProvider) Chat(
+	ctx context.Context,
+	messages []providers.Message,
+	tools []providers.ToolDefinition,
+	model string,
+	opts map[string]any,
+) (*providers.LLMResponse, error) {
+	return &providers.LLMResponse{
+		ToolCalls: []providers.ToolCall{{
+			ID:        "call_tool_limit_test",
+			Type:      "function",
+			Name:      "tool_limit_test_tool",
+			Arguments: map[string]any{"value": "x"},
+		}},
+	}, nil
+}
+
+func (m *toolLimitOnlyProvider) GetDefaultModel() string {
+	return "tool-limit-only-model"
+}
+
 // mockCustomTool is a simple mock tool for registration testing
 type mockCustomTool struct{}
 
@@ -442,6 +465,29 @@ func (m *mockCustomTool) Execute(ctx context.Context, args map[string]any) *tool
 	return tools.SilentResult("Custom tool executed")
 }
 
+type toolLimitTestTool struct{}
+
+func (m *toolLimitTestTool) Name() string {
+	return "tool_limit_test_tool"
+}
+
+func (m *toolLimitTestTool) Description() string {
+	return "Tool used to exhaust the iteration budget in tests"
+}
+
+func (m *toolLimitTestTool) Parameters() map[string]any {
+	return map[string]any{
+		"type": "object",
+		"properties": map[string]any{
+			"value": map[string]any{"type": "string"},
+		},
+	}
+}
+
+func (m *toolLimitTestTool) Execute(ctx context.Context, args map[string]any) *tools.ToolResult {
+	return tools.SilentResult("tool limit test result")
+}
+
 // testHelper executes a message and returns the response
 type testHelper struct {
 	al *AgentLoop
@@ -1083,6 +1129,89 @@ func TestAgentLoop_ContextExhaustionRetry(t *testing.T) {
 	}
 }
 
+func TestAgentLoop_EmptyModelResponseUsesAccurateFallback(t *testing.T) {
+	tmpDir, err := os.MkdirTemp("", "agent-test-*")
+	if err != nil {
+		t.Fatalf("Failed to create temp dir: %v", err)
+	}
+	defer os.RemoveAll(tmpDir)
+
+	cfg := &config.Config{
+		Agents: config.AgentsConfig{
+			Defaults: config.AgentDefaults{
+				Workspace:         tmpDir,
+				Model:             "test-model",
+				MaxTokens:         4096,
+				MaxToolIterations: 3,
+			},
+		},
+	}
+
+	msgBus := bus.NewMessageBus()
+	provider := &simpleMockProvider{response: ""}
+	al := NewAgentLoop(cfg, msgBus, provider)
+
+	response, err := al.ProcessDirectWithChannel(context.Background(), "hello", "empty-response", "test", "chat1")
+	if err != nil {
+		t.Fatalf("ProcessDirectWithChannel failed: %v", err)
+	}
+	if response != defaultResponse {
+		t.Fatalf("response = %q, want %q", response, defaultResponse)
+	}
+}
+
+func TestAgentLoop_ToolLimitUsesDedicatedFallback(t *testing.T) {
+	tmpDir, err := os.MkdirTemp("", "agent-test-*")
+	if err != nil {
+		t.Fatalf("Failed to create temp dir: %v", err)
+	}
+	defer os.RemoveAll(tmpDir)
+
+	cfg := &config.Config{
+		Agents: config.AgentsConfig{
+			Defaults: config.AgentDefaults{
+				Workspace:         tmpDir,
+				Model:             "test-model",
+				MaxTokens:         4096,
+				MaxToolIterations: 1,
+			},
+		},
+	}
+
+	msgBus := bus.NewMessageBus()
+	provider := &toolLimitOnlyProvider{}
+	al := NewAgentLoop(cfg, msgBus, provider)
+	al.RegisterTool(&toolLimitTestTool{})
+
+	response, err := al.ProcessDirectWithChannel(context.Background(), "hello", "tool-limit", "test", "chat1")
+	if err != nil {
+		t.Fatalf("ProcessDirectWithChannel failed: %v", err)
+	}
+	if response != toolLimitResponse {
+		t.Fatalf("response = %q, want %q", response, toolLimitResponse)
+	}
+
+	defaultAgent := al.registry.GetDefaultAgent()
+	if defaultAgent == nil {
+		t.Fatal("No default agent found")
+	}
+	route := al.registry.ResolveRoute(routing.RouteInput{
+		Channel: "test",
+		Peer: &routing.RoutePeer{
+			Kind: "direct",
+			ID:   "cron",
+		},
+	})
+	history := defaultAgent.Sessions.GetHistory(route.SessionKey)
+	if len(history) != 4 {
+		t.Fatalf("history len = %d, want 4", len(history))
+	}
+	assertRoles(t, history, "user", "assistant", "tool", "assistant")
+	if history[3].Content != toolLimitResponse {
+		t.Fatalf("final assistant content = %q, want %q", history[3].Content, toolLimitResponse)
+	}
+}
+
 // TestProcessDirectWithChannel_TriggersMCPInitialization verifies that
 // ProcessDirectWithChannel triggers MCP initialization when MCP is enabled.
 // Note: Manager is only initialized when at least one MCP server is configured