fix: harden rag answerability gate

internal/ai/runtime/executor/answerability_gate.go

@@ -11,6 +11,7 @@ import (
 	"cs-agent/internal/ai/runtime/internal/impl/factory"
 	"cs-agent/internal/ai/runtime/internal/impl/retrievers"
 	"cs-agent/internal/models"
+	"cs-agent/internal/pkg/utils"
 
 	"github.com/cloudwego/eino/components/model"
 	"github.com/cloudwego/eino/components/prompt"
@@ -178,10 +179,17 @@ func (g *KnowledgeAnswerabilityGate) retrieveKnowledge(ctx context.Context, stat
 	}
 	gate := g.withDefaults()
 	req := state.Input.Request
+	configuredKnowledgeIDs := utils.SplitInt64s(req.AIAgent.KnowledgeIDs)
 	retriever := gate.newRetriever(req.AIAgent)
 	if retriever == nil {
-		state.SkipGate = true
-		state.recordAnswerability(answerabilityStatusSkipped, "knowledge retriever unavailable", nil)
+		state.KnowledgeIDs = append([]int64(nil), configuredKnowledgeIDs...)
+		if len(configuredKnowledgeIDs) == 0 {
+			state.SkipGate = true
+			state.recordAnswerability(answerabilityStatusSkipped, "no knowledge configured", nil)
+			return state, nil
+		}
+		state.FallbackReply = resolveKnowledgeHumanSupportFallback(req.AIAgent)
+		state.recordAnswerability(answerabilityStatusUnanswerable, "knowledge retriever unavailable", nil)
 		return state, nil
 	}
 	knowledgeIDs := retriever.KnowledgeBaseIDs()
@@ -212,7 +220,7 @@ func (g *KnowledgeAnswerabilityGate) retrieveKnowledge(ctx context.Context, stat
 	}
 	if state.Input.Collector != nil && result != nil {
 		state.Input.Collector.SetRetrieverSummary(result.TraceSummary)
-		state.Input.Collector.Data.Retriever.Items = append(state.Input.Collector.Data.Retriever.Items, result.TraceItems...)
+		state.Input.Collector.AddRetrieverItems(result.TraceItems)
 	}
 	if result == nil || len(result.Hits) == 0 || strings.TrimSpace(result.ContextText) == "" {
 		state.FallbackReply = resolveKnowledgeHumanSupportFallback(req.AIAgent)
@@ -230,7 +238,7 @@ func (g *KnowledgeAnswerabilityGate) gradeAnswerability(ctx context.Context, sta
 		return state, nil
 	}
 	gate := g.withDefaults()
-	started := time.Now()
+	started := gate.now()
 	req := state.Input.Request
 	modelInstance, err := gate.newChatModel(ctx, req.AIConfig)
 	if err != nil {

internal/ai/runtime/executor/answerability_gate_test.go

@@ -95,6 +95,32 @@ func TestKnowledgeAnswerabilityGateEvaluateSkipsWhenNoKnowledgeConfigured(t *tes
 	}
 }
 
+func TestKnowledgeAnswerabilityGateEvaluateFallsBackWhenConfiguredRetrieverUnavailable(t *testing.T) {
+	collector := callbacks.NewRuntimeTraceCollector()
+	gate := newTestKnowledgeAnswerabilityGate(nil, nil)
+
+	state, err := gate.Evaluate(context.Background(), answerabilityGateInput{
+		Request:   newAnswerabilityGateRunInput("是否支持退款？", "1"),
+		Collector: collector,
+	})
+	if err != nil {
+		t.Fatalf("Evaluate returned error: %v", err)
+	}
+
+	if state.SkipGate {
+		t.Fatal("expected configured knowledge to fail closed, not skip")
+	}
+	if !strings.Contains(state.FallbackReply, "建议你联系人工客服进一步确认。") {
+		t.Fatalf("expected human-support fallback, got %q", state.FallbackReply)
+	}
+	if collector.Data.Answerability.Status != answerabilityStatusUnanswerable {
+		t.Fatalf("unexpected answerability status: %q", collector.Data.Answerability.Status)
+	}
+	if collector.Data.Answerability.Reason != "knowledge retriever unavailable" {
+		t.Fatalf("unexpected reason: %q", collector.Data.Answerability.Reason)
+	}
+}
+
 func TestKnowledgeAnswerabilityGateEvaluateFallsBackOnGrayZoneUnanswerableDecision(t *testing.T) {
 	collector := callbacks.NewRuntimeTraceCollector()
 	gate := newTestKnowledgeAnswerabilityGate(newAnswerabilityRetrieverWithHit(), &fakeAnswerabilityChatModel{

internal/ai/runtime/executor/context_builders.go

@@ -51,7 +51,7 @@ func appendRetrievedContext(ctx context.Context, req RunInput, summary *RunResul
 	}
 	if collector != nil {
 		collector.SetRetrieverSummary(retrieveResult.TraceSummary)
-		collector.Data.Retriever.Items = append(collector.Data.Retriever.Items, retrieveResult.TraceItems...)
+		collector.AddRetrieverItems(retrieveResult.TraceItems)
 	}
 	decision := buildKnowledgeGuardDecision(req.AIAgent, retrieveResult)
 	if len(decision.Instructions) > 0 {

internal/ai/runtime/internal/impl/callbacks/runlog_callback.go

@@ -106,6 +106,15 @@ func (c *RuntimeTraceCollector) SetRetrieverSummary(summary RetrieverTraceSummar
 	c.Data.Retriever.Policies = append([]RetrieverPolicyTraceItem(nil), summary.Policies...)
 }
 
+func (c *RuntimeTraceCollector) AddRetrieverItems(items []RetrieverTraceItem) {
+	if len(items) == 0 {
+		return
+	}
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	c.Data.Retriever.Items = append(c.Data.Retriever.Items, items...)
+}
+
 func (c *RuntimeTraceCollector) SetAnswerability(data AnswerabilityTraceData) {
 	c.mu.Lock()
 	defer c.mu.Unlock()

internal/ai/runtime/internal/impl/callbacks/runlog_callback_test.go

@@ -0,0 +1,20 @@
+package callbacks
+
+import "testing"
+
+func TestRuntimeTraceCollectorAddRetrieverItems(t *testing.T) {
+	collector := NewRuntimeTraceCollector()
+
+	collector.AddRetrieverItems([]RetrieverTraceItem{
+		{KnowledgeBaseID: 1, DocumentID: 10, DocumentTitle: "doc-1"},
+		{KnowledgeBaseID: 2, DocumentID: 20, DocumentTitle: "doc-2"},
+	})
+	collector.AddRetrieverItems(nil)
+
+	if got := len(collector.Data.Retriever.Items); got != 2 {
+		t.Fatalf("expected two retriever items, got %d", got)
+	}
+	if collector.Data.Retriever.Items[0].DocumentTitle != "doc-1" || collector.Data.Retriever.Items[1].DocumentTitle != "doc-2" {
+		t.Fatalf("unexpected retriever items: %#v", collector.Data.Retriever.Items)
+	}
+}