Agent-System/internal/intent/hard_negative_test.go

package intent_test

import (
	"context"
	"encoding/json"
	"os"
	"path/filepath"
	"runtime"
	"strings"
	"testing"
	"time"

	"chat-agent-system/internal/intent"
	"chat-agent-system/internal/maturity"
)

type goldenFile struct {
	Version string       `json:"version"`
	Cases   []goldenCase `json:"cases"`
}

type goldenCase struct {
	ID       string          `json:"id"`
	Input    string          `json:"input"`
	Snapshot *goldenSnapshot `json:"snapshot"`
	Pending  *goldenPending  `json:"pending"`
	ExpectP0 *goldenExpect   `json:"expect_phase0"`
	ExpectP1 *goldenExpect   `json:"expect_phase1"`
}

type goldenSnapshot struct {
	UserDocRole       string `json:"user_doc_role"`
	DocLifecycleState string `json:"doc_lifecycle_state"`
	ActiveDocID       string `json:"active_doc_id"`
	CurrentStage      string `json:"current_stage"`
}

type goldenPending struct {
	ExpectedIntent   string              `json:"expected_intent"`
	AllowAffirmative bool                `json:"allow_affirmative"`
	DefaultOption    *goldenExpectOption `json:"default_option"`
}

type goldenExpectOption struct {
	OptionID string            `json:"option_id"`
	Title    string            `json:"title"`
	Intent   string            `json:"intent"`
	Slots    map[string]string `json:"slots"`
}

type goldenExpect struct {
	LegacyIntent          string   `json:"legacy_intent"`
	Branch                string   `json:"branch"`
	RequiresClarifying    *bool    `json:"requires_clarifying"`
	MustNotWrite          bool     `json:"must_not_write"`
	DialogueAct           string   `json:"dialogue_act"`
	ForbiddenIntents      []string `json:"forbidden_intents"`
	MustNotIntent         string   `json:"must_not_intent"`
	MustNotBranch         string   `json:"must_not_branch"`
	PreferredIntent       string   `json:"preferred_intent"`
	ClarificationContains string   `json:"clarification_contains"`
	ReasoningContains     string   `json:"reasoning_contains"`
}

func loadGoldenV1(t *testing.T) goldenFile {
	t.Helper()
	_, file, _, ok := runtime.Caller(0)
	if !ok {
		t.Fatal("runtime.Caller failed")
	}
	path := filepath.Join(filepath.Dir(file), "testdata", "intent-golden-v1.json")
	raw, err := os.ReadFile(path)
	if err != nil {
		t.Fatalf("read golden file: %v", err)
	}
	var gf goldenFile
	if err := json.Unmarshal(raw, &gf); err != nil {
		t.Fatalf("parse golden file: %v", err)
	}
	if gf.Version != "intent-golden-v1" {
		t.Fatalf("unexpected golden version: %s", gf.Version)
	}
	return gf
}

func buildContextFromGolden(t *testing.T, snap *goldenSnapshot, pending *goldenPending) *intent.ContextSnapshot {
	t.Helper()
	ctx := &intent.ContextSnapshot{}
	if snap != nil {
		ctx.UserDocRole = snap.UserDocRole
		ctx.DocLifecycleState = snap.DocLifecycleState
		ctx.ActiveDocID = snap.ActiveDocID
		ctx.CurrentStage = snap.CurrentStage
	}
	if pending != nil {
		pe := &intent.PendingExpectation{
			ExpectedIntent:   intent.IntentType(pending.ExpectedIntent),
			AllowAffirmative: pending.AllowAffirmative,
			CreatedAt:        time.Now(),
			ExpiresAt:        time.Now().Add(time.Hour),
		}
		if pending.DefaultOption != nil {
			pe.DefaultOption = &intent.ExpectationOption{
				OptionID: pending.DefaultOption.OptionID,
				Title:    pending.DefaultOption.Title,
				Intent:   intent.IntentType(pending.DefaultOption.Intent),
				Slots:    pending.DefaultOption.Slots,
			}
		}
		ctx.PendingExpectation = pe
	}
	return ctx
}

func assertGoldenExpect(t *testing.T, caseID string, exp *goldenExpect, res *intent.IntentRecognitionResult) {
	t.Helper()
	if exp == nil {
		return
	}
	if exp.LegacyIntent != "" && string(res.TopIntent.Intent) != exp.LegacyIntent {
		t.Errorf("[%s] intent: want %s got %s", caseID, exp.LegacyIntent, res.TopIntent.Intent)
	}
	if exp.Branch != "" && string(res.Branch) != exp.Branch {
		t.Errorf("[%s] branch: want %s got %s", caseID, exp.Branch, res.Branch)
	}
	if exp.RequiresClarifying != nil && res.RequiresClarifying != *exp.RequiresClarifying {
		t.Errorf("[%s] requires_clarifying: want %v got %v", caseID, *exp.RequiresClarifying, res.RequiresClarifying)
	}
	if exp.MustNotIntent != "" && string(res.TopIntent.Intent) == exp.MustNotIntent {
		t.Errorf("[%s] must not intent %s", caseID, exp.MustNotIntent)
	}
	if exp.MustNotBranch != "" && string(res.Branch) == exp.MustNotBranch {
		t.Errorf("[%s] must not branch %s", caseID, exp.MustNotBranch)
	}
	if exp.PreferredIntent != "" && string(res.TopIntent.Intent) != exp.PreferredIntent {
		t.Errorf("[%s] preferred intent: want %s got %s", caseID, exp.PreferredIntent, res.TopIntent.Intent)
	}
	if exp.ClarificationContains != "" && !strings.Contains(res.TopIntent.ClarificationQuestion, exp.ClarificationContains) {
		t.Errorf("[%s] clarification should contain %q, got %q", caseID, exp.ClarificationContains, res.TopIntent.ClarificationQuestion)
	}
	if exp.ReasoningContains != "" && !strings.Contains(res.TopIntent.Reasoning, exp.ReasoningContains) {
		t.Errorf("[%s] reasoning should contain %q, got %q", caseID, exp.ReasoningContains, res.TopIntent.Reasoning)
	}
	for _, forbidden := range exp.ForbiddenIntents {
		if string(res.TopIntent.Intent) == forbidden {
			t.Errorf("[%s] forbidden intent %s", caseID, forbidden)
		}
	}
	if exp.MustNotWrite {
		writeIntents := map[intent.IntentType]bool{
			intent.IntentDocGovClaim:          true,
			intent.IntentDocGovSubmitReview:   true,
			intent.IntentDocGovReviewApprove:  true,
			intent.IntentDocGovReviewReject:   true,
			intent.IntentDocGovFinalApprove:   true,
			intent.IntentDocGovFinalReject:    true,
			intent.IntentDocGovCommentAdd:     true,
			intent.IntentDocGovCommentResolve: true,
			intent.IntentDocGovReopenRequest:  true,
			intent.IntentDocGovReopenApprove:  true,
		}
		if writeIntents[res.TopIntent.Intent] {
			t.Errorf("[%s] must not route to write intent %s", caseID, res.TopIntent.Intent)
		}
	}
	if exp.DialogueAct != "" && res.ShadowDialogueAct != exp.DialogueAct {
		t.Errorf("[%s] dialogue_act: want %s got %s", caseID, exp.DialogueAct, res.ShadowDialogueAct)
	}
}

// TestIntentGoldenV1_Phase1 运行 golden-v1 的 Phase 1 安全门禁期望
func TestIntentGoldenV1_Phase1(t *testing.T) {
	gf := loadGoldenV1(t)
	rec := intent.NewRecognizer()
	for _, c := range gf.Cases {
		t.Run(c.ID, func(t *testing.T) {
			ctx := buildContextFromGolden(t, c.Snapshot, c.Pending)
			res, err := rec.Recognize(context.Background(), c.Input, ctx)
			if err != nil {
				t.Fatalf("recognize failed: %v", err)
			}
			assertGoldenExpect(t, c.ID, c.ExpectP1, res)
		})
	}
}

// TestIntentGoldenV1_V2DualWrite 每条识别结果应双写 intent_frame_v2
func TestIntentGoldenV1_V2DualWrite(t *testing.T) {
	gf := loadGoldenV1(t)
	rec := intent.NewRecognizer()
	for _, c := range gf.Cases {
		t.Run(c.ID, func(t *testing.T) {
			ctx := buildContextFromGolden(t, c.Snapshot, c.Pending)
			res, err := rec.Recognize(context.Background(), c.Input, ctx)
			if err != nil {
				t.Fatalf("recognize failed: %v", err)
			}
			if res.IntentFrameV2 == nil {
				t.Fatal("expected intent_frame_v2")
			}
			if res.IntentFrameV2.LegacyIntent != res.TopIntent.Intent {
				t.Errorf("v2 legacy mismatch: %s vs %s", res.IntentFrameV2.LegacyIntent, res.TopIntent.Intent)
			}
		})
	}
}

// TestIntentGoldenV1_ShadowFields shadow 观测字段应被填充
func TestIntentGoldenV1_ShadowFields(t *testing.T) {
	rec := intent.NewRecognizer()
	ctx := &intent.ContextSnapshot{CurrentStage: string(maturity.StageV01_Triage)}
	res, err := rec.Recognize(context.Background(), "现在这个便携超声立项评估到哪一步了？", ctx)
	if err != nil {
		t.Fatalf("recognize failed: %v", err)
	}
	if res.ShadowScore <= 0 {
		t.Errorf("expected shadow_score > 0, got %v", res.ShadowScore)
	}
	if res.ShadowDialogueAct == "" {
		t.Errorf("expected shadow_dialogue_act to be set")
	}
	if res.ShadowDecisionHint == "" {
		t.Errorf("expected shadow_decision_hint to be set")
	}
}