feat(kol): enrich assist generate with URL references and template prompt

- Add kolAssistURLResolver that fetches reference articles via the Ark
  URL extractor when the user's description contains http(s) links, and
  merges up to kolAssistReferenceContentMaxRunes of extracted content
  into the generate description.
- Rewrite the generate user prompt into a strict template-authoring
  brief: output only the final prompt template in Chinese, parameterize
  variable info as {{placeholder}}, require at least 3 placeholders
  when references exist, and cover goal/topic/title/structure/style/
  interaction/constraints.
- Wire LLMConfig and logger through NewKolAssistService and the
  handler.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

server/internal/tenant/app/kol_assist_generate_context.go

@@ -0,0 +1,190 @@
+package app
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"regexp"
+	"strings"
+
+	"github.com/geo-platform/tenant-api/internal/shared/config"
+	"go.uber.org/zap"
+)
+
+const kolAssistReferenceContentMaxRunes = 6000
+
+var kolAssistURLPattern = regexp.MustCompile(`https?://[^\s]+`)
+
+type kolAssistURLResolver struct {
+	arkBaseURL string
+	arkAPIKey  string
+	httpClient *http.Client
+	logger     *zap.Logger
+}
+
+type kolAssistReferenceMaterial struct {
+	URL     string
+	Title   string
+	Content string
+}
+
+func newKolAssistURLResolver(llmCfg config.LLMConfig, logger *zap.Logger) *kolAssistURLResolver {
+	baseURL := strings.TrimSpace(llmCfg.BaseURL)
+	if baseURL == "" {
+		baseURL = defaultKnowledgeArkBaseURL
+	}
+
+	return &kolAssistURLResolver{
+		arkBaseURL: strings.TrimRight(baseURL, "/"),
+		arkAPIKey:  strings.TrimSpace(llmCfg.APIKey),
+		httpClient: &http.Client{Timeout: defaultKnowledgeURLParseTimeout},
+		logger:     logger,
+	}
+}
+
+func (r *kolAssistURLResolver) extract(ctx context.Context, rawURL string) (*kolAssistReferenceMaterial, error) {
+	if r == nil || r.httpClient == nil || strings.TrimSpace(r.arkAPIKey) == "" {
+		return nil, fmt.Errorf("url resolver is unavailable")
+	}
+
+	body, err := json.Marshal(arkToolExecuteRequest{
+		ActionName: "LinkReader",
+		ToolName:   "LinkReader",
+		Timeout:    int(defaultKnowledgeURLParseTimeout.Seconds()),
+		Parameters: map[string]any{
+			"url_list": []string{rawURL},
+		},
+	})
+	if err != nil {
+		return nil, fmt.Errorf("marshal webpage parser request: %w", err)
+	}
+
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, r.arkBaseURL+"/tools/execute", bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("create webpage parser request: %w", err)
+	}
+	req.Header.Set("Authorization", "Bearer "+r.arkAPIKey)
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := r.httpClient.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("execute webpage parser: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode >= http.StatusBadRequest {
+		data, _ := io.ReadAll(io.LimitReader(resp.Body, 4096))
+		return nil, fmt.Errorf("webpage parser failed: status=%d body=%s", resp.StatusCode, strings.TrimSpace(string(data)))
+	}
+
+	var parsed arkToolExecuteResponse
+	if err := json.NewDecoder(resp.Body).Decode(&parsed); err != nil {
+		return nil, fmt.Errorf("decode webpage parser response: %w", err)
+	}
+	if parsed.StatusCode != 0 && parsed.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("webpage parser returned status_code=%d", parsed.StatusCode)
+	}
+	if len(parsed.Data.ArkWebDataList) == 0 {
+		return nil, fmt.Errorf("webpage parser returned no content")
+	}
+
+	item := parsed.Data.ArkWebDataList[0]
+	if strings.TrimSpace(item.ErrorCode) != "" || strings.TrimSpace(item.ErrorMessage) != "" {
+		return nil, fmt.Errorf("webpage parser returned error: %s %s", strings.TrimSpace(item.ErrorCode), strings.TrimSpace(item.ErrorMessage))
+	}
+
+	content := truncateKolAssistReferenceContent(normalizeKnowledgeText(item.Content))
+	if content == "" {
+		return nil, fmt.Errorf("webpage parser returned empty content")
+	}
+
+	return &kolAssistReferenceMaterial{
+		URL:     strings.TrimSpace(rawURL),
+		Title:   strings.TrimSpace(item.Title),
+		Content: content,
+	}, nil
+}
+
+func (s *KolAssistService) enrichGenerateDescription(ctx context.Context, description string) string {
+	description = strings.TrimSpace(description)
+	urls := extractKolAssistURLs(description)
+	if len(urls) == 0 || s == nil || s.urlResolver == nil {
+		return description
+	}
+
+	references := make([]kolAssistReferenceMaterial, 0, len(urls))
+	for _, rawURL := range urls {
+		material, err := s.urlResolver.extract(ctx, rawURL)
+		if err != nil {
+			if s.logger != nil {
+				s.logger.Warn("kol assist url extraction failed",
+					zap.String("url", rawURL),
+					zap.Error(err),
+				)
+			}
+			continue
+		}
+		references = append(references, *material)
+	}
+
+	if len(references) == 0 {
+		return description
+	}
+
+	var builder strings.Builder
+	builder.WriteString(description)
+	builder.WriteString("\n\n以下是系统从参考网页提取到的正文内容。你必须基于这些内容学习文章的方法、结构、标题策略和风格，但不要复述具体事实，不要直接照抄原文。\n")
+	for index, item := range references {
+		builder.WriteString(fmt.Sprintf("\n[参考网页 %d]\nURL: %s\n", index+1, item.URL))
+		if item.Title != "" {
+			builder.WriteString("标题: ")
+			builder.WriteString(item.Title)
+			builder.WriteString("\n")
+		}
+		builder.WriteString("正文摘录:\n")
+		builder.WriteString(item.Content)
+		builder.WriteString("\n")
+	}
+
+	return builder.String()
+}
+
+func extractKolAssistURLs(input string) []string {
+	matches := kolAssistURLPattern.FindAllString(input, -1)
+	if len(matches) == 0 {
+		return nil
+	}
+
+	result := make([]string, 0, len(matches))
+	seen := make(map[string]struct{}, len(matches))
+	for _, match := range matches {
+		candidate := strings.TrimSpace(match)
+		candidate = strings.Trim(candidate, `"'<>[](){}`)
+		candidate = strings.TrimRight(candidate, ".,;，。；、)]}》」”")
+		if candidate == "" {
+			continue
+		}
+		if _, exists := seen[candidate]; exists {
+			continue
+		}
+		seen[candidate] = struct{}{}
+		result = append(result, candidate)
+	}
+	return result
+}
+
+func truncateKolAssistReferenceContent(content string) string {
+	content = strings.TrimSpace(content)
+	if content == "" {
+		return ""
+	}
+
+	runes := []rune(content)
+	if len(runes) <= kolAssistReferenceContentMaxRunes {
+		return content
+	}
+	return strings.TrimSpace(string(runes[:kolAssistReferenceContentMaxRunes]))
+}

server/internal/tenant/app/kol_assist_runtime.go

@@ -43,6 +43,9 @@ func (s *KolAssistService) Run(
 	if err != nil {
 		return nil, "", err
 	}
+	if req.Mode == kolAssistModeGenerate {
+		req.Description = s.enrichGenerateDescription(ctx, req.Description)
+	}
 
 	generateReq, err := BuildKolAssistGenerateRequest(req)
 	if err != nil {
@@ -88,7 +91,21 @@ func BuildKolAssistGenerateRequest(req AssistRequest) (llm.GenerateRequest, erro
 func BuildKolAssistUserPrompt(req AssistRequest) (string, error) {
 	switch strings.ToLower(strings.TrimSpace(req.Mode)) {
 	case kolAssistModeGenerate:
-		return req.Description, nil
+		return strings.TrimSpace(fmt.Sprintf(`
+你的任务是把“用户需求 + 参考材料”转换成“最终可直接给另一个 AI 使用的提示词模板”。
+
+严格要求：
+1. 只输出最终提示词模板，不要输出分析、总结、复盘、学习笔记、解释说明。
+2. 如果提供了参考文章或参考网页，只学习其选题切入、标题策略、结构框架、叙事节奏、风格语气、互动与转化方式，然后抽象成可复用规则。
+3. 任何会变化的信息都必须参数化成 {{变量名}} 占位符，不要把具体的人名、品牌、平台、标题、数字、时间、地区、产品、案例、链接、经历写死。
+4. 若用户基于具体文章/网页生成模板，结果中至少包含 3 个 {{}} 占位符。
+5. 结果必须是“生成同类内容”的模板，不是对参考文章的分析。
+6. 模板至少覆盖：创作目标、选题要求、标题要求、结构要求、风格要求、互动/转化要求、限制项。
+7. 输出中文，直接给出最终模板正文。
+
+用户需求与参考材料：
+%s
+`, req.Description)), nil
 	case kolAssistModeOptimize:
 		schemaJSON := "null"
 		if req.Schema != nil {

server/internal/tenant/app/kol_assist_service.go

@@ -8,10 +8,12 @@ import (
 	"github.com/google/uuid"
 
 	"github.com/geo-platform/tenant-api/internal/shared/auth"
+	"github.com/geo-platform/tenant-api/internal/shared/config"
 	"github.com/geo-platform/tenant-api/internal/shared/llm"
 	"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
 	"github.com/geo-platform/tenant-api/internal/shared/response"
 	"github.com/geo-platform/tenant-api/internal/tenant/repository"
+	"go.uber.org/zap"
 )
 
 const (
@@ -46,10 +48,12 @@ type KolAssistTaskResponse struct {
 }
 
 type KolAssistService struct {
-	profileSvc *KolProfileService
+	profileSvc  *KolProfileService
-	repo       repository.KolAssistRepository
+	repo        repository.KolAssistRepository
-	llm        llm.Client
+	llm         llm.Client
-	messaging  *rabbitmq.Client
+	messaging   *rabbitmq.Client
+	logger      *zap.Logger
+	urlResolver *kolAssistURLResolver
 }
 
 func NewKolAssistService(
@@ -57,12 +61,16 @@ func NewKolAssistService(
 	repo repository.KolAssistRepository,
 	llmClient llm.Client,
 	messagingClient *rabbitmq.Client,
+	llmCfg config.LLMConfig,
+	logger *zap.Logger,
 ) *KolAssistService {
 	return &KolAssistService{
-		profileSvc: profileSvc,
+		profileSvc:  profileSvc,
-		repo:       repo,
+		repo:        repo,
-		llm:        llmClient,
+		llm:         llmClient,
-		messaging:  messagingClient,
+		messaging:   messagingClient,
+		logger:      logger,
+		urlResolver: newKolAssistURLResolver(llmCfg, logger),
 	}
 }
 

server/internal/tenant/transport/kol_manage_handler.go

@@ -42,7 +42,7 @@ func NewKolManageHandler(a *bootstrap.App, assets *AssetHandler) *KolManageHandl
 		profileSvc: profileSvc,
 		pkgSvc:     app.NewKolPackageService(profileSvc, a.KolPackages, a.KolPrompts).WithCache(a.Cache),
 		promptSvc:  app.NewKolPromptService(a.DB, profileSvc, a.KolPackages, a.KolPrompts, a.KolSubscriptions, a.KolPromptAsset).WithCache(a.Cache),
-		assistSvc:  app.NewKolAssistService(profileSvc, a.KolAssists, a.LLM, a.RabbitMQ),
+		assistSvc:  app.NewKolAssistService(profileSvc, a.KolAssists, a.LLM, a.RabbitMQ, a.Config.LLM, a.Logger),
 		assets:     assets,
 	}
 }