feat(migrations): Harden task audit tracking and optimize article templates

- Added migration to harden task audit tracking by modifying audit_logs and related tables.
- Introduced operator_id to several tables for better tracking of actions.
- Updated article_templates with new prompt templates for various article types, enhancing content generation.
- Created prompt_rules and schedule_tasks tables to manage content generation rules and scheduling.
- Added foreign key constraints to articles for better data integrity.

server/internal/shared/llm/ark.go

@@ -25,7 +25,10 @@ type arkClient struct {
 	model           string
 	timeout         time.Duration
 	maxOutputTokens int64
+	webSearchLimit  int64
 	temperature     float64
+	topP            *float64
+	reasoning       *responses.ResponsesReasoning
 }
 
 func NewArkClient(cfg config.LLMConfig) Client {
@@ -53,17 +56,30 @@ func NewArkClient(cfg config.LLMConfig) Client {
 		maxOutputTokens = 4000
 	}
 
+	webSearchLimit := int64(cfg.WebSearchLimit)
+	if webSearchLimit <= 0 {
+		webSearchLimit = 3
+	}
+
 	temperature := cfg.Temperature
 	if temperature <= 0 {
 		temperature = 0.7
 	}
 
+	var topP *float64
+	if cfg.TopP > 0 {
+		topP = &cfg.TopP
+	}
+
 	return &arkClient{
 		client:          arkruntime.NewClientWithApiKey(cfg.APIKey, arkruntime.WithBaseUrl(baseURL)),
 		model:           model,
 		timeout:         timeout,
 		maxOutputTokens: maxOutputTokens,
+		webSearchLimit:  webSearchLimit,
 		temperature:     temperature,
+		topP:            topP,
+		reasoning:       resolveArkReasoning(cfg.ReasoningEffort),
 	}
 }
 
@@ -87,10 +103,21 @@ func (c *arkClient) Generate(ctx context.Context, req GenerateRequest, onDelta f
 		return nil, errors.New("prompt is required")
 	}
 
-	callCtx, cancel := context.WithTimeout(ctx, c.timeout)
+	timeout := c.timeout
+	if req.Timeout > 0 {
+		timeout = req.Timeout
+	}
+
+	maxOutputTokens := c.maxOutputTokens
+	if req.MaxOutputTokens > 0 {
+		maxOutputTokens = req.MaxOutputTokens
+	}
+
+	callCtx, cancel := context.WithTimeout(ctx, timeout)
 	defer cancel()
 
 	store := false
+	streamValue := true
 	inputMessage := &responses.ItemInputMessage{
 		Role: responses.MessageRole_user,
 		Content: []*responses.ContentItem{
@@ -105,11 +132,34 @@ func (c *arkClient) Generate(ctx context.Context, req GenerateRequest, onDelta f
 		},
 	}
 
+	var tools []*responses.ResponsesTool
+	if req.WebSearch != nil && req.WebSearch.Enabled {
+		webSearchLimit := c.webSearchLimit
+		if req.WebSearch.Limit > 0 {
+			webSearchLimit = int64(req.WebSearch.Limit)
+		}
+
+		tools = []*responses.ResponsesTool{
+			{
+				Union: &responses.ResponsesTool_ToolWebSearch{
+					ToolWebSearch: &responses.ToolWebSearch{
+						Type:  responses.ToolType_web_search,
+						Limit: &webSearchLimit,
+					},
+				},
+			},
+		}
+	}
+
 	stream, err := c.client.CreateResponsesStream(callCtx, &responses.ResponsesRequest{
 		Model:           c.model,
-		MaxOutputTokens: &c.maxOutputTokens,
+		MaxOutputTokens: &maxOutputTokens,
 		Store:           &store,
+		Stream:          &streamValue,
 		Temperature:     &c.temperature,
+		TopP:            c.topP,
+		Tools:           tools,
+		Reasoning:       c.reasoning,
 		Input: &responses.ResponsesInput{
 			Union: &responses.ResponsesInput_ListValue{
 				ListValue: &responses.InputItemList{
@@ -125,6 +175,9 @@ func (c *arkClient) Generate(ctx context.Context, req GenerateRequest, onDelta f
 		},
 	})
 	if err != nil {
+		if errors.Is(err, context.DeadlineExceeded) || errors.Is(callCtx.Err(), context.DeadlineExceeded) {
+			return nil, fmt.Errorf("ark response stream timeout after %s: %w", timeout, err)
+		}
 		return nil, fmt.Errorf("create ark response stream: %w", err)
 	}
 	defer stream.Close()
@@ -137,6 +190,9 @@ func (c *arkClient) Generate(ctx context.Context, req GenerateRequest, onDelta f
 			if errors.Is(err, io.EOF) {
 				break
 			}
+			if errors.Is(err, context.DeadlineExceeded) || errors.Is(callCtx.Err(), context.DeadlineExceeded) {
+				return nil, fmt.Errorf("ark response stream timeout after %s: %w", timeout, err)
+			}
 			return nil, fmt.Errorf("receive ark stream event: %w", err)
 		}
 
@@ -176,3 +232,20 @@ func (c *arkClient) Generate(ctx context.Context, req GenerateRequest, onDelta f
 		Model:   c.model,
 	}, nil
 }
+
+func resolveArkReasoning(value string) *responses.ResponsesReasoning {
+	switch strings.ToLower(strings.TrimSpace(value)) {
+	case "":
+		return nil
+	case "minimal":
+		return &responses.ResponsesReasoning{Effort: responses.ReasoningEffort_minimal}
+	case "low":
+		return &responses.ResponsesReasoning{Effort: responses.ReasoningEffort_low}
+	case "medium":
+		return &responses.ResponsesReasoning{Effort: responses.ReasoningEffort_medium}
+	case "high":
+		return &responses.ResponsesReasoning{Effort: responses.ReasoningEffort_high}
+	default:
+		return nil
+	}
+}

server/internal/shared/llm/client.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"strings"
+	"time"
 
 	"github.com/geo-platform/tenant-api/internal/shared/config"
 )
@@ -12,7 +13,10 @@ import (
 var ErrNotConfigured = errors.New("llm provider is not configured")
 
 type GenerateRequest struct {
-	Prompt string
+	Prompt          string
+	Timeout         time.Duration
+	MaxOutputTokens int64
+	WebSearch       *WebSearchOptions
 }
 
 type GenerateResult struct {
@@ -20,6 +24,11 @@ type GenerateResult struct {
 	Model   string
 }
 
+type WebSearchOptions struct {
+	Enabled bool
+	Limit   int32
+}
+
 type Client interface {
 	Validate() error
 	Generate(ctx context.Context, req GenerateRequest, onDelta func(string)) (*GenerateResult, error)