Version: 0.9.75.dev.260505
后端： 1.收口阶段 6 agent 结构迁移，将 newAgent 内核与 agentsvc 编排层迁入 services/agent - 切换 Agent 启动装配与 HTTP handler 直连 agent sv，移除旧 service agent bridge - 补齐 Agent 对 memory、task、task-class、schedule 的 RPC 适配与契约字段 - 扩展 schedule、task、task-class RPC/contract 支撑 Agent 查询、写入与 provider 切流 - 更新迁移文档、README 与相关注释，明确 agent 当前切流点和剩余 memory 迁移面
This commit is contained in:
Losita
908
backend/services/agent/node/chat.go
Normal file
908
backend/services/agent/node/chat.go
Normal file
@@ -0,0 +1,908 @@
|
||||
package agentnode
|
||||
|
||||
import (
|
||||
"context"
|
||||
"fmt"
|
||||
"io"
|
||||
"log"
|
||||
"strings"
|
||||
"time"
|
||||
|
||||
"github.com/cloudwego/eino/schema"
|
||||
"github.com/google/uuid"
|
||||
|
||||
agentmodel "github.com/LoveLosita/smartflow/backend/services/agent/model"
|
||||
agentprompt "github.com/LoveLosita/smartflow/backend/services/agent/prompt"
|
||||
agentrouter "github.com/LoveLosita/smartflow/backend/services/agent/router"
|
||||
agentstream "github.com/LoveLosita/smartflow/backend/services/agent/stream"
|
||||
llmservice "github.com/LoveLosita/smartflow/backend/services/llm"
|
||||
)
|
||||
|
||||
const (
|
||||
chatStageName = "chat"
|
||||
chatStatusBlockID = "chat.status"
|
||||
chatSpeakBlockID = "chat.speak"
|
||||
// chatHistoryKindKey 用于在 history 中打运行态标记，供 prompt 层做上下文分层。
|
||||
chatHistoryKindKey = "newagent_history_kind"
|
||||
// chatHistoryKindExecuteLoopClosed 表示"上一轮 execute loop 已正常收口"。
|
||||
// prompt 侧会据此把旧 loop 归档到 msg1，而不是继续占用 msg2 窗口。
|
||||
chatHistoryKindExecuteLoopClosed = "execute_loop_closed"
|
||||
)
|
||||
|
||||
type reorderPreference int
|
||||
|
||||
const (
|
||||
reorderUnknown reorderPreference = iota
|
||||
reorderAllow
|
||||
reorderDisallow
|
||||
)
|
||||
|
||||
// ChatNodeInput 描述聊天节点单轮运行所需的最小依赖。
|
||||
//
|
||||
// 职责边界：
|
||||
// 1. 只承载"本轮 chat"需要的输入，不负责持久化；
|
||||
// 2. RuntimeState 提供 pending interaction 与流程状态；
|
||||
// 3. ConversationContext 提供历史对话；
|
||||
// 4. ConfirmAction 仅在 confirm 恢复场景下由前端传入 "accept" / "reject"。
|
||||
type ChatNodeInput struct {
|
||||
RuntimeState *agentmodel.AgentRuntimeState
|
||||
ConversationContext *agentmodel.ConversationContext
|
||||
UserInput string
|
||||
ConfirmAction string
|
||||
ResumeInteractionID string
|
||||
Client *llmservice.Client
|
||||
ChunkEmitter *agentstream.ChunkEmitter
|
||||
CompactionStore agentmodel.CompactionStore // 上下文压缩持久化
|
||||
PersistVisibleMessage agentmodel.PersistVisibleMessageFunc
|
||||
}
|
||||
|
||||
// RunChatNode 执行一轮聊天节点逻辑。
|
||||
//
|
||||
// 核心职责：
|
||||
// 1. 恢复判定：有 pending interaction 则处理恢复；
|
||||
// 2. 路由分流：无 pending 时，调 LLM 判断复杂度并路由；
|
||||
// 3. direct_reply：简单任务，直接输出回复 → END；
|
||||
// 4. execute：中等任务，推 Execute ReAct；
|
||||
// 5. deep_answer：复杂问答，原地开 thinking 深度回答 → END；
|
||||
// 6. plan：复杂规划，推 Plan 节点。
|
||||
func RunChatNode(ctx context.Context, input ChatNodeInput) error {
|
||||
runtimeState, conversationContext, emitter, err := prepareChatNodeInput(input)
|
||||
if err != nil {
|
||||
return err
|
||||
}
|
||||
|
||||
// 1. 有 pending interaction → 纯状态传递，处理恢复。
|
||||
if runtimeState.HasPendingInteraction() {
|
||||
return handleChatResume(input, runtimeState, emitter)
|
||||
}
|
||||
|
||||
// 2. 无 pending → 路由决策（一次快速 LLM 调用，不开 thinking）。
|
||||
flowState := runtimeState.EnsureCommonState()
|
||||
if !runtimeState.HasPendingInteraction() && flowState.Phase == agentmodel.PhaseDone {
|
||||
terminalBefore := flowState.TerminalStatus()
|
||||
roundBefore := flowState.RoundUsed
|
||||
// 1. 只有"正常完成(completed)"才打 loop 收口标记：
|
||||
// 1.1 这样下一轮进入 execute 时，msg2 会只保留"当前活跃循环"窗口；
|
||||
// 1.2 异常收口（exhausted/aborted）不打标记，允许后续"继续"时沿用上一轮 loop 轨迹。
|
||||
if terminalBefore == agentmodel.FlowTerminalStatusCompleted {
|
||||
appendExecuteLoopClosedMarker(conversationContext)
|
||||
}
|
||||
flowState.ResetForNextRun()
|
||||
log.Printf(
|
||||
"[DEBUG] chat reset runtime for next run chat=%s round_before=%d terminal_before=%s",
|
||||
flowState.ConversationID,
|
||||
roundBefore,
|
||||
terminalBefore,
|
||||
)
|
||||
}
|
||||
nonce := uuid.NewString()
|
||||
messages := agentprompt.BuildChatRoutingMessages(conversationContext, input.UserInput, flowState, nonce)
|
||||
messages = compactUnifiedMessagesIfNeeded(ctx, messages, UnifiedCompactInput{
|
||||
Client: input.Client,
|
||||
CompactionStore: input.CompactionStore,
|
||||
FlowState: flowState,
|
||||
Emitter: emitter,
|
||||
StageName: chatStageName,
|
||||
StatusBlockID: chatStatusBlockID,
|
||||
})
|
||||
logNodeLLMContext(chatStageName, "routing", flowState, messages)
|
||||
|
||||
reader, err := input.Client.Stream(ctx, messages, llmservice.GenerateOptions{
|
||||
Temperature: 0.7,
|
||||
Thinking: llmservice.ThinkingModeDisabled,
|
||||
Metadata: map[string]any{
|
||||
"stage": chatStageName,
|
||||
"phase": "routing",
|
||||
},
|
||||
})
|
||||
if err != nil {
|
||||
log.Printf("[WARN] chat routing stream failed chat=%s err=%v", flowState.ConversationID, err)
|
||||
flowState.Phase = agentmodel.PhasePlanning
|
||||
return nil
|
||||
}
|
||||
|
||||
parser := agentrouter.NewStreamRouteParser(nonce)
|
||||
return streamAndDispatch(ctx, reader, parser, input, emitter, flowState, conversationContext)
|
||||
}
|
||||
|
||||
// appendExecuteLoopClosedMarker 在 history 中写入"execute loop 已正常收口"标记。
|
||||
//
|
||||
// 职责边界：
|
||||
// 1. 只负责写一个轻量 marker，供 prompt 分层；
|
||||
// 2. 不负责历史裁剪，不负责消息摘要；
|
||||
// 3. 若末尾已经是同类 marker，则幂等跳过，避免重复写入。
|
||||
func appendExecuteLoopClosedMarker(conversationContext *agentmodel.ConversationContext) {
|
||||
if conversationContext == nil {
|
||||
return
|
||||
}
|
||||
|
||||
history := conversationContext.HistorySnapshot()
|
||||
if len(history) > 0 {
|
||||
last := history[len(history)-1]
|
||||
if isExecuteLoopClosedMarker(last) {
|
||||
return
|
||||
}
|
||||
}
|
||||
|
||||
conversationContext.AppendHistory(&schema.Message{
|
||||
Role: schema.Assistant,
|
||||
Content: "",
|
||||
Extra: map[string]any{
|
||||
chatHistoryKindKey: chatHistoryKindExecuteLoopClosed,
|
||||
},
|
||||
})
|
||||
}
|
||||
|
||||
func isExecuteLoopClosedMarker(msg *schema.Message) bool {
|
||||
if msg == nil || msg.Extra == nil {
|
||||
return false
|
||||
}
|
||||
kind, ok := msg.Extra[chatHistoryKindKey].(string)
|
||||
if !ok {
|
||||
return false
|
||||
}
|
||||
return strings.TrimSpace(kind) == chatHistoryKindExecuteLoopClosed
|
||||
}
|
||||
|
||||
// streamAndDispatch 是流式路由分发的核心循环。
|
||||
//
|
||||
// 步骤说明：
|
||||
// 1. 从 StreamReader 逐 chunk 读取，喂给 StreamRouteParser 增量解析控制码；
|
||||
// 2. 控制码解析完成后，根据 route 进入对应的流式处理分支；
|
||||
// 3. 控制码解析超时或流异常结束 → fallback 到 plan。
|
||||
func streamAndDispatch(
|
||||
ctx context.Context,
|
||||
reader llmservice.StreamReader,
|
||||
parser *agentrouter.StreamRouteParser,
|
||||
input ChatNodeInput,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
flowState *agentmodel.CommonState,
|
||||
conversationContext *agentmodel.ConversationContext,
|
||||
) error {
|
||||
for {
|
||||
chunk, err := reader.Recv()
|
||||
if err == io.EOF {
|
||||
if !parser.RouteReady() {
|
||||
log.Printf("[WARN] chat stream ended before route resolved chat=%s", flowState.ConversationID)
|
||||
flowState.Phase = agentmodel.PhasePlanning
|
||||
return nil
|
||||
}
|
||||
break
|
||||
}
|
||||
if err != nil {
|
||||
log.Printf("[WARN] chat stream recv error chat=%s err=%v", flowState.ConversationID, err)
|
||||
flowState.Phase = agentmodel.PhasePlanning
|
||||
return nil
|
||||
}
|
||||
|
||||
content := ""
|
||||
if chunk != nil {
|
||||
content = chunk.Content
|
||||
}
|
||||
|
||||
visible, routeReady, _ := parser.Feed(content)
|
||||
if !routeReady {
|
||||
continue
|
||||
}
|
||||
|
||||
// 控制码解析完成，进入路由分发。
|
||||
decision := parser.Decision()
|
||||
|
||||
// 二次粗排硬闸门：若上下文已存在 rough_build_done 且用户未明确要求"重新粗排"，
|
||||
// 则强制关闭 needs_rough_build，避免"微调请求被误判成再次粗排"。
|
||||
if shouldDisableRoughBuildForRefine(conversationContext, input.UserInput, decision) {
|
||||
decision.NeedsRoughBuild = false
|
||||
decision.NeedsRefineAfterRoughBuild = false
|
||||
}
|
||||
// 首次粗排兜底：若用户未明确要求"只要初稿不优化"，则粗排后默认进入主动微调。
|
||||
if shouldForceRefineAfterFirstRoughBuild(conversationContext, input.UserInput, decision) {
|
||||
decision.NeedsRefineAfterRoughBuild = true
|
||||
}
|
||||
|
||||
log.Printf(
|
||||
"[DEBUG] chat routing chat=%s route=%s needs_rough_build=%v needs_refine_after_rough_build=%v allow_reorder=%v thinking=%v has_rough_build_done=%v task_class_count=%d raw=%s",
|
||||
flowState.ConversationID,
|
||||
decision.Route,
|
||||
decision.NeedsRoughBuild,
|
||||
decision.NeedsRefineAfterRoughBuild,
|
||||
decision.AllowReorder,
|
||||
decision.Thinking,
|
||||
hasRoughBuildDoneMarker(conversationContext),
|
||||
len(flowState.TaskClassIDs),
|
||||
decision.Raw,
|
||||
)
|
||||
|
||||
flowState.AllowReorder = resolveAllowReorder(input.UserInput, decision.AllowReorder)
|
||||
effectiveThinking := resolveEffectiveThinking(flowState.ThinkingMode, decision.Route, decision.Thinking)
|
||||
|
||||
switch decision.Route {
|
||||
case agentmodel.ChatRouteDirectReply:
|
||||
return handleDirectReplyStream(ctx, reader, input, emitter, conversationContext, flowState, effectiveThinking, visible)
|
||||
|
||||
case agentmodel.ChatRouteExecute:
|
||||
return handleRouteExecuteStream(reader, emitter, flowState, decision, input.UserInput, effectiveThinking, visible)
|
||||
|
||||
case agentmodel.ChatRouteDeepAnswer:
|
||||
return handleDeepAnswerStream(ctx, reader, input, emitter, conversationContext, flowState, effectiveThinking)
|
||||
|
||||
case agentmodel.ChatRoutePlan:
|
||||
return handleRoutePlanStream(reader, emitter, flowState, effectiveThinking, visible)
|
||||
|
||||
case agentmodel.ChatRouteQuickTask:
|
||||
// 关闭路由流，后续由 QuickTask 节点自行处理。
|
||||
_ = reader.Close()
|
||||
flowState.Phase = agentmodel.PhaseQuickTask
|
||||
return nil
|
||||
|
||||
default:
|
||||
flowState.Phase = agentmodel.PhasePlanning
|
||||
return nil
|
||||
}
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
// resolveEffectiveThinking 根据前端 ThinkingMode 和路由决策合并出最终 thinking 状态。
|
||||
//
|
||||
// 规则：
|
||||
// 1. "true"：前端强制开启，所有路由统一开；
|
||||
// 2. "false"：前端强制关闭，所有路由统一关；
|
||||
// 3. "auto"/""：按路由语义兜底；
|
||||
// 3.1 deep_answer 的语义本身就是"复杂问答 + 原地深度思考"，因此默认开启；
|
||||
// 3.2 execute 继续沿用路由模型给出的 decisionThinking；
|
||||
// 3.3 其余路由默认关闭，避免把轻量闲聊误升成高成本推理。
|
||||
func resolveEffectiveThinking(mode string, route agentmodel.ChatRoute, decisionThinking bool) bool {
|
||||
switch strings.TrimSpace(strings.ToLower(mode)) {
|
||||
case "true":
|
||||
return true
|
||||
case "false":
|
||||
return false
|
||||
default:
|
||||
if route == agentmodel.ChatRouteDeepAnswer {
|
||||
return true
|
||||
}
|
||||
return decisionThinking
|
||||
}
|
||||
}
|
||||
|
||||
// handleDirectReplyStream 处理闲聊回复。
|
||||
//
|
||||
// 两种模式：
|
||||
// 1. thinking=false：同一流续传，逐 chunk 推送；
|
||||
// 2. thinking=true：关闭路由流，发起第二次 thinking 流式调用。
|
||||
func handleDirectReplyStream(
|
||||
ctx context.Context,
|
||||
reader llmservice.StreamReader,
|
||||
input ChatNodeInput,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
conversationContext *agentmodel.ConversationContext,
|
||||
flowState *agentmodel.CommonState,
|
||||
effectiveThinking bool,
|
||||
firstVisible string,
|
||||
) error {
|
||||
if effectiveThinking {
|
||||
return handleThinkingReplyStream(ctx, reader, input, emitter, conversationContext, flowState)
|
||||
}
|
||||
return handleDirectReplyContinueStream(ctx, reader, input, emitter, conversationContext, flowState, firstVisible)
|
||||
}
|
||||
|
||||
// handleThinkingReplyStream 处理需要思考的回复：关闭路由流 → 第二次 thinking 流式调用。
|
||||
func handleThinkingReplyStream(
|
||||
ctx context.Context,
|
||||
reader llmservice.StreamReader,
|
||||
input ChatNodeInput,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
conversationContext *agentmodel.ConversationContext,
|
||||
flowState *agentmodel.CommonState,
|
||||
) error {
|
||||
_ = reader.Close()
|
||||
|
||||
deepMessages := agentprompt.BuildDeepAnswerMessages(flowState, conversationContext, input.UserInput)
|
||||
deepMessages = compactUnifiedMessagesIfNeeded(ctx, deepMessages, UnifiedCompactInput{
|
||||
Client: input.Client,
|
||||
CompactionStore: input.CompactionStore,
|
||||
FlowState: flowState,
|
||||
Emitter: emitter,
|
||||
StageName: chatStageName,
|
||||
StatusBlockID: chatStatusBlockID,
|
||||
})
|
||||
logNodeLLMContext(chatStageName, "direct_reply_thinking", flowState, deepMessages)
|
||||
deepReader, err := input.Client.Stream(ctx, deepMessages, llmservice.GenerateOptions{
|
||||
Temperature: 0.5,
|
||||
MaxTokens: 2000,
|
||||
Thinking: llmservice.ThinkingModeEnabled,
|
||||
Metadata: map[string]any{
|
||||
"stage": chatStageName,
|
||||
"phase": "direct_reply_thinking",
|
||||
},
|
||||
})
|
||||
if err != nil {
|
||||
log.Printf("[WARN] thinking reply stream failed chat=%s err=%v", flowState.ConversationID, err)
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
deepText, err := emitter.EmitStreamAssistantText(ctx, deepReader, chatSpeakBlockID, chatStageName)
|
||||
_ = deepReader.Close()
|
||||
if err != nil {
|
||||
log.Printf("[WARN] thinking reply emit error chat=%s err=%v", flowState.ConversationID, err)
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
deepText = strings.TrimSpace(deepText)
|
||||
if deepText != "" {
|
||||
conversationContext.AppendHistory(schema.AssistantMessage(deepText, nil))
|
||||
persistVisibleAssistantMessage(ctx, input.PersistVisibleMessage, flowState, schema.AssistantMessage(deepText, nil))
|
||||
}
|
||||
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
// handleDirectReplyContinueStream 处理无思考的闲聊：同一流续传。
|
||||
func handleDirectReplyContinueStream(
|
||||
ctx context.Context,
|
||||
reader llmservice.StreamReader,
|
||||
input ChatNodeInput,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
conversationContext *agentmodel.ConversationContext,
|
||||
flowState *agentmodel.CommonState,
|
||||
firstVisible string,
|
||||
) error {
|
||||
var fullText strings.Builder
|
||||
fullText.WriteString(firstVisible)
|
||||
|
||||
// 推送控制码之后的第一段内容。
|
||||
if strings.TrimSpace(firstVisible) != "" {
|
||||
if err := emitter.EmitAssistantText(chatSpeakBlockID, chatStageName, firstVisible, true); err != nil {
|
||||
return fmt.Errorf("闲聊回复推送失败: %w", err)
|
||||
}
|
||||
}
|
||||
|
||||
firstChunk := firstVisible == ""
|
||||
// 继续读同一个流，逐 chunk 推送。
|
||||
for {
|
||||
chunk, err := reader.Recv()
|
||||
if err == io.EOF {
|
||||
break
|
||||
}
|
||||
if err != nil {
|
||||
log.Printf("[WARN] direct_reply stream error chat=%s err=%v", flowState.ConversationID, err)
|
||||
break
|
||||
}
|
||||
if chunk == nil || chunk.Content == "" {
|
||||
continue
|
||||
}
|
||||
if err := emitter.EmitAssistantText(chatSpeakBlockID, chatStageName, chunk.Content, firstChunk); err != nil {
|
||||
return fmt.Errorf("闲聊回复推送失败: %w", err)
|
||||
}
|
||||
fullText.WriteString(chunk.Content)
|
||||
firstChunk = false
|
||||
}
|
||||
|
||||
text := fullText.String()
|
||||
if strings.TrimSpace(text) != "" {
|
||||
msg := schema.AssistantMessage(text, nil)
|
||||
conversationContext.AppendHistory(msg)
|
||||
persistVisibleAssistantMessage(ctx, input.PersistVisibleMessage, flowState, msg)
|
||||
}
|
||||
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
// handleRouteExecuteStream 处理工具调用路由：推送状态确认 → 设 PhaseExecuting。
|
||||
//
|
||||
// 说明：
|
||||
// 1. 关闭路由流（后续内容不需要）；
|
||||
// 2. 推送轻量状态通知；
|
||||
// 3. 设置流程状态，进入 Execute 或 RoughBuild。
|
||||
func handleRouteExecuteStream(
|
||||
reader llmservice.StreamReader,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
flowState *agentmodel.CommonState,
|
||||
decision *agentmodel.ChatRoutingDecision,
|
||||
userInput string,
|
||||
effectiveThinking bool,
|
||||
speak string,
|
||||
) error {
|
||||
// 关闭路由流。
|
||||
_ = reader.Close()
|
||||
|
||||
if strings.TrimSpace(speak) == "" {
|
||||
speak = "好的，我来处理。"
|
||||
}
|
||||
|
||||
// 推送轻量状态通知。
|
||||
_ = emitter.EmitStatus(chatStatusBlockID, chatStageName, "accepted", speak, false)
|
||||
|
||||
// 清空旧 PlanSteps 并设 PhaseExecuting。
|
||||
flowState.StartDirectExecute()
|
||||
|
||||
// 粗排开关逻辑。
|
||||
flowState.NeedsRoughBuild = false
|
||||
flowState.NeedsRefineAfterRoughBuild = false
|
||||
if decision.NeedsRoughBuild && len(flowState.TaskClassIDs) > 0 {
|
||||
flowState.NeedsRoughBuild = true
|
||||
flowState.NeedsRefineAfterRoughBuild = decision.NeedsRefineAfterRoughBuild
|
||||
}
|
||||
|
||||
flowState.ExecuteThinking = effectiveThinking
|
||||
flowState.OptimizationMode = resolveOptimizationMode(userInput, decision, flowState)
|
||||
|
||||
return nil
|
||||
}
|
||||
|
||||
// resolveAllowReorder 统一计算"本轮是否允许打乱顺序"。
|
||||
//
|
||||
// 步骤化说明：
|
||||
// 1. 后端先做显式语义判定：用户明确允许/明确禁止时，直接以后端判定为准；
|
||||
// 2. 若后端未识别到显式语义，再回退到路由模型的 allow_reorder 字段；
|
||||
// 3. 默认返回 false，确保"保持顺序"是系统默认行为。
|
||||
func resolveAllowReorder(userInput string, modelAllowReorder bool) bool {
|
||||
switch detectReorderPreference(userInput) {
|
||||
case reorderAllow:
|
||||
return true
|
||||
case reorderDisallow:
|
||||
return false
|
||||
default:
|
||||
return modelAllowReorder
|
||||
}
|
||||
}
|
||||
|
||||
// detectReorderPreference 识别用户是否"明确授权打乱顺序"。
|
||||
//
|
||||
// 职责边界：
|
||||
// 1. 只负责关键词级别的显式意图识别，不做复杂语义推理；
|
||||
// 2. 若同时命中"允许"与"禁止"，优先按"禁止"处理，避免误放开顺序约束；
|
||||
// 3. 未命中显式表达时返回 unknown，交给上层兜底策略。
|
||||
func detectReorderPreference(userInput string) reorderPreference {
|
||||
text := strings.ToLower(strings.TrimSpace(userInput))
|
||||
if text == "" {
|
||||
return reorderUnknown
|
||||
}
|
||||
|
||||
disallowPhrases := []string{
|
||||
"不要打乱顺序",
|
||||
"不允许打乱顺序",
|
||||
"保持顺序",
|
||||
"顺序不变",
|
||||
"按原顺序",
|
||||
"不要乱序",
|
||||
"别打乱",
|
||||
}
|
||||
if containsAnyPhrase(text, disallowPhrases) {
|
||||
return reorderDisallow
|
||||
}
|
||||
|
||||
allowPhrases := []string{
|
||||
"可以打乱顺序",
|
||||
"允许打乱顺序",
|
||||
"顺序不重要",
|
||||
"顺序无所谓",
|
||||
"顺序不限",
|
||||
"允许乱序",
|
||||
"可以乱序",
|
||||
"允许重排顺序",
|
||||
"reorder is fine",
|
||||
"any order",
|
||||
}
|
||||
if containsAnyPhrase(text, allowPhrases) {
|
||||
return reorderAllow
|
||||
}
|
||||
|
||||
return reorderUnknown
|
||||
}
|
||||
|
||||
// resolveOptimizationMode 统一确定当前 execute 的优化模式。
|
||||
func resolveOptimizationMode(
|
||||
userInput string,
|
||||
decision *agentmodel.ChatRoutingDecision,
|
||||
flowState *agentmodel.CommonState,
|
||||
) string {
|
||||
if decision != nil && decision.NeedsRoughBuild && flowState != nil && len(flowState.TaskClassIDs) > 0 {
|
||||
return "first_full"
|
||||
}
|
||||
if isExplicitGlobalReoptRequest(userInput) {
|
||||
return "global_reopt"
|
||||
}
|
||||
return "local_adjust"
|
||||
}
|
||||
|
||||
// isExplicitGlobalReoptRequest 识别用户是否明确要求全局重优化。
|
||||
func isExplicitGlobalReoptRequest(userInput string) bool {
|
||||
text := strings.ToLower(strings.TrimSpace(userInput))
|
||||
if text == "" {
|
||||
return false
|
||||
}
|
||||
keywords := []string{
|
||||
"全局优化",
|
||||
"整体优化",
|
||||
"全局重排",
|
||||
"整体重排",
|
||||
"重新优化全部",
|
||||
"重新优化整体",
|
||||
"全面优化",
|
||||
"整体体检",
|
||||
"全局体检",
|
||||
"重新体检",
|
||||
"global optimize",
|
||||
"global reopt",
|
||||
"overall optimize",
|
||||
}
|
||||
return containsAnyPhrase(text, keywords)
|
||||
}
|
||||
|
||||
func containsAnyPhrase(text string, phrases []string) bool {
|
||||
for _, phrase := range phrases {
|
||||
if strings.Contains(text, phrase) {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
// shouldDisableRoughBuildForRefine 判断是否应在 chat 路由阶段关闭"再次粗排"。
|
||||
//
|
||||
// 判定规则：
|
||||
// 1. 当前决策未请求粗排时，直接不干预；
|
||||
// 2. 上下文不存在 rough_build_done 时，不干预（首次粗排仍可走）；
|
||||
// 3. 若用户未明确要求"重新粗排/从头重排"，则关闭粗排开关，避免误触发。
|
||||
func shouldDisableRoughBuildForRefine(
|
||||
conversationContext *agentmodel.ConversationContext,
|
||||
userInput string,
|
||||
decision *agentmodel.ChatRoutingDecision,
|
||||
) bool {
|
||||
if decision == nil || !decision.NeedsRoughBuild {
|
||||
return false
|
||||
}
|
||||
if !hasRoughBuildDoneMarker(conversationContext) {
|
||||
return false
|
||||
}
|
||||
return !isExplicitRoughBuildRequest(userInput)
|
||||
}
|
||||
|
||||
// shouldForceRefineAfterFirstRoughBuild 判断是否应在"首次粗排"场景下强制开启 refine。
|
||||
//
|
||||
// 判定规则：
|
||||
// 1. 仅在当前决策仍然请求粗排时生效；
|
||||
// 2. 仅在首次粗排（上下文不存在 rough_build_done）时生效；
|
||||
// 3. 若用户明确表达"只要初稿/先不优化"，则不强制开启；
|
||||
// 4. 其余首次粗排场景一律开启，确保符合 PRD 的默认主动优化策略。
|
||||
func shouldForceRefineAfterFirstRoughBuild(
|
||||
conversationContext *agentmodel.ConversationContext,
|
||||
userInput string,
|
||||
decision *agentmodel.ChatRoutingDecision,
|
||||
) bool {
|
||||
if decision == nil || !decision.NeedsRoughBuild {
|
||||
return false
|
||||
}
|
||||
if hasRoughBuildDoneMarker(conversationContext) {
|
||||
return false
|
||||
}
|
||||
return !isExplicitNoRefineAfterRoughBuildRequest(userInput)
|
||||
}
|
||||
|
||||
func hasRoughBuildDoneMarker(conversationContext *agentmodel.ConversationContext) bool {
|
||||
if conversationContext == nil {
|
||||
return false
|
||||
}
|
||||
for _, block := range conversationContext.PinnedBlocksSnapshot() {
|
||||
if strings.TrimSpace(block.Key) == "rough_build_done" {
|
||||
return true
|
||||
}
|
||||
}
|
||||
return false
|
||||
}
|
||||
|
||||
// isExplicitRoughBuildRequest 识别用户是否明确要求"重新粗排/从头重排"。
|
||||
func isExplicitRoughBuildRequest(userInput string) bool {
|
||||
text := strings.ToLower(strings.TrimSpace(userInput))
|
||||
if text == "" {
|
||||
return false
|
||||
}
|
||||
keywords := []string{
|
||||
"重新粗排",
|
||||
"重做粗排",
|
||||
"从头排",
|
||||
"从头重排",
|
||||
"重新排一遍",
|
||||
"重新排课",
|
||||
"重排全部",
|
||||
"全部重排",
|
||||
"重置排程",
|
||||
"重置后重排",
|
||||
"重新生成初稿",
|
||||
"rebuild",
|
||||
"from scratch",
|
||||
}
|
||||
return containsAnyPhrase(text, keywords)
|
||||
}
|
||||
|
||||
// isExplicitNoRefineAfterRoughBuildRequest 识别用户是否明确要求"粗排后先不要自动微调"。
|
||||
func isExplicitNoRefineAfterRoughBuildRequest(userInput string) bool {
|
||||
text := strings.ToLower(strings.TrimSpace(userInput))
|
||||
if text == "" {
|
||||
return false
|
||||
}
|
||||
keywords := []string{
|
||||
"只要初稿",
|
||||
"先给初稿",
|
||||
"先排进去就行",
|
||||
"先排进去",
|
||||
"先不优化",
|
||||
"先别优化",
|
||||
"先不微调",
|
||||
"先别微调",
|
||||
"排完就收口",
|
||||
"粗排就行",
|
||||
"草稿就行",
|
||||
"draft only",
|
||||
"no refine",
|
||||
"no optimization",
|
||||
}
|
||||
return containsAnyPhrase(text, keywords)
|
||||
}
|
||||
|
||||
// handleDeepAnswerStream 处理复杂问答：关闭路由流 → 第二次流式调用。
|
||||
//
|
||||
// 步骤说明：
|
||||
// 1. 关闭第一个路由流；
|
||||
// 2. 发起第二次流式 LLM 调用（thinking 由 effectiveThinking 控制）；
|
||||
// 3. 真流式推送 reasoning + 正文；
|
||||
// 4. 完整回复写入 history。
|
||||
func handleDeepAnswerStream(
|
||||
ctx context.Context,
|
||||
reader llmservice.StreamReader,
|
||||
input ChatNodeInput,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
conversationContext *agentmodel.ConversationContext,
|
||||
flowState *agentmodel.CommonState,
|
||||
effectiveThinking bool,
|
||||
) error {
|
||||
// 1. 关闭第一个路由流。
|
||||
_ = reader.Close()
|
||||
|
||||
// 2. 第二次流式调用。
|
||||
thinkingOpt := llmservice.ThinkingModeDisabled
|
||||
if effectiveThinking {
|
||||
thinkingOpt = llmservice.ThinkingModeEnabled
|
||||
}
|
||||
deepMessages := agentprompt.BuildDeepAnswerMessages(flowState, conversationContext, input.UserInput)
|
||||
deepMessages = compactUnifiedMessagesIfNeeded(ctx, deepMessages, UnifiedCompactInput{
|
||||
Client: input.Client,
|
||||
CompactionStore: input.CompactionStore,
|
||||
FlowState: flowState,
|
||||
Emitter: emitter,
|
||||
StageName: chatStageName,
|
||||
StatusBlockID: chatStatusBlockID,
|
||||
})
|
||||
logNodeLLMContext(chatStageName, "deep_answer", flowState, deepMessages)
|
||||
deepReader, err := input.Client.Stream(ctx, deepMessages, llmservice.GenerateOptions{
|
||||
Temperature: 0.5,
|
||||
MaxTokens: 2000,
|
||||
Thinking: thinkingOpt,
|
||||
Metadata: map[string]any{
|
||||
"stage": chatStageName,
|
||||
"phase": "deep_answer",
|
||||
},
|
||||
})
|
||||
if err != nil {
|
||||
// 深度回答失败 → 降级返回。
|
||||
log.Printf("[WARN] deep answer stream failed chat=%s err=%v", flowState.ConversationID, err)
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
// 3. 真流式推送 reasoning + 正文。
|
||||
deepText, err := emitter.EmitStreamAssistantText(ctx, deepReader, chatSpeakBlockID, chatStageName)
|
||||
_ = deepReader.Close()
|
||||
if err != nil {
|
||||
log.Printf("[WARN] deep answer stream emit error chat=%s err=%v", flowState.ConversationID, err)
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
deepText = strings.TrimSpace(deepText)
|
||||
if deepText == "" {
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
// 4. 完整回复写入 history。
|
||||
msg := schema.AssistantMessage(deepText, nil)
|
||||
conversationContext.AppendHistory(msg)
|
||||
persistVisibleAssistantMessage(ctx, input.PersistVisibleMessage, flowState, msg)
|
||||
|
||||
flowState.Phase = agentmodel.PhaseChatting
|
||||
return nil
|
||||
}
|
||||
|
||||
// handleRoutePlanStream 处理规划路由：推送状态确认 → 设 PhasePlanning。
|
||||
func handleRoutePlanStream(
|
||||
reader llmservice.StreamReader,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
flowState *agentmodel.CommonState,
|
||||
effectiveThinking bool,
|
||||
speak string,
|
||||
) error {
|
||||
// 关闭路由流。
|
||||
_ = reader.Close()
|
||||
|
||||
if strings.TrimSpace(speak) == "" {
|
||||
speak = "好的，让我来规划一下。"
|
||||
}
|
||||
|
||||
_ = emitter.EmitStatus(chatStatusBlockID, chatStageName, "planning", speak, false)
|
||||
|
||||
flowState.Phase = agentmodel.PhasePlanning
|
||||
return nil
|
||||
}
|
||||
|
||||
// ─── 恢复处理（保持原有逻辑不变）───
|
||||
|
||||
// handleChatResume 处理 pending interaction 恢复。
|
||||
//
|
||||
// 职责边界：
|
||||
// 1. 只做状态传递：吞掉用户输入、写回历史、恢复 phase；
|
||||
// 2. 不生成 speak，真正的回复由下游 Plan / Execute 节点产出；
|
||||
// 3. 只推送轻量 status 通知前端"已收到回复，正在继续"。
|
||||
func handleChatResume(
|
||||
input ChatNodeInput,
|
||||
runtimeState *agentmodel.AgentRuntimeState,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
) error {
|
||||
pending := runtimeState.PendingInteraction
|
||||
flowState := runtimeState.EnsureCommonState()
|
||||
|
||||
if isMismatchedResumeInteraction(input.ResumeInteractionID, pending) {
|
||||
_ = emitter.EmitStatus(
|
||||
chatStatusBlockID, chatStageName,
|
||||
"stale_resume", "当前确认已过期，请刷新后重试。", false,
|
||||
)
|
||||
return nil
|
||||
}
|
||||
|
||||
// 用户输入在 service 层进入 graph 前已经统一追加到 ConversationContext。
|
||||
// 这里不再二次写入，避免 pending 恢复路径把同一轮 user message 追加两次。
|
||||
|
||||
switch pending.Type {
|
||||
case agentmodel.PendingInteractionTypeAskUser:
|
||||
// 用户回答了问题 → 恢复 phase，交给下游节点继续。
|
||||
runtimeState.ResumeFromPending()
|
||||
_ = emitter.EmitStatus(
|
||||
chatStatusBlockID, chatStageName,
|
||||
"resumed", "收到回复，继续处理。", false,
|
||||
)
|
||||
return nil
|
||||
|
||||
case agentmodel.PendingInteractionTypeConfirm:
|
||||
return handleConfirmResume(input, runtimeState, flowState, pending, emitter)
|
||||
|
||||
default:
|
||||
// connection_lost 等其他类型 → 直接恢复。
|
||||
runtimeState.ResumeFromPending()
|
||||
return nil
|
||||
}
|
||||
}
|
||||
|
||||
// handleConfirmResume 处理 confirm 类型恢复。
|
||||
//
|
||||
// 分支逻辑：
|
||||
// 1. accept → 恢复后 phase 设为 executing，下游 Execute 节点接管；
|
||||
// 2. reject + 有 PendingTool（工具确认）→ 回到 executing 让 Execute 节点换策略；
|
||||
// 3. reject + 无 PendingTool（计划确认）→ 清空计划，回到 planning 重新规划。
|
||||
func handleConfirmResume(
|
||||
input ChatNodeInput,
|
||||
runtimeState *agentmodel.AgentRuntimeState,
|
||||
flowState *agentmodel.CommonState,
|
||||
pending *agentmodel.PendingInteraction,
|
||||
emitter *agentstream.ChunkEmitter,
|
||||
) error {
|
||||
if isMismatchedResumeInteraction(input.ResumeInteractionID, pending) {
|
||||
_ = emitter.EmitStatus(
|
||||
chatStatusBlockID, chatStageName,
|
||||
"stale_resume", "当前确认已过期，请刷新后重试。", false,
|
||||
)
|
||||
return nil
|
||||
}
|
||||
|
||||
action := strings.ToLower(strings.TrimSpace(input.ConfirmAction))
|
||||
|
||||
switch action {
|
||||
case "accept", "approve":
|
||||
// 恢复前保存待执行工具，Execute 节点需要它。
|
||||
pendingTool := pending.PendingTool
|
||||
runtimeState.ResumeFromPending()
|
||||
// 将待执行工具放回临时邮箱，供 Execute 节点执行。
|
||||
if pendingTool != nil {
|
||||
copied := *pendingTool
|
||||
runtimeState.PendingConfirmTool = &copied
|
||||
}
|
||||
flowState.Phase = agentmodel.PhaseExecuting
|
||||
_ = emitter.EmitStatus(
|
||||
chatStatusBlockID, chatStageName,
|
||||
"confirmed", "已确认，开始执行。", false,
|
||||
)
|
||||
|
||||
case "reject", "cancel":
|
||||
runtimeState.ResumeFromPending()
|
||||
if pending.PendingTool != nil {
|
||||
// 工具确认被拒 → 回到 executing 换策略。
|
||||
flowState.Phase = agentmodel.PhaseExecuting
|
||||
} else {
|
||||
// 计划确认被拒 → 清空计划，回到 planning。
|
||||
flowState.RejectPlan()
|
||||
}
|
||||
_ = emitter.EmitStatus(
|
||||
chatStatusBlockID, chatStageName,
|
||||
"rejected", "已取消，准备重新规划。", false,
|
||||
)
|
||||
|
||||
default:
|
||||
_ = emitter.EmitStatus(
|
||||
chatStatusBlockID, chatStageName,
|
||||
"invalid_confirm_action", "未识别确认动作，请重试。", false,
|
||||
)
|
||||
}
|
||||
return nil
|
||||
}
|
||||
|
||||
func isMismatchedResumeInteraction(resumeInteractionID string, pending *agentmodel.PendingInteraction) bool {
|
||||
if pending == nil {
|
||||
return false
|
||||
}
|
||||
resumeID := strings.TrimSpace(resumeInteractionID)
|
||||
pendingID := strings.TrimSpace(pending.InteractionID)
|
||||
if resumeID == "" || pendingID == "" {
|
||||
return false
|
||||
}
|
||||
return resumeID != pendingID
|
||||
}
|
||||
|
||||
// prepareChatNodeInput 校验并准备聊天节点的运行态依赖。
|
||||
func prepareChatNodeInput(input ChatNodeInput) (
|
||||
*agentmodel.AgentRuntimeState,
|
||||
*agentmodel.ConversationContext,
|
||||
*agentstream.ChunkEmitter,
|
||||
error,
|
||||
) {
|
||||
if input.RuntimeState == nil {
|
||||
return nil, nil, nil, fmt.Errorf("chat node: runtime state 不能为空")
|
||||
}
|
||||
if input.Client == nil {
|
||||
return nil, nil, nil, fmt.Errorf("chat node: chat client 未注入")
|
||||
}
|
||||
|
||||
input.RuntimeState.EnsureCommonState()
|
||||
if input.ConversationContext == nil {
|
||||
input.ConversationContext = agentmodel.NewConversationContext("")
|
||||
}
|
||||
if input.ChunkEmitter == nil {
|
||||
input.ChunkEmitter = agentstream.NewChunkEmitter(
|
||||
agentstream.NoopPayloadEmitter(), "", "", time.Now().Unix(),
|
||||
)
|
||||
}
|
||||
return input.RuntimeState, input.ConversationContext, input.ChunkEmitter, nil
|
||||
}
|
||||
Reference in New Issue
Block a user