Version: 0.9.26.dev.260417
后端: 1. Prompt 层从 execute 专属骨架重构为全节点统一四段式 buildUnifiedStageMessages - 新增 unified_context.go:定义 StageMessagesConfig + buildUnifiedStageMessages 统一骨架,所有节点(Chat/Plan/Execute/Deliver/DeepAnswer)共用同一套 msg0~msg3 拼装逻辑 - 新增 conversation_view.go:通用对话历史渲染 buildConversationHistoryMessage,各节点复用,不再各自维护提取逻辑 - 新增 chat_context.go / plan_context.go / deliver_context.go:各节点自行渲染 msg1(对话视图)和 msg2(工作区),统一层只负责"怎么拼",不再替节点决定"放什么" - Chat/Plan/Deliver/Execute 的 BuildXXXMessages 全部从 buildStageMessages 切到 buildUnifiedStageMessages,移除旧路径 - 删除 execute_pinned.go:execute 记忆渲染合并到统一层 renderUnifiedMemoryContext - Plan prompt 不再在 user prompt 中拼装任务类 ID 列表和 renderStateSummary,改为依赖 msg2 规划工作区;Chat 粗排判断从"上下文有任务类 ID"改为"批量调度需求" - Deliver prompt 新增 IsAborted/IsExhaustedTerminal 区分,支持粗排收口和主动终止场景 2. Execute ReAct 上下文简化——移除归档搬运、窗口裁剪和重复工具压缩 - 移除 splitExecuteLoopRecordsByBoundary、findLatestExecuteBoundaryMarker、tailExecuteLoops、compressExecuteLoopObservationsByTool、buildEarlyExecuteReactSummary、trimExecuteMessage1ByBudget 等六个函数 - 移除 executeLoopWindowLimit / executeConversationTurnLimit / executeMessage1MaxRunes 等预算常量 - msg1 不再从历史中归档上一轮 ReAct 结果,只保留真实对话流(user + assistant speak),全量注入 - msg2 不再按 loop_closed / step_advanced 边界切分"归档/活跃",直接全量注入全部 ReAct Loop 记录 - token 预算由统一压缩层兜底,prompt 层不再做提前裁剪 3. 压缩层从 Execute 专属提升为全节点通用 UnifiedCompact - 删除 execute_compact.go(Execute 专属压缩文件) - 新增 unified_compact.go:UnifiedCompactInput 参数化,各节点(Plan/Chat/Deliver/Execute)构造时从自己的 NodeInput 提取公共字段,消除对 Execute 的直接依赖 - CompactionStore 接口扩展 LoadStageCompaction / SaveStageCompaction,各节点按 stageKey 独立维护压缩状态互不覆盖 - 非 4 段式消息时退化成按角色汇总统计,确保 context_token_stats 仍然刷新 4. Retry 重试机制全面下线 - dao/agent.go:saveChatHistoryCore / SaveChatHistory / SaveChatHistoryInTx 移除 retry_group_id / retry_index / retry_from_user_message_id / retry_from_assistant_message_id 四个参数,修复乱码注释 - dao/agent-cache.go:移除 ApplyRetrySeed 和 extractMessageHistoryID 两个方法 - conv/agent.go:ToEinoMessages 不再回灌 retry_* 字段到运行期上下文 - service/agentsvc/agent.go:移除 chatRetryMeta 及 resolveRetryGroupID / buildRetrySeed 等全部重试逻辑 - service/agentsvc/agent_quick_note.go:整个文件删除(retry 快速补写路径已无用) - service/events/chat_history_persist.go:移除 retry 参数传递 5. 节点层瘦身 + 可见消息逐条持久化 - agent_nodes.go 大幅简化:Chat/Plan/Execute/Deliver 节点方法移除 ToolSchema 注入、状态摘要渲染等逻辑,只做参数转发和状态落盘 - 新增 visible_message.go:persistVisibleAssistantMessage 统一处理可见 assistant speak 的实时持久化,失败仅记日志不中断主流程 - 新增 llm_debug.go:logNodeLLMContext 统一打印 LLM 上下文调试日志 - graph_run_state.go 新增 PersistVisibleMessageFunc 类型 + AgentGraphDeps.PersistVisibleMessage 字段 - service/agentsvc/agent_newagent.go 精简主循环,注入 PersistVisibleMessage 回调;agent_history.go 精简历史构建 - token_budget.go 移除 Execute 专属预算检查,统一到通用预算 前端: 1. 移除 retry 相关 UI 和类型 - agent.ts 移除 retry_group_id / retry_index / retry_total 字段及 normalize 逻辑 - AssistantPanel.vue 移除 retry 相关 UI 和交互代码(约 700 行精简) - dashboard.ts 移除 retry 相关类型定义 - AssistantView.vue 微调 2. ContextWindowMeter 压缩次数展示和数值格式优化 - 新增 formatCompactCount 工具函数,千位以上用 k 单位压缩(如 80k) - 新增压缩次数显示 3.修复了新对话发消息时,user和assistant消息被自动调换的bug 仓库:无
This commit is contained in:
@@ -12,20 +12,11 @@ import (
|
||||
)
|
||||
|
||||
const (
|
||||
// executeHistoryKindKey 用于在 history 中打运行态标记,供 prompt 分层识别。
|
||||
// 说明:loop_closed / step_advanced 等边界标记仍由节点层写入,但 prompt 层已不再消费它们——
|
||||
// 因为 msg1/msg2 已经按"真实对话流 + 当前活跃 ReAct 记录"重构,不再做 msg2→msg1 的归档搬运。
|
||||
executeHistoryKindKey = "newagent_history_kind"
|
||||
executeHistoryKindCorrectionUser = "llm_correction_prompt"
|
||||
executeHistoryKindLoopClosed = "execute_loop_closed"
|
||||
executeHistoryKindStepAdvanced = "execute_step_advanced"
|
||||
|
||||
// executeLoopWindowLimit 控制当轮 ReAct Loop 窗口最多保留多少条记录。
|
||||
executeLoopWindowLimit = 8
|
||||
|
||||
// executeTrimmedObservationText 是重复工具压缩后的 observation 占位文案。
|
||||
executeTrimmedObservationText = "当前工具调用结果已经被使用过,当前无需使用,为节省上下文空间,已折叠"
|
||||
|
||||
// executeConversationTurnLimit 控制 msg1 注入的最大对话轮数(user + assistant speak)。
|
||||
// 超出时保留最近的条目,早期部分由 ReAct 摘要兜底。
|
||||
executeConversationTurnLimit = 30
|
||||
)
|
||||
|
||||
type executeToolSchemaDoc struct {
|
||||
@@ -40,8 +31,6 @@ type executeLoopRecord struct {
|
||||
Observation string
|
||||
}
|
||||
|
||||
const executeMessage1MaxRunes = 1400
|
||||
|
||||
// buildExecuteStageMessages 组装 execute 阶段 4 条消息骨架。
|
||||
//
|
||||
// 消息结构(固定):
|
||||
@@ -82,87 +71,24 @@ func buildExecuteMessage0(stageSystemPrompt string, ctx *newagentmodel.Conversat
|
||||
return base + "\n\n" + toolCatalog
|
||||
}
|
||||
|
||||
// splitExecuteLoopRecordsByBoundary 按已收口标记拆分归档/活跃 ReAct 记录。
|
||||
//
|
||||
// 规则:
|
||||
// 1. 标记之前的记录归档到 msg1;
|
||||
// 2. 标记之后的记录作为活跃 loop 进入 msg2;
|
||||
// 3. 若没有标记,则全部视为活跃记录(兼容旧会话快照)。
|
||||
func splitExecuteLoopRecordsByBoundary(history []*schema.Message) (archived []executeLoopRecord, active []executeLoopRecord) {
|
||||
if len(history) == 0 {
|
||||
return nil, nil
|
||||
}
|
||||
|
||||
boundary := findLatestExecuteBoundaryMarker(history)
|
||||
if boundary < 0 {
|
||||
return nil, collectExecuteLoopRecords(history)
|
||||
}
|
||||
|
||||
if boundary > 0 {
|
||||
archived = collectExecuteLoopRecords(history[:boundary])
|
||||
}
|
||||
if boundary+1 < len(history) {
|
||||
active = collectExecuteLoopRecords(history[boundary+1:])
|
||||
}
|
||||
return archived, active
|
||||
}
|
||||
|
||||
func findLatestExecuteBoundaryMarker(history []*schema.Message) int {
|
||||
for i := len(history) - 1; i >= 0; i-- {
|
||||
msg := history[i]
|
||||
if msg == nil || msg.Extra == nil {
|
||||
continue
|
||||
}
|
||||
kind, ok := msg.Extra[executeHistoryKindKey].(string)
|
||||
if !ok {
|
||||
continue
|
||||
}
|
||||
switch strings.TrimSpace(kind) {
|
||||
case executeHistoryKindLoopClosed, executeHistoryKindStepAdvanced:
|
||||
return i
|
||||
}
|
||||
}
|
||||
return -1
|
||||
}
|
||||
|
||||
func trimExecuteMessage1ByBudget(content string) string {
|
||||
content = strings.TrimSpace(content)
|
||||
if content == "" {
|
||||
return ""
|
||||
}
|
||||
runes := []rune(content)
|
||||
if len(runes) <= executeMessage1MaxRunes {
|
||||
return content
|
||||
}
|
||||
if executeMessage1MaxRunes <= 3 {
|
||||
return string(runes[:executeMessage1MaxRunes])
|
||||
}
|
||||
return string(runes[:executeMessage1MaxRunes-3]) + "..."
|
||||
}
|
||||
|
||||
// buildExecuteMessage1V3 负责把真实对话流 + 上一轮 loop 归档并入 msg1,并统一做长度裁剪。
|
||||
// buildExecuteMessage1V3 只渲染"真实对话流 + 阶段锚点"。
|
||||
//
|
||||
// 改造说明:
|
||||
// 1. msg1 从人工提炼的摘要变为真实对话流,只注入 user + assistant speak;
|
||||
// 2. tool_call / observation 不在 msg1 中重复(已由 msg2 承载);
|
||||
// 3. 超出 executeConversationTurnLimit 的早期对话不注入,由 ReAct 摘要兜底。
|
||||
// 1. msg1 只保留 user + assistant speak 组成的真实对话历史,全量注入;
|
||||
// 2. tool_call / observation 一律由 msg2 承载,这里不再重复;
|
||||
// 3. 不再从历史中"归档"上一轮 ReAct 结果到 msg1——归档搬运逻辑已随 splitExecuteLoopRecordsByBoundary 一并移除;
|
||||
// 4. token 预算由统一压缩层兜底,prompt 层不做提前裁剪。
|
||||
func buildExecuteMessage1V3(ctx *newagentmodel.ConversationContext) string {
|
||||
lines := []string{"历史上下文:"}
|
||||
if ctx == nil {
|
||||
lines = append(lines,
|
||||
"- 对话历史:暂无。",
|
||||
"- 阶段锚点:按当前工具事实推进执行。",
|
||||
"- 历史归档 ReAct 摘要:暂无。",
|
||||
"- 历史归档 ReAct 窗口:暂无。",
|
||||
"- 当前循环早期摘要:暂无。",
|
||||
)
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
history := ctx.HistorySnapshot()
|
||||
|
||||
// 注入真实对话流(user + assistant speak),全量放入,不再限制轮数和单条长度。
|
||||
turns := collectExecuteConversationTurns(history)
|
||||
turns := collectExecuteConversationTurns(ctx.HistorySnapshot())
|
||||
if len(turns) == 0 {
|
||||
lines = append(lines, "- 对话历史:暂无。")
|
||||
} else {
|
||||
@@ -180,16 +106,15 @@ func buildExecuteMessage1V3(ctx *newagentmodel.ConversationContext) string {
|
||||
lines = append(lines, "- 阶段锚点:按当前工具事实推进,不做无依据操作。")
|
||||
}
|
||||
|
||||
archivedLoops, activeLoops := splitExecuteLoopRecordsByBoundary(history)
|
||||
lines = append(lines, "- 历史归档 ReAct 摘要:"+buildEarlyExecuteReactSummary(archivedLoops, executeLoopWindowLimit))
|
||||
lines = append(lines, renderArchivedExecuteLoopWindowForMessage1V3(archivedLoops))
|
||||
lines = append(lines, "- 当前循环早期摘要:"+buildEarlyExecuteReactSummary(activeLoops, executeLoopWindowLimit))
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
// buildExecuteMessage2V3 承载当前活跃 loop 的全部记录。
|
||||
// 若是新一轮刚开始(活跃 loop 为空),明确返回已清空状态。
|
||||
// 不再限制窗口大小,token 预算由 execute 层统一管理。
|
||||
// buildExecuteMessage2V3 承载当前会话中全部 ReAct Loop 记录。
|
||||
//
|
||||
// 改造说明:
|
||||
// 1. 不再按 execute_loop_closed / execute_step_advanced 边界切分"归档/活跃"两段;
|
||||
// 2. 直接从 history 提取全部 assistant tool_call + 对应 observation 作为当前 Loop 视图;
|
||||
// 3. 新一轮刚开始(尚未产生 tool_call)时返回明确占位,方便模型识别"干净起点"。
|
||||
func buildExecuteMessage2V3(ctx *newagentmodel.ConversationContext) string {
|
||||
lines := []string{"当轮 ReAct Loop 记录:"}
|
||||
if ctx == nil {
|
||||
@@ -197,31 +122,13 @@ func buildExecuteMessage2V3(ctx *newagentmodel.ConversationContext) string {
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
_, activeLoops := splitExecuteLoopRecordsByBoundary(ctx.HistorySnapshot())
|
||||
if len(activeLoops) == 0 {
|
||||
loops := collectExecuteLoopRecords(ctx.HistorySnapshot())
|
||||
if len(loops) == 0 {
|
||||
lines = append(lines, "- 已清空(新一轮 loop 准备中)。")
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
// 全量放入,不再限制窗口大小
|
||||
for i, loop := range activeLoops {
|
||||
lines = append(lines, fmt.Sprintf("%d) thought/reason:%s", i+1, loop.Thought))
|
||||
lines = append(lines, fmt.Sprintf(" tool_call:%s", renderExecuteToolCallText(loop.ToolName, loop.ToolArgs)))
|
||||
lines = append(lines, fmt.Sprintf(" observation:%s", loop.Observation))
|
||||
}
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
func renderArchivedExecuteLoopWindowForMessage1V3(records []executeLoopRecord) string {
|
||||
if len(records) == 0 {
|
||||
return "- 历史归档 ReAct 窗口:暂无。"
|
||||
}
|
||||
|
||||
windowLoops := tailExecuteLoops(records, executeLoopWindowLimit)
|
||||
windowLoops = compressExecuteLoopObservationsByTool(windowLoops)
|
||||
|
||||
lines := []string{"历史归档 ReAct 窗口(由上一轮 msg2 并入):"}
|
||||
for i, loop := range windowLoops {
|
||||
for i, loop := range loops {
|
||||
lines = append(lines, fmt.Sprintf("%d) thought/reason:%s", i+1, loop.Thought))
|
||||
lines = append(lines, fmt.Sprintf(" tool_call:%s", renderExecuteToolCallText(loop.ToolName, loop.ToolArgs)))
|
||||
lines = append(lines, fmt.Sprintf(" observation:%s", loop.Observation))
|
||||
@@ -525,51 +432,6 @@ func findExecuteThoughtBefore(history []*schema.Message, index int) string {
|
||||
return "(未记录)"
|
||||
}
|
||||
|
||||
func tailExecuteLoops(records []executeLoopRecord, limit int) []executeLoopRecord {
|
||||
if len(records) == 0 {
|
||||
return nil
|
||||
}
|
||||
if limit <= 0 || len(records) <= limit {
|
||||
result := make([]executeLoopRecord, len(records))
|
||||
copy(result, records)
|
||||
return result
|
||||
}
|
||||
result := make([]executeLoopRecord, limit)
|
||||
copy(result, records[len(records)-limit:])
|
||||
return result
|
||||
}
|
||||
|
||||
// compressExecuteLoopObservationsByTool 对窗口内重复工具做 observation 压缩。
|
||||
func compressExecuteLoopObservationsByTool(records []executeLoopRecord) []executeLoopRecord {
|
||||
if len(records) == 0 {
|
||||
return records
|
||||
}
|
||||
|
||||
latestIndexByTool := make(map[string]int, len(records))
|
||||
for i := len(records) - 1; i >= 0; i-- {
|
||||
key := strings.ToLower(strings.TrimSpace(records[i].ToolName))
|
||||
if key == "" {
|
||||
key = "unknown_tool"
|
||||
}
|
||||
if _, exists := latestIndexByTool[key]; !exists {
|
||||
latestIndexByTool[key] = i
|
||||
}
|
||||
}
|
||||
|
||||
result := make([]executeLoopRecord, len(records))
|
||||
copy(result, records)
|
||||
for i := range result {
|
||||
key := strings.ToLower(strings.TrimSpace(result[i].ToolName))
|
||||
if key == "" {
|
||||
key = "unknown_tool"
|
||||
}
|
||||
if latestIndexByTool[key] != i {
|
||||
result[i].Observation = executeTrimmedObservationText
|
||||
}
|
||||
}
|
||||
return result
|
||||
}
|
||||
|
||||
func renderExecuteToolCallText(toolName, toolArgs string) string {
|
||||
toolName = strings.TrimSpace(toolName)
|
||||
if toolName == "" {
|
||||
@@ -582,38 +444,6 @@ func renderExecuteToolCallText(toolName, toolArgs string) string {
|
||||
return toolName + "(" + toolArgs + ")"
|
||||
}
|
||||
|
||||
func buildEarlyExecuteReactSummary(records []executeLoopRecord, windowLimit int) string {
|
||||
if len(records) == 0 {
|
||||
return "暂无。"
|
||||
}
|
||||
if len(records) <= windowLimit {
|
||||
return "无(当前窗口已覆盖全部 ReAct 记录)。"
|
||||
}
|
||||
|
||||
early := records[:len(records)-windowLimit]
|
||||
toolCounts := make(map[string]int, len(early))
|
||||
for _, record := range early {
|
||||
key := strings.TrimSpace(record.ToolName)
|
||||
if key == "" {
|
||||
key = "unknown_tool"
|
||||
}
|
||||
toolCounts[key]++
|
||||
}
|
||||
|
||||
names := make([]string, 0, len(toolCounts))
|
||||
for name := range toolCounts {
|
||||
names = append(names, name)
|
||||
}
|
||||
sort.Strings(names)
|
||||
|
||||
parts := make([]string, 0, len(names))
|
||||
for _, name := range names {
|
||||
parts = append(parts, fmt.Sprintf("%s×%d", name, toolCounts[name]))
|
||||
}
|
||||
|
||||
return fmt.Sprintf("已折叠 %d 条旧记录,涉及:%s。", len(early), strings.Join(parts, "、"))
|
||||
}
|
||||
|
||||
func hasExecuteRoughBuildDone(ctx *newagentmodel.ConversationContext) bool {
|
||||
if ctx == nil {
|
||||
return false
|
||||
@@ -725,3 +555,12 @@ func renderExecuteTaskClassIDs(state *newagentmodel.CommonState) string {
|
||||
}
|
||||
return fmt.Sprintf("task_class_ids=[%s]", strings.Join(parts, ","))
|
||||
}
|
||||
|
||||
// renderExecuteMemoryContext 提取 execute 阶段要注入 msg3 的记忆文本。
|
||||
//
|
||||
// 1. 只读取统一的 memory_context,避免把其他 pinned block 误塞进 prompt。
|
||||
// 2. 为空时直接返回空串,保持 msg3 干净。
|
||||
// 3. 复用统一记忆渲染逻辑,保证各阶段记忆入口一致。
|
||||
func renderExecuteMemoryContext(ctx *newagentmodel.ConversationContext) string {
|
||||
return renderUnifiedMemoryContext(ctx)
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user