Version: 0.9.11.dev.260409
后端： 1. conv 并行迁移与切流接线（旧目录下沉到 newAgent/conv） - 新建 newAgent/conv/schedule_provider.go、schedule_state.go、schedule_preview.go、schedule_persist.go，保持原有排程转换/预览/持久化能力； - 删除旧目录 conv/schedule_provider.go、schedule_state.go、schedule_preview.go、schedule_persist.go； - 更新 cmd/start.go 与 service/agentsvc/agent_newagent.go，ScheduleProvider/SchedulePersistor 与 preview 转换统一切到 newAgent/conv； - 删除旧 conv/schedule_state_test.go（迁移期测试文件清理）。 2. execute 循环上下文收口增强（历史归档 + 当前轮清晰化） - 更新 node/chat.go：仅在 completed 收口时写 execute_loop_closed marker，供后续 prompt 分层归档； - 更新 prompt/execute_context.go：msg1/msg2 升级为 V3，按收口标记拆分“历史归档 loop / 当前活跃 loop”，并增加 msg1 长度预算裁剪； - 更新 node/execute.go：新增 execute 置顶上下文同步（execution_context/current_step），在轮次开始与 next_plan 后即时刷新； - 更新 prompt/execute.go + execute_context.go：补齐“当前计划步骤 + done_when”强约束，禁止未达成判定时提前 next_plan。 3. 图路由与执行策略微调 - 更新 graph/common_graph.go：Plan/Confirm 分支允许直接进入 Deliver 收口； - 更新 node/plan.go：always_execute 链路下补发计划摘要并写入历史，保证自动执行与手动确认文案一致； - 更新 model/common_state.go：DefaultMaxRounds 从 30 提升到 60。 4. 复合工具规划器重构（去重实现，复用 logic 公共能力） - 更新 tools/compound_tools.go：min_context_switch / spread_even 改为调用 backend/logic 规划器（PlanMinContextSwitchMoves / PlanEvenSpreadMoves）； - 新增 state_id↔logic_id 映射层，统一入参与回填，避免工具层与规划层 ID 语义耦合； - 删除 compound_tools 内部重复的规划/归一化/分组/打分实现，减少第三份复制逻辑。 5. 同步调试与文档 - 更新 newAgent/Log.txt 调试日志； - 新增 memory/记忆模块实施计划.md（面试优先版到产品可用版的落地路线）。 前端：无 仓库：无
This commit is contained in:
Losita
@@ -14,6 +14,7 @@ import (
|
||||
const (
|
||||
executeHistoryKindKey = "newagent_history_kind"
|
||||
executeHistoryKindCorrectionUser = "llm_correction_prompt"
|
||||
executeHistoryKindLoopClosed = "execute_loop_closed"
|
||||
|
||||
// executeLoopWindowLimit 控制“当轮 ReAct Loop 窗口”最多保留多少条记录。
|
||||
// 采用固定窗口能避免上下文无上限增长，且可保持“最近行为”可追踪。
|
||||
@@ -36,6 +37,8 @@ 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）：
|
||||
@@ -50,8 +53,8 @@ func buildExecuteStageMessages(
|
||||
runtimeUserPrompt string,
|
||||
) []*schema.Message {
|
||||
msg0 := buildExecuteMessage0(stageSystemPrompt, ctx)
|
||||
msg1 := buildExecuteMessage1(ctx)
|
||||
msg2 := buildExecuteMessage2(ctx)
|
||||
msg1 := buildExecuteMessage1V3(ctx)
|
||||
msg2 := buildExecuteMessage2V3(ctx)
|
||||
msg3 := buildExecuteMessage3(state, ctx, runtimeUserPrompt)
|
||||
|
||||
return []*schema.Message{
|
||||
@@ -140,6 +143,209 @@ func buildExecuteMessage2(ctx *newagentmodel.ConversationContext) string {
|
||||
}
|
||||
|
||||
// buildExecuteMessage3 生成当前执行状态与执行锚点。
|
||||
// buildExecuteMessage1V2 生成历史摘要：
|
||||
// 1. 已收口的 loop 归档到 msg1；
|
||||
// 2. 当前活跃 loop 只保留“早期摘要”；
|
||||
// 3. 最终对 msg1 做统一长度裁剪，控制 token 开销。
|
||||
func buildExecuteMessage1V2(ctx *newagentmodel.ConversationContext) string {
|
||||
lines := []string{"历史上下文（仅供参考）："}
|
||||
if ctx == nil {
|
||||
lines = append(lines,
|
||||
"- 用户目标：暂无可用历史输入。",
|
||||
"- 阶段锚点：按当前工具事实推进执行。",
|
||||
"- 历史归档 ReAct 摘要：暂无。",
|
||||
"- 当前循环早期摘要：暂无。",
|
||||
)
|
||||
return trimExecuteMessage1ByBudget(strings.Join(lines, "\n"))
|
||||
}
|
||||
|
||||
history := ctx.HistorySnapshot()
|
||||
firstUser, lastUser := pickExecuteUserInputs(history)
|
||||
switch {
|
||||
case firstUser == "":
|
||||
lines = append(lines, "- 用户目标：暂无可用历史输入。")
|
||||
case lastUser != "" && lastUser != firstUser:
|
||||
lines = append(lines, "- 用户目标："+firstUser+"；最近补充："+lastUser)
|
||||
default:
|
||||
lines = append(lines, "- 用户目标："+firstUser)
|
||||
}
|
||||
|
||||
if hasExecuteRoughBuildDone(ctx) {
|
||||
lines = append(lines, "- 阶段锚点：粗排已完成，本轮仅做微调，不重新 place。")
|
||||
} else {
|
||||
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, 0))
|
||||
lines = append(lines, "- 当前循环早期摘要："+buildEarlyExecuteReactSummary(activeLoops, executeLoopWindowLimit))
|
||||
return trimExecuteMessage1ByBudget(strings.Join(lines, "\n"))
|
||||
}
|
||||
|
||||
// buildExecuteMessage2V2 仅展示“当前活跃 loop”的窗口记录。
|
||||
func buildExecuteMessage2V2(ctx *newagentmodel.ConversationContext) string {
|
||||
lines := []string{"当轮 ReAct Loop 记录（窗口）："}
|
||||
if ctx == nil {
|
||||
lines = append(lines, "- 暂无可用 ReAct 记录。")
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
_, activeLoops := splitExecuteLoopRecordsByBoundary(ctx.HistorySnapshot())
|
||||
if len(activeLoops) == 0 {
|
||||
lines = append(lines, "- 暂无可用 ReAct 记录。")
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
windowLoops := tailExecuteLoops(activeLoops, executeLoopWindowLimit)
|
||||
windowLoops = compressExecuteLoopObservationsByTool(windowLoops)
|
||||
for i, loop := range windowLoops {
|
||||
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")
|
||||
}
|
||||
|
||||
// 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 := findLatestExecuteLoopClosedMarker(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 findLatestExecuteLoopClosedMarker(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
|
||||
}
|
||||
if strings.TrimSpace(kind) == executeHistoryKindLoopClosed {
|
||||
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“上一轮 loop 归档”并入 msg1，并统一做长度裁剪。
|
||||
func buildExecuteMessage1V3(ctx *newagentmodel.ConversationContext) string {
|
||||
lines := []string{"历史上下文（仅供参考）："}
|
||||
if ctx == nil {
|
||||
lines = append(lines,
|
||||
"- 用户目标：暂无可用历史输入。",
|
||||
"- 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 trimExecuteMessage1ByBudget(strings.Join(lines, "\n"))
|
||||
}
|
||||
|
||||
history := ctx.HistorySnapshot()
|
||||
firstUser, lastUser := pickExecuteUserInputs(history)
|
||||
switch {
|
||||
case firstUser == "":
|
||||
lines = append(lines, "- 用户目标：暂无可用历史输入。")
|
||||
case lastUser != "" && lastUser != firstUser:
|
||||
lines = append(lines, "- 用户目标："+firstUser+"；最近补充："+lastUser)
|
||||
default:
|
||||
lines = append(lines, "- 用户目标："+firstUser)
|
||||
}
|
||||
|
||||
if hasExecuteRoughBuildDone(ctx) {
|
||||
lines = append(lines, "- 阶段锚点：粗排已完成，本轮仅做微调，不重新 place。")
|
||||
} else {
|
||||
lines = append(lines, "- 阶段锚点：按当前工具事实推进，不做无依据操作。")
|
||||
}
|
||||
|
||||
// 1. 通过收口标记拆分“归档 loop / 当前活跃 loop”。
|
||||
// 2. 归档 loop 的窗口条目直接并入 msg1，满足“上一轮 msg2 挪入 msg1”。
|
||||
// 3. 当前活跃 loop 在 msg1 只保留早期摘要，详细窗口交给 msg2。
|
||||
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 trimExecuteMessage1ByBudget(strings.Join(lines, "\n"))
|
||||
}
|
||||
|
||||
// buildExecuteMessage2V3 仅承载“当前活跃 loop”的窗口。
|
||||
// 若是新一轮刚开始（活跃 loop 为空），明确返回“已清空”状态。
|
||||
func buildExecuteMessage2V3(ctx *newagentmodel.ConversationContext) string {
|
||||
lines := []string{"当轮 ReAct Loop 记录（窗口）："}
|
||||
if ctx == nil {
|
||||
lines = append(lines, "- 暂无可用 ReAct 记录。")
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
_, activeLoops := splitExecuteLoopRecordsByBoundary(ctx.HistorySnapshot())
|
||||
if len(activeLoops) == 0 {
|
||||
lines = append(lines, "- 已清空（新一轮 loop 准备中）。")
|
||||
return strings.Join(lines, "\n")
|
||||
}
|
||||
|
||||
windowLoops := tailExecuteLoops(activeLoops, executeLoopWindowLimit)
|
||||
windowLoops = compressExecuteLoopObservationsByTool(windowLoops)
|
||||
for i, loop := range windowLoops {
|
||||
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 {
|
||||
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 buildExecuteMessage3(state *newagentmodel.CommonState, ctx *newagentmodel.ConversationContext, runtimeUserPrompt string) string {
|
||||
lines := []string{"当前执行状态："}
|
||||
|
||||
@@ -169,6 +375,32 @@ func buildExecuteMessage3(state *newagentmodel.CommonState, ctx *newagentmodel.C
|
||||
if initialGoal != "" && initialGoal != currentGoal {
|
||||
lines = append(lines, "- 首轮目标来源："+initialGoal)
|
||||
}
|
||||
// 1. 有 plan 时，把当前步骤与完成判定强制写入 msg3。
|
||||
// 2. 该锚点用于约束模型“只推进当前步骤”，避免退化成泛化 ReAct。
|
||||
// 3. 当前步骤不可读时给出兜底指引，避免引用旧步骤。
|
||||
if state != nil && state.HasPlan() {
|
||||
current, total := state.PlanProgress()
|
||||
lines = append(lines, "计划步骤锚点（强约束）：")
|
||||
if step, ok := state.CurrentPlanStep(); ok {
|
||||
stepContent := strings.TrimSpace(step.Content)
|
||||
if stepContent == "" {
|
||||
stepContent = "（当前步骤内容为空）"
|
||||
}
|
||||
doneWhen := strings.TrimSpace(step.DoneWhen)
|
||||
if doneWhen == "" {
|
||||
doneWhen = "（未提供 done_when，需基于步骤目标给出可验证完成证据）"
|
||||
}
|
||||
lines = append(lines, fmt.Sprintf("- 当前步骤：第 %d/%d 步", current, total))
|
||||
lines = append(lines, "- 当前步骤内容："+stepContent)
|
||||
lines = append(lines, "- 当前步骤完成判定(done_when)："+doneWhen)
|
||||
lines = append(lines, "- 动作纪律1：未满足 done_when 时，只能 continue / confirm / ask_user，禁止 next_plan")
|
||||
lines = append(lines, "- 动作纪律2：满足 done_when 时，优先 next_plan，并在 goal_check 对照 done_when 给证据")
|
||||
lines = append(lines, "- 动作纪律3：禁止跳到后续步骤执行")
|
||||
} else {
|
||||
lines = append(lines, "- 当前计划步骤不可读；请先判断是否已完成全部计划")
|
||||
lines = append(lines, "- 若已完成全部计划，输出 done 并给出 goal_check 证据")
|
||||
}
|
||||
}
|
||||
if taskClassText := renderExecuteTaskClassIDs(state); taskClassText != "" {
|
||||
lines = append(lines, "- 目标任务类："+taskClassText)
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user