diff --git a/deps/agent-system-core b/deps/agent-system-core index 1b04a34..d437530 160000 --- a/deps/agent-system-core +++ b/deps/agent-system-core @@ -1 +1 @@ -Subproject commit 1b04a34bbc43143d3a90cea50e277964ba03e029 +Subproject commit d4375300cde71320cdd2a012304603efe9eb5cf9 diff --git a/internal/prompts/library/base-deliverables.md b/internal/prompts/library/base-deliverables.md index 01ef4a7..a3d76e3 100644 --- a/internal/prompts/library/base-deliverables.md +++ b/internal/prompts/library/base-deliverables.md @@ -1,7 +1,7 @@ --- id: base.deliverables layer: base -version: 1 +version: 2 title: 交付物工作流 description: 生成文档/表格/演示稿等文件交付物的通用工作流与配方（常驻基座） --- @@ -11,6 +11,8 @@ description: 生成文档/表格/演示稿等文件交付物的通用工作流 2. **登记**：调用 register_artifact登记生成文件（path 传相对路径）；登记后对话内自动出现可预览的文件卡片。 3. **收尾**：对话回复只给结论摘要与要点（几句话），全文进文件，不要在回复里复述文档内容；也不要只口头说文件名——卡片就是交付。 +没有调用工具、没有登记工件时，禁止说「已生成 / 已保存 / 已导出」。回复里不要写工件 ID（art_…）：登记成功后卡片会自动出现。 + 配方已迁入技能库（启动时种子到数据目录 skills/，完整 Python 配方见对应 SKILL.md）： - **docx-author**：python-docx——标题层级/表格/列表真实排版。 diff --git a/internal/runtime/llm.go b/internal/runtime/llm.go index 54d0838..ab7dabd 100644 --- a/internal/runtime/llm.go +++ b/internal/runtime/llm.go @@ -3,6 +3,7 @@ package runtime import ( "encoding/json" "fmt" + "regexp" "strings" "time" @@ -21,10 +22,27 @@ type LLMRunner struct { Prompts *prompts.Service // Stream 提供方边界注入点（nil = 按模型路由构造 agent-native 客户端）。 Stream agentllm.StreamFunc + // Now 注入时钟（nil = time.Now）。测试用它固定思考与工具耗时。 + Now func() time.Time } func (l *LLMRunner) Name() string { return "llm" } +func (l *LLMRunner) now() time.Time { + if l.Now != nil { + return l.Now() + } + return time.Now() +} + +// durationMs 把 [start, end] 换成毫秒。时钟未打点时返回 0。 +func durationMs(start, end time.Time) float64 { + if start.IsZero() || end.Before(start) { + return 0 + } + return float64(end.Sub(start).Microseconds()) / 1000 +} + const ( toolDelegate = "delegate_task" toolPlan = "update_plan" @@ -245,6 +263,7 @@ func (l *LLMRunner) Execute(rc RunCtx) error { } toolCallsMade := 0 thinkingIdx, textIdx := 0, 0 + deliverableGuardUsed := false // 交付物引用（对话内直接给出）：工具轮产生的工件不进过程折叠区， // 收集后作为内容部件钉在消息尾部（最终正文之后）——回复完给文件卡。 var deliverables []*core.ArtifactRef @@ -259,6 +278,7 @@ func (l *LLMRunner) Execute(rc RunCtx) error { var turnThinking, turnText strings.Builder var calls []agentllm.ToolCall var usage agentllm.Usage + var thinkStart, thinkEnd, textStart, textEnd time.Time ev := stream(rc.Context(), route.Model, msgs, decls, streamOpts) phaseEmitted := false for e := range ev { @@ -268,9 +288,19 @@ func (l *LLMRunner) Execute(rc RunCtx) error { rc.Emit(core.EvRunPhase, core.RunPhasePayload{Phase: core.PhaseResponding}) phaseEmitted = true } + now := l.now() + if textStart.IsZero() { + textStart = now + } + textEnd = now turnText.WriteString(e.Text) rc.Emit(core.EvMessageDelta, core.MessageDeltaPayload{Index: textIdx, Text: e.Text}) case agentllm.KindReasoningDelta: + now := l.now() + if thinkStart.IsZero() { + thinkStart = now + } + thinkEnd = now turnThinking.WriteString(e.ReasoningDelta) rc.Emit(core.EvThinkingDelta, core.ThinkingDeltaPayload{Index: thinkingIdx, Text: e.ReasoningDelta}) case agentllm.KindToolCallDelta: @@ -288,25 +318,44 @@ func (l *LLMRunner) Execute(rc RunCtx) error { // 记录本轮部件。 if turnThinking.Len() > 0 { - finalParts = append(finalParts, core.Part{Type: core.PartThinking, Text: turnThinking.String(), Process: true}) + finalParts = append(finalParts, core.Part{Type: core.PartThinking, Text: turnThinking.String(), Process: true, DurationMs: durationMs(thinkStart, thinkEnd)}) thinkingIdx++ } + turnBody := "" + textPartIdx := -1 if turnText.Len() > 0 { // 工具轮的文本是过程解说（终态随过程折叠）；终轮文本才是对用户的回复。 // 部分模型把推理以 内联混进 content 流：折离为思考部件（过程）， // 保证「最终输出 = 干净的对用户回复」契约。 body, think := splitThinkBlock(turnText.String()) if think != "" { - finalParts = append(finalParts, core.Part{Type: core.PartThinking, Text: think, Process: true}) + inlineDur := durationMs(thinkStart, thinkEnd) + if inlineDur == 0 { + inlineDur = durationMs(textStart, textEnd) + } + finalParts = append(finalParts, core.Part{Type: core.PartThinking, Text: think, Process: true, DurationMs: inlineDur}) thinkingIdx++ } if body != "" { + textPartIdx = len(finalParts) finalParts = append(finalParts, core.Part{Type: core.PartText, Text: body, Process: len(calls) > 0}) + turnBody = body textIdx++ } } if len(calls) == 0 { + // 谎称交付：没产生工件却报文件已生成，或引用了不存在的工件 ID。 + // 把这段正文降为过程，纠正一次，再给模型一轮补做或如实说明的机会。 + if !deliverableGuardUsed && claimsFalseDeliverable(turnBody, deliverables, rc) { + deliverableGuardUsed = true + if textPartIdx >= 0 { + finalParts[textPartIdx].Process = true + } + msgs = append(msgs, agentllm.Message{Role: agentllm.RoleAssistant, Content: turnText.String()}) + msgs = append(msgs, agentllm.Message{Role: agentllm.RoleUser, Content: deliverableGuardNote}) + continue + } // 终轮：无工具调用。 if len(finalParts) == 0 { finalParts = append(finalParts, core.Part{Type: core.PartText, Text: "（模型未返回内容）"}) @@ -329,6 +378,7 @@ func (l *LLMRunner) Execute(rc RunCtx) error { return fmt.Errorf("budget: 工具调用超过上限 %d（profile.budget.maxToolCalls）", maxTools) } rc.Emit(core.EvToolCall, core.ToolCallPayload{CallID: call.ID, Name: call.Name, Arguments: call.Arguments}) + callStart := l.now() var out string var err error @@ -484,13 +534,13 @@ func (l *LLMRunner) Execute(rc RunCtx) error { rc.Emit(core.EvToolResult, core.ToolResultPayload{CallID: call.ID, Status: "ok", Summary: summarize(out, err)}) } - summary := "" + summary := summarize(out, nil) if err != nil { summary = "error: " + err.Error() out = summary } finalParts = append(finalParts, - core.Part{Type: core.PartToolCall, Process: true, ToolCall: &core.ToolCallPart{ID: call.ID, Name: call.Name, Arguments: call.Arguments}}, + core.Part{Type: core.PartToolCall, Process: true, ToolCall: &core.ToolCallPart{ID: call.ID, Name: call.Name, Arguments: call.Arguments, Duration: durationMs(callStart, l.now())}}, core.Part{Type: core.PartToolResult, Process: true, ToolResult: &core.ToolResultPart{CallID: call.ID, Status: statusOf(err), Summary: summary}}, ) if artifactRef != nil { @@ -576,6 +626,42 @@ func orDefault(s, def string) string { return s } +const deliverableGuardNote = "你的上一段回复声称已经生成或登记了文件，但这一轮没有产生对应的工件。请二选一：用 run_python 写出文件并调用 register_artifact 登记，或如实说明这一轮没有生成文件。不要编造工件 ID，也不要在回复里写 art_ 开头的编号。" + +var ( + artIDRe = regexp.MustCompile(`art_[0-9a-f]{12}`) + fileClaimRe = regexp.MustCompile(`(?s)(?:已生成|已保存|已导出|已写入).{0,120}?\.(?:docx|xlsx|pptx|pdf|png|jpe?g|csv|md|txt|json)\b`) +) + +// claimsFalseDeliverable 判断终轮正文是否在没有对应工件的情况下声称交付。 +// 引用本轮已产出、或项目里确实存在的工件，不算谎称。 +func claimsFalseDeliverable(body string, produced []*core.ArtifactRef, rc RunCtx) bool { + if strings.TrimSpace(body) == "" { + return false + } + producedIDs := map[string]bool{} + for _, ref := range produced { + if ref != nil && ref.ArtifactID != "" { + producedIDs[ref.ArtifactID] = true + } + } + knownExisting := false + for _, id := range artIDRe.FindAllString(body, -1) { + if producedIDs[id] { + continue + } + if _, _, ok := rc.Artifact(id); ok { + knownExisting = true + continue + } + return true + } + if len(producedIDs) > 0 || knownExisting { + return false + } + return fileClaimRe.MatchString(body) +} + func summarize(out string, err error) string { if err != nil { return truncateRun(err.Error(), 160) diff --git a/internal/runtime/llm_test.go b/internal/runtime/llm_test.go index b996513..b2572d0 100644 --- a/internal/runtime/llm_test.go +++ b/internal/runtime/llm_test.go @@ -6,6 +6,7 @@ import ( "fmt" "strings" "testing" + "time" agentllm "agent-native/runtime/llm" core "agent-system-core" @@ -620,3 +621,167 @@ func TestLLMRunnerCardScopeProject(t *testing.T) { t.Errorf("项目级卡 ConversationID 应留空, got %q", rc.cards[0].ConversationID) } } + +// TestLLMRunnerRecordsThinkingAndToolDuration 验证：思考部件带上推理增量的时长， +// 工具调用带上执行时长，成功的工具结果也留下摘要（刷新后仍能展开）。 +func TestLLMRunnerRecordsThinkingAndToolDuration(t *testing.T) { + turns := [][]agentllm.Event{ + { + agentllm.Event{Kind: agentllm.KindReasoningDelta, ReasoningDelta: "先"}, + agentllm.Event{Kind: agentllm.KindReasoningDelta, ReasoningDelta: "想想"}, + agentllm.Event{Kind: agentllm.KindMessageEnd, Message: agentllm.Message{ToolCalls: []agentllm.ToolCall{ + {ID: "c1", Name: "update_todos", Arguments: "{}"}, + }}}, + }, + { + agentllm.Event{Kind: agentllm.KindTextDelta, Text: "完成了。"}, + agentllm.Event{Kind: agentllm.KindMessageEnd}, + }, + } + var tick int64 + r := &LLMRunner{ + Stream: scriptStream(turns, nil), + Now: func() time.Time { + tick++ + return time.Unix(0, tick*int64(time.Millisecond)) + }, + } + rc := &fakeRunCtx{ctx: context.Background(), prof: core.AgentProfile{Model: core.ModelRoute{Chat: "test-model"}}} + if err := r.Execute(rc); err != nil { + t.Fatalf("Execute: %v", err) + } + var thinking *core.Part + var call *core.Part + var result *core.Part + for i := range rc.assistant { + p := &rc.assistant[i] + switch p.Type { + case core.PartThinking: + thinking = p + case core.PartToolCall: + call = p + case core.PartToolResult: + result = p + } + } + if thinking == nil || thinking.DurationMs != 1 { + t.Fatalf("thinking durationMs = %v, want 1 (two reasoning deltas, 1ms apart)", thinking) + } + if call == nil || call.ToolCall == nil || call.ToolCall.Duration != 1 { + t.Fatalf("tool durationMs = %+v, want 1", call) + } + if result == nil || result.ToolResult == nil || result.ToolResult.Summary != "待办已更新" { + t.Fatalf("success summary = %+v, want 待办已更新", result) + } +} + +func textParts(parts []core.Part) []core.Part { + var out []core.Part + for _, p := range parts { + if p.Type == core.PartText { + out = append(out, p) + } + } + return out +} + +// TestLLMRunnerFalseDeliverableGuard 验证：终轮声称已生成文件但没有工件时， +// 这段正文降为过程，纠正一次；下一轮如实说明则作为最终回复。 +func TestLLMRunnerFalseDeliverableGuard(t *testing.T) { + turns := [][]agentllm.Event{ + { + agentllm.Event{Kind: agentllm.KindTextDelta, Text: "文档已生成：\n\n研发智能体方案评审_冲突点分析.docx\n工件 ID：art_4f9d2c1a8b3e"}, + agentllm.Event{Kind: agentllm.KindMessageEnd}, + }, + { + agentllm.Event{Kind: agentllm.KindTextDelta, Text: "这一轮没有生成文件。"}, + agentllm.Event{Kind: agentllm.KindMessageEnd}, + }, + } + var calls int + var lastUser string + base := scriptStream(turns, nil) + r := &LLMRunner{Stream: func(ctx context.Context, model string, msgs []agentllm.Message, tools []agentllm.ToolDecl, opts agentllm.StreamOptions) <-chan agentllm.Event { + calls++ + if n := len(msgs); n > 0 && msgs[n-1].Role == agentllm.RoleUser { + lastUser = msgs[n-1].Content + } + return base(ctx, model, msgs, tools, opts) + }} + rc := &fakeRunCtx{ctx: context.Background(), prof: core.AgentProfile{Model: core.ModelRoute{Chat: "test-model"}}} + if err := r.Execute(rc); err != nil { + t.Fatalf("Execute: %v", err) + } + if calls != 2 { + t.Fatalf("stream calls = %d, want 2", calls) + } + if !strings.Contains(lastUser, "register_artifact") { + t.Fatalf("correction missing, got %q", lastUser) + } + texts := textParts(rc.assistant) + if len(texts) != 2 || !texts[0].Process || texts[1].Process { + t.Fatalf("text parts = %+v, want claimed text process then honest reply", texts) + } +} + +// TestLLMRunnerFalseDeliverableGuardOnce 验证：纠正只发生一次，第二次仍谎称则接受为终轮。 +func TestLLMRunnerFalseDeliverableGuardOnce(t *testing.T) { + turns := [][]agentllm.Event{ + { + agentllm.Event{Kind: agentllm.KindTextDelta, Text: "文档已生成：甲.docx"}, + agentllm.Event{Kind: agentllm.KindMessageEnd}, + }, + { + agentllm.Event{Kind: agentllm.KindTextDelta, Text: "文档已保存：乙.docx"}, + agentllm.Event{Kind: agentllm.KindMessageEnd}, + }, + } + var calls int + base := scriptStream(turns, nil) + r := &LLMRunner{Stream: func(ctx context.Context, model string, msgs []agentllm.Message, tools []agentllm.ToolDecl, opts agentllm.StreamOptions) <-chan agentllm.Event { + calls++ + return base(ctx, model, msgs, tools, opts) + }} + rc := &fakeRunCtx{ctx: context.Background(), prof: core.AgentProfile{Model: core.ModelRoute{Chat: "test-model"}}} + if err := r.Execute(rc); err != nil { + t.Fatalf("Execute: %v", err) + } + if calls != 2 { + t.Fatalf("stream calls = %d, want 2 (guard fires once)", calls) + } + texts := textParts(rc.assistant) + if len(texts) != 2 || !texts[0].Process || texts[1].Process { + t.Fatalf("text parts = %+v, want only the first claim marked process", texts) + } +} + +// TestLLMRunnerExistingArtifactCitationNotGuarded 验证：引用项目里已有的工件不算谎称。 +func TestLLMRunnerExistingArtifactCitationNotGuarded(t *testing.T) { + turns := [][]agentllm.Event{ + { + agentllm.Event{Kind: agentllm.KindTextDelta, Text: "上次已生成的报告.docx 在这里：art_aaaaaaaaaaaa"}, + agentllm.Event{Kind: agentllm.KindMessageEnd}, + }, + } + var calls int + base := scriptStream(turns, nil) + r := &LLMRunner{Stream: func(ctx context.Context, model string, msgs []agentllm.Message, tools []agentllm.ToolDecl, opts agentllm.StreamOptions) <-chan agentllm.Event { + calls++ + return base(ctx, model, msgs, tools, opts) + }} + rc := &fakeRunCtx{ + ctx: context.Background(), + prof: core.AgentProfile{Model: core.ModelRoute{Chat: "test-model"}}, + execArtifact: &core.Artifact{ID: "art_aaaaaaaaaaaa", Name: "报告.docx", Kind: "document"}, + } + if err := r.Execute(rc); err != nil { + t.Fatalf("Execute: %v", err) + } + if calls != 1 { + t.Fatalf("stream calls = %d, want 1", calls) + } + texts := textParts(rc.assistant) + if len(texts) != 1 || texts[0].Process { + t.Fatalf("citation should stay the final reply, got %+v", texts) + } +}