Agent-System-Backend/internal/ctxasm/ctxasm.go

// Package ctxasm 实现上下文组装与预算管理（4.10）：分层管线 + 可观测裁剪记录。
package ctxasm

import (
	"encoding/json"
	"fmt"
	"regexp"
	"strings"
	"time"
	"unicode/utf8"

	agentllm "agent-native/runtime/llm"
	core "agent-system-core"
)

// Report 记录一次组装的实际构成（可观测：模型看到了什么、裁了什么）。
type Report struct {
	Items       []string  `json:"items"`
	Truncations []string  `json:"truncations,omitempty"`
	EstChars    int       `json:"estChars"`
	Sections    []Section `json:"sections,omitempty"`
	PrefixHash  string    `json:"prefixHash,omitempty"`
}

// DefaultMaxAttachmentChars 单个附件注入模型的默认字符上限（settings context.maxAttachmentChars 可调）。
// 16k 时代码值对会议纪要类长文档明显不足（正文大半被静默丢弃）；48k ≈ 主流 128k 上下文的安全占比。
const DefaultMaxAttachmentChars = 48000

// DefaultMaxCardChars 单张产物卡回填注入的默认字符上限（settings context.maxCardChars 可调）。
const DefaultMaxCardChars = 12000

// CardResolver 按 ID 取收件箱卡（产物卡全文回填；nil = 无卡服务/无此卡，跳过）。
type CardResolver func(id core.ID) *core.ActivityCard

// ArtifactResolver 按 ID 取工件及其内容（artifact_ref 正文回填；D46 与卡片回填
// 对称）。ok=false = 无工件服务/无此卡/内容不可读，部件留占位行。
type ArtifactResolver func(id core.ID) (art core.Artifact, content []byte, ok bool)

// AssembleInput 组装模型输入消息。sys 为提示词库组装出的 system 段（可为多段；
// 调用方经 prompts.Service.ResolveSystem 产出），promptObs 为对应可观测条目。
// prof.SystemPrompt 非空时作为画像自定义段追加（向后兼容旧画像）。
// resolveCard 用于 card 引用部件回填产物卡正文（submit_card；D35：卡片为产物权威
// 载体，下一轮模型不失忆），nil 时 card 部件只留占位。resolveArtifact 同理回填
// artifact_ref 工件正文（D46 统一；仅 UTF-8 文本，超 limit 截断留痕）。
func AssembleInput(prof core.AgentProfile, eff core.EffectiveSettings, history []core.Message, currentUser core.Message, attachments []core.ParsedDocument, knowledge []core.RetrievalHit, sys []agentllm.Message, promptObs []string, resolveCard CardResolver, resolveArtifact ArtifactResolver) ([]agentllm.Message, *Report) {
	rep := &Report{}
	msgs := make([]agentllm.Message, 0, len(sys)+1)
	msgs = append(msgs, sys...)
	rep.Items = append(rep.Items, promptObs...)
	if strings.TrimSpace(prof.SystemPrompt) != "" {
		msgs = append(msgs, agentllm.Message{Role: agentllm.RoleSystem, Content: prof.SystemPrompt})
		rep.Items = append(rep.Items, "system(画像自定义段)")
	}

	maxHist := prof.Context.MaxHistoryMessages
	if maxHist <= 0 {
		maxHist = int(eff.GetNumber("context.maxHistoryMessages", 40))
	}
	if maxHist <= 0 {
		maxHist = 40
	}
	history = visibleHistory(history)
	// 历史窗口（不含当前轮——当前轮单独注入附件与检索）。
	start := 0
	if len(history) > maxHist {
		start = len(history) - maxHist
		if prof.Context.SummarizeBeyond {
			summary := extractiveSummary(history[:start])
			if summary != "" {
				msgs = append(msgs, agentllm.Message{Role: agentllm.RoleSystem, Content: "[更早对话摘要]\n" + summary})
				rep.Items = append(rep.Items, fmt.Sprintf("summary(%d)", start))
			}
		}
		rep.Truncations = append(rep.Truncations, fmt.Sprintf("历史 %d 条超窗口 %d，仅保留最近部分", len(history), maxHist))
	}
	limitCard := int(eff.GetNumber("context.maxCardChars", DefaultMaxCardChars))
	if limitCard <= 0 {
		limitCard = DefaultMaxCardChars
	}
	cardsBackfilled := 0
	artsBackfilled := 0
	for _, m := range history[start:] {
		if m.ID == currentUser.ID {
			continue
		}
		if m.Role == core.RoleAssistant {
			msgs = append(msgs, renderAssistantTurns(m, resolveCard, limitCard, resolveArtifact, rep, &cardsBackfilled, &artsBackfilled)...)
			continue
		}
		content := renderMessage(m, resolveCard, limitCard, resolveArtifact, rep, &cardsBackfilled, &artsBackfilled)
		if content == "" {
			continue
		}
		msgs = append(msgs, agentllm.Message{Role: agentllm.RoleUser, Content: content})
	}
	rep.Items = append(rep.Items, fmt.Sprintf("history(%d)", len(history)-start))
	if cardsBackfilled > 0 {
		rep.Items = append(rep.Items, fmt.Sprintf("card(%d)", cardsBackfilled))
	}
	if artsBackfilled > 0 {
		rep.Items = append(rep.Items, fmt.Sprintf("artifact(%d)", artsBackfilled))
	}

	// 当前轮：文本 + 附件解析结果 + 知识检索注入。
	var sb strings.Builder
	for _, p := range currentUser.Parts {
		if p.Type == core.PartText {
			sb.WriteString(p.Text)
		}
	}
	if len(attachments) > 0 {
		limit := int(eff.GetNumber("context.maxAttachmentChars", DefaultMaxAttachmentChars))
		if limit <= 0 {
			limit = DefaultMaxAttachmentChars
		}
		sb.WriteString("\n\n[附件]（用户上传文档的解析全文，按原文顺序保留标题/列表/表格；超长时文末标注截断。请依据文档内容完成任务，输出完整不省略）\n")
		for _, doc := range attachments {
			text := doc.PlainText()
			if runes := []rune(text); len(runes) > limit {
				text = string(runes[:limit]) + "…（附件超长截断）"
				rep.Truncations = append(rep.Truncations, fmt.Sprintf("附件 %s 超 %d 字符截断（原文 %d 字符）", doc.Meta.Filename, limit, len(runes)))
			}
			fmt.Fprintf(&sb, "### %s\n%s\n", doc.Meta.Filename, text)
		}
		rep.Items = append(rep.Items, fmt.Sprintf("attachments(%d)", len(attachments)))
	}
	if len(knowledge) > 0 {
		sb.WriteString("\n\n[相关知识]\n")
		for i, h := range knowledge {
			fmt.Fprintf(&sb, "[%d] %s（来源 %s）\n%s\n", i+1, h.Doc.Title, h.Doc.Source, h.Doc.Chunk)
		}
		rep.Items = append(rep.Items, fmt.Sprintf("knowledge(%d)", len(knowledge)))
	}
	msgs = append(msgs, agentllm.Message{Role: agentllm.RoleUser, Content: sb.String()})

	fillEstimate(msgs, rep)
	return msgs, rep
}

// maxToolArgRunes 是回放进模型的单次工具参数上限。超限时截断字符串值，保持合法 JSON。
const maxToolArgRunes = 4000

// forgedMarkerLine 匹配模型在正文里模仿出来的调用/结果/工件行。回放时整行剔除，
// 避免下一轮把这种文本当成「以前就是这样调用工具」的格式。
var forgedMarkerLine = regexp.MustCompile(`(?m)^[^\n]*\[(?:调用工具|工具结果|产出工件) [^\n]*\n?`)

// renderMessage 把用户消息部件渲染为模型可读文本（思考部件不回灌；
// card 引用回填产物卡正文、artifact_ref 回填工件正文——截断记入 rep.Truncations）。
// 助手消息不走这里：工具调用必须保持提供方的原生消息，见 renderAssistantTurns。
func renderMessage(m core.Message, resolveCard CardResolver, limitCard int, resolveArtifact ArtifactResolver, rep *Report, cardsBackfilled, artsBackfilled *int) string {
	var sb strings.Builder
	for _, p := range m.Parts {
		switch p.Type {
		case core.PartText:
			sb.WriteString(stripForgedMarkers(p.Text))
			sb.WriteString("\n")
		case core.PartArtifactRef:
			sb.WriteString(renderArtifactRef(p.Artifact, resolveArtifact, limitCard, rep, artsBackfilled))
		case core.PartHITL:
			if p.Hitl != nil && p.Hitl.Resolved {
				fmt.Fprintf(&sb, "[用户确认：%s → %s]\n", p.Hitl.Question, p.Hitl.Answer)
			}
		case core.PartCard:
			sb.WriteString(renderCardPart(p.Card, resolveCard, limitCard, rep, cardsBackfilled))
		}
	}
	return strings.TrimSpace(sb.String())
}

// renderAssistantTurns 把一条助手消息切成提供方可接受的若干段：
// 前置文本与 tool_calls 组成一条 assistant，每个调用紧跟一条 tool 结果。
// 没有结果的调用补一条失败占位，避免模型把中断当成成功。孤立的结果直接跳过。
// 被守卫降级、且后面不是工具调用的过程文本不回放。
func renderAssistantTurns(m core.Message, resolveCard CardResolver, limitCard int, resolveArtifact ArtifactResolver, rep *Report, cardsBackfilled, artsBackfilled *int) []agentllm.Message {
	var out []agentllm.Message
	var text strings.Builder
	var calls []agentllm.ToolCall
	results := map[string]core.ToolResultPart{}
	seq := 0

	flush := func() {
		content := strings.TrimSpace(text.String())
		if content == "" && len(calls) == 0 {
			text.Reset()
			results = map[string]core.ToolResultPart{}
			return
		}
		out = append(out, agentllm.Message{Role: agentllm.RoleAssistant, Content: content, ToolCalls: calls})
		for _, c := range calls {
			body := "（无结果：运行中断）"
			isErr := true
			if res, ok := results[c.ID]; ok {
				body = res.Summary
				isErr = res.Status == "error"
			}
			out = append(out, agentllm.Message{
				Role:       agentllm.RoleTool,
				ToolCallID: c.ID,
				Name:       c.Name,
				Content:    body,
				IsError:    isErr,
			})
		}
		text.Reset()
		calls = nil
		results = map[string]core.ToolResultPart{}
	}

	for i, p := range m.Parts {
		switch p.Type {
		case core.PartText:
			if demotedProcessText(m.Parts, i) {
				continue
			}
			cleaned := stripForgedMarkers(p.Text)
			if strings.TrimSpace(cleaned) == "" {
				continue
			}
			if len(calls) > 0 {
				flush()
			}
			text.WriteString(cleaned)
			text.WriteString("\n")
		case core.PartToolCall:
			if p.ToolCall == nil {
				continue
			}
			id := p.ToolCall.ID
			if id == "" {
				id = fmt.Sprintf("hist_%s_%d", m.ID, seq)
				seq++
			}
			calls = append(calls, agentllm.ToolCall{
				ID:        id,
				Name:      p.ToolCall.Name,
				Arguments: compactToolArguments(p.ToolCall.Arguments),
			})
		case core.PartToolResult:
			if p.ToolResult == nil {
				continue
			}
			id := p.ToolResult.CallID
			if id == "" {
				for _, c := range calls {
					if _, ok := results[c.ID]; !ok {
						id = c.ID
						break
					}
				}
			}
			if id == "" || !hasCall(calls, id) {
				continue
			}
			results[id] = *p.ToolResult
		case core.PartArtifactRef:
			if len(calls) > 0 {
				flush()
			}
			text.WriteString(renderArtifactRef(p.Artifact, resolveArtifact, limitCard, rep, artsBackfilled))
		case core.PartHITL:
			if p.Hitl == nil || !p.Hitl.Resolved {
				continue
			}
			if len(calls) > 0 {
				flush()
			}
			fmt.Fprintf(&text, "[用户确认：%s → %s]\n", p.Hitl.Question, p.Hitl.Answer)
		case core.PartCard:
			if len(calls) > 0 {
				flush()
			}
			text.WriteString(renderCardPart(p.Card, resolveCard, limitCard, rep, cardsBackfilled))
		}
	}
	flush()
	return out
}

func hasCall(calls []agentllm.ToolCall, id string) bool {
	for _, c := range calls {
		if c.ID == id {
			return true
		}
	}
	return false
}

// demotedProcessText 是守卫降级后的谎称正文：过程文本，且后面紧跟的不是工具调用。
// 工具轮的解说也标了过程，但它后面就是 tool_call，要保留。
func demotedProcessText(parts []core.Part, i int) bool {
	if parts[i].Type != core.PartText || !parts[i].Process {
		return false
	}
	return i+1 >= len(parts) || parts[i+1].Type != core.PartToolCall
}

func stripForgedMarkers(text string) string {
	return strings.TrimSpace(forgedMarkerLine.ReplaceAllString(text, ""))
}

// compactToolArguments 在参数超限时截断其中的长字符串。解析失败或截断后仍超限时，
// 换成仍可被提供方接受的占位对象，而不是一段被截断的非法 JSON。
func compactToolArguments(raw string) string {
	if utf8.RuneCountInString(raw) <= maxToolArgRunes {
		return raw
	}
	var v any
	if err := json.Unmarshal([]byte(raw), &v); err != nil {
		return fmt.Sprintf(`{"_omitted":"参数 %d 字符"}`, utf8.RuneCountInString(raw))
	}
	truncateLongStrings(&v, 500)
	encoded, err := json.Marshal(v)
	if err != nil || utf8.RuneCountInString(string(encoded)) > maxToolArgRunes {
		return fmt.Sprintf(`{"_omitted":"参数 %d 字符"}`, utf8.RuneCountInString(raw))
	}
	return string(encoded)
}

func truncateLongStrings(v *any, keep int) {
	switch x := (*v).(type) {
	case map[string]any:
		for k, val := range x {
			truncateLongStrings(&val, keep)
			x[k] = val
		}
	case []any:
		for i := range x {
			truncateLongStrings(&x[i], keep)
		}
	case string:
		n := utf8.RuneCountInString(x)
		if n > keep {
			*v = string([]rune(x)[:keep]) + fmt.Sprintf("…（已省略 %d 字符）", n-keep)
		}
	}
}

// renderCardPart 渲染 card 部件：引用部件（cardId）回填产物卡全文（超限截断留痕）；
// 无 resolver 或无此卡时留占位（可见即信号，不静默丢弃）。
func renderCardPart(cp *core.CardPart, resolveCard CardResolver, limit int, rep *Report, cardsBackfilled *int) string {
	if cp == nil {
		return ""
	}
	if cp.CardID == "" || resolveCard == nil {
		return fmt.Sprintf("[卡片 %s（内容见收件箱）]\n", cp.Kind)
	}
	card := resolveCard(cp.CardID)
	if card == nil {
		return fmt.Sprintf("[卡片 %s（内容见收件箱）]\n", cp.Kind)
	}
	*cardsBackfilled++
	var sb strings.Builder
	fmt.Fprintf(&sb, "[产物卡片：%s]\n", card.Title)
	total := 0
	for _, n := range card.Body {
		switch n.Type {
		case "markdown":
			runes := []rune(n.Markdown)
			if total+len(runes) > limit {
				keep := limit - total
				if keep < 0 {
					keep = 0
				}
				rep.Truncations = append(rep.Truncations, fmt.Sprintf("卡片 %s 超 %d 字符截断", card.Title, limit))
				sb.WriteString(string(runes[:keep]))
				sb.WriteString("…（卡片超长截断）\n")
				return sb.String()
			}
			total += len(runes)
			sb.WriteString(n.Markdown)
			sb.WriteString("\n")
		case "kv":
			for _, kv := range n.KV {
				fmt.Fprintf(&sb, "%s：%s\n", kv[0], kv[1])
			}
		}
	}
	return sb.String()
}

// renderArtifactRef 渲染 artifact_ref 部件：有 resolver 且内容为 UTF-8 文本时
// 回填正文（与卡同 limit，超限截断留痕）；二进制/不可读/无 resolver 留占位行
// （可见即信号，不静默丢弃）。
func renderArtifactRef(a *core.ArtifactRef, resolve ArtifactResolver, limit int, rep *Report, backfilled *int) string {
	if a == nil {
		return ""
	}
	if resolve == nil {
		return fmt.Sprintf("[产出工件 %s]\n", a.Name)
	}
	art, content, ok := resolve(a.ArtifactID)
	if !ok {
		return fmt.Sprintf("[产出工件 %s]\n", a.Name)
	}
	if !utf8.Valid(content) {
		return fmt.Sprintf("[产出工件 %s（二进制内容，正文见工件查看器）]\n", art.Name)
	}
	*backfilled++
	text := string(content)
	if runes := []rune(text); len(runes) > limit {
		text = string(runes[:limit]) + "…（工件超长截断）"
		rep.Truncations = append(rep.Truncations, fmt.Sprintf("工件 %s 超 %d 字符截断", art.Name, limit))
	}
	return fmt.Sprintf("[产出工件 %s]\n%s\n", art.Name, text)
}

// extractiveSummary 把窗口外历史压成短摘要（不调用模型）。summarizeBeyond 打开时注入。
func extractiveSummary(dropped []core.Message) string {
	var lines []string
	for _, m := range dropped {
		var text string
		for _, p := range m.Parts {
			if p.Type == core.PartText && strings.TrimSpace(p.Text) != "" {
				text = strings.TrimSpace(p.Text)
				break
			}
		}
		if text == "" {
			continue
		}
		runes := []rune(text)
		if len(runes) > 80 {
			text = string(runes[:80]) + "…"
		}
		role := "用户"
		if m.Role == core.RoleAssistant {
			role = "助手"
		}
		lines = append(lines, role+"："+text)
		if len(lines) >= 12 {
			break
		}
	}
	return strings.Join(lines, "\n")
}

// InjectMemory 把未过期记忆作为系统段放到提示词之后。空列表不改。
func InjectMemory(sys []agentllm.Message, entries []core.MemoryEntry) []agentllm.Message {
	if len(entries) == 0 {
		return sys
	}
	var b strings.Builder
	b.WriteString("[记忆]\n")
	n := 0
	for _, e := range entries {
		if e.ExpiresAt != nil && e.ExpiresAt.Before(time.Now()) {
			continue
		}
		if strings.TrimSpace(e.Content) == "" {
			continue
		}
		fmt.Fprintf(&b, "- (%s) %s\n", e.Type, strings.TrimSpace(e.Content))
		n++
		if n >= 20 {
			break
		}
	}
	if n == 0 {
		return sys
	}
	out := make([]agentllm.Message, 0, len(sys)+1)
	out = append(out, sys...)
	out = append(out, agentllm.Message{Role: agentllm.RoleSystem, Content: b.String()})
	return out
}