Agent-Squad/squad/engine.go

package squad

import (
	"context"
	"crypto/sha256"
	"encoding/hex"
	"encoding/json"
	"errors"
	"fmt"
	"os"
	"path/filepath"
	"sort"
	"strings"
	"sync"
	"time"

	"error-code-system/errcode"
)

// Options 引擎组装选项（组合根装配；库内零产品依赖）。
type Options struct {
	DataDir    string // 数据根：runs/、roles.d/、blueprints.d/ 所在目录（必填）
	Registry   *RoleRegistry
	Blueprints *BlueprintRegistry
	Planner    Planner
	Backends   *BackendResolver
	Evaluator  StepEvaluator
	Checker    Checker
	Sink       ProgressSink
	// DefaultBackend 角色未声明 Backend 时的默认智能体名（默认 stub）。
	DefaultBackend string
	// StepTimeout 单步执行超时（默认 10 分钟）。
	StepTimeout time.Duration
	// HitlPolicy 全局 HITL 策略（角色/Goal 可覆盖）。
	HitlPolicy HitlPolicy
	// RepairBudget 质检失败自动修复次数（默认 1）。
	RepairBudget int
	// IntentGate 可选意图闸门：Run 入口一次确定性裁决（nil = 不启用，
	// 行为与原实现一致）。见 intent.go；实现方为 squadintent 适配器。
	IntentGate IntentGate
	// IntentCatalog 可选能力目录（squadintent 渲染）：注入 PlanInput 与 leader 提示词，
	// 并用于 intent 步骤的确定性校验。nil = 未启用目录驱动编排。
	IntentCatalog *IntentCatalog
	// IntentRunner 系统功能步执行器（squadintent 提供）。目录启用而 Runner 缺席时，
	// intent 步骤 fail-closed。
	IntentRunner IntentRunner
	// Router 可选分诊。direct = 单步 assistant，不进多角色；plan = leader 委派。
	// nil = 保持调用方传入的 RunOptions.Mode。
	Router Router
}

// engineDeps 驱动器依赖（由引擎从 Options 派生）。
type engineDeps struct {
	Registry     *RoleRegistry
	Backends     *BackendResolver
	Evaluator    StepEvaluator
	Checker      Checker
	Sink         ProgressSink
	Workspaces   *WorkspaceEngine
	StepTimeout  time.Duration
	HitlPolicy   HitlPolicy
	IntentRunner IntentRunner
}

// RunOptions 单次运行的可选项。
type RunOptions struct {
	// TaskKey 显式指定蓝图模板键；空则按 Goal.Text 匹配或自主规划。
	TaskKey string `json:"task_key,omitempty"`
	// Backend 运行级默认后端（覆盖引擎 DefaultBackend）。
	Backend string `json:"backend,omitempty"`
	// Mode 编排模式：standard（默认，一次性计划顺序执行）| leader
	// （leader-delegation：leader 角色逐步委派成员，成员产出后唤醒 leader 再评估）。
	Mode string `json:"mode,omitempty"`
	// LeaderRole leader 模式下的 leader 角色 key（必须已注册或在计划编制内）。
	LeaderRole string `json:"leader_role,omitempty"`
	// MaxLeaderTurns leader 模式每运行的 leader 轮次上限（默认 8，防失控）。
	MaxLeaderTurns int `json:"max_leader_turns,omitempty"`
}

// SquadEngine 小队引擎门面：Run / Respond / Resume / Inspect / ListRuns。
// 每个运行同一时刻只允许一个控制操作（Run/Respond 互斥）。
type SquadEngine struct {
	opts Options
	deps engineDeps
	ws   *WorkspaceEngine

	mu       sync.Mutex
	runLocks map[string]*sync.Mutex
}

// NewEngine 组装引擎并应用默认值。
func NewEngine(opts Options) (*SquadEngine, error) {
	if opts.DataDir == "" {
		return nil, fmt.Errorf("DataDir is required")
	}
	if opts.Registry == nil {
		opts.Registry = NewRoleRegistry()
	}
	if opts.Blueprints == nil {
		opts.Blueprints = NewBlueprintRegistry()
	}
	if opts.Planner == nil {
		opts.Planner = NewRulePlanner()
	}
	if opts.Backends == nil {
		opts.Backends = NewBackendResolver()
	}
	if opts.Evaluator == nil {
		opts.Evaluator = NoOpEvaluator{}
	}
	if opts.Checker == nil {
		opts.Checker = MinimalChecker{}
	}
	if opts.Sink == nil {
		opts.Sink = NopSink{}
	}
	if opts.DefaultBackend == "" {
		opts.DefaultBackend = "stub"
	}
	if opts.StepTimeout == 0 {
		opts.StepTimeout = 10 * time.Minute
	}
	if opts.RepairBudget == 0 {
		opts.RepairBudget = 1
	}
	if opts.HitlPolicy.Mode == "" {
		opts.HitlPolicy.Mode = HitlInteractive
	}
	ws := &WorkspaceEngine{DataDir: opts.DataDir}
	deps := engineDeps{
		Registry: opts.Registry, Backends: opts.Backends, Evaluator: opts.Evaluator,
		Checker: opts.Checker, Sink: opts.Sink, Workspaces: ws,
		StepTimeout: opts.StepTimeout, HitlPolicy: opts.HitlPolicy,
		IntentRunner: opts.IntentRunner,
	}
	return &SquadEngine{opts: opts, deps: deps, ws: ws, runLocks: map[string]*sync.Mutex{}}, nil
}

// Registry 暴露角色注册表（只读使用）。
func (e *SquadEngine) Registry() *RoleRegistry { return e.opts.Registry }

// Blueprints 暴露蓝图注册表。
func (e *SquadEngine) Blueprints() *BlueprintRegistry { return e.opts.Blueprints }

func (e *SquadEngine) runLock(runID string) *sync.Mutex {
	e.mu.Lock()
	defer e.mu.Unlock()
	l, ok := e.runLocks[runID]
	if !ok {
		l = &sync.Mutex{}
		e.runLocks[runID] = l
	}
	return l
}

// ErrRunBusy 表示该 run 的文件锁正被占用，且在 ctx 截止前没有拿到。
// 非 unix 平台没有 flock，只保留进程内互斥；跨进程互斥需要 unix。
var ErrRunBusy = errors.New("run is busy")

// acquire 进入 run 临界区：先校验 id，再拿进程内锁，然后按 ctx 轮询文件锁。
func (e *SquadEngine) acquire(ctx context.Context, runID string) (func(), error) {
	if !ValidID(runID) {
		return nil, invalidID(runID)
	}
	if ctx == nil {
		ctx = context.Background()
	}
	mu := e.runLock(runID)
	mu.Lock()
	release, err := lockRunFile(ctx, filepath.Join(e.opts.DataDir, "runs", runID))
	if err != nil {
		mu.Unlock()
		return nil, err
	}
	return func() {
		release()
		mu.Unlock()
	}, nil
}

// Run 发起一次运行：规划（含澄清/确认挂起）→ 执行 → 收口。
func (e *SquadEngine) Run(ctx context.Context, goal Goal, runOpts RunOptions) (*RunResult, error) {
	if strings.TrimSpace(goal.Text) == "" {
		return nil, fmt.Errorf("goal text is required")
	}
	runID := fmt.Sprintf("run_%s", time.Now().UTC().Format("20060102T150405.000000000"))
	release, err := e.acquire(ctx, runID)
	if err != nil {
		return nil, err
	}
	defer release()

	meta := RunMetadata{
		ID: runID, Goal: goal, Status: string(RunPlanning),
		Backend:   firstNonEmpty(runOpts.Backend, e.opts.DefaultBackend),
		Planner:   e.opts.Planner.Name(),
		CreatedAt: time.Now(), UpdatedAt: time.Now(),
		Metadata: map[string]string{
			"task_key": runOpts.TaskKey,
			"mode":     runOpts.Mode,
		},
	}
	if runOpts.Mode == "leader" {
		meta.Metadata["leader_role"] = runOpts.LeaderRole
		if runOpts.MaxLeaderTurns > 0 {
			meta.Metadata["max_leader_turns"] = fmt.Sprintf("%d", runOpts.MaxLeaderTurns)
		}
	}
	e.deps.Sink.Emit(Event{Type: EvRunStarted, RunID: runID, Text: goal.Text, At: time.Now()})

	// 意图闸门（可选，入口一次裁决；显式 TaskKey 优先，不重复过闸）。
	// nil 决策 = 不干预走原路径；闸门系统错误与守卫拒绝同样 fail-closed 成失败运行。
	if e.opts.IntentGate != nil && runOpts.TaskKey == "" {
		dec, derr := e.opts.IntentGate.Decide(ctx, goal, runOpts)
		if derr != nil {
			e.deps.Sink.Emit(Event{Type: EvRunFailed, RunID: runID, Text: derr.Error(), At: time.Now()})
			return e.failNew(&meta, derr)
		}
		if dec != nil {
			if rej := dec.Reject; rej != nil {
				if dec.IntentID != "" {
					meta.Metadata["intent_id"] = dec.IntentID
				}
				err := errcode.New(errcode.SquadIntentRefused, rej.Reason, "guidance", rej.Guidance)
				e.deps.Sink.Emit(Event{Type: EvRunFailed, RunID: runID, Text: err.Error(), At: time.Now()})
				return e.failNew(&meta, err)
			}
			if dec.BlueprintKey != "" {
				if _, ok := e.opts.Blueprints.Get(dec.BlueprintKey); !ok {
					err := errcode.New(errcode.SquadBlueprintNotFound,
						fmt.Sprintf("意图 %s 定向的蓝图 %s 不存在", dec.IntentID, dec.BlueprintKey))
					e.deps.Sink.Emit(Event{Type: EvRunFailed, RunID: runID, Text: err.Error(), At: time.Now()})
					return e.failNew(&meta, err)
				}
				runOpts.TaskKey = dec.BlueprintKey
				meta.Metadata["intent_id"] = dec.IntentID
				meta.Metadata["intent_blueprint"] = dec.BlueprintKey
			}
		}
	}

	st := &runState{Meta: meta, Approved: map[string]bool{}, RepairBudgetRemaining: e.opts.RepairBudget}
	return e.planAndExecute(ctx, st, runOpts)
}

// planAndExecute 规划 → 挂起或执行；st 可为已有运行状态（澄清续跑复用）。
func (e *SquadEngine) planAndExecute(ctx context.Context, st *runState, runOpts RunOptions) (*RunResult, error) {
	drv := &flowDriver{opts: e.deps}
	meta := &st.Meta

	if e.opts.Router != nil && runOpts.Mode == "" && meta.Metadata["route"] == "" {
		dec, rerr := e.opts.Router.Route(ctx, meta.Goal)
		if rerr != nil {
			return e.failNew(meta, rerr)
		}
		applyRoute(meta, dec, runOpts.Mode)
	}

	var planOut *PlanOutput
	var err error
	if meta.Metadata["route_direct"] == "1" {
		plan, perr := directAssistantPlan(e.opts.Registry, meta.Goal)
		if perr != nil {
			return e.failNew(meta, perr)
		}
		planOut = &PlanOutput{Plan: plan}
	} else {
		bpMatch, _ := e.opts.Blueprints.Match(meta.Goal.Text, runOpts.TaskKey)
		planOut, err = e.opts.Planner.Plan(ctx, PlanInput{
			Goal: meta.Goal, Blueprint: bpPtr(bpMatch), Registry: e.opts.Registry, RunID: meta.ID,
			IntentCatalog: e.opts.IntentCatalog,
		})
		if err != nil {
			return e.failNew(meta, err)
		}
	}
	// 目录驱动编排的确定性校验：intent 步骤命中目录、必填槽位齐、枚举合法（fail-closed）。
	if err := validateIntentSteps(planOut.Plan, e.opts.IntentCatalog); err != nil {
		e.deps.Sink.Emit(Event{Type: EvRunFailed, RunID: meta.ID, Text: err.Error(), At: time.Now()})
		return e.failNew(meta, err)
	}
	// 规划歧义：挂起澄清（auto/disabled 模式直接失败，不猜计划）。
	if planOut.Clarification != nil {
		if drv.policy(st).mode() != HitlInteractive {
			return e.failNew(meta, fmt.Errorf("planner requested clarification but hitl mode is %s", st.policy().mode()))
		}
		var qs []string
		for _, q := range planOut.Clarification.Questions {
			qs = append(qs, "- "+q.Text)
		}
		opts := []HumanOption{}
		if len(planOut.Clarification.Questions) == 1 && len(planOut.Clarification.Questions[0].Options) > 0 {
			opts = planOut.Clarification.Questions[0].Options
		}
		req := &HumanRequest{
			ID: newRequestID(), RunID: meta.ID, Kind: HitlClarification, Severity: SeverityActionRequired,
			Question:  "规划器无法确定如何执行该目标，请补充信息：",
			Options:   opts,
			Context:   fmt.Sprintf("原因：%s\n问题：\n%s", planOut.Clarification.Reason, strings.Join(qs, "\n")),
			CreatedAt: time.Now(), Status: "pending",
		}
		gate := &hitlGate{Request: *req, PrevStatus: RunPlanning}
		// 澄清挂起也需要工作区承载 pending.json。
		if err := e.materializeWorkspace(meta, st, &SquadPlan{Workflow: WorkflowPlan{WorkflowID: NewPlanID(), Status: "pending", Steps: nil}}); err != nil {
			return nil, err
		}
		if err := st.saveGate(gate); err != nil {
			return nil, err
		}
		st.Meta.Status = string(RunAwaitingClarification)
		_ = st.save()
		e.deps.Sink.Emit(Event{Type: EvHumanInputRequired, RunID: meta.ID, Text: req.Question, Data: marshalJSON(req), At: time.Now()})
		return &RunResult{RunID: meta.ID, Status: RunAwaitingClarification, Pending: req, WorkDir: st.Meta.WorkDir}, nil
	}

	// 计划确认：interactive 模式先挂起。leader 模式例外——规划步骤会被
	// leader 委派循环取代，派生编制本身不构成执行承诺，无需确认。
	if planOut.NeedsConfirmation && drv.policy(st).mode() == HitlInteractive && meta.Metadata["mode"] != "leader" {
		if err := e.materializeWorkspace(meta, st, planOut.Plan); err != nil {
			return nil, err
		}
		var lines []string
		for _, s := range planOut.Plan.Workflow.Steps {
			lines = append(lines, fmt.Sprintf("- %s [%s] %s（产出 %v）", s.ID, s.Role, s.Title, s.Targets))
		}
		var roles []string
		for _, a := range planOut.Plan.Squad.Agents {
			roles = append(roles, fmt.Sprintf("- %s（%s，backend=%s model=%s）", a.Role, a.Name, a.Backend, a.Model))
		}
		req := &HumanRequest{
			ID: newRequestID(), RunID: meta.ID, Kind: HitlConfirmation, Severity: SeverityActionRequired,
			Question: "计划已生成，需要人工确认后执行：",
			Options: []HumanOption{
				{Value: "approve", Label: "批准并执行"},
				{Value: "abort", Label: "放弃执行"},
			},
			Context: fmt.Sprintf("确认原因：%s\n编制：\n%s\n步骤：\n%s",
				planOut.ConfirmReason, strings.Join(roles, "\n"), strings.Join(lines, "\n")),
			CreatedAt: time.Now(), Status: "pending",
			Reason: "plan",
		}
		gate := &hitlGate{Request: *req, Plan: planOut.Plan, PrevStatus: RunPlanning}
		if err := st.saveGate(gate); err != nil {
			return nil, err
		}
		st.Meta.Status = string(RunAwaitingConfirmation)
		_ = st.save()
		e.deps.Sink.Emit(Event{Type: EvHumanInputRequired, RunID: meta.ID, Text: req.Question, Data: marshalJSON(req), At: time.Now()})
		return &RunResult{RunID: meta.ID, Status: RunAwaitingConfirmation, Pending: req, Plan: &planOut.Plan.Workflow, WorkDir: st.Meta.WorkDir}, nil
	}

	if err := e.materializeWorkspace(meta, st, planOut.Plan); err != nil {
		return nil, err
	}
	// leader-delegation 模式：忽略规划步骤，进 leader 委派循环。
	if meta.Metadata["mode"] == "leader" {
		e.initLeaderState(st)
		out, gate, err := e.runLeaderLoop(ctx, st)
		if err != nil {
			return nil, err
		}
		if gate == nil {
			e.finalizeEvent(out)
		}
		return out, nil
	}
	out, _, err := drv.advance(ctx, st)
	if err != nil {
		return nil, err
	}
	e.finalizeEvent(out)
	return out, nil
}

// initLeaderState 按运行选项初始化 leader 循环状态。
func (e *SquadEngine) initLeaderState(st *runState) {
	leaderRole := st.Meta.Metadata["leader_role"]
	if leaderRole == "" {
		if _, ok := findAgent(st.Plan.Squad.Agents, "leader"); ok {
			leaderRole = "leader"
		}
	}
	maxTurns := DefaultMaxLeaderTurns
	if v := st.Meta.Metadata["max_leader_turns"]; v != "" {
		if n := parseIntDefault(v, DefaultMaxLeaderTurns); n > 0 {
			maxTurns = n
		}
	}
	st.Leader = &LeaderState{LeaderRole: leaderRole, MaxTurns: maxTurns}
	st.saveLeader()
}

func parseIntDefault(s string, def int) int {
	n := 0
	for _, c := range s {
		if c < '0' || c > '9' {
			return def
		}
		n = n*10 + int(c-'0')
	}
	return n
}

// isLeaderRun 该运行是否处于 leader-delegation 模式。
func (e *SquadEngine) isLeaderRun(st *runState) bool {
	return st.Meta.Metadata["mode"] == "leader"
}

// resumeAfterHold 挂起解除后的统一续跑入口（按模式分发）。
func (e *SquadEngine) resumeAfterHold(ctx context.Context, st *runState) (*RunResult, error) {
	if e.isLeaderRun(st) {
		out, gate, err := e.runLeaderLoop(ctx, st)
		if err != nil {
			return nil, err
		}
		if gate == nil {
			e.finalizeEvent(out)
		}
		return out, nil
	}
	drv := &flowDriver{opts: e.deps}
	out, _, err := drv.advance(ctx, st)
	if err != nil {
		return nil, err
	}
	e.finalizeEvent(out)
	return out, nil
}

// materializeWorkspace 落盘工作区与子工作区（已存在则只更新计划并重建待执行子目录）。
func (e *SquadEngine) materializeWorkspace(meta *RunMetadata, st *runState, plan *SquadPlan) error {
	if st.Meta.WorkDir == "" {
		st.Meta = *meta
		wsDir, err := e.ws.PrepareWorkspace(&st.Meta, plan)
		if err != nil {
			return err
		}
		st.Template = plan.Template
		bb, _ := loadBlackboard(wsDir)
		st.BB = *bb
	} else {
		// 复用既有工作区：磁盘黑板为权威（澄清续跑/计划更新场景）。
		if bb, err := loadBlackboard(st.Meta.WorkDir); err == nil {
			st.BB = *bb
		}
		if st.Template == "" {
			st.Template = plan.Template
		}
	}
	st.Plan = *plan
	// 编制补全：步骤引用但编制缺失的角色从注册表解析（蓝图/派生计划常见）。
	if err := ensureSquadCoversSteps(&st.Plan, e.opts.Registry); err != nil {
		return err
	}
	if err := e.ws.BuildSubWorkspaces(st.Meta.WorkDir, &st.Plan, &st.BB); err != nil {
		return err
	}
	return st.save()
}

// ensureSquadCoversSteps 确保编制覆盖全部步骤角色。
func ensureSquadCoversSteps(plan *SquadPlan, reg *RoleRegistry) error {
	for _, step := range plan.Workflow.Steps {
		if _, ok := findAgent(plan.Squad.Agents, step.Role); ok {
			continue
		}
		spec, ok := reg.Get(step.Role)
		if !ok {
			return fmt.Errorf("step %s references role %s not registered", step.ID, step.Role)
		}
		plan.Squad.Agents = append(plan.Squad.Agents, spec)
	}
	return nil
}

func (e *SquadEngine) failNew(meta *RunMetadata, err error) (*RunResult, error) {
	return &RunResult{RunID: meta.ID, Status: RunFailed, Error: err.Error()}, nil
}

// Respond 应用人工决策并续跑。按请求 ID 幂等：重复回应同一请求返回当前状态。
func (e *SquadEngine) Respond(ctx context.Context, runID string, decision RespondDecision) (*RunResult, error) {
	release, err := e.acquire(ctx, runID)
	if err != nil {
		return nil, err
	}
	defer release()
	return e.respondLocked(ctx, runID, decision)
}

func (e *SquadEngine) respondLocked(ctx context.Context, runID string, decision RespondDecision) (*RunResult, error) {
	wsDir := filepath.Join(e.opts.DataDir, "runs", runID)
	st, err := loadRunState(wsDir)
	if err != nil {
		return nil, fmt.Errorf("load run %s: %w", runID, err)
	}
	gate, err := loadGate(wsDir)
	if err != nil {
		return nil, err
	}
	if gate == nil {
		return nil, fmt.Errorf("run %s has no pending human request", runID)
	}
	if decision.RequestID != "" && decision.RequestID != gate.Request.ID {
		// 幂等：回应已处理过的请求 → 返回当前状态。
		if gate.Request.Status != "pending" {
			return e.Inspect(runID)
		}
		return nil, fmt.Errorf("request id mismatch: expected %s", gate.Request.ID)
	}

	req := &gate.Request
	if strings.TrimSpace(decision.Choice) == "" && strings.TrimSpace(decision.Answer) != "" && choiceHoldKind(req.Kind) {
		return e.noteAndKeepHold(st, gate, decision)
	}
	req.Status = "answered"
	now := time.Now()
	req.AnsweredAt = &now
	if err := st.saveGate(gate); err != nil {
		return nil, err
	}
	e.deps.Sink.Emit(Event{Type: EvHumanResponded, RunID: runID, Text: firstNonEmpty(decision.Choice, decision.Answer),
		Data: marshalJSON(decision), At: time.Now()})

	res, err := e.applyDecision(ctx, st, gate, decision)
	if err != nil {
		req.Status = "pending"
		req.AnsweredAt = nil
		_ = st.saveGate(gate)
		return nil, err
	}
	if err := clearSameGate(st, req.ID); err != nil {
		return nil, err
	}
	return res, nil
}

func choiceHoldKind(kind HumanRequestKind) bool {
	switch kind {
	case HitlConfirmation, HitlPermission, HitlQualityReview:
		return true
	default:
		return false
	}
}

func (e *SquadEngine) noteAndKeepHold(st *runState, gate *hitlGate, decision RespondDecision) (*RunResult, error) {
	if err := st.addSteering(decision.Answer); err != nil {
		return nil, err
	}
	gate.Request.Status = "pending"
	gate.Request.AnsweredAt = nil
	if err := st.saveGate(gate); err != nil {
		return nil, err
	}
	return heldResult(st, gate), nil
}

func clearSameGate(st *runState, requestID string) error {
	current, err := loadGate(st.Meta.WorkDir)
	if err != nil || current == nil || current.Request.ID != requestID {
		return err
	}
	return st.saveGate(nil)
}

func (e *SquadEngine) applyDecision(ctx context.Context, st *runState, gate *hitlGate, decision RespondDecision) (*RunResult, error) {
	req := &gate.Request
	now := time.Now()
	drv := &flowDriver{opts: e.deps}
	switch req.Kind {
	case HitlClarification:
		return e.respondClarification(ctx, st, gate, decision)
	case HitlConfirmation:
		if req.Reason == "interrupted" {
			return e.respondInterrupted(ctx, st, req, decision)
		}
		if req.StepID != "" && gate.Plan == nil {
			return e.respondStepApproval(ctx, st, req, decision)
		}
		switch strings.ToLower(decision.Choice) {
		case "approve":
			if gate.Plan != nil {
				if err := e.materializeWorkspace(&st.Meta, st, gate.Plan); err != nil {
					return nil, err
				}
			}
			if e.isLeaderRun(st) {
				e.initLeaderState(st)
			}
			st.Meta.Status = string(RunExecuting)
			_ = st.save()
			return e.resumeAfterHold(ctx, st)
		default: // abort / reject
			return e.cancelRun(st, "人工拒绝计划"+appendAnswer(decision)), nil
		}
	case HitlGap:
		switch strings.ToLower(decision.Choice) {
		case "resolved", "":
			answer := decision.Answer
			for _, g := range st.BB.UnresolvedBlockingGaps() {
				st.BB.ResolveGapByID(g.ID, answerFor(g, answer))
			}
			if answer != "" {
				st.Clarifications = append(st.Clarifications, "缺口补充："+answer)
			}
		case "skip":
			for i := range st.BB.Gaps {
				st.BB.Gaps[i].Blocking = false
			}
		case "abort":
			return e.cancelRun(st, "人工中止（阻塞缺口）"), nil
		default:
			return nil, fmt.Errorf("unknown choice %q for gap request", decision.Choice)
		}
		st.Meta.Status = string(RunExecuting)
		_ = st.save()
		return e.resumeAfterHold(ctx, st)
	case HitlEscalation:
		switch strings.ToLower(decision.Choice) {
		case "abort":
			return e.cancelRun(st, "人工中止（leader 升级问人）"), nil
		default: // 自由文本回答 = 人的决定
			answer := strings.TrimSpace(decision.Answer)
			if answer == "" {
				for _, o := range gate.Request.Options {
					if strings.EqualFold(o.Value, decision.Choice) {
						answer = o.Label
					}
				}
			}
			if answer != "" {
				st.Clarifications = append(st.Clarifications, "升级问答："+answer)
				st.BB.Facts = append(st.BB.Facts, FactItem{
					ID: fmt.Sprintf("fact_%d", len(st.BB.Facts)+1), Claim: answer, Source: "human",
				})
				if st.Meta.WorkDir != "" {
					_ = e.ws.UpdateGoalFile(st.Meta.WorkDir, &st.Meta.Goal)
				}
			}
			st.Meta.Status = string(RunExecuting)
			_ = st.save()
			return e.resumeAfterHold(ctx, st)
		}
	case HitlAdjustment:
		switch strings.ToLower(decision.Choice) {
		case "approve":
			applyAdjustment(st, gate.Steps, gate.SkipIDs)
		case "reject", "":
			// 忽略调整
		case "abort":
			return e.cancelRun(st, "人工中止（拒绝动态调整）"), nil
		default:
			return nil, fmt.Errorf("unknown choice %q for adjustment request", decision.Choice)
		}
		st.Meta.Status = string(RunExecuting)
		_ = st.save()
		return e.resumeAfterHold(ctx, st)
	case HitlPermission:
		switch strings.ToLower(decision.Choice) {
		case "approve":
			// 授权记到角色（gate 上下文带角色名）。
			role := gate.Request.StepID
			for _, s := range st.Plan.Workflow.Steps {
				if s.ID == gate.Request.StepID {
					role = s.Role
				}
			}
			st.Approved["role:"+role] = true
			st.Meta.Status = string(RunExecuting)
			_ = st.save()
			return e.resumeAfterHold(ctx, st)
		case "reject":
			// 步骤失败（关键性语义由 failStep 处理）。
			for i := range st.Plan.Workflow.Steps {
				if st.Plan.Workflow.Steps[i].ID == gate.Request.StepID {
					st.Plan.Workflow.Steps[i].Status = "failed"
					st.Plan.Workflow.Steps[i].Error = "permission denied by human"
					st.Meta.Status = string(RunExecuting)
					_ = st.save()
					out, _, err := drv.failStepResume(ctx, st, i)
					if err != nil {
						return nil, err
					}
					e.finalizeEvent(out)
					return out, nil
				}
			}
			return e.cancelRun(st, "权限被拒绝且步骤不可定位"), nil
		default:
			return e.cancelRun(st, "人工中止（权限审批）"), nil
		}
	case HitlQualityReview:
		switch strings.ToLower(decision.Choice) {
		case "approve":
			return e.manualPass(st, decision)
		case "retry":
			// 追加一次修复预算并重跑交付步骤。
			st.RepairBudgetRemaining++
			st.Meta.Status = string(RunExecuting)
			_ = st.save()
			deliverStepID := lastCompletedStepID(st)
			out, _, err := drv.rerunForRepair(ctx, st, deliverStepID)
			if err != nil {
				return nil, err
			}
			e.finalizeEvent(out)
			return out, nil
		default:
			st.Meta.Status = string(RunFailed)
			st.Meta.CompletedAt = &now
			wf := st.Workflow()
			wf.Status = "failed"
			st.setWorkflow(wf)
			_ = st.save()
			e.deps.Sink.Emit(Event{Type: EvRunFailed, RunID: st.Meta.ID, Text: "人工裁决质检失败", At: time.Now()})
			return &RunResult{RunID: st.Meta.ID, Status: RunFailed, Plan: planPtr(st),
				Error: "quality review rejected by human", WorkDir: st.Meta.WorkDir}, nil
		}
	default:
		return nil, fmt.Errorf("unsupported request kind %q", req.Kind)
	}
}

// respondClarification 澄清回应：答案并入目标后重新规划。
func (e *SquadEngine) respondClarification(ctx context.Context, st *runState, gate *hitlGate, decision RespondDecision) (*RunResult, error) {
	choice := strings.ToLower(decision.Choice)
	if choice == "abort" {
		return e.cancelRun(st, "人工中止（澄清阶段）"), nil
	}
	answer := strings.TrimSpace(decision.Answer)
	if answer == "" && len(gate.Request.Options) > 0 {
		// 无自由文本时尝试取选项 Label。
		for _, o := range gate.Request.Options {
			if strings.EqualFold(o.Value, decision.Choice) {
				answer = o.Label
			}
		}
	}
	if answer != "" {
		st.Clarifications = append(st.Clarifications, answer)
		st.Meta.Goal.Text = st.Meta.Goal.Text + "\n【人工澄清】" + answer
		if st.Meta.WorkDir != "" {
			_ = e.ws.UpdateGoalFile(st.Meta.WorkDir, &st.Meta.Goal)
		}
	}
	if choice == "" || choice == "continue" || choice == "minimal" {
		if choice == "minimal" {
			// 人工选择最小编制：引导注册 4 个通用角色（backend 用运行默认）。
			e.bootstrapMinimalSquad(st)
		}
		st.Meta.Status = string(RunPlanning)
		_ = st.save()
		runOpts := RunOptions{TaskKey: st.Meta.Metadata["task_key"]}
		return e.planAndExecute(ctx, st, runOpts)
	}
	return e.cancelRun(st, "人工中止（澄清阶段）"), nil
}

// bootstrapMinimalSquad 注册通用最小编制（organizer/specialist/writer/leader）。
func (e *SquadEngine) bootstrapMinimalSquad(st *runState) {
	backend := st.Meta.Backend
	if backend == "" {
		backend = e.opts.DefaultBackend
	}
	for _, spec := range []RoleSpec{
		{Role: "organizer", Name: "梳理专员", Backend: backend, Keywords: []string{"整理", "梳理", "事实"},
			Responsibilities: []string{"梳理目标背景与输入，回填事实与缺口"}},
		{Role: "specialist", Name: "专项专员", Backend: backend, Keywords: []string{"分析", "推演", "方案"},
			Responsibilities: []string{"围绕目标开展专业分析与推演"}},
		{Role: "writer", Name: "撰写专员", Backend: backend, Keywords: []string{"撰写", "成文", "报告"},
			Responsibilities: []string{"汇总产出并撰写交付物"}},
		{Role: "leader", Name: "审核专员", Backend: backend, Keywords: []string{"审核", "终审", "把关"},
			Responsibilities: []string{"一致性终审与交付把关"}},
	} {
		_ = e.opts.Registry.Register(spec)
	}
}

// manualPass 人工放行质检：标记 ManualPass 并收口。
func (e *SquadEngine) manualPass(st *runState, decision RespondDecision) (*RunResult, error) {
	report := e.deps.Checker.Check(CheckRequest{
		RunID: st.Meta.ID, Workspace: st.Meta.WorkDir, Deliverable: "output.md",
		Plan: planPtr(st), Squad: &st.Plan.Squad, Template: st.Template,
	})
	report.ManualPass = true
	note := strings.TrimSpace(decision.Answer)
	if note != "" {
		report.Violations = append(report.Violations, "人工放行说明："+note)
	}
	status := RunCompleted
	if countStatus(st, "failed") > 0 || countStatus(st, "skipped") > 0 {
		status = RunDegraded
	}
	now := time.Now()
	st.Meta.Status = string(status)
	st.Meta.CompletedAt = &now
	wf := st.Workflow()
	wf.Status = string(status)
	st.setWorkflow(wf)
	_ = st.save()
	manifest := buildManifest(st, "output.md", *report)
	_ = writeJSON(filepath.Join(st.Meta.WorkDir, "manifest.json"), manifest)
	e.deps.Sink.Emit(Event{Type: EvRunCompleted, RunID: st.Meta.ID, Text: string(status), At: time.Now()})
	return &RunResult{RunID: st.Meta.ID, Status: status, Plan: planPtr(st),
		Deliverable: "output.md", Quality: report, WorkDir: st.Meta.WorkDir,
		Summary: summarize(st, *report, status) + "（质检人工放行）"}, nil
}

func (e *SquadEngine) cancelRun(st *runState, reason string) *RunResult {
	now := time.Now()
	st.Meta.Status = string(RunCancelled)
	st.Meta.CompletedAt = &now
	wf := st.Workflow()
	wf.Status = "cancelled"
	st.setWorkflow(wf)
	_ = st.save()
	e.deps.Sink.Emit(Event{Type: EvRunCancelled, RunID: st.Meta.ID, Text: reason, At: time.Now()})
	return &RunResult{RunID: st.Meta.ID, Status: RunCancelled, Plan: planPtr(st),
		Summary: reason, WorkDir: st.Meta.WorkDir}
}

// failStepResume 权限拒绝后的失败步骤处理（关键失败/非关键降级）。
func (d *flowDriver) failStepResume(ctx context.Context, st *runState, idx int) (*RunResult, *hitlGate, error) {
	return d.failStep(ctx, st, idx)
}

func (e *SquadEngine) respondStepApproval(ctx context.Context, st *runState, req *HumanRequest, decision RespondDecision) (*RunResult, error) {
	switch strings.ToLower(decision.Choice) {
	case "approve":
		st.Approved["step:"+req.StepID] = true
	case "skip":
		if !markStep(st, req.StepID, "skipped", "") {
			return nil, fmt.Errorf("step %s not found", req.StepID)
		}
	default:
		return e.cancelRun(st, "人工中止（步骤审批）"), nil
	}
	st.Meta.Status = string(RunExecuting)
	if err := st.save(); err != nil {
		return nil, err
	}
	return e.resumeAfterHold(ctx, st)
}

func (e *SquadEngine) respondInterrupted(ctx context.Context, st *runState, req *HumanRequest, decision RespondDecision) (*RunResult, error) {
	switch strings.ToLower(decision.Choice) {
	case "retry":
		if !markStep(st, req.StepID, "pending", "") {
			return nil, fmt.Errorf("step %s not found", req.StepID)
		}
	case "skip":
		if !markStep(st, req.StepID, "skipped", "") {
			return nil, fmt.Errorf("step %s not found", req.StepID)
		}
	default:
		return e.cancelRun(st, "人工中止（中断恢复）"), nil
	}
	st.Meta.Status = string(RunExecuting)
	if err := st.save(); err != nil {
		return nil, err
	}
	return e.resumeAfterHold(ctx, st)
}

func markStep(st *runState, id, status, errText string) bool {
	for i := range st.Plan.Workflow.Steps {
		if st.Plan.Workflow.Steps[i].ID != id {
			continue
		}
		st.Plan.Workflow.Steps[i].Status = status
		st.Plan.Workflow.Steps[i].Error = errText
		return true
	}
	return false
}

// recoverInterrupted 处理崩溃时停在 running 的步骤。
// interactive 挂起确认，不自动再执行；auto 允许一次重试；disabled 直接失败。
func (e *SquadEngine) recoverInterrupted(ctx context.Context, st *runState) (*RunResult, error) {
	for i := range st.Plan.Workflow.Steps {
		step := st.Plan.Workflow.Steps[i]
		if step.Status != "running" {
			continue
		}
		switch st.policy().mode() {
		case HitlDisabled:
			st.Plan.Workflow.Steps[i].Status = "failed"
			st.Plan.Workflow.Steps[i].Error = "interrupted"
			if err := st.save(); err != nil {
				return nil, err
			}
			drv := &flowDriver{opts: e.deps}
			out, _, err := drv.failStep(ctx, st, i)
			return out, err
		case HitlAuto:
			if step.Attempts >= 2 {
				st.Plan.Workflow.Steps[i].Status = "failed"
				st.Plan.Workflow.Steps[i].Error = "interrupted twice"
				if err := st.save(); err != nil {
					return nil, err
				}
				drv := &flowDriver{opts: e.deps}
				out, _, err := drv.failStep(ctx, st, i)
				return out, err
			}
			st.Plan.Workflow.Steps[i].Status = "pending"
			if err := st.save(); err != nil {
				return nil, err
			}
			return nil, nil
		default:
			drv := &flowDriver{opts: e.deps}
			out, _, err := drv.hold(st, HitlConfirmation, step.ID, step.Role,
				fmt.Sprintf("步骤「%s」在执行中中断。如何继续？", step.Title),
				[]HumanOption{
					{Value: "retry", Label: "重试"},
					{Value: "skip", Label: "跳过该步骤"},
					{Value: "abort", Label: "中止运行"},
				},
				"interrupted", "interrupted", nil)
			return out, err
		}
	}
	return nil, nil
}

// Resume 恢复运行：有挂起则返回挂起状态；未完成则继续推进（重启恢复入口）。
func (e *SquadEngine) Resume(ctx context.Context, runID string) (*RunResult, error) {
	release, err := e.acquire(ctx, runID)
	if err != nil {
		return nil, err
	}
	defer release()

	wsDir := filepath.Join(e.opts.DataDir, "runs", runID)
	st, err := loadRunState(wsDir)
	if err != nil {
		return nil, fmt.Errorf("load run %s: %w", runID, err)
	}
	if gate, _ := loadGate(wsDir); gate != nil && gate.Request.Status == "pending" {
		return &RunResult{RunID: runID, Status: RunStatus(st.Meta.Status), Plan: planPtr(st),
			Pending: &gate.Request, WorkDir: st.Meta.WorkDir}, nil
	}
	if RunStatus(st.Meta.Status).Terminal() {
		return e.Inspect(runID)
	}
	if out, err := e.recoverInterrupted(ctx, st); out != nil || err != nil {
		return out, err
	}
	return e.resumeAfterHold(ctx, st)
}

// Inspect 查看运行当前状态（含挂起请求与黑板）。
func (e *SquadEngine) Inspect(runID string) (*RunResult, error) {
	if !ValidID(runID) {
		return nil, invalidID(runID)
	}
	wsDir := filepath.Join(e.opts.DataDir, "runs", runID)
	st, err := loadRunState(wsDir)
	if err != nil {
		return nil, fmt.Errorf("load run %s: %w", runID, err)
	}
	out := &RunResult{
		RunID: runID, Status: RunStatus(st.Meta.Status), Plan: planPtr(st),
		Squad: &st.Plan.Squad, WorkDir: st.Meta.WorkDir, Summary: st.Meta.Status,
	}
	if gate, _ := loadGate(wsDir); gate != nil {
		out.Pending = &gate.Request
	}
	if data, err := os.ReadFile(filepath.Join(wsDir, "output.md")); err == nil && len(data) > 0 {
		out.Deliverable = "output.md"
	}
	if data, err := os.ReadFile(filepath.Join(wsDir, "manifest.json")); err == nil {
		var m Manifest
		if json.Unmarshal(data, &m) == nil {
			out.Quality = &m.QualityReport
		}
	}
	return out, nil
}

// Blackboard 查看运行黑板。
func (e *SquadEngine) Blackboard(runID string) (*BlackboardState, error) {
	if !ValidID(runID) {
		return nil, invalidID(runID)
	}
	wsDir := filepath.Join(e.opts.DataDir, "runs", runID)
	return loadBlackboard(wsDir)
}

// ListRuns 列出全部运行（按创建时间倒序）。
func (e *SquadEngine) ListRuns() ([]RunMetadata, error) {
	runsDir := filepath.Join(e.opts.DataDir, "runs")
	entries, err := os.ReadDir(runsDir)
	if err != nil {
		if os.IsNotExist(err) {
			return nil, nil
		}
		return nil, err
	}
	var out []RunMetadata
	for _, ent := range entries {
		if !ent.IsDir() {
			continue
		}
		st, err := loadRunState(filepath.Join(runsDir, ent.Name()))
		if err != nil {
			continue
		}
		out = append(out, st.Meta)
	}
	sort.Slice(out, func(i, j int) bool { return out[i].CreatedAt.After(out[j].CreatedAt) })
	return out, nil
}

// Pending 返回运行当前挂起请求（无则 nil）。
func (e *SquadEngine) Pending(runID string) (*HumanRequest, error) {
	if !ValidID(runID) {
		return nil, invalidID(runID)
	}
	gate, err := loadGate(filepath.Join(e.opts.DataDir, "runs", runID))
	if err != nil || gate == nil {
		return nil, err
	}
	return &gate.Request, nil
}

// Steer 向运行中的任务追加人工补充指令：在下一次步骤边界注入当前运行的
// prompt（步骤间注入，不中断正在执行的步骤），落盘 steering.json。
func (e *SquadEngine) Steer(runID, text string) error {
	if strings.TrimSpace(text) == "" {
		return fmt.Errorf("steering text is empty")
	}
	release, err := e.acquire(context.Background(), runID)
	if err != nil {
		return err
	}
	defer release()

	st, err := loadRunState(filepath.Join(e.opts.DataDir, "runs", runID))
	if err != nil {
		return fmt.Errorf("load run %s: %w", runID, err)
	}
	if RunStatus(st.Meta.Status).Terminal() {
		return fmt.Errorf("run %s already terminal (%s)", runID, st.Meta.Status)
	}
	st.Steering = append(st.Steering, strings.TrimSpace(text))
	_ = writeJSON(filepath.Join(st.Meta.WorkDir, "steering.json"), st.Steering)
	e.deps.Sink.Emit(Event{Type: "steered", RunID: runID, Text: truncate(text, 80), At: time.Now()})
	return nil
}

// SetBlackboardField 写入黑板字段（外部事实注入；可满足 field_set 等待条件）。
func (e *SquadEngine) SetBlackboardField(runID, key, value string) error {
	if strings.TrimSpace(key) == "" {
		return fmt.Errorf("field key is empty")
	}
	release, err := e.acquire(context.Background(), runID)
	if err != nil {
		return err
	}
	defer release()

	st, err := loadRunState(filepath.Join(e.opts.DataDir, "runs", runID))
	if err != nil {
		return fmt.Errorf("load run %s: %w", runID, err)
	}
	st.BB.Fields[key] = value
	if err := saveBlackboard(st.Meta.WorkDir, &st.BB); err != nil {
		return err
	}
	e.deps.Sink.Emit(Event{Type: "blackboard_field_set", RunID: runID,
		Text: key + "=" + truncate(value, 60), At: time.Now()})
	return nil
}

// SweepWakeups 扫描所有挂起运行：等待条件（run_done/field_set）满足时
// 合成自动决策续跑。护栏：同一请求自动唤醒次数达 MaxWakeupFires 即自动中止。
// 由 serve 模式周期调用，也可在任何运行终态后手动调用。返回唤醒数。
func (e *SquadEngine) SweepWakeups(ctx context.Context) int {
	runs, err := e.ListRuns()
	if err != nil {
		return 0
	}
	woken := 0
	for _, meta := range runs {
		if !RunStatus(meta.Status).IsHeld() {
			continue
		}
		gate, err := loadGate(filepath.Join(e.opts.DataDir, "runs", meta.ID))
		if err != nil || gate == nil || gate.Request.Status != "pending" || gate.Request.WaitCondition == nil {
			continue
		}
		release, err := e.acquire(ctx, meta.ID)
		if err != nil {
			continue
		}
		if e.fireWakeup(ctx, meta.ID) {
			woken++
		}
		release()
	}
	return woken
}

// SweepExpired 把超过 HITL 时限的挂起请求合成 fail 或 cancel 决策。
func (e *SquadEngine) SweepExpired(ctx context.Context, now time.Time) int {
	runs, err := e.ListRuns()
	if err != nil {
		return 0
	}
	n := 0
	for _, meta := range runs {
		if !RunStatus(meta.Status).IsHeld() {
			continue
		}
		gate, err := loadGate(filepath.Join(e.opts.DataDir, "runs", meta.ID))
		if err != nil || gate == nil || gate.Request.Status != "pending" {
			continue
		}
		decision, due := expiredDecision(gate.Request, e.deps.HitlPolicy, now)
		if !due {
			continue
		}
		if _, err := e.Respond(ctx, meta.ID, decision); err == nil {
			n++
		}
	}
	return n
}

// fireWakeup 在已持有 run 锁时处理一条挂起。只自动放行 gap 与 escalation。
func (e *SquadEngine) fireWakeup(ctx context.Context, runID string) bool {
	gate, err := loadGate(filepath.Join(e.opts.DataDir, "runs", runID))
	if err != nil || gate == nil || gate.Request.Status != "pending" || gate.Request.WaitCondition == nil {
		return false
	}
	met, fact := e.evaluateWaitCondition(runID, gate.Request.WaitCondition)
	if !met {
		return false
	}
	gate.Request.WakeupFires++
	if err := writeJSON(filepath.Join(e.opts.DataDir, "runs", runID, "pending.json"), gate); err != nil {
		return false
	}
	if gate.Request.WakeupFires >= MaxWakeupFires {
		_, err := e.respondLocked(ctx, runID, RespondDecision{RequestID: gate.Request.ID, Choice: "abort",
			Answer: "自动唤醒次数达上限，自动中止"})
		return err == nil
	}
	switch gate.Request.Kind {
	case HitlGap, HitlEscalation:
		decision := RespondDecision{RequestID: gate.Request.ID, Choice: "resolved", Answer: "[自动唤醒] " + fact}
		if _, err := e.respondLocked(ctx, runID, decision); err != nil {
			return false
		}
		e.deps.Sink.Emit(Event{Type: "wakeup_fired", RunID: runID, Text: fact, At: time.Now()})
		return true
	default:
		e.deps.Sink.Emit(Event{Type: "wakeup_held", RunID: runID, Text: fact, At: time.Now()})
		return false
	}
}

// evaluateWaitCondition 评估等待条件；返回是否满足与事实描述。
func (e *SquadEngine) evaluateWaitCondition(runID string, cond *WaitCondition) (bool, string) {
	switch cond.Type {
	case "run_done":
		if cond.RunID == "" || cond.RunID == runID {
			return false, "" // 自引用不触发
		}
		target, err := e.Inspect(cond.RunID)
		if err != nil {
			return false, ""
		}
		if target.Status.Terminal() {
			return true, fmt.Sprintf("等待的运行 %s 已进入终态 %s（唤醒报告事实，非业务完成）", cond.RunID, target.Status)
		}
	case "field_set":
		if cond.Field == "" {
			return false, ""
		}
		bb, err := e.Blackboard(runID)
		if err != nil {
			return false, ""
		}
		if v, ok := bb.Fields[cond.Field]; ok && strings.TrimSpace(v) != "" {
			return true, fmt.Sprintf("黑板字段 %s 已设置：%s", cond.Field, truncate(v, 80))
		}
	}
	return false, ""
}

func (e *SquadEngine) finalizeEvent(out *RunResult) {
	if out == nil {
		return
	}
	switch out.Status {
	case RunCompleted, RunDegraded:
		e.deps.Sink.Emit(Event{Type: EvRunCompleted, RunID: out.RunID, Text: out.Summary, At: time.Now()})
	case RunFailed:
		e.deps.Sink.Emit(Event{Type: EvRunFailed, RunID: out.RunID, Text: out.Error, At: time.Now()})
	}
}

// ---- 小工具 ----

func sha256File(path string) string {
	data, err := os.ReadFile(path)
	if err != nil {
		return ""
	}
	sum := sha256.Sum256(data)
	return hex.EncodeToString(sum[:])
}

func marshalJSON(v any) json.RawMessage {
	data, err := json.Marshal(v)
	if err != nil {
		return json.RawMessage(`{}`)
	}
	return data
}

func bpPtr(bp Blueprint) *Blueprint {
	if bp.Key == "" {
		return nil
	}
	return &bp
}

func planPtr2(p *SquadPlan) *WorkflowPlan {
	wf := p.Workflow
	return &wf
}

func answerFor(g GapItem, answer string) string {
	if answer == "" {
		return "人工确认按现状继续"
	}
	return answer
}

func appendAnswer(d RespondDecision) string {
	if strings.TrimSpace(d.Answer) == "" {
		return ""
	}
	return "（" + d.Answer + "）"
}

func lastCompletedStepID(st *runState) string {
	id := ""
	for _, s := range st.Plan.Workflow.Steps {
		if s.Status == "completed" {
			id = s.ID
		}
	}
	return id
}