Merge pull request #582 from Cai-Tang-www/feat/issue-578-ask-user-a

feat(runtime): implement ask_user tool for plan-mode user interaction

Author: pionxe

internal/app/bootstrap.go

@@ -230,6 +230,15 @@ func BuildGatewayServerDeps(ctx context.Context, opts BootstrapOptions) (Runtime
 		return RuntimeBundle{}, err
 	}
 
+	// Wire ask_user broker from runtime into the pre-registered ask_user tool.
+	if brokerAdapter := runtimeSvc.AskUserBrokerAdapter(); brokerAdapter != nil {
+		if askTool, err := toolRegistry.Get(tools.ToolNameAskUser); err == nil {
+			if setter, ok := askTool.(tools.AskUserBrokerSetter); ok {
+				setter.SetAskUserBroker(brokerAdapter)
+			}
+		}
+	}
+
 	// 注入记忆提取钩子：当 AutoExtract 启用且 memoSvc 可用时，ReAct 循环完成后异步提取记忆。
 	if memoSvc != nil && cfg.Memo.AutoExtract {
 		runtimeSvc.SetMemoExtractor(newMemoExtractorAdapter(
@@ -466,6 +475,7 @@ func buildToolRegistry(cfg config.Config) (*tools.Registry, func() error, error)
 	toolRegistry.Register(codebase.NewRead(repoSvc, cfg.Workdir))
 	toolRegistry.Register(codebase.NewSearchText(repoSvc, cfg.Workdir))
 	toolRegistry.Register(codebase.NewSearchSymbol(repoSvc, cfg.Workdir))
+	toolRegistry.Register(tools.NewAskUserTool(nil)) // broker injected after runtime creation
 	mcpRegistry, err := BuildMCPRegistry(cfg)
 	if err != nil {
 		return nil, nil, err

internal/app/bootstrap_test.go

@@ -2017,6 +2017,10 @@ func (s *stubRuntimeForBootstrap) ResolvePermission(context.Context, agentruntim
 	return nil
 }
 
+func (s *stubRuntimeForBootstrap) ResolveUserQuestion(context.Context, agentruntime.UserQuestionResolutionInput) error {
+	return nil
+}
+
 func (s *stubRuntimeForBootstrap) CancelActiveRun() bool {
 	return false
 }
@@ -2083,6 +2087,10 @@ func (s *stubRemoteRuntimeForBootstrap) ResolvePermission(context.Context, servi
 	return nil
 }
 
+func (s *stubRemoteRuntimeForBootstrap) ResolveUserQuestion(context.Context, services.UserQuestionResolutionInput) error {
+	return nil
+}
+
 func (s *stubRemoteRuntimeForBootstrap) CancelActiveRun() bool {
 	return false
 }

internal/cli/gateway_runtime_bridge.go

@@ -477,6 +477,19 @@ func (b *gatewayRuntimePortBridge) ResolvePermission(ctx context.Context, input
 	})
 }
 
+// ResolveUserQuestion 将网关 ask_user 回答转发到 runtime。
+func (b *gatewayRuntimePortBridge) ResolveUserQuestion(ctx context.Context, input gateway.UserQuestionAnswerInput) error {
+	if err := b.ensureRuntimeAccess(input.SubjectID); err != nil {
+		return err
+	}
+	return b.runtime.ResolveUserQuestion(ctx, agentruntime.UserQuestionResolutionInput{
+		RequestID: strings.TrimSpace(input.RequestID),
+		Status:    strings.TrimSpace(input.Status),
+		Values:    input.Values,
+		Message:   strings.TrimSpace(input.Message),
+	})
+}
+
 // CancelRun 转发 gateway.cancel 请求到 runtime 的 run_id 精确取消能力。
 func (b *gatewayRuntimePortBridge) CancelRun(_ context.Context, input gateway.CancelInput) (bool, error) {
 	if err := b.ensureRuntimeAccess(input.SubjectID); err != nil {

internal/cli/gateway_runtime_bridge_test.go

@@ -128,6 +128,10 @@ func (s *runtimeStub) ResolvePermission(_ context.Context, input agentruntime.Pe
 	return s.permissionErr
 }
 
+func (s *runtimeStub) ResolveUserQuestion(_ context.Context, input agentruntime.UserQuestionResolutionInput) error {
+	return nil
+}
+
 func (s *runtimeStub) CancelActiveRun() bool {
 	return s.cancelReturn
 }
@@ -238,6 +242,9 @@ func (r *runtimeWithoutCreator) ExecuteSystemTool(ctx context.Context, input age
 func (r *runtimeWithoutCreator) ResolvePermission(ctx context.Context, input agentruntime.PermissionResolutionInput) error {
 	return r.base.ResolvePermission(ctx, input)
 }
+func (r *runtimeWithoutCreator) ResolveUserQuestion(ctx context.Context, input agentruntime.UserQuestionResolutionInput) error {
+	return r.base.ResolveUserQuestion(ctx, input)
+}
 func (r *runtimeWithoutCreator) CancelActiveRun() bool {
 	return r.base.CancelActiveRun()
 }
@@ -310,6 +317,9 @@ func (r *runtimeWithoutCheckpointer) ExecuteSystemTool(ctx context.Context, inpu
 func (r *runtimeWithoutCheckpointer) ResolvePermission(ctx context.Context, input agentruntime.PermissionResolutionInput) error {
 	return r.base.ResolvePermission(ctx, input)
 }
+func (r *runtimeWithoutCheckpointer) ResolveUserQuestion(ctx context.Context, input agentruntime.UserQuestionResolutionInput) error {
+	return r.base.ResolveUserQuestion(ctx, input)
+}
 func (r *runtimeWithoutCheckpointer) CancelActiveRun() bool {
 	return r.base.CancelActiveRun()
 }

internal/cli/root_test.go

@@ -1051,6 +1051,10 @@ func (stubRuntimePort) ResolvePermission(context.Context, gateway.PermissionReso
 	return nil
 }
 
+func (stubRuntimePort) ResolveUserQuestion(context.Context, gateway.UserQuestionAnswerInput) error {
+	return nil
+}
+
 func (stubRuntimePort) CancelRun(context.Context, gateway.CancelInput) (bool, error) {
 	return false, nil
 }

internal/gateway/adapters/urlscheme/dispatcher_integration_unix_test.go

@@ -159,6 +159,13 @@ func (s *urlschemeIntegrationRuntimeStub) ResolvePermission(
 	return nil
 }
 
+func (s *urlschemeIntegrationRuntimeStub) ResolveUserQuestion(
+	context.Context,
+	gateway.UserQuestionAnswerInput,
+) error {
+	return nil
+}
+
 func (s *urlschemeIntegrationRuntimeStub) CancelRun(context.Context, gateway.CancelInput) (bool, error) {
 	return false, nil
 }

internal/gateway/bootstrap.go

@@ -1539,6 +1539,44 @@ func handleDeleteMCPServerFrame(ctx context.Context, frame MessageFrame, runtime
 	return MessageFrame{Type: FrameTypeAck, Action: FrameActionDeleteMCPServer, RequestID: frame.RequestID, Payload: map[string]any{"deleted": true, "id": input.ID}}
 }
 
+// handleUserQuestionAnswerFrame 处理 gateway.user_question_answer 请求。
+func handleUserQuestionAnswerFrame(ctx context.Context, frame MessageFrame, runtimePort RuntimePort) MessageFrame {
+	if runtimePort == nil {
+		return runtimePortUnavailableFrame(frame)
+	}
+	subjectID, subjectErr := requireAuthenticatedSubjectID(ctx)
+	if subjectErr != nil {
+		return errorFrame(frame, subjectErr)
+	}
+
+	input, err := decodeUserQuestionAnswerPayload(frame.Payload)
+	if err != nil {
+		return errorFrame(frame, NewFrameError(ErrorCodeInvalidAction, "invalid user_question_answer payload"))
+	}
+	input.SubjectID = subjectID
+	input.RequestID = strings.TrimSpace(input.RequestID)
+	if input.RequestID == "" {
+		return errorFrame(frame, NewMissingRequiredFieldError("payload.request_id"))
+	}
+
+	callCtx, cancel := withRuntimeOperationTimeout(ctx)
+	defer cancel()
+	if err := runtimePort.ResolveUserQuestion(callCtx, input); err != nil {
+		return runtimeCallFailedFrame(callCtx, frame, err, "user_question_answer")
+	}
+
+	return MessageFrame{
+		Type:      FrameTypeAck,
+		Action:    FrameActionUserQuestionAnswer,
+		RequestID: frame.RequestID,
+		Payload: map[string]any{
+			"request_id": input.RequestID,
+			"status":     input.Status,
+			"message":    "user question answered",
+		},
+	}
+}
+
 // handleResolvePermissionFrame 处理 gateway.resolvePermission 请求。
 func handleResolvePermissionFrame(ctx context.Context, frame MessageFrame, runtimePort RuntimePort) MessageFrame {
 	if runtimePort == nil {

internal/gateway/bootstrap_test.go

@@ -140,6 +140,10 @@ func (s *bootstrapRuntimeStub) ResolvePermission(ctx context.Context, input Perm
 	return nil
 }
 
+func (s *bootstrapRuntimeStub) ResolveUserQuestion(ctx context.Context, input UserQuestionAnswerInput) error {
+	return nil
+}
+
 func (s *bootstrapRuntimeStub) CancelRun(ctx context.Context, input CancelInput) (bool, error) {
 	if s != nil && s.cancelRunFn != nil {
 		return s.cancelRunFn(ctx, input)
@@ -4780,6 +4784,9 @@ func (runtimeOnlyStub) ListAvailableSkills(ctx context.Context, input ListAvaila
 func (runtimeOnlyStub) ResolvePermission(ctx context.Context, input PermissionResolutionInput) error {
 	return nil
 }
+func (runtimeOnlyStub) ResolveUserQuestion(ctx context.Context, input UserQuestionAnswerInput) error {
+	return nil
+}
 func (runtimeOnlyStub) CancelRun(ctx context.Context, input CancelInput) (bool, error) {
 	return false, nil
 }

internal/gateway/contracts.go

@@ -545,6 +545,20 @@ type DeleteProviderInput struct {
 	ProviderID string `json:"provider_id"`
 }
 
+// UserQuestionAnswerInput 表示一次 ask_user 回答的输入。
+type UserQuestionAnswerInput struct {
+	// SubjectID 是请求方身份主体标识。
+	SubjectID string `json:"subject_id,omitempty"`
+	// RequestID 是待回答的提问标识。
+	RequestID string `json:"request_id"`
+	// Status 是回答状态：answered / skipped。
+	Status string `json:"status"`
+	// Values 是用户选择的答案值。
+	Values []string `json:"values,omitempty"`
+	// Message 是用户的文本答案。
+	Message string `json:"message,omitempty"`
+}
+
 // SelectProviderModelInput 表示全局选择 provider/model 的输入。
 type SelectProviderModelInput struct {
 	// SubjectID 是请求方身份主体标识。
@@ -798,6 +812,8 @@ type RuntimePort interface {
 	ListAvailableSkills(ctx context.Context, input ListAvailableSkillsInput) ([]AvailableSkillState, error)
 	// ResolvePermission 向运行时提交一次权限审批决策。
 	ResolvePermission(ctx context.Context, input PermissionResolutionInput) error
+	// ResolveUserQuestion 向运行时提交一次 ask_user 回答。
+	ResolveUserQuestion(ctx context.Context, input UserQuestionAnswerInput) error
 	// CancelRun 按 run_id 精确取消运行态任务。
 	CancelRun(ctx context.Context, input CancelInput) (bool, error)
 	// Events 返回统一运行事件流。

internal/gateway/contracts_test.go

@@ -58,6 +58,10 @@ func (s *runtimePortCompileStub) ResolvePermission(_ context.Context, _ Permissi
 	return nil
 }
 
+func (s *runtimePortCompileStub) ResolveUserQuestion(_ context.Context, _ UserQuestionAnswerInput) error {
+	return nil
+}
+
 func (s *runtimePortCompileStub) CancelRun(_ context.Context, _ CancelInput) (bool, error) {
 	return false, nil
 }