add child-model agent_task flow with node model capability and stricter device contracts

README.md

@@ -50,6 +50,7 @@
 - `nodes` 工具支持设备快捷参数：`facing`、`duration_ms`、`command`
 - 设备动作响应统一：`ok/code/error/payload`（code 示例：`ok` `unsupported_action` `transport_error`）
 - 设备 `payload` 规范字段：`media_type` `storage` `url|path|image` `meta`
+- 支持 `agent_task`：主节点可向具备 `model` 能力的子节点下发任务，子节点返回执行结果
 
 实现位置：
 - `pkg/nodes/types.go`

README_EN.md

@@ -50,6 +50,7 @@ A `nodes` tool control-plane PoC is now available:
 - `nodes` tool supports device shortcuts: `facing`, `duration_ms`, `command`
 - unified device response envelope: `ok/code/error/payload` (code examples: `ok`, `unsupported_action`, `transport_error`)
 - device `payload` normalized fields: `media_type` `storage` `url|path|image` `meta`
+- supports `agent_task`: parent node can dispatch tasks to child nodes with `model` capability and receive execution results
 
 Implementation:
 - `pkg/nodes/types.go`

cmd/clawgo/cmd_status.go

@@ -162,19 +162,20 @@ func statusCmd() {
 		ns := nodes.DefaultManager().List()
 		if len(ns) > 0 {
 			online := 0
-			caps := map[string]int{"run": 0, "camera": 0, "screen": 0, "location": 0, "canvas": 0}
+			caps := map[string]int{"run": 0, "model": 0, "camera": 0, "screen": 0, "location": 0, "canvas": 0}
 			for _, n := range ns {
 				if n.Online {
 					online++
 				}
 				if n.Capabilities.Run { caps["run"]++ }
+				if n.Capabilities.Model { caps["model"]++ }
 				if n.Capabilities.Camera { caps["camera"]++ }
 				if n.Capabilities.Screen { caps["screen"]++ }
 				if n.Capabilities.Location { caps["location"]++ }
 				if n.Capabilities.Canvas { caps["canvas"]++ }
 			}
 			fmt.Printf("Nodes: total=%d online=%d\n", len(ns), online)
-			fmt.Printf("Nodes Capabilities: run=%d camera=%d screen=%d location=%d canvas=%d\n", caps["run"], caps["camera"], caps["screen"], caps["location"], caps["canvas"])
+			fmt.Printf("Nodes Capabilities: run=%d model=%d camera=%d screen=%d location=%d canvas=%d\n", caps["run"], caps["model"], caps["camera"], caps["screen"], caps["location"], caps["canvas"])
 		}
 	}
 }

pkg/agent/loop.go

@@ -79,7 +79,7 @@ func NewAgentLoop(cfg *config.Config, msgBus *bus.MessageBus, provider providers
 	toolsRegistry.Register(tools.NewProcessTool(processManager))
 	nodesManager := nodes.DefaultManager()
 	nodesManager.SetAuditPath(filepath.Join(workspace, "memory", "nodes-audit.jsonl"))
-	nodesManager.Upsert(nodes.NodeInfo{ID: "local", Name: "local", Capabilities: nodes.Capabilities{Run: true, Invoke: true, Camera: true, Screen: true, Location: true, Canvas: true}, Online: true})
+	nodesManager.Upsert(nodes.NodeInfo{ID: "local", Name: "local", Capabilities: nodes.Capabilities{Run: true, Invoke: true, Model: true, Camera: true, Screen: true, Location: true, Canvas: true}, Models: []string{"local-sim"}, Online: true})
 	nodesManager.RegisterHandler("local", func(req nodes.Request) nodes.Response {
 		switch req.Action {
 		case "run":
@@ -92,6 +92,8 @@ func NewAgentLoop(cfg *config.Config, msgBus *bus.MessageBus, provider providers
 				payload["command"] = parts
 			}
 			return nodes.Response{OK: true, Code: "ok", Node: "local", Action: req.Action, Payload: payload}
+		case "agent_task":
+			return nodes.Response{OK: true, Code: "ok", Node: "local", Action: req.Action, Payload: map[string]interface{}{"transport": "relay-local", "simulated": true, "model": req.Model, "task": req.Task, "result": "local child-model simulated execution completed"}}
 		case "camera_snap":
 			return nodes.Response{OK: true, Code: "ok", Node: "local", Action: req.Action, Payload: map[string]interface{}{"transport": "relay-local", "media_type": "image", "storage": "inline", "facing": req.Args["facing"], "simulated": true, "meta": map[string]interface{}{"width": 1280, "height": 720}}}
 		case "camera_clip":

pkg/nodes/manager.go

@@ -137,6 +137,8 @@ func (m *Manager) SupportsAction(nodeID, action string) bool {
 	switch action {
 	case "run":
 		return n.Capabilities.Run
+	case "agent_task":
+		return n.Capabilities.Model
 	case "camera_snap", "camera_clip":
 		return n.Capabilities.Camera
 	case "screen_record", "screen_snapshot":
@@ -162,6 +164,10 @@ func (m *Manager) PickFor(action string) (NodeInfo, bool) {
 			if n.Capabilities.Run {
 				return n, true
 			}
+		case "agent_task":
+			if n.Capabilities.Model {
+				return n, true
+			}
 		case "camera_snap", "camera_clip":
 			if n.Capabilities.Camera {
 				return n, true

pkg/nodes/transport.go

@@ -75,6 +75,8 @@ func actionHTTPPath(action string) string {
 		return "/run"
 	case "invoke":
 		return "/invoke"
+	case "agent_task":
+		return "/agent/task"
 	case "camera_snap":
 		return "/camera/snap"
 	case "camera_clip":

pkg/nodes/types.go

@@ -6,6 +6,7 @@ import "time"
 type Capabilities struct {
 	Run      bool `json:"run"`
 	Invoke   bool `json:"invoke"`
+	Model    bool `json:"model"`
 	Camera   bool `json:"camera"`
 	Screen   bool `json:"screen"`
 	Location bool `json:"location"`
@@ -23,6 +24,7 @@ type NodeInfo struct {
 	Token        string       `json:"token,omitempty"`
 	Capabilities Capabilities `json:"capabilities"`
 	Actions      []string     `json:"actions,omitempty"`
+	Models       []string     `json:"models,omitempty"`
 	RegisteredAt time.Time    `json:"registered_at,omitempty"`
 	LastSeenAt   time.Time    `json:"last_seen_at"`
 	Online       bool         `json:"online"`
@@ -32,6 +34,8 @@ type NodeInfo struct {
 type Request struct {
 	Action string                 `json:"action"`
 	Node   string                 `json:"node,omitempty"`
+	Task   string                 `json:"task,omitempty"`
+	Model  string                 `json:"model,omitempty"`
 	Args   map[string]interface{} `json:"args,omitempty"`
 }
 

pkg/tools/nodes_tool.go

@@ -22,10 +22,12 @@ func (t *NodesTool) Description() string {
 }
 func (t *NodesTool) Parameters() map[string]interface{} {
 	return map[string]interface{}{"type": "object", "properties": map[string]interface{}{
-		"action": map[string]interface{}{"type": "string", "description": "status|describe|run|invoke|camera_snap|camera_clip|screen_record|screen_snapshot|location_get|canvas_snapshot|canvas_action"},
+		"action": map[string]interface{}{"type": "string", "description": "status|describe|run|invoke|agent_task|camera_snap|camera_clip|screen_record|screen_snapshot|location_get|canvas_snapshot|canvas_action"},
 		"node":   map[string]interface{}{"type": "string", "description": "target node id"},
 		"mode":   map[string]interface{}{"type": "string", "description": "auto|p2p|relay (default auto)"},
 		"args":   map[string]interface{}{"type": "object", "description": "action args"},
+		"task":   map[string]interface{}{"type": "string", "description": "agent_task content for child node model"},
+		"model":  map[string]interface{}{"type": "string", "description": "optional model for agent_task"},
 		"command": map[string]interface{}{"type": "array", "description": "run command array shortcut"},
 		"facing":  map[string]interface{}{"type": "string", "description": "camera facing: front|back|both"},
 		"duration_ms": map[string]interface{}{"type": "integer", "description": "clip/record duration"},
@@ -95,12 +97,17 @@ func (t *NodesTool) Execute(ctx context.Context, args map[string]interface{}) (s
 			}
 			reqArgs["duration_ms"] = di
 		}
+		task, _ := args["task"].(string)
+		model, _ := args["model"].(string)
+		if action == "agent_task" && strings.TrimSpace(task) == "" {
+			return "", fmt.Errorf("invalid_args: agent_task requires task")
+		}
 		if action == "canvas_action" {
 			if act, _ := reqArgs["action"].(string); strings.TrimSpace(act) == "" {
 				return "", fmt.Errorf("invalid_args: canvas_action requires args.action")
 			}
 		}
-		resp, err := t.router.Dispatch(ctx, nodes.Request{Action: action, Node: nodeID, Args: reqArgs}, mode)
+		resp, err := t.router.Dispatch(ctx, nodes.Request{Action: action, Node: nodeID, Task: strings.TrimSpace(task), Model: strings.TrimSpace(model), Args: reqArgs}, mode)
 		if err != nil {
 			return "", err
 		}