fix(agent): refresh live status from AM in the agent LIST endpoint

GET /api/heicode/agents previously returned only the stored status, so a
freshly-deployed agent stayed "Pending"/"启动中" forever — the desktop client
polling the list never saw it go "running", and the web list only updated when
the user opened the detail panel (which polls /agents/{id}/status and was the
only path that refreshed from AM).

Now the list refreshes every non-terminal (non-"stopped") agent's live status
from AM before responding, concurrently under an 8s deadline. Per-user agents
are capped (≤8) so the fan-out is bounded and the list never hangs on a slow
AM. refreshAgentStatus now takes context.Context (callers pass the request ctx).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

heicode/controller/agent_template_handlers.go

@@ -1,9 +1,11 @@
 package controller
 
 import (
+	"context"
 	"crypto/subtle"
 	"strconv"
 	"strings"
+	"sync"
 	"time"
 
 	"github.com/gin-gonic/gin"
@@ -256,6 +258,14 @@ func HeicodeListAgents(c *gin.Context) {
 		agentError(c, "DEPLOYMENT_CONFLICT", "failed to list agents")
 		return
 	}
+	// Refresh live status from AM for every non-terminal agent BEFORE returning,
+	// so the list (and the desktop client that polls it) reflects reality without
+	// needing the detail panel to be opened. Previously only the detail/status
+	// endpoints refreshed, so a freshly-deployed agent stayed "Pending" in the
+	// list forever. Bounded: per-user agent count is capped (≤8) and the refreshes
+	// run concurrently under a short deadline, so the list never hangs on a slow AM.
+	refreshAgentStatusBatch(c.Request.Context(), rows)
+
 	items := make([]gin.H, 0, len(rows))
 	for _, row := range rows {
 		items = append(items, templateAgentResponse(row))
@@ -263,9 +273,31 @@ func HeicodeListAgents(c *gin.Context) {
 	common.ApiSuccess(c, gin.H{"items": items, "total": len(items)})
 }
 
+// refreshAgentStatusBatch concurrently refreshes the live status of every
+// non-terminal agent in rows (best-effort) under a single bounded deadline.
+// Each row is updated in place + persisted by refreshAgentStatus.
+func refreshAgentStatusBatch(ctx context.Context, rows []model.AgentDeployment) {
+	ctx, cancel := context.WithTimeout(ctx, 8*time.Second)
+	defer cancel()
+	var wg sync.WaitGroup
+	for i := range rows {
+		row := &rows[i]
+		if strings.TrimSpace(row.RuntimeDeploymentID) == "" ||
+			strings.EqualFold(strings.TrimSpace(row.Status), "stopped") {
+			continue
+		}
+		wg.Add(1)
+		go func(r *model.AgentDeployment) {
+			defer wg.Done()
+			refreshAgentStatus(ctx, r)
+		}(row)
+	}
+	wg.Wait()
+}
+
 // refreshAgentStatus best-effort pulls the live status from AM and persists it.
 // On any AM error it keeps the last-known status (never blocks the read).
-func refreshAgentStatus(c *gin.Context, row *model.AgentDeployment) {
+func refreshAgentStatus(ctx context.Context, row *model.AgentDeployment) {
 	if strings.TrimSpace(row.RuntimeDeploymentID) == "" {
 		return
 	}
@@ -275,7 +307,7 @@ func refreshAgentStatus(c *gin.Context, row *model.AgentDeployment) {
 	if strings.EqualFold(strings.TrimSpace(row.Status), "stopped") {
 		return
 	}
-	status, err := amGetAgentStatus(c.Request.Context(), row.RuntimeDeploymentID)
+	status, err := amGetAgentStatus(ctx, row.RuntimeDeploymentID)
 	if err != nil || strings.TrimSpace(status) == "" || status == row.Status {
 		return
 	}
@@ -299,7 +331,7 @@ func HeicodeGetAgent(c *gin.Context) {
 	if !ok {
 		return
 	}
-	refreshAgentStatus(c, &row)
+	refreshAgentStatus(c.Request.Context(), &row)
 	common.ApiSuccess(c, templateAgentResponse(row))
 }
 
@@ -310,7 +342,7 @@ func HeicodeGetAgentStatus(c *gin.Context) {
 	if !ok {
 		return
 	}
-	refreshAgentStatus(c, &row)
+	refreshAgentStatus(c.Request.Context(), &row)
 	common.ApiSuccess(c, gin.H{
 		"agent_id":   row.DeploymentID,
 		"status":     row.Status,