fix(acp): T10/T11/T12 observability + error semantics; revert S1 (broke main)

T10: gatewayRPCError marks OPENCLAW_GATEWAY_SOCKET_CLOSED with retryable=true,
poll=true so the client degrades to "background/reconnecting" + keeps polling
instead of hard-failing (feeds App T5).
T11: runId-tagged warn logs at the tasks.get unconfirmed-fallback and
run-deadline-interrupt sites, so a runId can be joined across App→bridge→plugin→gateway.
T12: process-level stability counters (gatewaySocketClosed, taskGetUnconfirmedFallback,
runDeadlineInterrupt) exposed via /api/ping.metrics.

Revert S1 (default expectedArtifactDirs): it set requiresExport=true / default dirs
for any artifact-inferring task, which made a gateway run that succeeds with NO
artifact hang "waiting for artifact export" (TestHTTPHandlerGatewayOpenClawHandlesFive
ConcurrentE2ECases + ...WithoutPromptHeuristic went red). The blocking is tied to
expectedArtifactDirs presence in openClawTaskGetRequiresArtifactExport; decoupling
scan-hint from block-on-export needs a careful, separately-tested change. Reverted to
keep main green; S1 to be redesigned (see docs/cases/06 §7).

Full internal/acp suite green.

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

internal/acp/http_handler.go

@@ -37,6 +37,7 @@ func (s *Server) Handler() http.Handler {
 				"commit":    info.Commit,
 				"version":   info.Version,
 				"buildDate": info.BuildDate,
+				"metrics":   bridgeStabilityMetricsSnapshot(), // T12
 			}
 			body, _ := json.Marshal(resp)
 			w.Header().Set("Content-Type", "application/json")

internal/acp/metrics.go

@@ -0,0 +1,29 @@
+package acp
+
+import "sync/atomic"
+
+// 关键稳定性指标（T12，docs/cases/06 §5）。
+//
+// 进程内累计计数，经 /api/ping 暴露，用于把「网关抖动 / run 超时」从靠用户截图
+// 变为可监控。三个计数对应三类已知的不稳定来源：
+//   - gatewaySocketClosed       : gatewayRPCError 命中 OPENCLAW_GATEWAY_SOCKET_CLOSED（连接断）
+//   - taskGetUnconfirmedFallback: tasks.get 走持久 run 仓兜底（gateway 无法确认 run，T7）
+//   - runDeadlineInterrupt      : run 超过 DeadlineAt 且 gateway 无法确认，回 interrupted（T9）
+var bridgeStabilityMetrics struct {
+	gatewaySocketClosed        atomic.Int64
+	taskGetUnconfirmedFallback atomic.Int64
+	runDeadlineInterrupt       atomic.Int64
+}
+
+func metricGatewaySocketClosedInc()        { bridgeStabilityMetrics.gatewaySocketClosed.Add(1) }
+func metricTaskGetUnconfirmedFallbackInc() { bridgeStabilityMetrics.taskGetUnconfirmedFallback.Add(1) }
+func metricRunDeadlineInterruptInc()       { bridgeStabilityMetrics.runDeadlineInterrupt.Add(1) }
+
+// bridgeStabilityMetricsSnapshot 返回当前计数快照，供 /api/ping 输出。
+func bridgeStabilityMetricsSnapshot() map[string]any {
+	return map[string]any{
+		"gatewaySocketClosed":        bridgeStabilityMetrics.gatewaySocketClosed.Load(),
+		"taskGetUnconfirmedFallback": bridgeStabilityMetrics.taskGetUnconfirmedFallback.Load(),
+		"runDeadlineInterrupt":       bridgeStabilityMetrics.runDeadlineInterrupt.Load(),
+	}
+}

internal/acp/openclaw_run_registry.go

@@ -1,6 +1,7 @@
 package acp
 
 import (
+	"log"
 	"strings"
 	"time"
 
@@ -125,11 +126,15 @@ func (s *Server) openClawTaskGetGatewayUnconfirmedFallback(params map[string]any
 		return s.markOpenClawRunDeadlineInterruptedLocked(sess, code, message)
 	}
 	// 仍在预算内：合成 running 句柄让客户端继续轮询，不因一次瞬时抖动硬失败。
+	metricTaskGetUnconfirmedFallbackInc() // T12
 	running := openClawRunningTaskResult(sess.openClaw)
 	running["transportDegraded"] = true
 	if strings.TrimSpace(code) != "" {
 		running["transportDegradedCode"] = strings.TrimSpace(code)
 	}
+	// T11：带 runId 的日志，便于与 App / 插件 / 网关四层按 runId 串联。
+	log.Printf("level=warn component=openclaw_run_registry event=tasks_get_unconfirmed_fallback runId=%q openclawSessionKey=%q code=%q",
+		sess.openClaw.RunID, sess.openClaw.SessionKey, strings.TrimSpace(code))
 	sess.lastResult = cloneMap(running)
 	return running
 }
@@ -143,6 +148,13 @@ func (s *Server) markOpenClawRunDeadlineInterruptedLocked(sess *session, code st
 	sess.task.ProgressStage = "interrupted"
 	sess.task.ProgressMessage = "OpenClaw run exceeded its budget and could not be confirmed"
 	sess.task.UpdatedAt = now
+	metricRunDeadlineInterruptInc() // T12
+	// T11：带 runId 的终态日志。
+	if sess.openClaw != nil {
+		log.Printf("level=warn component=openclaw_run_registry event=run_deadline_interrupt runId=%q openclawSessionKey=%q deadlineAt=%q code=%q",
+			sess.openClaw.RunID, sess.openClaw.SessionKey,
+			sess.openClaw.DeadlineAt.UTC().Format(time.RFC3339Nano), strings.TrimSpace(code))
+	}
 
 	result := map[string]any{
 		"ok":                 true,

internal/acp/orchestrator.go

@@ -850,12 +850,6 @@ type openClawArtifactContract struct {
 	SourceMessage          string
 }
 
-// defaultOpenClawExpectedArtifactDirs 是 agent 最常用的产物落盘目录。当任务期望产物却没有
-// 显式声明目录时，用作 openclaw-multi-session-plugins 的 workspace 根兜底扫描范围（S1）。
-func defaultOpenClawExpectedArtifactDirs() []string {
-	return []string{"reports/", "artifacts/", "exports/"}
-}
-
 func openClawArtifactContractForParams(params map[string]any, chatParams map[string]any) openClawArtifactContract {
 	metadata := shared.AsMap(params["metadata"])
 	taskLoadClass := strings.TrimSpace(shared.StringArg(metadata, "taskLoadClass", ""))
@@ -875,14 +869,6 @@ func openClawArtifactContractForParams(params map[string]any, chatParams map[str
 	if len(requiredExts) == 0 {
 		requiredExts = inferOpenClawRequiredArtifactExts(lowerMessage)
 	}
-	// S1（docs/cases/06 §7）：任务期望产物（需导出 或 已推断出 requiredExts）却没有显式声明
-	// expectedArtifactDirs 时，补一组缺省目录。openclaw-multi-session-plugins 在 task scope
-	// 目录为空时会回扫 workspace 根的 expectedArtifactDirs；该列表为空则兜底形同虚设，
-	// agent 写到 workspace 根（reports//artifacts//exports/）的产物就再也收不回（表现「暂无文件」）。
-	if len(expectedDirs) == 0 && (requiresExport || len(requiredExts) > 0) {
-		expectedDirs = defaultOpenClawExpectedArtifactDirs()
-		requiresExport = true
-	}
 	expectedFileCounts := normalizeOpenClawArtifactExtCountMap(shared.AsMap(contract["expectedFileCountByExtension"]))
 	if len(expectedFileCounts) == 0 {
 		expectedFileCounts = normalizeOpenClawArtifactExtCountMap(shared.AsMap(metadata["expectedFileCountByExtension"]))
@@ -1701,11 +1687,16 @@ func applyOpenClawConstraintDeliveryStatus(result map[string]any) {
 
 func gatewayRPCError(errorPayload map[string]any, fallback string) *shared.RPCError {
 	if isOpenClawRetryableGatewayError(errorPayload) {
+		metricGatewaySocketClosedInc() // T12
+		// T10：连接断属「可重试 / run 可能仍在后台、可续轮询」语义，而非 run 确实失败。
+		// 带 retryable/poll 提示，客户端据此降级为「后台续跑·重连中」(T5) 续轮询 tasks.get，而非硬失败。
 		return &shared.RPCError{
 			Code:    -32002,
 			Message: "OPENCLAW_GATEWAY_SOCKET_CLOSED: OpenClaw gateway connection closed during task execution",
 			Data: map[string]any{
 				"code":          "OPENCLAW_GATEWAY_SOCKET_CLOSED",
+				"retryable":     true,
+				"poll":          true,
 				"originalCode":  strings.TrimSpace(shared.StringArg(errorPayload, "code", "")),
 				"originalError": strings.TrimSpace(shared.StringArg(errorPayload, "message", "")),
 			},