fix: narrow daemon start not-loaded fallback (#54087) (thanks @hclsys)

Address Codex P1 + Greptile P2:
- Move config validation before the restart attempt so invalid config
  is caught in the stop→start path (not just the already-loaded path)
- Derive service.loaded from actual isLoaded() after restart instead
  of hardcoded true

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Signed-off-by: HCL <chenglunhu@gmail.com>

CHANGELOG.md

@@ -50,6 +50,7 @@ Docs: https://docs.openclaw.ai
 - Feishu/MSTeams message tool: keep provider-native `card` payloads optional in merged tool schemas so media-only sends stop failing validation before channel runtime dispatch. (#53715) Thanks @lndyzwdxhs.
 - Feishu/startup: keep `requireMention` enforcement strict when bot identity startup probes fail, raise the startup bot-info timeout to 30s, and add cancellable background identity recovery so mention-gated groups recover without noisy fallback. (#43788) Thanks @lefarcen.
 - Plugins: enforce terminal hook decision semantics for tool/message guards (#54241) Thanks @joshavant.
+- Gateway/daemon: keep `openclaw gateway start` failing on real restart errors for stopped-but-installed services, while still falling back to install hints after uninstall. (#54087) Thanks @hclsys.
 
 ## 2026.3.23
 

src/cli/daemon-cli/lifecycle-core.test.ts

@@ -210,4 +210,33 @@ describe("runServiceRestart token drift", () => {
     expect(payload.result).toBe("scheduled");
     expect(payload.message).toBe("restart scheduled, gateway will restart momentarily");
   });
+
+  it("fails start when restarting a stopped installed service errors", async () => {
+    service.isLoaded.mockResolvedValue(false);
+    service.restart.mockRejectedValue(new Error("launchctl kickstart failed: permission denied"));
+
+    await expect(runServiceStart(createServiceRunArgs())).rejects.toThrow("__exit__:1");
+
+    const payload = readJsonLog<{ ok?: boolean; error?: string }>();
+    expect(payload.ok).toBe(false);
+    expect(payload.error).toContain("launchctl kickstart failed: permission denied");
+  });
+
+  it("falls back to not-loaded hints when restart fails after uninstall", async () => {
+    service.isLoaded.mockResolvedValue(false);
+    service.restart.mockRejectedValue(new Error("launchctl bootstrap failed"));
+    service.readCommand.mockResolvedValue(null);
+
+    await runServiceStart({
+      serviceNoun: "Gateway",
+      service,
+      renderStartHints: () => ["openclaw gateway install"],
+      opts: { json: true },
+    });
+
+    const payload = readJsonLog<{ ok?: boolean; result?: string; hints?: string[] }>();
+    expect(payload.ok).toBe(true);
+    expect(payload.result).toBe("not-loaded");
+    expect(payload.hints).toEqual(["openclaw gateway install"]);
+  });
 });

src/cli/daemon-cli/lifecycle-core.ts

@@ -103,6 +103,14 @@ async function handleServiceNotLoaded(params: {
   }
 }
 
+async function isServiceInstallMissing(service: GatewayService): Promise<boolean> {
+  try {
+    return (await service.readCommand(process.env)) === null;
+  } catch {
+    return false;
+  }
+}
+
 async function resolveServiceLoadedOrFail(params: {
   serviceNoun: string;
   service: GatewayService;
@@ -208,18 +216,8 @@ export async function runServiceStart(params: {
   if (loaded === null) {
     return;
   }
-  if (!loaded) {
-    await handleServiceNotLoaded({
-      serviceNoun: params.serviceNoun,
-      service: params.service,
-      loaded,
-      renderStartHints: params.renderStartHints,
-      json,
-      emit,
-    });
-    return;
-  }
-  // Pre-flight config validation (#35862)
+  // Pre-flight config validation (#35862) — run for both loaded and not-loaded
+  // to prevent launching from invalid config in any start path.
   {
     const configError = await getConfigValidationError();
     if (configError) {
@@ -229,6 +227,44 @@ export async function runServiceStart(params: {
       return;
     }
   }
+  if (!loaded) {
+    // Service was stopped (e.g. `gateway stop` booted out the LaunchAgent).
+    // Attempt a restart, which handles re-bootstrapping the service. Without
+    // this, `start` after `stop` just prints hints and does nothing (#53878).
+    try {
+      const restartResult = await params.service.restart({ env: process.env, stdout });
+      const restartStatus = describeGatewayServiceRestart(params.serviceNoun, restartResult);
+      const postLoaded = await params.service.isLoaded({ env: process.env }).catch(() => true);
+      emit({
+        ok: true,
+        result: restartStatus.daemonActionResult,
+        message: restartStatus.message,
+        service: buildDaemonServiceSnapshot(params.service, postLoaded),
+      });
+      if (!json) {
+        defaultRuntime.log(restartStatus.message);
+      }
+      return;
+    } catch (err) {
+      // Only fall back to "not loaded" hints when the service install itself is
+      // gone. If restart failed while install artifacts still exist, surface
+      // the real error instead of reporting a misleading success.
+      if (await isServiceInstallMissing(params.service)) {
+        await handleServiceNotLoaded({
+          serviceNoun: params.serviceNoun,
+          service: params.service,
+          loaded,
+          renderStartHints: params.renderStartHints,
+          json,
+          emit,
+        });
+        return;
+      }
+      const hints = params.renderStartHints();
+      fail(`${params.serviceNoun} start failed: ${String(err)}`, hints);
+      return;
+    }
+  }
 
   try {
     const restartResult = await params.service.restart({ env: process.env, stdout });