test: stabilize docker live model suites

src/agents/models.profiles.live.test.ts

@@ -496,7 +496,10 @@ describeLive("live models (profile keys)", () => {
               throw new Error(msg || "model returned error with no message");
             }
 
-            if (ok.text.length === 0 && model.provider === "google") {
+            if (
+              ok.text.length === 0 &&
+              (model.provider === "google" || model.provider === "google-gemini-cli")
+            ) {
               skipped.push({
                 model: id,
                 reason: "no text returned (likely unavailable model id)",

src/gateway/gateway-models.profiles.live.test.ts

@@ -40,6 +40,11 @@ const THINKING_LEVEL = "high";
 const THINKING_TAG_RE = /<\s*\/?\s*(?:think(?:ing)?|thought|antthinking)\s*>/i;
 const FINAL_TAG_RE = /<\s*\/?\s*final\s*>/i;
 const ANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL = "ANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL";
+const GATEWAY_LIVE_DEFAULT_TIMEOUT_MS = 20 * 60 * 1000;
+const GATEWAY_LIVE_UNBOUNDED_TIMEOUT_MS = 60 * 60 * 1000;
+const GATEWAY_LIVE_MAX_TIMEOUT_MS = 2 * 60 * 60 * 1000;
+const GATEWAY_LIVE_MAX_MODELS = resolveGatewayLiveMaxModels();
+const GATEWAY_LIVE_SUITE_TIMEOUT_MS = resolveGatewayLiveSuiteTimeoutMs(GATEWAY_LIVE_MAX_MODELS);
 
 const describeLive = LIVE || GATEWAY_LIVE ? describe : describe.skip;
 
@@ -64,6 +69,27 @@ function toInt(value: string | undefined, fallback: number): number {
   return Number.isFinite(parsed) ? parsed : fallback;
 }
 
+function resolveGatewayLiveMaxModels(): number {
+  const gatewayMax = toInt(process.env.OPENCLAW_LIVE_GATEWAY_MAX_MODELS, -1);
+  if (gatewayMax >= 0) {
+    return gatewayMax;
+  }
+  // Reuse shared live-model cap when gateway-specific cap is not provided.
+  return Math.max(0, toInt(process.env.OPENCLAW_LIVE_MAX_MODELS, 0));
+}
+
+function resolveGatewayLiveSuiteTimeoutMs(maxModels: number): number {
+  if (maxModels <= 0) {
+    return GATEWAY_LIVE_UNBOUNDED_TIMEOUT_MS;
+  }
+  // Gateway live runs multiple probes per model; scale timeout by model cap.
+  const estimated = 5 * 60 * 1000 + maxModels * 90 * 1000;
+  return Math.max(
+    GATEWAY_LIVE_DEFAULT_TIMEOUT_MS,
+    Math.min(GATEWAY_LIVE_MAX_TIMEOUT_MS, estimated),
+  );
+}
+
 function capByProviderSpread<T>(
   items: T[],
   maxItems: number,
@@ -1144,7 +1170,7 @@ describeLive("gateway live (dev agent, profile keys)", () => {
       const useModern = !rawModels || rawModels === "modern" || rawModels === "all";
       const useExplicit = Boolean(rawModels) && !useModern;
       const filter = useExplicit ? parseFilter(rawModels) : null;
-      const maxModels = toInt(process.env.OPENCLAW_LIVE_GATEWAY_MAX_MODELS, 0);
+      const maxModels = GATEWAY_LIVE_MAX_MODELS;
       const wanted = filter
         ? all.filter((m) => filter.has(`${m.provider}/${m.id}`))
         : all.filter((m) => isModernModelRef({ provider: m.provider, id: m.id }));
@@ -1224,7 +1250,7 @@ describeLive("gateway live (dev agent, profile keys)", () => {
         logProgress("[minimax-anthropic] missing minimax provider config; skipping");
       }
     },
-    20 * 60 * 1000,
+    GATEWAY_LIVE_SUITE_TIMEOUT_MS,
   );
 
   it("z.ai fallback handles anthropic tool history", async () => {