ci: stabilize release live QA gates

qa/scenarios/config/config-restart-capability-flip.md

@@ -96,6 +96,9 @@ steps:
             - set: wakeMarker
               value:
                 expr: "`QA-CAPABILITY-${randomUUID().slice(0, 8)}`"
+            - set: wakeStartIndex
+              value:
+                expr: "state.getSnapshot().messages.length"
             - call: patchConfig
               args:
                 - env:
@@ -111,6 +114,9 @@ steps:
                             ref: originalImageGenerationModelPrimary
                   sessionKey:
                     ref: sessionKey
+                  deliveryContext:
+                    channel: qa-channel
+                    to: dm:qa-operator
                   note:
                     ref: wakeMarker
             - call: waitForGatewayHealthy
@@ -121,6 +127,16 @@ steps:
               args:
                 - ref: env
                 - 60000
+            - call: waitForOutboundMessage
+              saveAs: wakeReply
+              args:
+                - ref: state
+                - lambda:
+                    params: [candidate]
+                    expr: "candidate.direction === 'outbound' && candidate.conversation.id === 'qa-operator' && String(candidate.text ?? '').includes(wakeMarker)"
+                - expr: liveTurnTimeoutMs(env, 60000)
+                - sinceIndex:
+                    ref: wakeStartIndex
             - call: waitForCondition
               saveAs: afterTools
               args:

src/agents/subagent-announce.live.test.ts

@@ -494,7 +494,7 @@ describeLive("subagent announce live", () => {
         entry: spawnedRun,
         message: steerToken,
       });
-      expect(steerResult.status).toBe("accepted");
+      expect(["accepted", "done"]).toContain(steerResult.status);
 
       const steeredRun = await waitFor("steered child completion", () => {
         if (initialError) {

src/agents/tool-replay-repair.live.test.ts

@@ -72,11 +72,19 @@ function createNoopTools() {
   ];
 }
 
-function replayValidationTools(model: Model<Api>) {
-  // Responses-family providers may force a new tool call whenever tools are
-  // present. These live probes validate repaired historical transcript shape,
-  // not fresh tool invocation.
-  return isOpenAIResponsesFamily(model.api) ? undefined : createNoopTools();
+function replayValidationTools() {
+  return createNoopTools();
+}
+
+function disableResponsesReplayToolChoice(payload: unknown, model: Model<Api>): unknown {
+  if (!isOpenAIResponsesFamily(model.api) || !payload || typeof payload !== "object") {
+    return undefined;
+  }
+  const next = payload as { tool_choice?: unknown };
+  // Replay probes include historical tool calls to validate transcript repair,
+  // but they should not force a fresh noop tool call during the live request.
+  next.tool_choice = "none";
+  return next;
 }
 
 function buildReplayMessages(model: Model<Api>): AgentMessage[] {
@@ -270,12 +278,13 @@ describeLive("tool replay repair live", () => {
           {
             systemPrompt: "You are a concise assistant. Follow the user's instruction exactly.",
             messages: sanitized as never,
-            tools: replayValidationTools(model),
+            tools: replayValidationTools(),
           },
           {
             apiKey: requireApiKey(apiKeyInfo, model.provider),
             reasoning: "low",
             maxTokens: 96,
+            onPayload: disableResponsesReplayToolChoice,
           },
           120_000,
         );
@@ -345,12 +354,13 @@ describeLive("tool replay repair live", () => {
           {
             systemPrompt: "You are a concise assistant. Follow the user's instruction exactly.",
             messages: transformed as never,
-            tools: replayValidationTools(model),
+            tools: replayValidationTools(),
           },
           {
             apiKey: requireApiKey(apiKeyInfo, model.provider),
             reasoning: "low",
             maxTokens: 96,
+            onPayload: disableResponsesReplayToolChoice,
           },
           120_000,
         );

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

@@ -1082,10 +1082,10 @@ describe("getHighSignalLiveModelPriorityIndex", () => {
   it("prefers curated Google replacements over big-pickle", () => {
     expect(
       getHighSignalLiveModelPriorityIndex({ provider: "google", id: "gemini-3.1-pro-preview" }),
-    ).toBe(3);
+    ).toBe(2);
     expect(
       getHighSignalLiveModelPriorityIndex({ provider: "google", id: "gemini-3-flash-preview" }),
-    ).toBe(4);
+    ).toBe(3);
     expect(getHighSignalLiveModelPriorityIndex({ provider: "opencode", id: "big-pickle" })).toBe(
       null,
     );