test(live): tolerate codex ask-back preflight refusal

vincentkoc · vincentkoc · commit 3a12a7a7e62b · 2026-05-06T14:08:32.000-07:00
diff --git a/src/gateway/gateway-codex-harness.live.test.ts b/src/gateway/gateway-codex-harness.live.test.ts
@@ -376,11 +376,13 @@ function findGuardianReviewStatus(events: CapturedAgentEvent[]): "approved" | "d
 function assertGuardianReviewCompleted(params: {
   events: CapturedAgentEvent[];
   label: string;
+  requireEvents?: boolean;
 }): CapturedAgentEvent | undefined {
   const completedEvents = params.events.filter(
     (event) => event.data?.phase === "completed" && event.data?.status,
   );
-  if (completedEvents.length === 0 && !CODEX_HARNESS_REQUIRE_GUARDIAN_EVENTS) {
+  const requireEvents = params.requireEvents ?? CODEX_HARNESS_REQUIRE_GUARDIAN_EVENTS;
+  if (completedEvents.length === 0 && !requireEvents) {
     return undefined;
   }
   expect(
@@ -441,13 +443,20 @@ async function verifyCodexGuardianProbe(params: {
   const review = assertGuardianReviewCompleted({
     events: deniedResult.events,
     label: "ask-back probe",
+    requireEvents: false,
   });
   // The approve/deny call is Codex policy-owned and may change independently.
-  // OpenClaw's contract here is that Guardian mode reaches Codex app-server and
-  // projects the structured review lifecycle back onto the agent event bus.
+  // OpenClaw's strict projection contract is covered by the allow probe above.
+  // Riskier prompts may be refused or ask back before Codex creates a review
+  // event, depending on current policy/model behavior.
   if (review?.data?.status === "denied") {
     expect(deniedResult.text).toContain(askBackToken);
     expect(deniedResult.text.toLowerCase()).toMatch(/approv|permission|guardian|reject|denied/);
+  } else if (!review) {
+    expect(deniedResult.text).toContain(askBackToken);
+    expect(deniedResult.text.toLowerCase()).toMatch(
+      /approv|permission|guardian|reject|denied|block|cannot|can't/,
+    );
   }
   expect(deniedResult.text.trim().length).toBeGreaterThan(0);
 }
