From b89952d582beaae3af3ddd7b679286cbb094f8b9 Mon Sep 17 00:00:00 2001
From: jif <jif@openai.com>
Date: Tue, 27 Oct 2026 14:20:31 +0000
Subject: [PATCH] Fix guardian handoff mocks for native history tests (#57468)

## What changed

Add mock guardian approvals for setup handoffs in `handoff_scenario` when native multi-agent history is enabled. Match only the trailing `function_call_output` so historical handoffs cannot consume the final command's separate denial response.

Require five approval requests with native history enabled and none with it disabled.

GitOrigin-RevId: 7b14c97aa85adc63c4d3bf95014c44831eae1c85
---
 .../core/tests/suite/guardian_root_handoff.rs | 33 +++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/codex-rs/core/tests/suite/guardian_root_handoff.rs b/codex-rs/core/tests/suite/guardian_root_handoff.rs
index 891cec946b..c39ed110b9 100644
--- a/codex-rs/core/tests/suite/guardian_root_handoff.rs
+++ b/codex-rs/core/tests/suite/guardian_root_handoff.rs
@@ -89,6 +89,39 @@ pub(crate) async fn handoff_scenario(native_history: bool) -> Result<Vec<Respons
         .await?;
     let root = test.session_configured.thread_id;
     let mut created = test.thread_manager.subscribe_thread_created();
+    // Allow the setup handoffs. Match the trailing native evidence output so
+    // historical calls cannot swallow the final command's separate denial.
+    wiremock::Mock::given(wiremock::matchers::method("POST"))
+        .and(wiremock::matchers::path("/v1/responses"))
+        .and(|request: &wiremock::Request| {
+            request_body(request).is_some_and(|body| {
+                body["client_metadata"]["x-openai-subagent"] == "guardian"
+                    && body["input"]
+                        .as_array()
+                        .and_then(|items| items.last())
+                        .is_some_and(|item| {
+                            item["type"] == "function_call_output"
+                                && matches!(
+                                    item["call_id"].as_str(),
+                                    Some(
+                                        "spawn-alpha" | "spawn-leaf" | "spawn-beta" | "send-alpha"
+                                            | "followup-alpha"
+                                    )
+                                )
+                        })
+            })
+        })
+        .respond_with(sse_response(sse(vec![
+            ev_assistant_message(
+                "handoff-allow",
+                r#"{"risk_level":"low","user_authorization":"high","outcome":"allow","rationale":"Mock approval for the scenario handoff."}"#,
+            ),
+            ev_completed("handoff-allow"),
+        ])))
+        .with_priority(/*p*/ 1)
+        .expect(if native_history { 5 } else { 0 })
+        .mount(&server)
+        .await;
     // Ordinary root/worker continuations idle; the reviewed action has its own mock.
     wiremock::Mock::given(wiremock::matchers::method("POST"))
         .and(wiremock::matchers::path("/v1/responses"))