diff --git a/codex-rs/codex-api/src/endpoint/realtime_websocket/methods.rs b/codex-rs/codex-api/src/endpoint/realtime_websocket/methods.rs
index 0a8bac005..3c1fd7bd9 100644
--- a/codex-rs/codex-api/src/endpoint/realtime_websocket/methods.rs
+++ b/codex-rs/codex-api/src/endpoint/realtime_websocket/methods.rs
@@ -217,6 +217,7 @@ pub struct RealtimeWebsocketEvents {
 #[derive(Default)]
 struct ActiveTranscriptState {
     entries: Vec<RealtimeTranscriptEntry>,
+    last_handoff_entry_count: usize,
     new_input_entry: bool,
     new_output_entry: bool,
 }
@@ -459,7 +460,10 @@ impl RealtimeWebsocketEvents {
             }
             RealtimeEvent::HandoffRequested(handoff) => {
                 append_handoff_input(&mut active_transcript.entries, &handoff.input_transcript);
-                handoff.active_transcript = active_transcript.entries.clone();
+                handoff.active_transcript = active_transcript.entries
+                    [active_transcript.last_handoff_entry_count..]
+                    .to_vec();
+                active_transcript.last_handoff_entry_count = active_transcript.entries.len();
                 active_transcript.new_input_entry = true;
                 active_transcript.new_output_entry = true;
             }
diff --git a/codex-rs/core/tests/suite/realtime_conversation.rs b/codex-rs/core/tests/suite/realtime_conversation.rs
index dcf5e1f95..60167a155 100644
--- a/codex-rs/core/tests/suite/realtime_conversation.rs
+++ b/codex-rs/core/tests/suite/realtime_conversation.rs
@@ -2661,7 +2661,7 @@ async fn inbound_handoff_request_uses_active_transcript() -> Result<()> {
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn inbound_handoff_request_keeps_transcript_history_after_each_handoff() -> Result<()> {
+async fn inbound_handoff_request_sends_transcript_delta_after_each_handoff() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
     let api_server = start_mock_server().await;
@@ -2773,9 +2773,9 @@ async fn inbound_handoff_request_keeps_transcript_history_after_each_handoff() -
 
     let second_user_texts = requests[1].message_input_texts("user");
     assert!(second_user_texts.iter().any(|text| text
-        == "<realtime_delegation>\n  <input>second question</input>\n  <transcript_delta>user: first question\nuser: second question</transcript_delta>\n</realtime_delegation>"));
+        == "<realtime_delegation>\n  <input>second question</input>\n  <transcript_delta>user: second question</transcript_delta>\n</realtime_delegation>"));
     assert!(!second_user_texts.iter().any(|text| text
-        == "<realtime_delegation>\n  <input>second question</input>\n  <transcript_delta>user: first questionsecond question</transcript_delta>\n</realtime_delegation>"));
+        == "<realtime_delegation>\n  <input>second question</input>\n  <transcript_delta>user: first question\nuser: second question</transcript_delta>\n</realtime_delegation>"));
 
     realtime_server.shutdown().await;
     Ok(())