Enable instant interrupts by default (#51812)

## What changed

Enable `instant_interrupt` by default so new user input can preempt responses and yield foreground code-mode observations. Keep the feature marked as under development and update feature-policy tests to allow this exception.

Explicitly disable instant interrupts in core, TUI, and Python SDK tests that depend on responses completing before steering or processing subsequent input.

GitOrigin-RevId: a8cab2d6c445bfdbd10b0b2abcf7893d32ee8595
(cherry picked from commit 406b0c4446)

codex-rs/core/tests/suite/guardian_retained_context.rs

@@ -294,6 +294,11 @@ async fn streamed_question_precedes_reply_across_resume(
         .with_model("gpt-5.4")
         .with_history_mode(ThreadHistoryMode::Paginated)
         .with_config(|config| {
+            // This fixture retains the question by allowing its response to finish.
+            config
+                .features
+                .disable(Feature::InstantInterrupt)
+                .expect("disable InstantInterrupt feature");
             config.experimental_thread_store = ThreadStoreConfig::Local;
         })
         .build_with_streaming_server(&server)

codex-rs/core/tests/suite/pending_input.rs

@@ -1964,7 +1964,16 @@ async fn user_input_does_not_preempt_after_reasoning_item() {
     let (server, _completions) =
         start_streaming_sse_server(vec![first_chunks, response_completed_chunks("resp-2")]).await;
 
-    let codex = build_codex(&server).await;
+    let codex = test_codex()
+        .with_config(|config| {
+            config.update_plan_enabled = true;
+            config.features.disable(Feature::InstantInterrupt).unwrap();
+        })
+        .with_model("gpt-5.4")
+        .build_with_streaming_server(&server)
+        .await
+        .expect("build streaming Codex test session")
+        .codex;
 
     submit_user_input(&codex, "first prompt").await;
 
@@ -2390,6 +2399,7 @@ async fn steered_user_input_waits_for_model_continuation_after_mid_turn_compact(
     let codex = test_codex()
         .with_model("gpt-5.4")
         .with_config(|config| {
+            config.features.disable(Feature::InstantInterrupt).unwrap();
             config.model_provider.name = "OpenAI (test)".to_string();
             config.model_provider.supports_websockets = false;
             config.model_auto_compact_token_limit = Some(200);
@@ -2475,6 +2485,7 @@ async fn steered_user_input_follows_compact_when_only_the_steer_needs_follow_up(
     let codex = test_codex()
         .with_model("gpt-5.4")
         .with_config(|config| {
+            config.features.disable(Feature::InstantInterrupt).unwrap();
             config.model_provider.name = "OpenAI (test)".to_string();
             config.model_provider.supports_websockets = false;
             config.model_auto_compact_token_limit = Some(200);
@@ -2592,6 +2603,7 @@ async fn steered_user_input_waits_when_tool_output_triggers_compact_before_next_
     let test = test_codex()
         .with_model("gpt-5.4")
         .with_config(|config| {
+            config.features.disable(Feature::InstantInterrupt).unwrap();
             config.model_provider.name = "OpenAI (test)".to_string();
             config.model_provider.supports_websockets = false;
             config.model_auto_compact_token_limit = Some(200);

codex-rs/core/tests/suite/scenarios_guardian_heartbeat.rs

@@ -58,6 +58,8 @@ async fn heartbeat_repetitions_keep_later_human_authorization() -> anyhow::Resul
                 .features
                 .disable(Feature::EnableRequestCompression)
                 .unwrap();
+            // Keep the acknowledgement complete before the next heartbeat is processed.
+            config.features.disable(Feature::InstantInterrupt).unwrap();
             config.approvals_reviewer = ApprovalsReviewer::AutoReview;
             config.permissions.approval_policy = Constrained::allow_any(AskForApproval::OnRequest);
             config

codex-rs/features/src/lib.rs

@@ -1159,7 +1159,7 @@ pub const FEATURES: &[FeatureSpec] = &[
         id: Feature::InstantInterrupt,
         key: "instant_interrupt",
         stage: Stage::UnderDevelopment,
-        default_enabled: false,
+        default_enabled: true,
     },
     FeatureSpec {
         id: Feature::CodeModeOnly,

codex-rs/features/src/tests.rs

@@ -32,7 +32,8 @@ fn sleep_tool_config_rejects_unknown_mode() {
 #[test]
 fn under_development_features_are_disabled_by_default() {
     for spec in crate::FEATURES {
-        if matches!(spec.stage, Stage::UnderDevelopment) {
+        // Instant interrupts are enabled by default while retaining their development stage.
+        if matches!(spec.stage, Stage::UnderDevelopment) && spec.id != Feature::InstantInterrupt {
             assert_eq!(
                 spec.default_enabled, false,
                 "feature `{}` is under development and must be disabled by default",
@@ -121,7 +122,8 @@ fn codex_apps_mcp_protocol_can_be_enabled_independently_of_generic_mcp() {
 #[test]
 fn default_enabled_features_are_stable() {
     for spec in crate::FEATURES {
-        if spec.default_enabled {
+        // Instant interrupts are enabled by default while retaining their development stage.
+        if spec.default_enabled && spec.id != Feature::InstantInterrupt {
             assert!(
                 matches!(spec.stage, Stage::Stable | Stage::Removed),
                 "feature `{}` is enabled by default but is not stable/removed ({:?})",

codex-rs/tui/src/app/tests/safety_buffering.rs

@@ -540,6 +540,12 @@ async fn run_safety_retry(
     let model_catalog_path = codex_home.path().join("models.json");
     std::fs::write(&model_catalog_path, serde_json::to_vec(&model_catalog)?)?;
     let model_catalog_path = toml::Value::String(model_catalog_path.display().to_string());
+    // The committed-steer case counts usage from the response that finishes before steering.
+    let instant_interrupt_config = if committed_steer.is_some() {
+        "instant_interrupt = false"
+    } else {
+        ""
+    };
     std::fs::write(
         codex_home.path().join("config.toml"),
         format!(
@@ -557,6 +563,7 @@ stream_max_retries = 0
 
 [features]
 goals = true
+{instant_interrupt_config}
 "#,
             server.uri()
         ),

sdk/python/tests/test_app_server_turn_controls.py

@@ -20,7 +20,7 @@ def test_turn_steer_adds_follow_up_input(tmp_path) -> None:
         )
 
         with Codex(config=harness.app_server_config()) as codex:
-            thread = codex.thread_start()
+            thread = codex.thread_start(config={"features.instant_interrupt": False})
             turn = thread.turn("Start a steerable turn.")
             harness.responses.wait_for_requests(1)
             steer = turn.steer("Use this steering input.")