Stop mirroring Codex user input into realtime (#27116)

## Why

The realtime frontend model and the backing Codex thread should present
one coherent assistant. Raw typed messages, steers, and worker reports
belong to the orchestrator; the frontend model should receive the
orchestrator's user-facing result rather than a second copy of those
inputs.

Today normal `turn/start` input is automatically inserted into the
realtime conversation, while `turn/steer` is not. Besides creating
inconsistent context, this can make the frontend model react
independently before Codex has produced the response it should speak.

## What changed

- Remove automatic accepted-user-input mirroring into realtime
- Remove the mirror-only echo-suppression flag and dead V2 prefix helper
- Preserve explicit app-to-realtime text injection and FEM-to-Codex
delegation
- Replace the positive mirror tests and obsolete snapshots with a
negative routing regression test

## Test plan

- `cargo test -p codex-core
conversation_user_text_turn_is_not_sent_to_realtime`
- `cargo test -p codex-core
conversation_startup_context_is_truncated_and_sent_once_per_start`
- `cargo test -p codex-core inbound_handoff_request_starts_turn`

codex-rs/core/src/realtime_conversation.rs

@@ -739,10 +739,6 @@ fn prefix_realtime_text(text: String, prefix: &str, session_kind: RealtimeSessio
     format!("{prefix}{text}")
 }
 
-pub(crate) fn prefix_realtime_v2_text(text: String, prefix: &str) -> String {
-    prefix_realtime_text(text, prefix, RealtimeSessionKind::V2)
-}
-
 fn validate_realtime_voice(version: RealtimeWsVersion, voice: RealtimeVoice) -> CodexResult<()> {
     let voices = RealtimeVoicesList::builtin();
     let allowed = match version {

codex-rs/core/src/session/handlers.rs

@@ -15,10 +15,6 @@ use crate::session::session::Session;
 use crate::session::session::SessionSettingsUpdate;
 
 use crate::config::Config;
-use crate::realtime_context::REALTIME_TURN_TOKEN_BUDGET;
-use crate::realtime_context::truncate_realtime_text_to_token_budget;
-use crate::realtime_conversation::REALTIME_USER_TEXT_PREFIX;
-use crate::realtime_conversation::prefix_realtime_v2_text;
 use crate::review_prompts::resolve_review_request;
 use crate::session::spawn_review_thread;
 use crate::tasks::CompactTask;
@@ -54,9 +50,7 @@ use codex_protocol::request_user_input::RequestUserInputResponse;
 
 use crate::context_manager::is_user_turn_boundary;
 use codex_protocol::dynamic_tools::DynamicToolResponse;
-use codex_protocol::items::UserMessageItem;
 use codex_protocol::mcp::RequestId as ProtocolRequestId;
-use codex_protocol::user_input::UserInput;
 use codex_rmcp_client::ElicitationAction;
 use codex_rmcp_client::ElicitationResponse;
 use serde_json::Value;
@@ -91,14 +85,7 @@ pub async fn user_input_or_turn(
     op: Op,
     client_user_message_id: Option<String>,
 ) {
-    user_input_or_turn_inner(
-        sess,
-        sub_id,
-        op,
-        /*mirror_user_text_to_realtime*/ Some(()),
-        client_user_message_id,
-    )
-    .await;
+    user_input_or_turn_inner(sess, sub_id, op, client_user_message_id).await;
 }
 
 pub async fn update_thread_settings(
@@ -196,7 +183,6 @@ pub(super) async fn user_input_or_turn_inner(
     sess: &Arc<Session>,
     sub_id: String,
     op: Op,
-    mirror_user_text_to_realtime: Option<()>,
     client_user_message_id: Option<String>,
 ) {
     let Op::UserInput {
@@ -230,7 +216,7 @@ pub(super) async fn user_input_or_turn_inner(
     }
     sess.maybe_emit_unknown_model_warning_for_turn(current_context.as_ref())
         .await;
-    let accepted_items = match sess
+    match sess
         .steer_input(
             items.clone(),
             additional_context.clone(),
@@ -242,7 +228,6 @@ pub(super) async fn user_input_or_turn_inner(
     {
         Ok(_) => {
             current_context.session_telemetry.user_prompt(&items);
-            Some(items)
         }
         Err(SteerInputError::NoActiveTurn(items)) => {
             if let Some(responsesapi_client_metadata) = responsesapi_client_metadata {
@@ -256,7 +241,6 @@ pub(super) async fn user_input_or_turn_inner(
                 Some(sess.mcp_elicitation_reviewer()),
             )
             .await;
-            let accepted_items = items.clone();
             let additional_context_input = {
                 let mut state = sess.state.lock().await;
                 state.additional_context.merge(additional_context)
@@ -278,7 +262,6 @@ pub(super) async fn user_input_or_turn_inner(
                 crate::tasks::RegularTask::new(),
             )
             .await;
-            Some(accepted_items)
         }
         Err(err) => {
             sess.send_event_raw(Event {
@@ -286,33 +269,7 @@ pub(super) async fn user_input_or_turn_inner(
                 msg: EventMsg::Error(err.to_error_event()),
             })
             .await;
-            None
         }
-    };
-    if let (Some(items), Some(())) = (accepted_items, mirror_user_text_to_realtime) {
-        self::mirror_user_text_to_realtime(sess, &items).await;
-    }
-}
-
-async fn mirror_user_text_to_realtime(sess: &Arc<Session>, items: &[UserInput]) {
-    let text = UserMessageItem::new(items).message();
-    if text.is_empty() {
-        return;
-    }
-    let text = if sess.conversation.is_running_v2().await {
-        prefix_realtime_v2_text(text, REALTIME_USER_TEXT_PREFIX)
-    } else {
-        text
-    };
-    let text = truncate_realtime_text_to_token_budget(&text, REALTIME_TURN_TOKEN_BUDGET);
-    if text.is_empty() {
-        return;
-    }
-    if sess.conversation.running_state().await.is_none() {
-        return;
-    }
-    if let Err(err) = sess.conversation.text_in(text).await {
-        debug!("failed to mirror user text to realtime conversation: {err}");
     }
 }
 

codex-rs/core/src/session/mod.rs

@@ -1132,7 +1132,6 @@ impl Session {
                 additional_context: Default::default(),
                 thread_settings: Default::default(),
             },
-            /*mirror_user_text_to_realtime*/ None,
             /*client_user_message_id*/ None,
         )
         .await;