Expose assistant content token counts in completion notifications (#59041)

## What changed

Add optional `contentTokenCount` metadata to `item/completed` notifications, carrying the producer-supplied token count for each completed assistant message in default and plan modes. Counts include any sampled footer and are `null` when unavailable.

Keep this telemetry separate from stored thread items and omit it from subsequent model requests. Update the protocol schemas and generated TypeScript and Python types.

## Testing

Add coverage for per-message counts, including zero, in default and plan modes; missing and malformed metadata; backward-compatible notification decoding; and exclusion from subsequent model requests.

GitOrigin-RevId: 35973037a42650fa2d7569da8e88cb81215cf971

codex-rs/protocol/src/models.rs

@@ -1009,6 +1009,13 @@ pub struct InternalChatMessageMetadataPassthrough {
     #[schemars(skip)]
     #[ts(skip)]
     pub create_time: Option<serde_json::Number>,
+    /// Native content-token count for this sampled message, including its sampled footer.
+    /// Read from the producer only; do not replay telemetry into subsequent model requests.
+    #[serde_as(deserialize_as = "serde_with::DefaultOnError")]
+    #[serde(default, skip_serializing)]
+    #[schemars(skip)]
+    #[ts(skip)]
+    pub content_token_count: Option<u64>,
     /// Harness-owned classifications aligned with the item's content entries.
     #[serde_as(deserialize_as = "serde_with::DefaultOnError")]
     #[serde(default, skip_serializing_if = "Option::is_none")]

codex-rs/protocol/src/models/item_metadata_tests.rs

@@ -57,6 +57,41 @@ fn malformed_content_kinds_do_not_prevent_loading_response_item() -> Result<()>
     Ok(())
 }
 
+#[test]
+fn malformed_content_token_counts_do_not_prevent_loading_response_item() -> Result<()> {
+    for value in [json!(-1), json!(true), json!(1.5), json!("4"), json!(null)] {
+        let item: ResponseItem = serde_json::from_value(response_item_json(json!({
+            "content_token_count": value,
+        })))?;
+        assert_eq!(
+            item,
+            response_item(InternalChatMessageMetadataPassthrough::default())
+        );
+    }
+    Ok(())
+}
+
+#[test]
+fn content_token_counts_are_not_replayed_into_requests() -> Result<()> {
+    let item: ResponseItem = serde_json::from_value(response_item_json(json!({
+        "turn_id": "turn-1", "content_token_count": 23,
+    })))?;
+    let ResponseItem::Message {
+        internal_chat_message_metadata_passthrough: Some(ref metadata),
+        ..
+    } = item
+    else {
+        panic!("expected message metadata");
+    };
+    assert_eq!(metadata.content_token_count, Some(23));
+    let encoded = serde_json::to_value(item)?;
+    assert_eq!(
+        encoded["internal_chat_message_metadata_passthrough"],
+        json!({"turn_id": "turn-1"})
+    );
+    Ok(())
+}
+
 fn response_item(metadata: InternalChatMessageMetadataPassthrough) -> ResponseItem {
     ResponseItem::Message {
         status: None,

codex-rs/protocol/src/protocol.rs

@@ -2194,6 +2194,11 @@ pub struct ItemCompletedEvent {
     pub thread_id: ThreadId,
     pub turn_id: String,
     pub item: TurnItem,
+    /// Native tokens in the completed assistant message, including any sampled footer.
+    /// Completion telemetry only; absent for items without a producer-supplied count.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    #[ts(optional)]
+    pub content_token_count: Option<u64>,
     #[serde(default, skip_serializing_if = "Option::is_none")]
     #[ts(optional)]
     pub started_at_ms: Option<i64>,
@@ -5763,6 +5768,7 @@ mod tests {
     #[test]
     fn item_completed_event_from_image_generation_emits_end_event() {
         let event = ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             item: TurnItem::ImageGeneration(ImageGenerationItem {
@@ -5801,6 +5807,7 @@ mod tests {
     #[test]
     fn item_completed_event_from_file_change_emits_patch_end_event() {
         let event = ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             started_at_ms: Some(0),
@@ -5840,6 +5847,7 @@ mod tests {
     #[test]
     fn item_completed_event_from_mcp_tool_call_emits_end_event() {
         let event = ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             started_at_ms: Some(0),
@@ -5921,6 +5929,7 @@ mod tests {
             }),
         };
         let completed = ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             started_at_ms: Some(10),
@@ -5999,6 +6008,7 @@ mod tests {
             }),
         };
         let completed = ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             started_at_ms: Some(10),
@@ -6044,6 +6054,7 @@ mod tests {
     #[test]
     fn review_mode_item_completion_emits_legacy_events_with_ids() {
         let entered = ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             started_at_ms: Some(0),
@@ -6057,6 +6068,7 @@ mod tests {
             }),
         };
         let exited = ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             started_at_ms: Some(0),
@@ -6117,6 +6129,7 @@ mod tests {
     #[test]
     fn item_completed_event_defaults_missing_completed_at_ms() {
         let mut value = serde_json::to_value(ItemCompletedEvent {
+            content_token_count: None,
             thread_id: ThreadId::new(),
             turn_id: "turn-1".into(),
             item: TurnItem::UserMessage(UserMessageItem::new(&[])),