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
This commit is contained in:
aojiao-oai
2026-11-03 02:26:29 +00:00
committed by copyberry
parent e839b828fb
commit 16ddd538bd
48 changed files with 443 additions and 39 deletions
@@ -27,6 +27,7 @@ fn delegated_output_emits_only_first_preview_in_live_patch() {
);
for (index, text) in ["first task", "later task"].into_iter().enumerate() {
let item = RolloutItem::EventMsg(EventMsg::ItemCompleted(ItemCompletedEvent {
content_token_count: None,
thread_id,
turn_id: "turn".to_string(),
item: TurnItem::FunctionCallOutput(FunctionCallOutputItem {