feat(core): add metadata field to ResponseItem (#28355)

## Description

This PR adds an optional `metadata` field to `ResponseItem` for
Responses API calls. Only mechanical plumbing, no actual values
populated and sent yet. Turns out just adding a new field to
`ResponseItem` has quite a large blast radius already.

This change is backwards compatible because `metadata` is optional and
omitted when absent, so existing response items and rollout history
without it still deserialize and requests that do not set it keep the
same wire shape. For provider compatibility, we strip out `metadata`
before non-OpenAI Responses requests so Azure and AWS Bedrock never see
this field.

My followup PR here will actually make use of it to start storing and
passing along `turn_id`: https://github.com/openai/codex/pull/28360

## What changed

- Added `ResponseItemMetadata` with optional `turn_id`, plus optional
`metadata` on Responses API item variants and inter-agent communication.
- Preserved item metadata through response-item rewrites such as
truncation, missing tool-output synthesis, compaction history
rebuilding, visible-history conversion, rollout/resume, and generated
app-server schemas/types.
- Strip item metadata from non-OpenAI Responses requests while
preserving it for OpenAI-shaped requests.
- Updated the mechanical fixture/test construction churn required by the
new optional field.
This commit is contained in:
Owen Lin
2026-06-15 15:05:28 -07:00
committed by GitHub
Unverified
parent bef99f861b
commit 040dafa32d
85 changed files with 1637 additions and 92 deletions
+1
View File
@@ -181,6 +181,7 @@ use uuid::Uuid;
use crate::client::ModelClient;
use crate::codex_thread::ThreadConfigSnapshot;
#[cfg(test)]
use crate::compact::collect_user_messages;
use crate::config::Config;
use crate::config::Constrained;
@@ -296,7 +296,7 @@ impl Session {
// prompt shape.
// TODO(ccunningham): if we drop support for None replacement_history compaction items,
// we can get rid of this second loop entirely and just build `history` directly in the first loop.
let user_messages = collect_user_messages(history.raw_items());
let user_messages = compact::collect_user_messages(history.raw_items());
let rebuilt = compact::build_compacted_history(
Vec::new(),
&user_messages,
@@ -20,6 +20,7 @@ fn user_message(text: &str) -> ResponseItem {
text: text.to_string(),
}],
phase: None,
metadata: None,
}
}
@@ -31,6 +32,7 @@ fn assistant_message(text: &str) -> ResponseItem {
text: text.to_string(),
}],
phase: None,
metadata: None,
}
}
@@ -49,6 +51,7 @@ fn inter_agent_assistant_message(text: &str) -> ResponseItem {
text: serde_json::to_string(&communication).unwrap(),
}],
phase: None,
metadata: None,
}
}
+54 -22
View File
@@ -103,6 +103,7 @@ use codex_protocol::config_types::Settings;
use codex_protocol::models::BaseInstructions;
use codex_protocol::models::ContentItem;
use codex_protocol::models::ResponseItem;
use codex_protocol::models::ResponseItemMetadata;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::CodexErrorInfo;
use codex_protocol::protocol::CompactedItem;
@@ -189,6 +190,7 @@ fn user_message(text: &str) -> ResponseItem {
text: text.to_string(),
}],
phase: None,
metadata: None,
}
}
@@ -200,6 +202,7 @@ fn assistant_message(text: &str) -> ResponseItem {
text: text.to_string(),
}],
phase: None,
metadata: None,
}
}
@@ -259,6 +262,7 @@ fn skill_message(text: &str) -> ResponseItem {
text: text.to_string(),
}],
phase: None,
metadata: None,
}
}
@@ -1598,6 +1602,9 @@ async fn reconstruct_history_uses_replacement_history_verbatim() {
text: "summary".to_string(),
}],
phase: None,
metadata: Some(ResponseItemMetadata {
turn_id: Some("compact-turn".to_string()),
}),
};
let replacement_history = vec![
summary_item.clone(),
@@ -1608,6 +1615,7 @@ async fn reconstruct_history_uses_replacement_history_verbatim() {
text: "stale developer instructions".to_string(),
}],
phase: None,
metadata: None,
},
];
let rollout_items = vec![RolloutItem::Compacted(CompactedItem {
@@ -1669,32 +1677,35 @@ async fn resize_all_images_prepares_failures_before_history_insertion() {
]),
success: Some(true),
},
metadata: None,
};
session
.record_conversation_items(turn_context.as_ref(), std::slice::from_ref(&item))
.await;
let expected = vec![ResponseItem::FunctionCallOutput {
call_id: "call-1".to_string(),
output: FunctionCallOutputPayload {
body: FunctionCallOutputBody::ContentItems(vec![
FunctionCallOutputContentItem::InputText {
text: "before".to_string(),
},
FunctionCallOutputContentItem::InputText {
text: "image content omitted because it could not be processed".to_string(),
},
FunctionCallOutputContentItem::InputImage {
image_url: "https://example.com/image.png".to_string(),
detail: Some(ImageDetail::High),
},
]),
success: Some(true),
},
metadata: None,
}];
assert_eq!(
session.state.lock().await.clone_history().raw_items(),
&[ResponseItem::FunctionCallOutput {
call_id: "call-1".to_string(),
output: FunctionCallOutputPayload {
body: FunctionCallOutputBody::ContentItems(vec![
FunctionCallOutputContentItem::InputText {
text: "before".to_string(),
},
FunctionCallOutputContentItem::InputText {
text: "image content omitted because it could not be processed".to_string(),
},
FunctionCallOutputContentItem::InputImage {
image_url: "https://example.com/image.png".to_string(),
detail: Some(ImageDetail::High),
},
]),
success: Some(true),
},
}]
expected.as_slice()
);
}
@@ -1721,6 +1732,7 @@ async fn resize_all_images_prepares_resumed_history_before_installing_it() {
},
],
phase: None,
metadata: None,
};
session
@@ -1745,6 +1757,7 @@ async fn resize_all_images_prepares_resumed_history_before_installing_it() {
},
],
phase: None,
metadata: None,
}]
);
}
@@ -1860,7 +1873,8 @@ async fn resumed_history_injects_initial_context_on_first_context_update_only()
session
.record_context_updates_and_set_reference_context_item(&turn_context)
.await;
expected.extend(session.build_initial_context(&turn_context).await);
let initial_context = session.build_initial_context(&turn_context).await;
expected.extend(initial_context);
let history_after_seed = session.clone_history().await;
assert_eq!(expected, history_after_seed.raw_items());
@@ -7662,6 +7676,7 @@ async fn build_initial_context_omits_default_image_save_location_with_image_hist
status: "completed".to_string(),
revised_prompt: Some("a tiny blue square".to_string()),
result: "Zm9v".to_string(),
metadata: None,
}],
/*reference_context_item*/ None,
)
@@ -7914,6 +7929,7 @@ async fn handle_output_item_done_records_image_save_history_message() {
status: "completed".to_string(),
revised_prompt: Some("a tiny blue square".to_string()),
result: "Zm9v".to_string(),
metadata: None,
};
let mut ctx = HandleOutputCtx {
@@ -7944,7 +7960,8 @@ async fn handle_output_item_done_records_image_save_history_message() {
image_output_path.display(),
),
);
assert_eq!(history.raw_items(), &[image_message, item]);
let expected = vec![image_message, item];
assert_eq!(history.raw_items(), expected.as_slice());
assert_eq!(
std::fs::read(&expected_saved_path).expect("saved file"),
b"foo"
@@ -7969,6 +7986,7 @@ async fn handle_output_item_done_skips_image_save_message_when_save_fails() {
status: "completed".to_string(),
revised_prompt: Some("broken payload".to_string()),
result: "_-8".to_string(),
metadata: None,
};
let mut ctx = HandleOutputCtx {
@@ -7985,7 +8003,8 @@ async fn handle_output_item_done_skips_image_save_message_when_save_fails() {
.expect("image generation item should still complete");
let history = session.clone_history().await;
assert_eq!(history.raw_items(), &[item]);
let expected = vec![item];
assert_eq!(history.raw_items(), expected.as_slice());
assert!(!expected_saved_path.exists());
}
@@ -8128,6 +8147,7 @@ async fn record_context_updates_and_set_reference_context_item_reinjects_full_co
text: format!("{}\nsummary", crate::compact::SUMMARY_PREFIX),
}],
phase: None,
metadata: None,
};
session
.record_conversation_items(&turn_context, std::slice::from_ref(&compacted_summary))
@@ -8152,7 +8172,8 @@ async fn record_context_updates_and_set_reference_context_item_reinjects_full_co
let history = session.clone_history().await;
let mut expected_history = vec![compacted_summary];
expected_history.extend(session.build_initial_context(&turn_context).await);
let initial_context = session.build_initial_context(&turn_context).await;
expected_history.extend(initial_context);
assert_eq!(history.raw_items().to_vec(), expected_history);
}
@@ -8725,6 +8746,7 @@ async fn task_finish_emits_turn_item_lifecycle_for_leftover_pending_user_input()
text: "late pending input".to_string(),
}],
phase: None,
metadata: None,
};
assert!(
history.raw_items().iter().any(|item| item == &expected),
@@ -9157,6 +9179,7 @@ async fn abort_empty_active_turn_preserves_pending_input() {
text: "late pending input".to_string(),
}],
phase: None,
metadata: None,
};
let turn_state = {
let mut active = sess.active_turn.lock().await;
@@ -9415,6 +9438,7 @@ async fn tool_calls_reopen_mailbox_delivery_for_current_turn() {
namespace: None,
arguments: "{}".to_string(),
call_id: "call-1".to_string(),
metadata: None,
};
let mut ctx = HandleOutputCtx {
sess: Arc::clone(&sess),
@@ -9542,6 +9566,7 @@ async fn fatal_tool_error_stops_turn_and_reports_error() {
call_id: "call-1".to_string(),
name: "shell_command".to_string(),
input: "{}".to_string(),
metadata: None,
};
let call = ToolRouter::build_tool_call(item.clone())
@@ -9626,6 +9651,7 @@ async fn sample_rollout(
text: "first user".to_string(),
}],
phase: None,
metadata: None,
};
live_history.record_items(
std::iter::once(&user1),
@@ -9640,6 +9666,7 @@ async fn sample_rollout(
text: "assistant reply one".to_string(),
}],
phase: None,
metadata: None,
};
live_history.record_items(
std::iter::once(&assistant1),
@@ -9667,6 +9694,7 @@ async fn sample_rollout(
text: "second user".to_string(),
}],
phase: None,
metadata: None,
};
live_history.record_items(
std::iter::once(&user2),
@@ -9681,6 +9709,7 @@ async fn sample_rollout(
text: "assistant reply two".to_string(),
}],
phase: None,
metadata: None,
};
live_history.record_items(
std::iter::once(&assistant2),
@@ -9708,6 +9737,7 @@ async fn sample_rollout(
text: "third user".to_string(),
}],
phase: None,
metadata: None,
};
live_history.record_items(
std::iter::once(&user3),
@@ -9722,6 +9752,7 @@ async fn sample_rollout(
text: "assistant reply three".to_string(),
}],
phase: None,
metadata: None,
};
live_history.record_items(
std::iter::once(&assistant3),
@@ -9867,6 +9898,7 @@ while :; do sleep 1; done"#,
})
.to_string(),
call_id: "shell-cleanup-call".to_string(),
metadata: None,
};
let call = ToolRouter::build_tool_call(item)?
.expect("shell command response item should build a tool call");
@@ -536,6 +536,7 @@ async fn process_compacted_history_preserves_separate_guardian_developer_message
text: "stale developer message".to_string(),
}],
phase: None,
metadata: None,
},
ResponseItem::Message {
id: None,
@@ -544,6 +545,7 @@ async fn process_compacted_history_preserves_separate_guardian_developer_message
text: "summary".to_string(),
}],
phase: None,
metadata: None,
},
],
InitialContextInjection::BeforeLastUserMessage,
+1 -1
View File
@@ -1978,7 +1978,7 @@ async fn try_run_sampling_request(
| ResponseItem::WebSearchCall { .. }
| ResponseItem::ImageGenerationCall { .. }
| ResponseItem::Compaction { .. }
| ResponseItem::CompactionTrigger
| ResponseItem::CompactionTrigger { .. }
| ResponseItem::ContextCompaction { .. }
| ResponseItem::Other => false,
};
+1
View File
@@ -33,6 +33,7 @@ fn assistant_output_text(text: &str) -> ResponseItem {
text: text.to_string(),
}],
phase: None,
metadata: None,
}
}