mirror of
https://github.com/openai/codex.git
synced 2026-11-03 19:53:25 +00:00
initial draft
This commit is contained in:
@@ -391,6 +391,7 @@ async fn spawn_agent_can_fork_parent_thread_history() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: parent_spawn_call_id.clone(),
|
||||
metadata: None,
|
||||
};
|
||||
parent_thread
|
||||
.codex
|
||||
@@ -474,6 +475,7 @@ async fn spawn_agent_fork_injects_output_for_parent_spawn_call() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: parent_spawn_call_id.clone(),
|
||||
metadata: None,
|
||||
};
|
||||
parent_thread
|
||||
.codex
|
||||
@@ -550,6 +552,7 @@ async fn spawn_agent_fork_flushes_parent_rollout_before_loading_history() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: parent_spawn_call_id.clone(),
|
||||
metadata: None,
|
||||
};
|
||||
parent_thread
|
||||
.codex
|
||||
|
||||
@@ -133,6 +133,7 @@ async fn build_arc_monitor_request_includes_relevant_history_and_null_policies()
|
||||
namespace: None,
|
||||
arguments: "{\"old\":true}".to_string(),
|
||||
call_id: "call_old".to_string(),
|
||||
metadata: None,
|
||||
}],
|
||||
&turn_context,
|
||||
)
|
||||
@@ -161,6 +162,7 @@ async fn build_arc_monitor_request_includes_relevant_history_and_null_policies()
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
}],
|
||||
&turn_context,
|
||||
)
|
||||
|
||||
@@ -79,6 +79,7 @@ fn reserialize_shell_outputs(items: &mut [ResponseItem]) {
|
||||
call_id,
|
||||
name,
|
||||
input: _,
|
||||
..
|
||||
} => {
|
||||
if name == "apply_patch" {
|
||||
shell_call_ids.insert(call_id.clone());
|
||||
|
||||
@@ -147,6 +147,7 @@ fn reserializes_shell_outputs_for_function_and_custom_tool_calls() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-1".to_string(),
|
||||
@@ -158,6 +159,7 @@ fn reserializes_shell_outputs_for_function_and_custom_tool_calls() {
|
||||
call_id: "call-2".to_string(),
|
||||
name: "apply_patch".to_string(),
|
||||
input: "*** Begin Patch".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::CustomToolCallOutput {
|
||||
call_id: "call-2".to_string(),
|
||||
@@ -176,6 +178,7 @@ fn reserializes_shell_outputs_for_function_and_custom_tool_calls() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-1".to_string(),
|
||||
@@ -187,6 +190,7 @@ fn reserializes_shell_outputs_for_function_and_custom_tool_calls() {
|
||||
call_id: "call-2".to_string(),
|
||||
name: "apply_patch".to_string(),
|
||||
input: "*** Begin Patch".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::CustomToolCallOutput {
|
||||
call_id: "call-2".to_string(),
|
||||
|
||||
@@ -321,6 +321,8 @@ use codex_protocol::models::ContentItem;
|
||||
use codex_protocol::models::DeveloperInstructions;
|
||||
use codex_protocol::models::ResponseInputItem;
|
||||
use codex_protocol::models::ResponseItem;
|
||||
use codex_protocol::models::ResponseItemMetadata;
|
||||
use codex_protocol::models::ReviewDecisionMetadata;
|
||||
use codex_protocol::openai_models::ReasoningEffort as ReasoningEffortConfig;
|
||||
use codex_protocol::protocol::CodexErrorInfo;
|
||||
use codex_protocol::protocol::InitialHistory;
|
||||
@@ -1010,6 +1012,92 @@ fn stamp_user_message_type_on_input_item(item: &mut ResponseInputItem, kind: Use
|
||||
*metadata = Some(metadata_value);
|
||||
}
|
||||
|
||||
fn review_decision_to_metadata(decision: &ReviewDecision) -> ReviewDecisionMetadata {
|
||||
match decision {
|
||||
ReviewDecision::Approved => ReviewDecisionMetadata::Approved,
|
||||
ReviewDecision::ApprovedExecpolicyAmendment { .. } => {
|
||||
ReviewDecisionMetadata::ApprovedWithAmendment
|
||||
}
|
||||
ReviewDecision::ApprovedForSession => ReviewDecisionMetadata::ApprovedForSession,
|
||||
ReviewDecision::NetworkPolicyAmendment {
|
||||
network_policy_amendment,
|
||||
} => match network_policy_amendment.action {
|
||||
codex_protocol::protocol::NetworkPolicyRuleAction::Allow => {
|
||||
ReviewDecisionMetadata::ApprovedWithNetworkPolicyAllow
|
||||
}
|
||||
codex_protocol::protocol::NetworkPolicyRuleAction::Deny => {
|
||||
ReviewDecisionMetadata::DeniedWithNetworkPolicyDeny
|
||||
}
|
||||
},
|
||||
ReviewDecision::Denied => ReviewDecisionMetadata::Denied,
|
||||
ReviewDecision::Abort => ReviewDecisionMetadata::Abort,
|
||||
}
|
||||
}
|
||||
|
||||
fn response_item_tool_call_id(item: &ResponseItem) -> Option<&str> {
|
||||
match item {
|
||||
ResponseItem::LocalShellCall {
|
||||
call_id: Some(call_id),
|
||||
..
|
||||
} => Some(call_id),
|
||||
ResponseItem::FunctionCall { call_id, .. } => Some(call_id),
|
||||
ResponseItem::CustomToolCall { call_id, .. } => Some(call_id),
|
||||
_ => None,
|
||||
}
|
||||
}
|
||||
|
||||
fn stamp_tool_metadata_on_response_item(
|
||||
item: ResponseItem,
|
||||
metadata: ResponseItemMetadata,
|
||||
) -> ResponseItem {
|
||||
match item {
|
||||
ResponseItem::LocalShellCall {
|
||||
id,
|
||||
call_id,
|
||||
status,
|
||||
action,
|
||||
..
|
||||
} => ResponseItem::LocalShellCall {
|
||||
id,
|
||||
call_id,
|
||||
status,
|
||||
action,
|
||||
metadata: Some(metadata),
|
||||
},
|
||||
ResponseItem::FunctionCall {
|
||||
id,
|
||||
name,
|
||||
namespace,
|
||||
arguments,
|
||||
call_id,
|
||||
..
|
||||
} => ResponseItem::FunctionCall {
|
||||
id,
|
||||
name,
|
||||
namespace,
|
||||
arguments,
|
||||
call_id,
|
||||
metadata: Some(metadata),
|
||||
},
|
||||
ResponseItem::CustomToolCall {
|
||||
id,
|
||||
status,
|
||||
call_id,
|
||||
name,
|
||||
input,
|
||||
..
|
||||
} => ResponseItem::CustomToolCall {
|
||||
id,
|
||||
status,
|
||||
call_id,
|
||||
name,
|
||||
input,
|
||||
metadata: Some(metadata),
|
||||
},
|
||||
other => other,
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Clone)]
|
||||
pub(crate) struct SessionConfiguration {
|
||||
/// Provider identifier ("openai", "openrouter", ...).
|
||||
@@ -2904,6 +2992,10 @@ impl Session {
|
||||
match active.as_mut() {
|
||||
Some(at) => {
|
||||
let mut ts = at.turn_state.lock().await;
|
||||
ts.insert_pending_approval_call_id(
|
||||
effective_approval_id.clone(),
|
||||
call_id.clone(),
|
||||
);
|
||||
ts.insert_pending_approval(effective_approval_id.clone(), tx_approve)
|
||||
}
|
||||
None => None,
|
||||
@@ -2969,6 +3061,7 @@ impl Session {
|
||||
match active.as_mut() {
|
||||
Some(at) => {
|
||||
let mut ts = at.turn_state.lock().await;
|
||||
ts.insert_pending_approval_call_id(approval_id.clone(), call_id.clone());
|
||||
ts.insert_pending_approval(approval_id.clone(), tx_approve)
|
||||
}
|
||||
None => None,
|
||||
@@ -3254,7 +3347,20 @@ impl Session {
|
||||
}
|
||||
}
|
||||
|
||||
pub async fn record_approval_outcome(&self, approval_id: &str, decision: &ReviewDecision) {
|
||||
let mut active = self.active_turn.lock().await;
|
||||
let Some(at) = active.as_mut() else {
|
||||
return;
|
||||
};
|
||||
let mut ts = at.turn_state.lock().await;
|
||||
let call_id = ts
|
||||
.remove_pending_approval_call_id(approval_id)
|
||||
.unwrap_or_else(|| approval_id.to_string());
|
||||
ts.record_approval_outcome(call_id, review_decision_to_metadata(decision));
|
||||
}
|
||||
|
||||
pub async fn notify_approval(&self, approval_id: &str, decision: ReviewDecision) {
|
||||
self.record_approval_outcome(approval_id, &decision).await;
|
||||
let entry = {
|
||||
let mut active = self.active_turn.lock().await;
|
||||
match active.as_mut() {
|
||||
@@ -3275,6 +3381,53 @@ impl Session {
|
||||
}
|
||||
}
|
||||
|
||||
async fn stamp_tool_approval_metadata(
|
||||
&self,
|
||||
_turn_context: &TurnContext,
|
||||
response_item: ResponseItem,
|
||||
) -> ResponseItem {
|
||||
if !self.enabled(Feature::ItemMetadata) {
|
||||
return response_item;
|
||||
}
|
||||
let Some(call_id) = response_item_tool_call_id(&response_item) else {
|
||||
return response_item;
|
||||
};
|
||||
|
||||
let (outcome, has_pending_approval) = {
|
||||
let active = self.active_turn.lock().await;
|
||||
let Some(at) = active.as_ref() else {
|
||||
return response_item;
|
||||
};
|
||||
let ts = at.turn_state.lock().await;
|
||||
let outcome = ts.approval_outcome_for_call_id(call_id);
|
||||
let has_pending_approval = ts.has_pending_approval_for_call_id(call_id);
|
||||
(outcome, has_pending_approval)
|
||||
};
|
||||
|
||||
let mut metadata = match &response_item {
|
||||
ResponseItem::LocalShellCall { metadata, .. }
|
||||
| ResponseItem::FunctionCall { metadata, .. }
|
||||
| ResponseItem::CustomToolCall { metadata, .. } => metadata.clone().unwrap_or_default(),
|
||||
_ => return response_item,
|
||||
};
|
||||
|
||||
match outcome {
|
||||
Some(review_decision) => {
|
||||
metadata.is_tool_call_escalated = Some(true);
|
||||
metadata.review_decision = Some(review_decision);
|
||||
}
|
||||
None if !has_pending_approval => {
|
||||
metadata.is_tool_call_escalated = Some(false);
|
||||
metadata.review_decision = None;
|
||||
}
|
||||
None => {
|
||||
return response_item;
|
||||
}
|
||||
}
|
||||
|
||||
stamp_tool_metadata_on_response_item(response_item, metadata)
|
||||
}
|
||||
|
||||
pub async fn resolve_elicitation(
|
||||
&self,
|
||||
server_name: String,
|
||||
@@ -3769,6 +3922,10 @@ impl Session {
|
||||
turn_context: &TurnContext,
|
||||
response_item: ResponseItem,
|
||||
) {
|
||||
let response_item = self
|
||||
.stamp_tool_approval_metadata(turn_context, response_item)
|
||||
.await;
|
||||
|
||||
// Add to conversation history and persist response item to rollout.
|
||||
self.record_conversation_items(turn_context, std::slice::from_ref(&response_item))
|
||||
.await;
|
||||
@@ -4731,6 +4888,8 @@ mod handlers {
|
||||
}
|
||||
match decision {
|
||||
ReviewDecision::Abort => {
|
||||
sess.record_approval_outcome(&approval_id, &ReviewDecision::Abort)
|
||||
.await;
|
||||
sess.interrupt_task().await;
|
||||
}
|
||||
other => sess.notify_approval(&approval_id, other).await,
|
||||
|
||||
@@ -77,6 +77,7 @@ async fn forward_events_cancelled_while_send_blocked_shuts_down_delegate() {
|
||||
call_id: "call-1".to_string(),
|
||||
name: "tool".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
}),
|
||||
})
|
||||
|
||||
@@ -3920,6 +3920,137 @@ async fn task_finish_emits_prompt_queued_metadata_for_injected_user_input_when_f
|
||||
));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn review_decision_metadata_mapping_is_stable() {
|
||||
assert_eq!(
|
||||
review_decision_to_metadata(&ReviewDecision::Approved),
|
||||
codex_protocol::models::ReviewDecisionMetadata::Approved
|
||||
);
|
||||
assert_eq!(
|
||||
review_decision_to_metadata(&ReviewDecision::Denied),
|
||||
codex_protocol::models::ReviewDecisionMetadata::Denied
|
||||
);
|
||||
assert_eq!(
|
||||
review_decision_to_metadata(&ReviewDecision::Abort),
|
||||
codex_protocol::models::ReviewDecisionMetadata::Abort
|
||||
);
|
||||
assert_eq!(
|
||||
review_decision_to_metadata(&ReviewDecision::ApprovedForSession),
|
||||
codex_protocol::models::ReviewDecisionMetadata::ApprovedForSession
|
||||
);
|
||||
}
|
||||
|
||||
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
|
||||
async fn tool_call_metadata_stamps_escalated_review_decision_when_feature_enabled() {
|
||||
let (mut sess, tc, rx) = make_session_and_context_with_rx().await;
|
||||
Arc::get_mut(&mut sess)
|
||||
.expect("session should be uniquely owned in this test")
|
||||
.features
|
||||
.enable(crate::features::Feature::ItemMetadata)
|
||||
.expect("feature flag should be enabled for this test");
|
||||
|
||||
sess.spawn_task(
|
||||
Arc::clone(&tc),
|
||||
vec![UserInput::Text {
|
||||
text: "start".to_string(),
|
||||
text_elements: Vec::new(),
|
||||
}],
|
||||
NeverEndingTask {
|
||||
kind: TaskKind::Regular,
|
||||
listen_to_cancellation_token: false,
|
||||
},
|
||||
)
|
||||
.await;
|
||||
while rx.try_recv().is_ok() {}
|
||||
|
||||
sess.record_approval_outcome("call-1", &ReviewDecision::Denied)
|
||||
.await;
|
||||
sess.record_response_item_and_emit_turn_item(
|
||||
tc.as_ref(),
|
||||
ResponseItem::FunctionCall {
|
||||
id: None,
|
||||
name: "shell".to_string(),
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
)
|
||||
.await;
|
||||
|
||||
let event = tokio::time::timeout(std::time::Duration::from_secs(2), rx.recv())
|
||||
.await
|
||||
.expect("expected raw response item event")
|
||||
.expect("channel open");
|
||||
assert!(matches!(
|
||||
event.msg,
|
||||
EventMsg::RawResponseItem(ref ev)
|
||||
if matches!(
|
||||
&ev.item,
|
||||
ResponseItem::FunctionCall {
|
||||
metadata: Some(metadata),
|
||||
..
|
||||
} if metadata.is_tool_call_escalated == Some(true)
|
||||
&& metadata.review_decision
|
||||
== Some(codex_protocol::models::ReviewDecisionMetadata::Denied)
|
||||
)
|
||||
));
|
||||
}
|
||||
|
||||
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
|
||||
async fn tool_call_metadata_stamps_non_escalated_false_when_feature_enabled() {
|
||||
let (mut sess, tc, rx) = make_session_and_context_with_rx().await;
|
||||
Arc::get_mut(&mut sess)
|
||||
.expect("session should be uniquely owned in this test")
|
||||
.features
|
||||
.enable(crate::features::Feature::ItemMetadata)
|
||||
.expect("feature flag should be enabled for this test");
|
||||
|
||||
sess.spawn_task(
|
||||
Arc::clone(&tc),
|
||||
vec![UserInput::Text {
|
||||
text: "start".to_string(),
|
||||
text_elements: Vec::new(),
|
||||
}],
|
||||
NeverEndingTask {
|
||||
kind: TaskKind::Regular,
|
||||
listen_to_cancellation_token: false,
|
||||
},
|
||||
)
|
||||
.await;
|
||||
while rx.try_recv().is_ok() {}
|
||||
|
||||
sess.record_response_item_and_emit_turn_item(
|
||||
tc.as_ref(),
|
||||
ResponseItem::FunctionCall {
|
||||
id: None,
|
||||
name: "shell".to_string(),
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-2".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
)
|
||||
.await;
|
||||
|
||||
let event = tokio::time::timeout(std::time::Duration::from_secs(2), rx.recv())
|
||||
.await
|
||||
.expect("expected raw response item event")
|
||||
.expect("channel open");
|
||||
assert!(matches!(
|
||||
event.msg,
|
||||
EventMsg::RawResponseItem(ref ev)
|
||||
if matches!(
|
||||
&ev.item,
|
||||
ResponseItem::FunctionCall {
|
||||
metadata: Some(metadata),
|
||||
..
|
||||
} if metadata.is_tool_call_escalated == Some(false)
|
||||
&& metadata.review_decision.is_none()
|
||||
)
|
||||
));
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn steer_input_requires_active_turn() {
|
||||
let (sess, _tc, _rx) = make_session_and_context_with_rx().await;
|
||||
@@ -4119,6 +4250,7 @@ async fn fatal_tool_error_stops_turn_and_reports_error() {
|
||||
call_id: "call-1".to_string(),
|
||||
name: "shell".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
};
|
||||
|
||||
let call = ToolRouter::build_tool_call(session.as_ref(), item.clone())
|
||||
|
||||
@@ -281,6 +281,7 @@ fn for_prompt_strips_images_when_model_does_not_support_images() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-1".to_string(),
|
||||
@@ -300,6 +301,7 @@ fn for_prompt_strips_images_when_model_does_not_support_images() {
|
||||
call_id: "tool-1".to_string(),
|
||||
name: "js_repl".to_string(),
|
||||
input: "view_image".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::CustomToolCallOutput {
|
||||
call_id: "tool-1".to_string(),
|
||||
@@ -344,6 +346,7 @@ fn for_prompt_strips_images_when_model_does_not_support_images() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-1".to_string(),
|
||||
@@ -363,6 +366,7 @@ fn for_prompt_strips_images_when_model_does_not_support_images() {
|
||||
call_id: "tool-1".to_string(),
|
||||
name: "js_repl".to_string(),
|
||||
input: "view_image".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::CustomToolCallOutput {
|
||||
call_id: "tool-1".to_string(),
|
||||
@@ -536,6 +540,7 @@ fn remove_first_item_removes_matching_output_for_function_call() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-1".to_string(),
|
||||
@@ -560,6 +565,7 @@ fn remove_first_item_removes_matching_call_for_output() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-2".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
];
|
||||
let mut h = create_history_with_items(items);
|
||||
@@ -577,6 +583,7 @@ fn remove_last_item_removes_matching_call_for_output() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-delete-last".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-delete-last".to_string(),
|
||||
@@ -661,6 +668,7 @@ fn remove_first_item_handles_local_shell_pair() {
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-3".to_string(),
|
||||
@@ -817,6 +825,7 @@ fn remove_first_item_handles_custom_tool_pair() {
|
||||
call_id: "tool-1".to_string(),
|
||||
name: "my_tool".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::CustomToolCallOutput {
|
||||
call_id: "tool-1".to_string(),
|
||||
@@ -842,6 +851,7 @@ fn normalization_retains_local_shell_outputs() {
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "shell-1".to_string(),
|
||||
@@ -1052,6 +1062,7 @@ fn normalize_adds_missing_output_for_function_call() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-x".to_string(),
|
||||
metadata: None,
|
||||
}];
|
||||
let mut h = create_history_with_items(items);
|
||||
|
||||
@@ -1066,6 +1077,7 @@ fn normalize_adds_missing_output_for_function_call() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-x".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-x".to_string(),
|
||||
@@ -1084,6 +1096,7 @@ fn normalize_adds_missing_output_for_custom_tool_call() {
|
||||
call_id: "tool-x".to_string(),
|
||||
name: "custom".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
}];
|
||||
let mut h = create_history_with_items(items);
|
||||
|
||||
@@ -1098,6 +1111,7 @@ fn normalize_adds_missing_output_for_custom_tool_call() {
|
||||
call_id: "tool-x".to_string(),
|
||||
name: "custom".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::CustomToolCallOutput {
|
||||
call_id: "tool-x".to_string(),
|
||||
@@ -1121,6 +1135,7 @@ fn normalize_adds_missing_output_for_local_shell_call_with_id() {
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
}];
|
||||
let mut h = create_history_with_items(items);
|
||||
|
||||
@@ -1140,6 +1155,7 @@ fn normalize_adds_missing_output_for_local_shell_call_with_id() {
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "shell-1".to_string(),
|
||||
@@ -1188,6 +1204,7 @@ fn normalize_mixed_inserts_and_removals() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "c1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
// Orphan output that should be removed
|
||||
ResponseItem::FunctionCallOutput {
|
||||
@@ -1201,6 +1218,7 @@ fn normalize_mixed_inserts_and_removals() {
|
||||
call_id: "t1".to_string(),
|
||||
name: "tool".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
// Local shell call also gets an inserted function call output
|
||||
ResponseItem::LocalShellCall {
|
||||
@@ -1214,6 +1232,7 @@ fn normalize_mixed_inserts_and_removals() {
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
},
|
||||
];
|
||||
let mut h = create_history_with_items(items);
|
||||
@@ -1229,6 +1248,7 @@ fn normalize_mixed_inserts_and_removals() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "c1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "c1".to_string(),
|
||||
@@ -1240,6 +1260,7 @@ fn normalize_mixed_inserts_and_removals() {
|
||||
call_id: "t1".to_string(),
|
||||
name: "tool".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::CustomToolCallOutput {
|
||||
call_id: "t1".to_string(),
|
||||
@@ -1273,6 +1294,7 @@ fn normalize_adds_missing_output_for_function_call_inserts_output() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-x".to_string(),
|
||||
metadata: None,
|
||||
}];
|
||||
let mut h = create_history_with_items(items);
|
||||
h.normalize_history(&default_input_modalities());
|
||||
@@ -1285,6 +1307,7 @@ fn normalize_adds_missing_output_for_function_call_inserts_output() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-x".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-x".to_string(),
|
||||
@@ -1337,6 +1360,7 @@ fn normalize_adds_missing_output_for_custom_tool_call_panics_in_debug() {
|
||||
call_id: "tool-x".to_string(),
|
||||
name: "custom".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
}];
|
||||
let mut h = create_history_with_items(items);
|
||||
h.normalize_history(&default_input_modalities());
|
||||
@@ -1357,6 +1381,7 @@ fn normalize_adds_missing_output_for_local_shell_call_with_id_panics_in_debug()
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
}];
|
||||
let mut h = create_history_with_items(items);
|
||||
h.normalize_history(&default_input_modalities());
|
||||
@@ -1450,6 +1475,7 @@ fn normalize_mixed_inserts_and_removals_panics_in_debug() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "c1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "c2".to_string(),
|
||||
@@ -1461,6 +1487,7 @@ fn normalize_mixed_inserts_and_removals_panics_in_debug() {
|
||||
call_id: "t1".to_string(),
|
||||
name: "tool".to_string(),
|
||||
input: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::LocalShellCall {
|
||||
id: None,
|
||||
@@ -1473,6 +1500,7 @@ fn normalize_mixed_inserts_and_removals_panics_in_debug() {
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
},
|
||||
];
|
||||
let mut h = create_history_with_items(items);
|
||||
|
||||
@@ -45,10 +45,6 @@ const ALIASES: &[Alias] = &[
|
||||
legacy_key: "memory_tool",
|
||||
feature: Feature::MemoryTool,
|
||||
},
|
||||
Alias {
|
||||
legacy_key: "user_message_type_metadata",
|
||||
feature: Feature::ItemMetadata,
|
||||
},
|
||||
];
|
||||
|
||||
pub(crate) fn legacy_feature_keys() -> impl Iterator<Item = &'static str> {
|
||||
|
||||
@@ -142,12 +142,6 @@ fn collab_is_legacy_alias_for_multi_agent() {
|
||||
assert_eq!(feature_for_key("collab"), Some(Feature::Collab));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn user_message_type_metadata_is_legacy_alias_for_item_metadata() {
|
||||
assert_eq!(feature_for_key("user_message_type_metadata"), Some(Feature::ItemMetadata));
|
||||
assert_eq!(feature_for_key("item_metadata"), Some(Feature::ItemMetadata));
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn item_metadata_is_under_development_and_disabled_by_default() {
|
||||
assert_eq!(Feature::ItemMetadata.stage(), Stage::UnderDevelopment);
|
||||
|
||||
@@ -115,6 +115,7 @@ fn collect_guardian_transcript_entries_includes_recent_tool_calls_and_output() {
|
||||
namespace: None,
|
||||
arguments: "{\"path\":\"README.md\"}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-1".to_string(),
|
||||
@@ -444,6 +445,7 @@ async fn guardian_review_request_layout_matches_model_visible_request_snapshot()
|
||||
namespace: None,
|
||||
arguments: "{\"repo\":\"openai/codex\"}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
ResponseItem::FunctionCallOutput {
|
||||
call_id: "call-1".to_string(),
|
||||
|
||||
@@ -54,6 +54,7 @@ fn truncates_rollout_from_start_before_nth_user_only() {
|
||||
name: "tool".to_string(),
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
assistant_msg("a4"),
|
||||
];
|
||||
|
||||
@@ -10,6 +10,7 @@ use tokio_util::task::AbortOnDropHandle;
|
||||
|
||||
use codex_protocol::dynamic_tools::DynamicToolResponse;
|
||||
use codex_protocol::models::ResponseInputItem;
|
||||
use codex_protocol::models::ReviewDecisionMetadata;
|
||||
use codex_protocol::models::UserMessageType;
|
||||
use codex_protocol::request_permissions::RequestPermissionsResponse;
|
||||
use codex_protocol::request_user_input::RequestUserInputResponse;
|
||||
@@ -83,6 +84,8 @@ impl ActiveTurn {
|
||||
#[derive(Default)]
|
||||
pub(crate) struct TurnState {
|
||||
pending_approvals: HashMap<String, oneshot::Sender<ReviewDecision>>,
|
||||
pending_approval_call_ids: HashMap<String, String>,
|
||||
approval_outcomes_by_call_id: HashMap<String, ReviewDecisionMetadata>,
|
||||
pending_request_permissions: HashMap<String, oneshot::Sender<RequestPermissionsResponse>>,
|
||||
pending_user_input: HashMap<String, oneshot::Sender<RequestUserInputResponse>>,
|
||||
pending_elicitations: HashMap<(String, RequestId), oneshot::Sender<ElicitationResponse>>,
|
||||
@@ -102,6 +105,39 @@ impl TurnState {
|
||||
self.pending_approvals.insert(key, tx)
|
||||
}
|
||||
|
||||
pub(crate) fn insert_pending_approval_call_id(
|
||||
&mut self,
|
||||
approval_key: String,
|
||||
call_id: String,
|
||||
) -> Option<String> {
|
||||
self.pending_approval_call_ids.insert(approval_key, call_id)
|
||||
}
|
||||
|
||||
pub(crate) fn remove_pending_approval_call_id(&mut self, approval_key: &str) -> Option<String> {
|
||||
self.pending_approval_call_ids.remove(approval_key)
|
||||
}
|
||||
|
||||
pub(crate) fn record_approval_outcome(
|
||||
&mut self,
|
||||
call_id: String,
|
||||
decision: ReviewDecisionMetadata,
|
||||
) {
|
||||
self.approval_outcomes_by_call_id.insert(call_id, decision);
|
||||
}
|
||||
|
||||
pub(crate) fn approval_outcome_for_call_id(
|
||||
&self,
|
||||
call_id: &str,
|
||||
) -> Option<ReviewDecisionMetadata> {
|
||||
self.approval_outcomes_by_call_id.get(call_id).cloned()
|
||||
}
|
||||
|
||||
pub(crate) fn has_pending_approval_for_call_id(&self, call_id: &str) -> bool {
|
||||
self.pending_approval_call_ids
|
||||
.values()
|
||||
.any(|pending_call_id| pending_call_id == call_id)
|
||||
}
|
||||
|
||||
pub(crate) fn remove_pending_approval(
|
||||
&mut self,
|
||||
key: &str,
|
||||
@@ -111,6 +147,8 @@ impl TurnState {
|
||||
|
||||
pub(crate) fn clear_pending(&mut self) {
|
||||
self.pending_approvals.clear();
|
||||
self.pending_approval_call_ids.clear();
|
||||
self.approval_outcomes_by_call_id.clear();
|
||||
self.pending_request_permissions.clear();
|
||||
self.pending_user_input.clear();
|
||||
self.pending_elicitations.clear();
|
||||
|
||||
@@ -56,6 +56,7 @@ fn drops_from_last_user_only() {
|
||||
name: "tool".to_string(),
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
assistant_msg("a4"),
|
||||
];
|
||||
|
||||
@@ -139,6 +139,7 @@ async fn build_tool_call_uses_namespace_for_registry_name() -> anyhow::Result<()
|
||||
namespace: Some("mcp__codex_apps__calendar".to_string()),
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-namespace".to_string(),
|
||||
metadata: None,
|
||||
},
|
||||
)
|
||||
.await?
|
||||
|
||||
@@ -83,6 +83,7 @@ fn response_item_records_turn_ttft_for_first_output_signals() {
|
||||
namespace: None,
|
||||
arguments: "{}".to_string(),
|
||||
call_id: "call-1".to_string(),
|
||||
metadata: None,
|
||||
}
|
||||
));
|
||||
assert!(response_item_records_turn_ttft(
|
||||
@@ -92,6 +93,7 @@ fn response_item_records_turn_ttft_for_first_output_signals() {
|
||||
call_id: "call-2".to_string(),
|
||||
name: "custom".to_string(),
|
||||
input: "echo hi".to_string(),
|
||||
metadata: None,
|
||||
}
|
||||
));
|
||||
assert!(response_item_records_turn_ttft(&ResponseItem::Message {
|
||||
|
||||
@@ -366,6 +366,7 @@ async fn resume_replays_legacy_js_repl_image_rollout_shapes() {
|
||||
call_id: "legacy-js-call".to_string(),
|
||||
name: "js_repl".to_string(),
|
||||
input: "console.log('legacy image flow')".to_string(),
|
||||
metadata: None,
|
||||
};
|
||||
let legacy_image_url = "data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR4nGP4z8DwHwAFAAH/iZk9HQAAAABJRU5ErkJggg==";
|
||||
let rollout = vec![
|
||||
@@ -522,6 +523,7 @@ async fn resume_replays_image_tool_outputs_with_detail() {
|
||||
namespace: None,
|
||||
arguments: "{\"path\":\"/tmp/example.webp\"}".to_string(),
|
||||
call_id: function_call_id.to_string(),
|
||||
metadata: None,
|
||||
}),
|
||||
},
|
||||
RolloutLine {
|
||||
@@ -544,6 +546,7 @@ async fn resume_replays_image_tool_outputs_with_detail() {
|
||||
call_id: custom_call_id.to_string(),
|
||||
name: "js_repl".to_string(),
|
||||
input: "console.log('image flow')".to_string(),
|
||||
metadata: None,
|
||||
}),
|
||||
},
|
||||
RolloutLine {
|
||||
@@ -1887,6 +1890,7 @@ async fn azure_responses_request_includes_store_and_reasoning_ids() {
|
||||
namespace: None,
|
||||
arguments: "{}".into(),
|
||||
call_id: "function-call-id".into(),
|
||||
metadata: None,
|
||||
});
|
||||
prompt.input.push(ResponseItem::FunctionCallOutput {
|
||||
call_id: "function-call-id".into(),
|
||||
@@ -1903,6 +1907,7 @@ async fn azure_responses_request_includes_store_and_reasoning_ids() {
|
||||
env: None,
|
||||
user: None,
|
||||
}),
|
||||
metadata: None,
|
||||
});
|
||||
prompt.input.push(ResponseItem::CustomToolCall {
|
||||
id: Some("custom-tool-id".into()),
|
||||
@@ -1910,6 +1915,7 @@ async fn azure_responses_request_includes_store_and_reasoning_ids() {
|
||||
call_id: "custom-tool-call-id".into(),
|
||||
name: "custom_tool".into(),
|
||||
input: "{}".into(),
|
||||
metadata: None,
|
||||
});
|
||||
prompt.input.push(ResponseItem::CustomToolCallOutput {
|
||||
call_id: "custom-tool-call-id".into(),
|
||||
|
||||
@@ -262,11 +262,30 @@ pub enum UserMessageType {
|
||||
PromptQueued,
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq, JsonSchema, TS)]
|
||||
#[serde(rename_all = "snake_case")]
|
||||
#[ts(rename_all = "snake_case")]
|
||||
pub enum ReviewDecisionMetadata {
|
||||
Approved,
|
||||
Denied,
|
||||
Abort,
|
||||
ApprovedForSession,
|
||||
ApprovedWithAmendment,
|
||||
ApprovedWithNetworkPolicyAllow,
|
||||
DeniedWithNetworkPolicyDeny,
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone, Default, Serialize, Deserialize, PartialEq, Eq, JsonSchema, TS)]
|
||||
pub struct ResponseItemMetadata {
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
#[ts(optional)]
|
||||
pub user_message_type: Option<UserMessageType>,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
#[ts(optional)]
|
||||
pub is_tool_call_escalated: Option<bool>,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
#[ts(optional)]
|
||||
pub review_decision: Option<ReviewDecisionMetadata>,
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, JsonSchema, TS)]
|
||||
@@ -344,6 +363,9 @@ pub enum ResponseItem {
|
||||
call_id: Option<String>,
|
||||
status: LocalShellStatus,
|
||||
action: LocalShellAction,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
#[ts(optional)]
|
||||
metadata: Option<ResponseItemMetadata>,
|
||||
},
|
||||
FunctionCall {
|
||||
#[serde(default, skip_serializing)]
|
||||
@@ -358,6 +380,9 @@ pub enum ResponseItem {
|
||||
// Session::handle_function_call parse it into a Value.
|
||||
arguments: String,
|
||||
call_id: String,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
#[ts(optional)]
|
||||
metadata: Option<ResponseItemMetadata>,
|
||||
},
|
||||
ToolSearchCall {
|
||||
#[serde(default, skip_serializing)]
|
||||
@@ -391,6 +416,9 @@ pub enum ResponseItem {
|
||||
call_id: String,
|
||||
name: String,
|
||||
input: String,
|
||||
#[serde(default, skip_serializing_if = "Option::is_none")]
|
||||
#[ts(optional)]
|
||||
metadata: Option<ResponseItemMetadata>,
|
||||
},
|
||||
// `custom_tool_call_output.output` uses the same wire encoding as
|
||||
// `function_call_output.output` so freeform tools can return either plain
|
||||
|
||||
Reference in New Issue
Block a user