mirror of
https://github.com/openai/codex.git
synced 2026-11-05 08:28:29 +00:00
Test developer instruction inheritance for multi-agent workers (#35653)
## Testing - Verify full, compacted, and bounded history forks preserve developer messages while removing stale parent usage hints. - Add an app-server regression test that cold-resumes a root thread and confirms a roleless worker retains inherited developer instructions when lazily reloaded for follow-up work. GitOrigin-RevId: 40e454a2a2b721ea645cbc607b17a3ae6b617bd9
This commit is contained in:
Adam Perry @ OpenAI authored and copyberry committed
1 parent
462ed19a05
commit
9f82ffed1b
3 files changed
+347
No files matched your search
@@ -1209,6 +1209,7 @@ async fn spawn_agent_can_fork_parent_thread_history_with_sanitized_items() {
|
||||
let harness = AgentControlHarness::new().await;
|
||||
let mut parent_config = harness.config.clone();
|
||||
let _ = parent_config.features.enable(Feature::MultiAgentV2);
|
||||
parent_config.developer_instructions = Some("Parent developer instructions.".to_string());
|
||||
parent_config.multi_agent_v2.root_agent_usage_hint_text =
|
||||
Some("Parent root guidance.".to_string());
|
||||
parent_config.multi_agent_v2.subagent_usage_hint_text =
|
||||
@@ -1269,6 +1270,20 @@ async fn spawn_agent_can_fork_parent_thread_history_with_sanitized_items() {
|
||||
phase: None,
|
||||
internal_chat_message_metadata_passthrough: None,
|
||||
},
|
||||
ResponseItem::Message {
|
||||
id: None,
|
||||
role: "developer".to_string(),
|
||||
content: vec![
|
||||
ContentItem::InputText {
|
||||
text: "Parent developer instructions.".to_string(),
|
||||
},
|
||||
ContentItem::InputText {
|
||||
text: "Preserved developer context.".to_string(),
|
||||
},
|
||||
],
|
||||
phase: None,
|
||||
internal_chat_message_metadata_passthrough: None,
|
||||
},
|
||||
assistant_message("parent commentary", Some(MessagePhase::Commentary)),
|
||||
assistant_message("parent final answer", Some(MessagePhase::FinalAnswer)),
|
||||
assistant_message("parent unknown phase", /*phase*/ None),
|
||||
@@ -1333,8 +1348,24 @@ async fn spawn_agent_can_fork_parent_thread_history_with_sanitized_items() {
|
||||
let mut expected_final_answer =
|
||||
assistant_message("parent final answer", Some(MessagePhase::FinalAnswer));
|
||||
expected_final_answer.set_turn_id_if_missing(&turn_context.sub_id);
|
||||
let mut expected_developer_message = ResponseItem::Message {
|
||||
id: None,
|
||||
role: "developer".to_string(),
|
||||
content: vec![
|
||||
ContentItem::InputText {
|
||||
text: "Parent developer instructions.".to_string(),
|
||||
},
|
||||
ContentItem::InputText {
|
||||
text: "Preserved developer context.".to_string(),
|
||||
},
|
||||
],
|
||||
phase: None,
|
||||
internal_chat_message_metadata_passthrough: None,
|
||||
};
|
||||
expected_developer_message.set_turn_id_if_missing(&turn_context.sub_id);
|
||||
let expected_history = [
|
||||
expected_parent_seed,
|
||||
expected_developer_message,
|
||||
expected_final_answer,
|
||||
ResponseItem::Message {
|
||||
id: None,
|
||||
@@ -1393,6 +1424,17 @@ async fn spawn_agent_can_fork_parent_thread_history_with_sanitized_items() {
|
||||
!history_contains_text(no_hint_history.raw_items(), "Child subagent guidance."),
|
||||
"full-history forked child should not add empty subagent guidance"
|
||||
);
|
||||
assert!(
|
||||
history_contains_text(
|
||||
no_hint_history.raw_items(),
|
||||
"Parent developer instructions."
|
||||
),
|
||||
"full-history forked child should inherit parent developer instructions"
|
||||
);
|
||||
assert!(
|
||||
history_contains_text(no_hint_history.raw_items(), "Preserved developer context."),
|
||||
"full-history forked child should preserve unrelated developer fragments"
|
||||
);
|
||||
|
||||
let expected = (
|
||||
child_thread_id,
|
||||
@@ -1435,6 +1477,7 @@ async fn spawn_agent_fork_strips_parent_usage_hints_from_compacted_history() {
|
||||
let harness = AgentControlHarness::new().await;
|
||||
let mut parent_config = harness.config.clone();
|
||||
let _ = parent_config.features.enable(Feature::MultiAgentV2);
|
||||
parent_config.developer_instructions = Some("Parent developer instructions.".to_string());
|
||||
parent_config.multi_agent_v2.root_agent_usage_hint_text =
|
||||
Some("Parent root guidance.".to_string());
|
||||
parent_config.multi_agent_v2.subagent_usage_hint_text =
|
||||
@@ -1473,6 +1516,20 @@ async fn spawn_agent_fork_strips_parent_usage_hints_from_compacted_history() {
|
||||
phase: None,
|
||||
internal_chat_message_metadata_passthrough: None,
|
||||
},
|
||||
ResponseItem::Message {
|
||||
id: None,
|
||||
role: "developer".to_string(),
|
||||
content: vec![
|
||||
ContentItem::InputText {
|
||||
text: "Parent developer instructions.".to_string(),
|
||||
},
|
||||
ContentItem::InputText {
|
||||
text: "Preserved compacted developer context.".to_string(),
|
||||
},
|
||||
],
|
||||
phase: None,
|
||||
internal_chat_message_metadata_passthrough: None,
|
||||
},
|
||||
];
|
||||
parent_thread
|
||||
.session
|
||||
@@ -1532,6 +1589,17 @@ async fn spawn_agent_fork_strips_parent_usage_hints_from_compacted_history() {
|
||||
!history_contains_text(history.raw_items(), "Parent root guidance."),
|
||||
"forked child history should strip stale parent hints from compacted replacement history"
|
||||
);
|
||||
assert!(
|
||||
history_contains_text(history.raw_items(), "Parent developer instructions."),
|
||||
"forked child history should inherit compacted parent developer instructions"
|
||||
);
|
||||
assert!(
|
||||
history_contains_text(
|
||||
history.raw_items(),
|
||||
"Preserved compacted developer context."
|
||||
),
|
||||
"forked child history should preserve unrelated compacted developer fragments"
|
||||
);
|
||||
assert!(
|
||||
history_contains_text(history.raw_items(), "Child subagent guidance."),
|
||||
"full-history forked child should add the child subagent hint after compacted-history sanitization"
|
||||
@@ -1861,6 +1929,7 @@ async fn spawn_agent_fork_last_n_turns_strips_parent_usage_hints() {
|
||||
let harness = AgentControlHarness::new().await;
|
||||
let mut parent_config = harness.config.clone();
|
||||
let _ = parent_config.features.enable(Feature::MultiAgentV2);
|
||||
parent_config.developer_instructions = Some("Parent developer instructions.".to_string());
|
||||
parent_config.multi_agent_v2.root_agent_usage_hint_text =
|
||||
Some("Parent root guidance.".to_string());
|
||||
let mut child_config = harness.config.clone();
|
||||
@@ -1893,6 +1962,20 @@ async fn spawn_agent_fork_last_n_turns_strips_parent_usage_hints() {
|
||||
phase: None,
|
||||
internal_chat_message_metadata_passthrough: None,
|
||||
},
|
||||
ResponseItem::Message {
|
||||
id: None,
|
||||
role: "developer".to_string(),
|
||||
content: vec![
|
||||
ContentItem::InputText {
|
||||
text: "Parent developer instructions.".to_string(),
|
||||
},
|
||||
ContentItem::InputText {
|
||||
text: "Preserved bounded developer context.".to_string(),
|
||||
},
|
||||
],
|
||||
phase: None,
|
||||
internal_chat_message_metadata_passthrough: None,
|
||||
},
|
||||
spawn_agent_call(&parent_spawn_call_id),
|
||||
],
|
||||
)
|
||||
@@ -1940,6 +2023,14 @@ async fn spawn_agent_fork_last_n_turns_strips_parent_usage_hints() {
|
||||
!history_contains_text(history.raw_items(), "Parent root guidance."),
|
||||
"bounded fork should strip stale parent root hints before the child rebuilds startup context"
|
||||
);
|
||||
assert!(
|
||||
history_contains_text(history.raw_items(), "Parent developer instructions."),
|
||||
"bounded fork should inherit parent developer instructions from retained turns"
|
||||
);
|
||||
assert!(
|
||||
history_contains_text(history.raw_items(), "Preserved bounded developer context."),
|
||||
"bounded fork should preserve unrelated developer fragments"
|
||||
);
|
||||
|
||||
let _ = harness
|
||||
.control
|
||||
|
||||
Reference in new issue
Block a user