mirror of
https://github.com/openai/codex.git
synced 2026-11-08 08:37:30 +00:00
Distinguish forked sessions in session-start hooks (#44349)
## Why Forked threads reported `startup`, causing startup hooks to run again even when their context was inherited from the parent. Resuming with supplied history also reported `startup` instead of `resume`. ## What changed - Add `fork` as a `SessionStart` source and expose it in the hook input schema. - Report `fork` for histories with a fork parent and `resume` for supplied histories without one. - Continue dispatching `SubagentStart` for spawned subagents with either fresh or forked context. ## Testing Add regression coverage for fork hook matching without duplicate startup context, resume hooks with supplied history, and `SubagentStart` context injection for fresh and forked subagents. GitOrigin-RevId: 5ac1ae4ab8adcbbbceb8cb0ffb9b32b6402c585b
This commit is contained in:
1 parent
c6a59ef923
commit
e444aa99d7
7 files changed
+197
-16
No files matched your search
@@ -7,6 +7,7 @@ use anyhow::Result;
|
||||
use codex_config::HookStateToml;
|
||||
use codex_config::McpServerConfig;
|
||||
use codex_config::test_support::CloudConfigBundleFixture;
|
||||
use codex_core::ForkSnapshot;
|
||||
use codex_core::StartThreadOptions;
|
||||
use codex_core::TurnInput;
|
||||
use codex_core::TurnInputRequest;
|
||||
@@ -15,12 +16,14 @@ use codex_core::config::Config;
|
||||
use codex_core::config::Constrained;
|
||||
use codex_core::config::ThreadStoreConfig;
|
||||
use codex_features::Feature;
|
||||
use codex_history::InitialHistory;
|
||||
use codex_history::RolloutItem;
|
||||
use codex_model_provider_info::ModelProviderInfo;
|
||||
use codex_model_provider_info::built_in_model_providers;
|
||||
use codex_plugin::PluginHookSource;
|
||||
use codex_plugin::PluginId;
|
||||
use codex_protocol::items::parse_hook_prompt_fragment;
|
||||
use codex_protocol::mcp::ClientMcpExtensions;
|
||||
use codex_protocol::models::ContentItem;
|
||||
use codex_protocol::models::NetworkPermissions;
|
||||
use codex_protocol::models::PermissionProfile;
|
||||
@@ -1612,6 +1615,159 @@ async fn session_start_runs_before_user_prompt_submit_on_first_turn() -> Result<
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn forked_thread_matches_fork_session_start_without_repeating_startup_context() -> Result<()>
|
||||
{
|
||||
skip_if_no_network!(Ok(()));
|
||||
|
||||
let server = start_mock_server().await;
|
||||
let responses = mount_sse_sequence(
|
||||
&server,
|
||||
vec![
|
||||
sse(vec![ev_completed("resp-1")]),
|
||||
sse(vec![ev_completed("resp-2")]),
|
||||
sse(vec![ev_completed("resp-fork")]),
|
||||
],
|
||||
)
|
||||
.await;
|
||||
let mut builder = test_codex()
|
||||
.with_pre_build_hook(|home| {
|
||||
let script_path = home.join("session_start_hook.py");
|
||||
fs::write(
|
||||
&script_path,
|
||||
r#"import json
|
||||
import sys
|
||||
|
||||
source = json.load(sys.stdin)["source"]
|
||||
print(json.dumps({"hookSpecificOutput": {
|
||||
"hookEventName": "SessionStart",
|
||||
"additionalContext": source + " hook context"
|
||||
}}))
|
||||
"#,
|
||||
)
|
||||
.expect("write session start hook");
|
||||
let groups = ["^startup$", "^fork$"].map(|matcher| {
|
||||
serde_json::json!({
|
||||
"matcher": matcher,
|
||||
"hooks": [{
|
||||
"type": "command",
|
||||
"command": format!("python3 {}", script_path.display()),
|
||||
}],
|
||||
})
|
||||
});
|
||||
fs::write(
|
||||
home.join("hooks.json"),
|
||||
serde_json::json!({"hooks": {"SessionStart": groups}}).to_string(),
|
||||
)
|
||||
.expect("write hooks.json");
|
||||
})
|
||||
.with_config(trust_discovered_hooks);
|
||||
// Command hooks run on the host and require a host-native working directory.
|
||||
let test = builder.build(&server).await?;
|
||||
test.submit_turn("first prompt").await?;
|
||||
test.submit_turn("second prompt").await?;
|
||||
test.codex.flush_rollout().await?;
|
||||
|
||||
let forked = test
|
||||
.thread_manager
|
||||
.fork_thread(
|
||||
ForkSnapshot::TruncateBeforeNthUserMessage(1),
|
||||
test.config.clone(),
|
||||
test.codex.rollout_path().expect("parent rollout path"),
|
||||
/*thread_source*/ None,
|
||||
/*parent_trace*/ None,
|
||||
)
|
||||
.await?
|
||||
.thread;
|
||||
forked
|
||||
.start_or_steer_turn(TurnInputRequest::user_input(vec![UserInput::Text {
|
||||
text: "edited second prompt".to_string(),
|
||||
text_elements: Vec::new(),
|
||||
}]))
|
||||
.await?;
|
||||
wait_for_event(&forked, |event| matches!(event, EventMsg::TurnComplete(_))).await;
|
||||
|
||||
let requests = responses.requests();
|
||||
assert_eq!(requests.len(), 3);
|
||||
let hook_contexts = requests[2]
|
||||
.message_input_texts("developer")
|
||||
.into_iter()
|
||||
.filter(|message| {
|
||||
matches!(
|
||||
message.as_str(),
|
||||
"startup hook context" | "fork hook context"
|
||||
)
|
||||
})
|
||||
.collect::<Vec<_>>();
|
||||
assert_eq!(
|
||||
hook_contexts,
|
||||
vec!["startup hook context", "fork hook context"]
|
||||
);
|
||||
|
||||
forked.shutdown_and_wait().await?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn explicit_history_runs_resume_session_start_hook() -> Result<()> {
|
||||
skip_if_no_network!(Ok(()));
|
||||
|
||||
let server = start_mock_server().await;
|
||||
let response = mount_sse_once(&server, sse(vec![ev_completed("resp-resume")])).await;
|
||||
let mut builder = test_codex()
|
||||
.with_pre_build_hook(|home| {
|
||||
write_resume_and_compact_session_start_hook_with_context(
|
||||
home,
|
||||
"resume hook context",
|
||||
"compact hook context",
|
||||
)
|
||||
.expect("write resume session start hook");
|
||||
})
|
||||
.with_config(trust_discovered_hooks);
|
||||
// Command hooks run on the host and require a host-native working directory.
|
||||
let test = builder.build(&server).await?;
|
||||
let history = InitialHistory::Forked(vec![RolloutItem::ResponseItem(
|
||||
responses::user_message_item("supplied history").into(),
|
||||
)]);
|
||||
let resumed = test
|
||||
.thread_manager
|
||||
.resume_thread_with_history(
|
||||
test.config.clone(),
|
||||
history,
|
||||
test.thread_manager.auth_manager(),
|
||||
/*parent_trace*/ None,
|
||||
ClientMcpExtensions::default(),
|
||||
)
|
||||
.await?
|
||||
.thread;
|
||||
resumed
|
||||
.start_or_steer_turn(TurnInputRequest::user_input(vec![UserInput::Text {
|
||||
text: "continue".to_string(),
|
||||
text_elements: Vec::new(),
|
||||
}]))
|
||||
.await?;
|
||||
wait_for_event(&resumed, |event| matches!(event, EventMsg::TurnComplete(_))).await;
|
||||
|
||||
let hook_inputs = read_session_start_hook_inputs(test.codex_home_path())?;
|
||||
assert_eq!(
|
||||
hook_inputs
|
||||
.iter()
|
||||
.filter_map(|input| input.get("source").and_then(Value::as_str))
|
||||
.collect::<Vec<_>>(),
|
||||
vec!["resume"],
|
||||
);
|
||||
assert!(
|
||||
response
|
||||
.single_request()
|
||||
.message_input_texts("developer")
|
||||
.iter()
|
||||
.any(|message| message == "resume hook context"),
|
||||
);
|
||||
|
||||
resumed.shutdown_and_wait().await?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn async_hook_context_is_injected_into_the_active_turn() -> Result<()> {
|
||||
skip_if_no_network!(Ok(()));
|
||||
|
||||
@@ -279,7 +279,7 @@ print(json.dumps({{"systemMessage": "root stop complete"}}))
|
||||
}]
|
||||
}],
|
||||
"SubagentStart": [{
|
||||
"matcher": "worker",
|
||||
"matcher": "worker|default",
|
||||
"hooks": [{
|
||||
"type": "command",
|
||||
"command": format!("python3 {}", start_script_path.display()),
|
||||
@@ -607,20 +607,29 @@ async fn spawned_agent_uses_multi_agent_reasoning_effort_for_requests(
|
||||
Ok(())
|
||||
}
|
||||
|
||||
#[test_case(false; "fresh context")]
|
||||
#[test_case(true; "forked context")]
|
||||
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
|
||||
async fn subagent_start_replaces_session_start_and_injects_context() -> Result<()> {
|
||||
async fn subagent_start_replaces_session_start_and_injects_context(
|
||||
fork_context: bool,
|
||||
) -> Result<()> {
|
||||
skip_if_no_network!(Ok(()));
|
||||
|
||||
let server = start_mock_server().await;
|
||||
let agent_type = (!fork_context).then_some("worker");
|
||||
let expected_agent_type = agent_type.unwrap_or("default");
|
||||
let spawn_args = serde_json::to_string(&json!({
|
||||
"message": CHILD_PROMPT,
|
||||
"task_name": "child",
|
||||
"agent_type": "worker",
|
||||
"agent_type": agent_type,
|
||||
"fork_context": fork_context,
|
||||
}))?;
|
||||
|
||||
mount_sse_once_match(
|
||||
&server,
|
||||
|req: &wiremock::Request| body_contains(req, TURN_1_PROMPT),
|
||||
|req: &wiremock::Request| {
|
||||
body_contains(req, TURN_1_PROMPT) && !body_contains(req, SPAWN_CALL_ID)
|
||||
},
|
||||
sse(vec![
|
||||
ev_response_created("resp-turn1-1"),
|
||||
ev_function_call_with_namespace(
|
||||
@@ -640,7 +649,6 @@ async fn subagent_start_replaces_session_start_and_injects_context() -> Result<(
|
||||
body_contains(req, CHILD_PROMPT)
|
||||
&& body_contains(req, SUBAGENT_START_CONTEXT)
|
||||
&& !body_contains(req, "<subagent_notification>")
|
||||
&& !body_contains(req, SPAWN_CALL_ID)
|
||||
},
|
||||
sse(vec![
|
||||
ev_response_created("resp-child-1"),
|
||||
@@ -652,7 +660,9 @@ async fn subagent_start_replaces_session_start_and_injects_context() -> Result<(
|
||||
|
||||
let _turn1_followup = mount_sse_once_match(
|
||||
&server,
|
||||
|req: &wiremock::Request| body_contains(req, SPAWN_CALL_ID),
|
||||
|req: &wiremock::Request| {
|
||||
body_contains(req, SPAWN_CALL_ID) && !body_contains(req, SUBAGENT_START_CONTEXT)
|
||||
},
|
||||
sse(vec![
|
||||
ev_response_created("resp-turn1-2"),
|
||||
ev_assistant_message("msg-turn1-2", "parent done"),
|
||||
@@ -673,6 +683,7 @@ async fn subagent_start_replaces_session_start_and_injects_context() -> Result<(
|
||||
.enable(Feature::Collab)
|
||||
.expect("test config should allow feature update");
|
||||
})
|
||||
// Command hooks run on the host and require a host-native working directory.
|
||||
.build(&server)
|
||||
.await?;
|
||||
|
||||
@@ -686,7 +697,10 @@ async fn subagent_start_replaces_session_start_and_injects_context() -> Result<(
|
||||
)
|
||||
.await?;
|
||||
assert_eq!(start_inputs.len(), 1);
|
||||
assert_eq!(start_inputs[0]["agent_type"].as_str(), Some("worker"));
|
||||
assert_eq!(
|
||||
start_inputs[0]["agent_type"].as_str(),
|
||||
Some(expected_agent_type)
|
||||
);
|
||||
let spawned_id = wait_for_spawned_thread_id(&test).await?;
|
||||
assert_eq!(
|
||||
start_inputs[0]["agent_id"].as_str(),
|
||||
@@ -714,7 +728,10 @@ async fn subagent_start_replaces_session_start_and_injects_context() -> Result<(
|
||||
child_prompt_input["agent_id"].as_str(),
|
||||
Some(spawned_id.as_str())
|
||||
);
|
||||
assert_eq!(child_prompt_input["agent_type"].as_str(), Some("worker"));
|
||||
assert_eq!(
|
||||
child_prompt_input["agent_type"].as_str(),
|
||||
Some(expected_agent_type)
|
||||
);
|
||||
|
||||
let session_start_inputs = wait_for_hook_log(
|
||||
test.codex_home_path(),
|
||||
|
||||
Reference in new issue
Block a user