This commit is contained in:
Aaron Levine
2026-06-29 20:07:13 -07:00
parent ccdfb4f342
commit 317ebf64ce
6 changed files with 177 additions and 2 deletions

View File

@@ -769,6 +769,14 @@ pub(crate) async fn apply_bespoke_event_handling(
server_name: request.server_name.clone(),
request: request_body,
};
let pause_incremented =
match conversation.increment_out_of_band_elicitation_count().await {
Ok(_) => true,
Err(err) => {
error!("failed to pause timeout accounting for MCP elicitation: {err}");
false
}
};
let (pending_request_id, rx) = outgoing
.send_request(ServerRequestPayload::McpServerElicitationRequest(params))
.await;
@@ -781,6 +789,7 @@ pub(crate) async fn apply_bespoke_event_handling(
conversation,
thread_state,
permission_guard,
pause_incremented,
)
.await;
});
@@ -1772,8 +1781,14 @@ async fn on_mcp_server_elicitation_response(
conversation: Arc<CodexThread>,
thread_state: Arc<Mutex<ThreadState>>,
permission_guard: ThreadWatchActiveGuard,
pause_incremented: bool,
) {
let response = receiver.await;
if pause_incremented
&& let Err(err) = conversation.decrement_out_of_band_elicitation_count().await
{
error!("failed to resume timeout accounting after MCP elicitation: {err}");
}
resolve_server_request_on_thread_listener(&thread_state, pending_request_id).await;
drop(permission_guard);
let response = mcp_server_elicitation_response_from_client_result(response);

View File

@@ -218,6 +218,10 @@ impl Session {
};
}
let _pending_elicitation = turn_context
.turn_metadata_state
.track_pending_mcp_elicitation();
let (tx_response, rx_response) = oneshot::channel();
let prev_entry = {
let mut active = self.active_turn.lock().await;

View File

@@ -1145,6 +1145,10 @@ impl Session {
self.out_of_band_elicitation_paused.subscribe()
}
pub(crate) fn is_out_of_band_elicitation_paused(&self) -> bool {
*self.out_of_band_elicitation_paused.borrow()
}
pub(crate) fn set_out_of_band_elicitation_pause_state(&self, paused: bool) {
self.out_of_band_elicitation_paused.send_replace(paused);
}

View File

@@ -184,7 +184,7 @@ pub(super) async fn handle_runtime_response(
max_output_tokens: Option<usize>,
started_at: std::time::Instant,
) -> Result<FunctionToolOutput, String> {
let script_status = format_script_status(&response);
let script_status = format_script_status(exec, &response);
match response {
RuntimeResponse::Yielded { content_items, .. } => {
@@ -228,8 +228,14 @@ fn sanitize_runtime_image_detail(turn: &TurnContext, items: &mut [FunctionCallOu
sanitize_image_detail_items(can_request_original_image_detail(&turn.model_info), items);
}
fn format_script_status(response: &RuntimeResponse) -> String {
fn format_script_status(exec: &ExecContext, response: &RuntimeResponse) -> String {
match response {
RuntimeResponse::Yielded { cell_id, .. }
if exec.session.is_out_of_band_elicitation_paused()
|| exec.turn.turn_metadata_state.has_pending_mcp_elicitation() =>
{
format!("Script running with cell ID {cell_id}\nWaiting for user input")
}
RuntimeResponse::Yielded { cell_id, .. } => {
format!("Script running with cell ID {cell_id}")
}

View File

@@ -4,6 +4,7 @@ use std::sync::Arc;
use std::sync::Mutex;
use std::sync::RwLock;
use std::sync::atomic::AtomicBool;
use std::sync::atomic::AtomicUsize;
use std::sync::atomic::Ordering;
use serde_json::Value;
@@ -99,9 +100,21 @@ pub(crate) struct TurnMetadataState {
turn_started_at_unix_ms: Arc<RwLock<Option<i64>>>,
responsesapi_client_metadata: Arc<RwLock<BTreeMap<String, String>>>,
user_input_requested_during_turn: Arc<AtomicBool>,
pending_mcp_elicitations: Arc<AtomicUsize>,
enrichment_task: Arc<Mutex<Option<JoinHandle<()>>>>,
}
pub(crate) struct PendingMcpElicitationGuard {
count: Arc<AtomicUsize>,
}
impl Drop for PendingMcpElicitationGuard {
fn drop(&mut self) {
let previous = self.count.fetch_sub(1, Ordering::AcqRel);
debug_assert!(previous > 0, "pending MCP elicitation count underflowed");
}
}
impl TurnMetadataState {
#[allow(clippy::too_many_arguments)]
pub(crate) fn new(
@@ -142,6 +155,7 @@ impl TurnMetadataState {
turn_started_at_unix_ms: Arc::new(RwLock::new(None)),
responsesapi_client_metadata: Arc::new(RwLock::new(BTreeMap::new())),
user_input_requested_during_turn: Arc::new(AtomicBool::new(false)),
pending_mcp_elicitations: Arc::new(AtomicUsize::new(0)),
enrichment_task: Arc::new(Mutex::new(None)),
}
}
@@ -203,6 +217,17 @@ impl TurnMetadataState {
.store(true, Ordering::Relaxed);
}
pub(crate) fn track_pending_mcp_elicitation(&self) -> PendingMcpElicitationGuard {
self.pending_mcp_elicitations.fetch_add(1, Ordering::AcqRel);
PendingMcpElicitationGuard {
count: Arc::clone(&self.pending_mcp_elicitations),
}
}
pub(crate) fn has_pending_mcp_elicitation(&self) -> bool {
self.pending_mcp_elicitations.load(Ordering::Acquire) > 0
}
pub(crate) fn set_responsesapi_client_metadata(
&self,
responsesapi_client_metadata: HashMap<String, String>,

View File

@@ -3,6 +3,7 @@
use anyhow::Result;
use base64::Engine;
use base64::engine::general_purpose::STANDARD as BASE64_STANDARD;
use codex_config::types::AppToolApproval;
use codex_config::types::McpServerConfig;
use codex_config::types::McpServerTransportConfig;
use codex_core::config::Config;
@@ -12,6 +13,7 @@ use codex_features::CurrentTimeSource;
use codex_features::Feature;
use codex_login::CodexAuth;
use codex_models_manager::bundled_models_response;
use codex_protocol::config_types::ApprovalsReviewer;
use codex_protocol::config_types::WebSearchMode;
use codex_protocol::dynamic_tools::DynamicToolCallOutputContentItem;
use codex_protocol::dynamic_tools::DynamicToolFunctionSpec;
@@ -479,6 +481,125 @@ async fn run_code_mode_turn_with_rmcp_config(
Ok((test, second_mock))
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn code_mode_reports_pending_mcp_elicitation_on_yield_and_wait() -> Result<()> {
skip_if_no_network!(Ok(()));
let server = responses::start_mock_server().await;
let rmcp_test_server_bin = stdio_server_bin()?;
let mut builder = test_codex().with_config(move |config| {
let _ = config.features.enable(Feature::CodeMode);
let _ = config.features.enable(Feature::ToolCallMcpElicitation);
config.approvals_reviewer = ApprovalsReviewer::User;
let mut servers = config.mcp_servers.get().clone();
servers.insert(
"rmcp".to_string(),
McpServerConfig {
auth: Default::default(),
transport: McpServerTransportConfig::Stdio {
command: rmcp_test_server_bin,
args: Vec::new(),
env: None,
env_vars: Vec::new(),
cwd: None,
},
environment_id: "local".to_string(),
enabled: true,
required: false,
supports_parallel_tool_calls: false,
disabled_reason: None,
startup_timeout_sec: Some(Duration::from_secs(10)),
tool_timeout_sec: None,
default_tools_approval_mode: Some(AppToolApproval::Prompt),
enabled_tools: None,
disabled_tools: None,
scopes: None,
oauth: None,
oauth_resource: None,
tools: HashMap::new(),
},
);
config
.mcp_servers
.set(servers)
.expect("test mcp servers should accept any configuration");
});
let test = builder.build(&server).await?;
wait_for_mcp_server(&test.codex, "rmcp").await?;
responses::mount_sse_once(
&server,
sse(vec![
ev_response_created("resp-1"),
ev_custom_tool_call(
"call-1",
"exec",
r#"// @exec: {"yield_time_ms": 1000}
await tools.mcp__rmcp__sync({});"#,
),
ev_completed("resp-1"),
]),
)
.await;
let initial_yield = responses::mount_sse_once(
&server,
sse(vec![
ev_response_created("resp-2"),
responses::ev_function_call(
"call-2",
"wait",
&serde_json::to_string(&serde_json::json!({
"cell_id": "1",
"yield_time_ms": 10,
}))?,
),
ev_completed("resp-2"),
]),
)
.await;
let later_wait = responses::mount_sse_once(
&server,
sse(vec![
ev_assistant_message("msg-1", "waiting"),
ev_completed("resp-3"),
]),
)
.await;
test.submit_turn_with_approval_and_permission_profile(
"request confirmation from code mode",
AskForApproval::OnRequest,
PermissionProfile::Disabled,
)
.await?;
let initial_items = custom_tool_output_items(&initial_yield.single_request(), "call-1");
assert_eq!(initial_items.len(), 1);
assert_regex_match(
concat!(
r"(?s)\A",
r"Script running with cell ID 1\n",
r"Waiting for user input\n",
r"Wall time \d+\.\d seconds\nOutput:\n\z"
),
text_item(&initial_items, /*index*/ 0),
);
let wait_items = function_tool_output_items(&later_wait.single_request(), "call-2");
assert_eq!(wait_items.len(), 1);
assert_regex_match(
concat!(
r"(?s)\A",
r"Script running with cell ID 1\n",
r"Waiting for user input\n",
r"Wall time \d+\.\d seconds\nOutput:\n\z"
),
text_item(&wait_items, /*index*/ 0),
);
Ok(())
}
#[cfg_attr(windows, ignore = "no exec_command on Windows")]
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn code_mode_can_return_exec_command_output() -> Result<()> {