Stabilize Windows thread unsubscribe sleep test

codex-rs/app-server/tests/common/lib.rs

@@ -37,6 +37,7 @@ pub use responses::create_final_assistant_message_sse_response;
 pub use responses::create_request_permissions_sse_response;
 pub use responses::create_request_user_input_sse_response;
 pub use responses::create_shell_command_sse_response;
+pub use responses::create_shell_command_sse_response_from_command;
 pub use rollout::create_fake_rollout;
 pub use rollout::create_fake_rollout_with_source;
 pub use rollout::create_fake_rollout_with_text_elements;

codex-rs/app-server/tests/common/responses.rs

@@ -10,11 +10,31 @@ pub fn create_shell_command_sse_response(
 ) -> anyhow::Result<String> {
     // The `arguments` for the `shell_command` tool is a serialized JSON object.
     let command_str = shlex::try_join(command.iter().map(String::as_str))?;
-    let tool_call_arguments = serde_json::to_string(&json!({
-        "command": command_str,
+    create_shell_command_sse_response_from_command(&command_str, workdir, timeout_ms, call_id, None)
+}
+
+pub fn create_shell_command_sse_response_from_command(
+    command: &str,
+    workdir: Option<&Path>,
+    timeout_ms: Option<u64>,
+    call_id: &str,
+    login: Option<bool>,
+) -> anyhow::Result<String> {
+    // Use this when a test already has a shell command string. It fixes string
+    // quoting for those callers by avoiding a rebuild from argv with POSIX
+    // rules, which can change how Windows PowerShell parses the command; for
+    // sleep-based tests, nested parsing can add another PowerShell startup
+    // before the requested sleep even begins.
+    let mut tool_call_arguments = json!({
+        "command": command,
         "workdir": workdir.map(|w| w.to_string_lossy()),
         "timeout_ms": timeout_ms
-    }))?;
+    });
+    if let Some(login) = login {
+        tool_call_arguments["login"] = json!(login);
+    }
+
+    let tool_call_arguments = serde_json::to_string(&tool_call_arguments)?;
     Ok(responses::sse(vec![
         responses::ev_response_created("resp-1"),
         responses::ev_function_call(call_id, "shell_command", &tool_call_arguments),

codex-rs/app-server/tests/suite/v2/thread_unsubscribe.rs

@@ -4,7 +4,7 @@ use app_test_support::McpProcess;
 use app_test_support::create_final_assistant_message_sse_response;
 use app_test_support::create_mock_responses_server_repeating_assistant;
 use app_test_support::create_mock_responses_server_sequence_unchecked;
-use app_test_support::create_shell_command_sse_response;
+use app_test_support::create_shell_command_sse_response_from_command;
 use app_test_support::to_response;
 use codex_app_server_protocol::ItemStartedNotification;
 use codex_app_server_protocol::JSONRPCResponse;
@@ -129,13 +129,9 @@ async fn thread_unsubscribe_keeps_thread_loaded_until_idle_timeout() -> Result<(
 #[tokio::test]
 async fn thread_unsubscribe_during_turn_keeps_turn_running() -> Result<()> {
     #[cfg(target_os = "windows")]
-    let shell_command = vec![
-        "powershell".to_string(),
-        "-Command".to_string(),
-        "Start-Sleep -Seconds 1".to_string(),
-    ];
+    let shell_command = "Start-Sleep -Seconds 1";
     #[cfg(not(target_os = "windows"))]
-    let shell_command = vec!["sleep".to_string(), "1".to_string()];
+    let shell_command = "sleep 1";
 
     let tmp = TempDir::new()?;
     let codex_home = tmp.path().join("codex_home");
@@ -144,11 +140,12 @@ async fn thread_unsubscribe_during_turn_keeps_turn_running() -> Result<()> {
     std::fs::create_dir(&working_directory)?;
 
     let server = create_mock_responses_server_sequence_unchecked(vec![
-        create_shell_command_sse_response(
-            shell_command.clone(),
+        create_shell_command_sse_response_from_command(
+            shell_command,
             Some(&working_directory),
             Some(10_000),
             "call_sleep",
+            Some(false),
         )?,
         create_final_assistant_message_sse_response("Done")?,
     ])