Avoid shell kill cwd cleanup flakes (#1207)

* Avoid shell kill cwd cleanup flake

Run the long-lived shell kill command outside the fixture workspace so a surviving Windows child process cannot hold copilot-test-work-* open during cleanup.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Avoid shell kill cwd flakes across SDKs

Run mirrored long-lived shell kill E2E commands outside each fixture workspace so surviving Windows child processes cannot block temp workspace cleanup.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Wait for CLI stderr on stdio disconnect

When stdio startup fails quickly, wait briefly for the stderr reader before formatting the connection-lost exception so CLI argument errors are reported deterministically.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Include CLI stderr for stdio startup pipe failures

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* Revert "Include CLI stderr for stdio startup pipe failures"

This reverts commit b35e9f6.

* Revert "Wait for CLI stderr on stdio disconnect"

This reverts commit 344b18f.

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Author: stephentoub

dotnet/test/E2E/RpcShellAndFleetE2ETests.cs

@@ -34,7 +34,9 @@ public async Task Should_Kill_Shell_Process()
             ? "powershell -NoLogo -NoProfile -Command \"Start-Sleep -Seconds 30\""
             : "sleep 30";
 
-        var execResult = await session.Rpc.Shell.ExecAsync(command);
+        // On Windows, terminating the shell wrapper can briefly leave grandchildren alive.
+        // Keep this command outside the fixture workspace so that cleanup is not blocked by cwd handles.
+        var execResult = await session.Rpc.Shell.ExecAsync(command, cwd: Path.GetTempPath());
         Assert.False(string.IsNullOrWhiteSpace(execResult.ProcessId));
 
         var killResult = await session.Rpc.Shell.KillAsync(execResult.ProcessId);

go/internal/e2e/rpc_shell_and_fleet_e2e_test.go

@@ -63,7 +63,10 @@ func TestRpcShellAndFleetE2E(t *testing.T) {
 			command = "sleep 30"
 		}
 
-		exec, err := session.RPC.Shell.Exec(t.Context(), &rpc.ShellExecRequest{Command: command})
+		// On Windows, terminating the shell wrapper can briefly leave grandchildren alive.
+		// Keep this command outside the fixture workspace so cleanup is not blocked by cwd handles.
+		cwd := os.TempDir()
+		exec, err := session.RPC.Shell.Exec(t.Context(), &rpc.ShellExecRequest{Command: command, Cwd: &cwd})
 		if err != nil {
 			t.Fatalf("Failed to call session.shell.exec: %v", err)
 		}

nodejs/test/e2e/rpc_shell_and_fleet.e2e.test.ts

@@ -85,7 +85,9 @@ describe("Shell and fleet RPC", async () => {
                 ? `powershell -NoLogo -NoProfile -Command "Start-Sleep -Seconds 30"`
                 : "sleep 30";
 
-        const execResult = await session.rpc.shell.exec({ command });
+        // On Windows, terminating the shell wrapper can briefly leave grandchildren alive.
+        // Keep this command outside the fixture workspace so cleanup is not blocked by cwd handles.
+        const execResult = await session.rpc.shell.exec({ command, cwd: os.tmpdir() });
         expect(execResult.processId).toBeTruthy();
 
         const killResult = await session.rpc.shell.kill({ processId: execResult.processId });

python/e2e/test_rpc_shell_and_fleet_e2e.py

@@ -9,6 +9,7 @@
 
 import asyncio
 import sys
+import tempfile
 import uuid
 from pathlib import Path
 
@@ -75,7 +76,11 @@ async def test_should_kill_shell_process(self, ctx: E2ETestContext):
         else:
             command = "sleep 30"
 
-        exec_result = await session.rpc.shell.exec(ShellExecRequest(command=command))
+        # On Windows, terminating the shell wrapper can briefly leave grandchildren alive.
+        # Keep this command outside the fixture workspace so cleanup is not blocked by cwd handles.
+        exec_result = await session.rpc.shell.exec(
+            ShellExecRequest(command=command, cwd=tempfile.gettempdir())
+        )
         assert (exec_result.process_id or "").strip()
 
         kill_result = await session.rpc.shell.kill(