From 2c2bfbb4c7a60bbe80af11737eed91c1475b6ceb Mon Sep 17 00:00:00 2001
From: Aditya Kumar <adityaccds@gmail.com>
Date: Fri, 23 Oct 2026 23:38:28 +0530
Subject: [PATCH] Python: Report repeated native exit codes in persistent
 PowerShell sessions

A persistent PowerShell session reported exit code 0 when a native command
failed with the same code as the previous one. The exit-code check compared
$LASTEXITCODE by value, which does not change when the code repeats, and $?
reflects Invoke-Expression rather than the native command. Compare by
reference instead: PowerShell stores a newly boxed int on every native exit,
so the variable holds a different object exactly when the command ran a
native process. $LASTEXITCODE is still never written.
---
 .../agent_framework_tools/shell/_session.py   |  9 ++++--
 .../tools/tests/test_local_shell_tool.py      | 30 +++++++++++++++++++
 2 files changed, 37 insertions(+), 2 deletions(-)

diff --git a/python/packages/tools/agent_framework_tools/shell/_session.py b/python/packages/tools/agent_framework_tools/shell/_session.py
index f83f1c8e98e..8346aa4b171 100644
--- a/python/packages/tools/agent_framework_tools/shell/_session.py
+++ b/python/packages/tools/agent_framework_tools/shell/_session.py
@@ -379,7 +379,7 @@ def _build_script(self, command: str, sentinel: str) -> str:
                 # earlier command is still there when a cmdlet-only command
                 # runs. Snapshot it instead of clearing it: the variable stays
                 # readable by the user's own commands, which persistent mode is
-                # meant to preserve, and a value that differs afterwards can
+                # meant to preserve, and a different object afterwards can
                 # only have been written by this command.
                 " $__af_last = $LASTEXITCODE;"
                 " try {"
@@ -393,7 +393,12 @@ def _build_script(self, command: str, sentinel: str) -> str:
                 # $? has to be read on the very next statement: anything else
                 # in between overwrites it.
                 "   $__af_ok = $?;"
-                "   if ($LASTEXITCODE -ne $__af_last) { $__af_rc = $LASTEXITCODE }"
+                # Compare by reference, not by value: every native exit stores
+                # a newly boxed int, so a command that fails with the same code
+                # as the previous one still counts as a change. A value
+                # comparison would miss it, and $? cannot catch it either,
+                # because it reflects Invoke-Expression, which succeeded.
+                "   if (-not [object]::ReferenceEquals($LASTEXITCODE, $__af_last)) { $__af_rc = $LASTEXITCODE }"
                 "   elseif (-not $__af_ok) { $__af_rc = 1 }"
                 " } catch {"
                 "   [Console]::Error.WriteLine($_.ToString());"
diff --git a/python/packages/tools/tests/test_local_shell_tool.py b/python/packages/tools/tests/test_local_shell_tool.py
index 24ecbaaeb3b..5c591069f88 100644
--- a/python/packages/tools/tests/test_local_shell_tool.py
+++ b/python/packages/tools/tests/test_local_shell_tool.py
@@ -355,6 +355,36 @@ async def test_persistent_powershell_does_not_inherit_previous_exit_code() -> No
         )
 
 
+@pytest.mark.skipif(_POWERSHELL is None, reason="PowerShell is not installed")
+async def test_persistent_powershell_reports_repeated_native_exit_code(tmp_path: os.PathLike[str]) -> None:
+    """A native command failing with the same code as the previous one must still report it.
+
+    ``$LASTEXITCODE`` keeps the same value in that case, and ``$?`` reflects
+    ``Invoke-Expression`` rather than the native command, so the second
+    failure used to be reported as ``exit_code=0``.
+    Runs wherever PowerShell is installed, not just on Windows.
+    """
+    assert _POWERSHELL is not None
+    failing = f"& '{sys.executable}' -c 'import sys; sys.exit(3)'"
+    async with LocalShellTool(
+        mode="persistent",
+        shell=[_POWERSHELL, "-NoLogo", "-NoProfile", "-NonInteractive", "-Command", "-"],
+        approval_mode="never_require",
+        acknowledge_unsafe=True,
+        workdir=str(tmp_path),
+    ) as tool:
+        for attempt in range(1, 4):
+            result = await tool.run(failing)
+            assert result.exit_code == 3, f"attempt {attempt} reported rc {result.exit_code}"
+
+        succeeding = await tool.run("Write-Output ok")
+        assert succeeding.exit_code == 0
+
+        readback = await tool.run("Write-Output $LASTEXITCODE")
+        assert readback.exit_code == 0
+        assert readback.stdout.strip() == "3"
+
+
 @pytest.mark.skipif(sys.platform != "win32", reason="PowerShell-specific encoding")
 async def test_persistent_powershell_utf8_roundtrip() -> None:
     """Non-ASCII output should round-trip without mojibake."""