Python: [Bug]: Persistent PowerShell LocalShellTool reports exit_code 0 when a native command fails with the same code twice #9657

Observed Behavior

With mode="persistent" and PowerShell (the default shell on Windows), LocalShellTool reports exit_code=0 for a native command that failed, whenever it exits with the same code as the previous native command in the session. An agent that runs a failing test command, edits the code and reruns it with the same failure (exit 1 both times) is told that the second run succeeded. The same happens with repeated git errors (128) or any repeated exit N.

Since #8206, ShellSession._build_script (python/packages/tools/agent_framework_tools/shell/_session.py) works out the exit code like this:

$__af_last = $LASTEXITCODE;
...
Invoke-Expression $__af_cmd | Out-Default;
$__af_ok = $?;
if ($LASTEXITCODE -ne $__af_last) { $__af_rc = $LASTEXITCODE }
elseif (-not $__af_ok) { $__af_rc = 1 }

When the native command exits with the code that is already in $LASTEXITCODE, the value doesn't change, so the first branch doesn't run. The elseif doesn't run either, because $? reflects Invoke-Expression, which succeeded, rather than the native command inside it. That is the same on PowerShell 7.6.6 and Windows PowerShell 5.1:

cmd /c exit 3; $?                                  -> False
Invoke-Expression 'cmd /c exit 3'; $?              -> True
Invoke-Expression 'cmd /c exit 3' | Out-Default; $? -> True

So $__af_rc stays 0. The #8206 discussion mentioned this same-code case and expected the elseif to still report 1, but in practice it reports success.

Expected Behavior

Each failing run reports its own exit code: 1, 1, 1 and 128, 128 in the reproduction below. Commands that don't run a native process keep reporting 0, or 1 on a cmdlet error, and $LASTEXITCODE stays readable by the user's commands.

Steps to Reproduce

1. Create a LocalShellTool(mode="persistent") on Windows, where the default shell is PowerShell.
2. Run a native command that exits with a non-zero code, for example python -c "import sys; sys.exit(1)". It reports exit_code=1.
3. Run the same command again. It reports exit_code=0, although it failed with 1 again.

Minimal Reproduction

import asyncio
import sys

from agent_framework.tools import LocalShellTool

FAILING_PYTHON = f"& '{sys.executable}' -c 'import sys; sys.exit(1)'"
COMMANDS = [FAILING_PYTHON, FAILING_PYTHON, FAILING_PYTHON, "git rev-parse --git-dir", "git rev-parse --git-dir"]


async def main() -> None:
    # Default shell on Windows; run from a directory that is not a git repository.
    async with LocalShellTool(mode="persistent") as tool:
        for command in COMMANDS:
            result = await tool.run(command)
            label = "python -c 'import sys; sys.exit(1)'" if command == FAILING_PYTHON else command
            print(f"exit_code={result.exit_code:<4} <- {label}")


asyncio.run(main())

Output on main (Windows 11, PowerShell 7.6.6):

exit_code=1    <- python -c 'import sys; sys.exit(1)'
exit_code=0    <- python -c 'import sys; sys.exit(1)'
exit_code=0    <- python -c 'import sys; sys.exit(1)'
exit_code=128  <- git rev-parse --git-dir
exit_code=0    <- git rev-parse --git-dir

I get the same output with Windows PowerShell 5.1 (AGENT_FRAMEWORK_SHELL="powershell -NoLogo -NoProfile -NonInteractive -Command -").

Error Messages and Stack Traces

No response

Package Versions

agent-framework-tools: 1.0.0b261022, agent-framework-core: 1.23.0 (source checkout of main at 4982e4c)

Python Version

Python 3.13

Operating System

Windows

Regression

Yes

Additional Context

• Regression from Python: reset $LASTEXITCODE per command in persistent PowerShell sessions #8206 (the fix for .NET: Python: [Bug]: persistent PowerShell shell session reports a stale exit code for cmdlet-only commands #8205). Before it, the check was if ($LASTEXITCODE -ne $null) { $__af_rc = $LASTEXITCODE }, which reported a repeated failure correctly but also reported a stale code for cmdlet-only commands. Python: reset $LASTEXITCODE per command in persistent PowerShell sessions #8206 fixed the stale code and introduced this case. test_persistent_powershell_does_not_inherit_previous_exit_code runs one failing native command, so it doesn't cover running it twice.
• Suggested fix that still never writes $LASTEXITCODE (the constraint from the Python: reset $LASTEXITCODE per command in persistent PowerShell sessions #8206 review): check whether the variable now holds a different object rather than a different value. PowerShell stores a newly boxed int every time a native process exits, even when the code is the same, so [object]::ReferenceEquals($LASTEXITCODE, $__af_last) is False only if this command ran a native process or assigned the variable itself. I checked this on PowerShell 7.6.6 and Windows PowerShell 5.1 for exit codes 0, 1, 2, 3, 127, 255 and -1. Since this is a one-line change, I'm opening a PR with it and a regression test right after this issue.
• AI assistance: I used AI coding assistants (Claude Code and Antigravity) to help find this bug and to draft this report. I reviewed this report myself and checked the reproduction and its output above.

Acknowledgements

• I searched existing issues and did not find a duplicate.
• I personally verified this behavior and the reproduction details are authentic.
• I will wait for explicit maintainer agreement before starting implementation of a non-trivial change.