Fix: High idle CPU in RDS/Terminal Server environments (SCARD_E_INVALID_HANDLE) by DennisDyallo · Pull Request #445 · Yubico/Yubico.NET.SDK

DennisDyallo · 2026-04-01T16:19:54Z

Summary

Closes #434.

When an RDS/Windows 365/Terminal Server session is disconnected and reconnected, the Windows Smart Card Service invalidates existing SCARDCONTEXT handles. DesktopSmartCardDeviceListener was calling SCardGetStatusChange in a tight loop with the stale handle. WinSCard.dll internally raises and unwinds a C++ exception (CxxThrowException) for every such call — confirmed in 6 of 10 minidumps. This ran thousands of times per second, pegging one CPU core.

The error code returned to managed code (SCARD_E_INVALID_HANDLE = 0x80100003) was not handled by UpdateContextIfNonCritical. The code logged the error and immediately retried with no backoff and no context re-establishment.

Changes

ISCardInterop interface + SCardInterop concrete class — extracts the four SCard P/Invoke calls behind an injectable interface, enabling cross-platform unit tests without hardware or Windows
UpdateContextIfNonCritical — adds SCARD_E_INVALID_HANDLE, SCARD_E_SYSTEM_CANCELLED, ERROR_BROKEN_PIPE to the recovery switch (all three surface in RDS session transitions)
Thread.Sleep(1000) backoff — applied after every recovery attempt to prevent a secondary tight loop if SCardEstablishContext also fails while the service is transitioning
UpdateCurrentContext guard — checks SCardEstablishContext return value; on failure keeps existing context rather than replacing with a failed handle; explicitly disposes old handle
Default path backoff — unrecognised error codes also sleep 1000 ms (catch-all for unknown persistent errors)

See fix-rds-scard-invalid-handle.md for full root cause analysis, fix details, and test instructions.

Test plan

Cross-platform mock tests (run on any CI platform, no hardware needed)

dotnet test Yubico.Core\tests\Yubico.Core.UnitTests.csproj \
    --filter "FullyQualifiedName~DesktopSmartCardDeviceListenerSCardErrorTests" \
    --logger "console;verbosity=detailed"

WhenGetStatusChangeReturnsInvalidHandle_ContextIsReestablished — fails before fix, passes after
WhenGetStatusChangeAlwaysReturnsInvalidHandle_LoopDoesNotSpin — proves no tight loop (< 15 calls in 600ms)
WhenGetStatusChangeReturnsSystemCancelled_ContextIsReestablished
WhenContextReestablishmentFails_ListenerContinuesWithoutCrashing

Windows CPU regression test (requires Windows with SCardSvr running — no reader needed)

dotnet test Yubico.Core\tests\Yubico.Core.UnitTests.csproj \
    --filter "Category=WindowsOnly" \
    --logger "console;verbosity=detailed"

RealWinSCard_WhenHandleInvalidated_CpuDoesNotSpike uses reflection to invalidate the live SCARDCONTEXT handle (same as RDS disconnect), then measures Process.TotalProcessorTime over 3 seconds:

Before fix: ≈ 2500–3000 ms CPU consumed → FAIL
After fix: ≈ 30–100 ms CPU consumed → PASS

This test exercises the real WinSCard.dll C++ exception path reported by the OP.

🤖 Generated with Claude Code

…ard listener (#434) When an RDS/Terminal Server session is disconnected, the Windows Smart Card Service invalidates existing SCARDCONTEXT handles. DesktopSmartCardDeviceListener called SCardGetStatusChange in a tight loop with the stale handle — WinSCard internally raises a C++ exception (CxxThrowException) for each call, pegging a CPU core. Fix: - Add SCARD_E_INVALID_HANDLE, SCARD_E_SYSTEM_CANCELLED, ERROR_BROKEN_PIPE to the UpdateContextIfNonCritical recovery switch - Add Thread.Sleep(1000) backoff after recovery to prevent secondary tight loop - Guard UpdateCurrentContext against failed SCardEstablishContext - Extract ISCardInterop interface enabling cross-platform unit tests without hardware Tests: - DesktopSmartCardDeviceListenerSCardErrorTests: cross-platform mock tests (Track B) - DesktopSmartCardDeviceListenerWindowsTests: Windows CPU regression test (Track A) measuring TotalProcessorTime before/after handle invalidation via real WinSCard Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Copilot

Pull request overview

Fixes a high-idle-CPU regression in DesktopSmartCardDeviceListener seen in RDS/Terminal Server reconnect scenarios by handling invalidated smart-card contexts, adding recovery backoff, and improving testability via an injectable SCard interop abstraction.

Changes:

Extracts the WinSCard/PCSC calls behind ISCardInterop with a production SCardInterop implementation to enable deterministic unit testing.
Extends non-critical error recovery to include SCARD_E_INVALID_HANDLE, SCARD_E_SYSTEM_CANCELLED, and ERROR_BROKEN_PIPE, and adds a 1s recovery backoff to prevent tight retry loops.
Adds cross-platform mock tests and Windows-only integration/regression tests (including CPU consumption measurement) covering the RDS invalid-handle scenario.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
`Yubico.Core/src/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListener.cs`	Adds injectable SCard interop, expands recoverable error handling, and introduces recovery backoff + safer context refresh logic.
`Yubico.Core/src/Yubico/PlatformInterop/Desktop/SCard/ISCardInterop.cs`	Introduces an internal abstraction for the subset of SCard APIs the listener uses.
`Yubico.Core/src/Yubico/PlatformInterop/Desktop/SCard/SCardInterop.cs`	Provides the production `ISCardInterop` implementation delegating to `NativeMethods`.
`Yubico.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerSCardErrorTests.cs`	Adds cross-platform mock-based tests to validate recovery and throttling behavior.
`Yubico.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerWindowsTests.cs`	Adds Windows-only tests that invalidate a real context and validate CPU + recovery + disposal behavior.
`fix-rds-scard-invalid-handle.md`	Documents root cause analysis, fix details, and how to run/validate tests.

Copilot · 2026-04-01T16:26:54Z

Yubico.Core/src/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListener.cs

+        /// Internal constructor that accepts a test double for the SCard API surface.
+        /// </summary>
+        internal DesktopSmartCardDeviceListener(ISCardInterop scard)
+        {


The injected scard dependency is assigned without a null check. If a caller accidentally passes null (including future internal callers/tests), this will produce a NullReferenceException later in the listener thread and be harder to diagnose. Consider validating scard up-front (e.g., throw ArgumentNullException) before assigning to _scard.

Suggested change

{

{

if (scard is null)

{

throw new ArgumentNullException(nameof(scard));

}

Copilot · 2026-04-01T16:26:54Z

Yubico.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerWindowsTests.cs

+            SCardContext context = GetListenerContext(listener);
+            // SCardReleaseContext with the raw IntPtr tells WinSCard the handle is gone.
+            // Subsequent SCardGetStatusChange calls using this handle will fail immediately
+            // with SCARD_E_INVALID_HANDLE and trigger WinSCard's internal C++ exception path.
+            NativeMethods.SCardReleaseContext(context.DangerousGetHandle());
+        }


InvalidateListenerContext ignores the return value from SCardReleaseContext. If the call fails (e.g., context already replaced/disposed, service transition), the tests may proceed without actually invalidating the handle and could give false confidence. Consider checking the returned error code and either Assert success or Skip with a message when invalidation cannot be performed.

Copilot · 2026-04-01T16:26:54Z

Yubico.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerWindowsTests.cs

+
+            Skip.IfNot(ListenerIsActive(listener),
+                "Smart Card service (SCardSvr) is not running on this machine.");
+
+            Thread.Sleep(300);
+            InvalidateListenerContext(listener);
+
+            // Let recovery fire once (1000ms sleep inside the listener thread).
+            Thread.Sleep(1500);
+
+            // Now dispose — must complete well within 8 seconds.
+            var stopwatch = Stopwatch.StartNew();
+            var exception = Record.Exception(() => listener.Dispose());
+            stopwatch.Stop();
+
+            Assert.Null(exception);
+            Assert.True(
+                stopwatch.ElapsedMilliseconds < 5000,
+                $"Dispose took {stopwatch.ElapsedMilliseconds}ms after handle invalidation. " +
+                "Expected < 5000ms. The listener thread may be blocked in the recovery sleep.");


In RealWinSCard_WhenHandleInvalidatedThenDisposed_DisposalCompletesCleanly, the listener is not in a using (or try/finally). If Skip.IfNot(ListenerIsActive(listener), ...) triggers, the test exits via exception and the listener is never disposed, leaking the background thread/context for the rest of the test run. Consider creating the listener with using var and still measuring a manual Dispose() call (double-dispose should be safe), or wrapping the listener in a try/finally that disposes it on all paths.

Suggested change

Skip.IfNot(ListenerIsActive(listener),

"Smart Card service (SCardSvr) is not running on this machine.");

Thread.Sleep(300);

InvalidateListenerContext(listener);

// Let recovery fire once (1000ms sleep inside the listener thread).

Thread.Sleep(1500);

// Now dispose — must complete well within 8 seconds.

var stopwatch = Stopwatch.StartNew();

var exception = Record.Exception(() => listener.Dispose());

stopwatch.Stop();

Assert.Null(exception);

Assert.True(

stopwatch.ElapsedMilliseconds < 5000,

$"Dispose took {stopwatch.ElapsedMilliseconds}ms after handle invalidation. " +

"Expected < 5000ms. The listener thread may be blocked in the recovery sleep.");

try

{

Skip.IfNot(ListenerIsActive(listener),

"Smart Card service (SCardSvr) is not running on this machine.");

Thread.Sleep(300);

InvalidateListenerContext(listener);

// Let recovery fire once (1000ms sleep inside the listener thread).

Thread.Sleep(1500);

// Now dispose — must complete well within 8 seconds.

var stopwatch = Stopwatch.StartNew();

var exception = Record.Exception(() => listener.Dispose());

stopwatch.Stop();

Assert.Null(exception);

Assert.True(

stopwatch.ElapsedMilliseconds < 5000,

$"Dispose took {stopwatch.ElapsedMilliseconds}ms after handle invalidation. " +

"Expected < 5000ms. The listener thread may be blocked in the recovery sleep.");

}

finally

{

listener.Dispose();

}

…try/finally leak fix - Add ArgumentNullException guard for injected ISCardInterop in internal constructor - Check SCardReleaseContext return value in test helper and Skip on failure - Wrap disposal test listener in try/finally to prevent resource leak on Skip Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

… CPM override - Replace Assert.NotEqual(..., string) with Assert.True(..., string) — xunit 2.9.3 has no string-message overload for NotEqual (that's xunit 3.x only) - Add Directory.Packages.props to .gitignore (local-only workaround that prevents a parent project's CPM config from breaking restore in this branch) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

github-actions · 2026-04-01T16:51:54Z

Test Results: Windows

2 files 2 suites 29s ⏱️
4 060 tests 4 038 ✅ 22 💤 0 ❌
4 062 runs 4 040 ✅ 22 💤 0 ❌

Results for commit a9964fe.

♻️ This comment has been updated with latest results.

github-actions · 2026-04-01T16:52:03Z

Test Results: Ubuntu

2 files 2 suites 55s ⏱️
4 052 tests 4 030 ✅ 22 💤 0 ❌
4 054 runs 4 032 ✅ 22 💤 0 ❌

Results for commit a9964fe.

♻️ This comment has been updated with latest results.

github-actions · 2026-04-01T16:52:12Z

Test Results: MacOS

4 files 4 suites 58s ⏱️
4 034 tests 4 031 ✅ 3 💤 0 ❌
4 036 runs 4 033 ✅ 3 💤 0 ❌

Results for commit a9964fe.

♻️ This comment has been updated with latest results.

Copilot

Pull request overview

Copilot reviewed 6 out of 7 changed files in this pull request and generated 4 comments.

Copilot · 2026-04-01T20:26:27Z

Yubico.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerWindowsTests.cs

+using Yubico.PlatformInterop;
+
+namespace Yubico.Core.Devices.SmartCard.UnitTests
+{


[Collection("WindowsOnlyTests")] groups these tests, but without a corresponding CollectionDefinition (e.g. DisableParallelization = true) or assembly-level parallelization settings, other test collections can still run concurrently and skew this file's CPU-sensitive Process.TotalProcessorTime assertions. Consider adding a CollectionDefinition("WindowsOnlyTests", DisableParallelization = true) in the test project or otherwise ensuring these tests run non-parallel.

Suggested change

{

{

[CollectionDefinition("WindowsOnlyTests", DisableParallelization = true)]

public class WindowsOnlyTestsCollection

{

}

Copilot · 2026-04-01T20:26:27Z

Yubico.Core/src/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListener.cs

            // Non-critical errors that need context update
            if (UpdateContextIfNonCritical(result))
            {
                _log.LogInformation("GetStatusChange indicated non-critical status {Status:X}.", result);
                return true;


When UpdateContextIfNonCritical(result) triggers UpdateCurrentContext(), _readerStates is refreshed, but the current CheckForUpdates call continues using the pre-recovery newStates clone and later assigns _readerStates = newStates. This can overwrite the refreshed reader list and keep polling with stale state after recovery. Consider short-circuiting the current update iteration when recovery occurs (restart with a fresh clone of _readerStates).

Copilot · 2026-04-01T20:26:27Z

Yubico.Core/src/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListener.cs

+            // Sleep briefly to prevent a tight loop if this error persists (e.g. unknown
+            // persistent error codes not yet classified as recoverable).
+            _log.SCardApiCall(nameof(NativeMethods.SCardGetStatusChange), result);
            _log.LogInformation("Reader states:\n{States}", states);


_log.LogInformation("Reader states:\n{States}", states) will log the array type name rather than per-reader details. Since SCARD_READER_STATE has an overridden ToString(), consider formatting the array into a joined string (or logging entries individually) so this diagnostic output is actionable when investigating SCard errors.

Suggested change

_log.LogInformation("Reader states:\n{States}", states);

string formattedStates = string.Join(Environment.NewLine, states.Select(s => s.ToString()));

_log.LogInformation("Reader states:\n{States}", formattedStates);

Copilot · 2026-04-01T20:26:28Z

...co.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerSCardErrorTests.cs

+                context = new SCardContext(IntPtr.Zero);
+
+                if (_establishContextFailAfterFirstCall && callNum > 1)
+                {
+                    return ErrorCode.SCARD_E_NO_SERVICE;
+                }
+


FakeSCardInterop.EstablishContext returns SCARD_S_SUCCESS but sets context to new SCardContext(IntPtr.Zero) (an invalid handle). This diverges from real WinSCard behavior and can mask bugs where production code assumes a successful establish yields a non-invalid context. Consider returning a distinct non-zero handle when returning success (and only using IntPtr.Zero on failure).

Suggested change

context = new SCardContext(IntPtr.Zero);

if (_establishContextFailAfterFirstCall && callNum > 1)

{

return ErrorCode.SCARD_E_NO_SERVICE;

}

if (_establishContextFailAfterFirstCall && callNum > 1)

{

context = new SCardContext(IntPtr.Zero);

return ErrorCode.SCARD_E_NO_SERVICE;

}

// On success, simulate a valid, non-zero context handle (as WinSCard would).

context = new SCardContext(new IntPtr(callNum));

- Short-circuit CheckForUpdates after context recovery at all three GetStatusChange call sites to prevent stale newStates from overwriting the freshly refreshed _readerStates (Copilot review finding). - Format SCARD_READER_STATE[] in diagnostic logging so individual reader entries are printed instead of the array type name. - Return distinct non-zero handles from FakeSCardInterop.EstablishContext on success, matching real WinSCard behavior. - Extract DRY helper for context-reestablishment test assertions. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

github-actions · 2026-04-02T15:26:32Z

Package	Line Rate	Branch Rate	Complexity	Health
Yubico.Core	52%	42%	1551	➖
Yubico.YubiKey	50%	46%	7180	➖
Summary	50% (12840 / 25478)	45% (3085 / 6882)	8731	➖

Minimum allowed line rate is 40%

Copilot

Pull request overview

Copilot reviewed 6 out of 7 changed files in this pull request and generated 1 comment.

Copilot · 2026-04-02T15:36:14Z

...co.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerSCardErrorTests.cs

+                // Return a distinct non-zero handle on success, matching real WinSCard behavior.
+                context = new SCardContext(new IntPtr(callNum));


FakeSCardInterop.EstablishContext returns new SCardContext(new IntPtr(callNum)) for “successful” calls. SCardContext is a SafeHandle whose ReleaseHandle() P/Invokes NativeMethods.SCardReleaseContext, so disposing the listener in these mock tests will still invoke native smart-card APIs with a fake handle. That weakens the isolation of these tests (they can fail due to native/PCSC availability or native error behavior rather than the recovery logic). Consider returning IntPtr.Zero (so release is effectively a no-op) or using a test-only SCardContext implementation that avoids native calls in ReleaseHandle().

Suggested change

// Return a distinct non-zero handle on success, matching real WinSCard behavior.

context = new SCardContext(new IntPtr(callNum));

// Use IntPtr.Zero so disposing the context in tests does not invoke native APIs.

context = new SCardContext(IntPtr.Zero);

Copilot

Pull request overview

Copilot reviewed 6 out of 7 changed files in this pull request and generated 2 comments.

Copilot · 2026-04-02T18:53:45Z

Yubico.Core/tests/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListenerWindowsTests.cs

+    [Collection("WindowsOnlyTests")]
+    public class DesktopSmartCardDeviceListenerWindowsTests


[Collection("WindowsOnlyTests")] does not prevent this class from running in parallel with other test collections, but the assertions measure whole-process CPU (Process.TotalProcessorTime) and the file header says the tests should run in isolation. This can make the CPU threshold assertions flaky when xUnit parallelization is enabled.

Consider adding a [CollectionDefinition("WindowsOnlyTests", DisableParallelization = true)] fixture in the test project (or otherwise disabling parallelization for these tests) so they won’t run concurrently with other collections.

Copilot · 2026-04-02T18:53:45Z

Yubico.Core/src/Yubico/Core/Devices/SmartCard/DesktopSmartCardDeviceListener.cs

@@ -31,6 +29,7 @@ internal class DesktopSmartCardDeviceListener : SmartCardDeviceListener
    {
        private static readonly string[] readerNames = new[] { "\\\\?\\Pnp\\Notifications" };


The special PC/SC PnP notification reader name literals look inconsistent and likely incorrect:

readerNames uses "\\\\?\\Pnp\\Notifications" (plural Notifications)

GetReaderStateList prepends "\\\\?PnP?\\Notification" (different escaping/placement)

WinSCard/PCSC uses the virtual reader \\?\\PnP\\Notification (singular). Consider replacing both with a single shared constant using the canonical value to avoid relying on fallback behavior and to make behavior consistent across platforms.

Suggested change

private static readonly string[] readerNames = new[] { "\\\\?\\Pnp\\Notifications" };

private const string PnpNotificationReaderName = @"\\?\PnP\Notification";

private static readonly string[] readerNames = new[] { PnpNotificationReaderName };

DennisDyallo requested a review from Copilot April 1, 2026 16:21

Copilot started reviewing on behalf of DennisDyallo April 1, 2026 16:21 View session

Copilot AI reviewed Apr 1, 2026

View reviewed changes

DennisDyallo and others added 2 commits April 1, 2026 18:29

DennisDyallo requested a review from Copilot April 1, 2026 20:16

Copilot started reviewing on behalf of DennisDyallo April 1, 2026 20:17 View session

Copilot AI reviewed Apr 1, 2026

View reviewed changes

DennisDyallo requested a review from Copilot April 2, 2026 15:29

Copilot started reviewing on behalf of DennisDyallo April 2, 2026 15:30 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

DennisDyallo requested a review from Copilot April 2, 2026 18:46

Copilot started reviewing on behalf of DennisDyallo April 2, 2026 18:47 View session

Copilot AI reviewed Apr 2, 2026

View reviewed changes

-        {
+        {
+            if (scard is null)
+            {
+                throw new ArgumentNullException(nameof(scard));
+            }

-{
+{
+    [CollectionDefinition("WindowsOnlyTests", DisableParallelization = true)]
+    public class WindowsOnlyTestsCollection
+    {
+    }

	_log.LogInformation("Reader states:\n{States}", states);
	string formattedStates = string.Join(Environment.NewLine, states.Select(s => s.ToString()));
	_log.LogInformation("Reader states:\n{States}", formattedStates);

		// Return a distinct non-zero handle on success, matching real WinSCard behavior.
		context = new SCardContext(new IntPtr(callNum));

		[Collection("WindowsOnlyTests")]
		public class DesktopSmartCardDeviceListenerWindowsTests

		@@ -31,6 +29,7 @@ internal class DesktopSmartCardDeviceListener : SmartCardDeviceListener
		{
		private static readonly string[] readerNames = new[] { "\\\\?\\Pnp\\Notifications" };

	private static readonly string[] readerNames = new[] { "\\\\?\\Pnp\\Notifications" };
	private const string PnpNotificationReaderName = @"\\?\PnP\Notification";
	private static readonly string[] readerNames = new[] { PnpNotificationReaderName };

Conversation

DennisDyallo commented Apr 1, 2026

Summary

Changes

Test plan

Cross-platform mock tests (run on any CI platform, no hardware needed)

Windows CPU regression test (requires Windows with SCardSvr running — no reader needed)

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results: Windows

Uh oh!

github-actions bot commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results: Ubuntu

Uh oh!

github-actions bot commented Apr 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results: MacOS

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 1, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions bot commented Apr 2, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 2, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Milestone

Development

Uh oh!

2 participants

github-actions bot commented Apr 1, 2026 •

edited

Loading

github-actions bot commented Apr 1, 2026 •

edited

Loading

github-actions bot commented Apr 1, 2026 •

edited

Loading