Reword pending-resume skip messages to match actual runtime contract

The previous wording ('Pending runtime fix') implied a runtime change was being tracked separately. That isn't accurate: runtime 1.0.56 (copilot-agent-runtime PR #9040) intentionally cleans up the session when the last RPC client disconnects, and there is no in-flight fix planned.

The OLD tests model same-process ForceStop+resume and rely on the old behavior where the session stayed alive in the runtime process. They need to be redesigned to either keep an owner connected (warm resume) or to model true process-restart resume from persisted session state.

No code change beyond the skip-reason text; tests remain skipped while the redesign is figured out.

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

Author: stephentoub

dotnet/test/E2E/PendingWorkResumeE2ETests.cs

@@ -19,12 +19,14 @@ public class PendingWorkResumeE2ETests(E2ETestFixture fixture, ITestOutputHelper
     private static readonly TimeSpan PendingWorkTimeout = TimeSpan.FromSeconds(60);
     private const string SharedToken = "pending-work-resume-shared-token";
 
-    // TODO: Re-enable once the runtime restores warm-resume behavior for pending permission
-    // requests after the original client disconnects. Runtime PR #9040 (commit b8e1220b45)
-    // now cleans up session state when the last RPC owner disconnects, causing
-    // HandlePendingPermissionRequest to return success=false on resume. A runtime-side
-    // fix is being tracked separately.
-    [Fact(Skip = "Pending runtime fix: cold cleanup contract makes HandlePendingPermissionRequest return success=false after disconnect+resume.")]
+    // Skipped after the runtime 1.0.56 bump. Runtime PR #9040 (commit b8e1220b45) changed
+    // SDKServer.handleConnectionClosed to tear down the session when the last RPC client
+    // disconnects, so the in-memory pending permission request is gone before the resumed
+    // client can satisfy it and HandlePendingPermissionRequest returns success=false. This
+    // test models same-process ForceStop+resume; it needs to be redesigned to either keep
+    // an owner connected (warm resume) or to model a true process restart against the
+    // persisted session state.
+    [Fact(Skip = "Runtime 1.0.56 cleans up the session on last-client disconnect (copilot-agent-runtime PR #9040), so the in-memory pending request is gone before resume can satisfy it. Test needs redesign.")]
     public async Task Should_Continue_Pending_Permission_Request_After_Resume()
     {
         var originalPermissionRequest = new TaskCompletionSource<PermissionRequest>(TaskCreationOptions.RunContinuationsAsynchronously);
@@ -93,10 +95,12 @@ static string ResumePermissionTool([Description("Value to transform")] string va
             $"ORIGINAL_SHOULD_NOT_RUN_{value}";
     }
 
-    // TODO: Re-enable once the runtime restores warm-resume behavior for pending external
-    // tool calls after the original client disconnects. Same root cause as
-    // Should_Continue_Pending_Permission_Request_After_Resume above.
-    [Fact(Skip = "Pending runtime fix: cold cleanup contract makes HandlePendingToolCall return success=false after disconnect+resume.")]
+    // Skipped for the same reason as Should_Continue_Pending_Permission_Request_After_Resume:
+    // runtime 1.0.56 (copilot-agent-runtime PR #9040) tears down the session when the last
+    // RPC client disconnects, so the in-memory pending external tool call is gone before
+    // the resumed client can satisfy it. Needs redesign to keep an owner connected (warm)
+    // or to model true process-restart resume from persisted state.
+    [Fact(Skip = "Runtime 1.0.56 cleans up the session on last-client disconnect (copilot-agent-runtime PR #9040), so the in-memory pending tool call is gone before resume can satisfy it. Test needs redesign.")]
     public async Task Should_Continue_Pending_External_Tool_Request_After_Resume()
     {
         var originalToolStarted = new TaskCompletionSource<string>(TaskCreationOptions.RunContinuationsAsynchronously);

go/internal/e2e/pending_work_resume_e2e_test.go

@@ -25,12 +25,14 @@ func TestPendingWorkResumeE2E(t *testing.T) {
 	ctx := testharness.NewTestContext(t)
 
 	t.Run("should continue pending permission request after resume", func(t *testing.T) {
-		// TODO: Re-enable once the runtime restores warm-resume behavior for pending
-		// permission requests after the original client disconnects. Runtime PR #9040
-		// (commit b8e1220b45) now cleans up session state when the last RPC owner
-		// disconnects, causing HandlePendingPermissionRequest to return Success=false
-		// on resume. A runtime-side fix is being tracked separately.
-		t.Skip("Pending runtime fix: cold cleanup contract makes HandlePendingPermissionRequest return Success=false after disconnect+resume.")
+		// Skipped after the runtime 1.0.56 bump. Runtime PR #9040 (commit b8e1220b45)
+		// changed SDKServer.handleConnectionClosed to tear down the session when the
+		// last RPC client disconnects, so the in-memory pending permission request is
+		// gone before the resumed client can satisfy it and HandlePendingPermissionRequest
+		// returns Success=false. This test models same-process ForceStop+resume; it
+		// needs to be redesigned to either keep an owner connected (warm resume) or to
+		// model a true process restart against the persisted session state.
+		t.Skip("Runtime 1.0.56 cleans up the session on last-client disconnect (copilot-agent-runtime PR #9040), so the in-memory pending request is gone before resume can satisfy it. Test needs redesign.")
 
 		ctx.ConfigureForTest(t)
 
@@ -145,10 +147,13 @@ func TestPendingWorkResumeE2E(t *testing.T) {
 	})
 
 	t.Run("should continue pending external tool request after resume", func(t *testing.T) {
-		// TODO: Re-enable once the runtime restores warm-resume behavior for pending
-		// external tool calls after the original client disconnects. Same root cause as
-		// "should continue pending permission request after resume" above.
-		t.Skip("Pending runtime fix: cold cleanup contract makes HandlePendingToolCall return Success=false after disconnect+resume.")
+		// Skipped for the same reason as "should continue pending permission request
+		// after resume": runtime 1.0.56 (copilot-agent-runtime PR #9040) tears down
+		// the session when the last RPC client disconnects, so the in-memory pending
+		// external tool call is gone before the resumed client can satisfy it. Needs
+		// redesign to keep an owner connected (warm) or to model true process-restart
+		// resume from persisted state.
+		t.Skip("Runtime 1.0.56 cleans up the session on last-client disconnect (copilot-agent-runtime PR #9040), so the in-memory pending tool call is gone before resume can satisfy it. Test needs redesign.")
 
 		ctx.ConfigureForTest(t)
 

nodejs/test/e2e/pending_work_resume.e2e.test.ts

@@ -166,11 +166,13 @@ describe("Pending work resume", async () => {
         return `localhost:${port}`;
     }
 
-    // TODO: Re-enable once the runtime restores warm-resume behavior for pending
-    // permission requests after the original client disconnects. Runtime PR #9040
-    // (commit b8e1220b45) now cleans up session state when the last RPC owner
-    // disconnects, causing handlePendingPermissionRequest to return success=false
-    // on resume. A runtime-side fix is being tracked separately.
+    // Skipped after the runtime 1.0.56 bump. Runtime PR #9040 (commit b8e1220b45)
+    // changed SDKServer.handleConnectionClosed to tear down the session when the
+    // last RPC client disconnects, so the in-memory pending permission request is
+    // gone before the resumed client can satisfy it and handlePendingPermissionRequest
+    // returns success=false. This test models same-process ForceStop+resume; it needs
+    // to be redesigned to either keep an owner connected (warm resume) or to model
+    // a true process restart against the persisted session state.
     it.skip(
         "should continue pending permission request after resume",
         { timeout: TEST_TIMEOUT_MS },
@@ -244,9 +246,12 @@ describe("Pending work resume", async () => {
         }
     );
 
-    // TODO: Re-enable once the runtime restores warm-resume behavior for pending
-    // external tool calls after the original client disconnects. Same root cause as
-    // "should continue pending permission request after resume" above.
+    // Skipped for the same reason as "should continue pending permission request
+    // after resume": runtime 1.0.56 (copilot-agent-runtime PR #9040) tears down the
+    // session when the last RPC client disconnects, so the in-memory pending external
+    // tool call is gone before the resumed client can satisfy it. Needs redesign to
+    // keep an owner connected (warm) or to model true process-restart resume from
+    // persisted state.
     it.skip(
         "should continue pending external tool request after resume",
         { timeout: TEST_TIMEOUT_MS },

python/e2e/test_pending_work_resume_e2e.py

@@ -128,15 +128,18 @@ async def _safe_force_stop(client: CopilotClient) -> None:
 
 
 class TestPendingWorkResume:
-    # TODO: Re-enable once the runtime restores warm-resume behavior for pending
-    # permission requests after the original client disconnects. Runtime PR #9040
-    # (commit b8e1220b45) now cleans up session state when the last RPC owner
-    # disconnects, causing handle_pending_permission_request to return success=False
-    # on resume. A runtime-side fix is being tracked separately.
+    # Skipped after the runtime 1.0.56 bump. Runtime PR #9040 (commit b8e1220b45)
+    # changed SDKServer.handleConnectionClosed to tear down the session when the last
+    # RPC client disconnects, so the in-memory pending permission request is gone
+    # before the resumed client can satisfy it and handle_pending_permission_request
+    # returns success=False. This test models same-process force_stop+resume; it
+    # needs to be redesigned to either keep an owner connected (warm resume) or to
+    # model a true process restart against the persisted session state.
     @pytest.mark.skip(
         reason=(
-            "Pending runtime fix: cold cleanup contract makes "
-            "handle_pending_permission_request return success=False after disconnect+resume."
+            "Runtime 1.0.56 cleans up the session on last-client disconnect "
+            "(copilot-agent-runtime PR #9040), so the in-memory pending request "
+            "is gone before resume can satisfy it. Test needs redesign."
         )
     )
     async def test_should_continue_pending_permission_request_after_resume(
@@ -219,13 +222,17 @@ def resumed_tool_handler(args):
         finally:
             await _safe_force_stop(server)
 
-    # TODO: Re-enable once the runtime restores warm-resume behavior for pending
-    # external tool calls after the original client disconnects. Same root cause as
-    # test_should_continue_pending_permission_request_after_resume above.
+    # Skipped for the same reason as
+    # test_should_continue_pending_permission_request_after_resume: runtime 1.0.56
+    # (copilot-agent-runtime PR #9040) tears down the session when the last RPC
+    # client disconnects, so the in-memory pending external tool call is gone before
+    # the resumed client can satisfy it. Needs redesign to keep an owner connected
+    # (warm) or to model true process-restart resume from persisted state.
     @pytest.mark.skip(
         reason=(
-            "Pending runtime fix: cold cleanup contract makes "
-            "handle_pending_tool_call return success=False after disconnect+resume."
+            "Runtime 1.0.56 cleans up the session on last-client disconnect "
+            "(copilot-agent-runtime PR #9040), so the in-memory pending tool call "
+            "is gone before resume can satisfy it. Test needs redesign."
         )
     )
     async def test_should_continue_pending_external_tool_request_after_resume(