[Blind-Striker]

📝 Description

What does this PR do?
It adds nine real-server tests for the local credential, integration and MCP catalog operations, the last of #380's sweep. They run on the existing simulated-drive fixture, with no new fixture, provider, support type or dependency.

ID	Operation	Test
C01	RemoveCredentialAsync	CredentialsClientLiveTests.RemoveCredentialAsync_Should_Remove_The_Owned_Row_When_It_Exists
C02	CreateCredentialAsync	CredentialsClientLiveTests.CreateCredentialAsync_Should_Persist_The_Row_When_The_Id_Is_Unused
C03	ActivateCredentialAsync	CredentialsClientLiveTests.ActivateCredentialAsync_Should_Select_The_Older_Row_When_A_Newer_Row_Is_Active
C04	ListCredentialsAsync	CredentialsClientLiveTests.ListCredentialsAsync_Should_Return_The_Owned_Metadata_When_A_Row_Was_Created
C05	UpdateCredentialAsync	CredentialsClientLiveTests.UpdateCredentialAsync_Should_Replace_The_Label_When_A_Row_Exists
I01	ConnectWithKeyAsync	IntegrationClientLiveTests.ConnectWithKeyAsync_Should_Store_A_Connection_When_The_Seeded_Key_Method_Is_Used
I02	IntegrationClient.GetAsync	IntegrationClientLiveTests.GetAsync_Should_Return_The_Seeded_Integration_When_Its_Id_Is_Requested
I03	ListIntegrationsAsync	IntegrationsClientLiveTests.ListIntegrationsAsync_Should_Include_The_Seeded_Provider_When_Plugins_Are_Activated
M01	ListMcpServersAsync	McpServersClientLiveTests.ListMcpServersAsync_Should_Report_The_Disabled_Entry_When_It_Was_Added_At_Runtime

Secrets. Credential values are inert sentinels the tests create. No test prints, compares, serializes or dumps a value, a response body or RawBody. Every list is filtered to the test's own IDs or labels, with no global counts and no host or environment credential touched.

Ownership. Each credential, connection and MCP entry has its cleanup registered before it is created, so a lost response still cleans up. M01's MCP entry is typed, disabled, OAuth-off, and points at an inert .invalid URL, so no connection is attempted.

With this PR, #380's sweep is complete: PRs 1–4 cover every reachable operation the issue named, and the out-list with reasons is on the issue.

Part of #380 (PR 4 of 4).

🔄 Type of Change

• 🧪 Test improvements

🎯 Protocol & API Surface

• No change to the public API surface
• Nothing under external/, spec/ or src/ was touched

🧪 Testing

Each test is shown to catch the missing or wrong operation. It is green, then red at its oracle, then green again; the mutations are not committed.

Test	Mutation	Failure
C01	Success without removing	the owned ID remains
C02	Valid success fabricated without inserting	the owned ID is missing
C03	Success without activating	the older row stays inactive
C04	Empty successful list	the owned ID is missing
C05	Success without updating	the label stays the initial one
I01	Success without connecting	owned rows: expected 1, found 0
I02	Sent to /api/integration/opencode	expected sim, received opencode
I03	Empty successful catalog	sim is missing
M01	Empty successful catalog	the owned MCP name is missing, before any event wait

Every failure is an assertion, never a hang guard or timeout. M01 checks the list before it waits for the status event, so a no-op cannot stall it.

Review. A fresh read-only review found no defects. Its one note was a null-forgiving request.Id! without a stated invariant (slop-review R7). The test helper now returns the ID it assigned, so the operator is gone. The resulting arrange copy between C02 and C04 moved into a private method in the class; the Contains(id) oracle each mutant failed on is unchanged.

Merge-queue ejection, and the fix. The first merge group failed the clone gate. #423 (PR 3, merged just before) and this PR each carried the same owned-MCP removal helper:

• ExperimentalClientLiveTests:185-193
• McpServersClientLiveTests:59-68

Each PR was clone-free on its own. Two slices now use the disabled-MCP registration and its owned removal, so per testing-style.md they live in the root Support/. The new helper Support/DisabledMcpServer.cs holds both, and the two classes call it instead of private copies. The helper's removal still tolerates only the exact already-missing refusal. The fix is rebased onto #423's merge, 3cbb8c14.

✅ Completion Gate

Local check (quality-gates.md) on the first head, base cff6057f, idle host. Every step's own exit code was read:

• slopwatch: zero findings
• dotnet build --configuration Release: 0 warnings, 0 errors
• dotnet format whitespace / dotnet format style --severity warn: no changes
• verify-clones: passed, 0 new (tests 5.68%, under the 6% ceiling); no baseline change
• OpenCode.Sdk.Tools.Tests (all but NativeAot): 1995, 0 failed
• OpenCode.Sdk.Tests (all but ServiceElection/NativeAot): 8317, 0 failed

After review, a narrow re-check ran on that head. The head 65db67ae changes only CredentialsClientLiveTests.cs (the R7 fix and its private method). On it:

• the SDK test project builds on all target frameworks with 0 warnings;
• both format checks report no changes;
• verify-clones reports 0 new;
• CredentialsClientLiveTests passes 15, 0 failed;
• the RepositoryGates guards pass 93, 0 failed.

Final head b66bc866 (rebased onto 3cbb8c14, with the shared helper):

• dotnet build --configuration Release: 0 warnings, 0 errors
• dotnet format style and whitespace: no changes
• slopwatch: zero findings
• verify-clones: passed, 0 new (tests 5.66%)
• OpenCode.Sdk.Tools.Tests (all but NativeAot): 1995, 0 failed
• OpenCode.Sdk.Tests (all but ServiceElection/NativeAot): 8344, 0 failed

The host runs Bun 1.4.3, a maintainer-named off-pin host (OPENCODE_SDK_TESTS_BUN_MISMATCH=allow); the pin is 1.4.2.

✅ Code Quality Checklist

• No new suppression
• No tests skipped, weakened, or deleted to make the suite pass