deps: bump MCP 1.2 -> 1.3 + e2e bootstrap

.agents/routine-last-run.md

@@ -71,3 +71,61 @@ so the next run can resume from the same line of work without re-discovering it.
 
 **Handoff:** none. No work that belongs in `qyl-unit-tests`,
 `qyl-integration-tests`, or `qyl-e2e-tests` was deferred from this run.
+
+## qyl-e2e-tests 2026-05-17 02:45
+
+**Outcome:** bootstrap PR opened.
+
+First-ever run of `qyl-e2e-tests` on a workstation with the full toolchain
+present (dotnet 10.0.300, Docker 29.4.0). The previous run (2026-05-16)
+exited as a no-op because the remote container lacked `dotnet`.
+
+Per the skill's bootstrap section, this run produced the **infrastructure-only
+PR** (project + topology fixture + Nuke target + central-package pins). **No
+scenario tests** were added; the next routine run picks up from here and adds
+the first scenario.
+
+**Changes shipped:**
+- `tests/qyl.e2e.tests/qyl.e2e.tests.csproj` — new test project, `ANcpLua.NET.Sdk.Test`.
+- `tests/qyl.e2e.tests/E2ECollection.cs` — `[CollectionDefinition("E2E", DisableParallelization = true)]`.
+- `tests/qyl.e2e.tests/Topology/QylTopologyOptions.cs` — image tags + startup timeout.
+- `tests/qyl.e2e.tests/Topology/QylTopologyFixture.cs` — programmatic Testcontainers
+  topology: bridge network → `qyl-collector:latest` → `qyl-mcp:latest`; WireMock
+  process-local; containers reach it via `host.docker.internal`.
+  `WithImagePullPolicy(_ => false)` enforces local-image-only.
+- `tests/qyl.e2e.tests/Bootstrap/WireMockLlmSeamTests.cs` — `Category=E2EBootstrap`
+  smoke (no Docker) — proves the WireMock seam roundtrips a scripted
+  `/v1/chat/completions` and shows up in `LogEntries`.
+- `eng/build/BuildTest.cs` — new `E2ETests` target depending on
+  `IDocker.DockerImageBuild`, filters `Category=E2E`. Default `Test` excludes
+  `Category=E2E` (bootstrap tests stay in).
+- `qyl.slnx` — registered the project.
+- `Version.props` + `Directory.Packages.props` — added `WireMock.Net 2.6.0` and
+  `Testcontainers 4.11.0`. Split-pinned `OpenTelemetry.Instrumentation.AspNetCore`
+  to 1.15.2 (forced by WireMock.Net 2.6.0 transitive); Http + Runtime stay at
+  the umbrella 1.15.1 (no 1.15.2 release exists for them).
+
+**Verification:**
+- `dotnet build qyl.slnx` — 0 errors, 1454 warnings (baseline 1393; ~61 new
+  warnings are all `MultipleGlobalAnalyzerKeys` from the dual-`.globalconfig`
+  worktree setup — benign, present on every worktree build).
+- `dotnet test tests/qyl.e2e.tests --filter-trait Category=E2EBootstrap` — 3
+  consecutive runs, all green (~1s each, 0 flakes).
+- `nuke E2ETests` target wired up but **not executed** this run (would require
+  rebuilding all four qyl Docker images — out of scope for the bootstrap PR).
+
+**Overlap with `Smoke`:** `eng/smoke/run.sh` (Nuke target `Smoke`) is the
+PRD #173 quality gate using real Ollama + real qyl Compose stack. E2E uses
+**WireMock** for deterministic LLM stubbing. The two don't overlap — Smoke
+covers "does the stack work against a real model"; E2E covers "does the
+stack route data correctly given a known-bad/redacted LLM response".
+
+**Gaps for the next run:** add the first real scenario. Highest-value candidates:
+1. Agent submits chat → trace arrives at downstream sink with credentials redacted.
+2. MCP HTTP session reconnect after transient collector failure.
+3. Cost rollup updates after a single chat completion.
+
+Pick exactly one. Add a sink container (e.g. an OTel collector configured to
+write to a file volume) to `QylTopologyFixture` for scenario 1.
+
+**Handoff:** bootstrap PR opened; next cycle picks up from here.

Directory.Packages.props

@@ -20,8 +20,10 @@
     <PackageVersion Include="OpenTelemetry.Extensions.Hosting" Version="$(OpenTelemetryVersion)"/>
     <!-- Instrumentation packages ship on a slower cadence than the SDK — pin
          them to the shared instrumentation version (1.15.1) rather than the
-         SDK version (1.15.3). -->
-    <PackageVersion Include="OpenTelemetry.Instrumentation.AspNetCore" Version="$(OpenTelemetryAspNetCoreInstrumentationVersion)"/>
+         SDK version (1.15.3). AspNetCore is bumped to 1.15.2 because
+         WireMock.Net 2.6.0 (consumed by tests/qyl.e2e.tests) transitively
+         requires it; Http and Runtime have no 1.15.2 release, hence the split. -->
+    <PackageVersion Include="OpenTelemetry.Instrumentation.AspNetCore" Version="1.15.2"/>
     <PackageVersion Include="OpenTelemetry.Instrumentation.Http" Version="$(OpenTelemetryAspNetCoreInstrumentationVersion)"/>
     <PackageVersion Include="OpenTelemetry.Instrumentation.Runtime" Version="$(OpenTelemetryAspNetCoreInstrumentationVersion)"/>
     <!-- Microsoft.Extensions -->
@@ -108,6 +110,9 @@
     <PackageVersion Include="YamlDotNet" Version="$(YamlDotNetVersion)"/>
     <!-- Terminal UI -->
     <PackageVersion Include="Spectre.Console" Version="$(SpectreConsoleVersion)"/>
+    <!-- E2E test topology (tests/qyl.e2e.tests only): deterministic LLM stub + Docker topology -->
+    <PackageVersion Include="WireMock.Net" Version="$(WireMockNetVersion)"/>
+    <PackageVersion Include="Testcontainers" Version="$(TestcontainersVersion)"/>
   </ItemGroup>
   <ItemGroup>
     <PackageDownload Include="GitVersion.Tool" Version="[$(GitVersionToolVersion)]"/>

Version.props

@@ -47,7 +47,7 @@
 
     <OpenAIVersion>2.10.0</OpenAIVersion>
     <SystemClientModelVersion>1.11.0</SystemClientModelVersion>
-    <ModelContextProtocolVersion>1.2.0</ModelContextProtocolVersion>
+    <ModelContextProtocolVersion>1.3.0</ModelContextProtocolVersion>
     <MicrosoftIdentityModelVersion>8.17.0</MicrosoftIdentityModelVersion>
     <NukeVersion>10.1.0</NukeVersion>
     <NetArchTestVersion>1.3.2</NetArchTestVersion>
@@ -57,5 +57,9 @@
     <ReportGeneratorVersion>5.5.1</ReportGeneratorVersion>
     <VersionizeVersion>2.5.0</VersionizeVersion>
     <AwesomeAssertionsVersion>9.4.0</AwesomeAssertionsVersion>
+    <!-- E2E test topology: WireMock for deterministic LLM stubbing, Testcontainers for
+         Docker-driven topology spin-up. Used only by tests/qyl.e2e.tests. -->
+    <WireMockNetVersion>2.6.0</WireMockNetVersion>
+    <TestcontainersVersion>4.11.0</TestcontainersVersion>
   </PropertyGroup>
 </Project>

eng/build/BuildTest.cs

@@ -176,6 +176,12 @@ interface IQylTest : ITest, IHazSourcePaths
             AssertAtLeastOneTestExecuted("Functional");
         });
 
+    Target E2ETests => d => d
+        .Description("Run end-to-end tests (full Docker topology against freshly built images)")
+        .DependsOn<ICompile>(static x => x.Compile)
+        .DependsOn<IDocker>(static x => x.DockerImageBuild)
+        .Executes(() => RunFilteredE2ETests());
+
     Target TestSummary => d => d
         .Unlisted()
         .Description("Generate Markdown test summary from MTP TRX reports")
@@ -198,10 +204,12 @@ interface IQylTest : ITest, IHazSourcePaths
             .ReportTrx($"{project.Name}.trx")
             .IgnoreExitCode(8);
 
-        // Heavy/opt-in tests (Category=regen — shell out to Weaver, Category=integration etc.)
-        // are excluded from the default Test run; pass --IQylTest.TestFilter to include them.
+        // Heavy/opt-in tests (Category=regen — shell out to Weaver; Category=E2E — full
+        // Docker topology via DockerImageBuild) are excluded from the default Test run;
+        // pass --IQylTest.TestFilter to include them, or run the dedicated sub-target
+        // (E2ETests). E2EBootstrap-traited tests (no Docker) intentionally stay in.
         if (TestFilter is { Length: > 0 } f) mtp.FilterQuery(f);
-        else mtp.FilterNotTrait("Category", "regen");
+        else mtp.FilterNotTrait("Category", "regen").FilterNotTrait("Category", "E2E");
 
         if (StopOnFail == true) mtp.StopOnFail();
         if (LiveOutput == true || IsLocalBuild) mtp.ShowLiveOutput();
@@ -236,6 +244,41 @@ sealed void RunFilteredTests(string namespaceFilter, string trxSuffix, bool need
             }), completeOnFailure: true);
     }
 
+    sealed void RunFilteredE2ETests()
+    {
+        EnsureTestcontainersConfigured();
+
+        var e2eProjects = TestProjects
+            .Where(static p => p.Name.Equals("qyl.e2e.tests", StringComparison.OrdinalIgnoreCase))
+            .ToArray();
+
+        if (e2eProjects.Length is 0)
+        {
+            Log.Warning("E2ETests: no qyl.e2e.tests project found; skipping");
+            return;
+        }
+
+        DotNetTasks.DotNetTest(s => s
+            .SetNoBuild(true)
+            .SetNoRestore(true)
+            .SetResultsDirectory(TestResultsDirectory)
+            .CombineWith(e2eProjects, (ss, project) =>
+            {
+                var mtp = MtpExtensions.Mtp()
+                    .ReportTrx($"{project.Name}.E2E.trx")
+                    .IgnoreExitCode(8)
+                    .FilterTrait("Category", "E2E");
+
+                if (StopOnFail == true) mtp.StopOnFail();
+                if (LiveOutput == true || IsLocalBuild) mtp.ShowLiveOutput();
+
+                var projectPath = project.Path ??
+                                  throw new InvalidOperationException($"Project '{project.Name}' has no path");
+                string[] args = ["--project", projectPath.ToString(), .. mtp.BuildArgs().Prepend("--")];
+                return ss.SetProcessAdditionalArguments(args);
+            }), completeOnFailure: true);
+    }
+
     sealed void EnsureTestcontainersConfigured()
     {
         if (IsServerBuild)

qyl.slnx

@@ -45,6 +45,7 @@
   <Folder Name="/tests/">
     <Project Path="tests/qyl.collector.integration.tests/qyl.collector.integration.tests.csproj" />
     <Project Path="tests/qyl.collector.tests/qyl.collector.tests.csproj" />
+    <Project Path="tests/qyl.e2e.tests/qyl.e2e.tests.csproj" />
     <Project Path="tests/qyl.mcp.tests/qyl.mcp.tests.csproj" />
   </Folder>
 

services/qyl.mcp/Formatting/ErrorFormatter.cs

@@ -11,10 +11,19 @@ public static string FormatForLlm(Exception error, McpTransportMode transport) =
             TaskCanceledException { InnerException: TimeoutException or null } =>
                 "**Timeout:** The collector did not respond in time. Retry or check if qyl collector is running.",
             OperationCanceledException opEx => FormatOperationCancelled(opEx),
+            // ModelContextProtocol 1.3 throws IOException (incl. ClientTransportClosedException)
+            // for transport connect/closure failures that 1.2 wrapped as InvalidOperationException.
+            // This arm must precede InvalidOperationException — order matters in C# pattern switches.
+            IOException ioEx => FormatTransportError(ioEx, transport),
             InvalidOperationException configEx => FormatConfigError(configEx, transport),
             _ => FormatUnknown(error, transport)
         };
 
+    private static string FormatTransportError(IOException ex, McpTransportMode transport) =>
+        transport is McpTransportMode.Stdio
+            ? $"**Connection Error**\n\n{ex.Message}\n\nCheck if the MCP server process is running."
+            : $"**Connection Error**\n\n{ex.Message}\n\nCheck the endpoint URL and network reachability.";
+
     private static string FormatHttpError(HttpRequestException ex, McpTransportMode transport)
     {
         var (category, hint) = ex.StatusCode switch

tests/qyl.e2e.tests/Bootstrap/WireMockLlmSeamTests.cs

@@ -0,0 +1,48 @@
+using System.Net.Http.Json;
+using System.Text.Json;
+using WireMock.RequestBuilders;
+using WireMock.ResponseBuilders;
+using WireMock.Server;
+
+namespace Qyl.E2E.Tests.Bootstrap;
+
+[Trait("Category", "E2EBootstrap")]
+public sealed class WireMockLlmSeamTests
+{
+    [Fact]
+    public async Task ScriptedChatCompletion_RoundtripsAndIsRecordedInLogEntries()
+    {
+        using var llm = WireMockServer.Start();
+
+        llm.Given(Request.Create().WithPath("/v1/chat/completions").UsingPost())
+            .RespondWith(Response.Create()
+                .WithStatusCode(200)
+                .WithHeader("Content-Type", "application/json")
+                .WithBodyAsJson(new
+                {
+                    id = "chatcmpl-bootstrap",
+                    choices = new[]
+                    {
+                        new { message = new { role = "assistant", content = "Bearer secret-token-12345" } },
+                    },
+                }));
+
+        var ct = TestContext.Current.CancellationToken;
+        using var client = new HttpClient { BaseAddress = new Uri(llm.Url!) };
+        var request = new
+        {
+            model = "gpt-4o-mini",
+            messages = new[] { new { role = "user", content = "summarize this" } },
+        };
+        using var response = await client.PostAsJsonAsync("/v1/chat/completions", request, ct);
+
+        response.IsSuccessStatusCode.Should().BeTrue(
+            "the WireMock stub must respond 200 for the configured route");
+
+        using var payload = await JsonDocument.ParseAsync(await response.Content.ReadAsStreamAsync(ct), cancellationToken: ct);
+        payload.RootElement.GetProperty("id").GetString().Should().Be("chatcmpl-bootstrap");
+
+        llm.LogEntries.Should().ContainSingle(
+            "WireMock must record the single POST /v1/chat/completions for assertion replay");
+    }
+}

tests/qyl.e2e.tests/E2ECollection.cs

@@ -0,0 +1,7 @@
+namespace Qyl.E2E.Tests;
+
+[CollectionDefinition(Name, DisableParallelization = true)]
+public sealed class E2ECollection : ICollectionFixture<Topology.QylTopologyFixture>
+{
+    public const string Name = "E2E";
+}