fix: the Await bridge moves to a PRODUCT assembly — product code deadlocks too

.github/copilot-instructions.md

@@ -42,7 +42,7 @@ gave you no synchronous overload. Flag it and name the boundary it should have c
 - **Test projects** (`test/**`). 🚨 `.ToTask()` is FORBIDDEN here too (2026-08-30): Rx completes its
   `TaskCompletionSource` without `RunContinuationsAsynchronously`, so the awaiter resumes INLINE on the
   signalling thread inside Rx's trampoline and every later `await` inherits that scheduler. Tests await
-  through `ObservableAwait.Await(ct)` (MeshWeaver.Fixture) or `ReactiveCompletion.ObserveCompletion`.
+  through `ObservableAwait.Await(ct)` (MeshWeaver.Messaging, in MeshWeaver.Messaging.Hub — product and test code alike) or `ReactiveCompletion.ObserveCompletion`.
 - **A one-line SDK/MCP surface adapter** whose body is reactive, e.g.
   `public Task<string> Patch(...) => ops.Patch(...).FirstAsync().ObserveCompletion(Report, ct);` — an external
   signature may demand a `Task`; it is still never produced with `.ToTask()`.

scripts/type-forwards.allow

@@ -25,3 +25,14 @@
 # $(MeshWeaverRoot) (core, MeshWeaver.Plugins, MeshWeaver.SocialMedia — the latter pinned
 # pre-split). A forwarder is impossible here: the base would have to reference the AI assembly,
 # which references the base (the documented reference-cycle case).
+
+# #2771 — ObservableAwait leaves MeshWeaver.Fixture for MeshWeaver.Messaging.Hub (namespace
+# MeshWeaver.Fixture → MeshWeaver.Messaging), so product code gets the deadlock-safe bridge. No
+# shipped package can hold the TypeRef: the type was added 2026-08-30 by #2750, one day AFTER the
+# last release (v3.0.0-rc8, 2026-08-29), and the shipped MeshWeaver.Fixture 3.0.0-rc8 nupkg was
+# pulled from nuget.org and inspected — it contains no ObservableAwait (control: HubTestBase,
+# TestBase, ConfigureMesh all present in the same DLL). Fixture is in no module bundle or bake
+# surface; its only consumers are TestBase assemblies and test projects that compile from source.
+# A forwarder is impossible without keeping `namespace MeshWeaver.Fixture` inside a product
+# assembly, which would tell product code to `using MeshWeaver.Fixture` — the wrong direction.
+MeshWeaver.Fixture:MeshWeaver.Fixture.ObservableAwait  #2771 — added by #2750 after v3.0.0-rc8; never shipped

src/MeshWeaver.Fixture/ObservableAwait.cs → src/MeshWeaver.Messaging.Hub/ObservableAwait.cs

@@ -1,10 +1,17 @@
 using System.Reactive.Linq;
 
-namespace MeshWeaver.Fixture;
+namespace MeshWeaver.Messaging;
 
 /// <summary>
-/// The ONE way a test waits for an observable — and the reason <c>.ToTask()</c> does not appear in
-/// this repository (maintainer, 2026-08-30: <i>"no ToTask ever"</i>).
+/// The ONE way anything in this codebase waits for an observable — and the reason <c>.ToTask()</c>
+/// does not appear in it (maintainer, 2026-08-30: <i>"no ToTask ever"</i>, <i>"strictly"</i>).
+///
+/// <para>🚨 It lives in a PRODUCT assembly, not the test fixture, because the rule is not a test
+/// convention. The bridges that actually deadlock a hub are the ones in product code — an SDK
+/// boundary we must implement (a Microsoft.Agent.Framework <c>…Async</c> override, an
+/// <c>IHostedService</c>, ASP.NET middleware) whose signature is not ours to change. Those had no
+/// sanctioned bridge to move to while this helper was test-only, which is why
+/// MeshWeaver.Plugins still carries 80 of them in <c>src/</c> while core carries none.</para>
 ///
 /// <para>🚨 Rx's own <c>ToTask()</c> completes its <c>TaskCompletionSource</c> WITHOUT
 /// <c>RunContinuationsAsynchronously</c>, so the awaiting test resumes INLINE on whichever thread

test/Directory.Build.props

@@ -62,6 +62,22 @@
          covers Should()/asserts + the Seconds()/Milliseconds()/… time helpers, so
          test files need no per-file assertion import. -->
     <Using Include="MeshWeaver.Reactive.Assertions" />
+    <!-- ObservableAwait.Await — the ONE sanctioned observable→Task bridge — moved from
+         MeshWeaver.Fixture (a TEST assembly) into MeshWeaver.Messaging.Hub, because the bridges
+         that actually deadlock a hub are in PRODUCT code: an SDK boundary we must implement
+         (a Microsoft.Agent.Framework ...Async override, IHostedService, ASP.NET middleware) cannot
+         reference the test fixture, so those had nowhere to go and stayed on Rx's .ToTask().
+         Imported globally here so the ~330 existing call sites need no churn, and so exactly ONE
+         implementation exists — a forward in the old namespace made every .Await(ct) ambiguous
+         (CS0121) in the files that import both. -->
+    <!-- Excluded: the three test projects with no messaging dependency at all (they assert JSON,
+         image packing and browser E2E). A global using cannot resolve a namespace the project does
+         not reference, so naming them is honest; adding a ProjectReference just to satisfy a using
+         would be the wrong direction. -->
+    <Using Include="MeshWeaver.Messaging"
+           Condition="'$(MSBuildProjectName)' != 'MeshWeaver.Json.Test'
+                  and '$(MSBuildProjectName)' != 'MeshWeaver.PluginImage.Test'
+                  and '$(MSBuildProjectName)' != 'MeshWeaver.Portal.E2E.Test'" />
   </ItemGroup>
 
   <ItemGroup>

test/MeshWeaver.Documentation.Test/ObservableToTaskBridgeGuard.cs

@@ -202,11 +202,17 @@ private static readonly (string File, string Why)[] SanctionedBridges =
             + "merge into a path-rename conflict. When the move lands, this entry goes stale and "
             + "the register test above will say so."),
 
-        ("src/MeshWeaver.Fixture/ObservableAwait.cs",
-            "The test tree's ONE wait, introduced by #2750 when 1,538 `.ToTask(` call sites were "
+        ("src/MeshWeaver.Messaging.Hub/ObservableAwait.cs",
+            "The ONE wait, introduced by #2750 for the test tree when 1,538 `.ToTask(` call sites "
+            + "were swept, and moved by #2771 from MeshWeaver.Fixture into the messaging assembly so "
+            + "PRODUCT code has a deadlock-safe bridge too: the sibling repos carried 65 raw "
+            + "`.ToTask(` sites in product source that could not be swept while the only safe "
+            + "wrapper lived in a test assembly. Same category as ReactiveWait: infrastructure whose "
+            + "entire PURPOSE is to hand a caller a Task to await, so it cannot be bridge-free. "
+            + "(The type-forwards gate was consulted: the type is one day younger than the last "
+            + "release, so no shipped package binds the old name — see scripts/type-forwards.allow.) "
             + "swept. Same category as ReactiveWait: infrastructure whose entire PURPOSE is to hand "
-            + "a test a Task to await, so it cannot be bridge-free, and it ships from a production "
-            + "root because MeshWeaver.Fixture is where the test bases live. It is a FAITHFUL "
+            + "It is a FAITHFUL "
             + "ToTask (last value, faults on an empty sequence) — deliberately, because settling on "
             + "the first notification would have silently changed 462 call sites that do not reduce "
             + "to a single element; only the continuation scheduling differs. Safe form. "