Fix zero-state chat trap + auto-approve own-node sub-pairing

src/OpenClaw.Chat/ChatModels.cs

@@ -123,7 +123,30 @@ public record ChatDataSnapshot(
     IReadOnlyDictionary<string, ChatTimelineState> Timelines,
     string? DefaultThreadId,
     string? ConnectionStatus,
-    string[] AvailableModels);
+    string[] AvailableModels,
+    ChatComposeTarget ComposeTarget);
+
+/// <summary>
+/// Describes where the UI may send the next chat message. Distinct from
+/// <see cref="ChatDataSnapshot.Threads"/> because, in protocols like the
+/// OpenClaw gateway, there is always a canonical "main" target whose session
+/// row may not have been materialized yet (zero sessions on fresh install).
+/// The UI uses this to decide whether the composer should be enabled even
+/// when <see cref="ChatDataSnapshot.Threads"/> is empty.
+/// </summary>
+/// <param name="SessionKey">
+/// The provider-resolved canonical session key for the default send target,
+/// or <c>null</c> when not yet known (e.g. handshake incomplete).
+/// </param>
+/// <param name="IsReady">
+/// True when the provider is connected, has resolved a canonical send target,
+/// and accepts <see cref="IChatDataProvider.SendMessageAsync"/> calls keyed by
+/// <see cref="SessionKey"/>.
+/// </param>
+public sealed record ChatComposeTarget(string? SessionKey, bool IsReady)
+{
+    public static ChatComposeTarget NotReady { get; } = new(null, false);
+}
 
 public sealed class ChatDataChangedEventArgs(ChatDataSnapshot snapshot) : EventArgs
 {
@@ -157,7 +180,10 @@ public interface IChatDataProvider : IAsyncDisposable
     event EventHandler<ChatProviderNotificationEventArgs>? NotificationRequested;
 
     Task<ChatDataSnapshot> LoadAsync(CancellationToken cancellationToken = default);
-    Task<ChatThread> CreateThreadAsync(string? initialMessage = null, CancellationToken cancellationToken = default);
+    // Note: there is intentionally no CreateThreadAsync. The gateway protocol
+    // has no "create new session" RPC; the canonical send target is exposed via
+    // ChatDataSnapshot.ComposeTarget and the first SendMessageAsync against it
+    // implicitly materializes the session on the server.
     Task SendMessageAsync(string threadId, string message, CancellationToken cancellationToken = default);
     Task SendMessageAsync(string threadId, string message, CancellationToken cancellationToken, IReadOnlyList<OpenClaw.Shared.ChatAttachment>? attachments) =>
         SendMessageAsync(threadId, message, cancellationToken);

src/OpenClaw.Connection/GatewayConnectionManager.cs

@@ -239,6 +239,22 @@ private async Task ConnectCoreAsync(string? gatewayId = null)
                 _gatewayNeedsV2Signature = true;
             };
 
+            // Operator-side auto-approve: the node-side WindowsNodeClient
+            // only sees its own pairing state. But when the gateway broadcasts
+            // a node.pair.requested event, the OPERATOR client gets a fresh
+            // NodePairListUpdated push. If our own node deviceId shows up in
+            // that pending list and we have approval scopes, approve it
+            // immediately — otherwise the user is stuck looking at a connected
+            // node with empty capabilities until they manually approve.
+            // This complements (not replaces) the node-side flow on line ~810
+            // which still handles the case where the node knows it's pending
+            // before any operator event lands.
+            lifecycle.DataClient.NodePairListUpdated += (s, info) =>
+            {
+                if (Interlocked.Read(ref _generation) != gen) return;
+                _ = TryOperatorAutoApproveOwnNodePairAsync(info, gen);
+            };
+
             // If we already know this gateway needs v2, tell the client upfront
             if (_gatewayNeedsV2Signature)
                 lifecycle.DataClient.UseV2Signature = true;
@@ -841,17 +857,23 @@ private async void OnNodePairingStatusChanged(object? sender, PairingStatusEvent
             _transitionSemaphore.Release();
         }
 
-        // Auto-approve node pairing if operator has admin/pairing scope
+        // Auto-approve node pairing if operator has admin/pairing scope.
+        // _autoApproveInFlight is a CAS guard scoped to JUST the approve RPC —
+        // we release it before the reconnect delay so unrelated approvals
+        // (different requestIds) aren't starved while we wait for the gateway
+        // and node-reconnect handshake to settle (which can take 5–30s on
+        // first connect via WSL cold-start).
         if (e.Status == PairingStatus.Pending && !string.IsNullOrWhiteSpace(e.RequestId)
             && e.RequestId != _lastAutoApprovedRequestId)
         {
-            // Atomic guard: only one approval in-flight at a time.
-            // If another approval is already running, skip this one entirely.
             if (Interlocked.CompareExchange(ref _autoApproveInFlight, e.RequestId, null) != null)
             {
                 return;
             }
 
+            var approvalGeneration = Interlocked.Read(ref _generation);
+            bool attemptedApprove = false;
+            bool approved = false;
             try
             {
                 var operatorClient = _activeLifecycle?.DataClient;
@@ -865,18 +887,10 @@ private async void OnNodePairingStatusChanged(object? sender, PairingStatusEvent
                         _diagnostics.Record("node", $"Auto-approving node pairing (requestId={e.RequestId})");
                         try
                         {
-                            var approved = await operatorClient.NodePairApproveAsync(e.RequestId);
-                            if (approved)
-                            {
-                                _lastAutoApprovedRequestId = e.RequestId;
-                                _diagnostics.Record("node", "Node pairing auto-approved — reconnecting node");
-                                await Task.Delay(1000); // brief delay for gateway to process
-                                await StartNodeConnectionAsync();
-                            }
-                            else
-                            {
+                            attemptedApprove = true;
+                            approved = await operatorClient.NodePairApproveAsync(e.RequestId);
+                            if (!approved)
                                 _diagnostics.Record("node", "Node auto-approval failed");
-                            }
                         }
                         catch (Exception ex)
                         {
@@ -888,8 +902,117 @@ private async void OnNodePairingStatusChanged(object? sender, PairingStatusEvent
             }
             finally
             {
+                // Only dedupe after an actual approve attempt. If the operator
+                // client was disconnected or lacked scope, the operator-side
+                // NodePairListUpdated path must still be able to approve this
+                // same requestId once the operator is ready.
+                if (attemptedApprove && Interlocked.Read(ref _generation) == approvalGeneration)
+                    _lastAutoApprovedRequestId = e.RequestId;
+                Interlocked.Exchange(ref _autoApproveInFlight, null);
+            }
+
+            // Post-approve reconnect happens OUTSIDE the CAS guard so it
+            // doesn't block unrelated approvals.
+            if (approved)
+            {
+                _diagnostics.Record("node", "Node pairing auto-approved — reconnecting node");
+                await Task.Delay(1000); // brief delay for gateway to process
+                if (Interlocked.Read(ref _generation) == approvalGeneration)
+                    await StartNodeConnectionAsync();
+            }
+        }
+    }
+
+    /// <summary>
+    /// Operator-side auto-approve. When the gateway pushes
+    /// <see cref="OpenClawGatewayClient.NodePairListUpdated"/> and there is a
+    /// pending entry for our OWN node's deviceId, approve it. The node-side
+    /// auto-approve at <see cref="OnNodePairingStatusChanged"/> handles the
+    /// case where the node already knows it is pending; this method handles
+    /// the case where the node is device-paired (its WindowsNodeClient sees
+    /// itself as Paired) but its node-sub-pairing hasn't been approved yet —
+    /// the only signal for that case is the operator-side broadcast.
+    /// </summary>
+    private async Task TryOperatorAutoApproveOwnNodePairAsync(PairingListInfo? info, long gen)
+    {
+        if (info?.Pending == null || info.Pending.Count == 0) return;
+
+        var ownNodeId = _nodeConnector?.NodeDeviceId;
+        if (string.IsNullOrWhiteSpace(ownNodeId)) return;
+
+        var operatorClient = _activeLifecycle?.DataClient;
+        if (operatorClient?.IsConnectedToGateway != true) return;
+        if (!OperatorScopeHelper.CanApproveDevices(operatorClient.GrantedOperatorScopes)) return;
+
+        // Track whether ANY approve succeeded so we know to schedule one
+        // reconnect at the end (rather than reconnecting per-entry, which
+        // would race with itself).
+        string? lastApprovedRequestId = null;
+
+        foreach (var req in info.Pending)
+        {
+            if (Interlocked.Read(ref _generation) != gen) return;
+            if (string.IsNullOrWhiteSpace(req.RequestId)) continue;
+            if (req.RequestId == _lastAutoApprovedRequestId) continue;
+            if (!string.Equals(req.NodeId, ownNodeId, StringComparison.OrdinalIgnoreCase)) continue;
+
+            // CAS guard scoped to JUST the approve RPC. Release before the
+            // post-approve reconnect so unrelated approvals are not starved
+            // (e.g. another own-node pending with a different requestId in
+            // the same or next snapshot).
+            if (Interlocked.CompareExchange(ref _autoApproveInFlight, req.RequestId, null) != null)
+                continue;
+
+            bool approved = false;
+            try
+            {
+                _diagnostics.Record("node", $"Operator-side auto-approving own node pairing (requestId={req.RequestId})");
+                try
+                {
+                    approved = await operatorClient.NodePairApproveAsync(req.RequestId);
+                    if (!approved)
+                        _diagnostics.Record("node", "Operator-side node pair approval rejected by gateway");
+                }
+                catch (Exception ex)
+                {
+                    _logger.Warn($"[ConnMgr] Operator-side node auto-approve failed: {ex.Message}");
+                    _diagnostics.Record("node", $"Operator-side auto-approve error: {ex.Message}");
+                }
+            }
+            finally
+            {
+                // Always record the requestId — both on success (prevent
+                // re-approving the same id after the gateway re-broadcasts)
+                // and on failure (prevent a spin loop on a rejected id).
+                // Re-check generation: if a reconnect happened during the
+                // await above, DisposeActiveClient already cleared
+                // _lastAutoApprovedRequestId for the new generation; we must
+                // not overwrite that null with a stale id from the old gen.
+                if (Interlocked.Read(ref _generation) == gen)
+                    _lastAutoApprovedRequestId = req.RequestId;
                 Interlocked.Exchange(ref _autoApproveInFlight, null);
             }
+
+            if (approved)
+            {
+                lastApprovedRequestId = req.RequestId;
+                // Continue scanning so a second own-pending in the same
+                // snapshot (e.g. stale requestId from prior session) also
+                // gets attempted — broken only by an explicit failure to
+                // approve, which we still try the next entry for.
+            }
+            // On failure/rejection, fall through to next own-pending entry
+            // rather than break — the gateway may not re-broadcast if the
+            // approve frame round-tripped and was rejected mid-flight.
+        }
+
+        // Single reconnect after the snapshot is fully processed.
+        if (lastApprovedRequestId is not null)
+        {
+            _diagnostics.Record("node", $"Operator-side approved {lastApprovedRequestId} — reconnecting node so caps propagate");
+            await Task.Delay(1000);
+            if (Interlocked.Read(ref _generation) == gen)
+                await StartNodeConnectionAsync();
         }
     }
 

src/OpenClaw.Shared/IOperatorGatewayClient.cs

@@ -41,6 +41,10 @@ public interface IOperatorGatewayClient
     string? OperatorDeviceId { get; }
     IReadOnlyList<string> GrantedOperatorScopes { get; }
     bool IsConnectedToGateway { get; }
+    /// <summary>Canonical main session key resolved from hello-ok; <c>null</c> until handshake.</summary>
+    string? MainSessionKey { get; }
+    /// <summary>True once the hello-ok handshake has been processed.</summary>
+    bool HasHandshakeSnapshot { get; }
 
     // ─── Connection events (from WebSocketClientBase) ───
     event EventHandler<ConnectionStatus>? StatusChanged;

src/OpenClaw.Shared/OpenClaw.Shared.csproj

@@ -9,6 +9,7 @@
 
   <ItemGroup>
     <InternalsVisibleTo Include="OpenClaw.Shared.Tests" />
+    <InternalsVisibleTo Include="OpenClaw.Connection.Tests" />
   </ItemGroup>
 
   <ItemGroup>