fix(gateway): implement proper OpenClaw WebSocket handshake protocol

- Send JSON-RPC connect request after WebSocket opens (required by OpenClaw Gateway) - Use OpenClaw protocol format: { type: "req" } instead of { jsonrpc: "2.0" } - Include proper ConnectParams: client info, auth token, protocol version - Handle OpenClaw response format: { type: "res", ok: true/false } - Handle OpenClaw event format: { type: "event", event: "..." } - Wait for handshake completion before marking connection as running - Improve error handling for connection failures The Gateway was rejecting connections because: 1. ClawX wasn't sending the required "connect" handshake message 2. The protocol format was incorrect (standard JSON-RPC vs OpenClaw format)
2026-02-06 02:17:43 +08:00
parent 931fea3735
commit 0f8e6f3f9e
3 changed files with 224 additions and 17 deletions
--- a/build_process/commit_15_gateway_ws_handshake.md
+++ b/build_process/commit_15_gateway_ws_handshake.md
@@ -0,0 +1,92 @@
+# Commit 15: Fix Gateway WebSocket Handshake
+
+## Summary
+
+Fixed the Gateway WebSocket connection instability by implementing the proper OpenClaw WebSocket handshake protocol.
+
+## Problem
+
+The Gateway connection was failing with errors like:
+- `token_mismatch` - Authentication failing despite correct tokens
+- `invalid handshake: first request must be connect` - Missing handshake
+- Repeated connect/disconnect cycles
+
+Root causes:
+1. ClawX was not sending the required `connect` handshake message after WebSocket opens
+2. Using standard JSON-RPC 2.0 format instead of OpenClaw's custom protocol format
+3. Config file had hardcoded token overriding CLI arguments
+
+## Solution
+
+### 1. Implement Proper Connect Handshake
+
+After WebSocket opens, send a proper connect request:
+
+```typescript
+const connectFrame = {
+  type: 'req',  // OpenClaw uses 'req' not 'jsonrpc: 2.0'
+  id: connectId,
+  method: 'connect',
+  params: {
+    minProtocol: 3,
+    maxProtocol: 3,
+    client: {
+      id: 'gateway-client',
+      displayName: 'ClawX',
+      version: '0.1.0',
+      platform: process.platform,
+      mode: 'ui',
+    },
+    auth: {
+      token: gatewayToken,
+    },
+    caps: [],
+    role: 'operator',
+    scopes: [],
+  },
+};
+```
+
+### 2. Handle OpenClaw Protocol Format
+
+OpenClaw Gateway uses a custom protocol format:
+- Request: `{ type: "req", id, method, params }`
+- Response: `{ type: "res", id, ok, payload, error }`
+- Event: `{ type: "event", event, payload }`
+
+Updated `handleMessage` to parse these formats correctly.
+
+### 3. Remove Hardcoded Config Token
+
+The `~/.openclaw/openclaw.json` file had a hardcoded token that was overriding the CLI token. Updated to remove the auth.token field so the environment variable takes precedence.
+
+## Files Changed
+
+### `electron/gateway/manager.ts`
+- **connect()**: Added proper handshake flow with connect request
+- **handleMessage()**: Parse OpenClaw protocol response/event formats
+- **handleProtocolEvent()**: New method to handle OpenClaw events
+- **rpc()**: Use OpenClaw request format `{ type: "req" }`
+
+## Protocol Flow
+
+```
+Before (broken):
+  1. Open WebSocket
+  2. Immediately mark as "running" ❌
+  3. Send RPC requests (rejected - no handshake)
+
+After (fixed):
+  1. Open WebSocket
+  2. Send connect handshake with auth token
+  3. Wait for response { type: "res", ok: true }
+  4. Mark as "running" ✓
+  5. Send RPC requests (accepted)
+```
+
+## Testing
+
+After this fix:
+- Gateway connects successfully
+- WebSocket stays connected without constant reconnection
+- RPC calls work correctly
--- a/build_process/process.md
+++ b/build_process/process.md
@@ -20,6 +20,7 @@
 * [commit_12] Real API key validation - OpenRouter support, actual API calls to verify keys
 * [commit_13] Remove channel setup step - Simplified onboarding, channels moved to Settings
 * [commit_14] Auto-install skills step - Replace skill selection with auto-installation progress UI
+* [commit_15] Fix Gateway WebSocket handshake - Implement proper OpenClaw protocol

 ### Plan:
 1. ~~Initialize project structure~~ ✅
@@ -50,6 +51,7 @@ All core features have been implemented:
 - Real API key validation via actual API calls (Anthropic, OpenAI, Google, OpenRouter)
 - Simplified setup wizard (channel connection deferred to Settings page)
 - Auto-install skills step with real-time progress UI (no manual skill selection)
+- Fixed Gateway WebSocket connection with proper OpenClaw handshake protocol

 ## Version Milestones

--- a/electron/gateway/manager.ts
+++ b/electron/gateway/manager.ts
@@ -208,6 +208,7 @@ export class GatewayManager extends EventEmitter {
  
  /**
   * Make an RPC call to the Gateway
+   * Uses OpenClaw protocol format: { type: "req", id: "...", method: "...", params: {...} }
   */
  async rpc<T>(method: string, params?: unknown, timeoutMs = 30000): Promise<T> {
    return new Promise((resolve, reject) => {
@@ -231,9 +232,9 @@ export class GatewayManager extends EventEmitter {
        timeout,
      });
      
-      // Send request
+      // Send request using OpenClaw protocol format
      const request = {
-        jsonrpc: '2.0',
+        type: 'req',
        id,
        method,
        params,
@@ -442,20 +443,73 @@ export class GatewayManager extends EventEmitter {
    const gatewayToken = await getSetting('gatewayToken');
    
    return new Promise((resolve, reject) => {
-      // Include token in WebSocket URL for authentication
-      const wsUrl = `ws://localhost:${port}/ws?auth=${encodeURIComponent(gatewayToken)}`;
+      // WebSocket URL (token will be sent in connect handshake, not URL)
+      const wsUrl = `ws://localhost:${port}/ws`;
      
      this.ws = new WebSocket(wsUrl);
+      let handshakeComplete = false;
      
-      this.ws.on('open', () => {
-        console.log('WebSocket connected to Gateway');
-        this.setStatus({
-          state: 'running',
-          port,
-          connectedAt: Date.now(),
+      this.ws.on('open', async () => {
+        console.log('WebSocket opened, sending connect handshake...');
+        
+        // Send proper connect handshake as required by OpenClaw Gateway protocol
+        // The Gateway expects: { type: "req", id: "...", method: "connect", params: ConnectParams }
+        const connectId = `connect-${Date.now()}`;
+        const connectFrame = {
+          type: 'req',
+          id: connectId,
+          method: 'connect',
+          params: {
+            minProtocol: 3,
+            maxProtocol: 3,
+            client: {
+              id: 'gateway-client',
+              displayName: 'ClawX',
+              version: '0.1.0',
+              platform: process.platform,
+              mode: 'ui',
+            },
+            auth: {
+              token: gatewayToken,
+            },
+            caps: [],
+            role: 'operator',
+            scopes: [],
+          },
+        };
+        
+        console.log('Sending connect handshake:', JSON.stringify(connectFrame));
+        this.ws?.send(JSON.stringify(connectFrame));
+        
+        // Store pending connect request
+        const connectTimeout = setTimeout(() => {
+          if (!handshakeComplete) {
+            console.error('Connect handshake timeout');
+            reject(new Error('Connect handshake timeout'));
+            this.ws?.close();
+          }
+        }, 10000);
+        
+        this.pendingRequests.set(connectId, {
+          resolve: (result) => {
+            clearTimeout(connectTimeout);
+            handshakeComplete = true;
+            console.log('WebSocket handshake complete, gateway connected');
+            this.setStatus({
+              state: 'running',
+              port,
+              connectedAt: Date.now(),
+            });
+            this.startPing();
+            resolve();
+          },
+          reject: (error) => {
+            clearTimeout(connectTimeout);
+            console.error('Connect handshake failed:', error);
+            reject(error);
+          },
+          timeout: connectTimeout,
        });
-        this.startPing();
-        resolve();
      });
      
      this.ws.on('message', (data) => {
@@ -467,8 +521,13 @@ export class GatewayManager extends EventEmitter {
        }
      });
      
-      this.ws.on('close', () => {
-        console.log('WebSocket disconnected');
+      this.ws.on('close', (code, reason) => {
+        const reasonStr = reason?.toString() || 'unknown';
+        console.log(`WebSocket disconnected: code=${code}, reason=${reasonStr}`);
+        if (!handshakeComplete) {
+          reject(new Error(`WebSocket closed before handshake: ${reasonStr}`));
+          return;
+        }
        if (this.status.state === 'running') {
          this.setStatus({ state: 'stopped' });
          this.scheduleReconnect();
@@ -477,7 +536,9 @@ export class GatewayManager extends EventEmitter {
      
      this.ws.on('error', (error) => {
        console.error('WebSocket error:', error);
-        reject(error);
+        if (!handshakeComplete) {
+          reject(error);
+        }
      });
    });
  }
@@ -486,7 +547,38 @@ export class GatewayManager extends EventEmitter {
   * Handle incoming WebSocket message
   */
  private handleMessage(message: unknown): void {
-    // Check if this is a JSON-RPC response
+    if (typeof message !== 'object' || message === null) {
+      console.warn('Received non-object message:', message);
+      return;
+    }
+    
+    const msg = message as Record<string, unknown>;
+    
+    // Handle OpenClaw protocol response format: { type: "res", id: "...", ok: true/false, ... }
+    if (msg.type === 'res' && typeof msg.id === 'string') {
+      if (this.pendingRequests.has(msg.id)) {
+        const request = this.pendingRequests.get(msg.id)!;
+        clearTimeout(request.timeout);
+        this.pendingRequests.delete(msg.id);
+        
+        if (msg.ok === false || msg.error) {
+          const errorObj = msg.error as { message?: string; code?: number } | undefined;
+          const errorMsg = errorObj?.message || JSON.stringify(msg.error) || 'Unknown error';
+          request.reject(new Error(errorMsg));
+        } else {
+          request.resolve(msg.payload ?? msg);
+        }
+        return;
+      }
+    }
+    
+    // Handle OpenClaw protocol event format: { type: "event", event: "...", payload: {...} }
+    if (msg.type === 'event' && typeof msg.event === 'string') {
+      this.handleProtocolEvent(msg.event, msg.payload);
+      return;
+    }
+    
+    // Fallback: Check if this is a JSON-RPC 2.0 response (legacy support)
    if (isResponse(message) && message.id && this.pendingRequests.has(String(message.id))) {
      const request = this.pendingRequests.get(String(message.id))!;
      clearTimeout(request.timeout);
@@ -503,7 +595,7 @@ export class GatewayManager extends EventEmitter {
      return;
    }
    
-    // Check if this is a notification (server-initiated event)
+    // Check if this is a JSON-RPC notification (server-initiated event)
    if (isNotification(message)) {
      this.handleNotification(message);
      return;
@@ -513,6 +605,27 @@ export class GatewayManager extends EventEmitter {
    this.emit('message', message);
  }
  
+  /**
+   * Handle OpenClaw protocol events
+   */
+  private handleProtocolEvent(event: string, payload: unknown): void {
+    // Map OpenClaw events to our internal event types
+    switch (event) {
+      case 'tick':
+        // Heartbeat tick, ignore
+        break;
+      case 'chat':
+        this.emit('chat:message', { message: payload });
+        break;
+      case 'channel.status':
+        this.emit('channel:status', payload as { channelId: string; status: string });
+        break;
+      default:
+        // Forward unknown events as generic notifications
+        this.emit('notification', { method: event, params: payload });
+    }
+  }
+  
  /**
   * Handle server-initiated notifications
   */