fix(gateway): flush throttled delta before emitChatFinal

visionik7 · visionik7 · commit 949e83d8c8fa · 2026-02-23T17:46:03.000-05:00
The 150ms throttle in emitChatDelta can suppress the last text chunk
before emitChatFinal fires, causing streaming clients (e.g. ACP) to
receive truncated responses. The final event carries the complete text,
but clients that build responses incrementally from deltas miss the
tail end.

Flush one last unthrottled delta with the complete buffered text
immediately before sending the final event. This ensures all streaming
consumers have the full response without needing to reconcile deltas
against the final payload.
diff --git a/src/gateway/server-chat.ts b/src/gateway/server-chat.ts
@@ -335,6 +335,27 @@ export function createAgentEventHandler({
     const text = normalizedHeartbeatText.text.trim();
     const shouldSuppressSilent =
       normalizedHeartbeatText.suppress || isSilentReplyText(text, SILENT_REPLY_TOKEN);
+    // Flush any throttled delta so streaming clients receive the complete text
+    // before the final event.  The 150 ms throttle in emitChatDelta may have
+    // suppressed the most recent chunk, leaving the client with stale text.
+    if (text && !shouldSuppressSilent) {
+      const lastSent = chatRunState.deltaSentAt.get(clientRunId) ?? 0;
+      if (lastSent > 0) {
+        const flushPayload = {
+          runId: clientRunId,
+          sessionKey,
+          seq,
+          state: "delta" as const,
+          message: {
+            role: "assistant",
+            content: [{ type: "text", text }],
+            timestamp: Date.now(),
+          },
+        };
+        broadcast("chat", flushPayload, { dropIfSlow: true });
+        nodeSendToSession(sessionKey, "chat", flushPayload);
+      }
+    }
     chatRunState.buffers.delete(clientRunId);
     chatRunState.deltaSentAt.delete(clientRunId);
     if (jobState === "done") {
