Fix WhatsApp image delivery and Chat duplicate attachment display.

Enqueue deliverable tool attachments on the WhatsApp queue path, and dedupe the same attachment_id within one Chat bubble (prefer deliverable).

Co-authored-by: Cursor <cursoragent@cursor.com>

interfaces/admin/static/chat.js

@@ -557,14 +557,70 @@ function _foldProcessTextRedundant(processText, collapsedItems) {
   return false;
 }
 
+/**
+ * Prefer a single render of each attachment_id inside one aggregated bubble.
+ * When both generate + save_deliverable_attachment persist the same id, keep the
+ * deliverable row (or the first occurrence if none is marked deliverable).
+ */
+function _preferredAttachmentOwnerById(items) {
+  const preferred = new Map();
+  (Array.isArray(items) ? items : []).forEach((it, itemIdx) => {
+    const list = parseAttachments(it && it.attachments);
+    for (const att of list) {
+      if (!att || typeof att !== "object") continue;
+      const aid = String(att.attachment_id || att.attachmentId || "")
+        .trim()
+        .toLowerCase();
+      if (!aid) continue;
+      const prev = preferred.get(aid);
+      if (prev === undefined) {
+        preferred.set(aid, itemIdx);
+        continue;
+      }
+      if (att.deliverable === true) {
+        const prevAtts = parseAttachments((items[prev] && items[prev].attachments) || null);
+        const prevAtt = prevAtts.find((a) => {
+          const id = String((a && (a.attachment_id || a.attachmentId)) || "")
+            .trim()
+            .toLowerCase();
+          return id === aid;
+        });
+        if (!(prevAtt && prevAtt.deliverable === true)) preferred.set(aid, itemIdx);
+      }
+    }
+  });
+  return preferred;
+}
+
+function _attachmentsForBubbleItem(raw, itemIdx, preferredById) {
+  const list = parseAttachments(raw);
+  if (!list.length) return null;
+  const kept = [];
+  for (const att of list) {
+    if (!att || typeof att !== "object") continue;
+    const aid = String(att.attachment_id || att.attachmentId || "")
+      .trim()
+      .toLowerCase();
+    if (aid && preferredById instanceof Map) {
+      const owner = preferredById.get(aid);
+      if (owner != null && owner !== itemIdx) continue;
+    }
+    kept.push(att);
+  }
+  return kept.length ? kept : null;
+}
+
 async function _buildAggregatedAssistantBubble(tsIso, items) {
   const inner = el("div", { class: "chat-msg chat-msg--assistant chat-msg--rich" });
   const collapsedItems = [];
-  for (const it of items || []) {
+  const preferredAttOwner = _preferredAttachmentOwnerById(items);
+  const itemList = Array.isArray(items) ? items : [];
+  for (let itemIdx = 0; itemIdx < itemList.length; itemIdx++) {
+    const it = itemList[itemIdx];
     const kind = String((it && it.kind) || "");
     const text = String((it && it.text) || "");
-    const parsedAtts = parseAttachments(it && it.attachments);
-    const hasInlineAtt = parsedAtts.length > 0;
+    const filteredAtts = _attachmentsForBubbleItem(it && it.attachments, itemIdx, preferredAttOwner);
+    const hasInlineAtt = Array.isArray(filteredAtts) && filteredAtts.length > 0;
     if (
       !text.trim() &&
       !(kind === "tool_result" && hasInlineAtt) &&
@@ -582,11 +638,11 @@ async function _buildAggregatedAssistantBubble(tsIso, items) {
             collapsedItems.push({ title: t("reasoning.processNotes"), text: t0 });
           }
         }
-        const attsAssistant = await renderAttachmentsEl(it && it.attachments);
+        const attsAssistant = await renderAttachmentsEl(filteredAtts);
         if (attsAssistant) inner.appendChild(attsAssistant);
       } else {
         _appendAssistantTextSegments(inner, text, collapsedItems);
-        const attsAssistant = await renderAttachmentsEl(it && it.attachments);
+        const attsAssistant = await renderAttachmentsEl(filteredAtts);
         if (attsAssistant) inner.appendChild(attsAssistant);
       }
     } else if (kind === "reasoning") {
@@ -597,7 +653,8 @@ async function _buildAggregatedAssistantBubble(tsIso, items) {
       if (adminChatShowToolOutput) collapsedItems.push({ title: _toolSummaryTitle("tool"), text });
       // 支持两类附件：base64（image/input_image）和引用型（image_ref）。
       // image_ref 需要异步拉取 blob，因此改为走 renderAttachmentsEl()。
-      const attsEl = await renderAttachmentsEl(it && it.attachments);
+      // Same attachment_id from generate + save_deliverable is shown once (prefer deliverable).
+      const attsEl = await renderAttachmentsEl(filteredAtts);
       if (attsEl) inner.appendChild(attsEl);
     } else {
       inner.appendChild(el("div", { class: "chat-msg__md", html: renderMarkdownHtml(text) }));

runtime/application/gateway/inbound_service.py

@@ -1454,11 +1454,19 @@ def process_inbound_payload(payload: dict[str, Any]) -> dict[str, Any]:
                                         )
                                         channel_turn_uuid = str(turn_result.turn_uuid or "").strip()
                                         turn_reply = str(turn_result.reply_text or "").strip()
+                                        # Prefer attachments on the matching assistant text row.
+                                        # Fallback: deliverable tool media from this user turn
+                                        # (e.g. save_deliverable_attachment / image generate).
                                         turn_atts = _collect_reply_attachments_from_history(
                                             store=store,
                                             session_id=str(session_id),
                                             reply_text=turn_reply,
                                         )
+                                        if not turn_atts:
+                                            turn_atts = _collect_recent_tool_attachments(
+                                                store=store,
+                                                session_id=str(session_id),
+                                            )
                                     except Exception as e:
                                         turn_reply = f"抱歉，处理消息时出错：{type(e).__name__}: {e}"
                                         turn_atts = []

tests/test_whatsapp_inbound_queue_cancel.py

@@ -121,6 +121,63 @@ class WhatsappInboundSerialQueueTests(unittest.TestCase):
         self.assertEqual(len(pending), 1)
         self.assertEqual(pending[0].get("text"), "only once")
 
+    def test_whatsapp_queue_includes_deliverable_tool_image(self) -> None:
+        """Regression (session 8d2b…): image on tool row must still enqueue to WhatsApp.
+
+        Final assistant_text often has no attachments; deliverable lives on
+        ``save_deliverable_attachment`` tool_result. Queue path used to skip it.
+        """
+        reply_text = "done, here is a cute puppy"
+
+        def _handle_turn(**kwargs):
+            msg = kwargs.get("msg")
+            sid = str(getattr(msg, "session_id", "") or "")
+            self.store.add_message(sid, "user", "draw a puppy", event_type="user_text")
+            self.store.add_message(
+                sid,
+                "tool",
+                json.dumps({"ok": True, "attachment_id": "img-puppy", "deliverable": True}),
+                attachments=[
+                    {
+                        "type": "image_ref",
+                        "attachment_id": "img-puppy",
+                        "name": "cute_puppy.png",
+                        "mime": "image/png",
+                        "deliverable": True,
+                    }
+                ],
+                event_type="tool_result",
+            )
+            # Final assistant body intentionally has no attachments (real gateway behavior).
+            self.store.add_message(sid, "assistant", reply_text, event_type="assistant_text")
+            return _FakeTurn(reply_text)
+
+        with self._patch_common(), mock.patch("runtime.gateway.OclawGateway") as gw_cls, mock.patch(
+            "runtime.orchestration.group_ingest.should_process_group_inbound", return_value=True
+        ), mock.patch(
+            "runtime.application.gateway.whatsapp_inbound_access.handle_whatsapp_access",
+            return_value=None,
+        ), mock.patch.object(
+            inbound_mod,
+            "_maybe_expand_reply_attachments_for_channel",
+            # Keep image_ref as-is so we can assert attachment_id without a real asset blob.
+            side_effect=lambda reply: None,
+        ):
+            gw = gw_cls.return_value
+            gw.handle_turn.side_effect = lambda **kw: _handle_turn(**kw)
+            out = inbound_mod.process_inbound_payload(self._payload("draw a puppy", stanza="img1"))
+
+        self.assertEqual(out.get("delivery"), "queued")
+        pending = self.store.list_pending_channel_outbound_messages(
+            channel="whatsapp", account_id="wa-default", limit=5
+        )
+        self.assertEqual(len(pending), 1)
+        self.assertEqual(pending[0].get("text"), reply_text)
+        atts = pending[0].get("attachments") or []
+        self.assertTrue(atts, "outbound must include deliverable image attachment")
+        self.assertEqual(atts[0].get("attachment_id"), "img-puppy")
+        self.assertTrue(atts[0].get("deliverable") is True)
+
     def test_busy_inbound_is_accepted_queued_then_merged(self) -> None:
         release = threading.Event()
         seen_texts: list[str] = []