diff --git a/dist/fabric/src/inbound.js b/dist/fabric/src/inbound.js
index e163985..317bd73 100644
--- a/dist/fabric/src/inbound.js
+++ b/dist/fabric/src/inbound.js
@@ -233,8 +233,21 @@ export class FabricInbound {
                 },
                 onRecordError: (err) => this.log.warn(`fabric: session record failed agent=${agentId}: ${String(err)}`),
                 onDispatchError: (err, info) => this.log.warn(`fabric: ${info.kind} dispatch failed agent=${agentId}: ${String(err)}`),
-                // Fabric has no length limit: deliver the whole reply as ONE message.
-                replyOptions: { disableBlockStreaming: true },
+                // - disableBlockStreaming: Fabric has no length limit, deliver the
+                //   whole reply as ONE message.
+                // - sourceReplyDeliveryMode 'automatic': OpenClaw defaults group
+                //   chats to "message_tool_only", which SUPPRESSES auto-delivery of
+                //   the agent's text reply (it expects the agent to call a message
+                //   tool). Fabric already gates *when* an agent speaks via the
+                //   per-recipient wakeup flag, so once a turn is dispatched the
+                //   reply must always flow back through `deliver`. Forcing
+                //   'automatic' overrides the group default so the reply is
+                //   delivered. (source-reply-delivery-mode: a truthy `requested`
+                //   wins unless it's message_tool_only with no tool available.)
+                replyOptions: {
+                    disableBlockStreaming: true,
+                    sourceReplyDeliveryMode: 'automatic',
+                },
             });
             this.log.info(`fabric: dispatch returned agent=${agentId} channel=${channelId}`);
         }
diff --git a/src/inbound.ts b/src/inbound.ts
index 3f5fc18..0d82a57 100644
--- a/src/inbound.ts
+++ b/src/inbound.ts
@@ -279,8 +279,21 @@ export class FabricInbound {
           this.log.warn(`fabric: session record failed agent=${agentId}: ${String(err)}`),
         onDispatchError: (err: unknown, info: { kind: string }) =>
           this.log.warn(`fabric: ${info.kind} dispatch failed agent=${agentId}: ${String(err)}`),
-        // Fabric has no length limit: deliver the whole reply as ONE message.
-        replyOptions: { disableBlockStreaming: true } as never,
+        // - disableBlockStreaming: Fabric has no length limit, deliver the
+        //   whole reply as ONE message.
+        // - sourceReplyDeliveryMode 'automatic': OpenClaw defaults group
+        //   chats to "message_tool_only", which SUPPRESSES auto-delivery of
+        //   the agent's text reply (it expects the agent to call a message
+        //   tool). Fabric already gates *when* an agent speaks via the
+        //   per-recipient wakeup flag, so once a turn is dispatched the
+        //   reply must always flow back through `deliver`. Forcing
+        //   'automatic' overrides the group default so the reply is
+        //   delivered. (source-reply-delivery-mode: a truthy `requested`
+        //   wins unless it's message_tool_only with no tool available.)
+        replyOptions: {
+          disableBlockStreaming: true,
+          sourceReplyDeliveryMode: 'automatic',
+        } as never,
       });
       this.log.info(`fabric: dispatch returned agent=${agentId} channel=${channelId}`);
     } catch (err) {