[Feature]: Allow agent response opt out

agent/prompt_builder.py

@@ -178,6 +178,15 @@ def _strip_yaml_frontmatter(content: str) -> str:
     "Skills that aren't maintained become liabilities."
 )
 
+SILENT_RESPONSE_GUIDANCE = (
+    "If you have nothing meaningful to add in a conversation, you can suppress delivery "
+    "by responding with the exact string '[SILENT]'. This must be the ONLY content "
+    "of your response; any additional characters, whitespace, emojis, or commentary "
+    "will cause the message to leak and be delivered to the user. The transcript entry "
+    "is preserved for continuity. This works for the gateway's final_response "
+    "delivery path — direct send_message tool calls are not affected."
+)
+
 KANBAN_GUIDANCE = (
     "# Kanban task execution protocol\n"
     "You have been assigned ONE task from "

agent/system_prompt.py

@@ -35,6 +35,7 @@
     OPENAI_MODEL_EXECUTION_GUIDANCE,
     PLATFORM_HINTS,
     SESSION_SEARCH_GUIDANCE,
+    SILENT_RESPONSE_GUIDANCE,
     SKILLS_GUIDANCE,
     STEER_CHANNEL_NOTE,
     TASK_COMPLETION_GUIDANCE,
@@ -102,6 +103,9 @@ def build_system_prompt_parts(agent: Any, system_message: Optional[str] = None)
     # Pointer to the hermes-agent skill + docs for user questions about Hermes itself.
     stable_parts.append(HERMES_AGENT_HELP_GUIDANCE)
 
+    # [SILENT] response marker — always available, regardless of tools/platform.
+    stable_parts.append(SILENT_RESPONSE_GUIDANCE)
+
     # Universal task-completion / no-fabrication guidance.  Applied to ALL
     # models regardless of tool_use_enforcement gating — the failure modes
     # this targets (stopping after a stub; fabricating output when a real

gateway/platforms/base.py

@@ -24,6 +24,12 @@
 
 logger = logging.getLogger(__name__)
 
+# Sentinel: when the agent's response is exactly this marker, delivery is
+# suppressed. The agent can use this to opt out of replying (e.g. in group
+# channels when there's nothing meaningful to add). Output is still persisted
+# in the transcript for continuity.
+SILENT_MARKER = "[SILENT]"
+
 # Audio file extensions Hermes recognizes for native audio delivery.
 # Kept in sync with tools/send_message_tool.py and cron/scheduler.py via
 # should_send_media_as_audio() below.
@@ -4094,6 +4100,15 @@ async def _stop_typing_task() -> None:
             # post-send block can schedule the deletion.
             response, _ephemeral_ttl = self._unwrap_ephemeral(response)
 
+            # [SILENT] marker: agent opted out of replying. Suppress delivery
+            # but keep the transcript entry for continuity.
+            if response and response.strip() == SILENT_MARKER:
+                logger.info(
+                    "[%s] Agent returned %s — suppressing delivery for %s",
+                    self.name, SILENT_MARKER, event.source.chat_id,
+                )
+                response = None
+
             # Send response if any.  A None/empty response is normal when
             # streaming already delivered the text (already_sent=True) or
             # when the message was queued behind an active agent.  Log at

gateway/run.py

@@ -1168,6 +1168,7 @@ def _reload_runtime_env_preserving_config_authority() -> None:
     EphemeralReply,
     MessageEvent,
     MessageType,
+    SILENT_MARKER,
     _reply_anchor_for_event,
     merge_pending_message_event,
 )
@@ -14883,18 +14884,27 @@ async def _notify_long_running():
                     )
                     first_response = result.get("final_response", "")
                     if first_response and not _already_streamed:
-                        try:
+                        # Skip resend if the response is [SILENT] — it was
+                        # intentionally suppressed and shouldn't leak here.
+                        if first_response.strip() == SILENT_MARKER:
                             logger.info(
-                                "Queued follow-up for session %s: final stream delivery not confirmed; sending first response before continuing.",
+                                "Queued follow-up for session %s: first response is %s, skipping resend.",
                                 session_key or "?",
+                                SILENT_MARKER,
                             )
-                            await adapter.send(
-                                source.chat_id,
-                                first_response,
-                                metadata=_status_thread_metadata,
-                            )
-                        except Exception as e:
-                            logger.warning("Failed to send first response before queued message: %s", e)
+                        else:
+                            try:
+                                logger.info(
+                                    "Queued follow-up for session %s: final stream delivery not confirmed; sending first response before continuing.",
+                                    session_key or "?",
+                                )
+                                await adapter.send(
+                                    source.chat_id,
+                                    first_response,
+                                    metadata=_status_thread_metadata,
+                                )
+                            except Exception as e:
+                                logger.warning("Failed to send first response before queued message: %s", e)
                     elif first_response:
                         logger.info(
                             "Queued follow-up for session %s: skipping resend because final streamed delivery was confirmed.",

hermes_cli/tips.py

@@ -192,6 +192,7 @@
     "Voice messages on Telegram, Discord, WhatsApp, and Slack are auto-transcribed.",
 
     # --- Gateway & Messaging ---
+    "If the agent's response is exactly [SILENT], delivery is suppressed — useful in group channels when there's nothing to add.",
     "Hermes runs on 21 messaging platforms: Telegram, Discord, Slack, WhatsApp, Signal, Matrix, IRC, Microsoft Teams, email, and more.",
     "hermes gateway install sets it up as a system service that starts on boot.",
     "DingTalk uses Stream Mode — no webhooks or public URL needed.",