hermes - 💡(How to fix) Fix [Feature] Option to queue incoming messages instead of interrupting running tasks [1 comments, 1 participants]

Description

Currently, when the Hermes gateway agent is executing a task and the user sends a new message, the gateway immediately interrupts the running agent (aborting in-flight tool calls) and responds with "⚡ Interrupting current task (X min elapsed, iteration N/90, running: terminal). I'll respond to your message shortly."

This interrupt-then-restart pattern is destructive:

• Long-running tasks (builds, deployments, multi-step operations) lose all progress
• The agent re-processes the new message from scratch, often re-executing the same expensive commands
• Creates a frustrating loop where the user sees repeated interruptions with no actual progress

Requested Behavior

Add a configurable option (per platform or global) to queue incoming messages instead of interrupting:

gateway:
  message_queue_mode: queue  # "interrupt" (current default) | "queue"
  # or per-platform:
  feishu:
    message_queue_mode: queue

When queue mode is enabled:

1. New messages are silently queued (with a brief acknowledgment like "📋 Queued (position 1), I'll get to it after finishing the current task")
2. The agent continues executing its current task uninterrupted
3. Once the current task completes (or the agent loop exits), the queued messages are dequeued and processed in order
4. No tool calls are aborted, no progress is lost

Use Case

Messaging platform users (Feishu, Telegram, Discord) frequently send follow-up messages while the agent is working — corrections, additional context, or unrelated questions. In most cases, these do not require immediate interruption of the current task. The user just wants the agent to handle them when it's done.

The current interrupt-only behavior is particularly harmful for:

• Terminal commands with background=true or long timeouts
• Multi-step workflows (build -> test -> deploy)
• Tasks that have already consumed significant API context/tokens

Environment

• Hermes Agent v0.9.x (gateway mode)
• Platform: Feishu/Lark (but affects all messaging platforms)
• Related: #11631 (no feedback during long tasks)

Suggested Implementation

The gateway already has _pending_messages infrastructure via merge_pending_message_event(). The change would be:

1. Add a config flag gateway.message_queue_mode (default: "interrupt" for backward compatibility)
2. In _handle_busy_session(): instead of calling running_agent.interrupt(), just store the event in the pending queue
3. Send a lightweight queued acknowledgment instead of the interrupt message
4. After run_agent() completes for the current task, check the pending queue and start a new run with the queued messages