hermes - 💡(How to fix) Fix Feature Request: Context-aware task switching for WeChat/messaging platforms

Feature Request: Context-aware task switching for messaging platforms

Problem

Currently on WeChat (and other messaging platforms), when a task is running and the user sends a new message, the behavior is:

1. If agent is busy — message queues, waits for current turn to finish
2. If agent is idle — new message interrupts immediately

This creates a pain point: users often want a middle ground:

• Don't interrupt when I'm just adding info or making a casual comment
• But naturally switch when I'm trying to redirect/correct the agent

The current workaround (/stop + new task) works but feels forced and unnatural.

Desired Behavior

A "context-aware" task switching behavior:

The key insight: the agent should recognize when a user's message indicates a context shift (e.g., "stop that", "wait, wrong approach"), and respond to that naturally rather than blindly queuing behind the current task.

Why This Matters

• Messaging is inherently conversational — users expect natural flow
• Explicit /stop breaks the conversation rhythm
• Current behavior treats all messages the same regardless of intent
• Makes the agent feel less like a smart assistant and more like a rigid command queue

Implementation Approaches

Option A: Intent detection in preprocessing Add a lightweight intent classifier in the message preprocessing layer:

• "supplementary" messages → queue
• "redirect/correct/stop/abandon" signals → interrupt current turn
• Hard keywords: "stop", "wait", "actually", "no", "wrong", "instead", "forget that" Option B: Explicit redirect marker Add a /redirect command or #redirect tag that user can include:
• More explicit, predictable behavior
• But still requires user to know about the command Option C: Priority queue with context awareness Keep a queue but add a priority field:
• Normal messages: priority = 1 (queue)
• Redirect signals: priority = 10 (preempt)
• Agent checks queue priority before each tool call

Constraints / Considerations

• Should not break existing /stop behavior (that's the explicit override)
• Must work across all messaging platforms (WeChat, Telegram, Discord, etc.)
• False positives (interrupting when user didn't want to) are worse than false negatives
• Should be configurable — some users may want aggressive interrupt, others want complete quiet