openclaw - 💡(How to fix) Fix Discord typing indicator is delayed despite typingMode=instant [1 pull requests]

Summary

After upgrading to OpenClaw 2026.5.22, Discord typing indicators appear noticeably later than before even when agents.defaults.typingMode is configured as "instant".

This is most visible in Discord channel/thread conversations: the inbound message is accepted, but the user does not see typing until after pre-run work such as session handling, memory recall, queue checks, or preflight logic has already run.

Environment

• OpenClaw: 2026.5.22
• Gateway: LaunchAgent on macOS
• Channel: Discord
• Conversation shape: Discord channel / thread-topic style conversation
• Config:

{
  "agents": {
    "defaults": {
      "typingMode": "instant"
    }
  }
}

Expected behavior

For typingMode: "instant", Discord should emit a typing indicator as soon as the inbound message is accepted for handling / queued for a run.

In other words, the UX expectation is:

1. Discord message received
2. OpenClaw accepts it as an actionable turn
3. Discord typing indicator is sent immediately
4. Session/memory/queue/preflight/model work proceeds
5. Keepalive continues while the run is active

Actual behavior

The typing indicator appears to start later, after the agent reply runner has progressed to run-start logic.

On a real 2026.5.22 gateway with typingMode: "instant", logs showed pre-run work adding visible delay before the user saw typing, including memory recall and occasional event-loop delay. Example timing observed:

• memory-tdai recall: ~0.6s to ~2.8s on some turns
• prior post-update recovery window also showed event_loop_delay and queued active runs

The important point is not memory specifically; any pre-run work before signalRunStart() makes "instant" feel non-instant from the Discord user's perspective.

Source-level observation

In the packaged 2026.5.22 dist, typingMode: "instant" resolves to immediate typing behavior, but the trigger appears tied to reply-run start rather than Discord inbound acceptance.

Relevant packaged files observed locally:

• dist/typing-mode-*.js
  • resolveTypingMode(...)
  • createTypingSignaler(...)
  • signalRunStart() calls typing.startTypingLoop() for mode === "instant"
• dist/agent-runner.runtime-*.js
  • typingSignals.signalRunStart() is called after active-run queue handling and just before preflight compaction / memory flush / model-run work
• Telegram has an early typing cue path, but Discord appears to rely on the later run-start path.

Why this matters

For human users, "instant" is interpreted as "the bot is alive and has accepted my message". If typing is delayed until after recall/preflight/queue work, it feels like the bot is unresponsive, especially after update/restart or when memory/session work is slower.

This also makes post-update smoke tests misleading: gateway status and eventual reply can be OK, while the conversation still feels stalled for several seconds.

Suggested fix

Add an early Discord typing cue at inbound acceptance / queue admission time, similar in spirit to Telegram's early typing cue.

Suggested behavior:

• If the message is actionable and not suppressed by policy, emit one Discord typing indicator immediately.
• Continue using the existing run-start typing loop as keepalive once the reply runner starts.
• Preserve existing suppressions for heartbeat/system/internal/silent paths.
• Ensure thread/topic Discord targets use the correct conversation target when sending the early typing indicator.

Notes

This is not a config issue: agents.defaults.typingMode was already set to "instant" and the schema accepts never | instant | thinking | message.