openclaw - 💡(How to fix) Fix Bug: subagent announce delivery delay (~11min) when main session is busy

Summary

When a subagent completes and sends an announce event while the main session is processing another message, the announce enters a pending queue but is only scanned once and dropped. The delivery is eventually retried, but with a significant delay (observed: ~11 minutes).

Reproduction Steps

1. Spawn subagent A (executor)
2. Before A completes, spawn subagent B (qa) from the announce handler of a previous subagent
3. Subagent A completes at T+0 → announce enters pending queue (main session busy handling B's spawn)
4. Main session eventually becomes idle
5. Announce from A is delivered ~11 minutes after A actually completed

Observed Timeline

• Subagent (executor) completed: 2026-05-12 01:05:34 GMT+8 (timestamp: 1778519134211)
• User received the forwarded announce message: 2026-05-12 01:16 GMT+8
• Delay: ~11 minutes

Expected Behavior

Pending announces should be retried on a short interval (e.g., every 5–10 seconds) until the main session is available. An 11-minute delay makes multi-agent workflows feel broken from the user's perspective.

Environment

• OpenClaw version: OpenClaw 2026.5.7 (eeef486)
• OS: Darwin 25.3.0 (arm64)
• Channel: Feishu (direct message)
• Agent setup: main + executor + qa + researcher (4-agent multi-agent setup)

Additional Context

This appears to be a delivery-recovery gap: the pending queue scanner runs once when the announce arrives, finds the session busy, and does not schedule a follow-up retry. The announce is only picked up when the next unrelated event triggers another queue scan.

Workaround: After spawning subagents, main actively polls subagents list rather than relying on announce delivery.