Summary

When a user message triggers an internal sessions_spawn subagent run, the main session may already send a successful reply, but Telegram continues showing typing for a long period (e.g. ~65s) until the internal subagent finishes. In many cases, no second user-visible message is sent.

This creates a confusing UX:

• user receives a valid reply
• typing continues as if assistant is still composing another reply
• typing disappears without additional output

Environment

• OpenClaw: 2026.2.24 (stable)
• Channel: Telegram (direct chat)
• Gateway: local loopback (127.0.0.1:18789)
• Observed on Linux host

Timeline (example)

• 08:10:00 user sends message
• 08:10:09 main session triggers sessions_spawn (internal sub-session created)
• 08:10:22 main session reply delivered to user
• 08:10:22 ~ 08:11:21 Telegram typing indicator remains active
• 08:11:21 internal sub-session completes; typing disappears
• no second message is delivered

Expected behavior

After the main user-visible reply is delivered, typing should be cleared unless there is an explicit pending user-visible follow-up message.

Actual behavior

Typing remains active while internal background/subagent work is still running, even if no follow-up message will be sent.

Impact

• Misleads users into waiting for a second message
• Feels like "stuck typing" / "ghost reply"
• Especially bad in production Telegram workflows

Suspected area

• Typing lifecycle around sessions_spawn / follow-up queue
• Idle/cleanup path not decoupled from internal background run state

Suggested fixes

1. Decouple typing state from internal subagent lifecycle once a final user-visible message is sent.
2. Add hard timeout for typing indicator renewal in Telegram path.
3. Only keep typing alive if a concrete outbound user message is pending.
4. Add structured logs for typing start/renew/clear with sessionKey + runId for easier diagnosis.

Notes

This is reproducible in real usage where main chat dispatches internal subagent runs for operational tasks.
Related: #27317 (sub-agent silent message drop when output exceeds Telegram 4096-char limit) — that issue covers the "no reply at all" case; this issue covers the "typing indicator leak" even when a reply was successfully sent.