openclaw - 💡(How to fix) Fix Auto-announce completion events not reliably delivering to parent sessions [1 comments, 2 participants]

Problem

Subagent and backgrounded exec sessions with yieldMs are not reliably firing completion events back to the parent session. The parent yields and waits, but the completion message never arrives, requiring manual polling or user intervention.

Observed Patterns

1. Subagent completions: Sometimes arrive as user messages (working), sometimes timeout silently with no message (broken)
2. Backgrounded exec with yieldMs: Almost never announce completion automatically — parent has to poll with process to check status
3. Timeout handling: When a subagent times out, no explicit "timed out" message surfaces — just silence

Impact

• Parent agent promises "I'll check back in X minutes" but has no mechanism to actually do so
• Creates a frustrating user experience where follow-ups require manual prodding
• Defeats the purpose of async work patterns — parent can't trust that completions will arrive

Expected Behavior

• When a subagent completes (success or timeout), parent session should receive a message with the result
• When an exec session with yieldMs exits, parent should receive a completion event
• Timeout events should be explicit, not silent failures

Reproduction

1. Spawn a subagent with sessions_spawn and context: "fork"
2. Parent calls sessions_yield and waits
3. Subagent completes successfully
4. Expected: Parent receives completion message automatically
5. Actual: Completion message sometimes arrives, sometimes doesn't

Similar pattern with exec(command, yieldMs: 120000) — completion often doesn't announce.

Environment

• OpenClaw version: 2026.5.7 (eeef486)
• Runtime: Mac Studio M4 Max, macOS
• Channel: Telegram direct chat

Suggested Fix

The auto-announce delivery routing needs to be more reliable:

• Ensure all subagent completions (success/timeout/error) trigger delivery to parent
• Ensure backgrounded exec completions fire events when process exits
• Add diagnostic logging when delivery fails so we can see where messages get dropped