Bug

Sub-agent task results are silently lost when the announcement delivery to the parent session exceeds the hardcoded 60-second gateway timeout. The sub-agent completes successfully, but the user never sees the result.

Evidence

Gateway logs showing repeated announce failures:

Feb 15 02:20:47: Subagent announce failed: Error: gateway timeout after 60000ms
Feb 14 18:52:22: Subagent announce failed: Error: gateway timeout after 60000ms
Feb 13 12:51:39: announce queue drain failed for agent:main:main: Error: gateway timeout after 60000ms
Feb 13 12:55:45: announce queue drain failed for agent:main:main: Error: gateway timeout after 60000ms

Root Cause

src/agents/subagent-announce.ts → sendAnnounce() calls callGateway(... timeoutMs: 60_000 ...) — hardcoded, no retry, no backoff. On first timeout the announcement is silently dropped.

Steps to Reproduce

1. Spawn a sub-agent via sessions_spawn during moderate gateway load (e.g., concurrent heartbeats, WhatsApp reconnects, multiple sessions active)
2. Sub-agent completes its work successfully
3. Announcement delivery times out at 60s
4. User never receives the result — no error shown, no retry attempted
5. Sub-agent session gets cleaned up, leaving no trace

Impact

• Users think sub-agents "forgot" to report back
• Results are lost despite successful execution
• No indication to the user that announcement failed
• Transcript still exists but user has no way to know to check it

Expected Behavior

• Announcement delivery should retry with exponential backoff (at least 3 attempts)
• Timeout should be configurable via agents.defaults.subagents.announceTimeoutMs (or similar)
• Failed announcements should be logged visibly and queued for retry
• Completed sub-agent sessions should persist longer for recovery

Suggested Fix

1. Make timeout configurable: agents.defaults.subagents.announceTimeoutMs (default: 120000)
2. Add retry with backoff: 3 retries at 60s/120s/240s before giving up
3. Persist on failure: If all retries fail, store the announcement payload for manual recovery via /subagents log
4. Surface failures: Show a visible warning to the user when announce fails (e.g., "⚠️ Sub-agent completed but delivery failed — use /subagents log <id> to view results")

Workarounds

• Lower agents.defaults.subagents.maxConcurrent to reduce gateway contention
• Manually check results via /subagents list + /subagents log <id>
• Re-trigger announcement by asking the main agent to check sessions_history for the sub-agent session

Environment

• OpenClaw v2026.2.14
• Channel: WhatsApp
• OS: Ubuntu Linux (AWS EC2, m7i-flex.large)
• Config: maxConcurrent=4, subagents.maxConcurrent=8

Related

Confirmed by community member in Discord — code path identified in src/agents/subagent-announce.ts.