Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Subagent completion direct announce still fails when the requester wake path is stale: logs show repeated Subagent completion direct announce failed errors after queue delivery reports transcript_commit_wait_unsupported and no_active_run.

Steps to reproduce

1. Run a background subagent workflow where the requester session has already completed or can no longer accept transcript-commit wake delivery.
2. Let the child subagent complete and trigger completion announce delivery back to the requester.
3. Observe the gateway log emit queue-message failures followed by Subagent completion direct announce failed with no visible reply.

Expected behavior

Completed subagent results should still be surfaced to the requester through the supported delivery path. If transcript-commit wake delivery is unsupported and the requester run is stale, OpenClaw should recover with a message-tool handoff instead of repeatedly reporting no visible completion.

Actual behavior

In the observed 2026-05-18 UTC log window, completed subagent runs repeatedly failed direct announce delivery after stale requester wake attempts. The prior related fix attempt, #82804, is closed and unmerged, so current mainline still has the failing behavior.

OpenClaw version

NOT_ENOUGH_INFO

Operating system

Windows host observed in gateway logs; exact Windows version NOT_ENOUGH_INFO

Install method

pnpm dev / local gateway logs

Model

NOT_ENOUGH_INFO

Provider / routing chain

NOT_ENOUGH_INFO

Additional provider/model setup details

NOT_ENOUGH_INFO

Logs, screenshots, and evidence

Scan window: 2026-05-18T03:33:35Z through 2026-05-18T15:33:35Z.
announce_total=131 unique_runs=44
queue_fail_by_reason={"transcript_commit_wait_unsupported":277,"no_active_run":131}

queue message failed: sessionId=[redacted requester session id] reason=no_active_run
Subagent completion direct announce failed for run [redacted run id]: completion agent did not produce a visible reply

queue message failed: sessionId=[redacted requester session id] reason=transcript_commit_wait_unsupported
queue message failed: sessionId=[redacted requester session id] reason=transcript_commit_wait_unsupported
queue message failed: sessionId=[redacted requester session id] reason=no_active_run
Subagent completion direct announce failed for run [redacted run id]: completion agent did not produce a visible reply

Per-run retry pattern: one affected run failed direct announce 3 times from 2026-05-18T04:08:45.158Z through 2026-05-18T04:08:49.050Z. Another affected run failed direct announce 3 times from 2026-05-18T15:24:15.367Z through 2026-05-18T15:24:19.039Z.

Related public state: #82803 was closed by the previous attempt, but #82804 is closed with mergedAt=null, so the fix did not land.

Impact and severity

Affected: background subagent completion delivery back to requester sessions.
Severity: High, because completed child work can fail to surface to the requester and operators see repeated gateway errors.
Frequency: 131 direct-announce errors across 44 unique runs in the observed 12-hour log window.
Consequence: missed completion messages and noisy repeated gateway failures after subagents complete.

Additional information

The failure is grounded in current code in src/agents/subagent-announce-delivery.ts: completion direct delivery first tries to wake the requester session with transcript-commit waiting, then treats an empty direct-agent response as completion agent did not produce a visible reply. It did not recover by forcing message-tool delivery after the wake path proved stale.