Summary

A subagent completion can fail to reach the user when the requester session has already been evicted/inactive and the completion fallback path tries to write through the requester-agent/session transcript path. In a Feishu direct-message session, this surfaced as the generic user-facing error:

Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

Observed behavior

On [email protected], the relevant log sequence was:

per-chat task exceeded 300000ms cap ... evicting from queue
Active requester session could not be woken ... reason=no_active_run
SessionWriteLockTimeoutError: session file locked (timeout 60000ms)
Embedded agent failed before reply: session file locked

The child/subagent had completed, but the parent/requester completion delivery failed after the requester run was no longer active and the requester session file was locked.

Expected behavior

For top-level subagent completions with a usable direct-message delivery target:

• if the requester run is inactive or active wake fails with no_active_run, completion delivery should still send a concise direct/background message to the user when possible;
• long child result text should not be mirrored wholesale into the requester transcript on this fallback path;
• the fallback should avoid depending on a fresh requester-agent turn when the requester session is already inactive/locked.

Environment

• OpenClaw version: 2026.6.1 (2e08f0f)
• Install kind: npm global package
• OS: macOS arm64
• Gateway mode: local LaunchAgent
• Channel observed: Feishu direct message

Local hotfix tested

A local dist-only hotfix was tested against dist/subagent-announce-delivery-DhVEJzi6.js:

1. Add a capped direct text fallback for subagent task completion results.
2. If active requester wake fails or requester is inactive, try direct text completion delivery before the requester-agent handoff.
3. Keep the existing announce-agent fallback if direct text delivery is unavailable.
4. Cap direct text fallback output to a concise head/tail preview to reduce lock-amplifying transcript writes.

Syntax checks passed:

node --check dist/subagent-announce-delivery-DhVEJzi6.js
node --check dist/subagent-announce-HPJnGzNW.js
node --check dist/agent-harness-task-runtime-DiaBUagI.js

Suggested upstream fix

In the source equivalent of src/agents/subagent-announce-delivery.ts:

• route top-level subagent_announce completions with direct-message targets to direct/background text delivery when the requester is inactive or resolveActiveWakeWithRetries returns no_active_run;
• cap or artifact-link long child result text before any user-visible fallback or transcript mirror;
• consider treating SessionWriteLockTimeoutError from the direct announce-agent path as eligible for direct text fallback, not only generated-media fallback.

This appears related to the documented completion-delivery path where OpenClaw tries active requester wake first, then requester-agent handoff. The failure mode is when both the active requester path and transcript write path are unavailable, even though an external direct-message route exists.