Bug type

Crash (process/app exits or hangs)

Beta release blocker

No

Summary

We are seeing a recurring lifecycle failure in OpenClaw where a subagent or exec child still emits stdout/update events after the parent run/session has already been ended, patched away, or deleted.

The first hard error is:

Agent listener invoked outside active run

After that, the failure cascades into symptoms such as:

• subagent announce timeout
• gateway websocket handshake timeout / closed before connect
• Subagent orphan run pruned ... reason=missing-session-entry

In at least one observed case, a Claude Code child process continued running after its wrapper/supervision chain disappeared and was reparented to PID 1. This suggests the runtime lost ownership of the child before stdout/update draining and completion handoff were fully finished.

This does not appear to require a plugin update or a gateway restart to trigger. The same error chain also appears in incident windows without a nearby restart.

Steps to reproduce

1. Start a subagent or exec-backed child process that can continue emitting stdout near the end of a run.
2. Let the parent run/session reach a completion or cleanup edge quickly, for example:
   • subagent completion
   • agent.wait
   • a run ending with stopReason=toolUse
   • a no-reply/completion path
   • session patch/delete/cleanup while child completion is still in flight
3. Allow the child/supervisor to emit stdout or update events slightly after the parent active-run/session state has already been cleared.
4. Observe that the runtime throws Agent listener invoked outside active run, then downstream subagent announce/resume/gateway behavior starts failing.

Expected behavior

If a child process emits late stdout/update events near the end of a run:

• the runtime should drain those updates before clearing active-run ownership, or
• safely ignore/drop late post-run updates without throwing

Subagent completion/announce should remain reliable, and child processes should not become orphaned.

Actual behavior

Late child/supervisor stdout reaches the runtime after the parent active run has already been cleared, causing:

• Agent listener invoked outside active run

This then cascades into:

• subagent announce timeout
• websocket handshake timeout / closed before connect
• Subagent orphan run pruned ... reason=missing-session-entry

In at least one case, the underlying Claude child process continued running after its wrapper/parent disappeared and was observed with PPID=1.

OpenClaw version

2026.4.5

Operating system

macOS 26.4

Install method

npm

Model

openai-codex/gpt-5.4

Provider / routing chain

openclaw->openai-codex/gpt-5.4

Additional provider/model setup details

Environment

• OpenClaw 2026.4.5
• macOS 26.4
• launchd-managed gateway
• local websocket/gateway traffic on 127.0.0.1
• reproduced during real subagent + Claude Code workflows

Logs, screenshots, and evidence

2026-04-06T14:13:41.117+07:00 [openclaw] Unhandled promise rejection: Error: Agent listener invoked outside active run
    at Agent.processEvents (file:///opt/homebrew/lib/node_modules/openclaw/node_modules/@mariozechner/pi-agent-core/src/agent.ts:533:10)
    at file:///opt/homebrew/lib/node_modules/openclaw/node_modules/@mariozechner/pi-agent-core/src/agent.ts:380:21
    at Object.onUpdate (file:///opt/homebrew/lib/node_modules/openclaw/node_modules/@mariozechner/pi-agent-core/src/agent-loop.ts:539:7)
    at emitUpdate (file:///opt/homebrew/lib/node_modules/openclaw/dist/exec-defaults-uj0McX2k.js:1524:8)
    at handleStdout (file:///opt/homebrew/lib/node_modules/openclaw/dist/exec-defaults-uj0McX2k.js:1546:4)
    at Object.onSupervisorStdout [as onStdout] (file:///opt/homebrew/lib/node_modules/openclaw/dist/exec-defaults-uj0McX2k.js:1610:3)
    at file:///opt/homebrew/lib/node_modules/openclaw/dist/exec-defaults-uj0McX2k.js:1007:21
    at Socket.<anonymous> (file:///opt/homebrew/lib/node_modules/openclaw/dist/exec-defaults-uj0McX2k.js:568:4)
    at Socket.emit (node:events:519:28)
    at addChunk (node:internal/streams/readable:561:12)

Impact and severity

High severity.

User-visible impact:

• subagent completions fail to announce reliably
• websocket/gateway failures surface as abnormal closure symptoms
• runs may appear lost or partially completed

Operational impact:

• child processes can become orphaned and continue running without ownership
• session/run state becomes inconsistent (missing-session-entry)
• completion handoff becomes unreliable
• long-running automation workflows become fragile around cleanup/completion timing

This looks like a lifecycle race in parent run/session cleanup versus late child stdout/update delivery. Even if local wrappers are hardened to reduce orphan symptoms, the upstream runtime sequencing issue remains.

Additional information

No response