OpenClaw merged PR #132890 late Saturday, restoring a small but important automation path: cron and exec-completion system-event turns can start subagents again on CLI-backed agent runtimes.
The fix matters because cron jobs are often where OpenClaw users put the work that should happen without a person staring at the terminal. A nightly digest, a periodic repository audit, or an exec-completion follow-up may need to split work into child runs. Before this change, those internal event turns could hit Gateway validation before the model work began.
What Broke
The pull request says the failure came from internal source identifiers. Cron and exec-completion turns used existing identifiers, cron-event and exec-event, but the shared channel policy did not recognize them as internal non-delivery channels.
That made the runtime treat the identifiers as unknown channels when a system-event turn tried to start a child run. In practical terms, an automation could have the right authority and still fail because the source label was not accepted by the Gateway validation path.
The PR frames the user impact directly: cron system-event and exec-completion turns can now use sessions_spawn without failing with an unknown-channel error.
The Fix
The merged change updates the shared internal non-delivery channel policy so it recognizes both cron-event and exec-event.
The important boundary is that this is not a broader channel-routing change. The PR states that event provenance is preserved, the existing Gateway validation path can accept child runs, and delivery routing plus external channel behavior are unchanged.
That is the right shape for an automation fix. Cron and exec events are internal sources, not user-facing destinations. They need to be valid places for work to originate, without becoming new places where replies are delivered.
Why Automation Users Should Care
Subagents are useful when a scheduled task has natural parallelism or needs isolation. Examples include:
- A cron job that checks several sources and asks child runs to summarize each source.
- A build-completion hook that spawns a follow-up diagnosis task.
- A nightly maintenance job that separates discovery, verification, and reporting into smaller runs.
Without this fix, those patterns could fail at the system-event boundary even when the same spawn worked from a regular chat or another accepted channel.
Verification
The PR includes focused evidence for the exact boundary. Before the fix, owner-boundary reproduction showed both cron-event and exec-event rejected. After the fix, both were accepted.
The tests also extend the existing classification table and Gateway acceptance table for both channel and replyChannel. Targeted formatting and git diff --check passed.
This is a small PR by line count, but it lands in a high-leverage part of OpenClaw: automation that can safely delegate work without pretending to be a normal message channel.