OpenClaw changed its agent cron system on July 1, adding an on-exit schedule in commit 68bfa42b9b so a gateway supervisor can watch a command, persist one-shot completion, and fire the existing wake pipeline after the process exits.
For agent operators, the watch item is whether background work becomes an auditable runtime contract: the source now separates a turn-owned command from a gateway-owned watcher that survives the turn.
Facts
- Commit
68bfa42b9badds theon-exitschedule kind across CLI flags, protocol schema, cron storage, gateway watchers, and UI presentation. - The design note says agent-spawned background processes die with turn cleanup, so the watcher moves under the gateway supervisor instead.
- The new watcher persists completion before firing the job and fails closed if storing that terminal state fails.
Evidence
The public sources are the OpenClaw commit, its DESIGN-cron-on-exit.md note, the gateway watcher implementation, and the gateway-protocol cron schema. Together they show the trigger, the delivery path, and the caveat that the first version is a one-shot command-exit watcher, not a general process monitor.
Context
This is page-one because it changes the operator model. A long build, deploy, or test can become a first-class wake condition instead of a fragile shell trick inside an agent turn. Builders should watch whether the same pattern appears around PID watching, repeated exits, and delivery diagnostics.
Limits
The commit does not prove production adoption, cross-project standardization, or a released user interface. What to watch: whether OpenClaw documents examples for real operator workflows and whether the deferred PID-watch variant lands without weakening the fail-closed replay guard.