openclaw - 💡(How to fix) Fix [Feature] activeUntil / expiresAt for recurring cron jobs [1 participants]

Summary

Add a per-job activeUntil (or expiresAt) field for recurring cron schedules (--every, --cron). When the configured timestamp passes, the scheduler stops triggering the job and (configurably) removes it from ~/.openclaw/cron/jobs.json.

Motivation

OpenClaw cron one-shots (--at) auto-delete after success. Recurring jobs have no expiration knob — they run forever unless an operator manually openclaw cron removes them.

Real use cases that hit this gap:

• "Run this dashboard refresh every 10m until 2026-05-01" — temporary operational duty during a launch window.
• "Generate a weekly digest every Monday 09:00 for the duration of the EU summer pilot, ending 2026-09-30."
• "Poll this incident channel every 5m for the next 24h" — incident-bound recurring task.

Today the only options are:

1. Add the job and rely on the operator to remember to remove it. Ops debt accumulates; orphaned jobs run forever.
2. Build a wrapper script that writes a kill date elsewhere and calls openclaw cron remove itself. Custom-per-deployment.

A native activeUntil collapses this into the scheduler.

Proposed schema

Adding a field to recurring cron jobs:

{
  "name": "priority-dash-refresh",
  "schedule": { "kind": "every", "interval": "10m" },
  "activeUntil": "2026-05-01T00:00:00Z",   // optional ISO-8601 timestamp
  "onExpire": "remove",                     // "remove" (default) | "disable"
  // ...rest of job definition unchanged
}

CLI:

openclaw cron add --name priority-dash-refresh --every 10m \
    --message "..." --active-until 2026-05-01

openclaw cron list would show a column with the remaining lifetime / expiration date.

Behavior

• Before each scheduled run, the scheduler checks activeUntil against now(). If expired, it skips the run and applies onExpire:
  • remove (default): delete the job from jobs.json. Mirrors the auto-cleanup behavior of --at one-shots.
  • disable: mark the job enabled: false but keep the definition for audit / re-enable.
• A timezone-naive activeUntil would be interpreted in the gateway's local TZ, matching croner's existing behavior.
• Non-recurring (--at) jobs reject activeUntil (their schedule already encodes a single execution).
• The audit ledger logs cron-expired at the moment of removal/disable.

Backwards compatibility

activeUntil is optional; existing job definitions need no migration. Default behavior matches today's semantics (run forever).

Why this is worth doing in core

Every operator who runs OpenClaw with non-trivial recurring tasks rebuilds this. Reference example, just from the past week: a multi-agent fleet operator added "refresh dashboard until launch date" and had to either accept the orphan-job risk or write external lifecycle plumbing. Native activeUntil removes that burden and matches the lifecycle ergonomics already present for --at one-shots.

Happy to draft a PR if the maintainers are open to it.